Free-Space Laser Communications

OPTICAL AND FIBER
COMMUNICATIONS REPORTS
Editorial Board : A. Bjarklev, Lyngby
D. Chowdhury, Coming
A. Majumdar, Agoura Hills
M. Nakasawa, Sendai-shi
CG. Someda , Padova
H.-G. Weber, Berlin
OPTICAL AND FIBER
COMMUNICATIONS REPORTS
The Opti cal and Fiber Communications Report s (OFCR) book series provides a sur-
vey of selected topics at the forefront of research. Each book is a topical collection of
contributions from leading research scientists th at gives an up-to-date and broad-
spectrum overview of various subjects.
The main topics in this expanding field will cover for example :
• specialty fibers (periodic fibers, holey fibers, erbium-doped fibers)
• broadband lasers
• optical switching (MEMS or others)
• polarization and chromatic mode dispersion and compen sation
• long-haul transmission
• optical networks (LAN, MAN, WAN)
• protection and restoration
• further topi cs of contemporary interest.
Including both general information and a highly technical presentation of the results,
this series satisfies the needs of experts as well as graduates and researchers starting in
the field. Books in this series establish themselves as comprehensive guides and ref-
erence texts following the impressive evolution of this area of scienceand technology.
The editors encou rage prospective authors to correspond with them in advance of
submitting a manuscript. Submission of manuscripts should be made to one of the
editors. See also http://springeronline.comlseries/481O.
Editorial Board Masat aka Nakasawa

Laboratory of Ultrahigh-speed Optical
Anders Bjarklev Communication
COM , Technical University of Denmark Research Institute of Electrical
DTU Building 345V Communication
2800 Ksg. Lyngby, Denmark Tohoku University
Email: ab@com.dtu .dk 2-1-1 Katahira, Aoba-ku
Sendai-shi 980-8577, Japan
Dipak Chowdhury
Email: nakazawa @riec.tohoku.ac.jp
Corning Inc.
SP-TD-01-2 Carlo G. Someda
Corning, NY 14831, USA Dfil-Universita di Padova
Email: chowdhurdq@corning .com Via Gradenigo 6/A
35131 Padova , Italy
Arun K. Majumdar
Email: someda@dei .unipd.it
LCResearch, Inc.
30402 Rainbow View Drive Hans-Georg Weber
Agoura Hills, CA 91301 Heinrich-Hertz Institut
Email: a.majumdar@IEEE.org Einsteinufer 37
10587 Berlin, Germany
Email: hgweber@hhi.de
Arun K. Majumdar Jennifer C. Ricklin
Free-Space Laser
Communications
Principles and Advances
~ Springer
Arun K. Majumdar Jennifer C. Ricklin
LCResearch, Inc. AMSRL-CI-CN
30402 Rainbow View Drive 2800 Powder Mill Rd.
Agoura Hills, CA 91301 Adelphi, MD 20783
a.majumdar@IEEE.org jricklin@darpa.mil
ISBN-13: 978-0-387-28652-5 e-ISBN-13: 978-0-387-28677-8

Library of Congress Control Number: 2006939943
© 2008 Springer Science+Business Media, LLC

All rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (Springer Science+Business Media LLC, 233 Spring Street ,
New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly
analysis. Use in connection with any form of information storage and retrieval , electronic
adaptation, computer software, or by similar or dissimilar methodology now known or hereafter
developed is forbidden.
The use in this publication of trade names , trademarks, service marks, and similar terms, even
if they are not identified as such, is not to be taken as an expression of opinion as to whether or
not they are subject to proprietary rights.
Printed on acid-free paper.
9 8 7 6 543 2 I
springer.com
Preface
Free-space laser communications, also referred to as optical communica-

tions, is a popular subject in today's technological marketplace. A number
of conferences on this subject have been organized by professional societies
such as SPIE (the International Society of Photo Optical and Instrumenta-
tion Engineering) , OSA (Optical Society of America), and IEEE (Institute of
Electrical and Electronics Engineers) . The evolving technology of free-space
laser communications is emerging as an appealing alternative to RF com-
munications for links between satellites, as well as a promising addition to
terrestrial applications such as video or computer linkups between buildings.
There is a pressing need for more information on laser communications that
is comprehensive enough to provide in-depth knowledge of free-space com-
munications, and that can satisfy the current demands of the research and
commercial needs.
This book has been designed to provide a comprehensive, unified tutorial
to further understanding of the fundamental techniques for laser communi-
cations through the earth's atmosphere. The driving force behind free-space
laser communications is the continuous demand for higher bandwidth to
deliver high-capacity voice, data, and images to the customer. Free-space
propagation distances include ranges that encompass a few millimeters (for
example between optical interconnects in a computer using photonics to
replace metal interconnects), a few meters (such as indoor communications),
a few kilometers (between buildings, campuses, and hospitals), and even up to
thousands of kilometers (such as from an aircraft or satellite to the ground).
Although the concept of free-space laser communications systems has been
around as long as the laser itself, it has not been as widely implemented in
fielded systems as one might expect. There have been several reasons for this,
the principal one being the lack of a demonstratable need for extremely high-
bandwidth wireless systems, and the cost, complexity, and reliability of the
technology involved. There have also been complications with several of the
attempts to field large-scale operational systems. But with the unprecedented
growth of the internet, multimedia, and related data services and image trans-
fer, there has been a tremendously growing appetite for immed iately available
information at high bandwidth. This has given free-space laser communica-
tions a unique niche in the communications market.
v
VI Preface
The need for bandwidth for both commercial and military applications is
growing faster than the ability of the local infrastructure to provide them,
and also faster than the rate at which the infrastructure can be augmented.
Technology development in the areas of higher modulation capability of laser
sources, fast and sensitive detectors, and optical components such as DWDM
(Dense Wavelength Division Multiplexing) and optical amplifiers, all have
helped to achieve higher bandwidth. However, at optical wavelengths, the
interaction of electromagnetic waves with the atmosphere is much stronger
than that at microwave frequencies, and hence weather conditions such as
turbulence and fog limit the performance of optical links.
This book provides a comprehensive overview for readers who require
information about the basics of free-space laser communications, as well as
up-to-date advanced knowledge of the state-of-the-art technologies available
today. We have tried to make this book as useful as possible as a resource
for engineers, scientists, researchers, and students interested in understanding
and designing laser communication systems for atmospheric channels. The
reader will further benefit from the knowledge of how adaptive optics and
related technologies such as MEMS (Micro-Electro-Mechanical Systems)
can be integrated into free-space laser communications. There is continuity
between chapters so that the information and learning from anyone chapter
is connected to that of the other chapters.
Chapter 1 is an introduction to atmospheric laser communication and dis-
cusses the overall laser communications architecture. This chapter serves as a
general background for the concepts presented in subsequent chapters.
Chapter 2 describes the impact that atmospheric effects have on laser com-
munication systems. This chapter describes the physical processes underly-
ing optical extinction and optical turbulence-induced signal losses. Recent
advances in the modeling of optical scintillation under weak-to-strong fluc-
tuations associated with both terrestrial links and satellite/ground links have
been addressed . This and other information given in this chapter enable cal-
culation of the anticipated system performance over a wide range of channel
characteristics for ground-to-ground, ground-to-air/air-to-ground, and satel-
lite uplink/downlink optical communication scenarios.
Chapter 3 provides a theoretical treatment and analysis of the perform-
ance of a free-space laser communication system under a given communi -
cation format scheme. This chapter develops a communication link budget
model relating the atmospheric channel effects and the main parameters of
the transmitter and receiver stations. Numerous examples are given for spe-
cific system parameters in order to design a specificsystem. Various scenarios
of practical importance are described: horizontal (terrestrial) links, and slant
paths including downlink (satellite-to-ground) and uplink (ground-to-satel-
lite). This system performance analysis will be useful for designers in optimiz-
ing the optical station main parameters in order to establish a data link with
adequate performance under various atmospheric conditions.
Preface Vll
Chapter 4 provides an overview of state-of-the-art, as well as future optical

transmitter and receiver,designs that are particularly well suited for free-space
laser communications. Emphasis is placed on high-performance photon-effi-
cient systems and technologies suitable for operation in both deep space and
terrestrial optical links. This chapter focuses on practical design options for
the transmitter and receiver blocks. Practical tradeoffs and implementation
issues that arise from using various technologies are presented, and the inter-
play between modulation format, transmitter waveform, and receiver design
are discussed.
Wave-front distortions of optical waves can be mitigated, in principle,
with adaptive optics. Chapter 5 discusses an adaptive optics approach that
has significant advantages for free-space laser communication terminals. This
technique does not require wave-front measurements, which are difficult to
obtain under the strong scintillation conditions typical for many communica-
tion scenarios. It is based on the direct optimization of a performance quality
metric, e.g., the communication signal strength. An experimental adaptive
optics system is also described.
Chapter 6 provides a broadband optical wireless solution for closing the
"last mile" connectivity gap throughout metropolitan networks. This chapter
discusses various network architectures and the usage of "optical repeaters."
Point-to-point and point-to-multipoint solutions are explained for construct-
ing different network architectures. Terrestrial applications for short and long
ranges, as well as space applications, are described. Important design criteria
for connecting the user to the "backbone" by free space optics techniques are
explained.
Error-control codes can help to mitigate turbulence-induced signal fading in
laser communications through atmospheric turbulence. Chapter 7 provides an
in-depth treatment of this critical subject through the use of techniques based
on the statistical properties of turbulence-induced signal intensity fading as a
function of both temporal and spatial coordinates. An optimal maximum-like-
lihood detection scheme is described for correlated spatial-diversity reception.
This chapter also describes an experimental demonstration of the effectiveness
of two temporal domain techniques in a terrestrial link using on-off keying.
The three remaining chapters cover the recent development of longer-
wavelength free-space laser communications. A longer wavelength source
can provide significantly improved transmission through the atmosphere,
as compared to shorter wavelengths. Chapter 8 discusses optical parametric
oscillation devices used for mid-IR (MWIR) data communication through
low visibility conditions. This chapter also describes the design of a MWIR
transceiver that provides high peak power and short pulses for the Multiple
Integrated Laser Engagement System (MILES) program.
Chapter 9 discusses the high-speed modulation properties of a quantum
cascade laser (QCL) suitable for free-space laser communications . A com-
parison of experiments between a QCL and a near-infrared laser-based link
VIII Preface
in controlled low-visibility experiments is discussed. A recent application of

a QCL-based link used for transmission of satellite TV data streams is also
presented in this chapter.
Chapter 10 discusses technology for all-weather long-wavelength infrared
(LWIR) free-space optical communications. Superior transmission through
atmospheric phenomena such as fog, clouds, and smoke, coupled with improve-
ments in LWIR laser and modulator design, makes possible reliable optical
replacements for radio and microwavecommunication links for many applica-
tions. This chapter summarizes the components and techniques developed for
high-speed, full-duplex all-weather infrared communications systems.
We express our sincere appreciation to Dr. Hans I Koelsch, the Springer
Editorial Director for Physics, for his encouragement and interest from the
very start of this project. We also thank all of the contributors, as well as
Ms. Virginia Lipsey, Associate Editor for Physics, and Ms. Vasudha Gandhi,
Editorial Assistant, for their continuous help throughout the preparation of
this book. A.K. Majumdar wishes to thank his daughter Sharmistha for her
help in checking the texts of his sections, his wife Gargi for her encourage-
ment and support, and also his feline friends, Rocky and Sasha, who provided
unconditional companionship for the entire duration of his work in complet-
ing this project. He also wishes to pay tribute to the memory of his brother,
Dr. Nirmal Majumdar of London, who was a constant source of inspiration
throughout his writing of this book . IC. Ricklin thanks her husband Mikhail
for his support and encouragement, as wellas the gray family cat who kept the
monitor's mouse icon under continuous strong observation and control.
Arun K. Majumdar
Agoura Hills, California
Jennifer C. Ricklin
Laurel, Maryland
Contributors
Dr. UHa Birnbacher Dr. Stephen Hammel

Department of Communications Atmospheric Propagation Branch
and Wave Propagation 2558 Space and Naval Warfare
Graz Technical University Systems Center
A-8010 Graz, Inffeldgasse 12, Austria San Diego, CA 92152
birnbacher@inw.tu-graz.ac.at hammel@spawar.navy.mil
43-3 I6-873-7442 (619) 553-4578
Dr. David Caplan Dr. Donald P. Hutchinson

MIT Lincoln Laboratory Engineering Science and Technology
Optical Communications Division
Technology Oak Ridge National Laboratory
244 Wood St., Mail Stop C-245, Bethel VaHey Road, Oak Ridge,
Lexington, MA 02420 TN 37831
doc@li.mit.edu hutchinsondp@ornl.gov
(781) 981-5025; CeH: (978) 852-3888 (865) 574-4730; CeH: (865)
382-8155
Dr. Frank D. Eaton
AFRLlDESA, Air Force Research Prof. Joseph M. Kahn
Laboratory Department of Electrical
3550 Aberdeen Ave. SE, Kirtland Engineering
AFB , NM 87117 Stanford University
frank .eaton@kirtland.af.mil 372 Packard Building, 350 Serra
(505) 853-1091 MaH, Stanford, CA 94305-9515
jmk@ee.stanford.edu
Dr. Michael Gebhart (650) 724-9584
Department of Communications
and Wave Propagation Dr. Svetlana L. Lachinova
Graz Technical University Institute for Systems Research
A-8010 Graz, Inffeidgasse 12, Austria University of Maryland
gebhart@inw.tu-graz.ac.at College Park , MD 20742
43-3 I6-873-7442 sIIachin@maii.umd.edu
IX
x Contributors
Dr. Erich Leitgeb Dr. Jennifer C. Ricklin

Department of Communications Defense Advanced Research Projects
and Wave Propagation Agency (DARPA), Advanced
Graz Technical University Technology Office
A-8010 Graz, Inffeldgasse 12, Austria 3701 N. Fairfax Drive, Arlington,
leitgeb@inw.tu-graz.ac.at VA 22203
43-316-873-7442 Jennifer.Ricklin@darpa.mil
(703) 526-4751
Dr. Arun K. Majumdar
LCResearch, Inc. Dr. Mikhail A. Vorontsov
30402 Rainbow View Drive, Agoura Army Research Laboratory
Hills, CA 91301 AMSRL-CI-S
a.majumdar@IEEE.org 2800 Powder Mill Road, Adelphi,
(818) 706-8788; Cell: (818) 620-7918 MD 20783-1197
mvorontsov@arl.army.mil
Prof. Rainer Martini (30I) 394-0214
Department of Physics and
Engineering Physics Dr. Thomas Weyrauch
Stevens Institute of Technology Intelligent Optics Laboratory
Hoboken, NJ 07030 Inst itute for Systems Research
rmartini@stevens-tech.edu University of Maryland
(20 I) 216-5634 A.V. Williams Bldg., Mail Stop
1103, College Park, MD 20742
Dr. Narasimha S. Prasad weyrauch@umd.edu
NASA Langley Research Center,
468/LEOB Prof. Edward A. Whittaker
5 North Dryden Street , B1202 Department of Physics and
Hampten, VA 23681-2199 Engineering Physics
n.s.prasad@larc.nasa.gov Stevens Institute of Technology
(757) 864-9403 Hoboken, NJ 07030
ewhittak@stevens-tech.edu
Dr. Roger K. Richards
Engineering Science and Technology Dr. Xiaoming Zhu
Division QUALCOMM, Inc.
Oak Ridge National Laboratory 675 Campbell Technology Parkway
Bethel Valley Road , Oak Ridge, Suite 200 Campbell, CA 95008
TN 37831 xzhu@qualcomm.com
richardsrk@ornl.gov (408) 423-0346
Contents
Preface v
Arun K. Majumdar and Jennifer C. Ricklin
Contributors IX
Introduction
Arun K. Majumdar
Atmospheric channel effects on free-space laser communication 9

Jennifer C. Ricklin, Stephen M. Hammel, Frank D. Eaton and Svetlana L. Lachinova
Free-space laser communication performance in the atmospheric channel 57

Arun K. Majumdar
Laser communication transmitter and receiver design 109

David O. Caplan
Free-space laser communications with adaptive optics :

Atmospheric compensation experiments 247
Thomas Weyrauch and Mikhail A. Vorontsov
Optical networks, last mile access and applications 273

E. Leitgeb, M. Gebhart, and U. Birnbacher
Communication techniques and coding for atmospheric turbulence channels 303

Xiaoming Zhu and Joseph M. Kahn
Optical communications in the mid-wave IR spectral band 347

Narasimha S. Prasad
Quantum cascade laser-based free space optical communications 393

R. Martini and E.A. Whittaker
All-weather long-wavelength infrared free space optical communications 407

D.P. Hutchinson and R.K. Richards
© 2007 Springer Science + Business Media Inc.
Introduction
Arun K. Majumdar
LCResearch, Inc.
30402 Rainbow View Drive
Agoura Hills, California 9130 I
Email: a .majumdar@IEEE.org
Abstract. This book covers various topics relevant to free-space laser communica-
tions (FSO lasercom). The purpose of this initial chapter is to provide the technical
background needed to understand and appreciate the following chapters. as well as
to establish the connections and continuity among them. This chapter will present
the most fundamental physical equations and parameters needed to characterize and
analyze free-space laser communications.
1. Introduction
Recent rapid progress in information and communication technologies has exceeded

our expectations for meeting the requirements of multimedia society in the 21st cen-
tury.Free-spaceoptical communication is considered to be one of the key technologies
for realizing very-high-speed multi-gigabit-per-second (multi-Gb/s) large-capacity
aerospace communications. Using lasers as signal carriers, FSO lasercom can pro-
vide a line-of-sight, wireless, high-bandwidth, communication link between remote
sites. Rapidly growing use of the Internet and multimedia services has created con-
gestion in the telecommunications networks and placed many new requirements on
carriers. Laser transmitters offer an intermediate, low-risk means to introduce desired
network functionalities with extremely high bandwidth. The wireless aspect of FSO
lasercom can be a crucial advantage, particularly in local area networks (LANs) and
metropolitan area networks (MANs) where in cities the laying of fiber optic cables is
expensive. When the WorldTrade Centers collapsed on September l lth, 200 I, so too
did the networks of all of the corporations located within the buildings. Hundreds of
thousandsof dollars were lost due to the down-time that these companies experienced.
With the help of a few wireless broadband providers using FSO technology, these
2 Arun K. Majumdar
compan ies were able to get their data networks up and running in a fraction of the time
and considerably less money that it would have cost them to reinstall their wire lines,
be it a fiber or cable network.
FSO lasercom offers substantial advantages over conventional RF wireless com-
munications technology, including higher data rates, low probability of intercept, low
power requirements, and much smaller packaging . FSO lasercom systems have proven
to be a viable alternative to optical fiber based systems in several applications, as the
technology comes closer and closer to providing the 5-nines (99.999%) service that
many corporations require of their data networks .
/./. Brief History
Communications using light is not a recent science . Old Roman records indicate that
polished metal plates were occasionally used as mirrors to reflect sunlight for long-
range signaling. The U.S. military used similar sunlight-powered devices to send tele-
graph information from mountain top to mountain top in the early 1800's. Blinking
lights have been used to send messages from one ship to another for many centurie s.
Alexander Graham Bell performed an experiment with his "Photophone" where he
used sunlight reflected off of a vibrating mirror and a selenium photo cell to send
telephone-like signals over a range of 600 feet. A complete history of optical and laser
communications can be found in the literature [1-2].
In 1895, Sir J.c. Bose ofCalcutta, India gave his first public demonstration of radio
communication, using electromagnetic waves to ring a bell remotely and to explode
some gunpowder [3-5]. As reported by the Daily Chronicle of England : "The inventor
(J.c. Bose) has transmitted signals to a distance of nearly a mile and herein lies the
first and obvious and exceedingly valuable application of this new theoretical marveL"
Invited by Lord Rayleigh , Bose reported on his microwave/millimeter-wave experi-
ments to the Royal Institution and other societies in England [6]. The first successful
wireless signaling experiment by Marconi was not until 1897. Thus, the inventor of
Radio Communication was not Marconi but Sir J.c. Bose. The author appreciates
this opportunity to acknowledge a fellow scientist from Calcutta for his pioneering
work leading to the development of modem wireless laser communications. Interested
readers can read about Bose's contributions further in [7].
During the years 1930-1950, increased carrier frequencies were considered to
achieve greater information capacitylbandwidth: Microwaves of 3-30 GHz, and Mil-
limeter Waves of 30-300 GHz.
Following the first demonstration ofa working laser in 1960 at the Hughes Research
Laboratories in Malibu, California, the first serious development of FSO lasercom was
called the "Lightphone". NASA and the Air Force were especially interested and
forced some new technological developments. During the boom period of optical fiber
installation, civil FSO lasercom technology lay dormant ; but in military and space
laboratories the development never really stopped.
During the last thirty years, great advances have been made in electro-optics and
opto-electronics, and incorporated into today's laser communications systems, mostly
for defense applications. The aerospace/defense activity established a strong founda-
tion upon which today's commercial optical wireless systems are based. Systems have
been developed for ground-to-ground, ground-to-aircraft, ground-to-satellite , satellite -
to-satellite, and even satellite-to-submarine applications .
Introduction 3
1.2. Applications
Some typical scenarios where FSO lasercom can be used are

LAN-LAN connections on campuses at Fast-Ethernet or Gigabit-Ethernet Speeds
LAN-to-LAN connection in a city
Speedy service delivery of high-bandwidth access to fiber networks
Temporary network installation (for special events or other purposes)
Re-establishing high-speed connection quickly (Disaster RecoverylEmergency
Response)
Communications between ground and spacecraft, or between spacecrafts, includ-
ing elements of a satellite constellation
Interstellar communication
Ship-to-ship communications with high data rates providing complete security
Figure I illustrates free-space laser communications architectures for ground-to-
ground, ground-to-space, space-to-ground and space-to-space links. These links may
include, for example , a metropolitan network using air-to-air laser communication be-
tween office buildings, a long-distance optical link between an optical ground station
(with adaptive optics) and a building or satellite platform, or an inter-satellite link.
Another area of interest is the broadband wireless solution for closing the so-called
last mile connectivity gap throughout metropolitan networks. Different network archi-
tectures and the usage of "Optical Repeaters ", Point-to-Point, and Point-to-Multipoint
solutions are explained in the sixth chapter ["Optical networks, last mile access and
applications", by E. Leitgeb, M. Gebhart and U. Birnbacher (001 1O.1007/s10297-
004-0025-x)].
1.3. Advantages and Challenges
High data rates

High transmission security
No Federal Communications Commission (FCC) license or frequency allocation
required
Light weight, small volume, and lower power consumption, providing a potential
edge over RF communication
Portability and quick deployment
Increased security due to the laser's narrow beam-ideal for the wireless transfer
of financial, legal, military, and other sensitive information
1.4. Limitations
For many cases of practical interest, the constant presence of optical turbulence in
the atmospheric channel is the limiting factor in reliable free-space laser communica-
tion performance [8]. Other atmospheric effects that can degrade performance include
4 Arun K. Majumdar
Int ers atellite

link
Aircraft.U'A'v'
platform
~
Merroplitnn link Sate llite-

ground link
~~o /
g~ ~
..----- DD
00
.D ~
~
~ Optical gro und
o 0 --------. stat ion w ith
Lon g distan ce
opticnl link
Fig.T, Free-Space LaserCommunications Architectures for ground-to-ground, ground-to-space,

space-to-ground, and space-to-space terminal links.
absorption and scattering of the laser beam. Another limitation of light beam commu-
nications is the requirement of a clear line-of-sight path between the laser transmitter
and the receiver, since light cannot penetrate trees, hills, buildings, etc. A further \im-
itation, sometimes overlooked, applies to the position of the sun relative to the laser
transm itter and receiver, to avoid having the receiver look into the sun.
1.5. Basics ofOperation
The object of the communication system is to transfer data or information from a

transmitter at one point to a receiver at another through the intervening atmospheric
channel withan acceptable error rate while providing high reliability. The transmitter
consists basically of a laser, a modulator (with a selected modulation scheme), a laser
driver (to provide necessary power supply), and a telescope. Depending on the modu-
lation format, the modulator converts bits of information into an electrical signal and
modulates the laser to generate an optical signal. The telescope expands the optical
beam to reduce diffractive effects that spread the beam out.
The optical signal propagates through the atmospheric channel and is collected
by the receiver to recover the data . The receiver consists of a telescope, a detector,
and a decoder to get the signal out. The telescope collects the optical signal and
contracts the beam size small enough to fit onto the detector. The detector converts
the optical signal back to an electrical signal. Generally for a direct detection system ,
the receiver consists of an avalanche photodiode (APD) photodetector, a matched filter
implemented in the form of a moving integrator, and a clocked comparator. The clocked
comparator performs a threshold test at bit interval boundaries.
Introduction 5
2. Understanding Free-Space Laser Communications Systems Performance
The chapters of this book describe and explain different aspects of lasercom; but all
address one common question : How can we use optical wavelengths to accomplish
high data rate performance under different scenarios and atmospheric conditions?
2. J. Bit Error Rate
One standard measure for laser communication system performance is the available
post-detection signal-to-noise ratio (SNR). While this is most relevant to describe ana-
log systems, another measure is more meaningful for digital communications. When
we send information by digital transmis sion, the desired message is converted to binary
symbols (bits). The transmission of digital bits over the optical link is then done on
a bit-by-bit basis (binary encoding). Note that optical bit modulation can be accom-
plished using either a pulsed laser beam or a continuous-wave (CW) laser. In either
case, the real performance measure in digital communications is provided not directly
by the SNR, but rather by the probability of error, also referred to as bit error rate
(BER). We can define the BER as the probability that an error may occur in a bit
in the pulse train, i.e., a "I " bit turns into a "0" bit or vice versa. For commercial
telecommunications applications BER = 10- 9 is a typical required value.
In general , the BER is given by
BER = Po /2 + Pl/2 , (I)
where Po is the probability of mistaken "0" for" I", and PI is the probability of
mistaken "I" for "0" . The 1/2 multiplicative factor comes from the fact that, in a
general digital communication system, "0" and " I" are equally likely.
2.2. Fundamental Limit of Light Detection
One of the principal limitations in optical communications is the power sensitivity of the
receivers . As detection devices based on semiconductor materials are technologically
improved, the minimum amount of light power required in the detection process is
reduced . A natural question is, what is the minimum amount of photons we can have
so that the detector detects a "I " bit with acceptable precision? Even if we could grow
absolutely perfect semiconductor materials and could suppress any possible noise in the
detection circuit , there is still a fundamental limitation which is entirely independent of
the detection device or system : the quantum fluctuation of light. This manifests itself
as random or pseudorandom change s in the number of photons present in the optical
signal, yielding intensity fluctuations which can produce a bit error. The quantum
fluctuations, as do any other noise source present in the system, set a lower limit on
the BER of the overall communication link.
We can use Eq. (I) to estimate the BER due to quantum fluctuation. The photon
occupancy number P( n) gives the probability of detecting n photons . For low optical
signal levels, we can assume that the photon occupancy number distributes with Poisson
statistics expressed as
P(n) = N; exp( -Np)/n! , (2)
6 Arun K. Majumdar
where N p is the average number of photons sent in the signal for a single bit of
data . In the case of "0", N p = 0, and in the case of "I", N p = I or more. The
Poisson distribution assumption is reasonable because the light being detected is being
produced by a stimulated emission which distributes as a coherent field state. The
probability of detecting zero photons provided that an average of N p photons have
been sent is therefore :
P(n = 0) = exp( -Np ) . (3)
Note that P(n > 1) = 0 for the case of N p = 0, since a total lack of photons can
never give anything other than a "0". We are considering the best possible detection
systems where thermal fluctuations have been neglected . Solving for N p from Eqs. (3)
and (I), and using a typical required BER, yields
Np = -In(BER) = -In(1O- 9 ) ~ 20 photons. (4)
This is the minimum number of photons that have to reach the detector to decide" I"
with an acceptable BER of 10- 9 • This is a fundamental limitation that comes directly
from the statistical nature of the radiation and sets a limit to any improvement in
semiconductor detectors.
2.3. BERfor a Random Stochastic Communication Channel
When the communication channel is a turbulent atmosphere with random variations

(both spatial and temporal) of refractive index, system performance is degraded by
turbulence-induced beam spreading, angle-of-arrival fluctuations, and scintillation.
In the presence of these temporal fluctuations, the HER is considered a conditional
probability that must be averaged over the probability density function (PDF) of the
random signals to determine the unconditional HER. For example , in the case of on-off
keying (OOK) for pulsed modulation in binary direct detection receiver, the HER is
given by
BER = -211 a
00
PI (s)erfc ( (SNR)S)
J2 ' ds ,
2 2(1 s )
(5)
where (SNR) is the mean SNR in the presence of atmospheric turbulence , (is) is the
mean output signal current , and PI (s) is the PDF of the received fluctuating signal at
the receiver that depends on the level of turbulence strength and the propagation path.
In the above equation, erfc(x) is the complementary error function. Often-realistic
theoretical models for the PDF proposed by various researchers include Log-Normal
and Gamma-Gamma distributions. Recently, a method of reconstructing an unknown
PDF of intensity fluctuations directly from the measured higher-order moments has
been reported [9], eliminating the need to assume applicability of theoretical PDF
models. The mean (BER) for a turbulent atmospheric channel is of course always
larger than that of a no-turbulence channel :
(BER )Turbulence > (BER )NoTurbulence. (6)
We describe below some of the techniques (described further in the book) which
have been used to mitigate the atmospheric turbulence effects and hence to improve
BER, that is, to decrease BER.
Introduction 7
2.4. Wavelength Selection Criteria
The choiceof the transmitting laserwavelength willdependuponthe atmospheric prop-

agation characteristics, optical background noise, and the technologies developed for
lasers, detectors, and spectral filters. For a long atmospheric channel, the wavelength
will generallyneed to be restrictedto spectralregionsof very lowatmospheric absorp-
tion.Anotherconsiderationis intensityfluctuation (scintillation) due to turbulence. The
strengthof intensityfluctuations decreasesas A-7/ 6; thus, scintillation and henceBER
can bedecreasedby usinga longerwavelength. Free-spacelaser communications using
longerIRwavelengths aredescribedin theeighth["Opticalcommunications in the mid-
waveIR spectralband", by NarasimhaS. Prasad(DOl 1O.1007/s10297-005-0057-x)],
nineth["Quantum cascadelaser-based freespaceopticalcommunications", by R. Mar-
tini and E.A. Whittaker(DOl 10.1007Is 10297-005-0052-2)], and tenth ["All-weather
long-wavelength infrared free space optical communications", by D.P. Hutchinson
and R.K. Richards (DOl 1O.1007/sI0297-006-0078-0)] chapters as follows: 3-5 J-Lm
range using a solid-state, compact, optical parametric oscillator device; 8 J-Lm using
a Quantum Cascade Laser (QCL); and around 10.6 J-Lm using a compact, RF-driven
waveguide C02 laser.
2.5. Adaptive Optics for Free-Space Laser Communications
Refractive-index inhomogeneities of the turbulent atmosphere cause wave-front dis-

tortions of optical wave and are responsible for severe scintillation which can lead
to power losses at the receiver and eventually to fading of the received signal be-
Iowa prescribed threshold. In principle, wavefront distortions can be mitigated with
adaptive optics, therebyimproving BER. The fifth chapter ["Free-space lasercommu-
nications with adaptive optics: Atmospheric compensation experiments", by Thomas
Weyrauch and Mikhail A. Vorontsov (DOl 10.I007/s 10297-005-0033-5)] discusses a
non-conventional adaptiveoptics approachthat has certain advantages with respect to
its incorporation into the free-space optical communication channel.This techniqueis
based on directoptimizationof a performance quality metric,e.g., the communication
signal strength, with a stochastic parallel gradient descent (SpaD) algorithm. Thus,
BER can be improved for a laser communication system with adaptive optics.
2.6. Coding for Atmospheric Channel
Error control coding can be used over free-space laser communication links to miti-
gate turbulence-induced fading. The techniques are generally based on the statistical
properties of turbulence-induced signal intensity fading, as functions of both tem-
poral and spatial coordinates. The seventh chapter ["Communication techniques and
coding for atmospheric turbulence channels", by XiaomingZhu and Joseph M. Kahn
(DOl I0.1007/s I0297-006-0077-1)] discussesin detail the error-probability bound for
coded OOK free-space optical communication through atmospheric turbulence chan-
nels. This is also a very effective way to improveBER through a randomchannel.
8 Arun K. Majumdar
3. Summary
In this chapter, a general introductory discussion is outlined to help the readers un-
derstand the overall concepts and the objectives for free-space laser communications.
This background material will be useful in making smooth connection s among all of
the chapters in this book.
Acknowledgement
The author would like to thank Dr. Bradley Bobbs for reviewing this chapter and his
useful comments.
References
I. http://www.eee.strath .ac.uklug-infoIl9984/comohl .pdf

2. http://www.ballaerospace.comllasercomm_history_1950.html
3. "The Work of Jagadis Chandra Bose: 100 years of mm-wave Research," by D.T. Emerson
of the National Radio Astronom y Observatory, Tucson, Arizona . It is based on material
presented at the IEEE-MTT-S Internat ional Microwave Sympo sium in Denver, Colorado,
June 8-13,1997.
4. IEEE-MTT-S Internation al Microwave Sympo sium Digest, Volume 2, ISSN 0149-645X,
pp.553-556,1997.
5. D.T. Emerson , IEEE Trans. Microwave Theory Tech. 45 (12),2267-2273 (December 1997).
6. J.e. Bose, Collected Physical Papers (New York: Longmans, Green and Co., 1927).
7. Probir K. Bondyopadhya, "Sir J.c. Bose's Diode Detector Received Marconi 's First Transat-
lantic Wireless Signal of December 190I," (The "Italian Navy Cohere" Scandal Revisited),
Proc. IEEE. 86, I (January 1998).
8. Arun K. Majumdar and Jennifer C. Ricklin, "Effects of the atmospheri c channel on Free-
space laser Communications," Proc. SPIE, Free-Space Laser Communication V,Vol. 5892,
2005.
9. Arun K. Majumdar, Carlos E. Luna, and Paul S. Idell, "Reconstruction of Probability Density
Function of Intensity Fluctuation s Relevant to Free-Space Laser Communication through
Atmospheric Turbulence", Proc. SPIE, Free- Space Laser Communic ations VI, Vol. 6304,
2006.
© 2006 SpringerScience + BusinessMedia Inc.
DOl: 1O.1007/sI0297-005-0056-y
Originallypublishedin J. Opt. Fiber. Commun. Rep.3,111-158 (2006)
Atmospheric channel effects on free-space laser communication
Jennifer C. Ricklin 1, Stephen M. Hammel", Frank D. Eaton", and Svetlana L.

Lachinova"
1 Defense Advanced Research Projects Agency, Advanced Technology Office

370 I N. Fairfax Drive, Arlington , VA 22203
Email: Jennifer.Ricklin@darpa.mil
2 Atmospheric Propagation Branch, 2858
Space and Naval Warfare Systems Center, San Diego, CA 92152
Email: stephen.doss-hanunel@navy .mil
3 AFRLlDESA, Air Force Research Laboratory
3550 Aberdeen Ave. SE, Kirtland AFB, NM 87117
Email: frank .eaton@kirtland.af .mil
4 Intelligent Optics Laboratory, University of Maryland
2107 Technology Ventures Building , College Park, MD 20742
Email: sllachin@mail .umd .edu
Abstract. Free-space laser commun ication offers an attractive alternative for trans-
ferring high-bandwidth data when fiber optic cable is neither practic al nor feasible .
However, there are a variety of deleterious features of the atmospheric channel that
may lead to seriou s signal fading, and even the compl ete loss of signal altogether.
Physical obstructions-such as birds, insects, tree limbs, or other factors--ean tem-
porarily or permanently block the laser line-of-sight. Platform/building motion due
to wind, different ial heating and cooling, or ground motion over time can result in
serious misalignment of fixed-position laser communication systems. But most im-
portantly of all, absorption and scatterin g due to particulate matter in the atmosphere
may significantly decrease the transmitted optical signal, while random atmospheric
distortions due to optical turbulence can severely degrade the wave-front quality of a
signal-carrying laser beam, causing intensity fading and random signal losses at the
receiver.
1. Introduction
Fog, rain, dust, snow, smoke, and other aerosol particulate matter primarily attenu-
ate the signal-carrying laser beam, which to a certain extent can be compensated by
10 Ricklin et al.
increasing the signal gain. Molecular absorption may be minimized by appropriate

selection of the optical wavelength. In contrast, the random fluctuations in the atmo-
sphere's refractiveindex that cause optical turbulencealwaysresult in increased system
bit error rates, especially for near-horizontal propagation paths through long stretches
of the Earth's atmosphere. Optical turbulence-induced signal losses increase as the
transmitter to receiver distance is increased, there is no wavelength "window" where
these effects are avoided (although longer wavelengths fare better), and turning up the
signal gain does not necessarily improve laser beam quality. For many cases of practi-
cal interest, the constant presence of optical turbulence in the atmospheric channel is
the limiting factor in reliable wireless optical communication link performance.
In this paper we attempt to provide some understandingof the physical processes
underlying optical extinction and optical turbulence-induced signal losses, with an
eye towards providing the reader with the necessary tools to evaluate and predict
anticipated laser communication system performance. Optical extinction is caused by
absorption and scattering by molecules and aerosol particles, and we begin with a
description of the basic physical principles behind these processes.There are different
models in use for molecularextinction and for aerosol extinction, and we survey those
that are the most useful in the prediction of signal loss. Next, atmospheric properties
that lead to the formation of optical turbulence are discussed. A comprehensive set
of optical turbulence models is provided to enable estimation of optical turbulence
strength in all levels of the atmosphere. Behavior of the propagating signal beam
in optical turbulence is discussed, which leads to expressions for estimating optical
communication system bit error rates as a function of both communicationsystem and
optical turbulence parameters. Sufficientinformation is provided to allow both system
designers and users to calculate anticipated system performance over a wide range of
channel characteristicsfor ground-to-ground,ground-to-air/air-to-ground, and satellite
uplink/downlink optical communication scenarios.
2. Beam Extinction Due to Atmospheric Aerosols and Molecules
Extinction is defined as a reduction in the intensity of a propagating laser beam.There

are two processes that cause extinction: absorption, and scattering. Both of these pro-
cesses remove energy from the forward propagating beam direction. Absorption and
scattering processes can be further subdivided into two size domains: a molecular
regime, and a regime of larger particles (aerosols). Here we examine the physics that
drives these two processes, and review existing models for describing the effects of
absorption and scattering on laser beam propagation in the atmospheric channel.
2. J. Extinction
Extinction describes the attenuation of a laser beam as it passes through a medium

containing atoms, molecules,and particles. Goody andYung [I] definethe fundamental
law of extinction as that of Lambert, which states that the extinction process is linear in
both the intensity of radiation and in the amount of matter, provided that the physical
state (i.e., temperature, pressure, composition) is held constant.
For the interaction of a monochromatic propagating beam with a non-vacuous
medium over an incremental path length ds, the change in intensity is given by
Atmospheric channel effects on free-space laser communication II
dl; = -Iv{3ext,vds, (I)
where l.. is intensity and {3ext ,v an extinction coefficient, both at the frequency u. Since
the extinction process is linear, this implies that
f3ext,v = {3sca,v + {3ab s, v , (2)
i.e., the total extinction {3v consists of the linear sum of extinction coefficients {3e xt, v
due to absorption and scattering. Furthermore, the various terms resulting from differ-
ent constituents of a medium can be summed. As an example, the absorption due to
specific absorption lines is expressed as
n
{3a bs ,v =L (3abs ,v (i). (3)

i= 1
The optical thickness, or optical depth, of a medium between two points hI and h 2 is
defined as
J
h2
Tv(hl, h2) = {3ext,v (s ) d s . (4)

hi
With knowledge of the optical thickness Tv and the initial intensity 10, the resultant
reduction in intensity after propagation between the two points hI and h2 is given by
(5)
It is important to determine the explicit absorption {3ab s,v for a given frequency v.
Molecules absorb energy in discrete quanta, and this process alters the translational,
vibrational, and/or rotational state of the molecule . The resulting spectral line for the
absorption coefficient can be described by its central frequency, with a shape that is
dependent on several line-broadening features.
2.2. Molecular Extinction
Molecular absorption is a primary factor in beam attenuation. At wavelengths above I

Mm, atoms do not couple strongly with the electromagnetic field and their contributions
to the extinction {3ext,v can be neglected . Furthermore, molecular extinction at these
wavelengths is predominantly due to absorption of incident radiation with only minor
scattering contributions.
Molecules absorb energy in discrete quanta, and this process alters the elec-
tronic, vibrational, and/or rotational state of the molecule . The absorption spectrum of
molecules therefore consists of a series of discrete absorption lines. The bands for rota-
tional spectra are from 10 em wavelengths to 100 Mm; vibrational spectra are typically
from 100 to I Mm. Electronic transitions are in the visible and ultraviolet bands. [2]
For wavelengths of interest to free-space laser communication the vibrational spectra
are of the greatest relevance.
The nominal frequency and strength of a discrete vibrational energy transition in a
molecule can be calculated with quantum mechanical models . However, this discrete
transition is rarely observed since the vibrational transition is generally coupled with
12 Ricklin et al.
rotational transitions, which results in considerable broadening of the line. The broad-
ened line can be described as the sum of individual transitions between a rotational
state of the initial vibration state, and a rotational state of the final vibrational state.
The approach to describe such a broadband line is then to determine the individual
line-profile functions as a function of frequency. These line-profile functions have been
represented as a variety of analytic expressions in different theories. The 'natural' line
shapes apply to the vacuum. In the atmosphere, the line-profile functions are modified
due to mechanisms such as Doppler broadening, or pressure broadening. Additional
broadening occurs if molecules become aggregated or dissolve in water droplets .
As a consequence, determination of the shape and strength of the absorption spec-
trum, and hence the extinction of molecules, from first principles is a complicated
task. For many molecules, these absorption spectra (both the broadband vibration-
rotational transitions and the individual underlying rotational fine structure) have been
measured experimentally under laboratory conditions and have been made available in
tabular form. With the use of these tables, the extinction {3ext ,v for a specific mixture
of molecules (such as encountered in the atmosphere) may be evaluated.
2.3. Molecular Transmittance Codes
The Air Force Geophysical Laboratory (AFGL) has developed three models that de-
scribe transmittance losses due to molecular absorption: LOWTRAN, MODTRAN,
and FASCODE. [3,4] LOWTRAN and MODTRAN are both band models, and FAS-
CODE is a line-by-line model.
The line-by-line model FASCODE determines the optical depth as a function of
either wave number or wavelength. An absorption line has three primary characteristics:
the central position of the line Vo; the strength or intensity of the line S; and the shape
function for the absorption line. The natural line shape (or Lorentz profile) is given by
[I]
8
f(v - vo) = 1l'[(v _ vo)2 + 8 2 J' (6)
where 8 is the line width. This profile is most important in the lower atmosphere under
conditions of relatively high pressure.
A line-by-line model computes the line for each absorbing gas, which necessarily
includes the wings of lines at various distances from the monochromatic line of interest.
The result of such a calculation is the layer optical depth, or an optical thickness. The
transmittance T of a region is the ratio of incident power to emergent power, and the
relation between transmittance and optical depth is implicitly defined by Eq. (5):
Total transmittance for a path of interest is then determined by finding the trans-
mittances for each homogeneous layer or region that composes the path, and then
multiplying them together to get the resultant total transmittance.
A band model such as MODTRAN takes a fundamentally different approach to
the determination of transmittance. A band model calculates the spectrally averaged
atmospheric transmittance. Such a model does not obey the Lambert law, except for the
very narrow spectral windows over which an absorption coefficient can be considered
to be constant, which is much less than a line width. This characteristic of spectrally
Atmospheric channel effectson free-space laser communication 13
averaged models is reflected in the fundamental output of the MODTRAN model: the
full-path transmittance, rather than the optical depth or an extinction coefficient.
The MODTRAN code has evolved over the last 30 years from an initial low-
resolution transmittance code called LOWTRAN . It has been upgraded and augmented
by numerous auxiliary programs, including aerosol modules to predict the extinction
induced by aerosol particles. MODTRAN is available as a FORTRAN code, but using
it without a user interface requires extensive knowledge of the input field formats and
output conventions . Many user interfaces have been developed to make the MOD-
TRAN code more accessible . [5]
2.4. Aerosol Extinction
The previous section considered the problem of modeling molecular extinction. Here
we consider the effects of particles . An aerosol particle is larger than a molecule, but
still small enough to remain suspended in the atmosphere for an extended period. The
diameter range d of atmospheric aerosols is roughly 0.002 J.Lm < d < 100 J.Lm.
There is a large variability in the source of aerosols in the atmosphere as can be
seen from Table I.
Table 1. Estimates of particles smaller than 20 micron radius emitted into or formed in the
atmosphere (106 metric tons/year) [SMIC, 1971) (reproduced from (3)).
Natural
Soil and rock debris* 100-500
Forest fires and slash-burning debris* 3-150
Sea salt 300
Volcanic debris 25-150
Particles formed from gaseousemissions:
Sulfatefrom H2S 130-200
Ammonium salts from NH3 80-270
Nitratefrom NOs 60-430
Hydracarbons from plant exudations 75-200
Subtotal 773-2200
Man-made
Particles (direct emissions) 10-90
Particles formedfrom gaseous emissions:
Sulfate from S02 130-200
Nitratefrom NOs 30-35
Hydracarbons 15-90
Subtotal 185-415
Total 958-2615
*Includesunknown amounts of indirectman-madecontributions.
Atmospheric aerosols are generally size-classified in two primary modes, each

mode having a different genesis . The fine mode consists of particles with diameters
less than d < 2.5 J.Lm. This mode is again divided into an Aitken mode, with particles
d < 0.1 J.Lm created by condensation or nucleation of molecules that grow larger by
14 Ricklin et al.
Chemical Conversion
IHot tpor l of Gases to Low
Volatility Vapors
Condensation
Homogeneous
Nucleation
+
Condensation 1.-------.
Growth of Nuclei Wind Blown Dust
+
Emissions
+
Sea Spray
+
Volcanoes
+
Plant Particles
Diameter, urn
_ Transient Nuclei or +Accumulation Mechanically Generated
Aitken Nuclei Range Range Aerosol Range -
- - - - - - F i n e Particles:-----l---Coarse Panicles--
Fig. 1. Definition and genesis of fine mode and coarse mode for aerosol particle size (6).
coagulation; and an accumulation mode, with particle diameters 0.1 pm < d < 2.5
pm that do not evolve into larger particles . The coarse mode with diameter d > 2.5
pm consists primarily of particles generated by mechanical processes, such as wind
blowing over soil, sand, or water (sea spray).
Particle lifetimes are dependent upon particle size. If the effects of turbulence are
neglected and the only forces on a particle are the gravitational force and the drag force,
then a steady-state or terminal settling velocity can be determined. Let TlOo represent
the time for a particle of diameter d to fall 100m. Then Two ~ 3.6 years for d =0.1
pm, TlOo ~ 33 days for d = 1.0 pm, TlOo ~ 8.5 hours for d = 10.0 pm, and Two ~
7 minutes for d = 100.0 pm. Thus within the first 10m above the sea or land surface
one can find strong gradients in large particles since the current particle population is
heavily determined by the production mechanisms of the very recent past (see Fig. I).
Aerosol composition for open-ocean condition s can be classified into two types :
sea-salt particles, and a smaller (residual) concentration of continental aerosol particles.
Sea-salt particles are produced from sea-spray droplets that have evaporated , with
Atmospheric channel effects on free-spacelasercommunication 15
the particles then serving as condensation nuclei within the high humidity oceanic
environment. These particles can grow in size in regions of high humidity, and shrink
by evaporation as humidity decreases.
As will be seen in the next section, beam extinction caused by a single aerosol
particle depends on the particle's size and composition. Thus in order to evaluate
aerosol extinction for a laser beam propagating through the atmosphere, we need to
know the number of aerosols present, as well as their sizes and composition. The
aerosol particle size distribution (PSD) describes the number of particles as a function
of size. The PSD can be measured by a number of instruments utilizing a variety
of mechanical , aerodynamic, and electromagnetic principles . It is more difficult to
infer the composition of the particles . One technique analyzes the particles once they
have passed through the instrument measuring the PSD, but can only yield general
information about the composition. For chemical characterization, it is more customary
to collect material on filters and analyze the contents in the laboratory. This yields a
detailed composition, but usually with only a very coarse size resolution .
There are considerable experimental difficulties that may be encountered in carry-
ing out measurements of PSD and composition, which makes these experimental tech-
niques less suitable in an operational environment. Hence, models have been developed
describing aerosol concentration in the atmosphere as a function of meteorological or
climatological parameters. A few of these models are detailed in a subsequent section,
but first we tum our attention to Mie theory, which relates aerosol size and composition
to extinction.
2.5. Mie Theory
The models and measurement techniques described in the previous section result in
an aerosol particle size distribution. To determine the effect of atmospheric aerosols in
the channel on laser beam propagation, it is necessary to use particle size distribution
to determine extinction for a given wavelength. Beam extinction is the sum of the
absorption and scattering of laser energy, and both can be calculated using Mie theory.
Extinction characteristics of aerosol particles are computed almost exclusively
for either spherical, or (oblate or prolate) spheroid particles. Although other simple
geometries can be addressed (e.g., ellipsoids, cylinders, and disks), the fundamental
Mie calculation determines the absorption and scattering of an incident plane wave
by a homogeneous sphere [7]. In the lower marine atmosphere and at wavelengths of
interest, major contributions to aerosol extinction result from sea-spray droplets. These
droplets are spherical, implying that Mie theory can be used reliably. For particles that
are roughly the same size or larger than the wavelength of interest, scattering is usually
the dominant term in the extinction equation. For particles substantially smaller than
the wavelength, absorption is dominant.
In Mie theory, the critical input parameters are wavelength and particle size. A key
parameter is the dimensionless size parameter x :
27rr
x = kr = - (7)
A
for a particle of radius r , wave number k, and radiation wavelength A. The dimen-
sionless size parameter x illustrates the nature of the Mie calculation: it is only the
ratio between the particle size and radiation wavelength that matters. For incident light
16 Ricklin et al.
4.,.-----------------,
o 5 10 15 20 x 25
Fig. 2. Extinction efficiency computed for the refractive index n = 1.29 + iO.05. This value
represents a wet sea salt particle near the 10.6 micron region.
of intensity 10 , the total scattered energy due to a single particle with radius rand
refractiveindex n can be determined. This total energy can be equated to the incident
energy on a definedarea C sca , the scatteringcross-section. Likewise, the total incident
energy absorbed by the particle can be equated to an area Cab s, called the absorption
cross-section. It follows that the beam extinction Ce x t , or the total energy removed
from the beam, is given by
Ce x t = Csca + C abs . (8)
The particle size, beam wavelength, and extinction characteristics are related by a
quantity called the extinctionefficiency. A sphericalparticleof radius r hasa geometric
cross-section of 7fT 2 so that
_ Ce x t
Qext - 7rr2 (9)
defines the effective extinction cross-section. [7]
Next we examine the relation between the size parameter x and the effective
cross-section Q ext. In Fig. 2 we show a Mie calculation for the refractiveindex n =
1.29 + iO.05. Computationswere done with a program for Mie scatteringcoefficients
called BHMIEwrittenby Bohrenand Huffman[8]and adaptedfor the IDL languageby
P. 1. Flatau. This refractiveindex is typical for a wet sea salt particle near a wavelength
of A = 10.6 J-Lm. The relative value of Q abs can be deduced in this figure from
Q abs = Q ext - Q sca; absorption is a substantial fraction of the overall extinction for
this combination of real and imaginary parts of the refractive index.
Plots of the extinction cross-section vs. the size parameter are useful to display
the relative change in cross-section as size parameter increases. But it is important to
understand a point emphasized by Bohren and Huffman [8]: refractiveindex and size
parameterare not truly independentvariablesin the sensethat if we varythe wavelength
A to effectivelychange z, the refractiveindex changes as well; the refractive index of
a given substance is a function of wavelength. On the other hand, the particle size r
can be changed to vary x without difficulty.
2.6. Aerosol Models
It is important to be able to assess the aerosol particle concentration with some accu-
racy before invoking a Mie calculation. There are a number of aerosol models that have
been developed to study extinction phenomena. Technical challenges render a detailed
experimental assessment in operational scenarios difficult, so aerosol models are used
instead to predict particle concentration on the basis of climatology or actual meteo-
rological conditions. There is no aerosol model that can accurately predict extinction
under all conditions. This is particularly the case when local meteorological conditions
dominate the aerosol source function. We present modeling possibilities that permit
the use of local geographical and meteorological information to more closely approxi-
mate the actual aerosol conditions. These models are presented in order of increasingly
detailed requirements for local environmental conditions as input data.
The aerosol particle number concentration c(r) is defined as the number ofparticles
per unit volume per radius increment: c(r) = dN jdr. Deirmendjian [9] proposed a
modified gamma distribution that models the number of particles per radius increment
and per unit volume as
c(r) = ~ = arC> exp( -br7) (10)
with
Q
b=-7-' (II)
"I r mod
Here r mod is the mode radius (in /Lm) that corresponds to the radius at which c(r) is
a maximum, and N is the total number density in particles per ern". Note that a, b, Q,
and "I are four adjustable parameters that allow the function to be fit to a large variety
of models. N (r) is a continuous function, but if data from particle size counters are
available (generally consisting of counts per discrete bin size), a histogram of N as
a function of particle size r can be constructed. The concentration N is found by
integrating Eq. (10), and the values of Q and "I can be estimated by a curve fit, leaving
the value of b to be found from Eq. (II).
The Junge function [10] is another well-known distribution function:
dN -v
( ) =--=pr
cr , (12)
dlogr
where c(r) is now defined as the number of particles per unit volume and per loga-
rithm of the radius. Determination of both p and v is dependent upon the particular
characteristics of the aerosol. The exponent v can vary over the range 2 < v < 4, with
the lower value v ::::; 2 typically found in fog conditions. If field experiments provide
particle concentrations as a function of size, then p can be determined from Eq. (12).
It is important to realize that Eqs. (10) and (12) have been suggested for aerosol con-
centrations over land, with only a minimal contribution from sea-salt particles. For the
Junge function, these sea-salt particles cause a deviation from the predicted profile for
the size range I - 10 /Lm, which includes many of the wavelengths of interest for laser
communication applications.
Four models (rural, urban, maritime, and tropospheric) developed by Shettle and
Fenn [3] are based upon either a single log-normal distribution, or the sum of two
log-normal distributions:
18 Ricklin et al.
dN(r) 2
c(r) = ~ = LNdi(r) , (13)
i=1
where l i(r) is the log-normal distribution defined by

2
.( ) _ 1 {_ [In (r/ Pi) ] }
f , r - ra, v ~exp
27f
2'
2a i
(14)
where Pi is the mode radius (radius for which the distribution is maximized), and a,
is a distribution width parameter. These four models are available in the MODTRAN
4.0 code.
The tropospheric model contains only a single log-normal distribution, while the
other three models combine two log-normal distributions . The normalized number
density N, is constrained so that N 1 + N2 = 1, and actual distributions can be scaled
to provide extinction coefficients or visibilities in accord with observations.
The rural model [3] is designed to represent aerosols that have no urban or indus-
trial sources . The aerosol is a mixture of water soluble components and "dust-like"
components. The complex refractive index is determined by the relative weighting
of the components in the mixture. The urban model [3] modifies the aerosol mix by
including combustion products (e.g., soot) and other industrial products . Note that the
refractive index for soot has a relatively large imaginary part (especially in comparison
with other aerosol types for the visible wavelengths).
Aerosols created over the open ocean are different from continental aerosols . Ma-
rine aerosols consist to a large extent of sea-salt particles that are produced from sea-
spray and bursting bubbles, which then aggregate and condense in the high-humidity
environment of the near-sea surface. The maritime aerosol model has two components:
(I) a sea-salt component, and (2) a continental component. The continental component
is identical with the mixture defined in the rural aerosol model except that the largest
particles are eliminated due to assumed fallout by the time the air mass has transported
out to the open ocean .
These four models represent the large range in variability of aerosol attenuation. It
is often desirable to tie a model more closely to the local environmental conditions. The
Navy Aerosol Model (NAM) [II] represents an effort to develop a maritime aerosol
model that accounts for the production of near-surface aerosol by white-capping and sea
spray, and which will permit particle size changes due to changes in relative humidity.
NAM is widely used to model the effects of aerosol extinction in marine environments,
and because NAM is available in the MODTRAN environment it has become one of
the default marine aerosol models.
Similar to the four models developed by Shettle and Fenn based upon the log-
normal distribution, in NAM [11,12] the aerosol distribution is assumed to be repre-
sentable as a sum over four lognormally distributed modes:
dN 3
-d = LNdi(r), (15)
r i=O
where N, is amplitude, and the shape function Ii(r) is a lognormal distribution nor-
malized to unity:
f ,.(r ) -- 1 exp ( - [In(r / Pi

2
)]2) • (16)
../'hai exp(a;/2)Pi 2a i
Atmospheric channel effects on free-space lasercommunication 19
For historical reasons the amplitudes for each mode are defined by {A i} i=0 •. . .3' where
Ai = 10glO Ni , Each shape function j i(r) is characterized by the width a, and the
center radius Pi. For i = 0 , ... ,3 a, = JQ.5 and the center radii Pi are relative
humidity-based corrections to the nominal center radii:
R;(Gf - RHlO /lOO)

(17)
Pi = Gf(l - RHlO /lOO) ,
where Rll n, is relative humidity measured at 10m above the surface, R o = 0.03, R 1 =
0.03, R 2 = 0.24, and R 3 = 2.0 . The coefficients Gf and Gf are the Gerber coeffi-
cients. [13]
NAM has some well-known limitations. NAM was designed to determine ex-
tinction at a single height above the ocean surface, namely IO m. In addition, it was
designed to represent aerosol extinction conditions that occur over open-ocean condi-
tions, and not necessarily the actual conditions (i.e., coastal environments). Because of
the need for an accurate marine aerosol model that depends upon local meteorological
conditions, the Advanced Navy Aerosol Model (ANAM) [12) was developed to cor-
rect some of the limitations and deficiencies found in NAM [II]. ANAM introduces
a height dependence for aerosol particle concentrations from the surface up to several
tens of meters.
The ANAM provides an additional height-dependent mode (index i = 4) for larger
particles (r > 5 jlm) :
dN 4
-d = LN;fi(r) , (18)
T i=O
where j i is defined in Eq. (14), a4 = 0.5, R4 = 8.0, and N 4 = lOA4 with
A4 = -2.5 + 0.07VlO - 0.04h, (19)
where h is height above the water surface, and VlO is wind speed measured at a lO-m
height.
Additional resources are available to permit the definition of a custom aerosol
scenario and thence calculate aerosol extinction. The software package Optical Prop-
erties of Aerosols and Clouds (OPAC) provides optical properties of various aerosol
species within the spectral range of 0.25 to 40.0 jlm. OPAC [14] contains a database
of microphysical properties, as well as the resulting optical properties of cloud and
aerosol components for different wavelengths and humidity conditions. The optical
properties that are provided as output include absorption, scattering, and extinc-
tion coefficients. OPAC also contains a FORTRAN program that allows the user
to extract data from this database, to calculate additional optical properties, and
to calculate optical properties of mixtures of the stored clouds and aerosol com-
ponents. This model includes aerosol components drawn from a variety of sources
[14-16]. OPAC is available on the World Wide Web at http://www.meteo .physik.uni-
muenchen.de/strahlung/aerosol/aerosol.html.
2.7. Atmospheric Attenuation of Laser Power
The transmittance T( s) is given by

20 Ricklin et al.
- /11(8) _ - f3v s
T.11 (8
) -/ - e , (20)
11,0
where the total extinction coefficient {311 comprises the aerosol scattering, aerosol
absorption, molecular scattering, and molecular absorption terms:
{3v = «:
SC 3,V + {3aer
a b s ,v +
{3mol + (3mol
SC3 , V abs ,v · (21)
The total extinction coefficient is defined in units of km - 1.

In a link budget calculation it is necessary to determine the loss La due to atmo-
spheric extinction:
La = 10 log 10 [T(8)]. (22)
For typical clear air conditions, {311 :::::: 0.1; for haze conditions (with a visibility of
roughly 4 km), {311 :::::: 1.0 ; in fog conditions, {311 ~ 10, and for a dense fog with a
visibility of 10 m, {311 :::::: 391. Assume the path length of the desired optical length is 2
km, and assume the atmospheric conditions are hazy with (311 = 1.0 at the wavelength
of interest. Transmittance is then given by T = e- f3v s = e- I X2 = e- 2 , so that the
loss is La = 1OIog Io(e- 2) :::::: 8.7 dB.
3. Channel Effects Due to Optical Thrbulence
Random variations in the refractive index of the Earth's atmosphere-commonly re-

ferred to as optical turbulence-are responsible for random fluctuations in the signal-
carrying laser beam intensity (irradiance) called 'scintillations'. Related turbulence-
induced effects include beam spreading beyond the spreading predicted by diffraction
(which decreases spatial power density incident on the receiver), and a continuous ran-
dom motion of the beam centroid about the receiver ('beam wander'). These effects
work in tandem to create random signal losses at the receiver, resulting in increased
system bit error rates due to signal fading. They can also result in the complete loss
of signal altogether. We review the atmospheric phenomenology giving rise to optical
turbulence, and discuss the various models available to describe and quantify optical
turbulence in the Earth's atmosphere.
3.1. Refractive Index Structure Parameter C~
Optical turbulence occurs when turbulent winds in the Earth's atmosphere mix the
alway s-present vertical moisture and temperature gradients caused by the Sun's heat-
ing of the Earth's surface . This creates irregularities in the refractive index of the atmo-
sphere in the form of eddies, or cells, called 'optical turbules ' . In 1941 Kolmogorov
and Obukhov [17-20] simultaneously applied dimensional analysis to an isotropic
velocity field, and derived inertial-subrange predictions for the forms of the velocity
spectra and velocity structure functions that now bear Kolmogorov 's name [21]. They
found that a subclass of optical turbules within the inertial subrange, defined as the
range of optical turbule sizes bounded above by the turbulent outer scale L; and below
by the turbulent inner scale lo, share a degree of statistical consistency that permits a
meaningful theoretical treatmen t. Here we briefly summarize these results .
The stochastic field describing fluctuations of the refractive index n(p) at a vector
location p in the turbulent atmosphere is characterized by a structure function . Let
n(PI) and n(P2) be values of the refractive index at the vector locations PI and P2'
respectively. We can describe fluctuations in the refractive index using the refractive
index structure function D n , defined as
D n (Pll P2) = ( In(PI ) - n(P2) n, (23)

where (. . . ) represents statistical averaging . For locally homogeneous and isotropic
turbulence the refractive index structure function depends only on the modulus of the
vector separation P = PI - P2 so that D n (PI' P2) = D n (p).
Within the inertial subrange the refractive index structure function is described by
the Kolmogorov-Obukhov two-thirds power law [22]:
o; (p) = C~(h)p2 /3 , La < p < La, (24)

where the proportionality constant C~ (h), called the refractive index structure param-
eter, is a function of wavelength, atmospheric pressure, and atmospheric temperature,
and varies strongly with the height above ground h. The "strength" of optical turbu-
lence is characterized by the value of the refractive index structure parameter, which
' can vary from 10- 17 m- 2 / 3 or less when optical turbulence is "weak", up to 10- 12
m- 2 / 3 or more when it is "strong".
The refractive index n(p) can be characterized by the spatial power spectral density
tP n (1),) , where I>, is a three-dimensional wave vector. The turbulence spectrum links
random fluctuations in the Earth 's refractive index with optical turbulence effects on
laser propagation . From the two-thirds power law, it can be deduced that the associated
power spectral density for refractive index fluctuations is given by the well-known
Kolmogorov spectrum :
(25)
where I>,f = 271"/ La and 1>,0 =

271"/ La are the boundaries of the spatial spectrum
corresponding to the inertial subrange .The spectrum wavenumber", (not to be confused
with the wavenumber k = 271"/ A associated with electromagnetic propagation) is
defined as I>, = 271"/£, where £ represents turbulent eddy size. Small turbulent eddies
correspond to large spectrum wavenumbers, and large turbulent eddies correspond to
small spectrum wavenumbers.
To perform calculations using the Kolmogorov model requires assuming that the
turbulent outer scale La is infinite while the turbulent inner scale La is negligibly small
(typically La is on the order of mm, while La is on the order of meters). Note also that
the Kolmogorov spectrum has a singularity at I>, = O. Tatarskii [23] suggested using
a Gaussian function to truncate the Kolmogorov spectrum in the dissipation range
I>, > 1>,0 :
tP n ( I>,) _
- 0.033 C2n I>,
-11 / 3 exp ( 2/
-I>,
2 )
I>,m , 1>,0 < I>, < 00, (26)
where « ,« = 5.92/Lo is the cutoff at high wavenumbers . However, the Tatarskii spec-
trum also has a singularity at I>, =O.
In the modified von Karman spectrum , the Tatarskii spectrum is altered so that it
is finite and isotropic for I>, < "'0:
22 Ricklinet at.
10 X E_14C,2 [mO
m]
8 a
6
4
2
0
0 12 24 36 48 60 hours 72
Fig. 3. Diurnal C; profile over a three-day period in August 2002. Typical daytime value is
represented by 'a' , and neutral event(turbulence minimum) by 'b' [courtesy Mikhail Vorontsov].
Other atmospheric spectral modelshavebeen proposedthat accountfor a "bump"

seen at high wavenumbers near 1/L; that appears in measurements of temperature data
[24,25]. Sincetherefractiveindexobeysthesamespectrallawas temperature, it follows
that the bump should also appear in the spectrum of refractive-index fluctuations .
However, for many engineering applications the Kolmogorov model, which does not
include consideration of turbulentinner and outer scale effects, is sufficient.
For purposesof understanding the behaviorof optical turbulence it is convenient to
considerseveral distinctatmospheric layerswith uniquefeatures: the surfaceboundary
layer, the convective scaling layer, the tropopauseregion, and the lower stratosphere.
The surfaceboundarylayer is influenced by the aerodynamic roughness of the Earth's
surface features, and temperature gradients due to solar heatingand radiative cooling.
Energydrivingthe opticalturbulence mechanism in this layercomesfromtwosources:
mechanical energydue to windshear, and buoyantenergyfromtemperature gradients.
The vertical extentof thesurfaceboundarylayerisdeterminedby theinterplaybetween
mechanical and buoyancy contributions to the turbulent energy budget, and varies
diurnally from tens to hundreds of meters at night, to perhaps as high as a kilometer
or more on warm, sunnydays. In the surfaceboundarylayer mechanical contributions
dominate and the greatest extremes in optical turbulence behavior are found. In the
tropopause there is often a peak in turbulence as a result of wind shear and the strong
inversionfoundin this region. The couplingof gravitywavesand turbulence from 2 to
20 km above.mean sea level, and a climatology of C~ profiles over the same altitude
region, were described by Nastromand Eaton [26,27].
In the convective scaling layer, mechanical energy no longer dominates the tur-
bulent energy budget and instead buoyancy becomes the driving mechanism. Optical
turbulence is therefore less dependent on the aerodynamic roughness of the Earth's
surface in this layer. The surface boundary layer and the convective scaling layer to-
gether define the planetaryboundary layer, whichextends upward to the height of the
lowest temperature inversion. Above the planetary boundary layer (h rv 10 km) the
Earth's surface features no longerinfluence optical turbulence. In the tropopause there
is often a peak in turbulence as a result of wind shear. Above this level turbulence
strength rapidly decreases; turbulence effects disappearat altitudes around 25 km.
Figure 3 illustratestypical diurnal behaviorfor C~ over a three-day period in Au-
gust 2002.Thisdata wasobtainedat theArmyResearch Laboratory'sA..LOT testbedin
Adelphi,Marylandover a path representative of many free-space lasercommunication
scenarios: a laser transmitterlocatedon top of a two-story building, with the receiver
Fig. 4. Propagation path for C~ data shown in Fig. 3. The receiversystem is on a water tower,
seen here as the small whitedot above the tree line locatedabove the laser transmitter.
located on a water tower 2.5 km away (see Fig. 4). Atmospheric turbulence measure-
ments taken at this site over a one-year period indicate that atmospheric turbulence
strengths for such a path typically vary between 10- 13 and 10- 14 m - 2/3 .
3.2. Optical Turbulence Models
For most terrestrial sites within the first 100m of the Earth's atmosphere , optical tur-
bulence as described by C;(h) is typically within the range 10- 13 _ 10- 15 m- 2 / 3 .
However, it is often desirable to have a more accurate estimate for local turbulence
strength. The refractive index structure parameter C; (h) not only varies as a func-
tion of altitude, but also according to local conditions such as terrain type, geographic
location, cloud cover, and time of day. Several dozen turbulence profile models have
been developed from experimental measurements made at a variety of locations (e.g.,
see [28,29]). Most of these predict similar results, although with the exception of the
PAMELA model they generally represent average values with little or no modification
for local conditions. Most are designed based on the structure of the optical turbu-
lence profile, with the units of measurement for hand C;(h) being m and m- 2 / 3 ,
respectively. Some of the more commonly used models are presented below.
3.2.1. PAMELA Model
The PAMELA model provides estimates of C; within the surface boundary layer. The
required inputs are latitude, longitude, date, time of day, percent cloud cover, and ter-
rain type, as well as a single measurement (or estimate) of atmospheric temperature ,
pressure and wind speed at the desired height h of the C; estimate. For slant paths
24 Ricklin et al.
with a variable height, measurements corresponding to the height of the laser receiver
should be used for the most accurate C~ estimate [30]. PAMELA was adapted from
earlier, more complicated similarity-based optical turbulence models [31-33] . C~ es-
timates predicted by these earlier models were found to compare reasonably well with
experimental data taken at different sites under widely varying environmental condi-
tions [34]. The theory and algorithm upon which the PAMELA model is based are
provided here, and a sample version of PAMELA in Mathcad is given in Appendix A.
Mechanical and buoyancy contributions to the turbulent energy budget are de-
scribed by the process [35]
dE
-=M+B-€ (28)
dt '
where E is mean kinetic energy of turbulence per unit mass, M is rate of energy
production due to wind shear, B is rate of energy production due to buoyancy, and e is
the energy dissipation rate. From the Kolmogorov hypothesis of universal equilibrium,
for steady-state conditions [35]
€= M + B. (29)
In Eq. (29) the rate of energy production due to wind shear is given by
2
8v 1 (30)
M=Km 8h '
I
where K m is the turbulent exchange coefficient for momentum, and v is average wind
speed. The rate of energy production due to buoyancy can be obtained from
B=-Kh (j (g) Bh:'

80 (31)
Here g is acceleration due to the Earth's gravity (9.8 m1s 2 ) , Ki, is the turbulent
exchange coefficient for heat, and 0 is mean potential temperature in degrees Kelvin.
The potential temperature 0 and temperature T are related by the expression () =
T +0.0098h.
The flux profile relationships for the vertical gradients of average wind velocity
and mean potential temperature have similar forms, and are given by [36-38]
8(} (
8h = kv h
T.) (h)L
<Ph ' (32)
where k; is von Karman's constant taken to be 0.4, 1L. is friction velocity, T. is charac-
teristic (or scaling) temperature, <Pm (h / L) is dimensionless wind shear, <Ph (h / L) is
dimensionless temperature gradient (lapse rate), and L is the Monin-Obukhov length.
The Monin-Obukhov length is fundamental in describing the onset of turbulence and
the vertical structure ofturbulence-induced flux profiles in the Earth's surface boundary
layer.
The form of the flux profile relationships varies depending on atmospheric stability
as characterized by the Pasquill stability category P. Stable atmospheric conditions
occur when the Earth's surface begins to cool. Potential temperature increases with
height, so that a parcel of air shifted vertically will return to its original position. Neutral
Atmospheric channeleffects on free-space laser communication 25
atmospheric conditions are associated with the absence of a temperature gradient. If

a parcel of air is displaced it will remain in its new position . Unstable atmospheric
conditions exist when the Earth's surface is being heated so that potential temperature
decreases with height. An air parcel displaced upwards will continue to rise.
The Pasquill stability category P can be estimated using [31,39,40]
1
P=--(4-c
2 w + c)
r, (33)
where Cr is radiation class, and the wind speed class Cw is obtained from
C w -- ~
2' Va < 8 mis,
(34)
Cw = 4, Va> 8 ui]« ,
In Eq. (34) Va is the average wind speed measured (or estimated) at the reference
height h. During the day the radiation class c; is defined by
R
c; = 300 ' (35)
where R is the solar irradiance incident on a unit horizontal area of the Earth's surface
in W/m 2 . A technique for estimating the solar irradiance R from latitude, longitude,
date, time of day, and percent cloud cover is provided in Appendix B. At night, if the
cloud cover is greater than 50% c- = -I; otherwise, c, = -2.
Once the Pasquill stability parameter P is determined it is possible to estimate the
Monin-Obukhov length L [31 ,39,40]:
2
h-
L= [
(0.00435P + 0.0037P3 ) r
( O.5 - 0 .23 IPI+O .0 32 5p ) ] -1 .
(36)
The surface roughness length h; appearing in Eq. (36) is related to the effective
aerodynamic roughness of the Earth's surface features. Shear forces arising from wind
flow over the Earth's surface introduce a vertical gradient into the wind velocity. Above
a height of approximately ten times the surface roughness length the vertical wind
velocity profile is logarithmic . The surface roughness length is related to the average
height in em of the Earth's surface features h f by [41]
hr~exp [-2 .85+1.2ln(hf)] , hf <700 cm. (37)

Table 2 lists measured surface roughnes s length values for typical urban and rural
conditions. [42-45]
In the unstable atmosphere (negative P) flux profiles for the dimensionless wind
shear t[Jm (hi L) and the dimensionless temperature gradient t[Jh (hi L) are given by
[38]
and when the atmosphere is stable (positive P) they take the form
26 Ricklinet al.
Table 2. Surface roughness length valuesfor typicallyencounteredsurface types.
Surface Type Roughness Length (em)

Village 40
Town 55
Light density residential 108
Park 127
Office 175
Central business district 321
Heavydensity residential 370
Grass (5-6 ern) 0.75
Alfalfa 2.7
Long grass 3
Grass (60-70 em) 11.4
Open brush or scrub 16
Wheat 22
Dense brush or scrub 25
Forestclearings or cutovers 32-48
Com 74
Coniferous forest 110
Citrus orchard 198
Fir forest 283
«; (~) =0.74+4.7 (~). (39)
The friction velocity u, is given by [38]
(40)
u. = In (hh ) - 'l/Jm'
r
with the diabatic influence function for momentum 'l/Jmobtained from
-5(h/L) , P>O ,
'l/Jm=
{ In [e~l) (.!¥)2] - 2arctan(y) + ~ ,y = [1-16(h/L)]1/4 , p:s o.
(41)
The characteristic temperature T. is defined as
-H
T. = - - - . (42)
Cpp u .
In Eq. (42) cp is the specific heat at constant pressure given by 1004 J/(kg OK),
and p is atmospheric density in kg/m", The atmospheric density can be obtained from
p = PaI(2.9T) where Pa is atmospheric pressure in millibars and T is atmospheric
temperature in degrees Kelvin, both measured (or estimated) at the height h. An algo-
rithm for estimating the sensible heat flux H is provided in Appendix B. Note that the
characteristic temperature becomes undefined at very low wind speeds . Equating wind
speeds of less than 3 to 3 m/s avoids this singularity in the model while still retaining
reasonable accuracy [30].
Atmosphericchannel effectson free-space laser communication 27
With knowledge of the friction velocity u. it is possibleto estimate the turbulent

exchangecoefficient for heat K h [38]:
K _ kvu.h
(43)
h - <Ph (~)'
Now consider the atmospheric refractive index, which at optical frequencies de-
pendson fourquantities: atmospheric temperature, atmospheric pressure,opticalwave-
length, and specific humidity. Foropticalfrequencies, humidityoverdry land generally
contributes less than I% to the refractive index fluctuations and is typically ignored.
The refractive indexof air is often writtenin the form n = n + Lln, where the average
valueof the refractiveindex n is slightly greater than one and Lln represents random
deviations from this average value. Optical turbulence-induced deviations of the re-
fractive index of air from its average valueat optical frequencies can be approximated
by [46]
3)
Ll = 77.6 x 1O-6Pa (1 7.52 x 10- (44)
n T +).2 '
where). is wavelength in microns. For standardatmospheric temperature and pressure

at sea level(T =288 0 K, Pa =1013 millibars) and a wavelength of one micron,variation
in the refractive index dn is on the order of 3 x 10- 4 •
By assuming that C~ has a representation analogous to that of the potential tem-
perature structure function we can write [35]
2
2 bK h 8n
Cn = £,1 / 3 ( 8h ) ' (45)
where b is a constantcommonly approximated by 2.8. [47]

By ignoring the small contribution to the total differential from fluctuations in
atmospheric pressure, we can differentiate Eq. (44) with respect to the potential tem-
perature e and obtain
8n (-77.6 x 1O-6Pa) (1 + 7.52 x 1O- 3.x- 2 )

(46)
8e (e-0.0098 h) 2
From the chain rule
8n 8n 8e
8h ~ 8e 8h ' (47)
and using Eq. (32) while ignoringthe small contribution due to wavelength it follows
that
8n rv (-77.6 x 1O- 6Pa) T.<Ph (~)

(48)
8h = kvhT2
This completesthe set of equations requiredto estimate the refractiveindexstruc-
ture parameter C~ givenby Eq. (45). An exampleof the PAMELA modelprogrammed
in Mathcadis given in Appendix A.
In Fig.5(a-f) diurnal C~ values estimatedusingthePAMELA modelarccompared
with scintillometervalues measuredover the path depicted in Fig. 4 at ARL'sA-LOT
facility [30]. Foreach diurnal plotan averagedvalueof atmospheric pressure was used.
28 Ricklin et al.
10- 16 10- 16
hours hours
10- 17 10- 17
5 9 13 17 21 25 29 5 9 13 17 21 25 29
13
10-
C:lm- 213
j
14
10-
10- 15
10- 16 10- 16
hours hours
10- 17 10- 17
5 9 13 17 21 25 29 5 9 13 17 21 25 29
10- 13
C: lm- 213
j e 10- 13
C:lm- 213
j
10- 14 N"~J
10- 15
10- 16 -
hours
10- 17 10- 17
5 9 13 17 21 25 29 5 9 13 17 21 25 29
Fig. 5. Comparison of measured and predicted C~ values for (a) 15 December 2002, (b) 16
December 2002, (c) 17 December 2002, (d) 8 February 2003, (e) 12 February 2003, and (f) 13
February 2003.
Three averaged values of wind speed were used, one for after midnight, one for daytime
conditio ns, and one for late evening. Together with a smoothed running average of the
atmospheric temperatu re measuremen ts, these formed the necessary meteorological
inputs to the PAMELA model. In each case the predicted average C~ values obtained
using the PAMELA model compare favorably with the measured C~ values.
3.2.2. G u rvich Model
As shown in many experiments, the dependence of C~ on the altitude h in the surface

boundary layer follows the power law C~ <X h -q , where q is 4/3, 2/3, orO for unstable ,
neutral, or stable atmospheric conditions, respectively [48]. Based on this law, Gurvich
et. al. [49] suggested the following phenomenological model to obtain profiles of C~
in the atmosphere. In this model profiles are classified into four groups dependent on
the measured (or estimated) value ofC~ at the height h =2.5 m.Analytical expressions
that approximate the experimen tal data are given as follows [48,50]:
1. For C~ /2 .5 m > 10- 13 m- 2 / 3 (strong turbulence) :
10- 13..,.-- - - - - - - - - - - - -----,

2 213
C. [m- )
10- 14
10-16'-J---...:::::.......,...-"":>O'~- __
10-184-~~~~~~~~-~~~____l
10 100 1000 h [m) 10000
Fig. 6. Atmospheric turbulence profiles for the Gurvich model for four different values of
C~12 .5 m - Top to bottom: (I) C~12. 5 m = 1.1 X 10- 13 m- 2/ 3; (2) C~12 . 5 m = 10- 14
m- 2/ 3; (3) C~12 . 5 m = 10- 15 m- 2/ 3; (4) C~I2.5 m = 10- 16 m- 2/ 3.
c~ (h) = {2 C~ 12.5 m (h/2 .5)-4/3 l 2.5 m ::; h ::; 1000 m

(49)
Cn 11000 m exp [-(h - 1000)/9000] 1000 m < h
C; 12.5 m (h /2 .5)-2/3 l 2.5 m ::; h ::; 50 m

C~ (h) = C; 150m (h /50) -4 /3 , 50 m :S h :S 1000 m (50)
{
C; 11000 m exp [-(h - 1000)/9000J l 1000 m < h
3. For 6.5 X 10 - 15 m- 2/ 3 ?: C~ 12.5 m > 4.3 X 10- 16 m- 2/ 3,
c~ (h) = { 2 C; 12.5 m (h /2 .5)-2/3 , 2.5 m :S h:S 1000 m (51)

Cn 11000 m exp [-(h - 1000)/9000], 1000 m < h
4. For 4.3 x 10 - 16 m- 2/ 3 ?: C; 12.5 m(weak turbulence):
C2 (h) _ { C; 12.5 m l 2.5 m :S h :S 1000 m (52)

n - C; 11000 m exp [-(h - 1000)/9000]' 1000 m < h
Sample calculations for C;

profiles corresponding to the Gurvich model are shown
in Fig. 6.
Insubsequent studies, Gracheva and Gurvich [51] developed the so-called ' simple'
models of turbulence, obtained from a large number of experimental measurements:
A. Model for the 'best' atmospheric conditions (weak turbulence) :
log 10 [Cn2rrnn
. (h) - 5 . 19 X 10- 16 x 10-0.00086 h]
(53)
= -18.34 + 2.9 x 1O- 4h - 2.84 x 1O- 8h 2 + 7.4 3 X 1O- 13h3
B. Model for the 'worst' atmospheric conditions (strong turbulence) :
30 Ricklin et al.
10-13,,-==::::::---------~
W
C.2 [m- ]
10-'4 _
r
IO- IS : t - - - -__
10-'8+--,---.-r""TT'l.,..,.,---r---,-r-.-rTTTr-'-'-""""""'~
10 100 1000 h [m] 10000
Fig. 7. Atmospheric turbulence profiles for the Gurvich ' simple' models .
Jog10 [Cn2 max (h) - 9 . 5 X 10- 14 X 1O -0.0 0209h]

(54)
= -14.39 + 1.7 X 1O- 4 h - 3.48 X 8 2
1O- h + 9.59 X 1O- 13 h 3
C. Modelfor 'average' atmospheric conditions (geometrical meanvalueof the two
previousprofiles):
Jog10 [C~ av (h)] = ~ {Jog10 [C~ max (h)] + Jog10 [C~ min (h)]} . (55)
Sample calculationsfor C~ min ' C ; max' and C; av are presentedin Fig. 7.
3.2.3. SLC-Day Model
The SubmarineLaser Communication-Day (SLC-Day) model given by [52,53]
0 0 m < h < 19 m
!
4.008 X 1O-13h-1.054 19 m < h < 230 m
C~ (h) = 1.300 X 10-
15
230 m < h < 850 m (56)
6.352 X 1O-7h-2.966 850 m < h < 7000 m
6.209 X 10- 16 h -0 .622 9 7000 m < h < 20000 m
is frequently used to describe daytime turbulence conditions at inland sites.
3.2.4. Hufnagel-Valley Model
The Hufnagel-Valley (HV) model[54,55] is one of the mostpopularmodelsfor inland

sites and daytime viewingconditions since it permits variations in high-altitude wind
speed and near-ground turbulence levels. In this model C~(h) is assigned a sum of
threeexponential decayterms corresponding to a surfaceboundary layer,a strong layer
caused by the high-altitude jet stream, and a background tropopause layer, i.e.,
C~ (h) = A exp (-h/100) + 5.94 x 10- 53 (v / 27)2 h 10 exp (-h/1000) (57)

+2 .7 x 10- 16 exp (-h/1 500) ,
10-1)
. a
10-1)
b
C.2 [m-2IJ] " , C.2 [m-
2IJ
]
\ A=1.7xI0-llm-2JJ
10-14 \ 10-14
\ v =21 m1s
\
\
10-1> \
10-1>
\
\
10-16 10-16
10-11 10-11
10-11 10-11
10 100 1000 10000 10 100 1000 10000
Fig. 8. Altitude turbulence profiles associated with the Hufnagel-Valley model for different
values of (a) structure parameter A, and (b) high-altitude rms wind speed v .
where A is the nominal value of C~ at the ground and v is the estimated high-altitude
rms wind speed in m/s in the 5-20-km altitude range [29]. Commonly used values are
A = 1.7x 10- 14 m- 2/ 3 and v = 21 m/s ("HV-21 model") . To illustrate the effect of
changing these two parameters, two values of nominal ground -turbulence levels and
two values of moderate-to-strong high-altitude wind speeds were used to calculate the
C~ profiles shown in Fig. 8. Note that the ground-turbulence level has no visible effect
above I km.
3.2.5. HV-Night Model
Night-time turbulence conditions at good seeing locations have been modeled using
a version of the Hufnagel-Valley profile modified to fit observations made during
experiments at the Air Force Maui Optical Station [56]. The modified Hufnagel-Valley
(or HV-Night) profile is given by [53]
c~ (h) = 1.9 x 10- 15 exp (-h /100) + 8.16 x 1O- 54h lO exp (-h/1000) (58)
+3 .02 x 10- 17 exp (-h/1500) .
3.2.6. Greenwood Model
Another nighttime turbulence model developed for astronomical imaging from moun-
taintop sites is given by [28,57]
c~ (h) = [2.2 x 10- 13 (h + 10)- 1.3 + 4.3 x 10- 17] exp (- h/ 4000). (59)
The C~ profile models discu ssed above are compared in Fig. 9. As shown, turbu-
lence in the surface layer is typically stronger for the day model s than for the night
models.
32 Ricklin et al.
10- 13..,..-- - - - - - - - - - - - - --,

2 213
C. [m- )
10- 14 t -.......... _
10-18+--,..-.-..-ncrrnc---r-T""T""T"TT,.,..---,.---....-nrrrn........q
10 100 1000 h [m) 10000
Fig. 9. Atmospheric turbulence profiles associated with two day (solid lines) and two night (dashed
lines) models : (I) HV-21 model; (2) SLC-Day model ; (3) HV-Night model; (4) Greenwood
model.
3.2.7. Other Thrbulence Models
Van Zandt et al. [58,59] developed a sophisticated theoretical Aeronomy Laboratory

model based on relationships proposed by Tatarski [23] to calculate profiles of C~
from archival rawinsonde data :
c~ = b 0:.' L~ /3 M 2 , (60)
where b is a constant commonly approximated by 2.8, 0:.' is a ratio of eddy diffusivities
usually taken to be unity, and M is the generalized index of refraction involving atmo-
spheric pressure, absolute temperature, potential temperature, and specific humidity-
all of which can be obtained from rawinsonde data. The motivation was to create a
model that could be used anywhere rawinsonde data exists. One difficulty with this
approach is to relate or parameterize small-scale fluctuations in terms of large-scale
variables that can be measured . In [58] L; was treated unrealistically as a constant
within turbulent layers. Within each slab defined by the vertical resolution of rawin-
sonde data, the value of C~ was calculated using L o = 10m. Another important part
of the model is to specify the probability of turbulence occurrence for the fraction F of
the slab that is expected to be turbulent, where F is determined from a statistical cal-
culation that accounts for stability and mean shear of each slab. Since 1978 the model
has evolved considerably, and now contains a more realistic probabilistic treatment of
L; and F [60]. Model results have shown good agreement with radar measurements.
The AFRL Radiosonde Model [61] is similar in many respects to the Aeron-
omy Laboratory Model in that it uses the relationship between C~ and L o , and the
Richardson's number. However, it is simpler in construction. It assumes that micros-
hears cause turbulence, but the shear used in the Richardson number calculation is not
data-dependent as in the Aeronomy model calculation. Instead, in the AFRL model
a "model atmosphere" is used for calculation. Two separate models are used by the
AFRL model, one for the troposphere and one for the stratosphere. These models are
based on high-resolution wind profiles obtained using measurements of rocket trails.
Atmospheric channel effects onfree-space laser communication 33
y y'
r p
receiver plane
Fig. 10. Propagation geometry.
3.3. Free-Space Laser Communication in Optical Turbulence
Withthe use of anyof the previously describedoptical turbulence modelsit is possible

to estimatethe anticipated rangeof optical turbulence levels fora particulargeographic
location and time. But to fully characterize the effects of optical turbulence on free-
space laser communication, it is also necessary to consider the actual propagation of
the signal-carrying laser beam through this turbulent atmosphere. We begin here by
examining the propagation of a laser beam in free space, and then use this framework
to develop the necessary expressions for predictinglaser communication system per-
formance in a turbulent channel for various communication scenarios. The approach
presented hereenables system users and developers to considerthe variousdesign and
site location trade-offs involved in maximizing system performance.
3.3.1. Free-Space Laser Beam Propagation
Consider a simple free-space Gaussian beam wave model with parabolic wavefront
phase envelope that allows complete characterization of the focusing or diverging
characteristics of a laser beam [62,63]. Since it was first proposed this beam wave
modelhas beenusedin a numberof studiesconcerninglaser beampropagation through
random media. [64-67]
At z = 0 the free-space electric field of a unit amplitude, lowest-order paraxial
Gaussian beam propagating predominantly along the z-axis can be represented in the
form
U (r, 0) = exp [- (~5 + l~o) r

2
] , (61)
where W o is the transmitterbeam radius (beam size), R; is the radius of curvatureof

the phasefront, k = 21r/A is theopticalwavenumber, and r = (x 2 +y2)1 /2 is transversal
distance from the beam center in the transmitter plane. The propagation geometry is
shown in Fig. 10.
After propagating a distance z from the transmitterthe optical field becomes
U(
p, Z
)=exp(jkz)
r + j z exp [- 1 (1
r+ i z W5 + 2R o
jk)
p
2] , (62)
34 Ricklinet aI.
where p = (X'2 + y'2)1 /2 is transversal distance from the beam optical axis in the
receiver plane. In Eq. (62) we have used the transmitter beam parameters [66,67f
' ( ) _ R; - z , z
r z - -R-- , z = -;;- , (63)
o Zd
where the normalized focusing parameter f characterizes focusing properties of the

beam in terms of deviation of the wavefront curvature from the condition of optimal
focusing R; = z, and Zd = kW~/2 is diffractivedistance. In this notation convergent
(focused) beams are indicated by positive R; and divergent beams by negative R o . At
the beam waist R; is infinite and the beam size takes its smallest possible value. In
terms of the focusing parameter convergent beams are defined by f < I , collimated
beams by f = I, and divergent beams by f > 1. In the following text, the functional
dependence of f and Z on propagation distance z is to be understood.
Free-space laser beam characteristics at the receiver are defined by
, f R( z) + z , () Z Z
(64)
rree(Z) = f 2 + Z2 = R(z) , Zree ".: f2+ Z2 = O.5kw 2(z) '
As with the transmitter beam parameters, the functional dependence of free and
Zree on z is to be understood.
The free-space beam size w(z) and phase front radius of curvature R(z) at the
receiver plane can be expressed in terms of the transmitter beam parameters as
1 Z (f2 + Z2)
w ( z) = ,2
Wo (r + z'2) -2, R()
Z = ' (1 (65)
r - r' ) - z'2'
For a collimated beam with the beam waist at the transmitter aperture z = 0 and
f = 1, so that we obtain the well-known expressions describing beam size and phase
front radius of curvature as a function of z:
In terms of the beam size w(z) the average intensity at the receiver is expressed
as
w~
I(p , z) = w 2(z) exp
(_2
w
p2 )
2(z). (67)
Other relations involving these parameters are discussed in [63].
3.3.2. Laser Beam Propagation in Optical Turbulence
Using the notationdevelopedabove, the beam size (radius)after propagatinga distance

zin optical turbulence is given by [66,67]
W (Z ) -
_ Wo
(,2+,.,2)1
r ." z /2 , (68)
5 Other notations used for these quantities include z = D or z = A o• and f = Do or f = e o.

Atmospheric channeleffectson free-space lasercommunication 35
40-,------------...,..-----,
w[cm]
30
20
10
1;. = 1 (coherent)
0+--,---,----.--.,--,--,---.---1
o 500 1000 1500 z [m] 2000
Fig. 11. Collimated beamsize as a function of rangefor several choices of phasediffuser (>.. =
0.785 pm, W o = 2.5 em, and C~ = 10- 14 m- 2 / 3 ) .
where Po(Z) = (0.55 C~k2 z) - 3 / 5 is the coherence length of a spherical wave prop-
agating in optical turbulence, ( is the global coherence parameter, and (. describes
spatial coherence properties of the signal-carrying laser beam as it exits the transmitter
(s = 1 for a coherent beam, (s > 1 for a partially coherent beam).
Equation (68) provides a simple way to calculate the desired transmitter diffuser
characteristics in order to obtain the optimal beam footprint at the receiver for a set of
anticipated turbulence conditions. For example, as shown in Fig. II after propagating
=
2 km through moderate optical turbulence a collimated beam (J\ 0.785 j.tm) with a
transmitter beam size of 2.5 cm will have a beam footprint (diameter) of only 9.6 cm.
With such a small beam footprint the probability of pointing errors is quite high. By
placing a phase diffuser directly in front of the laser transmitter aperture it is possible
to increase the beam size at the receiver to accommodate potential misalignment or
pointing errors. If we choose (8 = 100 the beam footprint after propagating 2 km
through moderate optical turbulence will be 41 cm, and for (8 =500 the beam footprint
will be 90 cm . However, since increasing the beam footprint also reduces the power
incident on a fixed-size receiver, it is important to not significantly overestimate the
necessary receiver beam size.
The average intensity for a unit amplitude laser beam in optical turbulence is
provided by
w~
(I (p)) = w 2 (z) exp
(_2 p2 )
w (z) ,
2
(69)
where w(z) is given by Eq. (68).

The phase front radius of curvature R(z) for a laser beam propagating in optical
turbulence is
z(f 2 + ( Z2)
R( z) = A,- ;-- 2 -2 ' (70)
'+'Z -.,z - T
and the receiver beam parameters are given by
_ R( z) + z Z
Trec (Z) = R(z) , Zrec(z ) = 0.5kw2(z) (71)
36 Ricklin et al.
Finally,the complexdegree of coherence for a laser beam in turbulenceis available

from
where Pc is coherence length of the optical field at the receiver:
2 ",2 2 ) - 1/2
Po 'f/ Po
pc = Po ( 1 + 2w~ Z2 - 2w2(z) (73)
3.3.3. Scintillation Index and Aperture Averaging
The scintillation index afn S describes fluctuations (scintillations) in optical power as

measured by a point receiver. It is related to the log-amplitude variance a~ through
[68]
2 (12 (p, z)) ( 2)

alnS= 2 -l=exp 4a x -1. (74)
(I(p ,z) )
In Eq. (74) I (p, z) is the intensity received at a point on the receiver aperture
after propagating a distance z through the turbulent channel, and a~ (p, z) is the log-
amplitude variance. In the weak fluctuation regime where 4a~(p,z) < I we can ap-
proximate
(75)
The use of Eq. (75) leads to the followingexpression for the scintillation index of
a Gaussian beam wave [62,64]:
2
afn S s::! 4.42 ar Z~fc6 w f (z) + 3.86 ar {0.4 [(1 + 2frcc) 2 + 4z;ec] 5/ 12
(76)
x cos [~6 arctan (1 +2Z2frec
rec ) ] _ .!.! z5/6 }
16 rec ,
where ar(z) = 1.23 C;e /6 Z 11/6 is the Rytov variancefora plane wave.The receiver
beam parameters are given above in Eq. (71). Since the effects of optical turbulence
are accounted for in Eq. (76) by the Rytov variance ar, only the source coherence
parameter (s (not the global coherence parameter () should be used when calculating
R(z) and w(z) for use in Eq. (76). When 4a~(p,z) » 1 (strong fluctuation regime),
results obtained using Eq. (76) tend to overestimate the level of signal fading and the
resulting bit error rates.
Figure 12 illustrates the scintillation index afn S Ip=O as a function of channel
distance z for two optical turbulencestrengths C;.
Note that as turbulencestrength in-
creases by half an order of magnitudefrom 10- 14 to 5 x 10- 14 m- 2 / 3 , the scintillation
index increases dramatically.
After propagating through optical turbulence in the channel, the quality of the
beam footprint at the receiverdeteriorates and begins to break up into random regions
1.5....-----------------,
,
,,
1.0 .
C 2 = 5xlO-14m-21,J ,, "
,
,
,,
,,
0.5
o 500 1000 1500 z [m] 2000
Fig. 12. Scintillation index for a collimated (1' = 1), partially coherent «s
= 100) laser
beam as a function of channel length for different optical turbulence strengths (>' = 1.55/Lm,
1110 = 2.5 ern),
Fig. 13. Typical beam footprintof a laser beam after propagationthroughoptical turbulence.
of high and low intensity (see Fig. 13). For sufficiently long path lengths and/or large
optical turbulence strengths, the loss in beam quality at the receiver causes random,
fluctuating regions of low (or no) signal intensity, leading to noticeable signal fades.
Increasing the size of the receiver aperture relative to the size of these regions of low
intensity "averages" the signal fluctuations, and thus decreases signal fading. This
phenomenon is known as "aperture averaging" .
To describe the effects of aperture averaging on the received signal it is necessary
to multiply the scintillation index by the aperture-averaging factor A [23,67]:
(77)
where A is given by [67]

38 Ricklin et al.
1.0 1.0
\
a
A I A
0.8 \
, 0.8
, \
0.6 0.6
, \
0.4 ,, \
0.4
\ C.2 = 5xlO-14m 213
-
0.2
\
,, 0.2
.... D[cm]
0.0 --- 0.0
0 5 10 15 20 25 0 5 10 15 20 25
Fig. 14. Aperture-averaging factor as a function of the receiver lens diameter for (a) >. = 1.55
11m and (b) >. =0.785 11m (f = I, (s = 100, Wo =2.5 em, Z = 2000 m).
A= - 16/
1r
1
xdx exp {
-D2x2
- - [2+ - -,- - -
p; w; Z2
p~ p~
</>2 ] }
-
w 2 (z)
o
x [arccos (x) - x~] . (78)
In Eq. (78) x = pi D, and D is the receiver aperture diameter. An example of

how to calculate the aperture-averaged scintillation index using MathCad is given in
Appendix C.
The aperture-averaging factor A is shown as a function of the receiver lens di-
ameter D for different beam wavelengths and optical turbulence strengths in Figs.
14 (a, b). Note that the degree of beam divergence does not significantly affect the
aperture-averaging factor A. Also note that more aperture averaging occurs when opti-
cal turbulence is stronger (compare solid lines for C~ = 10- 14 m -2 /3 with the dashed
lines for C~ =5 X 10- 14 m-2 / 3) . This is because for stronger optical turbulence the
coherence length p; decreases relative to the size of the receiver aperture, resulting
in additional aperture averaging. Similarly, the smaller wavelength in Fig. 14 results
in a smaller coherence length po relative to the size of the receiver aperture, and thus
additional aperture averaging occurs.
The effects of aperture averaging on the scintillation index can be seen from Figs.
15 (a, b). The scintillation index for a point corresponds to a lens diameter of zero. As
receiver aperture size increases, the scintillation index decreases significantly. How-
ever, for the case considered here where ,\ = 1.55 /lm , aperture sizes larger than about
6 ern do not lead to any significant addition in aperture averaging. Decreasing the
wavelength by a factor of two to ,\ = 0.785 /lm roughly halves the aperture size re-
quired to "average" the majority of scintillations. It is clear that aperture averaging
can provide a powerful advantage in reducing atmospheric turbulence-induced signal
fades. However,depending on the channel distance z, optical turbulence strength, and
Atmosphericchannel effects on free-space laser communication 39
3. 5 - . - - - - - - - - - . . ,
~a--' 3.5
b
,2
alnS \
\
2.5- 2.5 \
2.0- 2.0
,
\
I
\
1.5 \ .
\ C' = 5xlO-14m-2IJ
\
1.0- 1.0 \ C' = 1O-14m-2IJ
0.5-
D[cm]
0.5 .>
... D[cm]
I I I I
0.0 -+--.-.---'--:;:::=;;::=;==;:=;==;=0=1
o 2 4 6 8 10 0 2 4 6 8 10
Fig. 15. Aperture-averaged scintillation index as a function of receiver lens diameter [1' = I, ( s
= 100, W o =2.5 em, Z =2000m; (a) >. = 1.55 J.tm, and (b) >. =0.785 J.tm] .
0.35-,-----------------,
C.' = 1O-14m-2IJ
C.' = 5xlO-'4m-2IJ
0.25
0.20
0.15
0.10
0.05
O.OO+--.,-..::;..-,.=~:::::::::::;=.--,.---,--..!:::...::.~~
o 500 1000 1500 z [m] 2000
Fig. 16. Aperture-averaged scintillation index as a function of channel distance for different
optical turbulencestrengths and receiver lens diameters [1' = I, ( s = 100, W o = 2.5 em, >. = 1.55
J.tm].
operating wavelength, there is a maximum size beyond which increasing the receiver
aperture does not lead to significant increases in scintillation reduction .
The aperture-averaged scintillation index (J;~ sIP= W(Z ) / 2 is shown in Fig. 16 as
a function of channel distance for 4 ern and 8 em receiver diameters . In both cases
scintillations increase with path length, and for the 4 em diameter receiver these can
be quite severe. Doubling the receiver aperture size decreases scintillations by about a
factor of two.
40 Ricklin et at.
3.3.4. Beam Wander
When a laser beam propagates through optical turbules that are smaller than the di-
ameter of the beam, the quality of the beam wavefront is degraded resulting in laser
beam scintillation (see Fig. 13). However, optical turbules larger than the beam di-
ameter instead tend to cause the entire beam to be deflected, resulting in a random
wander of the beam centroid about the aimpoint called "beam wander" . Beam wander
and scintillation are both important factors for laser communication satellite uplinks,
since the laser beam diameter is smaller than many of the optical turbules encoun-
tered. Satellite downlinks are less affected by beam wander because the laser beam
diameter will have spread somewhat before entering the Earth's atmosphere [69]. For
near-horizontal paths beam wander can be a significant factor in the weak fluctuation
regime, which is defined by the condition
2
0"1 < 1 an d 2 A5/6
0"1 Zrec < 1. (79)
Beyond the weak fluctuation regime, the magnitude of beam wander quickly be-
gins to diminish as the beam wavefront loses coherence due to the accumulation of
wavefront distortions. The effects of beam wander can be compensated by the use of an
adaptive tip-tilt mirror system as discussed in "Free-space laser communications with
adaptive optics: Atmospheric compensation experiments" by Weyrauch and Vorontsov
(DOl: 10. I007/s 10297-005-0033-5) in this publication .
If the beam footprint is recorded over a sufficiently long period oftime, the resulting
beam width is the summation of the short-term beam width in turbulence w(z) given by
Eq. (66), plus any additional widening due to random beam motion about the centroid
position [70]:
W(Z)long =w(z)+<.8~>, <.8~ >=1.52C~Z3W;;-1/3 . (80)
In Eq. (80) the term <.8~ > is the mean square lateral displacement of the beam
from its centroid position .
3.3.5. Bit Error Rate Determination for a Direct-Detection Binary Optical

Communication Link
The most important question for a laser communication system operated in the presence
of turbulence is, how does optical turbulence affect the bit error rate (BER) of the sys-
tem? Here we review a model for a direct-detection binary optical communication link
consisting of a laser transmitter, atmospheric channel containing optical turbulence,
and a maximum likelihood receiver [67]. The effects of the atmospheric channel are
described by the aperture-averaged scintillation index O";~ s - The maximum likelihood
receiver consists of an avalanche photodiode (APD) photodetector, a matched filter im-
plemented in the form of a moving integrator, and a clocked comparator. The clocked
comparator performs a threshold test at the bit interval boundaries .
Assume the photodetector is an APD that uses current gain to reduce the effects
of thermal (Boltzmann) noise, which correspond ingly increases noise of the output
photocurrent above the Poisson shot noise limit. For an average current gain G this
excess noise is characterized by the factor F [71]:
Atmospheric channel effectson free-space lasercommunication 41
F = keffG + (1 - keff)(2 - 1/G) , (81)
where keff is the ratio of the ionization coefficient for holes to electrons. If the APD
were instead a p-i-n photodiode, the value of both the current gain G and excess noise
factor F would be unity.
When more than a few hundred photons are absorbed during a single bit interval,
the APD output photocurrent can be modeled as a Gaussian stochastic process . The
probability density function (PDF) for the output of the moving integrator at t = n
is described by the Gaussian PDF:
_
P(YTb ) - . /2
1
2 exp
(-(YTb- < YTb
2 2
»2) •
(82)
V tt aYTb aYTb
In the absence of atmospheric fading the mean (YTb) and variance a~T in Eq.
• b
(82) are grven by [71]
(YTb) = fj-GPopt = eGnq"

(83)
2
a YTb = 2B (6FG
hf
2 P.
opt
+ 2KBT)
RL '
where nq, =TJPopt/hf is photon flux absorbed by the detector in photons per second,
2B = lin is effective bandwidth (data rate) of the moving integrator, TJ is quantum
efficiency (probability that an incident photon is absorbed), e is electric charge in
coulombs, hf is the photon energy, and Popt is the instantaneous optical power incident
on the receiver in watts. The quantity 2KBTlRL represents thermal noise of the load
resistor as seen by the APD, where K B = 1.38 X 10- 23 J/K is Boltzmann's constant,
T is effective noise temperature in K, and RL is load resistor input to the amplifier in
ohms. For simplicity, background light and photodetector dark current are neglected.
Both experimental and theoretical evidence have shown that, in the weak fluctu-
ation regime, turbulence-induced optical power fluctuations can be considered log-
normal with a PDF given by [68,72]:
P(S)dS
=
~exp
1 (-(lns+~ar~s)2)
2/2
.!.dS
S'
(84)
V 27ra1n S a 1n S
where S is a dimensionless random variable describing optical turbulence-induced

fading in the received signal, and the average of In S is (In S) = _1/2
ar~ s-
For on-off signaling, the optical power levels at the receiver can be represented
under the two signaling hypotheses as either a digital I (ON state) given by
P~Pt = S(Popt), (85)
or a digital 0 (OFF state) given by
P~Pt = cS(Popt) , (86)
where e represents residual laser light in the OFF state due to the source modulator
having an imperfect extinction ratio. Background light and photodetector dark current
noise have been ignored . The mean and variance of the moving integrator output
representing the Gaussian APD photocurrent under the HI (ON state) are
42 Ricklin et al.
(yit,) = e G S i!J (popt) = e G S n¢

(87)
opt ) +
2S(P'
CJ2 =
1 -
Var(ylTb ) -- 2B (6FG
hi 2KBT)
RL
and under HO (OFF state):
(Y~b) = eGcSi!J(popt) = e Ge Sru.

(88)
opt ) +
2cS(P'
o=
CJ2 - Var(yOTb ) -- 2B (6FG
hi 2KBT)
RL
In Eqs. (87) and (88) the variance considers the effects of both excess shot noise
and thermal noise from the APD load resistor.
The maximum likelihood receiver [73] consists of a circuit that implements the
decision rule
Decide HI
> (89)
<
Decide HO
which yields the following quadraticequation for the decision threshold value YTh:
(90)
In the absence of fading, the average receiverprobability of error (bite error rate,
or HER), is given by the well-known expression:
Pr (errorjno fading) = ~ (PFalseAlarm + PMiss)

= Q (YTh - (Y~b)) +Q (YTh - (yit,)) (91)
2CJo 2CJl
where Q(x) is the Gaussian tail integral definedas
2
1 /00 ( 2_t 1 erfc ( x )
Q(x) =,fh exp )
dt = 2 y'2 . (92)
x
The effects of optical turbulence-induced log-normal fadingon receiverHER are

included by averaging Eq. (91) over the log-normal PDF (85):
Atmospheric channel effects on free-space laser commun ication 43
10°' . - - - - - - - - - - - , . . . , 10"
a b
10- 1 10- 1
10-2 10-2
s 10-3
~ 10-
3
6 10- 4
10-4
t:
U.l -I
,':: 10 10-1
III
6
10- 10-<\
10-7 10-7
10-''+---.----,-.,-1--.--+..:...,-,.-, 10-'+--.--,----,-'-, -..-, ---r- l
-50 -40 -30 -20 -10 -50 -40 -30 -20 -10
Popi [dBm] pop. [dBm]
Fig. 17. BER as a function ofreceived power for; (a) A = 1.55 Jlm, W o =2.5 em, Z = 1000 m;
and (b) A = 1.55 Jlm, Wo =2.5 ern, Z =2000 m. Left to right: (I) D =8 em, C~ = 10 - 14 m- 2 / :J;
(2) D =8 ern, C~ =5x 10- 14 m- 2 / 3 ; (3) D =4 cm, C~ = 10- 14 m- 2 / :J; (4) D =4 em, C~
= 5x 10- 14 m- 2 / 3 .
(93)
Figures 17(a,b) illustrate typical effects of optical turbulence on the average BER
for a direct- detection binary optical communication link. These calculations assumed
a current gain of G = 20, ketf = 0.03, 2B = lin
= I Gbps, load resistance RL =
50 ohms , T = 300 K, quantum efficiency TJ = 0.8, and a 1000; I modulator extinction
ratio (E = 0.001). The aperture-averaged scintillation index ar~ s characterizing op-
tical turbulence effects for this receiver system was obtained using eqns . (76)-(78),
where we assumed an average value for arn
s lp=w(z )/2. The average optical power
at the receiver P opt is expressed in dB rnilliwatts, with a I milliwatt optical power
corresponding to 0 dBm . The optical power is obtained from the photon flux n ,p using
the relationship P o p t =hf n ,p /TJ. We assume there are no pointing errors or beam mo-
tion/wander, and that the transmitter/receiver optical system is perfectly aligned . Once
the average optical power is determined, the optical power in dBm can be obtained
from
P opt [dBm] = 10 10glO (Po p t [mW]) . (94)

44 Ricklin et aI.
10°
10-1
10~2
....tU
II)
3
p::: 10-
....
0
t:: 10-4
~
.... 10-5
iii
10-6
10-7
Popt [dBm]
10-8
-65 -60 - 55 -50 -45 -40 -35 -30 -25
Fig. 18. Effect of optical turbulence on BERas a functionof receivedoptical power for collimated
(f= 1) coherent «(8 = I) and partially coherent «(8 = 10) beams (>' =0.785 j.tm, W o =2.5 em,
z =2000 m, D = 10em). Left to right: (I ) (8 = 10, C; = 10- 15 m- 2 / 3 (Ao-rn 8 =0.013); (2)
(8 = 10, C; C;
= 1.2x 10- 14 m- 2 / 3 (AO'rn S =0.05); (3) (s = I, = 10- 15 m- 2 / 3 (AO'rn 8
C;
=0.045); (4) ( 8 = I, = 1.2x 10- 14 m- 2 / 3 (AO'rn S =0.118).
3.3.6. Reducing Optical Turbulence Effects
There are a variety of techniques available to mitigate the effects of optical turbu-
lence. For example, increasing the size of the receiveraperture offers an effective and
simple way to reduce turbulence-induced signal fades. A related approach is to use a
phase diffuser as an inexpensive beamexpander, thus reducing scintillations and signal
fades through an "artificial" aperture averaging effect [67]. Both of these techniques
reduce the correlation radius of the laser beam with respect to the receiver aperture
size. As seen in Fig. 18, scintillation reduction due to "artificial" aperture averaging
can directly translate to a significant reduction in BER. Another related approach is
the use of multiple transceiver apertures to reduce intensity fluctuations [74]. While
these three techniques can be helpful in mitigating BER losses in weak-to-moderate
optical turbulenceregimes, they begin to lose theireffectiveness as turbulence strength
increases. It then becomes necessary to incorporate adaptive optical control into the
laser communication systemitself (75- 78]. This approach is discussed in greater detail
in "Free-space lasercommunications with adaptive optics:Atmospheric compensation
experiments" by Weyrauch and Vorontsov (DOl: lO.lO07/s10297-005-0033-5) in this
publication.
Acknowledgments
The authors thankArnoldTunickfor his carefulreviewof the PAMELA model, as well

as his many helpful suggestions. Jennifer C. Ricklin was an employee of the Army
Research Laboratory when this work was performed.
Appendix A: Mathcad Version of PAMELA Model
PAMELA inputs
lat - latitude in degrees (positive north of the equator)

long -longitude in degrees (positive west of Greenwich)
Nd - number of day in the year [for I Jan, Nd = I, etc.]
gmt - Greenwichmean time (24 hour clock - 15.20 is twenty minutesafter 3 pm gmt)
tstart - define start of 24 hour period for diurnal calculations, in terms of GMT (24
hour clock)
cc - percent cloud cover in eighths (2 indicates 2/8, etc.)
Pa - atmospheric pressure in millibars
Tf - atmospheric temperature in degrees Fahrenheit
vo,v - average wind speed in mls
hr - surface roughness length in meters, describes terrain features
h - height above ground in meters (height of wind, temperature, pressure measure-
ments). This height will also be the height of the Cn2 estimate
Calculate Solar Insolation I, Irradiance R, Sensible Heat Flux H
d - angular fraction of a year for a particular date (degrees)

M - true solar noon (time in hours of the meridian passage, 24 hour clock)
ht - solar hour angle (degrees)
o- solar declination
0: - solar elevationangle
slz - solar zenith angle (degrees)
Tc - transmission coefficient of the atmosphere
A2
I - Solar Insolation in W/m

R - Solar Irradiance in W/m A2
A2
H - Sensible Heat flux in W/m
Calculate Pasquill Stability Class
cr - radiationclass
cw - wind speed class
P - Pasquill stability category
L - Monin-Obukhov length in meters
Calculate Flux Profile Relationships
<Pm - dimensionless wind shear

<Ph - dimensionless temperaturegradient
Pm - diabatic influence function for momentum
Ustar - friction velocity
p - atmospheric density
Tstar - characteristic temperature
Kh - turbulentexchange coefficient for heat
46 Ricklin et al.
Calculate Cn2
dndz - gradient for refractive index fluctuations
E>: eddy dissipation rate
M - rate of energy production due to wind shear
B - rate of energy production due to buoyancy
A_2/3
Cn2 - refractive index structure parameter, in m
Model Inputs
Date, time,and location inputs

Nd := 151 lat := 32 long := 106 tstart := 6
A Greenwich mean time converter can be found at http://greenwichmeantime.com/.
http://www.infoplease.com/ipa/AOOOI769.htmlgives latitudes and longitudes of major
world cities .
Meteorological and Terrain Inputs

cc:=3 v :=2 hr :=O.1 h :=4 P a :=IQOO Tf := 85
3 if v < 3
vo := { v otherwise
Change degrees Fahrenheit to Kelvin
T := ~ . (Tf - 32) + 273 .15

T = 302 .594
Loop over one diurnal cycle
gmt := tstart , tstart + O.l..tstart + 24 .0

tstart = 6 t start + 0 .1 = 6.1 tstart + 24 = 30
A
Calculate Solar Insolation I (Wlm 2)
d:= (Nd-1)· 3::.~ 4

,:= 279.93 + d
M:= 12 + 0.12357sin(d·deg) - 0.004289cos(d·deg) + . I53809sin(2·d·deg)
+ 0.060783cos(2·d·deg)
ht(gmt) := 15(gmt - M) -long
{3:+ , + OA087sin(,.deg) + 1.8724cos(, ·deg) - 0.0 I82sin(2·,.deg)
+ 0.OO83cos(2 ·,.deg)
J:= asin(0.39785sin({3·deg»
a(gmt):= asintsimlat-degj-sintd-rad) + costlat-degj -Id-radj-cosfhttgmtj -deg) )
slz(gmt) := 90.57 - a(gmt) ·57.29
Tc(gmt):= 0.57 + 0.0045slz(gmt)
Itemp(gmt):= I353cos(slz(gmt)·deg) ·Tc(gmt)sec(slz(gmt) -de g )
I(gmt) := 0 if Itemp(gmt) < 0
Itemp(gmt) otherwise
1(18) = 819.659
Calculate Solar Irradiance R and Sensible Heat Flux H (W/m '2)
Bo,o:= 1.07 B5,0:= 0.67

B1,0:= 0.89 B6,0:= 0.59
B2,0:= 0.81 B7,0:= 0.45
B3,0:= 0.76 B8 ,0:= 0.23
B4 ,0:= 0.72
b:= for i E O..ee
x ~ Bi,o if i = ee
x
b=0.76
R(gmt):= b-Itgmt)
H(gmt):= O.4(R(gmt) - 100)
H(gmt):= O.4(R(gmt) - 100) if I(gmt) > 0
-40 otherwise
H(l8) = 209.176
Calculate Pasquill Stability Class
Calculate radiation class Cr

ertemp(gmt):= x ~ R~go~t) if H(gmt) > 0
x ~ -I ifH(gmt) ~ 0 1\ ee 2: 4
x ~ -2 if H(gmt) ~ 0 1\ ee < 4
x
er(gmt):= 3 if ertemp(gmt) 2: 0
ertemp(gmt) otherwise
er(l8) = 2.076
Calculate wind speed class cw

ew:= x ~ Y2° ifvo ~ 8
x ~ 4 ifvo > 8
x
Pt(gmt):= x ~ 0.5 . (4 - ew + I cngmrjl) if H(gmt) ~ 0
x ~ -0.5 · (4 - ew + I er(gmt)l) if H(gmt) > 0
x ~ 0 ifee = 8
x
P(18)= -2.288
Calculate Flux Profile Relationships
Calculate Monin-Obukhov Length L

L(gmt):= (0.004349P(gmt)
+ O.003724P(gmt)3).hr-(O.5034-0.231 IP( gmt)I +0.0325.P(gmt)2) -1
1J(gmt):= L(::nt)
L(l8) =-13.124
48 Ricklin et al.
Estimatedimensionless wind shear pm

pm(gmt):= x t- (I - 16·«gmt))-·25 if P(gmt) :s:: 0
x t- I + 5'«gmt) if P(gmt) > 0
x
pm( 18)=0.642
tJim(gmt):= y t- (I - 16·«(gmt))·25 if P(gmt) :s:: 0

2 2
x t-In l+l . l-tl - 2· atan (y) + ~ if P(gmt) :s:: 0
x t- -5 ·«(gmt) ifP(gmt) >0
x
tJim(l8)=0.6
Estimatedimensionless temperature gradient ph

ph(gmt):= x t- .74(1 - 9· «gmt)) -·5 ifP(gmt):S:: 0
x t- .74 + 4.7·« gmt) ifP(gmt) > 0
x
ph(l8)=0.382
Estimatefriction velocityUstar
Ustar(gmt):= O.4VQ
Inf.-t[Fm(gmt)
Ustar( 18) = 0.389
Estimatecharacteristic temperature Tstar

. Pa
P.= 2 .87·T
-H(gmt)
T star( ) :=
gmt l004p .Ust ar( gmt)
Tstar( 18) = 0.466
Estimate turbulent exchange coefficients for heat Kh
Kh(gmt):= O.4·Usta r (grn t) ·h

4>h(gmt )
Kh(18) = 1.625
Calculate Refractive Index Structure Parameter Cn2
Estimategradientfor refractive index fluctuations

·= -(77 .6 .1Q- 6) .Pa .Tstar( grnt) .4>h(grnt)
dndz(gmt) . OA h .T 2
dndz(l8) = 9.435 x /0-8

Atmospheric channel effects on free-space lasercommunication 49
Estimate eddy dissipation rate e
€(gmt):= Usta~.~hmt)3 . (<Pm (gmt) - ((gmt))
€( 18) =0.035
Cn2(gmt):= 2 .8 .Kh(gmt).dn z(gmt)2
rd
E(g m t )3
Cn2(18) = 1.242 x 10- 13
Appendix B: Calculating Solar Iirradiance and Sensible Heat Flux
The incident solar irradiance R and sensible heat flux H can be estimated from the
latitude, longitude, date, time of day, and cloud cover as follows. The solar elevation
angle (sun's zenith distance) 0: can be obtained from [79]
sino: = sin</> sin <5 + cos</> cos d coelu ; (BI)

where </> is the latitude, <5 is the solar declination, and li , is the solar hour angle . The
following method for calculating the solar declination angle accounts for the slight
asymmetry due to the ellipticity of the Earth's orbit:
sin <5 = sin(23AO) sin {3, (B2)

where {3 is given in degrees by
{3 =, + OA087 sin, + 1.8724cos,- 0.0182sin(2,) - 0.0083cos(2,). (B3)
In (B3)
, = 279 .93 +d . (B4)
The angle d is the angular fraction of a year for a particular date, and is given by
d = (number of d ays in a year - 1) x (3::.~4) . (B5)
For example, on January 1st d = 0, and on March 21st d = 79x(0.985) = 77.815.

The solar hour angle b, is a measure of the longitudinal distance from the sun to the
point of interest on the ground, and can be obtained from
h., = 15(T - M) - TJ. (B6)

where T is Greenwich mean time (GMT) , TJ is longitude (positive west of Greenwich),
and M is the time in hours of the meridian passage (true solar noon) given by
M = 12+0.12357sind-0.004289 cosd+0.1538sin(2d)+.06078cos(2d) . (B7)
For time periods sufficiently short that all quantities can be considered constant,
the solar irradiance per unit time incident on a unit horizontal area can be found from
[79]
50 Ricklin et al.
1= 1353 W /m 2 cos (90 .57 _ a)a se C(90 .5 7 - a) , (B8)

where a is the transmission coefficient of the atmosphere :
a = 0.57 + 0.0045(90.57 - a) . (B9)

The value of the solar irradiance R is obtained from the incident solar irradiance
I weighted to account for cloud cover [80]:
R=bI, (BIO)
where
% cloud cover : 0 1/8 2/8 3/8 4/8 5/8 6/8 7/8 1

(B II)
b: 1.07 0.89 0.81 0.76 0.72 0.67 0.59 0.45 0.23 '
The sensible heat flux in W/m 2 can then be estimated using [81]
H = 0.4 (bI - 100) . (BI2)
Appendix C: Calculation of Aperture-Averaged Scintillation Index Using

Mathcad
This program calculates the aperture-averaging factor A as a function of the receiver

aperture diameter D, the scintillation index as a function of the radial distance from
the beam center p, and the aperture-averaged scintillation index for fixed D and p.
This approach includes effects due to source coherence. [see 1. C. Ricklin and F.
M. Davidson, "Atmospheric optical communication with a Gaussian Schell beam," J.
Opt. Soc. Am. A, 20(5), 856-866(2003)].
L:= 1000 path length (m)

A:= 785· 10- 6 wavelength (m)
wo:= 0.025 transmitter beam size (radius) (m)
k:= 2~1r wavenumber
(s:= I source coherence: I for coherent beam, > I for partially coherent beam
Cn2:= 1.0.10- 14
Rytov variance:
0'1:= 1.23·Cn2 ·ki ·L ¥
0'1 = 0.44
focusing parameter:
r.hat.> I r.hat < I focused beam
r.hat = I collimated beam, plane wave
r.hat > I divergent beam (spherical wave is limiting case at infinity)
diffractive parameter:
zJJat:= k 2woL2 o
0
D:= 0.00I, 0.0015..0.5 diameter (m) of receiving aperture

2 -3
po:= (0.55oCn2 ok oL)T coherence length of a propagatingspherical wave
A.._ r_hat _ z_hat ow o 2
,/-,. - z.ha t po2
(:= (s + 2~:~2 global coherence parameter
beam size (radius) at the receiver:

w:= wo-rr.har' + (-zJJat2)O o5
Aperture-averaging Factor A:
10 r -- -- -,-- -- - . , - -- - --,
Scintillation Index as a Function of p
We will need to calculate the beam parameters without considering optical turbulence
effects because these are included by the Rytov variance. Only partial coherence ef-
fects are included. We need to use the receiver parameters rather than the transmitter
parameters.
</Js := r_hat no turbulence effects
r.h at 2 2
Lo(r_hat
RS := z_hato¢s- r_hat +<soz.hat ) be C • L( . )
2 - <soZ-llat2 am recusing parameter at receiver
ws := wo · (z-hat + (s . z_hat )Oo5
2 2
beam size at the receiver
ws = 0.027
A 0 - 2 0L
beam diffractive parameter (receiver)

.- k w 2o
e := 1 + ~s beam focusing parameter(receiver)

Calculate Log-Intensity Variance as a function of radial distance from the beam
center (p=O)
[from W. B. Miller, 1. C. Ricklin, and L. C. Andrews, "Log-amplitude variance and

wave structure function : a new perspective for Gaussian beams," JOSA A 10(4),661-
672 (1993); modified as described above in Ricklin et all.
52 Ricklin et al.
p := 0,0.001..0.025
<Pl (p) := 4.42 . 0-1 . 11 5 / 6 . - ; ; + 3.86 · 0-1 ·0.4 . (1 + 2 . 8 )2
+ 4 . 11 2 5 / 12 . cos!!. . atan 1+2 ·6 _ 11 .115 / 6
6 2 ·11 16
0.8
Q) 1(p;o.6
0.4
0.2 ' - 0 - - - ' - - - - ' - - - - ' - - - - ' - - - - '

0.005 0.01 0.015 0.02
Calculation ofAperture-Averaged Log-Intensity Variance f or Fixed Receiver

Aperture size
D := 0.01 receiver aperture diameter in m

1 2 2 2 2 2
A := 16 f exp - D· x .2+ 2 po 2 _ qr·po - . acos(x) - x . (1 _ X 2)O.5 . xdx
11" Jo ----,;;:;r- wo .z_hat ~
p := 0.25 . ws radial location on beam (how far from the beam' s center)
<PI := 4.42 . 0-1 . - ; ; + 3.86 . 0-1 · 0.4 . (1 + 2 . 8 )2
. 11 5 / 6
+4.11 2 5 / 12 • cos!!. . atan 1+2 ·6 _ 11 .115 / 6
6 2 ·11 16
A = 0.855
<PI = 0.302
Aperture-averaged scintillation index:
A . <PI = 0.258
References
I. R.M. Goody and Y.L. Yung, Atmospheric Radiati on: Theoretical Basis, 2nd ed. (Oxford
University Press, 1989).
2. M.E. Thomas and D.O. Duncan, Atmospheric Transmission, The Infrared and Electro-
Optical System s Handb ook, Vol. 2 (SPIE Press, 1993).
3. A.S. Jursa, Ed., Handbook ofGeophysics and the Space Environment (Air Force Geophysics
Laboratory, 1985).
4. G.P. Anderson, EX . Kneizys, J.H. Chetwind, J. Wand, M.L. Hoke, L.S. Rothman , L.M.
Kimball, R.A. McClatchey, E.P. Shettle, SA Clough, W.O. Gallery, L.w. Abreu, J.E.A .
Selby, "FASCODFJMODTRANILOWTRAN: Past/Present/Future", 18th Annual Review
Conference on Atmospheric Transmission Models, 1995.
5. ONTAR Corporation (http://www.ontar.com!).
6. K.T. Whitby and B. Cantrell, "Atmospheric aeroso ls--characteristics and measurements,"
Proc. Intern. Conf. on Environ. Sensing and Assessment, Las Vegas, Nevada, 14-19 Septem-
ber (IEEE, New York, 1976).
7. H.C. van de Hulst, Light Scattering by Small Particles (John Wiley & Sons, 1957).
8. c.P. Bohren and D.R. Huffman, Absorption and Scattering ofLight by Small Particles (John
Wiley & Sons, 1983).
9. D. Deirmendjian, Electromagnetic Scattering on Spherical Polydispersions (Elsevier, 1969).
10. E.J. McCartney, Optics of the Atmosphere; Scattering by Molecules and Particles. (John
Wiley & Sons, 1976).
II. S.G. Gathman, "Optical properties of the marine aerosol as predicted by the Navy aerosol
model," Opt. Eng. 22, 57-62 (1983) .
12. A.M.J. van Eijk, "The ANAM-3.0 development," (TNO-The Netherlands Organisation for
Applied Scientific Research Report FEL-01-CI87, 2001).
13. A. Deepak and H.E. Gerber, "Report of the experts meeting on aerosols and their climatic
effects", WCP-55 , 1983.
14. M. Hess, P. Koepke, I. Schult, "Optical Properties of Aerosols and Clouds: The Software
Package OPAC", Bull. Amer. Met. Soc. 79, 831-844 (1998).
15. E.P. Shettle and R.W. Fenn, "Models for the aerosol s of the lower atmosphere and the effects
of humidity variations on their optical properties", AFGL- TR-79-0214, 1979.
16. G.A. d' Almeida, P. Koepke, and E.P. Shettle, Atmospheric Aerosols: Global Climatology
and Radiative Characteristics (A. Deepak Publishing, 1991)
17. A.N. Kolmogorov, "The local structure of turbulence in an incompressible viscous fluid for
very large Reynolds numbers," C. R. Acad. Sci. V.R.S.S. 30, 301-305 (1941) .
18. A.N. Kolmogorov, "Dissipation of energy in the locally isotropic turbulence," C. R. Acad.
Sci. V.R.S.S. 32,16-18 (1941) .
19. A.M. Obukhov, "On the energy distribution in the spectrum of a turbulent flow, C. R. Acad.
Sci. V.R.S.S. 32,19-21 (1941) .
20. A.M. Obukhov, "On the energy distribution in the spectrum of a turbulent flow," Izv., Akad.
Nauk SSSR, Ser. Geogra . Geofiz. 5,453-466 (1941) .
21. A.M. Yaglom, "Laws of small-scale turbulence in atmosphere and ocean (in commemoration
of the 40th anniversary of the theory of locally isotropic turbulence)," Izv., Atmos . Oceanic
Phy. 17,919-935 (1981) .
22. A.N. Kolmogorov, "A refinement of previous hypotheses concerning local structure of tur-
bulence in a viscous incompressible fluid at high Reynolds number," J. Fluid Mech. 13,
82-85 (1962) .
23. V.I. Tatarskii, The Effects of the Turbulent Atmosphere on Wave Propagation (National
Technical Information Service , Springfield, Va., 1971).
24. R.J. Hill and S.P. Clifford, "Modified spectrum of atmospheric temperature fluctuations and
its application to optical propagation," J. Opt. Soc. Am. 68, 892-899 (1978) .
25. L'C . Andrews, "An analytical model for the refractive index power spectrum and its appli-
cation to optical scintillations in the atmosphere," J. Mod. Opt. 39, 1849-1853 (1992).
26. G.D . Nastrom and P.O. Eaton, "The coupling of gravity waves and turbulence and turbulence
at White Sands, New Mexico from radar observations," J. Appl. Meteorol., 32 (I), 81-87
(1993) .
27. G.D. Nastrom and P.O. Eaton, "Variations of winds and turbulence seen by the 50-MHz
radar at White Sands Missile Range, New Mexico," J. Appl. Meteorol., 34 (10), 2135-2148
(1995).
54 Ricklin et al.
28. M.C. Roggemann and B.M. Welsh, Imaging Through Turbulence (CRC Press, Boca Raton,
FL, 1996), p. 320.
29. J.W. Hardy, Adaptive Optics for Astronomical Telescope s (Oxford University Press, New
York, NY, 1998), p. 448.
30. Y. Han Oh, J.e. Ricklin, E. Oh, S. Doss-Hammel and ED . Eaton, "Estimating optical tur-
bulence effects on free-space laser communication: modeling and measurements at ARL's
A-LOT facility," SPIE, 5550, 247-255 (2004) .
31. WB . Miller, J.e. Ricklin and W.J. Stewart, An Optical Turbulence Code for the Surface
Boundary Layer (Army Research Laboratory, ASL-TR-0220, 1987).
32. WB . Miller and J.C. Ricklin , A Module for Imaging Through Optical Turbulence (Army
Research Laboratory, ASL-TR-022 1-27, 1990).
33. H. Rachele and A. Tunick, "Energy balance model for imagery and electromagnetic propa-
gation," J Appl Meteorol, 33, 964-976 (1994) .
34. R.W Smith, J.e. Ricklin, K.E. Cranston and J.P. Cruncleton, "Comparison of a model
describing propagation through optical turbulence (PROTURB) with field data," SPIE, 2222,
780-789 (1994).
35. A. Ishimaru, Wave Propagation and Scattering in Random Media (Academic, New York,
1978), Vol. II, Appendix C.
36. J.A. Businger, "Turbulent Transfer in the Atmospheric Surface Layer," in Workshop on
Micrometeorology, D. Haugen, Ed. (American Meteorological Society, Boston, MA, 1973).
37. N. Busch, "On the Mechanics of Atmospheric Turbulence ," in Workshop on Micrometeo-
rology, D. Haugen, Ed. (American Meteorolog ical Society, Boston, MA, 1973).
38. H.A. Panofsky and J.A. Dutton, Atmospheric Turbulence: Models and Methods for Engi -
neering Applications (John Wiley & Sons, New York, 1984).
39. M. Liu, D. Durran, P. Mundkur, M. Yocke and J. Ames, The Chemistry, Dispersion and
Transport ofAir Pollutants Emitted From Fossil Fuel Plants in California : Data Analysis
and Emission Impact Model (National Technical Information Service, PB-264-822, 1976).
40. EV. Hansen , A Fractional Stability Category Scheme (Atmospheric Sciences Laboratory
Technical Report ASL-TR-0275 , 1990).
41. E.C. Kung, "Climatology of aerodynamic roughness parameter and energy dissipation in the
planetary boundary layer of the northern hemisphere," in Studies ofthe Effects ofVariation s
in Boundary Conditions on the Atmospheric Boundary Layer (Annual Report, Contract
DA-36-039-AMC-00878, Dept. of Meteorology, Univ, of Wisconsin, Madison, Wisconsin,
1963).
42 . Characterization ofWind Speed in the Lowest Layers ofthe Atmosphere Near the Ground :
Strong Winds (Engineering Sciences Data Unit Ltd, London , UK, EDSU 72026, 1972).
43. L.O. Myrup and J. Ranziere , A Consistent Scheme for Estimating Diffusivities to be Used
in Air Quality Models (NTIC, Springfield , VA, PB-272-284 , 1976).
44. A.S. Smedman-Hogstrom and U. Hogstrom , "Practical method of determining wind fre-
quency distributions for the lowest m from routine meteorological data," J. Appl. Meteorol.,
17 (7), 942-954 (1977).
45. e.D. MacArthur and P.A. Hains, The Roughness Length Associated With Regions ofHetero-
geneous Vegetation and Elevation (Army Research Laboratory, ASL-CR-82-0206-I, 1982).
46. S.E Clifford, "The classical theory of wave propagation in a turbulent medium," in Laser
Beam Propagation in the Atmosphere, J. Strohbehn , ed. (Springer, New York, 1978).
Atmosphericchannel effects on free-space laser communication 55
47. A.S. Monin and A.M. Yaglom, Statistical Fluid Mechanics (MIT Press, Cambridge, MA,
1971).V.L. Mironov, Laser beam propagation in turbulence atmosphere (Nauka, Moscow,
1981), p. 246.
48. V.L. Mironov, Laser beam propagation in turbulence atmosphere (Nauka, Moscow, 1981),
p.246.
49. A.S. Gurvich, A.I. Kon, V.L. Mironov, and S.S. Khmelevtsov, Laser radiation in turbulent
atmosphere (Nauka, Moscow, 1976), p. 227.
50. V.A. Banakh,ed., High-power laser beams in a randomly inhomogeneous atmosphere (SO
RAN Publ., Novosibirsk, 1998),p. 341.
51. M.E. Gracheva and A.S. Gurvich, "Simple models of turbulence," Izv. Akad. Nauk SSSR,
Fiz. Atm. Okeana 16(10), 1107-11I 1 (1980).
52. M.G. Miller and P.L. Zieske, Turbulence environmental characterization (Rome Air Devel-
opment Center, GriffissAir Force Base, NY, RADC-TR-79-131, 1979).
53. R.R. Parenti and R.J. Sasiela, "Laser-guide-star systems for astronomical applications," J.
Opt. Soc. Am. A, 11(I), 288-309 (1994).
54. R.E. Hufnagel, "Variations of atmospheric turbulence," in Digest of topical meeting on
optical propagation through turbulence (Optical Society of America, Washington, D.e.,
1974), p. WAI.
55. G.e. Valley, "Isoplanatic degradation of tilt correction and short-term imaging systems,"
Appl. Opt. 19(4),574-577 (1980).
56. B.G. Zollars, "Atmospheric-turbulence compensationexperimentsusingsynthetic beacons,"
in Selected Papers on Adaptive Optics for Atmospheric Compensation (SPIE Press, 1994),
Vol. MS92, pp. 564-588.
57. D.P. Greenwood, "Bandwidth specification for adaptive optics systems," J. Opt. Soc. Am.,
67(3),390-393 (1977).
58. T.E. Van Zandt, , J.L. Green, K.S. Gage, and W.L. Clark, Vertical profiles of refractivity
turbulence structure constant: Comparison of observations by the Sunset radar with a new
theoretical model, Radio Science, 13, 819-829 (1978).
59. T.E. Van Zandt, K.S. Gage, and J.M. Warnock, An improve model for the calculation of
profilesof wind, temperature, and humidity, PreprintsTwentiethConference on Radar Me-
teorology,American Meteorological Society, Boston MA, 129-135 (1981).
60. J.M. Warnock, T.E. Van Zandt, and J.L. Green, A statistical model to estimate mean values
of parametersof turbulence in the free atmosphere, Preprints7th symposium on turbulence
and diffusion, Nov 12-15, 1985, Boulder, CO, AMS, Boston, MA, 156-159.
61. E.M. Dewan, R.E. Good, B. Beland, and J. Brown, A model for Cn2 (opti cal turbulence)
profiles using radiosonde data, (PL-TR-93-2043, Environmental ResearchPapers, No. 1121 ,
Phillips Laboratory, HanscomAFB, MA, 1993).
62. W.B. Miller, J.e. Ricklin and LiC. Andrews, "Log-amplitude variance and wave structure
function: a new perspective for Gaussian beams; ' J. Opt. Soc. Am. A, 10 (4), 661-672
(1993).
63. LiC, Andrews, W.B. MillerandJ.e. Ricklin, "Geometrical representationof Gaussianbeams
propagating through complex optical systems," Appl. Opt., 32 (30), 5918-5929 (1993).
64. LiC, Andrews and R.L. Phillips, Laser Beam Propagation Through Random Media (SPIE
Press, Bellingham, 1998).
65. LC. Andrews, R.L. Phillips, and C.Y. Hopen, Laser Beam Scintillation With Applications
(SPIE Press, Bellingham, 2001).
56 Ricklin et al.
66. J.e. Ricklin and EM. Davidson, "Atmospheric turbulence effects on a partially coherent
Gaussian beam: implications for free-space laser communication," J. Opt. Soc. Am. A, 19
(9), 1794-1803 (2002).
67. J.e. Ricklin and EM . Davidson, "Atmospheric optical communication with a Gaussian
Schell beam," J. Opt. Soc. Am . A, 20 (5), 856-866 (2003).
68. VI . Tatarskii, Wave Propagation in a Turbulent Medium (McGraw-Hill Book Company,
Inc., New York, 1961).
69. E Dios, J.A. Rubio, A. Rodriguez and A. Comeron , "Scintillation and beam-wander analysis
in an optical ground station-satellite uplink," Appl. Opt ., 43 (12), 3866-3873 (2004).
70. R.L. Fante, "Electromagnetic beam propagation in turbulent media: an update," Proc . IEEE,
68 (11),1424-1443 (1980).
71. RJ . McIntyre, "The distribution of gains in uniformly multiplying avalanche photodiodes,'
IEEE Trans. Electron. Dev., 19 (6), 703-713 (1972).
72. J.H. Churnside and C.M. McIntyre, "Averaged threshold receiver for direct detection of
optical communications through the lognormal atmospheric channel," Appl. Opt ., 16 (10),
2669-2676 (1977).
73. R.M. Gagliardi and S. Karp, Optical Communications (Wiley-Interscience, New York,
1995).
74. A.R. Weeks, J. Xu, R.R. Phillips, L.C . Andrews, C.M. Stickley, G. Sellar, J.S. Stryjewsky and
J.E. Harvey, "Experimental verification for an eight-element multiple-aperture equal-gain
coherent laser receiver for laser communications," Appl. Opt., 37 (21), 4782-4788 (1998).
75. R.K. Tyson, "Adaptive optics and ground-to-space laser communications," Appl.d Opt ., 3S
(19),3640-3646 (1996).
76. B.M. Levine, E.A. Martinsen, A. Wirth, A. Jankevics, M. Toledo-Quinones, F. Landers and
T.L. Bruno , "Horizontal line-of-sight turbulence over near-ground paths and implications
for adaptive optics corrections in laser communications," Appl. Opt ., 37 (21), 4553-4560
(1998).
77. R.K. Tyson, "Bit-error rate for free-space adaptive optics laser communications," J. Opt.
Soc. Am . A, 19 (4), 753-758 (2002).
78. R.K. Tyson, "Indirect measurement of a laser communications bit-error-rate reduction with
low-order adaptive optics," Appl. Opt. , 42 (21),4239-4243 (2003).
79. RJ. List, Smithsonian Meteorological Tables (Smithsonian Institute, Washington, DC,
1951).
80. EB . Smith, "A Scheme for estimating the vertical dispersion of a plume from a source
near the ground," Proc. ofthe ThirdMeeting ofthe ExpertPanelon Air Pollution Modeling
(NATO Committee on the Challenges of Modern Society, Paris, France, 2-3 October, 1972).
81. EB . Smith, "The relation between Pasquill stability P and Kazanski-Monin stability (in
neutral and unstable conditions," Atmos . Environ ., 13 (6), 879-881 (1979).
© 2005SpringerScience+ Business MediaInc.
001: 10. I007/s10297-005-0054-0
Originally published in J. Opt. Fiber. Commun. Rep. 2, 345-396 (2005)
Free-space laser communication performance in the atmospheric

channel
Arun K. Majumdar
LCResearch, Inc.
30402 Rainbow View Drive
Agoura Hills, California 91301
Email :a.majumdar@IEEE.org
Abstract. In spite of the tremendous technical advancement of available components,

the major limitation of free-space laser communication (lasercom) performance is due
to the atmosphere, because a portion of the atmospheric path always includes turbu-
lence and multiple scattering effects . Starting from a fundamental understanding of
the laser communications system under diverse weather conditions, this chapter pro-
vides a comprehensive treatment of the evaluation of parameters needed for analyzing
system performance. The significance of higher-order statistics of probability density
functions of irradiance fluctuations due to turbulence to performance analysis is ex-
plained . Starting from link analysis, the necessary expressions relating link margin,
bit-error-rate, signal-to-noise-ratio, and probability of fade statistics are presented.
Results for laboratory-simulated atmospheric turbulence and multiple scattering are
presented. Example numerical results for simulations of lasercom systems operat-
ing under various atmospheric conditions are presented for various scenarios such as
uplink-downlink (e.g., between ground and satellite, aircraft or UAV) and horizontal
(terrestrial) link. Both turbulence and multiple scattering effects have been included in
the analysis with both on-off keying and pulse-position modulation schemes . Statisti-
cal estimation and computation of communication parameters presented in this chapter
will be useful in designing and optimizing lasercom systems that are reliable under all
weather conditions.
1. Introduction
There has been tremendous technical advancement of available components such

as laser/LED transmitter, high sensitivity optical receiver offering extremely high
bandwidth, efficient modulation techniques, improvement in low powerconsumption,
58 Arun K. Majumdar
weight, and size. Inspite of many such technological development , the major limi-
tation of free-space laser communication (Iasercom) performance is the atmosphere.
Atmospheric condition ultimately determines the laser communications systems per-
formance not only ofterrestrial applications but also for uplink-downlink (e.g., between
ground and satellite, aircraft or UAV terminals), because a portion of the atmospheric
path always includes turbulence and multiple scattering effects.
This paper provides a comprehensive theoretical treatment and analysis of the
performance of a laser communication (hereafter in short, "lasercom") system under
a given communication format scheme. Numerous examples are given for specific
system parameters, and the necessary tools are developed to design a system fulfilling
specified requirements. The performance of a lasercom system is generally quantified
by the "link margin", i.e., the ratio, expressed in dB, of the signal power received to
the signal power required to achieve a specified data rate with a specified acceptable
probability of error. The link margin calculation is therefore essential to design an
acceptable system. Atmospheric conditions affect system performance and so need
to be accounted for in the calculation. A link budget model has been developed that
includes dependence on the atmospheric channel that the signal passes through as well
as on the transmitter and receiver stations. This model aids designers in optimizing
the optical station main parameters to be able to establish a data link with adequate
performance.
The main advantage in communicating, i.e., transferring information from one
point to another, with optical frequencies is the potential increase in information trans-
fer rate. This rate is directly related to the bandwidth of the modulated carrier, which is
generally limited to a fixed fraction of the carrier frequency itself. Therefore, increasing
carrier frequency from that of, say, RF or microwaves to that of optical waves increases
the information capacity of a communication system by many orders of magnitude .
A usable bandwidth at an optical frequency of rv200 THz will be about 105 times
that of a carrier in the RF range. A second advantage of using optical frequencies is
the station compactness it allows. A 6-inch-diameter lens "antenna" for use with an
optical frequency of 6 x 1014 Hz has an effective gain of 122 dB. At an RF frequency
of I GHz to generate about 60 dB gain, it would require an antenna with a diameter of
over 200 feet. This is due to the fact that the beam divergence is roughly proportional
to V D, where A is the carrier wavelength and D is the aperture diameter, and varies
inversely with antenna gain. Thus, a longer wavelength necessitates a proportionately
larger antenna to achieve the same gain.
Because of these advantages, optical carrier frequencies can accommodate very
high data rates of 100 GHz or more. With the advent of the new technology of wave-
length division multiplexing (WDM), we can further increase the data rate per optical
beam. As optical technology matures, the advantages of practical free-space laser com-
munications systems can be realized with much architecture.
We here divide lasercom systems into several distinct application classes : (a) paths
between terrestrial locations, e.g. between buildings in metropolitan areas, mountain -
to-mountain, horizontal link between two ground stations, (b) paths between ground
and near space, e.g., ground-to-satellite (Uplink) or satellite-to-ground (Downlink),
(c) paths between airborne aircraft , (d) Space-shuttle-to-ground, (e) UAV-to-ground,
(f) indoor communications with paths <rv 100m, usually using infrared frequencies,
and (g) free-space communications in optical interconnects on optical chips, with path
lengths ranging from hundreds of microns up to rv I ern. Besides the differences in
Free-space laser communication performance in the atmosphericchannel 59
path lengths, these application classes also differ in data-rate regimes (multi-Gbit/s to
a few MbitJs or less), line-of-sight requirements for the links, and the impact, if any,
of weather and other atmospheric effects.
The design of a complex engineering system such as a lasercom system requires a
thorough understanding of the technology building blocks, taking into account of the
state-of-the-artdevelopmentof the subsystems and devices,as wellas an understanding
of the complex phenomena of the physics of optical wave propagation through the
atmosphere. The ability to predict system performance is essential for the design of a
practical,optimal, cost-effectivesystem.The lasercom system mustbe able to establish
a communication link between transmittingand receivingstations with a specifieddata
rate and a probability of error lower than a specified bit-error rate (BER).
Section 2 of this paper describes the procedure for computing link budget and ex-
plains theequations for data rate, link margin,and BER for an atmosphericchannel with
absorption and scattering. Numerical results are presented to show how the relevant
communication parameters depend on range and required power. Section 3 describes
the theoretical background of scintillation effects on signal-to-noise-ratio (SNR), and
the procedures for computing BER for on-off keying (OOK) and pulse-position mod-
ulation (PPM) schemes. This section also discusses the probabilitydistribution of fade
statistics for lasercom system. In section 4, example numerical results for different
communication scenarios such as uplink, downlink, and horizontal (terrestrial) link
in the presence of atmospheric turbulence and multiple scattering medium are pro-
vided. Section 5 discusses other types of examples of free-space optical and laser
communications which include indoor optical communications and free-space optical
interconnects. Multiple transmitters/receivers approach for lasercom is the subject of
section 6, and section 7 provides some concluding remarks.
2. Basics of Laser Communication Link Analysis
2. J. Communication Channel Characterization
The primary factors characterizing an atmospheric communication channel include

atmospheric attenuation (from both scattering and absorption) and scintillation. Atten-
uation of the optical wave (i.e., reduction of its power) by atmosphere can be caused
in several ways, including absorption of light by gas molecules and Rayleigh or Mie
scattering by gas molecules or aerosol particles suspended in the air. Optical trans-
mission spectrum of the U.S. Standard Atmosphere over the various communication
windows is available in the literature. [I] The individual absorption lines are caused
by the vibrational-rotational absorption lines of water vapor, C02, CH4 , and other
gases of minor constituents in the atmosphere. Scattering from aerosols and particu-
lates such as in fog, clouds, smoke, and dust also contribute to the total attenuation
of the optical beam. Some typical values of near-ground (terrestrial path) attenuations
with corresponding visibilities are [2,3]:
Very clear: 0.19 dB/km (50 km visibility), fog: 6.9 dB/km (1.9 km),
moderate fog: 28.9 dB/km (500 m visibility).
60 Arun K. Majumdar
Scintillation is the fluctuations in the detector signal as a result of random varia-

tions in the refractive index of the turbulent atmosphere along the channel. The amount
of scintilIation can change by more than an order of magnitude during the course of a
day, with the most severe effects around midday. The noise produced by scintillation in
the received power results in increased BER and reduced performance. Scintillation-
induced fades can ultimately cause loss of the communication signal altogether. Re-
search for the past 40 years has alIowed us to estimate the effects of turbulence on
optical beam irradiance fluctuations by relating them to the refractive -turbulence struc-
ture parameter, C~ . The turbulence strength of the atmosphere is usualIy expressed in
terms of a scintillation index, explained in a later section. We will cover communi -
cation performance for three different types of scenarios as examples later on in this
paper: horizontal (terrestrial) link, downlink (satelIite-to-ground), and uplink (ground -
to-satellite).
2.2. Transmitterand Receiver System
There are several atmospheric transmission windows within the 700-1 O,OOO-nm wave-
length range that have an attenuation of <0.2 dB/km and so are nearly transparent.
The majority of free-space lasercom systems are designed to operate in the windows of
780-850 and 1520-1600 nm. Around 850 nm, reliable, inexpensive, high-performance
transmitter and detector components are readily available. These are often used in net-
work and transmission equipment. A highly sensitive silicon (Si) avalanche photodiode
(APD) and an advanced vertical-cavity surface-emitting laser (VeSEL) are available
for operation near 850 nm. For the wavelength range of 1520-1600 nrn, high quality
transmitter and detector components are also readily available. InGaAs is the most
commonly used detector material for this longer wavelength; it exhibits high band-
width capability combined with a high spectral response. This material is extremely
sensitive because of an internal amplification (avalanche) process. These wavelengths
are also suitable with erbium-doped fiber amplifier (EDFA) technology for high-power
generation (>500 mW) and high-data rate (>2.5 Gbit/s) . This wavelength region is
also important from a laser eye safety point of view. The American National Standard
for Eye Safe Use of Lasers Institute (ANSI) standards for the maximum permissible
exposure level (MPE) for a lO-sec exposure is about I mW/cm 2 for an 800-nm wave-
length and about 100 mW/cm 2 for a 1550-nm wavelength . The longer wavelength
system is thus readily designed to be eye safe. Other benefits of the 1550-nm wave-
length region include a reduced solar background and reduced scattering attenuation ,
because of the Rayleigh dependence of inverse wavelength to the fourth power.
2.3. Link Analysis
The overalI system performance of a lasercom is quantified using a link budget derived
from the range equation, which combines attenuation and geometrical aspects to cal-
culate the received power. The process of finding the link margin through the system
link calculation is the topic of this section. The system link calculation is carried out as
shown in the Fig.l. We will show some examples oflink calculations for different com-
munication scenarios of practical importance for various communication parameters
Received Signal = ( PT ) (OT ) (t T) (t . ..) (A/4ltL)' (lt O/).)' (t. )
~ -----~.-.1 "" ..I "- - - - ~
TRANSMITIER PROPAGATION MEDIUM RECEIVER
Fig. 1. Optical signal link calculations.
such as transmitter power, propagation losses, receiver sensitivity. The receiver's sen-
sitivity determines the amount of received optical power needed to achieve the required
signal-to-noise ratio (SNR) for a given expected communication performance.
The purpose of this section is to develop the parameters necessary to calculate
the performance of an optical communication link . We shall consider the situation of
optical propagation between points in free-space. Consider a laser transmitter antenna
with gain GT transmitting a total power PT at the wavelength. The signal power
received at the communications detector can be expressed (from the range equation)
as
Received Signal, PREC = PTGTTTTATMSGRTR, (I)
where TT is the transmitter optical efficiency, TATM is the value of the atmospheric
transmission at the laser transmitter wavelength, S is the free-space loss, G R is the
receiver antenna gain, and TR is the receiver optical efficiency. The transmitter gain,
free-space loss, and receiver antenna gain are given by G T = 16/0} (where OT is
the full transmitting divergence angle), S = (>.. /47fL) 2 (where L is the range), and
G R = (7f D / >..)2 (where D is the receiver diameter). TATM may be written in terms of
the atmospheric attenuation factor a given by -10 10g(TATM) / L .We can then express
the received signal as
PREc(Received Signal) = PTGTTTTATM(>../47fL)2(7fD />..)2 TR. (2)
Equation (2) can be re-written as

2/O}L2)TTTATMTR.
PREC = PT(D (3)
If we write TATM in terms of atmospheric attenuation factor, a (dBlkm) at the

wavelength, >.., the received power is then given by
FREC = PT(D 2/O}L2)TTlO<- oL/ IO )TR . (4)
Normally an optical link typically consists of two tran sceivers, each made up
of one (or more) transmitting laser( s) and receiving photo detector(s). Transmitting
62 Arun K. Majumdar
optics (telescope, lenses, mirrors) shape the transmitted laser beam which is collected
by the receiver optics so that the received signal is focused onto the photo detector.
The parameters of the communications are chosen so that sufficient signal from the
lasers on one transceiver reaches the photo detector on the other transceiver through
the atmosphere to differentiate ones (signal) and zeros (no signal) with negligible error
(i.e., very small specified BER).
2.3.1. Data Rate
Given a laser transmitter power PI. with transmitter divergence of ()t, receivertelescope
area A, transmit and receive optical efficiency Topt, the achievable data rate R can be
obtained from
R - PtToptTATMA 5
- 1r(()t/ 2 )2U E p N b ' ( )
where E p = he/ A is the photon energy and N b is the receiver sensitivity in # pho-
tonslbit.
2.3.2. Link Margin

Another important parameter in optical communications link analysis is "Link Mar-
gin", which is the ratio of available received power to the receiver power required to
achieve a specified BER at a given data rate. Note that the "required" power at the
receiver PREQ (watts) to achieve a given data rate, R (bits/sec) , and receiver sensi-
tivity, Nb (# photonslbit), is related by PREQ = NbRhv = NbRhe/ A, where u is
the frequency of the laser light of wavelength (h = Planck 's constant, e = velocity of
light). The parameter Ns is usually defined as receiver sensitivity and can be expressed
as # photonslbit, or in dBm (taking the logarithm to the base IO and multiplied by IO
with base as I mW). Finally, we can define the link margin M as
M = Received Power/Required Power.
Combining the above equations yields the expression
(6)
2.3.3. Bit Error Rate in Presence of Atmospheric Absorption and Scattering
A laser beam propagating through the atmosphere is attenuated by absorption and

scattering due to the presence of aerosols, dust, smoke, fog, clouds, rain, snow, and
atmospheric molecules. In this section, we address only the optical propagation caus-
ing absorption and scattering. The detailed analysis of lasercom performance due to
turbulence-induced atmospheric scintillation will be discussed in later section. We first
calculate the received signal power for a lasercom system for a given range and ex-
tinction (combined absorption and scattering) . The detector noise that comes from the
background power and the inherent detector noise are then evaluated . For the trans-
mission of both a binary I and a binary 0, we can then calculate the distribution of
signal plus additive Gaussian noise. Finally, the error probability associated with binary
detection (bit error rate) can be determined by the procedure described below.
The photocurrent induced by the received optical signal is given by
. GPRECAq
Z= he ' (7)
where PREC is the received laser power given by Eq.(4), q = 1.601 X 10- 19 C is
the electronic charge, and G is the avalanche photo detector (APD) gain . For a PIN
Photodiode, we take G = 1. The background power and the detector noise power can
be written as [4]
(8)
where Hbk g is the background radiance, BW i f is the receiver filter's optical band-
width , A r ec is the receiver area , <Pdet is the detector's solid angle field-of-view (FaY),
and TJre cy is receiver optical efficiency. The detector noise power can be expressed as
Pdet = JBwifNEPdet , (9)
where the values for the bandwidth and the noise-equivalent power (NEP) of the
detector are typically supplied by the manufacturer. The mean-square fluctuations in
the signal current isig can be expressed as [5,6]
(isig ~oise ) = 2qM F i sigBWdet , (10)
where F is the excess-noise factor for the APD (F = 1 for a PIN photodiode). We can
similarly write the expressions for the mean-square fluctuations in the currents due
to background and the detector noise. Assuming the mean-square fluctuations in the
photo detector current due to signal, background, and detector are all independent and
uncorrelated, we can calculate the total rms noise for a binary I or 0 in the following
way:
For transmission of binary 1 : 0" 1 = (i sig ~oise ) + (i bkg ~oise ) + (i det ~oise )'
(I I)
For transmission of binary 0 :ao = (i;i g noise(ry)} + (i bkg ~oise ) + (i det ~oise ) '
(12)
where isig noise( TJ) reflects the extinction ratio of the laser signal and thus takes into
account of the incomplete modulation of the laser from an ON state to an OFF state
with a typical value [4] of modulation depth of", 90 %.
The probability of detecting either a I or a 0 in error can be expressed as
0 .51 exp[-(i - 1)2/2ai] .

error! = ~ di , (13)
/ - 00 V 27rai
exp[- (i)2/2aa]d '

error O=
1 00
0 .51 J27raa
z, (14)
The overall BER is the mean of these two probabilities:
BER = (errorl + err orO)/2. (15)

64 Arun K. Majumdar
' ,10"
~:---._
-,
- a = O.2 dBlkm
. ,-./ .::-.::::.=:=---- ------
<. -
<, ----------== ~-
_...t?:..~_...~~ EJi'\( Ill .
~--
<,
<; a =21 d Blkm
1-10"
<:»:<," ,
-
I 'IO~
"" "-,,
4
1_10
~ ~ 0 I~M . 1M I~~ l~ I~~ ~ M~ ~l ~ ~~
L i nk ran~e(m)
Fig. 2. Communication data rate as a function of range for three atmospheric conditions: clear
(0 = 0.2 dBIkm), haze (0 = 4 dB/km) and fog (0 = 21 dB/km).
2.3.4. Example Numerical Results
Data Rate Versus Range The range equation can be used to generate the communi-
cations data rate versus range for varying atmospheric conditions. Suppose a 2.5-km
link is desired at 1.25 Gbit/s. Furthermore, the system has to be eye safe, so that we
use a 1.55-J.lm laser with 70 mW power. We assume a 13-cm receiver aperture and
transmitting beam divergence of 8.5 mrad FWHM . Receiver sensitivity for 1.25 Gbit/s
was taken to be -36 dBm, which is equivalent to 1568 photons/bit at this wavelength
and data rate. Based on these assumptions, link margins were calculated for signal at-
tenuation due to weather events [3] suc h as clear (visibility = 23 km, 0 = 0.2 dB/km),
haze (visibility = 2 km, 0: = 4 dB/km), and fog (visibility = 0.5 krn, 0: = 21 dB/km) .
The transmitter and receiver optics efficiencies are both assumed to be equal to 0.5.
Data rates versus range is shown in Fig. 2. As can be seen, a data rate of 3 Gbitls can be
achieved for a range of 2.4 km under clear condition , for 1.4 km under haze condition,
and for 0.6 km under fog condition . The system is desired to have communication
capabili ty for the above ranges and atmospheric conditions . Under night-time condi-
tions, the ranges would be increased further. Increased data rate is possible if a higher
power laser, or a larger aperture, or both, is used.
Available Optical Link Margin (dB) Versus Link Range Figure 3 illustrates avail-
able optical link margin (dB) versus link range for the same system parameters such as
laser power, transmitter beam divergence, aperture diameter, and operating wavelength
of 1.55 J.lm and the above visibility conditions. As seen in this figure, 10 dB of link
margin is available for clear condition for the system operating over a 1.2-km span,
for haze over a 0.885 km span, and for fog just under a 0.5-km span for the above
conditions of 1.25 Gbit/s at 1.55 J.lm. The curve shows how much margin a given
system has at a given range to compensate for scattering, absorption, and scintillation
losses. Under system operation, parameters defining channel capacity can be varied
to accommodate link availability under varying atmospheric condition s. The link mar-
gin curves fully encompass almost all basic aspects of laser communication system
performance.
I~
"§
~ \\\Q/ =0.2 dBlk m
0. '" 4dBlkm
8- 20
n = 2ldBlkm
500 1000 1500 2000 2500

Link Range (m )
Fig. 3. Link margin versus link range for three atmospheric conditions: clear, haze and fog.
Link Budget Computation To give an idea of the various components of link budget,
Table 1 shows an overall calculation of the link performance of a downlink (satellite-to-
ground) laser communication system The link analysis is performed for a satellite-to-
ground range of 4.83 x 105 m, transmitter laser operating at wavelength A = 0.635 m
at the data rate of 3 Gbit/s using a detector (receiver) sensitivity Ni, = 100 photons/bit
and a receiver of diameter of 1.4 m. The table shows the laser power required for a
specified overall link margin of 6.0 dB for this link.
Table 1. Link analysis example of a satellite-to-ground laser communication system.
Parameter Value/Factor dB
Wavelength (A ) 0.635 p.m

Range (L) 4.83 x 10° m
Data Rate 3 Gbps
Receiver Diameter (D) 1.4 m
Transmitter Divergence Angle (OT) 2.07 X 10- 4 rad
TransmiuerAntennaGain rGj- = 16/(OT)2) 3.73 X 108 +85.72
Transmitter Optical Loss 0.1 -10.0
Space Loss (8 = (A/47rL)2) 1.09 X 10- 26 -259.61
ReceiverAntenna Gain ICjj = (7rD/A)2) 47.974 X 1012 +136.81
Receiver Optical Loss 0.1 -10.0
SYSTEM LOSS -57.08
Atmospheric Turbulence Margin -11 .30
Clear Air Transmission Loss - 2.08
TOTAL LINK LOSS - 70.46
LINK MARGIN -6.00
DESIGN LOSS - 76.46
Required Received Signal at 3 Gbps 9.36 x 10 - 8 w -70.29 (= 10 tosio 9.36 x 10 - 8)
Required Laser Power at 3 Gbps =Required 4.14 W (= 106.17/1°) -70.29 + 76.46 = 6.17
received signal - Design Loss
66 Arun K. Majumdar
10
1 ~------
\ '\, "'--...,
I
\ \
\ \
"'
~ u =l3dBlkm
.... _-_. -
\ \ "
\ \
: /l dBlkrn -,
\ \ -,
-,
\ \ "--,
\ --\- ~ ~ 9 dBik;;; -,
\ \
0.05 0.1 0.15 0.2 025 0.3 0.35 0.4
REQUIRED roW ER ( W)
Fig. 4. Biterror rate(BER)as a function of requiredlaser powerfor threeatmospheric conditions

: thin fog ( = 9 dB/km), light fog ( = 11 dB/km ) and moderate fog ( = 13dB/km).
Bit Error Rate Versus Required Power for Absorption And Scattering Numerical
results for laser communication system operating under various atmospheric conditions
are depicted in Fig. 4. The BER of a high-bandwidth (570 Mbits/s) lasercom system
as a function of required laser power (to achieve that BER) for a one-way path of 2
km is shown in Fig. 4. Some of the parameters were similar to the ones used for a
retro-reflected experiment reported by other researchers [4]: laser wavelength A =
810 nm, transmitter efficiency rr = 0.5, receiver optical efficiency TR = 0.5, receiver
diameter D = 13 em, beam divergence (full) Or = 1 mrad, detector field-of-view
(FaY) 4> det = 200 rnrad, background radiation Hbkg = 0.2 W m - 2 nm -1 sr- 1 , filter
optical bandwidth BW if = 20 nm, detector bandwidth BWdet = 422 MHz, detector
NEP: NEPd e t = 0.19 pW / Hz 1 / 2 , APD gain G = 135, effective ionization coefficient
K eff = 0.005. In this work presented in this paper, we varied the required power from
about 4 to 350 mW and calculated the corresponding BER for three different foggy
atmospheric conditions : the atmospheric attenuation factor, a = 9 dB/km (thin fog,
about 1.5 km visibility), a = II dB/km(light fog, about 1.3 km visibility), and a = 13
dB/km (lightly moderate fog, I km visibility). It is seen from the Fig. 4 that to achieve
a BER of 10- 9 we need the laser powers of about 60, 140, and 350 mW, respectively,
for operating under the above various foggy conditions specified by 9, II, and 13
dB/km attenuation, respectively. There are also other parameters one can vary such
as transmitter beam divergence angle, receiver diameter, etc. to change the required
power for achieving a given BER. The designer can then trade-off these parameters
in designing the lasercom system to operate optimally under various atmospheric
scattering conditions.
Low-Visibility Atmospheric Optical Communication Link Using Pulse Position

Modulation (PPM) In a low visibility atmosphere, multiple scattering due to partic-
ulates such as haze, rain, snow, clouds, fog, dust, and smoke can seriously degrade the
performance of the atmospheric commun ication system by limiting the percentage of
time the communication link would be available for data transmission . By adapting
an optical receiver to increase the field-of-view (FaY), the performance degrada-
tion can be somewhat compensated because multiple scattering increases the beam
divergence. In the example below we have analyzed the communication link using
multiple-forward-scattered (MFS) radiation. The analysis is based on simulating vari-
ous physical factor s that are encountered in the actual atmospheric channel. Parameters
that affect signal-to-noise fluctuation at the receiver end, such as multiple scattering
of the laser beam off particles in clouds and the atmosphere, sky background, and
the effects of direct illumination by the solar flux, are evaluated. Typically, a turbid
atmosphere exhibits strong forward-scattering peaks with significant wide-angle scat-
ter and backscattered components. Therefore, we can exploit scattered radiation in
addition to the un-scattered component to improve communication through a low-
visibility atmosphere (by additionally collecting some portion of the scattered signal
with increased field-of-view) . The atmosphere is characterized by the optical depth
T = f3 L > 1, where f3 is the atmospheric attenuation coefficient and L is the propaga-
tion path length . The available signal power received through the entire atmospheric
link is given by utiliz ing the MFS propagation model [7]:
(16)
where P; = cw laser power at wavelength A, p R = receiver aperture radius, 8r =

transmitter initial beam divergence (half-cone angle), «
= effective attenuation coef-
ficient = f3 - f3~, with f3 = f3a + f3s (f3a being the absorption coefficient and f3s being
the scattering coefficient of the medium), {3~ = {3s¢, ¢ being the forward-scattering
efficiency, 80 = 21 / 2 /kpo (k is the wave number = 21r/ A and the atmospheric chan-
nel coherence length po = (3/f3~Lk28})), 8F = rms forward scatter angle, 8R =
field-of-view half cone angle, i.e., FOV = 28 R .
The unscattered component of the received signal is given by
(17)
Note that for the multiple-scattering regime, [(PA)SC/(PA)UNS > Ij.where

(PA)sC is the scattered component of radiation, given by (PA) from the Eq. (16).
Photo detector noise plus fluctuations in the channel and background noise cause er-
rors to some of the demodulated information. The background noise power can be
written as
(18)
where T a and r; are atmospheric and receiver transmissivities, .1A is the input optical
filter bandwidth (which is the same as BWif used in earlier equation), and N(A) is
the spectral radiance of the background at wavelength A. A typical value of N(A)in
the daytime is 10 W/m 2 sr /-Lm. The ratio of scattered-to-unscattered power with the
same transmitter and the same propagation distance can be expressed as
(PA) sc atter ed exp(¢{3s L )

~---;-'------=.,--------=~"";:-------~ (19)
(PA)unsca ttered
[1+ ~ 3
/3s </>
11 2
T
LI}] [1 + ~ /3s</> LII}] .
3 11 2
R
Figure 5 shows the result of SNR in dB as a function of FOV for a monodisperse

and polydisperse (containing a number of different monodisperse particles) medium
from an experiment described in detail elsewhere. [8, I0] A scattering cell containing
68 Arun K. Majumdar
.
, -221
o (!n OlSPfllSl
1"
IOW AfH ll, O.3 TO l. 11,.., 1
WOtoCllllSt1;RSE
lO",toIiU Il -J·U II"'"
"
· ,: -----'---'---'----"-:--
0,25 0,5
........_ ----
fIELO OHOl'lll FOVl.2'RIlEGRHI
,
0.15
---- '--
t.,
Fig. 5. Signal-to-noise-ratio as a function of receiver FOV for a monodi sperse and polydisperse
medium (T = optical depth) (reproduced with permission from Optical Society of America) .
Fig.6. Pulse position modulation (PPM) format (M is the total number of time slots and T is
the slot width) (reproduced with permission from Optical Society of America) .
both polydisperse and monodisperse particles of polystyrene latex spheres (of diam-
eters ranging from 0.3 to 5.17 p,m) in water suspension simulated the low-visibility
(cloudy) atmosphere a few kilometers thick in the laboratory using a l2.7-cm-long
scattering cell. A narrowband interference filter with .1,\ ~ 15 A was used. A diode
laser transmitter (,\ = 0.8486 p,m) in the experiment had a cw power output of 7 mW
and a full-angle divergence of 15 0. The optical thickness T characterizing the medium
was determined by extrapolating to the ratio of 10 (intensity without medium) and 1
(intensity including medium) a zero FOY using the relation
T = [In(Io/I)]FQv->o- '
In Fig. 5 notice that by increasing the FOY beyond 0.25 degrees, we will not gain much
in SNR of an atmospheric optical communication system for a polydisperse medium,
and actually lose SNR for a monodisperse medium.
Free-space lasercommunication performance in the atmospheric channel 69
Signal requirements for MFS-modeled communication channel When an airborne

laser transmitter terminal (previously called "station" ) passes between the sun and
the receiver terminal (which may be another airborne system) , the signal becomes
masked to a degree by the spectral component of the sun's background near the laser
wavelength . By varying the receiver FOV, it is possible to control both the amount
of information signal impinging on the detector and background noise. For a given
data rate and background condition, there exists an optimal FOV to maximize the
system margin. Pulse position modulation (PPM) is a highly efficient modulation
technique for communication through clouds, haze, dust, smoke, rain, snow, etc. in
which multiple scattering is involved with or without direct solar background. The
PPM format employs a synch pulse to determine a major time division often called
a window, which is then divided into time slots defined by the particular data value
depending on which slot the pulse falls in. Figure 6 illustrates the PPM format. The
SNR (assuming no pulse stretch effect in time domain) is derived from [9]
F BE = FpA + (1 - FD), (20)
1 2
Fp A = f(C exp( -~ / 2), (21)
y21l'~
1 2
FD = f(C exp( -( / 2), (22)
y21l'(
where
qr (_ qs - qr
~= J(q'iv) , - J(q'jy ) '
In the above equation, FBE = probability of bit error, F pA = probability of false alarm,
FD= probability of pulse detection, qs= signal charge, qr= charge threshold for pulse
detection, (q'jy)= rms noise charge. The voltage signal-to-noise ratio is then given by
(NS) v =~+(= J as(q'jy }' (23)
Power SNR is obtained from SNR = [( tT)

v] 2.
Let us denote k = number of bits per laser pulse, and the number of time slots
is M = 2 k • In this analysis , M = 32, and k = 5 were assumed . Assuming that the
probability of error FB E is 10- 6 , and since thirty-one slots can have a false alarm rate,
then
31FpA + (1 - FD) = 10- 6 .

If we assume equal probability so that
(1- FD) = !1O-

2
6
=5 X 10- 7 .
Solving for FD and FpA and from Eqs. (??) and (22), the parameters ~ and ( can
now be determined as, in this example, ~=5 .5341 and (= 4.8991, so that ~ + (= I0.4332
yielding power SNR= 108.852 (= 20.37 dB). Following the method as outlined in an
70 Arun K. Majumdar
earlier paper, [10] the optical power P; required at the receiver for a specified bandwidth
and SNR is given by
r; = (h~~~m) [(SNR) + (SNR)2 + 4(NB + Nv) (~) (SNR) ,] (24)
where m is the bit rate.

The system margin is defined as the ratio of the available power received power to
the signal power required to achieve a given bit rate m. Note that for the PPM coding,
the bandwidth B is given by B = Ij(2x slot width) . The slot width 7 is then given
by 7 = kj(Mm) , so that the bandwidth can be written as
B=~=Mm . (25)
27 2k
The system margin 8M can be now expressed in dB as a function of total FOV (=
20R ) :
(26)
The system margin was determined for a concentration of the latex spheres suspended
in water for optical depth 7 =4.23, and two values of the spectral radiance generated
by a quartz-halogen lamp were used: N(A) = 700 W/m 2 sr Jlm and N(A) = 7000
W/m 2 sr Jlm at the wavelength A =0.8486 Jlm. Figure 7 is a plot of the system margin
vs. FOV for 7 = 4.23 and background radiances of N(A) = 700 W/m sr Jlm and
2
2
N( A) = 7000 W/m sr Jlm. For a given data rate, the system margin increases initially
with FOV,but levels off. so that at some point further increasing FOV does not increase
the system margin further. Notice that the system margin is zero under some conditions,
resulting in excess error rate (for example. in the ranges of FOV of 0.043 ° to 0.18 ° for
the data rate of 327.86 Mbps and for the range of 0.043 ° to 0.27 ° for 655.36 Mbps).
This implies that for the system margin above zero, communication can be maintained
only when the system margin is positive. The data also sugge sts that the limiting FOV
is really a minimum FOV, but from practical point of view, the system margin for
large-angle FOV would be of greatest interest.
The multiple-forward scattering (MFS) model propagation model is applicable to
provide parameter values needed for communication system design under most atmo-
spheric conditions such as haze, dust, smoke, clouds, rain, snow etc. with or without
direct solar background. Although the results from a laboratory-simulation scattering
medium are presented here, the results and conclusions are useful in analyzing commu-
nication performance for a real scattering atmosphere. The multiple-forward-scattered
radiation can be utilized for the potential use of optical communications such as be-
tween aircraft or ground-to-UAV in presence of scattering medium such as dust, fog,
or smoke.
2.4. Optical Link Reliability
As we have seen that connection in free space is accomplished by means of a narrow

optical beam with low divergence. For a successful and reliable installation of optical
link, it is therefore necessary to know the steady parameters for standard atmosphere
T - 4.23
- - M\J - 7DOWInl2.sl .1dO
- - - - M),i• 7000 W;m2.$I. j{lTl
60
55
50
45
_ -- - - - 4 KBps
:: ~- - - -- -- - ..~:
15 ( ~ 116 Ma,,,
20
_ _ _ _ _ _ 3Z7.68 MBps
_ _ _ _ _ _ ___ 655.36 MBps
o 0.1 0.2 0.3 0.4 0.5 0.8 0.1 0.8 0.1 1.0
REl[).(lF·V1EW. 29}t(OEGREfj
Fig.7. System margin vs. FOY foropticaldepth =4.23for various datarates andtwo different
background noise radiances (reproduced with permission from Optical Society of America).
and statistical character of the weather in a given locality. The link budget includes all
average losses of optical power Pe [dBm] transmitted by the laser source, the received
power P; [dBm], receiversensitivity Ni, [dBm] and propagation loss L p [dBm]. If we
express link margin LM in dBm, we can write
(27)
LM is a basic link parameter that serves to express the reliability of optical link.
We can define the dynamic range as the interval of acceptable power in which the
link function is guaranteed with a definite error rate. The receiver is saturated when
P; > Ps ( a specified saturation value), and the required signal-to-noise ratio is not
provided when P; < Ns . Link reliability is quantified by the availability, which is
the percentage of time T av (%), when the data transmission bit rate is more than its
required value.The link availabilitycan equivalently be defined as the probability that
additional power losses LA caused by atmospheric effects (including absorption and
scattering) are less (in dB) than the link margin LM. The link availability can now be
expressed by means of a probability density p(aA) of an attenuation coefficient a A
(dB/km) as
f
UAlim
T av = 100% p(aA)daA , (28)

a
where aAlim is the limiting attenuation coefficient value, given by
72 Arun K. Majumdar
QAlim = LM(L)1000/L (29)
for a range L. The probability distribution of p(QA) can be determined from long-time
monitoring of the received signal level from a real measuring link, or using the data
collected over a local area in the past for a long time . In airports or meteorological sta-
tions, the visibility V(km) is monitored and is then converted to attenuation coefficient
as [3]
3 .91 ( A )
dB/ k m ) -- 4. 434 V (30)
QA (
-q
550
[the factor 4.34 comes from the relationship Q (dBlkm) =(10 loglOe) Q (km - 1 ) =4.34
1
Q (km- ) ] where the parameter q = 1.6 @ V >50, q = 1.3 @6< V <50, q = 0.16
V+0.34 @ 1< V <6, q = V- 0.5 @ 0.5 < V < I, q=O @V < 0.5. By evaluating the
above equation for T av for given limiting attenuation coefficients, the link availability
can be obtained, if we know the probability distribution of QA for a local area.
3. Laser Communication Performance Prediction and Analysis Under

Scintillation Conditions
Laser radiation propagating through turbulence develops both temporal and spatial
fluctuations ofirradiance, which is defined as scintillation. A lasercom system generally
integrates the signal spatially at the plane of a receiver, but will still see temporal
fluctuations in intensity. Scintillation is a serious issue for laser communications for
both terrestrial as well as ground-to-space or space-to-ground data links, as it can
produce large transient dips in the signal. The fading of the received signal below a
prescribed threshold temporarily degrades, or even annihilates, the link performance.
The probability density function (pdf) of the randomly fading irradiance signal due
to atmospheric turbulence determines the reliability of a communication link. Over
the years, many pdf models have been predicted for both weak and strong turbulence
cases. In this section we will summarize some of the proposed POFs that are relevant
to optical communications. We will evaluate the performance oflasercom system s that
operate under various atmospheric conditions. Each turbulent regime will be assumed
homogeneous and isotropic (can be spatially varying) with refractive-index structure
parameter C~ and inner scale size lo.Results from the literature for the pdfs for both
plane wave and spherical wave will be considered.
We now discuss some of the pdf models for irradiance fluctuations which will be
used for evaluating the performance of various lasercom systems such as horizontal
(terrestrial) link and slant paths involving downlink (from space-to-ground) and uplink
(ground-to-space) through atmospheric turbulence. One other parameter of interest is
log-intensity variance (scintillation index) . Expressions for point receiver and aperture-
averaged (for finite size detector) irradiance variance will be outlined.
For laser communications system applications it is desirable to have available
quantitative estimates of various statistical quantities that are associated with atmo-
spheric turbulence-induced irradiance fluctuations (scintillation)
Mo st common measure of the magnitude of scintillation is the scintillation index
defined by
2
CJ[ = (31)
where I is the irradiance of the optical wave and ( ) denotes the ensemble average
which is also equal to long-time averageassumingthe process to be ergodic. In weak
fluctuation theory, the scintillation index is proportional to the Rytov variance given
by
ai= 1.23C~e / 6 L 11/6 . (32)
C~ is thestrengthof turbulence, k =21ftA, is theoptical wavenumber, and L is the path

length between the communication transmitterand the receiver. For a horizontal path
of up to a fewkilometer, the valueof C~ is takento be a constant, whereasfor downlink
(space-to-ground) or uplink (ground-to-space), the C~ profile model as a function of
altitude h has to be used. For vertical or slant path, the so-called Hufnagel-Valley
(HV517) model is generally regarded as representative of continental conditions. The
value of 5 and 7 in the HV517 profile refer to the valuesof the atmospheric coherence
length (ro) in cm, and the isoplanatic angle (80) in microradians, respectively at the
wavelength of 0.55 JLm. At an wavelength of 1.315 JLm, the corresponding values are
14 em and 20 wad, respectively. [II]
3.0.1. Hufnagel-Valley Model
C~(h) = 0.00594(v/27)2(1O- s h)1O exp( -h/1000)

+ 2.7 X 10- 16 exp ( -h/1500) + A exp( -h/100) , (33)
where A = C~ (0) is the ground-level value of C~, and v is rms wind speed modeled
by
v= [15: lO' f,~O' (V'h+ + 30exp [- (h ~8~~OO)'])' dhr

v,
(34)
In Eq. (??), v g is theground windspeedand v» is the beamslewrate. For our numerical
calculations for various examples in this paper, we have assumed v = 21 m/s and
A = 1.7 X 10- 14 for the C~(h) model.
In some special cases such as tactical directed energy applications, it is necessary
to have a maritime C~(h) model supported by long-termdata. In contrast to HV517
model,there is no "standard" maritime C~ (h) model. A turbulence profilethat closely
approximates the low-altitude component of a maritime model can be written as the
following form:
C~(h) = C1 + C2 exp( -h/ C3) + C4 exp( -h / cs)(o :::; h:::; 6000m) , (35)
where the values of the constants are given [11] in Table 2 for three maritime cases
(best, median and worst):
In many practical cases either the HV517 or median maritime profile is used for
computational purposes.
3.i. Scintillation index: Point Receiver and Aperture Averaged Variance
After the optical beam propagates a sufficiently long path and/or due to large atmo-
sphericturbulence strengths, beamquality at the receivertends to breakup into regions
74 Arun K. Majumdar
Table 2. Values of the constants for maritime C~ profiles.
Case C);(h = 0) C1 C2 C3 C4 C5
(m- 2 / 3 ) (m- 2 / 3 ) (m- 2 / 3 ) (m) (m - 2/ 3 ) (m)
Best - 10% 1.0E - 16 9.8286E - 18 7.1609E - 17 100 1.9521E - 17 1500
Median - 50% 8.0E - 16 9.8583E - 18 4.9877 E - 16 300 2.9228E - 16 1200
Worst - 10% 1.0E - 14 9.2002E - 18 9.4387 E - 15 800 6.7328E - 16 1000
of high and low intensity which leads to significant signal fades. This is the case typi-
cally happens for a point receiver. If we now increase the size of the receiver aperture
relative to the size of theses "patches" of high and low intensity, this will basically
"average" these signal fluctuationsover the aperture size. This is called "aperture aver-
aging". It is therefore necessary to multiply the scintillation index for a point aperture
(i.e., diameter, D = 0) by the aperture averaging factor A to obtain the scintillation
index for a specific finite aperture (of size D) :
A= a; (D)/a';(O). (36)
The factor A is given by
A
16 fD0 br(p)
= 1rD2 [-1 (P)
cos D -; (1- ;22) t/2] pdp, (37)
where p is the separation distance between two points and b/(p) is the normalized
covariance function.
In this paper, we will consider basically three types of optical waves for generating
numerical examples of evaluatingcommunication performance:Plane Wave,Spherical
Wave and Beam Wave. For the space-to-ground communications downlink, the beam
is distorted as it enters down through the atmosphere. The ground-level scintillation
near the center of the received wave can be accurately modeled by a plane wave. [12]
For an uplink, where the atmospheric turbulence begins just outside the transmitting
aperture, we can assume a spherical wave. For a horizontal path communication link,
we can assume that the C~ is constant (for shorter paths, 10km or less), and we assume
Gaussian beam wave.
We follow the treatment given in [13] for the expressions of scintillation index
for plane and spherical waves including the aperture averaging effect for a finite size
aperture. For Gaussian beam wave, we will use the expressions given by [14].
Let us assume that the received irradiance 1 can be modeled as a product of two
terms, x (arises from large-scale turbulent eddies, and y (arises from small-scale ed-
dies): 1 = xy. Weassume that the two processes x and y to be statistically independent.
The received irradiance of the optical wave can thus be modeled as a modulation pro-
cess in which small-scale (diffractive) fluctuations are multiplicatively modulated by
large-scale (refractive) fluctuations. For normalized irradiance, (1) = 1, and we can
write the scintillation index as
(38)
Next we take into account of the finite size aperture effect, which is the factor A
as defined earlier. We can then finally write the expressions for aperture-averaged
scintillation index for negligible inner scale as given by [13].
3.1.1. Plane Wave
2 [ OA9ar
a/(D) = exp 7/ 6
(1 + 0.65d 2 + 1.1la~2/5)
0.5Ia12(1 + 0.69a 12/5) _5/6]
1
+ -1 (39)
1 + 0.90d 2 + 0.62d2a~2/5 '
where ar
is the Rytov variance for plane wave. Note that for constant C~, the Ry-
tov variance is given by Eq. (??), but for slant path (downlink or uplink), altitude-
dependenceof C~ (h) has to be taken into account. The parameter d is defined by
_J
d-
k D2
4L ' (40)
where L is the path length.
3.1.2. Spherical Wave
Similar expressions for the aperture-averaged scintillation index for spherical wave
are given by [13]
2
adD) = exp
[ OA9d 7 /6
(1 + O.I8d2 + 0.56a~2/5)
0.5Ia22(1 + 0.69a212/5) -5/6]
+ -1 (41)
1 + O.90d 2 + 0 .62d2a~2 / 5
where a~ is the Rytov variancefor spherical wave(note: a~ = 0.4 ar). Again note that
for slant path (downlink or uplink), altitude-dependence of C~ (h) has to be taken into
account in calculating a~ . We will use these scintillation index values for analyzing
the communication performance examples later in this paper.
3.1.3. Gaussian beam wave [14]
For the Gaussian beam wave, the beam size (radius) at the receiver is given by
(42)
76 Arun K. Majumdar
=
where Wo is the transmitter beam size (radius), 1'( L) (R o - L) / R o is the normalized
transmitter focusing parameter, R o is the phase front radius of curvature at the trans-
mitter, Z = L / Zd is the normalized distance, Zd = kW5/2 is diffractive distance, L
is the communication channel propagation distance. The global coherence parameter
( is defined by
(= ( s + -2W5
2. (43)
Po
In the above equation (s is the coherence of the source beam «(s = I for coherent
beam) and po(L) = (0.55C;k2L)-3/5 is the coherence length of a spherical wave
propagating in the atmospheric turbulence.
The scintillation index for the Gaussian beam wave for a point receiver is given
by [14]
2
a} ~ 4.42arz~lc6 W;(L) + 3.86ar {0.4 [(1 + 21'rec)2 + 4Zrect/12
[5
x cos -6 tan
-1 [1 +2i\ec]
2-
Zrec
11_
- -6 Zrec
1
5/ 6] }
, (44)
where ar
= 1.23C;e / 6 L 11 / 6 is the Rytov variance for a plane wave. The receiver
beam parameters are given by [15]
_ R(L) + L L
rrec(L) = R(L) and zrec(L) = 0.5kw 2(L) · (45)
The phase front radius of curvature at the receiver is defined by [14]

2
R(L) = L(1' + ( Z2) and </> == ~ _ Zw~ . (46)
</>z - (Z2 - 1'2 Z Po
As before, the aperture-averaged scintillation index for Gaussian beam wave case
can be obtained by multiplying the point-receiver scintillation index by the aperture
average factor, Aa for Gaussian beam wave case given by [15]
2 2
16 jl {_D X ( 2 + w5Z2
P5 P5</>2 )}
Aa = -; x dx exp --pr- - w 2(L)
o
x [cos - (x) - x(l - X2)1 /2] .
1
(47)
In the above equation x = piD with D being the aperture diameter.
3.2. Probability Density Function (pdf) Models
Free-space laser communications have the potential for offering extremely high data
rate links and are very suitable where fiber optic links are not practical, such as between
buildings across cities and space-to-ground or ground-to-space data links. To establish
high reliable free-space laser communications links quantitative estimates of various
statistical quantities that are associated with atmospheric turbulence-induced irradi-
ance fluctuations (scintillation) are absolutely necessary. The signal fading severely
Free-spacelaser communication performance in the atmosphericchannel 77
degrades the link performance and the bit error probability of the received information
signal is increased above an unacceptable limit. Scintillation can thus cause losses in
received power which can lead to ultimate break in communications. The form of the
probability density function (pdf) of the randomly fading irradiance signal determines
the extent of reliability of an optical communication system. An accurate pdf is there-
fore necessary in designing a reliable lasercom system . Understanding optical propaga-
tion through turbulence is extremely complex because of the complicated nature of the
generation of atmospheric turbulence occurring in different situations. There is no one
single pdf which can describe correctly the random fluctuations through every type of
turbulent condition. Most of the treatment available for optical propagation are based
on statistical theory, and is valid for isotropic or homogeneous turbulence following
Kolmogorov's law. Assumptions of stationarity and ergoticity help to simplify the cal-
culation (because the ensemble average in that case is the same as long-time average).
It was pointed out that the higher-order moments can give more accurate informa-
tion concerning the contributions from the tails of the probability density function. 16
Detection and fade probabilities are primarily based on the tails of the pdf .and there-
fore underestimating this region significantly affects the accuracy of communication
performance analysis .
The moments, central moments and other higher-order coefficients up to eighth
order (non-dimensional) in statistical analysis of irradiance fluctuations with a given
pdf were obtained using the equations [17,18]:
!
00
kp(1)dI
kth moments : m k = I , (48)
o
kth central moments : /-Lk = ((1 - (I ) )k ) , (49)

where I stands for irradiance, p(1) is the pdf of I .and () indicates an ensemble average.
(50)
Excess : r 4 = /-L4/ /-L~ - 3, (51)

r
Superskewness : 5 = /-L5/ /-L3/-L2 - 10, (52)
Super excess : r6 = /-L6/ /-L~ - 15, (53)
Hyperskewness : n = /-L7 / /-L3/-L~ - 105, (54)
Hyper ex cess : n = /-Ls / /-L~ - 105. (55)
Note that, for a Gaus sian pdf, the odd-order coefficients nand n
do not exist
and become indeterminate, wherea s the even-order coefficients n
= r 6 = = o. n
Therefore any nonzero value of the even-order parameters indicates that the random
process deviates from a Gaussian distribution.
The higher-order coefficients (up to eighth order) defined above can give much
more detailed information concerning the contribution from the tails of non-Gaussian
distributions and were utilized in [19] to compare model distributions (such as K, log-
normal , Rice-Nakagami and a universal statistical model distribution) with the mea-
sured data of 0.6328 /-Lm laser-irradiance fluctuations propagated through a laboratory-
simulated turbulence. [16,17] The laboratory atmospheric turbulence generator is de-
scribed in details elsewhere. [20,21]
78 Arun K. Majumdar
From practical point of view, it is desirable to have available a tractable pdf model
for the irradiance fluctuations so that we can predict the communication performance
with an acceptable accuracy. Log-normal distribution is the most widely used model
for the probability density function of the irradiance due to its simplicity : this pdf
however is only applicable to weak turbulen ce regime . As the strength of turbulence
increases, multiple scattering effects become important and significant deviations from
log-normal statistics are exhibited in experimental data. Log-normal pdf underesti-
mates the behavior in the tails as compared with measurement results . For communi-
cation systems , accurate detection and fade probabilities primarily depend on the tails
of the pdf, hence underestimating this region significantly affects the accura cy of the
communication performance. As mentioned earlier, many statistical models have been
proposed over the years to describe scintillation characteristics under a wide range of
turbulence conditions, e.g., K distribution, 1 - K distribution, and log-normal Rician
channel. In a recent series of paper a gamma-gamma pdf has been proposed [13,22]
based on work by Nakagami. [23] This model is a two-parameter distribution which
is based on a doubly stochastic theory of scintillation, and assumes that small-scale
irradiance fluctuations are modulated by large-scale irradiance fluctuations of the prop-
agating wave, both governed by independent gamma distributions. The gamma-gamma
distribution can be directly related to the atmospheric parameters defining turbulence
conditions. In this chapter, we will consider mainly the two pdfs, namely, log-normal
and gamma-gamma: log-normal distribution is applicable to weak turbulence case and
the gamma-gamma distribution is assumed to valid for both weak and strong turbu-
lence regimes . These two pdfs which are used in this paper for computational purpose
of evaluating communication performance are given by
Log-normal distribution:
1 { [In (6)) + ~aJ(D)r} 1> 0, (56)

p(I) = 1aI(D)v'27i exp - 2ai(D) ,
where aI(D) is the aperture-averaged scintillation index as defined earlier, and (1) is
the mean intensity.
Gamma-gamma probability density function:
2(afJ)( a+ 13l/ 2 (0+13 )/2 _ 1 ( )

p(I) = r(a)r(fJ) 1 K a - 13 2JafJ1 , (57)
where K a ( ) is the modified Bessel function of the second kind ofordera. Here oand d
are the effective number of small -scale and large scale eddies of the turbulent environ-
ment. These parameters are directly related to the atmospheric turbulence conditions
as [13,22]
Plane Wave with aperture-averaged scintillation index :
- 1
OA9ar
a = exp 7/6 - 1 , (58)
{ [
(1 + 0.65d 2 + 1.1la~2/5) ] }
0.510"12 ( 1 + 0.690" 112/ 5)-5/ 6] } -1

(3 = exp - 1 (59)
{ [ 1 + 0.90d 2 + 0.62d20"~2 /5
Spherical Wave with aperture-averaged scintillation index:

_ 1
0.490"~
0= exp 7/6 -1 (60)
{ [
(1 + 0.18d2 + 0.560"~2/5) ] }
- 1
0.510"22 ( 1 + 0.690"212/ 5) _5/ 6]
(3 = exp -1 , (61)
{ [ 1 + 0.90d2 + 0 .62d20"~2 /5 }
where the parameters 0"1, 0"2and d were defined earlier.
3.3. Received Signal-to-Noise -Ratio (SNR) and Bit-Error-Rate (BER)
In almost every area of measurements, the ultimate limit to the detectability of a weak
signal is set by noise-unwanted signals that obscure the desired signal. The same is
true for free-space laser communications systems where the shot noise, background
noise and thermal noise contribute to the total noise at the receiver. The goal of a
lasercom system for digital communication is to transmit the maximum number of
bits per second over the maximum possible range with the fewest errors. Electrical
data signals are converted to optical signals via a modulator. A "I " is transmitted
as a pulse of light while a "0" has no light output. The number of "I 's" and "O's"
transmitted per second determ ines the speed of the link (bit rate). At the receiving end
of the link, the optical signal is detected by an optical-to-electrical converter (e.g., a
photo detector). A decision circuit then identifies the " I's" and "O's" in the signal, and
thus recovers the information sent.
For digital communication system, information is sent over an optical link as digital
symbols . This is accompanied by encoding the source information into binary symbols
(bits) and transmitting the bits as some type of coded optical field, for example, by
encoding on a bit-by-bit basis (binary encoding). Each bit is then sent individually
by transmitting one of two optical fields to represent each bit. In this paper we will
consider only direct detection (DD) system in which the standard binary procedure is
to pulse an optical source (e.g., a laser or an LED) on or off depending on data bit.
This encoding is referred to as on-off keying (OaK). At the receiver OaK decoding is
based on a decision as to whether the pulse slot time has high enough field energy or
not. The selected threshold determines the best performance in decoding the correct
signal with the lowest probability of making a bit decision error and thus the bit error
rate (BER) can be obtained. The other modulation format of pulse position modulation
(PPM) will also be considered.
3.3.1. Relationship between SNR and BER

The performance of the lasercom system can be predicted by computing the bit-error-
rate (BER) of the system which depends on the modulation format, and the signal-to-
noise ratio (SNR) where the noise contributions come from all possible sources which
80 Arun K. Majumdar
include signal shot noise, dark current noise, thermal/Johnson noise in the electronic s
following the photo detector, and the background noise. Assuming a Gaussian distri-
bution of noise, the SNR at the output of the photodetctor in the absence of turbulence
is given by [13]
Ps
SNRo = - - ; = = = = = = = = = = = = (62)
( 2h~ B ) (Ps + PB) + ( ~~) 2 ( 4k~N B)
where Ps is the signal power of the optical transmitter and PB is the background
noise, both in watts, 'T] is the detector quantum efficiency, e is the electronic charge
in coulomb s, h is the Planck's constant, v is the optical frequency in hertz, k is the
Boltzmann constant, B is the bandwidth of the (detector) filter, TN is the effective
noise temperature and R is the effective input resistance to the amplifier of the detector.
For shot noise limited operation, the background noise and the thermal noise can be
ignored.
In presence of turbulence, the SNR is a fluctuating term (i.e., an instantaneous
value) and the average (mean) value is to be taken The mean SNR can be expressed as
[ 13]
(SNR) = SNRo , (63)
J (Iso) + a 7(D) SNRJ
where SNR o is the signal-to-noise-ratio in absence of turbulence defined earlier, Pso
is the signal power in the absence of atmospheric effects, the mean input signal power
(Ps) (i.e., the mean of the instantaneou s input signal power Ps ), and a;(D)is the
aperture-averaged scintillation index previously discussed.
Let us now under stand the communication performance criteria defined by the
parameter BER. The signal entering the decision circuit fluctuates due to the various
noise mechani sms. The bit-error-rate (BER) is the probabilit y of incorrect bit identifi-
cation by the decision circuit. The design of optical communication systems relies on
the estimation of performan ce figure of merit such as bit-error-rate (BER). One of the
most commonly used parameters is the receiver sensitivity defined as the minimum
averaged received optical power required to achieve a given HER. In order to simulate
the characteristics of optical links or sub-systems, it is therefore essential to be able
to compute the BER in an effective and reliable way. It is important to investigate
the degree to which the optical communication may be expected to be degraded by
atmospheric turbulence, by calculating the expected increase in bit error probability
due to turbulence, including the effect of a finite aperture receiver.
3.3.2. Why Atmospheric Turbulence Increases the Bit Error Rate?
For any type of detector, we can assume that the bit error probability (i.e., the relative
frequency of errors) is functionally related to the signal strength (denoted by s) which
can be defined as the number of signal photons per bit incident on the detector. The bit
error probability can be written as E (s) for a signal pulse of strength s. If the signal
strength fluctuates randomly due to atmospheric turbulence, s becomes a random
variable with a probability distribution P (s). Therefore the bit error probability will
also become a random variable. Most of the statistical information about E is contained
in the first two moments of the distribution (the mean and the variance), which can be
expressed as
!
00
(E ) = E( s)P( s)ds, (64)

o
var(E) = !
o
00
(E - (E ) )2 ds, (65)
where (E) denotes ensemble average and "var" indicates the variance . If we express
E(s) as a finite Taylor formula about (s), with a second degree error term, we can
write
!
00
(E ) = E((s)) + 1/2 E(2)((S) + 9(s)(s- (s))(s - (S)))2P( s)ds, (66)

o
where E(2) is the second derivative of E(s) , and 9(s) is between oand I for each value
of s, but is otherwise completely unknown. It is evident from the above equation that
(E) is always greaterthan E( (s)) if the second derivative of E(s) is positive for all s.
Therefore a sufficient condition for atmospheric turbulence to causean increasein the
bit error rate is that the second derivative of the bit errorprobability function, E( s)
be everywhere positive. This can be understood intuitively by observing the typical
"BER" curves vs. signal strength which is decreasing with s, and highly nonlinear,
with a positive second derivative. As the signal s fluctuates about its mean (s ), value
of E given by E( s) oscillates along the curve, to either side of the point ((s), E( (s)).
But since the curve is decrea sing with s, but more slowly as s increases, the increase
in E as the point moves to the left is greater in magnitude than the decrease in E as
the point moves to the right. Thus the average value of E is biased upward from its
value at (s) .
In this paper we will assume lognormal and gamma-gamma distributions (depend -
ing on the strength of turbulence and the communication link scenario) arising from
atmospheric turbulence .
3.3.3. Bit-Error Rate Computation for OOK Modulation
One of the primary considerations in the design of a communication system is the

performance of the system when it is corrupted by noise. We have discussed the various
sources of noise in a lasercom system including fluctuating random noise introduced by
the turbulence environment. In a binary on-off keying (OaK) system, the transmitter
sends a pulse of light into a channel to represent a "I" and does not send any light
for a "0". When signal plus noise is present at the receiver input, there are two ways
in which errors can occur. The receiver decides a "0" was sent when in fact a "I"
was transmitted . Also, the receiver may decide a "I" was transmitted when a "0" was
actually sent. The probability of error or the bit-error rate (BER) can be expressed as
r ; = P( 11 O)P(O) + P( Oll)P(l), (67)
where P(O) and P(l) are the probability ofa binary "0" and "I" respectively, and
P( 11 0) and P( 011) are the conditional probabilities. Assuming the transmitter is
82 Arun K. Majumdar
sending"I "s and "O"s withequal probability, each has a probability equaled to 1h, so
the probability of bit error is then given by
(68)
We also assume the Gaussian distribution for both noise as well as signal plus noise,
the probability of error in absence of turbulence is given by
BERD =!2 (SNRD)

2V2
. (69)
In presenceof turbulence, the probability of error is given by [13,24]

00
BER = ! !pI(s)erfc ( (~~)s) ds, (70)

2 2 2(zs)
o
wherePI (s) is the probability distribution of irradiance, is is the instantaneous signal
current whosemean valueis givenby (is) = TJe(Ps) [lu/, (Ps) being the meansignal
power, and (SNR) is the meanSNR in presenceof turbulence definedearlier. Depend-
ing on the level of turbulence and the communication link scenario, various forms of
the probability distributions can be modeled. In this chapter we will compute BER
using lognormal and gamma-gamma probability distributions. Note that the proba-
bility distribution contains the parameter s which is related to the aperture-averaged
scintillation index. Some of the numericalresults showingBER computations will be
presented later in the paper.
3.3.4. Bit-Error Rate Computation for Pulse Position Modulation(PPM) :

Some Basics
We presentedthe conceptof pulse position modulation (PPM) in the earlier section to
evaluateperformanceof laser communication system whichhas to operate under scat-
tering medium. In that case we assumed a non fluctuating medium, i.e. no turbulence
mediumwas assumed.In manysituationsof practicalimportance, we need to take into
account of the turbulence environment. It has been shown that M -ary pulse-position
modulation is the optimum modulation format for background-limited optical com-
munication. In this type of modulation, the numberof bits sent per PPM wordor pulse
is 10gzM = k and the PPM assumesa mechanism for achieving time synchronization
having an averagedata rate equal to
10gzM
(71)
R = (M + N)Lj.T
In a basic time period called the interpulseinterval, one has to determinein whichone
of the M slots, each of width fJ.T seconds, the received pulse resides. The difference
between the interpulseinterval and M fJ.T, the word time, is called the dead time and
is denoted N fJ.T, where N in most situations is not an integer. M is chosen in order
to maximizethe averagedata rate, so that the maximum data rate is given by [25]
logz Mo p t ]
R m a x = R pm a x [ 1 + aMo p t '
(72)
where Rpmax is the maximum pulse repetition rate of the system which can be ex-
pressed in terms of pulse repetition rate R p as follows
1 i1T
Rp = ---,-------..,.- for a= -
TD(l + aM) TD
1
RPmax = TD for a = 0, (73)
where i1T is the slot width, TD is the dead time, a for maximumdata rate is givenby
(74)
In order to transmitat a lowerdata rate, one can use a lower Rpmax or a smaller value
of M. If one has to operate at a higher data rate, one would require either a higher
value for RPmax or the use of multiple frequencies to increase M .
The probability of making an error in demodulating the PPM symbol, PE , is
expressed as [25]
vk f
00
PE = 1 - exp {-(x - SNR)2/ 2} (1- erfc(x))M-tdx , (75)

- 00
where erfc(x) is a complimentary error function .

The above equation can be approximated as
(M -1)
P E ~ (1l"8NR)1 /2 exp( -SNR/4) , (76)
2
where the signal-to-noise ratio, SNR is SNR = ~ with Ks is the signal count and
K n is the noise count. Finally, the bit error rate, BER can be related to PE in the
following way [25]:
2k - 1
BER= -k 2 PE, (77)
-1
where k is the numberof bits sent per PPM word or pulse, as definedearlier.
Note that in the above derivations, we have assumed that we have ignored pulses
which overlap more than one slot, i.e., pulse broadeningby the optical scatter channel
is assumedless than i1T, with perfectsynchronization. However, we havealso studied
the effectof pulsestretchingand will be discussedin details in numerical examplelater
on in this paper.
3.4. Probability of Fade
The object of designingany communication system is to obtain uninterrupted commu-

nicationsbetweenthe transmitterand the receiverduring the exchangeof information
of sending and receiving data. We have discussed earlier that because of the presence
of randomly varying atmospheric channel, the received signal fluctuates so that it is
possible for the received signal to fall below an acceptable detection level and error
rate. The fading probability for an optical communication link can be determined from
84 Arun K. Majumdar
the knowledge of statistics of intensity fluctuations (pdt) and the scintillation index.
The reliability of a communication system thus depends on the fading probability.
Let us denote I (t) as the instantaneous value of irradiance normalized to its mean.
We need to obtain the fraction of time I(t) 2: Jr, where Ir is the threshold level
of intensity specified by the communication system. Because of the assumption of an
ergodic process for the statistics of optical propagation in our case, the time averages
are equal to ensemble averages, so that we can also write the fraction of time of a fade
where I ::; IT as
f
IT
Fraction(I::; Jr) = CPI(I:::; IT) = PI(I)dI, (78)

o
where C PI is the cumulative probability for irradiance and is assumed to be either
a lognormal (for weak turbulence) or a Gamma-Gamma distribution (for weak to
strong turbulence) model, for the numerical examples described in this papter. The
fade threshold parameter FT ofa signal (in decibels) can be defined by [13,26]
F T = 1OIog lO ((I (0, L))/ Jr ), (79)
where (1(0, L)) is the on-axis mean intensity at a communication distance L. We can
also write
In [(I(~~L) )] = -0.23FT. (80)
In the numerical examples described in the following section , we show the probability
of fade as a function of the threshold parameter F T (given in dB below the mean on-
axis intensity at the receiver) for various communication scenarios, such as downlink,
uplink and horizontal links and for two model pdfs (lognormal and Gamma-Gamma).
4. Example Numerical Results
In this section we present numerical results generated by computer programs to predict

the communication performance for various specific scenarios of laser communication
links.
4.1. Example I: Uplink (Spherical Wave/OaK)
For an uplink lasercom link, i.e., from a ground-based terminal transmitting to a satel-
lite, the atmospheric turbulence begins just outside the transmitter aperture, and we
assume a spherical wave. The aperture-averaged scintillation index (17 (D) is calculated
from the parameters of the propagation path (propagation model, satellite altitude, and
zenith angle) and the wavelength. For computational purpose, we assume a Hufnagel-
Valley (H-V) turbulence profile discussed earlier with A = I.7 X 10- 14 m - 2 / 3 , and
v = 2 I m/s to represent the H-V517 model conditions. We predict lasercom system
performance by first computing the BER. The parameters used for the OOK modu-
lation format for uplink are the following : wavelength A = 1.55 JIm, satellite height
H = 3.8x 105 m, transmitter divergent angle = 17.45 JIrad, and a data rate of2 Gbit/s.
----
J
O. J
TTPLTNK ... I
0.0 I
~ I I
ffi 1.10- 3 '\""'J I
~ -
~ 1 . 10 4 \ "" , I D -O
~
] 1.10- 5
\ 1 ,\, ~ D = O.4 m
\71/=o,m
::a 1·10- (,
\1
~ 1.10- 7 \ \ }{
II \ 1\
_ ~oTurbulenc~ 1\ ~-~
I I\ I \\---:-:--
'.
10 15 20
<S~ R> (dB)
Fig. 8. Uplink Mean bit-error-rate foron-off-keying direct detection as afunction ofmean signal-
to-noise-ratio for various receiver aperture diameters [D = 0 (point receiver), and D = 0.4 m,
0,6 m]. The zenith angle is 45 degrees. The no-turbulence case is also shown.
The BERcalculationis based on a Gamma-Gamma probability density functionof the

intensity fluctuations and the procedurewas outlined earlier.
Figure 8 shows the mean BER as a function of average signal-to-noise ratio
((SNR)) for a point receiver (D = 0) and a receiver with diameter D = 40 em
and D = 60 cm, with zenith angle = 45°, and gamma-gamma PDF. The effect of
atmospheric turbulenceon BER is clearly apparent: at (SNR) of about II dB, the BER
increases from a value of 10- 9 for no turbulence case by severalorders of magnitude
to about 1.9 x 10- 4 for a point receiveruplink communication with turbulence. Also
seen from the figure is that increasingthe aperture diametercan improveBER as pre-
dicted (for example, for a BER of 10- 9 one needs average (SNR) of approximately
16.97 dB for point receiver, 16.32 dB for a diameter of D = 40 ern, and 15.74 dB for
D = 60 ern), For a communication systemwithavailable(SNR) of 16.35 dB, the BER
is decreasedfrom about 3.6 x 10- 9 for a point receiverto 10- 9 for a receiverdiameter
of 40 cm. Increasingthe aperture diameter can thus improveBER as predicted.
Figure9 shows the mean BER as a function of meanSNR for three zenith angles:
0, 45 and 60 degrees. A gamma-gamma distribution with spherical wave model was
assumedand the effects of inner scale were ignored. The effect of turbulence on mean
bit-error-rate is clear: for example for a mean signal-to-noise-ratio of 10 dB the mean
bit-error-rate increasesfrom a valueof 2.5 x 10- 7 (no turbulence) to about 5 x 10- 4
(severalorders of magnitude) for a zenith angle of 60 degrees. The figure shows that
an increase in zenith angle requires a greater SNR for a given BER For example, to
achieve a BER of 10- 9 , the required SNR is increased from 15.42 dB (0 degree) to
16.06 dB (60 degree).
Nextwecalculatethe probability of fadefor the sameuplinksphericalwavemodel
to the satelliteundersimilarconditionsfor two PDFs,namely, gamma-gamma and log-
normaldistributions, andfor botha pointreceiver(D = 0) andfor a numberof receivers
with diameters of D = 20, 40, and 60 cm. Figure 10 shows the probability-of-fade
curves as a function of threshold level F T below the mean irradiance. A lower proba-
bility of fade is predicted for the log-normal PDF than for gamma-gamma. The fade
86 Arun K. Majumdar
0.0 I
- I
~
I
TTPT.1'N'1(
"",-
I
I
I
-, I \ -, I
) ·10
I \ "'- I
-,
1· 10
I \ \. I
I "~ 1
1· 10' ..s"~
I l'I:oTu rbu lenc;c........ ~ Al
1
\ OdCI!-r«'
-~~g= 1
1 \
'0
< SI\;R> /lilt) " '0
Fig.9. UplinkMeanbit error rate for on-offkeyingdirectdetectionas a function of mean signal-

to-noise-ratio for three zenith angles: 0, 45 and 60 degrees with aperture diameter D = 60 em.
and for the no-turbulence case.
0
I ~ -!·!..?-!. . .~-I( I
o. I
..-------- I I I
0.0 I
I -:) - O ~' 40 '", _ I - - -
I"-,,",- .1
~ 1·10- ·\ I ~~'" 1/ O_" .'m_ 1 - - -
~ 1 ,10-" I I "'-"-': ! / / '"'"'""'~;._ -

~ I ' I O-~ ~D - O~~{:'/
J: 1 ·10- /'1 I ~.:":.-nc:::~"" .... .:
I D- 60 , m' I Y/;
J-Y\ "
1·10-'
1 0- 11
01
I ~'N"mm1 " "',
1 10'
0
I I I 1\ " -,
I I I \ \ -,
1 .10- 10
, 10 12
Thn:shokl !.c vc1l<HlI
Fig. 10. UplinkProbability of fade vs. threshold level for two POFs: Gamma-Gammaand Log-
Normaldistributions and for three receiverdiameters D and zenithangleof 45 degrees. The case
D = 0 represents a point receiver.
probability is also reduced by increasing the aperture . For gamma -gamma distribution
with a 6-dB threshold level, for example, the fade probability is decreased from 2.2 x
10- 4 for a point receiver to about 2 x 10- 5 for an aperture diameter of 60 em. The
similar conclusion is drawn for the log-normal distribution where the fade probability
is decreased from 4.5 x 10- 5 for a point receiver to about 1.8 x 10- 6 for a diameter
of60cm.
Decreasing the zenith angle also improves the predicted fade probability in the
presence of atmospheric turbulence . Figure II depicts the probability of fade as a
function of threshold level for three zenith angles: 0, 45 and 60 degrees, for both
gamma-gamma and log-normal distributions. To achieve a probability of fade of 10- 9
with a gamma-gamma distribution, for example, the threshold levels required increase
from 8.96 dB for 0 degree to 9.7 dB for 60 degrees. The similar conclusion is valid also
~,I
o.
0.0 I
I r'..' ":a. I
TTP T.1N K
I
j I ' ~;~ " I
1·IO-
I I .:\:" I Odcgrec
...~ , .rc-' --
I
'''~ ''
4Sdc gfl..
'CS
~ I 'I O-~
/
j I
~'\:/ V I "'de""",
1 _IO- to
-"''' ' I h '~~·. J
45 dcgrccs • .11 -
', ... .- Gamma.fja mma
I
I - Ode.~"~tr2 "'yr " /
_ Log.Nunna l
I I I '. .....\ ...

•
Thrcsbold Level (dB)
io
"
Fig. 11. Uplink Probability of fade vs. threshold level for three zenith angles: 0.45 and 60
degrees. and for two PDFs: Gamma-Gamma and Log-Normal distributions.
for log-normal distribution. The log-normal distribution needs lower threshold levels
than those for gamma-gamma distribution for a fixed probability of fade value.
4.2. Example 2: Downlink (Plane Wave/OaK)
For a downlink path from a satellite to a ground terminal, the ground-level scintillation
near the center of the received wave can be accurately modeled by a plane wave [12].
The Rytov variance in this case depends mostly on high-altitude turbulence, and is
consistent with weak-fluctuation theory except for the case of very large zenith angle.
We evaluated the system performance by again calculating the average bit error rate
(BER) for a plane wave using OOK modulation scheme . The same values of the
parameters A and v for the H-V atmospheric model as used for the uplink case were
taken. The fixed system parameters were also taken to be the same as for the uplink case,
except for transmitter divergence = 5.1 wad, data rate = 2 Gbit/s , receiver aperture
diameters D = 0 (point receiver), D = 10 and 40 cm. Figure 12 shows the numerical
results of the mean bit error rate as a function of average signal-to-noise-ratio (SNR).
As in the uplink case, it is again apparent that the atmospheric turbulence degrades
the system performance: for (SNR) of 10.8 dB, the BER increases from 10- 9 (no
turbulence case) to about 10- 4 with D = 40 em, and to about 2 x 10- 4 with a
point receiver (D = 0). Increasing the aperture helps in improving the system BER,
for example, one needs approximately 16.21 dB available (SNR) while 16.77 dB of
(SNR) is required for a point receiver to achieve the same mean bit error rate (BER)
of 10- 9 • Figure 13 shows the values of (BER) as a function of (SNR ) for different
zenith angles : ~ = 0°, 45° and 60° . A fixed aperture diameter of D = 40 cm was
assumed. The BER is increased from I 0-9 to approximately 2.8 x 10- 9 if we increase
the zenith angle from 0° to 60° for available (SNR) of 15.75 dB . Also in order to
achieve the same (BER), say, 10- 9, the (SNR) has to be increased from 15.75 dB at
0° zenith angle to 16.38 dB at 60° zenith angle .
The probability of fade variation is similar to the uplink case. Figure 14 shows the
probability of fade as a function of threshold level in dB for two aperture diameters,
88 Arun K. Majumdar
0. 1
r--- nnw N'T.1NK
<,....
-,.......
\ ', " . n =o
'.:'
\
\ .,.:,.A- 04 m
-
~ ' J,
-,
No T urbnlr- nr-r- .... \ \
I ~ " \
\ "
~ "
10 15 20
<SNR> (dB)
Fig. 12. Downlink Mean bit-error-ratefor on-off keyingdirect detection as a function of mean
signal-to-noise-ratio fortworeceiver aperture diameters and the zenith angle of 45 degrees. The
no-turbulence case is also shown.
~ 1-10- )
~ 4
0.0 1
) _10-
I
0, I
-- "' n n w N T. Th1K
........
" '..........
\
I
1
'f',.
I
I
I
I
,:1
g 1,10- ' \ 1\ . I
"'
c; 1.10- (, \1 \ \ . II 60 degrees
\
~ 1.10'
I .IO-I!
1\ ~d"
., ' ' '
)· 10'
So Turbult'nct'4
\ 1:\-..
..--"" \'
odegrec: , ...
1.10- 10
I\
<SS R> (dU)
'0 15 '0
Fig. 13. Downlink Mean bit-error-rate for on-off keying direct detection vs. mean signal-to-
noise-ratio for three different zenithangles: 0,45 and60degrees, andaperture diameter D = 40
cm.The no-turbulence case is also shown.
D = 0 cm (point receiver), D = 40 em. A fixed value of the zenith angle ~ =

45° was assumed. The two probability density functions for irradianee fluctuations
were assumed: Gamma-Gamma and log-normal. The bit error rate values seem to be
lower for log-normal distribution than the Gamma-Gamma distribution. For example,
for a threshold level of 8 dB and for a 40-cm-diameter aperture, the mean bit error
rate is increased from 2 x 10- 9 for log-normal distribution to the value 1.6 x 10- 7
if one assumes Gamma-Gamma distribution. Increasing the aperture diameter also
decreases the probability of fade, as expected, for example using a Gamma-Gamma
distribution model, and for a fixed threshold level of 9.6 dB, the probability of fade is
decreased from 3 x 10- 8 for a point receiver (D = 0) to a value of 10- 9 if we use a
I
- ,- :-...•.,. = DOW NLINK
1
~,,::..~-
I
0 01
j
~~ ;~I= - - --
I ' I O-
~ - -. ."1..
- - -I- - -I------..-:-1'"
~ 1· )0 ~
-:-""~ - D'O ---
E 1.10- 5
=
Z
! I.IU-t> -, ~
.- :1 D -O
l..:' A f Gamma-
1
1·10'
-. .. D - 4 0cm ,:- ~ L
Gamma
1·10 _ .
- I Log-Normal ) L
L.-
:u :s::=•
====i ~
, -.;-
10 12
Thl':Shokl Level (0,111)
Fig. 14. Downlink Probability of fade vs. threshold level for two PDFs: Gamma-Gamma and
Log-Normal distributions, two receiver diameters and zenith angle 45 degrees.
0.1 ~!I , I
I
n nw
I
NT.1NK
001
... I ·I O-l
I ~ I
~ 1 · \0- 4
I I,,~ I - Odo!t1 _
f
_ _ __
I I "\~ / 4St.lc:grecs
E l ·IO- S
i 1'10- "
I o ,,;.,.ys ··~y //',---
I 4~dc:grc~ -.'\'_ j
_"....,lor.. fN': _ ~ ·
1·10'
I ·I O- M
I I I \ ...• " Gamma.(Jamma
1·10'
I I I \ · ,,"~ 17
~'"
Log -I\.'ormat --..:.
I I , '?(
1.10- 10
, '0 12
ThH:shold l evel hllJ)
Fig. 15. Downlink Probability of fade vs. threshold level for three zenith angle s: 0, 45 and 60
degrees , and for two PDFs : Gamma-Gamma and Log-Normal distributions. The receiver diamter
D =40cm.
diameter of 40 em. The similar trend follows for log-normal distribution. Finally, Fig.
15 shows the probability of fade as a function of threshold level in dB for three zenith
angles ~ = 0 0 ,45 0 and 60 0 • The effect of using both Gamma-Gamma and log-normal
distributions are also depicted in the same figure. The similar trends are also observed :
for a fixed threshold level, increas ing the zenith angle causes the probability of fade
to increase (approximately even by an order of magnitude) for both the distributions,
thus degrading the system performance at higher zenith angles.
4.3. Example 3: Horizontal: Terrestrial Link (Gaussian Beam Wave/OOK)
We provide here results for a numerical example of the average bit error rate (BER) of a
horizontal link such as between buildings or mountaintops. This direct-detection binary
lasercom link consists of a laser transmitter, atmospheric channel containing clear-air
turbulence, and a maximum-likelihood receiver as described by other researchers.
90 Arun K. Majumdar
[14] The maximum-likelihood receiver consists of an avalanche photodiode (APD)

detector, a matched filter implemented in the form of a moving integrator, and a clock
comparator which performs a threshold test at bit interval boundaries. The excess noise
for the APD is characterized by a factor, F given by
F = keff . G + (1 - keff)(2 - 1/G) , (81)
where G is the average current gain, keff is the ratio of the ionization coefficient for
holes to electrons. The expression for BER for no turbulence is given by [13]
1
NoThrbulence : BER = 2" (PFAlSEAlarrn + PMISS)
= Q (YTh- < Y~b » + Q (YTh- < Y~b » ,(82)
2ao 2al
where Q(x) is the Gaussian tail integral defined by
Q(x) = ~!exp (-t) dt = ~erfc (~) . (83)

x
The decision threshold value YTh above is solved from a quadratic equation:
(:~ - 1) Y~h + 2 ((Y~b) - :~ (Y~b) ) YTh - a~ In (:~)

a~ 0 2 1 2
+ 2"
a
(YTb) - (YTb) = 0, (84)
o
where (Y~b) and a? are the mean and variance respectively of the moving integrator
output of the Gaussian APD photocurrent under ON state, and (Y~b) and a6 are the
same parameters for OFF state. The BER for turbulence is given by [14]
1
Turbulence : BER = 2" (PFAlSEAlarm + A.nss) ,
where
YTh - eGcnq, 1
PFAISEAlarm = 00 Q ( )
[ J2B(nq,ce 2FG2 + 2KBTIRL) SJ21rAa?nS
- (In S + ~Aa?n S)2)

x exp ( 2A 2
a lnS
dS
P, YTh - eGcnq, 1
MISS = 00 Q( )
[ J2B(nq,e 2FG2 + 2KBTIRL) SJ21rAa?nS
- (In s + ~Aa?ns)2) (85)

x exp ( 2A 2 dS.
a 1nS
Figures 16 and 17 depict the effects of turbulence on the BER for a direct-detection
binary optical detection link. The following parameters were chosen for generating
1.0E+OO
1.0E.Q1
, I I l zSOOO m. 0 .83 urn, 0 =10 em~ f
No T",buIon<.
IY\
l -500 0 m. O.83 urn, 0 -25 em
1.0E.Q2 - ;- "
1.0E-03 - c
l'l
...'"
0:
1.0E.Q4 -- - -- , 0 10cm
-
§ 1.0e.QS - \
- - - - - -
OJ
.;l
1.0e.Q6 ---- . -
\
III 1.0e .Q7 -- - ~ -~ - -
NO "TURBULENCE \ 0=-;5 em
1.0e.Qe
1.0e.Q9
1.0E-1 0
-- - I -I -\ -
-60.0 -50.0 -40.0 -30.0 -20.0 -10.0 0.0
R ece a-ed Power d Bm I
Fig. 16. BER vs. received powerfor L = 5 km, = 0.83 rn, and for two aperturediameters D =
10em, D = 25 em.
S lE-<13
e
0:
le -oc -
B , E-<l5
OJ
.;l
III 'E-<l6 - - - .--r..!"-
NO TURBULENCE
le ~7
l e-oe
' E-<l9 1---l----+--'--I---l---~-__1
-«100 ·50.00 ·30 00 -20.00 . 10 00 000
Received Pow e r t!B m )
Fig. 17. BER vs. received powerfor L = 5 km, = 1.55 rn, and for two aperture diameters D =
10em, D = 25 em.
the numerical results shown in the Figures 16 and 17: two laser wavelengths .A =
0.830JIm and .A = 1.55JIm, two aperture diameters D = 10 em and D = 25 em,
communication range = 5 krn, transmitter beam size Wo = 2.5 em, data rate =1.25
Obit/sec, efficiency 'f/= 0.8, the strength-of-turbulence parameter C~ = 1.2 X 10- 14
m- 2 / 3 (assumed constant over the horizontal propagation path), current gain G =
20, K eff = 0.03, load resistance RL = 50 ohm, temperature T = 300 K, and WOO: I
modulator extinction ratio (10 = 0.00 I). The aperture-averaged scintillation index a~, s
was calculated from the equations discussed in earlier section. The figures show the
BER versus the average optical power Po p t at the receiver. As seen previously for
uplink and downlink examples, the BER is again clearly increased by atmospheric
turbulence and by decreasing the aperture diameter. For example, at the wavelength of
0.830JIm, the BER increases from approximately 10- 9 (no turbulence value) to 10- 4
92 Arun K. Majumdar
for a 25-cm-diameter aperture, and even to about 10- 2 for a 1O-cm-aperturediameter.

These values are taken for the received powerof -30 dBm. For the wavelength of 1.55
jlm, compared to no turbulence BER of 10- 9 , the BER increases to approximately 5
x 10- 4 for a 25-cm-diameteraperture, and 1.5 x 10- 2 for a lO-cm-diameter aperture,
all the values at the received power of - 35 dBm. Atmospheric turbulence is therefore
the dominant cause for the system degradation affecting BER to increase severalorder
of magnitudes. The curves will be helpful for designing the communicationsystem to
operate under atmospheric turbulent conditions.
4.4. Example 4: Downlink PPM Laser Communication in Presence ofAtmospheric

Turbulence and Multiple Scattering Media
4.4.1. Theory and Formulation of BER for PPM with Stretched Pulses for
Pulse Position Modulation
In this example, numerical examples of degradation in the bit-error rate (BER) of a

multi-Gigabit, pulse position modulation (PPM) laser downlink due to turbulenceand
multiple scattering media is presented. Multiple scattering of optical wave occurs for
a variety of atmospheric conditions, such as cloud, haze, fog and rain. The greatest
degradations are scintillation which causes temporal fluctuations of the signal count at
the receiver, pulse stretching of the initially transmitted short pulse and random prop-
agation time delays. Severe turbulence and multiple scattering in the communication
channel may lead to unacceptable symbol error probabilities. The impulse responses
of both the turbulent and multiple scattering channels have been computed and their
effects on the data rate are developed. An example of a solid state laser transmitterand
a high-speed detector shows the engineering feasibility and discusses practicallimita-
tions and problems associated with a multi-Gigabitdata rate for downlinkatmospheric
laser communication channel.
Pulse position modulation (PPM) in atmospheric optical communication has been
used in the past. [25,27-29] The PPM's primary advantage is the ability to communi-
cate at a high data rate over noisy channels. For multi-Gigabit downlink, very short,
highly peaked laser pulses arc transmitted which encounter the followingeffects after
propagating through the atmosphere: pulse spreading (temporal broadening),scintilla-
tion and the random arrivaltimes. These will result in increasingBER for data encoded
in the PPM format in a space-to-ground downlink. The receiver on the ground has to
decode the pulsesby detectingthe correctslot positionin a given word. Figure 18 shows
a M 'ary PPM format where there are M different waveforms, each is represented by
a pulse in each of M time slots corresponding to distinct symbols. The stretched (and
distorted) pulse is shown in the slot q. In vacuum, the pulses should fall exactly within
each time slot without any overlap into other slots or frames. Becauseof the spread due
to atmospheric effects, an overlap into other time slots may thus produce inter-symbol
interference, and an overlap into other frames may produce inter-frame interference.
The data rate, R is given by
R = log2 M , (86)
MTs +Td
where Ts is the duration of any of the M time slots, Tdis the frame dead time after
the last slot. In order to estimate the correct slot location of the transmitted pulse and
correspondingsymbol, the receiverdetects a "greatest of' photoelectroncounts in each
T, _ _PPM FRAME /
LLLL",
SLOT
ft ",LU;J
q M-I M
r,
Fig. 18. Expected distorted receive pulse in the qth time slot of a PPM frame, showing the
expected slot pulse areas.
of the M time slots over this time Ts . When the pulse is stretched or distorted, the
counts in the qth slot will decrease and the counts in the adjacent slots will be increased .
This will cause the receiver to incorrectly decode transmitted symbols resulting in an
increase in BER.
We now evaluate the BER performance for such PPM lasercom system . The exact
procedure and the necessary equations are described in details elsewhere . [30,31] With
a matched filter at the receiver to be a pure slot integrator, the random sampled output
count for the j th slot is
(87)
(j -l)TS
where P(t) is the instantaneous temporal pulse shape of the received signal at the
detector output , and nj is the combined background and receiver noise count in this
slot for a particular sampling . Assuming the pulse is always at the qth time slot, a long
series of frames is sent. Neglecting inter-frame and edge effects at the first and Mth
slots, the result will be equivalent to randomly placing the pulses in individual frames .
Each slot will have some expected count that will depend on the expected pulse shape,
pulse amplitude and arrival time. If we denote these expected slot counts be
mj, j = 1,2, . . . .q, . . . ,M. (88)
When the turbulence and scattering effects are significant, these expected counts are
time varying, giving rise to conditional counting processes in each time slot. Thus
the expected counts mj will be random variables, in general. In particular frames, the
counts in individual slots will usually deviate from their expected values. A symbol
error will occur if
rj > r q , j = 1,2, ... ,M, j =1= q. (89)
The probability of symbol error (PSE) in a frame can then be expressed as
M
PSE = L Prob(r, > r q ) . (90)
Tj = I ,j#q
94 Arun K. Majumdar
Assuming the mean count is a random variation in all the slots and is continuous over
the range [0,00], the total probability of the count in a slot j exceeding the count in
slot q is
f f
00 00 00 rj-l
Prob(rj> r q) = P(m q) P(mj) L P(rj Imj) L P(rq Im q ) dmjdmq.

o 0 rj=1 rq=O
(91)
In the above equation, P( m q) and P( mj) are the probability densities of the expected
slot counts over the continuous range . Similar conditional Poisson count was used by
Churnside and Hill. [32] Substituting Eq. (??) into Eq. (??), we obtain
r ·-1
f f
M ~ ~
PSE = . L
J=1,J",q 0
P(m q)
0
P(mj) f=
rj=1
P(rj Imj) t
rq=O
P(r q Im q ) dmjdmq.
(92)
The BER can now be obtained from
BER = M log2 M PSE. (93)

2(M -1)
If the discrete counting process is a conditional Poisson, given by
rk
P(rk Imk) = m\
rk .
e- m k
k = 1,2, .. . , M. (94)
We can now write the expression for BER as

M
BER = M log2 M ""
2(M - 1) j=f::",q
Next we determine the functional forms of the mean count probability densities result-
ing from turbulence and multiple scattering effects and then evaluate communication
performance for optical downlink.
In order to evaluate the temporal spreading of the initially very short transmitted
pulse, we need to know the transfer function characteristics of the atmospheric channel.
This transfer function can actually be obtained from the convolution of G 1 (t - z I c)
with G 2 (t - zlc) where G 1 (t - zlc) corresponds to the high-frequency component
of the transfer function , and G 2 (t - zIe) corresponds to the low-frequency one. The
reason for the separation of the G function into these two components comes from
the fact that the output pulse can be expressed in terns of the two-frequency mutual
coherence function (MCF), F, given by [33]
(96)
The impulse response function of the random medium and the MCF is related by
f
oo
G(t - t') = 2~ r(Wd)e-jwd(t-ndwd, (97)

-~
where Wd represents the difference between the angular frequencies, WI and W2.
The expected instantaneous output power resulting from a given input pulse, Pi(t)
is
!
00
Po(t) = Pi(t')G(t - t')dt'. (98)

- 00
The expected impulse response can be expres sed as (for optical propagation along z
direction)
!
00
G(t - z/c) = GI(t - z/ c) ®G2(t - z/c) = GI(t - z tc - t')G2(t')dt' , (99)

- 00
where c is the velocity of light, and ® represents a convolution.

The first function is [31,33]
GI(t-z/c) = 4T
1r
I
L o (-1)
00 n { 21r2(t-Z /C)}
(2n+1)exp -(2n+1) 16 T
I
'
(100)
where TI = ( 1.~8J C~2/5 k~/5 L 11 / 5, with L being the propagation length with tur-
bulence , ko= 21r1>' is the wave number of the optical signal.
The second function to be convolved to get the impulse response is
1 {(t-z / C)2} (101)

G2(t - z/c ) = y1iT exp - T:j '
2
where
(102)
with L o = outer scale size of turbulence. The parameters T I and T 2 were calculated
for a given turbulence strength C~ . Thus GI and G 2 functions were evaluated. We
will discuss the impulse response function for multiple scattering media later in this
section.
4,4.2. BET-Error-Rate (BER) and Slot Count Statistics
The received pulse power over the effective aperture area of the telescope is a random
function which depends upon the impulse response G(t)and the background noise of
the receiver. Let the received pulse power per unit area be
P(t) = IS(t) , (103)
where I is a random intensity function and S(t) is defined as
!
00
S(t) = (]) Pi(t')G(t - t')dt' , (104)

- 00
where Pi(t') is the input pulse power, I is a parameter that depends on the full width
transmitter beam angle, Or, and the range, L, and is defined as
96 Arun K. Majumdar
(105)
(1) is the average or expected mean value of the intensity random process, 1. The
expected instantaneous count for the slot q, which contains the information signal will
then be
m q = aA r ee 7
(q-l)Ts
5
1 S(t) dt + nq , (106)
where a = il; is a constant, TJ is the detector quantum efficiency, h is the Planck's

constant, v is the optical frequency, A r ee is the receiver aperture area, and n q is the
combined background and receiver noise count at the qth slot. Following the procedure
as shown in [30], we can obtain the following expression for the pdf of mj (if we know
the pdffor P i with the noise mean pdf given by P nq(n q) as shown below):
(107)
The term A q can be expressed as [31]
Aq = (J) 7I 5
(q-l)Ts - 00
P;(t')G(t - t')dt'dt . (108)
After some mathematical manipulation the area A q can be simplified to
A = 'Y Pt t p (109)
q 2(1) '
where t p is the duration of the pulse with a constant power Pt within this pulse duration .
Assuming the expected noise count , n q, in each slot is constant, then the pdf, P n q (n q)
must be a delta function of the form e5( m q - n q ) , so that we can finally write
(110)
which implies that

m q - n q)
P(
m)q = PI ( A A . (III)
a ree q
4.4.3. The Impulse Response Function in the Case of Multiple Scattering
While the turbulent medium is considered to be a continuous random medium, multi-

ple scattering occurs for a discrete random medium whereas the medium consists of
scatterers of random sizes, shapes , orientations, velocities and densities. For multiple
scattering processes, the mutual coherence function can be written as
(112)
where Wo is the albedo of a scatterer, 7 = pnasZ is the optical distance for a path z,
and a s is the scattering cross section. Note that albedo is defined as Wo = ~ where at
° t
is the total cross section of the random medium which is equal to (a s + a a) where a a
is the absorption cross section of the randomly distributed particles . Since G 1 (t - z/ c)
r
depends on 1 (Wd) only, it remains same for multiple scattering . G 2 (t- z/ c), however,
depends on the exponential term of the above equation, so that we can write
1f {[I - we] }
00
G2 (t - z / C) -- - exp - --- 7 e- i Wd(t- Z/ C)dWd , (113)

27r Wo
- 00
G2(t - z/c) = J(t - z/ c) exp { _ [1 :;0] 7} (114)
for discrete random media involving multiple scattering. [33,34]

The impulse response function, G (t - z / c) applicable to multiple scattering media
can be finally written as
7r exp {[I-Wo]
G(t - z/ c) = 4T1 - ~ 7 }~oo
L.... (-1) n (2n + 1)
o
x exp { - (2n+ I) -
2 7r2 (t-2Z/C)} . (115)
16 T1
The expected pulse area will be different for this multiple scattering case and is given
by
Aq = 1 exp { _ [~;)tp
1
:;0] 7} . (116)
The BER is obtained as outlined in [31] and is given by

Mlog ~
BER = 2(M _2M
1) . L.... exp [-(aAreclav(Aj + Aq))J
J=l ,J"oq
where I av = I q = I j is the average constant power per unit area amplitude of the
intensity random process. The equation for BER was evaluated numerically and the
results are shown below.
4.4.4. Numerical Results

The Gs, G 2 , and G impulse response functions for both turbulence and multiple
scattering media and BER computation. Based on the theoretical background de-
scribed earlier, the parameters T 1 and T2 were calculated for both weak (C~ = 10- 18
m -2/ 3) and strong (C~ = 10- 14 m -2/ 3) turbulence cases. Using these values, the G 1
and G 2 functions were evaluated for both turbulence and multiple scattering media.
Figure 19 shows the functions G 1 , G 2 and G for strong turbulence case and Fig. 20
shows the function G I functions for various atmospheric conditions of dense rain,
light fog, fog and cloud . The corresponding T 1 values are also shown in the figures.
To calculate G 2 for multiple scattering media, we need to know the albedo Wo and
the optical depth 70f the scattering media whose values were assumed as follows: fog
(wo= 0.6,7= 4) and cloud (wo= 0.8,7= 20).
98 Arun K. Majumdar
g 1.71)::6
5
~ 1,703.
1,7[132
\ r' w,::"===",::-"",::-eu.E==:"'
. =':::'' i' :::,m:::,'':::" =
'":':'==1
G,::'
Fig. 19. The functions Cl. C2 and C for a strong turbulence case (en = 10- 7 m- 2 / 3 ) .
(a) (b)
Fig. 20. The C 1 functionsfor multiplescatteringmedia.The figures showfor variousatmospheric

conditionsof (a) dense rain and light fog. (b) fog and cloud.
4.4.5. BER Computation for Both Turbulence and Multiple Scattering Media
The following parameters were used for calculating BER for both turbulence and
multiple- scattering media: laser wavelength A = 1.06 /Lm, detector efficiency Tf =
0.2, effective receiver aperture area A r ec = 0.78554 m2 , T s = 167 ps for 2 Gbit/s data
rate, transmitter power Pi = I W, optical path transmissivity T a = 0.8, transmitter
optical efficiency TT = 0.9, receiver optical efficiency TR = 0.9, pointing loss TL =
I, transmit beam angle (It = 15 wad, range L = 3.6 X 107 m. The input Gaussian
to
.--
B T EFlIlQIlRATE TUFlBULEflK:E INPUrP ULSE E p~e< TE TURBULENCE INPUTPIJ..SE 9J ~.ec
SITERRORR....
.t
.a
.J
''-,
;,::~ T .,
slrong:/
Turbulence
-s Turbulence
", ,
", ' "
.' '~
f ts ,
., Weak
r uro urence
" Weal;.
Turbulence
" "
' ! " . 'f ,f
" II ,,~,
!"
rs D
,~ • ,if
TR.iJIlSMITP()Io\'ER iV',., I. ) TRANSMIT POWEI<(Wall.)
(a) (b)
Fig. 21. BER VS, required transmitter power for weak and strong turbulence cases and for two
input laserpulses: (a) 10psec, (b) 50 psec. For both turbulence cases, the parameters usedfor the
calculations are: Q = 1.0668 X 10 18 , A r ec = 0.7854, = 15 x 10- 6 rad, range L = 3.6 X 107 e
m.
BER: MULTIPLE SCATIERING, INPUT PULSE = 50psec
'~ "-., • •
·2
·4
·6
-8
IX
ur
IE. -10
o
i\ \
o
...J -12 \ \
·14
\ \
\
· 16
..- FOG:Wo - O.6 , tau = 4 I
· 18 ------ CLOUD: Wo = 0.8, tau = 20
2
10° 10' 10 10'
TRANSMITPOWER ('Natts)
Fig. 22. BER VS. requiredtransmitterpowerfor multiple scatteringmedia(fog and cloud) for an
input laser pulse of 50 psec. The parametersused for the calculationsare Q = 1.0668 X 1018 ,
A r ec = 0.7854, e = 15 x 10- 6 rad, L = 3.6 X 107 m.
pulse with FWHM = lOps and 50 ps for two turbulence cases and 50 ps for multiple
scattering media were assumed . Figure 21 shows the BER as a function of required
transmitter power for both weak and strong turbulence, and Fig. 22 shows the same
plots for fog and cloud using an input laser pulse of 50 ps. For the weak turbulence
case using a IO-pslaser pulse, the required laser transmitter power to achieve a BER of
10- 7 is approximately 3 W of laser power while for strong turbulence case the same
power will achieve much increased BER of 2 x 10- 3 . When a laser input pulse of 50
ps is used, it requires about 700 mW of laser power for weak turbulence case, whereas
with this power, a BER of 3 x 10- 4 can be achieved for strong turbulence case. For
multiple scattering media with an input laser pulse of 50 ps, it requires a laser power
100 Arun K. Majumdar
of approximately 12W for fog and 120 W for cloud to achieve the same BER of 10- 8 .
A high-data-rate lasercom system downlink with transmitting short laser pulse of 50
ps and reasonable laser power is thus feasible with the PPM scheme described in this
scheme.
5. Other Examples/Scenariosof Free-Space Optical and Laser

Communications
5.1. Ground to Space Shuttle Link
Researchers have presented results from experimental demonstration using a very

lightweight optical wavelength communication without a laser in space (LOWCAL)
[35] between ground based telescope and a Space Shuttle . This link established an up-
link/downlink carrier at 852-nm wavelength and 852-nm signal beam. Both the trans-
mitting laser and the return link receiver were located on the ground terminal, while the
optical elements located on the spacecraft were the liquid crystal retro-modulator and a
simple forward link receiver. Some of the specifications for the experiment were: range
=640 km, data rate = 10 kbps, telescope diameter =0.6 m, modular weight =2-4 kg,
and the retro-reflector area of 70-180 ern". The diode laser transmitter power was 0.2
W. The transmitted linearly polarized laser beam was directed through a quarter-wave
plate converting the beam into a circularly polarized beam. The beam is then directed
to the retro-reflector type modulator which directs the beam incident on the space-
craft directly back to the transmitting telescope. For downlink a differential circular
polarization keying (DCPK) format was utilized where the binary data was encoded
in the polarization of the beam. The uplink communication link used sub-carrier FSK
modulation . The downlink receiver achieved aSNRof 138, and a BER < 1.67 x 10- 7 ,
whereas the uplink receiver had SNR of 1000, and HER < 1.67 x 10- 7 . The system
('V
thus had unique features such as no laser in space required, all pointing was on the
ground and full-duplex operation with one laser beam.
5.2. UAV-to-Ground Lasercom Link
A group ofresearchers reported the design and development of acquisition, tracking and
pointing subsystem for Unmanned -Aerial- Vehicle (UAV) to ground free-space optical
communications link. [36]. The communication link was developed from a UAV to
stationary ground stations located at Wrightwood, California and Maui, Hawaii . The
range of the UAV was 50 km. The downlink laser transmitter wavelength was 1550
nm and had a power of 200mW for developing a 2.5 Gigabit per second data rate
communication for a BER of IE - 9. Analysis presented to demonstrate that the
system met the requirements of a jitter error of 18.6 wad and a bias error of 2.2 J1.rad .
5.2.1. Indoor Optical Communication
All of the design and performance characteristics described so far for long distance
apply equally to short distance communication systems with one obvious exception-
the effect of atmospheric loss is almost none at all for indoor systems. The link budget
for power is therefore determined almost entirely by the transmitter power, free space
loss, and receiver sensitivity. The basic operating principles of indoor point-to-point
systems are not different from the outdoor communication links, the design s are very
different to accommodate various requirements. For example, the optical source must
be eye safe, which generally means that it is an LED, which in tum limits the data
transfer capacity to typically a few Mbitls . On the other hand, indoor systems do not
have to face atmospheric effects as the outdoor systems require, so that they can be
produced very cost effectively. Finally, the indoor applications are basically confined
to short distance spans and hence are appropriate to optical wireless systems that utilize
wide diverging beams rather than the narrow beams of point-to-point systems . Each
"cell" created by an optical "telepoint" base station can be shared by many users. In
some cases, diffuse paths are utilized, where the beams, radiating over a wide angle,
are also allowed to reflect off walls, ceiling, floor, furniture, etc. The penalty of diffuse
transmission is a much reduced data rate capacity compared to line-of-sight systems .
The short-range indoor wireless communication is receiving increasing attention
because of its role in the emerging technologies involving portable computing and
multimedia terminals at work and everyday environment. Some ofthe typical portable
devices may include laptop computers, personal digital assistants, and portable tele-
phones, while the base stations are usually connected to a computer with other net-
worked connections. Many indoor communication systems employing infrared LED
wireless links have been reported . [37-39] Using pure diffuse links, a high-speed,
power-efficient indoor wireless infrared communication using code combining is re-
ported [40] where a multiple transmitter link design was used with a narrow field-
of-view direction diversity receiver. The design goal was to eliminate the effect of
inter-symbol interference (lSI) so that power-efficient signaling schemes such as PPM
can be employed at very high data rates. Recently, a portable transceiver for indoor
wireless link is reported. [41] The system employs a transmitter of eye-safe infrared
LEOs and a receiver of photodiode arrays with multi-channel trans-impedance-summer
architecture. A wide field of view (up to 55 deg half-angle) and high speed (up to 35
MHz) was attained to support different intensity modulation schemes . The received
signal achieved a HER of 10- 4 at a plane of 2 m away from the transmitter, even at a
point ± 50 deg off the transmitter's vertical axis. The bit rate of the transceiver was up
to 40 Mbitlsec in an indoor non-directed infrared wireless link to be extended to 100
Mbitls using LEOs with higher cut frequency. A receiver angle diversity design for
high-speed diffuse indoor wireles s communication is reported [42] where the effects
of multipath distortion was compensated but at the expense of an increa sed path loss.
A (6m x 3m x 3.4 m) upper illuminated room was used with a receiver placed I
m above the floor was used. Their receiver structure consisted of 25 non-overlapping
18-degree FOV receiver elements, each element having an effective collection area of
40 cm 2 , using I cm 2 detectors . With a 60 mW diffuse optical source, and utilizing
non-return-to-zero (NRZ) on-off keying modulation scheme at 50 Mbitls, the perfor-
mance achieved was HER of > 10- 4 for normal illumination and HER of < 10- 7 for
upper illumination across most of the coverage area.
5.2.2. Free-space Optical Interconnect
This paper will be incomplete without the mention of one other type of extremely
short range optical communication which uses free-space, such as data links between
chips, multi-modules and boards. Optics is becoming a mature technology for inter-
connects and free-space optical interconnects are seen as well-suited for the chip-level
and board-level interconnects. Several interconnect architectures based on arrays of
VCSELs have been proposed . Some of the problems to overcome include minimiza-
tion of optical channel crosstalk arising from the laser beam diffraction, and the mis-
alignment between VCSELs, microlenses and photo detectors . Chip-to-chip optical
communications through the air could boost data throughput between chips by a factor
of a thousand . [43] Optical communications will the computer, connecting one circuit
board to another, and chip-to-chip communications will the market. Free space op-
tical interconnects links operating at 2.5 Gbitls between VCSEL arrays and suitable
detector arrays have been demonstrated. [44] The researchers developed and demon-
strated 'the operation of a fully packaged optical interconnect system for multi-chip
interconnections capable of sustaining channel data rates as high as 800 Mbitls.
6. Multiple TransmittersIReceivers Approach for Lasercomm
The importance of using multiple transmitters and receivers was pointed out [45] many
years ago, with an example of laser communications between aircraft showing that the
system performance can be improved by using wide divergent beams. Wide divergent
beams can be used to eliminate the need for complex gimbaled pointing and tracking
mountings. For some very long range communication systems , the light from one
single source (LED or a laser diode) may not be enough to have the necessary link
margin to achieve the required BER. A large lens used in conjunction with a single
light source may result in a light beam that is too narrow to be practical in some cases.
The divergence angle may be so small that keeping the transmitted light aimed at
the distant receiver may not be practical. To launch more light at the distant receiver,
sometimes it is desirable to use multiple light sources. The multichannel approach was
chosen because it can increase the range of communications in presence of background
radiation more effectively than using a single-channel system with the same total FOV.
[45] The multichannel approach can also help in the alignment procedure in a situation
such as laser communications between aircraft where multiple LEDs were used, each
with its own lens. With multiple light sources and lenses, it is also easier to aim at
the distant receiver. For laser communication between a satellite and a ground-based
terminal , uplink scintillation is a serious issue. Fluctuations on the uplinked beacon
and the communication lasers can be minimized by transmitting multiple independent
lasers from separate apertures which then sum incoherently at the satellite . In order to
mimic the atmospheric effects for an uplink slant path to a satellite, experimental results
from a horizontal link (for distances 1.2 and 10.4 km) were reported . [46] 1,2,4,8, and
16 laser transmitters were located on an 18-inch-diameter circle at the two ranges of
1.2 and 10.4 km. For all laser separation and ranges, a reduction in received intensity
fluctuations due to turbulence were observed as the number of laser transmitters was
increased from I to 16. Figure 23 shows the histograms of the probability of intensity
curves vs. intensity for the two paths (integrated probability was normalized to the
value I and the mean intensity was taken as I). As the number of laser transmitters was
increased, the reduction in fluctuations was observed as can be seen from the decrea ses
in variances (FWHM) of the individual histograms.
Probability of R=xdcd 1Dlenolty: 1.21:m Kanse
18 inell Clrcul" Laser Spacing

2 inell Receive Ap<mue
-- ILascr
- -..-...-.
.... 2Lascn
4Lascn
._ ._ .- . 8Lasen
-- 16Lasen
1.5 2 2.5 4
1Dlenslty (normalized to mean)
(a)
Probability orRecorded Intenslty: 10.4 I<mRange
18 inch Circular Laser Spacing

2 inch ReceiveAperture
- - ILaser
- - - - 2Lasers
4 Lasers
. - . _ . _ . 8Lasers
- - 16Lasers
0.5 1.5 2 2.5

InlellSily (normaIlud 10 mean)
(b)
Fig. 23. Probability density function of intensity for varying number of laser transmitters (I , 2,
4,8 and 16 lasers), and for two ranges (a) 1.2 and (b) 10.4 km. ([46), reprinted with permission
from SPIE) .
The reduction of scintillation by the use of multiple uplink beams, each incoherent
with the rest, was demonstrated by a group of researchers. [47] The objective of the
experiment was to establish a l-Mbps optical communication link between a satellite
and a ground terminal. The link was established by first transmitting a beacon from
the ground station to the satellite and the scintillation effects were investigated by
launching multiple beams during uplink. Figure 24 shows the expected pdf when the
total laser power was equally distributed in I, 2,4,8, or 16 beams. With increasing
number of beams, the mean varied little, but the variance drops significantly with
additional beams. The method of dividing the laser beam into a number of beams will
help in the presence of strong scintillation to avoid deep fades and to improve BER
statistics.
0.020 r--.,.--r--.,.--r --.,.---,- -,-----r- --.----,
0.018
NUMBEROF BEAMS:
0.016 - - - 1
--- 2
- -4
0.014 -- 6
_ · - - -- - 16
0 012
0.010
o.ooa
0.006
0004
0.002
60 80 100 120 140 160

NORMALI2ED RECEIVED INTENSITY S
Fig.24. The probability density functions of the received signal intensity for varying number
of beams showing theimprovement in variance with multiple uplink beams. Thetotal power is
equally distributed among I, 2,4, 8, or 16 beams and hence the PDFs are normalized bythe total
power.«(47l, reprinted withpermission fromlPL).
The advantageof using multiplereceiverswasalreadypointedout earlierto demon-

strate the optical communications concept between aircraft. [45] The advantage of
using a number of detectors each having small FOVs,and each with very narrowband
optical filters effectively reduce the overall background noise as compared to using
a single large FOV with a single detector. Thus multiple detectors can increase the
received SNR. A theoretical analysis of an optical array receiver also showed [48]
that an array receiverconcept was a viable concept alternativeto using a single large-
aperture to collect the optical signal energy for high data rate applications. In presence
of atmospheric turbulence and background interference, the multi-element telescope
arrays observing pulse position modulatedoptical signals will be very desirable.
7. Conclusion
In this paper, we have provided the basics of free-space laser/optical communication

system performancewith a comprehensive treatmentof the theoreticalanalysis needed
for system design. We have investigated various communication scenarios of practi-
cal importance: horizontal (terrestrial) link and slant paths including downlink (from
satellite-to-ground), and uplink (ground-to-satellite). This chapter discusses with de-
tailed theoretical analysis the basic parameters for designing a laser communication
system that has to operate under variousatmosphericconditions,such as fog and turbu-
lence. Numericalexamples for variouscommunication links under variousconditions
are given to illustrate the trends in how design parameters affect performance and to
demonstrate how system design optimization is accomplished.
This paper has not considered adaptive optics for atmosphericturbulence mitiga-
tion or coding schemes, which can also significantly enhance system performance in
many cases, but are beyond the scope of this chapter. We have also not covered the
areas of heterodyne detection schemes and pointing and tracking.
It is obviousthatthe laser/optical communication is heretostay. Withour increased

knowledge of atmospheric effectsand the physicsof turbulence generation and inter-
actions with optical waves, and the continuous push for developing more sensitive
receivers, lasercommunication systemperformance willcontinueto improve, with in-
creasingly higher data rates, accuracy, and availability under all-weatheratmospheric
conditions.
Acknowledgments
The author wishes to thank Jennifer C. Ricklin for providing data for bit-error-rate
calculations for Gaussian beamwave. The author wouldlike to acknowledge LarryC.
Andrews, RonaldL. PhillipsandRobertK.Tyson for veryhelpful technical discussions
to clarify uplink-downlink theoretical equations. The author is deeply indebted to
FotiosP. Kourouniotis for assistance in thecomputation andsimulation of data. Helpful
discussions with Laurence Chen and a critical reading of the manuscript by Bradley
Bobbs are also acknowledged.
References
I. O.K. Killinger, J.H. Chumside, and L.S. Rothman, Atmospheric Optics, OSA Handbook of
Optics, edited by M. Bass (1995), Chapter 44.
2. EJ. McCartney, Optics of the Atmosphere (Wiley, New York, 1969).
3. 1.1. Kim, B. McArthur, and E. Korevaar, Comparison of laser beam propagation at 785 nm
and 1550 nm in fog and haze for optical wireless communications, Proc. SPIE, Vol. 4214,
p. 26-37, Optical Wireless Communications III, edited by Eric J. Korevaar (2000).
4. Brian R. Strickland, Michael J. Lavan, Eric Woodbridge, and Victor Chan, Effects of fog
on the bit-error rate of a free-space laser communication system , Appl. Opt. , 38, 424-431
(1999).
5. S.G. Lambert and w.L. Casey, Laser Communications in Space (Artech, Boston, 1995).
6. R.H. Kingston, Optical Sources, detectors, and Systems : Fundamentals and Applications
(Academic, San Diego, 1995).
7. W.S. Ross, W.P. Jaeger, J. Nakai, T.T. Nguyen and J.H. Shapiro, Atmospheric Optical
Propagation-an Integrated Approach, Opt. Eng., 21, 775 (1982).
8. A.K. Majumdar, Laboratory-Simulation Experiment for Optical Communication through
Low-Visibility Atmosphere using a diode laser, IEEE J. Quantum Electron ., QE-20, 919
(1984).
9. R.W. Svorec, Parametric Performance Analysis of Spaceborne Laser Communication Sys-
tems, Proc. SPIE, Vol. 295, 66 (1981).
10. A. K. Majumdar, Optical Communication between aircraft in low-visibility atmosphere using
a diode laser, Appl. Opt., 24 (21), 3659-3665 (1985).
II. Alan J. MacGovern, DavidA. Nahrstedt, and Michael M. Johnson, Atmospheric propagation
for tactical directed energy application, Proc . SPIE, Vol. 4034, pp. 128-139, Laser Weapons
Technology, Todd D. Steiner, Paul H. Merritt, Eds. (2000).
12. V.1. Tatarskii, The Effects of the Turbulent Atmosphere on Wave Propagation (translated
from Russian by IPST) (Available from the U.S. Dept. of Commerce, NTIS, Springfield,
VA 22151),1971.
13. Larry e. Andrews, Ronald L. Phillips, and Cynthia Y. Hopen, Laser Beam Scintillation with
Applications, SPIE Press, Bellingham, Washington (2001) .
14. Jennifer e. Ricklin, Stephane Bucaille, and Frederic M. Davidson, Performance loss factors
for optical communication through clear air turbulence , Proc. SPIE, 5160,1-12 (2003).
15. J.e. Ricklin and F. M. Dadidson, Atmospheric turbulence effects on a partially coherent
Gaussian beam: implications for free-space laser communications, J. Opt. Soc. Am. A,
19(9), 1794-1803 (2002) .
16. Arun K. Majumdar, Higher-order skewness and excess coefficients of some probability
distributions applicable to optical propagation phenomena, J. Opt. Soc. Arn., 69 (I), 199-
202 (1979) .
17. Arun K. Majumdar, Uniqueness of statistics derived from moments of irradiance fluctuations
in atmospheric optical propagation, Opt. Commun ., 50 (1),1-7 (1984) .
18. Arun K. Majumdar, Higher-order statistics of laser-irradiance fluctuations due to turbulence,
1. Opt. Soc. Am. A, 1, 1067-1074 (1984) .
19. Arun K. Majumdar and Hideya Garno, Statistical measurements of irradiance fluctuations
of a multipass laser beam propagated through laboratory-simulated atmospheric turbulence ,
Appl. Opt ., 21 (12), 2229-2235 (1982) .
20. Hideya Gamo and Arun K. Majumdar, Atmospheric turbulence chamber for optical trans-
mission experiment: characterization by thermal method, Appl. Opt., 17 (23), 3755-3761
(1978).
21. Arun K. Majumdar, John A. DiUbaldo, and Alenka Brown-VanHoozer, Laboratory Simu-
lation of Atmospheric Turbulence for Laser propagation : design and characterization, Proc.
SPIE. Vol. 3432 (1998).
22. M.A. Al-Habash, L'C , Andrews and R. L. Phillips, Mathematical model for the irradiance
probability density function of a laser beam propagating through turbulent media, Optical
Eng., 40 (8), 1554-1562 (2001) .
23. M. Nakagami, The m distribution-a general formula of intensity distribution of rapid
fading, in Statistical Methods in Radio Wave Propagation, edited by WC. Hoffman, 3-36
(Pergamon , New York, 1960).
24. Robert K. Tyson, Bit-error-rate for free-space adaptive optics laser communication, J. Opt.
Soc. Am. A, 19 (40), 753-758 (2002) .
25. Sherman Karp, Robert M. Gagliardi, Steven E. Moran and Larry B. Stotts, Optical Channels
(Plenum Press, New York, 1988).
26. H.T. Yura and W.G. McKinley, Optical Scintillation Statistics for IR ground-to-space laser
communic ation systems, Appl. Opt., 22 (21), 3353-3358 (1983).
27. H.G. Salfren, Effect of atmospheric turbulence on the bit error probability of a space to
ground neaf infrared laser communications link using binary pulse position modulation and
an avalanche photodiode detector, NASA Technical Memorandum 100699, Scientific and
Technical Information Branch, 1987.
28. R.M. Gagliardi and S. Karp, M -ary Poisson detection and optical communication, IEEE
Trans . Commun . Tech., COM·17, 208-216 (1969).
29. Robert M. Gagliardi and Sherman Karp, Optical Communi cations (Robert E. Krieger Pub-
lishing Company, Malabar, Florida, 1988).
30. Arun K. Majumdar and William C. Brown, Atmospheric Turbulence effecys on the Per-
formance of Multi-Gigabit Downlink PPM Laser Communications, Proc. SPIE, Vol. 1218,
568-584, Free-Space Laser Communication Technologies II (1990) .
31. Fotios Panagiotis Kourouniotis , Atmospheric Turbulence and Multiple Scattering effects on
the Bit Error Rate of the Optical Receiver for PPM MultiGigabit Laser Communications,
Master of Science Thesis, University of Colorado (1991) .
32. J.H. Churnside and R. J. Hill, Probability density of irradiance scintillations for strong
path-integrated refractive turbulence , J. Opt. Soc. Am. A., 4 (4), 727-733 (1987) .
33. Akira Ishimaru, Wave Propagation and Scatteringin RandomMedia, Vol.2 (Academic Press,
1978).
34. Shin Tsy Hong, I. Sreenivasiah, and Akira. Ishimaru, Plane Wave Pulse Propagatioin
Through Random Media, IEEE Trans. Antennas Propagation, AP-25, 6 (1997) .
35. T.M. Shay, D.A. Hazzard, J. MacCannell, G. Lee, C.D. Garrett, J. A. Payne, N. Dahlstrom
and S. Horan, First Experimental Demonstration of Full-Duplex optical Communications
on a Single Laser Beam, AIAA 15th Annual AIAAlUtah State University Small Satellite
Conference, August 13-16, 200 I.
36. TGerardo G. Ortiz , Shinhak Lee, Steve Monacos, Malcom Wright and Abhijit Biswas, De-
sign and development of a robust ATP subsystem for the Altair UAV-to-Ground Lasercomm
2.5 Gbps Demonstration, In Free-Space Laser Communication Technologies XV, G. Stephen
Mecherle, Editor, Proc. SPIE, Vol. 4975 (2003).
37. J.M. Kahn and J. R. Barry, Wireless infrared communications, Proc. IEEE . 85 (3), 265-298
(1997) .
38. G.W. Marsh and J. M. Kahn, Performance evaluation of experimental 50-Mbpss diffuse
infrared wireless link using on-off keying with decision-feedback equalization, IEEE Trans.
Commun .,44 (II), 1496-1504 (1996) .
39. A.P. Tang and J. M. Kahn, Wireless infared communication links using multi-beam trans-
mitter and imaging receivers, Proc. IEEE IntI. Conf. Commun ., pp. 180--186 (1996).
40. Koorosh Akhavan, Mohsen Kavehrad and Svetla Jivkova, High-speed power-efficient indoor
wireless infrared communication using code combining-Part I, IEEE Trans. Commun., 50
(7),1098-1109 (2002).
41. Lijun Jiang, George Chung Kit Chen , Ye Yang, New Wei Lee, Allen Yeo and Hongga
Li, Light-emitting-diode-based eide-field-of-view transceiver for indoor optical infrared
wireless communication, Opt. Eng., 43 (4), 918-923 (2004) .
42. Soo Hee Khoo, Wenwei Zhang, Grahame E. Faulkner, Dominic C. O'Brien and David J.
Edwards , Receiver angle diversity design for high-speed diffuse indoor wireless commu-
nications, Proc. SPIE, Vol. 4530 , I 16-124, Optical Wireless Communication IV, edited by
Eric J. Korevaar (200 I).
43. Neil Savage, Linking with light, IEEE Spectrum, 39 (8), 32-35 (2002) .
44. Sadik Esener and Philippe Marchand, Present and future needs of free-space optical inter-
connects, internet website : http://ipdps .eece.unm.edu/2oo0/wocsIl8ooI109.pdf
45. Arun K. Majumdar and George H. Fortescue, Wide-beam atmospheric optical communica-
tion for aircraft application using semiconductor diodes , Appl. Opt ., 22 (16), 2495-2504
(1983).
46. Isaac I. Kim, Harel Hakakha, Prasanna Adhikari, Eric Korevaar and Arun K. Majumdar,
Scintillation reduction using multiple transmitters , in Free-Space Laser Communications
Technologies IX, Proc. SPIE, Vol. 2990, 1997.
47. M. Jeganathan, K.E. Wilson, and J.R. Lesh, Preliminary analysis of fluctuations in the
received Uplink-Beacon-Power data obtained from the GOLD experiments, JPL's TDA
ProgressReport 42-124,20-32 (1996).
48. V.Vilnrotter, C.w. La,M. Srinivasan, R. Mukai, andK.Andrews, Anopticalarrayreceiver for
deep-spacecommunication through atmospheric turbulence, JPL Publication: IPNProgress
Report, Report 42-154, 1-21 (2003).
© 2007 SpringerScience + Business MediaInc.
001 : 1O.1007/sI0297-006-0079-z
Laser communication transmitter and receiver design
David 0 , Caplan
MIT Lincoln Laboratory, Lexington, Massachusetts
Abstract. Free-space laser communication systems have the potential to provide flex-
ible, high-speed connectivity suitable for long-haul intersatellite and deep-space links.
For these applications, power-efficient transmitter and receiver designs are essential
for cost-effective implementation. State-of-the-art des igns can leverage many of the
recent advance s in optical communication technologies that have led to global wide-
band fiber-optic networks with multiple Tbitls capacities. While spectral efficiency has
long been a key design parameter in the telecommunications industry, the many THz
of excess channel bandwidth in the optical regime can be used to improve receiver
sensitivities where photon efficiency is a design driver. Furthermore, the combination
of excess bandwidth and average-power-lim ited opt ical transmitt ers has led to a new
paradigm in transmitter and receiver design that can extend optimi zed performanc e of
a single receiver to accommodate multipl e data rates.
This paper discusses state-of-the-art optical transmitter and receiver designs that are
particul arly well suited for average-power-limited photon-starved links where chan-
nel bandwidth is readily available . For comp arison, relatively simple direct-detection
systems used in short terrestrial or fiber optic links are discu ssed, but emph asis is
placed on mature high-p erformance photon-efficient systems and commercially avail-
able technologies suitable for operation in space. The fundamental characteristics of
optical sources, modulators, amplifiers, detectors, and associated noise sources are re-
viewed along with some of the unique properties that distingui sh laser communication
systems and comp onent s from their RF count erparts. Also addressed is the interplay
between modulation format , transmitter waveform, and receiver design , as well as
practical tradeoffs and implementation considerations that arise from using various
technolog ies.
1. Introduction
Optical communications has provided unprecedented capacity in modem networks .

The rapid growth of the Internet has led to investment in wide-band fiber-optic net-
works that now span the planet. The optical communications revolution has also led
110 David O. Caplan
to the development of new and enabling technologies that can be applied outside of
conventional telecom applications. Free-space laser communications, also known as
lasercom, is one area that can benefit from these recent advances . Free-space optical
networks have the potential to surround the planet with flexible and agile wide-band
lasercom that could extend to the moon, Mars, [1,2] and beyond-capable of bridg-
ing billion kilometer links at Megabit/sec (Mbps) data rates. Suchcapabilities require
photon-efficient transmitters and receivers, the main thrust of this paper, which can ex-
tend unrepeatered links over ultra-long-haul distances exceeding the net fiber deployed
on Earth today.
Free-space laser communication performance directly depends on the efficiency
and sensitivity of optical transmitters and receivers. However, until recently, these fac-
tors were not the driving factors in the buildup of fiber-optic networks. But as the appar-
ent demand for bandwidth has approached the limitations of deployed fiber-optic links,
more sensitive receivers have become a means of improving network performance in
terms of power and bandwidth efficiency. Consequently fiber- and free-space-based
technology requirements have started to converge. High-sensitivity photon-efficient
transmitter and receiver designs can reduce mid-span amplifier requirement s, dimin-
ish nonlinear impairments, and extend link distances in fiber networks, but they are
especially beneficial for free space optical (FSO) communication, since improvements
in receiver sensitivity directly reduce transmitted power requirements which in tum ,
can lead to significant size, weight, power (SWAP) and cost reduction s.
The availability of modular high-reliability high-performance commercial-off-the-
shelf (COTS) technologies already developed for and widely used by the telecommuni-
cation industry is of great value to the development of future free-space laser commu-
nication systems [3]. Not only can the tremendous telecom investment in technology
development be leveraged in many cases, but the field-proven heritage of such tech-
nologies can provide increa sed confidence in reliability and life-time estimates, as well
as the component manufacturing process. This knowledge can be used to accelerate
the process of integrating cutting-edge technologies into reliable designs, and reduce
the expense of starting costly qualification programs from scratch . This is especially
important in space-based systems, where repair is impractical, and both reliability
and performance are critical design drivers. The value added from heritage telecom
technologies which require TeIcordia (formerly known as Bellcore) qualification , is of
additional value since these standards (e.g., [4-9]) often test to mechanical and ther-
mal levels that are similar to many of the environmental requirements for space-based
platforms .
Free-space optical links have some notable advantages over both radio-frequency
(RF) and fiber-optic links that allow greater flexibility in transmitter and receiver de-
sign and optimization. These include the absence of channel dispersion and nonlineari -
ties, and virtually unlimited channel bandwidth, especially in space-based applications
where atmospheric absorption is nonexistent. Such characteri stics allow FSO transmit-
ter (TX) and receiver (RX) designs to leverage the average-power-limited properties
of optical transmitters to generate optimized signaling waveforms [10-15], which,
when used in combination with photon-efficient modulation formats, can contribute
substantially to overall link efficiency, 1Jlink. A practical measure of 1Jlin k is energy
required per bit-received, [Jlbit].
Lasercommunication transmitterand receiverdesign III
/./. Background
The theories that govern both RF and optical communications over a noisy channel are
well established, dating back to 1949 to the work of Claude Shannon [16,17] in which he
established the channel capacity theorem which states that error-free communications
are possible up to rate C (bits/s) over a channel of bandwidth B (Hz) with a signal of
average power, P (W) perturbed by additive white Gaussian noise (AWGN) of power
spectral density of N o/2 (W/Hz) . The capacity of such a link is given in Eq. (I), which
highlight s the interplay between the three key system parameters: bandwidth, average
signal power, and noise power spectral density [17-19]:
C = Blog2(1 + SNR) = B log2 (1 + ::B) = B lOg2 (1 + ~:) [bits/s].

(I)
The SNR is simply the ratio of the average signal power (P), to noise power,
(NoB). For a data rate R b (bits/s), the average signal power is RbEb, where Eb is
the energy-per-bit (J/bit). Substituting spectral efficiency r = Rb/ B [(bit/s)/Hz] , we
obtain the right-hand side of Eq. (I), where Eb/No is the energy-per-bit to noise
power spectral density ratio. For shot-noise-limited optical signals the Eb/ No term
can be replaced by the average number of signal photons-per-bit (PPB) in most cases
(see section 2.3) and corresponds to the SNR over the observation window, with the
exception of coherent-homodyne RXs, where the quantum-limited SNR =PPB/2 [20].
The Shannon Limit shown in Fig. I, achieved when Rb = C, is the optimum tradeoff
between spectral-efficiency, r, and photon-efficiency. In the limiting case of optimum
bandwidth expansion, with modulation and coding, Shannon-limited sensitivity of
0.7 PPB (or ,,-,0.35 PPB for homodyne) corresponding to a -1.6-dB SNR can be
achieved for the AWGN channel [17,18,21-23] . A detailed discussion of Shannon-
limited capacities for optical communication systems is given in [23].
Also highlighted in Fig. I is uncoded performance at the 10- 9 bit-error-rate
(BER) and Shannon-limited capacities for on-off-keying (OOK), differential-phase-
shift-keying (DPSK), and M -ary orthogonal-keying modulation formats such as M-
ary pulse-position-modulation (M -PPM) and frequency- shift-keying (M -FSK) with
hard-decision coding . DPSK has received significant interest from the FSO commu-
nity and the telecom industry since it is both energy-efficient and spectrally-efficient.
For a quantum-limited optically preamplified DPSK receiver with optimal coding,
Shannon-limited performance approaches 3 PPB with ,,-,0.5 bit/s/Hz efficiency. High-
rate optically-preamplified receivers using 24.6% and 7% low-overhead forward error
correction (FEe) with 0.8 and 0.935 bits/s/Hz efficiencies have demonstrated 7 and 9
PPB receiver sensitivities at 10 and 40 Gbit/s data rates, respectively [24,25].
While spectral-efficiency has long been a key design parameter in the telecom
industry, in many optical communication links excess channel bandwidth is available,
and can be used to improve performance where photon efficiency is the design driver.
In such links, the available bandwidth can be used to improve receiver sensitivities
by ,,-,5 dB by using, for example , coded M-ary orthogonal modulation formats such
as M -PPM. As shown in Fig. I, the quantum-limited sensitivity (pre-amplified, direct
detection) for uncoded I024-PPM in which each symbol carries 10 bits of information ,
is "-'6 dB-PPB (",4 PPB), with IOOx(M / lOg2 M) bandwidth expansion. With the
addition of "-'50% overhead (OH) hard-decision FEC, '" 1.5 PPB sensitivity can be
112 DavidO. Caplan
U;.ce:I.~J: ~~:
f 1~9.B ~ ~ : ~ a
I :; .... I
4- Uncod..d2..PPM
=~i~n;o~~:::~~~~s' on
- OOK, AWGN, Hard Oeclslon
· U~.,.m~~,d.;·. j.f.:~S. f,],.1

""
~ J::O~d~W~J . ~:~~~: ::~:: ~:;: ~:~::::::
• 256· P PM, AWGN, Hard Decision
~
@ io" BER : : : :: • 10U.P PM, AWGN, Hard Decision
.•...~ ..:_.; .;
Z
!!?.
10 .;.
~ !!1 i "l-ill
. ; ; [ ;...; .;.:,;.
1
_._ -
~ j ~ ••••• - - -
M· PPM Opflmum. AWGN , Hard Decision
M· PPM Optimum. AWGN. 50 ft Decision
:·, - ,f~~~=::::~'~::~_~~:Eti~I
;;;
~ 5 .
l:
o
'0
J:
,',!'
11. o ., ......;.....;...;.., ;..;.::
-1.6 .:, . ' '::'. .. . . . . .... . . . : ' ::: I
······ ..T Stlaiinoii lImi((oi AWGNChaiin. , ; : :;' . . ... " " " ;, .,: I
-5 . .. , , i iii .
10" 10° 10' 10' 10'
Band width Expan sion Factor, [(bit/sI/Hzr '
Fig. 1. The tradeoffbetweensensitivity (photon-efficiency)as measured in photonsper bit, and

bandwidth expansion (spectral-efficiency) for OOK, DPSK, and M -ary orthogonal, as wellas the
ultimateShannon limit for AWGN systems [22). Reprinted with permission of IEEE. (@2006
IEEE.)
achieved, and this can be extended to nearly I PPB by implementing optimal soft-
decision decoding [22).
For the Poisson channel dominated by quantum noise, the Shannon-limited sen-
sitivity per-photon can be made arbitrarily low, with a lower bound given by the ratio
of thermal-noise-photon energy to signal-photon energy [26-28]:
B kBTln(2)
(2)
Cph Quantum-limit = hu [photons/bit] ,
where B is the receiver bandwidth in Hz, C p h is the capacity in bits/sec per photon , kB
is Boltzmann's constant (1.38 x 10- 23 J/K), h is Planck's constant (6.63 x 10- 34 J
s), and T is temperature in Kelvin (K). For the case of 1.5 j,tm signal photons at 300K,
this yields a limit of 0.022 PPB (-16.6 dB SNR) or 46 bitslphoton [27,28], which in
principle, can be further improved by reducing the receivertemperature to 3K. Although
this is not practically realizable, the promise of such capabilities has made the research
and development of efficient wide-band photon-counting detector technologies an area
of active research [1,2,29-32], especially since Gigabit-per-second (Gbps) class high-
sensitivity communications employing super-conducting photon-counting detectors
have recently been demonstrated [33-35].
A block diagram of the communication link is shown in Fig. 2, and the focus of
this paper will be to optimize overall link performance (1]link) through the design of
elements within the transmitter and receiver subsystems.
Functionally the free-space optical transmitter typically includes a laser source, a
modulator that can impart a variety of modulation formats, a desired bit rate, coding,
and a high power amplifier such as an Erbium-doped fiber amplifier (EDFA) with
I Receiver
TJRx
r
o";;ta
Fig. 2. Block diagram of the optical communications link, with the input being the data and the
output being anestimate of the input data. Overall link performance 7]link = 7]TX 7]ch 7]RX .
an electrical to optical conversion efficiency which is often the dominant component

of the net transmitter efficiency (1}TX) . For transmission through the channel (1}ch),
we can include transmitteroptical (telescope) transmission efficiency (1}Txopt), free
space transmission efficiency (1}space), which is reduced by diffraction losses, atmo-
sphericabsorption,scattering,turbulence and scintillation, and interferencefrom back-
ground radiance. e.g., loss and additivenoisecollectively (1-1}atm). Also includedare
pointing, acquisition and trackingefficiency (1}PAT), and lastly the receivercoupling
efficiency (1}RXOpt). The net receiverefficiency (1}RX) includes the quantum-limited
number of photons-per-bit to achieve the desired bit-error-rate (PPBQL). associated
implementation penalty (1}QL), and potentialcoding gain (Geode) from forward error
correction (FEC). These terms are dependent on many factors such as the detection
scheme, which may be direct, optically preamplified, homodyne or heterodynedetec-
tion, and the demodulation approach, which can be either incoherent or coherent and
can occur in either the optical or electrical domain.
High-rateopticalcommunication systemsgenerallyutilizePIN photodiodes which
are commercially availablewith electrical bandwidths exceeding50 GHz. Avalanche
photodiodes are detectorswith internalgain thatcan be used to improveRX sensitivity,
although this generallycomes at the expense of bandwidthand limiteddynamic range
of operation.A decodercircuit maybeemployedto processreceived FECencodeddata,
and the final figure of meritisthe BER,whichisdeterminedbycomparingthe inputdata
(D) to the output estimate (D) . For systems which use FEC, the BER is often readily
available and can be used as a feedback parameter to improve overall performance.
Typically receiver sensitivities are measured in terms of the power received or the
number of photons-per-bit required to achieve a particular BER, typically 10- 9 (I
error per billion bits received), but the target can change dependingon the application
and system requirement.
Optical receiver sensitivities are ultimately limited by choice of modulation for-
mat, coding, and fundamental quantum fluctuations. This quantum noise is commonly
referred to as "shot noise", whichcomes from the randomness in photon arrivals, and
imposes theoretical limits on receiverperformance. For high-rate high-sensitivity op-
tical communication systems, optically preamplified receivers are the most practical.
being widely used in the telecom industry and demonstrating the best performance at
Gbitls rates and beyond (see sections 4 and 5).
The theoretical (quantum limited) sensitivity is dependent on the modulation and
receivertype and is often measured in termsof the numberof PPB requiredto achieve
a 10- 9 BER. For optically-preamplified binary intensity modulation (1M) formats
such as OOKand binary-PPM, the quantum limitedsensitivity is ",,40 photons-per-bit
[20,36-39]. More complex orthogonal signaling formats such as M -PPM or M -FSK
114 David O. Caplan
?9
10
__ - - - - - --:;:;:::;'1 100 ~
90 £
···· "fog;( M)"······· ·· ···~.~ ::::::::::::::::::::: ::::::::: ::::::::::::: o
'-0E
;r; 6
8
7
.. ]
»:
........ .. ....... .. ........ ... . ...... .. . ........ .. . .... .
. .
80 ~
70
60 .~
g
g,
i ~ .{ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .;,I1~~,(r;).·••·• • • • • ·
50
40 ~
30 ;S
"C
20 .~
I - :-k = Blls /Symbol
1 . ,.... ...... Bandwldlh Expans ion Faclor 10 -g
o o ~
o 200 400 600 800 1000
Constell ation Size, (M)
Fig.3. Bits persymbol (left axis) and bandwidth expansion factor (right axis) for M -ary orthog-
onal signaling.
can be used to further improvethe theoretical sensitivity at the expenseof bandwidth,

which degradesspectral efficiency by a factorof (M/ 10g2 M) as shown in Fig. 3.
For example256-PPM improves receiversensitivity by about a factorof 6, to ",,6
PPB, but requires a factor of 32 expansion in bandwidth (see also Fig. I). For high-
rate bandwidth-limited applications, optically-preamplified Differential-Phase-Shift-
Keying (DPSK) has demonstrated the best sensitivity, providing a ",,3dB uncoded
benefitover OOK and 2-PPM,corresponding to a theoretical sensitivity of ",,20PPB,
although this comes at the expense of a more complicated receiverdesign.
Forward error correction coding is another powerful tool available to the sys-
tem designer that is often a cost-effective means of improving receiver sensitivity
[17,18,40]. FEe effectively operates by converting the HER of an input signal which
contains redundant code bits to an improved output HER. As illustrated in Fig. I,
coding is needed in order to extend uncoded-quantum-limited performance to the ul-
timate Shannon-limited sensitivities. The use of FEC reduces spectral efficiency and
increasesTX and RX electroniccomplexity, requiring the additional encodingand de-
coding hardware, butcan providesignificant codinggain withrelatively littleoverhead.
For example, with commonly used enhanced Reed-Solomon 255/239 FEC coding, a
",,6dB sensitivity improvement is achieved by converting a ",,3 x 10- 3 input BER to a
10-9 output BERwithonly7% overhead, i.e.,additional bandwidth expansion relative
to the rate. Givenan opticallypreamplified DPSK receiverwith near-quantum-limited
performance this could improvesensitivities to ",,6PPB, within ",,3 dB of the AWGN
Shannon limit for DPSK.The benefits of such low-complexity FEC codes have been
demonstrated at 10 [41] and 40 Gbps [25] data rates. They have also become com-
monly used in the telecom industryas a means of diminishing nonlinearimpairments
and increasingoverall system throughput.
More powerful turbocodes with 100% overhead (rate ~ ) havebeendevelopedthat
can closely approach Shannon-limited performance [42], however the complexity of
implementing such serially concatenated codes has presently limited real-time appli-
cation to low data rates < ",,50 Mbps [2,43-45]. But with lessoverhead(25%,rate 0.8)
and complexity, block turbo codes havedemonstrated >- IadB coding gain at rates up
to 10 Gbps [24]. For the AWGN channel, with high-constellation orthogonal modu-
lation formats such as M-PPM with optimal soft-decision decoding, '" I photonlbit
sensitivity can be achieved in principle[22].
1.2. Scope
This paper will expand upon the topics introduced above, providing an overview of
state-of-the-artand futureoptical transmitterand receiverdesigns that are particularly
wellsuited for FSOcommunication. Forcomparison, relatively simpledirectdetection
systems used in short terrestrial or fiber optic links will be discussed, but emphasis
will be placedon maturehigh-performance photon-efficient systemsand technologies
suitable for operation in deep space optical links.
While overallsystem level performance is dependent on many factors such as TX
and RX aperture size, wavefront quality, and pointing acquisition and tracking (PAT)
[3,46--48] this section will focuson practicaldesign optionsfor the TX and RX blocks
of the communication link shown in Fig. 2. The remaining link elements-the cou-
plingopticsandchannelconsiderations are discussedin detailin"Atmospheric channel
effects on free-space laser communication", by Ricklin et al. (DOl 10.1007/sI0297-
005-oo56-y) and "Free-space laser communication performance in the atmospheric
channel", by Arun K. Majumdar (001 1O.1007/sI0297-oo5-0054-0) in this publica-
tion.
The fundamental characteristics of optical sources, modulators, amplifiers, detec-
tors, and associated noise sources will be discussed along with some of the unique
properties that distinguish laser communication systems and components from their
RF counterparts. Practical tradeoffs and implementation issues that arise from using
various technologies, and the interplay between modulation format, transmitterwave-
form, and receiverdesign and optimization will be presented.
This paper is intended to complementthe rich subject of transmitterand receiver
design, by highlighting practicaldesign considerations and recentdevelopments in the
state-of-the-art high-performance FSO systems. While introductory and background
material is included for clarity, the reader is directed towards the numerous papers,
patents, and texts referenced at the end of this paper for a broader view of this subject.
1.3. Historical Perspective
Thereare manyexamplesof free-space opticalcommunications rangingfromthe useof

mirrorsand sunlight by the ancient Greeks, use of fire beacons by the Chinese (",800
BC) and later by Romans, and smoke signals by American Indians. Other notable
examples include lighthouses, flags on sailing vessels, and Paul Revere's famous use
oflantems, "One ifby land twoifby sea" on April 18, 1775. Suchearly demonstrations
of free-space optical communications were enabled by the use of widely available
terahertzelectromagnetic receivers, namelythe humaneye with its ability to detect the
electromagnetic visiblespectrumat wavelengths from about400 to 750 nm.Alexander
Graham Bell and SumnerTainterdemonstrated the first FSO telephonemessageusing
the patented"photophone" [49-53], illustrated Fig. 4. The photophone usedsunlightas
a source, modulated by reflecting off a vibratingmirror, and photoconductive selenium
as the receiveron June 3, 1880.
116 David O. Caplan
Fig. 4. :(Left), The Bell photophone transmitter and (right) photophone receiver. Rerinted with
permission of Lucent Technologies. Inc./BellLabs.
However, despite these early advances, optical communications was soon over-
taken by rapid development in wireline and wireless RF communications pioneered
by the likes of Hertz [54], Bell[55] , Edison [56], Marconi [57], Loomis [58], and Fes-
senden [59] in the late 1800s and early 1900s. During the following 100 years RF
communications became a refined commodity with applications ranging from radio,
television, cellular, and satellite communications. As the established means of wire-
less communications, the maturity and sophistication of existing RF capabilities make
them worthy of comparison to up-and-coming FSO systems.
Present day FSO systems have significant potential to improve upon RF capabilities
in areas where bandwidth or regulatory limitations exist, applications where security
is important, and in ultra-long-haul space-based links where size, weight and power
(SWAP) are at a premium and diffraction losses dominate. Since the invention of the
laser in 1960, there have been significant developments in the critical technologies
needed to tap the potential of photonic communications summarized in the list below.
Significant lasercom milestones

1960 Invention of the laser, [60-63]
1962 Invention of semiconductor diode laser, (GaAs) [64,65]
1964 Invention of fiber optical amplifier (Nd:glass I p,m) [66,67]
1970 First continuous-wave room-temperature semiconductor lasers [68]
1970 Development of low loss «20 dB/km) glass fibers [69-74]
1980 First commercial optical fiber system at 45 Mb/s (AT&T)
1987 Development of Erbium-doped fiber laser amplifier EDFA (1.5p,m)
[75-77]
1988 First trans-Atlantic fiber cable (280 Mb/s, 1.3 p,m) [78]
1989 First undersea tests of optically amplified (EDFA) fiber system
1992 First trans-Atlantic fiber cable using 1.5p,mtechnology (560 Mb/s) [79]
1993 Mel purchases 500 EDFAs and begins installation in terrestrial network
1995 Bidirectional Ground -to-Orbit Lasercom Demonstration (GOLD),
I Mbps up- and down-link transmissions @ 0.514 and 0.830 p,m on
Engineering Test Satellite-VI (ETS-VI) in elliptical GEO transfer orbit
[80-83]
Lasercommunication transmitterand receiver design 117
1996 First trans-Atlantic fiber cable using EDFAs (5 Gb/s) [84]

1996 1 Thit/s demon strated over fiber [85-88]
2000 360 million km of fiber world wide [89]
200 I Capacity of commercial WDM systems exceeds 1.6 This [89]
200 I 10 This demonstrated over fiber [90-94]
200 1 Geosynchronous Lightweight Technology Experiment (GeoLITE) [95-
97], successful demonstration of bidirectional laser communications
between a satellite in geosynchronous orbit (GEO), ground, and aircraft
2001 Semiconductor Intersatellite Link Experiment (SILEX) [98-101],
demonstrating bidirectional GEO-LEO and GEO-ground laser com-
munications between the Advanced Relay Technology Mission Satel-
lite (ARTEMIS) in GEO, with a IOmW (average) 2 Mbps (2-PPM)
directly-driven semiconductor laser TX @ 0.8 p,m and a 50 Mbps Si-
APD based NRZ-OOK RX; the SPOT-4 imaging satellite in low earth
orbit (LEO), and the optical ground station (OGS) [102,103] .
2003 Mars Laser Communication Demonstration (MLCD) program initiated
(targeting the ", 300,000,000 km Mars to Earth link, '" 1-30 Mbit/s)
using 5W MOPA TX, coded M -PPM (M = 16 - 64) and photon-
counting detectors @ 1.064 p,m [1,2,104]. Critical technologies demon-
strated but program cancelled in 2005.
2005 Successful bidirectional inter orbit lasercom link between Optical Inter-
orbit Communications Engineering Test Satellite (OICETS) in LEO and
ARTEMIS in GEO at 2 and 50 Mbps [105]
It is likely that within the next decade, we will see the first deployments of opera-
tional free-space laser communications systems with global and interplanetary reach .
2. General Wavelength Considerations
2.1. Carrier Characteristics
Electromagnetic (EM) waves can be described as an oscillating field with frequency,

vs that travel at the speed of light, c. The inverse relationship between the EM carrier
frequency (v) and wavelength (A) is
c = AV (3)
where the constant c = 3.0 x 108 m/s is the speed of light in vacuum. Quantized
energy is proportional to frequency, and is given by Planck's relation, which defines
the energy-per-photon to be
E = hv = he/A [J], (4)

where h =6.63 X 10- 34 J s is Planck 's constant. The illustration of the electromag-
netic spectrum in Fig. 5 spans 15 order s of magnitude from kilometer long RF waves
to picometer long gamma radiation and gives perspective into the broad range of wave-
lengths, frequencies , and photon-energies that can be used for communications.
As discussed in further detail in section 2.3, these basic EM properties profoundly
impact the transmitter and receiver designs . For example, as we transition from using
118 David O. Caplan
THE ELECTROMAGNETIC SPECTRUM

10 -1 10.2 lU I 10~ 1U - ~ 10 · 10. 7 1U·3 10 · 10 If' 10,11 10 ,11
I I I I I I I
Sl,r ul ..
......VC"
k l'(]th .@ ~
, .. '!>-<
;;0,
........
Cnmmon
nlmemwoWt'
0..·1.."......
(~'..~.~;;: _ _.L...---'
\ «nncf)
J
.-
.' ~
10*' 10 ' 10:1

-'-------'-,------'--_.L...----'-_--'--_.L...----'_--'--,----..L...----'-:---'--_
10 9 10 101 lO l l 10 11 10 ' I
I
10 14
M.~I .1" "':~I f-" " ,
o,;,;~~ __-.l-
-
t _ ..L...----'_--"-_..L...----I_--"-_-'-----I'---'-_-'-_L..--'-_-'-_L..--'-_
(e1et l roll orult\ )
Fig. S. The electromagnetic spectrum is shown in terms of wavelength along with correspond-
ing frequency, and energy-per-photon. Conventional names of the spectral bands and example
sources are also depicted. Courtesy of the Ad vanced Light Source, Lawrence Berkel ey National
Laboratory.
RF carriers towards higher frequency optical waves for communication, the carrier
wavelength shortens . This directly impacts both transmit and receive antenna (aper-
ture) size and diffraction, which is proportional to wavelength. The diffraction limited
transmitter beam angle is approximately [47,106]
e = AI DTX [rad] , (5)
where DTX is the transmitter aperture diameter. Thus using shorter wavelengths allows
for smaller divergence angles, or equivalently, better directivity for a given aperture
size. Note that this also increases the precision required to hit the target receiver
and therefore comes at the expense of increased difficulty in pointing, acquisition,
and tracking . From Eq. (5), the far-field on-axis intensity (W/m 2 ) can be estimated ,
and for a receiver with collection area (ARX) the power delivered to the RX can be
calculated. Diffraction losses reduce the free space transmission efficiency, Tlspa ce,
defined as the ratio of on-axis received power and transmitted power, which in the
far-field (AL » Dh) is approximately given by [106],
1r DRX DTX ) 2 "

1]spa ce ~
( 4AL = 1 - Diffraction Loss , (6)
and decreases with the square of the carrier wavelength (A) and link distance (L) .
Carrier frequency directly impacts the fractional bandwidth available for modu-
lation, and therefore limits the maximum rate at which that information can be trans-
mitted. For instance , if the modulation bandwidth is limited to 10% of the carrier
frequency, this corresponds to 30 THz available bandwidth for a l/-tm optical carrier
versus 3 GHz bandwidth for a lcm RF carrier. This highlights a key difference be-
tween high-rate RF and FSO communications, since RF systems need to be spectrally
E
Fig. 6. A time varying electro-magnetic signal, s (t, A, v, cP, p), depicting the potential for
amplitude, frequency, phase, and/or polarization modulation .
efficient in high-rate applications at the expense of energy or photon efficiency (see

Fig. I), whereas FSO systems can often use bandwidth as a flexible design parameter.
Channel properties also vary substantially with wavelength. Clouds, fog, rain, tur-
bulence, scattering (e.g., Raleigh scattering ex: 1/,X4), atmospheric absorption, etc. all
have a strong wavelength dependence. Owing to the longer wavelength, RF commu-
nications are generally more robust in atmospheric channels where these factor s are a
limitation [107-109].
Lastly, photon energy is proportional to the quantum or shot noise in the transmitted
signal , and therefore can impact the receiver sensitivity. As the photon energy (in
Joules), increases, the number of photons per second decreases for a given power (in
W, or J/s). This in tum increases the "graininess" or shot noise of the signal, which
drives high-sensitivity optical receiver design as described further in section 2.3.
2.2. Electromagnetic Signaling Options
B
A general expres sion for a time varying electro-magnetic (EM) signal s (t, A, v, cP,
p) is given in Eq. (7) and illustrated in Fig. 6 where variables A, v, cP, and = ai +p
b} represent the carrier amplitude, frequency, phase , and polarization field properties
respectively, and can be time-dependent, e.g., cP = ¢ (t ):
B s(t , A, v, cP, p) = A(t) COS(27Tvt + cP)p [Vj m], (7)
where P(t) = IA(tW j (2zo) is the power in the field envelope in W, and Zo is the
impedance of the medium, which is 300 [l in free-space. Each of these basic field
properties can be modulated either alone or in combination depending on the system
needs . The additional dimensions can carry information, for use in multi-dimensional
symbol constellations, or simply improve the modulation format.
For example, in narrowband intensity modulated optical systems where stimulated
Brillouin scattering (SBS) nonlinearities are a limitation, carrier phase, polarization,
andlor frequency can be modulated to expand the optical spectrum beyond the SBS
bandwidth of ",50 MHz to mitigate this effect [110,111]. Even if unused, these pa-
rameters may need to be controlled, measured, or known at the receiver to achieve
optimum performance. For instance, FSO receivers are often single polarization with
120 David O. Caplan
N ~,
... ~ Q (.INavg )~
P BE ( OOK ) <' - - 2-
Gl
(a) ~ PB":t;OK-;ount :::: e - 2 N~,
o
a. o •••
...
Gl
DPSK symbo l 't, = 1 PBE ( DPSK ) = t e - N ~,
(b) ~ Coh
p BE " '" ' _ f4'/
Q ('N' )-- e. - 2 N . "
O -l--
a. () +'-- .._- t---1I---I- -I- -t-- -1
... (PSK ) - a"gH
Time M-PPM sym bol

sl ot = 'r.J4 Ts = 2 Ppeak= M* P avg = 4 Pavg
,...., ~
i
PBE ( 2 - PPM )
P BE ( M - PPM )
...
... PBE (2 - PSK)
(d) ~ P EE ( M - PSK )
a. ...
PolSK symbol
't, = 1
PBE ( 2- POISK )
p
Time
Fig.7. Commonly used modulation formats (a) OOK, (b) DPSK, (c) 4-PPM, (d)4-FSK, (e) PolSK
with same data, data rate and average power. Also shown on the right-hand side is the approximate
bit-error probably for the modulation formats (a)-(e) with ideal optically preamplified receivers
[except for photon-counting OOK in (a) and coherent PSK in (b)) .
tight spectral filtering to allow for efficient diplexing of transmit and receive signals
[46], and/or to reduce detected noise. This requires proper alignment of transmit and
receive wavelength and polarization, either by stable open loop control of both the TX
and RX, tracking and compensation by the RX, feedback between the TX and RX, or
a combination thereof.
2.2.1. Overview of FSO Modulation Formats and Sensitivities
While there are many modulation possibilities, the most common modulation formats
considered for FSO links include on-off-keying (OOK), differential-phase-shift-keying
(DPSK) , phase-shift-keying (PSK), and orthogonal modulation formats such as M-ary
pulse-position modulation (M -PPM) and frequency-shift-keying (M -FSK) illustrated
in Fig. 7.
Lasercommunication transmitter and receiverdesign 121
2.2.1.1. On-Off-Keying (OOK)
The predominant form of signaling used in optical communication systems today is

on-off key (OaK) because the transmitter and receiver hardware are relatively simple
and fiber optic networks generally operate at high signal-to-noise ratios with small
dynamic range requirements and well controlled signal levels at the receivers. Also
known as binary amplitude-shift-keying, OaK is a form of intensity modulation (1M)
in which binary information is represented by the presence or absence optical signal
energy within the symbol. At the receiver, the "I" or "0" logical decision is determined
by the received symbol energy being above or below a predetermined threshold. The
optimum threshold balances the probability of 0 and I errors, and is dependent on
the received signal power and noise statistics, requiring adaptive thresholds for best
performance over a fluctuating channel [20,39,112,113]. Regardless of the accuracy of
optimum OaK threshold predictions based on theoretical calculations, the threshold
and other operational parameters can be optimized with near-real-time BER feedback
that is readily available with commercial FEe hardware [114,115]. Alternatively, mod-
ulation formats that inherently establish a threshold, such as antipodal or orthogonal
signaling (e.g., PSK, DPSK, FSK, PPM) can be used to eliminate the need for adaptive
thresholding.
For an ideal shot-noise-limited optically-prearnplified receiver, the OaK error
probability is approximately given by [18,20,23,37-39]
PB E(OOK) ~
r,;r-) ~ 21 e
Q ( V Navg -Na v / 2
g, (8)
where N a v g is the SNR, equal to the average number of photons-per bit (PPB) received,
and the Q-function is
Q (z) = v'211r" !
oo
z
2
1 Erfc (
e _x dx = 2 z)
v'2 1 ( 1 - Erf (
= 2 z )) .
v'2 (9)
The Erf( ) and Erfct) functions are the error function and complementary error
function, respectively [18,20]. The 10- 9 bit-error rate corresponds to 36 PPB, slightly
lower than the right-hand approximation in Eq. (8) which results in ",40 PPB. When
implemented with non-return-to-zero (NRZ) waveforms with an equal probability of
I 's and O's, the peak power is equal to twice the average. Pulsed return-to-zero or RZ
waveforms can be used with all the modulation formats mentioned above , with the
peak -to-average power ratio varying inversely with the duty-cycle.
Note that the analytical BER expressions in Eqs. (8)-(11) and (13)-(14) assume
Gaussian noise statistics [18,39,116-118], which do not accurately reflect the Poisson
noise statistics of the received input signal (see section 2.3.5), nor the statistics of the
preamplified signal detected by the square-law photodetector. The statistics for a sinu-
soidal electric field with narrow-band additive white Gaus sian noise after square-law
detection for 0 and I signal s are no longer Gaussian, they are Rayleigh and Rician,
respectively [116,119,120]. However, detailed numerical evaluation of preamplified
OaK performance assuming these distributions yields ",38 PPB at 10- 9 BER [36,
38], compared to ",36 PPB with the Gaussian approximation. An additional correc-
tion includes the exact stati stic s based on the quantum-mechanical description of the
122 David O. Caplan
physical photon-detection process which are the degenerate Bose-Einstein probabil-

ity distribution for O's and the noncentral negative binomial distribution (NNB) also
called the Laguerre distribution for I's [47,121-126] . Nevertheless, the Gaussian ap-
proximation yields straightforward analytic BER expressions that provide a reasonable
estimate of receiver sensitivity, with SNR estimates for a particular BER accurate to
within>- I dB relative to calculations based on with exact statistics [126,127].
2.2.1.2. Differential-Phase-Shift -Keying (DPSK)
Differentially encoded PSK [DPSK, see Fig. 7(b)] has received considerable attention
by the FSO community and the telecom industry due to a ",,3 dB sensitivity improve-
ment over commonly used on-off-keying, and reduced peak power which mitigates
nonlinear effects [24,25,41 ,128-133]. NRZ-DPSK can be implemented with a con-
stant envelope, so that the peak power is equal to the average. Binary information is
conveyed with two orthogonal symbols represented by the relative phase between two
differentially encoded bits: a "0" represented by no-phase change and a "I " by a 1f
phase difference, (or visa-vera). This phase modulation also tends to suppress the opti-
cal carrier, which can be advantageous in systems limited by slow optical nonlinearities
such as SBS (see section 3.5.5). Typically, adjacent bits are differentially encoded with
a time-separation r« of one symbol duration or a bit period, 7b it , but this can gener-
ally be extended to an integer number n of symbol periods, i.e., re = n7b it. This
can provide some flexibility in implementing simplified multi-rate and multi-channel
receivers (see, e.g., [134-136]), a subject discussed further in section 5.2.
DPSK's utility has been established with many long-haul fiber-optic experiments
demonstrating multiple Tbitls over r- I0,000 km fiber spans with hundreds of WDM-
DPSK channels (AS), e.g., [137-141]. Despite significant benefits, DPSK has not yet
been widely deployed in operational systems, in part because the benefits of DPSK
come at the cost of increased complexity over OOK, requiring a phase modulator
and differential precoding in the transmitter, and an optical delay-line interferometer
(01) demodulator and balanced detection in the receiver in order to derive maximum
benefit. The DPSK RX design is discussed further in section 5.2. Performance can
also be degraded significantly by differential chirp [132,142] . However, DPSK is much
easier to implement than coherent PSK, which requires a stable, narrow-linewidth local
oscillator phase-locked to the received signal necessary to implement the homodyne
receiver. In contrast, DPSK can be demodulated through a self-homodyne process of
comparing the relative phase of the differentially encoded symbols. This eliminates
the need for the LO and associated phase-lock ing challenges, and relaxes the need
for narrow linewidth laser sources, since the self-homodyne process only requires
phase-coherence for the time duration between the differentially encoded symbol. The
theoretical error-rate for optically preamplified DPSK is
I -Na v g
P B E(DPSK) = 2e , (10)
corresponding to 20 PPB @ 10- 9 BER [38]. The nearly 3 dB sensitivity benefit of

DPSK over OOK can be viewed as a result of the differential encoding, which utilizes
energy from the two symbols to determine the relative phase for one bit of information .
Optical differential quadrature phase shift keying (DQPK) [132,143-147] and
duobinary modulation [148-152] are spectrally efficient modulation formats related
to DPSK that have recently been considered for use over long-haul fiber links, where
dispersion is often a dominantlimitation. However, dispersionand channel bandwidth
are generally not limiting factors in free-space links, so the combination of reduced
photon-efficiency and increased complexity make these modulation formats less at-
tractive than DPSK for use in free-space applications.
2.2.1.3. Phase-Shift-Keying (PSK)
As noted above, coherent PSK has challenging LO and signal alignment requirements
[20,153-155] needed to determine the absolute phase of the optical field. In addition,
there are practical tradeoffs between PSK RX bandwidth and sensitivity, which are
especially noticeable since, unlike preamplifier receivers, coherent receivers are not
easily scaled to higher data rates via wavelength division multiplexing (WDM). For
the additional complexity, coherent (binary) PSK provides among the best theoretical
RX sensitivity, with
PB E(P SK) = Q ( vi 4Nav g ) :::::: e- 2N

avg
, (II)
corresponding to 9 PPB RX sensitivity @ 10- 9 BER [20,156]. But in practice the

high-sensitivity potential of optical PSK has not been realized. As seen in Fig. 61,
the best PSK demonstration reported uncoded sensitivity about 2.5 dB from theory
at 4 Mbps [157]. In the Gbit/s regime the best PSK performancedegraded to '"'-'6 dB
from theory at '"'-'6 Gbps and '"'-'9 dB from theory at '"'-'8 Gbps [158], providing little
benefit over optically preamplified DPSK, which is a WDM-scalable approach with
numerous demonstrationsof comparable or better performancein terms of sensitivity
and data rate [24,25,41,128-130,133,159].The nearly 4 x sensitivitybenefit PSK has
over OOK is in part due to binary PSK being an antipodal format, in which a "0" is
represented by signal field of -I (11" phase) and a " I " is represented by a I (phase of
0), so that 2-PSK has twice the signal-distance of OOK. The other 2 x advantage can
be viewed as a result the coherent detection process, which detects the peak power
of the received optical carrier, that is twice the average power measured in intensity
modulation formats.
2.2.1.4. M -ary Orthogonal Modulation
To improve receiver sensitivity, high-order M -ary orthogonal formats [18] such as

M-PPM, M-FSK [160-162], and hybrid combinations can be used (see also section
5.3). While not spectrally efficient, these formats can significantly improve photon-
efficiency. Two symbols m and n are considered orthogonal if
J
o
sm(t)sn(t)dt = Omn, (12)
where T s is the symbol period and Omn is the Kronecker delta function. The param-
eter M refers to the symbol alphabet or constellation size, with the information per
symbol increasing as k = log2(M) and bandwidth increasing with M . This results
in a spectral efficiency [Bit/sIHz] that varies as log., (M) / M shown in Fig. 3, which
clearly illustrates the trade between bandwidth (spectral efficiency) and SNR (photon
124 DavidO. Caplan
efficiency) noted by Shannon (see Fig. 1). For large M,

ideal performance approaches
Shannon-limited capacity [I7 , 18,23]. Orthogonal modulation formats also establish
their own threshold, with the optimum decision based on the largest sample within the
symbol set. This enables the system to operate optimally over wide and rapid changes
in signal level, without the need for adaptive thresholds that are necessary for OOK.
The bit-error rate for optically-preamplified binary-orthogonal modulation, e.g.,
2-PPM is given by
1 _ N av g
P B E(2 -0RTH) = 2e 2 (13)
with sensitivity comparable to OOK and exactly twice that of DPSK, corresponding
to 40 PPB at 10- 9 BER . For M-ary orthogonal modulation, no simple closed form
expression exists. An accurate but computationally challenging expression for the ideal
optically preamplified symbol-error-rate is given by[l63]
where the Rician distribution [119]
(15)
is the probability density function for sample Xj, where X l repre sents a sample with
both signal and noise, and Xo represents the samples with noise only, and Nsp is the
spontaneous emission factor related to the noise figure of the amplifier. For an ideal am-
plifier, N sp = I. The function 10 is the modified Bessel function of the first kind of zero
order. Robinson, in [163] provides an excellent derivation of the optically-preamplified
symbol-error rate in Eq. (14) along with simplified techniques for evaluating the ex-
pression.
For each symbol error, there is still a chance that some of the bits within the
deocoded symbol are correct. Assuming that each symbol is transmitted with equal
probability, the bit-error probability is
P B E(M-ORTH) = (MM~21) PSE =}

M---+large
PSE(M -ORTH)
2
(16)
Qualitatively, the M -ary orthogonal bit-error probability can be related to the

binary BER expression in (13) by
l_~
PBE(M -O RTH) ~ 2e 2 (17)
where keff is related to the bits-per-symbol given in Eqs. (18)-(20):
keff = k ( k-l)
1- kN
c
' (18)
(19)
Lasercommunication transmitter andreceiver design 125
- M= 2
- M=4
- M=8
- M = 16
- M = 32
• •• •• M = 64
••• • • M =12 8
0::
W 3 . •• •• M = 256
m ••• - - M =51 2
c; 4 -- --- M =1024
....o .......... M = 2048
.......... M =4096
6 -2-PSK
7
8
9
10
11
12 0L..-L_.l.____L_-'----l.._.L--.l.....::,.::..L...:~..:._:oL.:>."_J~_,L_~ .,.L---J..::"---:.l.----L-J?__--,J
4 5 6 7 8 9 10 11 12 13 18
Receiver Input Energy, PPB [dB]
Fig. 8. BERcurvesfor 2·PSK and optically preamplified M -aryorthogonal modulation.
Tk = 0.4 (1 - 1 _ e~(k-5) ) . (20)
For small Mand large received signal (Ns ) , k eff -+ k , yielding a receiver sen-
sitivity of about 40 PPB/log 2 M or 40lk at 10- 9 BER. This simple approximation
overestimates the RX sensitivity by less than >- 1.25 PPB, which for M < 64 at 10- 9
BER is accurate to '" ~ dB with the accuracy improving at lower error rates. For
increased accuracy without much computational complexity, Eq. (17) can be used to
calculate M -ary othogonal BER performance with "'0.1 dB accuracy.
Bit-error rate curves as a function of photonslbit (in dB) are shown in Fig. 8
for ideal 2-PSK and optically premplified M -ary orthogonal modulation. Note that
theoretical preamplified OaK and DPSK BER performance is similar to binary and
4-ary othogonal modulation, respectively . Without additional coding, 32-orthogonal
modulation has comparable performance to ideal 2-PSK at 10- 9 BER . At higher
10- 3 BER, near the FEe threshold for enhanced 255/329 Reed-Solomon codes, 2·
PSK RX sensitivity is as good as to 1024-orthogonal. Above 10- 2 BER, ideal2-PSK
has a clear sensitivity advantage over preampli fied M -ary orthogonal modulation.
However in practice, at high rates above a Gbitls (see Fig. 61), reported coherent-
PSK [158,164-166] suffers ",5 dB greater implementation penalties than preamplified
RXs [13,15,128-130,133,135,159], eliminating much of the theoretical benefit. While
improved PSK designs may achieve the anticipated sensitivities, based on existing
demonstrations, preamplified RX designs have exhibited superior sensitivity and data-
rate, as wel1 as WDM scalability [135,138,139,149,167].
2.2.1.5. M-ary Pulse-Position Modulation (M-PPM)
Pulse-Position Modulation (PPM) is a form of orthogonal signaling that uses the same
transmitter and receiver hardware used for OaK. In M-PPM [2,15,26,43 ,168-171],
k-bits of information are encoded by the position of an optical pulse within an M -slot
symbol, (Fig. 7c).A s a result, M-PPM waveforms have a low duty cycle, equal to 11M,
126 DavidO. Caplan
making it well suited for use with EDFA-based average-power-limited transmitters, but
a poor choice foruse with semiconductor-based peak-power-limited transmitters [172].
For M -PPM, the bandwidth expansion directly impacts the modulation bandwidth and
associated transmit/receive electronic bandwidth
M RO a t a = M R Oata = S (21)
BE(M -PPM) = log2 M k '
with associated symbol rate
RO a t a S
R sym(M -P PM) = log2 M = M' (22)
While electrical bandwidth limitations may limit the maximum slot-rate S for a
single M-PPM channel, the high-speed 10 and 40 Gbps electronics developed for the
telecom industry make it relatively easy to implement at reasonably high data-rates.
For example, by transmitting 16-PPM at a slot-rate S = 10 Gslot/s, 2.5 Gbps can be
delivered with symbol rate of 625 Msym/sec.
The low duty cycle of M -PPM waveforms can also lead to optical nonlinearities,
which can limit the peak transmit power, a subject discussed further in section 3.5.5. On
the receive side, M-PPM requires two clocks to be acquired, a symbol clock (R sy m)
and slot clock at rate S = M R sy m [173]. Clock acquisition can be challenging for
large M since received power at the clock frequencie s varies as rv(1IM2 ) , which may
require embedded synchronization bits within the data as M gets large [44].
M -PPM has several practical advantages over OOK and DPSK. As a result of the
low duty cycle, PPM has less spectral content at low frequencies and consequently
has a smaller fractional bandwidth than OOK and DPSK. Fractional bandwidth (B f)
is a factor used to classify signals as narrowband , wideband, or ultra-wideband and
is often defined by the ratio of bandwidth at -10 dB points to center frequency [174]
given by
B BW lOd B 2 (fh - /l)
f = (fh + /l) /2 'J» + fl) , (23)
where fh and h are the highest and lowest cutoff frequencies (at the -10 dB point)
of the spectrum. For RF signals , B f > 20% is considered ultra-wide band. However
for optical communications signals, the spectrum of the RF drive spectrum typically
has B f > rv200% as shown in Fig. 9. For such broadband signals, a more relevant
parameter is the high-low bandwidth ratio,
(24)
For a pseudo-random bit sequence (PRBS) of 10 Gbps NRZ-OOK waveforms,

B r extends from a practical lower bound of rv10kHz to rv10 GHz, or six decades . In
contrast, the spectra for M -PPM waveforms operating at the same data rate span less
than two decades, reducing B; by over 4 orders of magnitude , despite having more
high-frequency content (Fig. 9).
This relaxes the performance requirements on wide-band electronic amplifiers and
drivers . In addition, since the longest string of consecutive I 's is two (from two adjacent
PPM symbols) , pattern-dependencies in transmit and receive hardware are reduced ,
making it easier to generate and receive high-quality waveforms. These benefits can
also extend to the ultra-high -speed domain , where the reduced pattern dependence
f.5 ; • • • • t;D.·.f !lr<J".]n~Jt(~:'1~! I .~,:; i T Ii

·15 ..... NRZ·OOK 10 kHz 7.6 GHz 7.6 GHz 2 GHz 6 ... , .;.~ :.:. . ~
. _:.. ..:.... :... :
8. 2-P PM 0 .8 GHz 7.6 GHz 6.8 GHz 1.6 GHz 1 ~ ~ :': • \ /I : ,, ~
I:: ~ :
~ ·20 ..... •••• 4-P PM 0.35 GHz 6.5 GHz 6.1 GHz 1.8 GHz 1.3 :.. ·H..f f -~ - " '::r: ' I:' A
" I ~: jif
~ -_ .• 16-PPM 0 .1 GHz 6.9 GHz 6.8 GHz 1.9 GHz 1.9 : :: ' ," : - : :i:, ~1
o -25 ,'I
11. 10.2
Fig. 9. Calculated spectra forsquare waveforms used in OOK and 2-PPM,4-PPM, and 16-PPM
modulation forwith a fixed 100 psec pulse width [175). The M -PPM waveforms have a smaller
fractional-bandwidth (B f) and asignificantly smaller bandwidth ratio, tB; = !h/ fd than OOK.
For the OOK spectrumjj is set to 10 kH z, which is a common low-frequency specification for
applicable broad-band electronics. Note that for a constant data rate, the M -PPM spectra are
broadened bya factor of M / log2 (M), which increases BW lOdB but doesnotimpact B«
has improved the performanceof all-optical demultiplexers [163,176] . M-PPM also

benefits from the sequential nature of the symbol set, which enables a single-chain of
drive electronics and associated filters to generate and receive the complete symbol
set. This simplifies and improves decision process, since it is easier to make a fair
comparison of the M -samples within a symbol to determine which is the largest.
2.2.1.6. M-ary Frequency-Shift Keying (M-FSK)
The FSK symbol set consistsof M frequencies or wavelengths and like PPM, conveys
k = log., (M) bits of information per symbol (Fig . 7d). Since FSK transmitsa different
frequency for each symbol, modulation bandwidth and associated transmit/receive
electronic bandwidth requirements are actually less than or equal to the data rate:
Ro a t a
B E(M-FSK) = -I
M = R sym(M- FSK ) . (25)
og2
The bandwidth expansion in M -FSK instead comes from the spread of M-
frequenciesused, whichgenerally requires a parallel M -channel receiverdesign.This
adds some complexityover the serially generated and receivedPPM waveformsmen-
tioned above, since any imbalances in power or waveform fidelity in the M TX or
RX parallel channels can degrade performance. As with DPSK, FSK waveforms can
have a 100% duty cycle, making it well suited for use with average or peak power-
limited transmitters. It can be implemented using a single frequency-modulated (FM)
source with binary- [162,177-179] or M-ary modulation [160,180], or using external
modulation and subcarriermultiplexing(SCM) [181,182]. These approachesgenerally
require relatively tight wavelength spacing due to transmitter or receiver limitations
[161]. However, with the use of independently-modulated wavelength-multiplexed-
sources, or fast-tunable wideband DBR sources [180], arbitrarily wide wavelength
spacing can be used. For parallel TX and RX designs, the duty-cycle per channel is
128 David O. Caplan
roughly 11M, assumingall symbolsare equally likely, which leads to a similar reduc-
tion in the bandwidth-ratio and associated benefits that were described for M -PPM
above.
2.2.1.7. Polarization-Shift-Keying (PoISK)
Binary-polarization-shift keying(2-PoISK) [183] can be used to extendthe symbolset

in the M -ary orthogonal formats described above by a bit (one bit per symbol), i.e.,
M -+ 2M and k -+ k + 1, without increasing peak power or electrical bandwidth
requirements per channel. The PolSK symbols encode binary information in two-
orthogonalpolarization bases, such as left and right-hand-circular polarizations or "s"
and "p" linear polarizations, which can be implemented using combinations of phase
modulatorand parallelintensitymodulators [184]. The performance ofgeneralizedM-
PolSK has recently been evaluated for hybrid combinations with M -FSK [185] and
multidifferential phase(DPSK)and amplitudemodulation (ASK) [186,187] . However,
foroptimalsensitivity, an orthogonal M -arysignalsetas definedin Eq. (12) is required.
Like FSK and DPSK, PolSK is well suited for use with averageor peak powerlimited
transmitters, since it can operate witha 100% duty cycle. Furtherdiscussionof hybrid-
modulation formats, performance, and implementation benefits is givenin section 5.3.
2.3. Comparison ofRF and Optical Properties
Until this point, the description of the EM field has been qualitative. However, there
are many distinguishing characteristics between the RF carrier and the optical carrier
that becomeapparentas the specificwavelengths are compared.A summary of values
for a I-em microwave carrier and a l-um optical carrier at the same power level is
shown in Table I .
Table 1. Comparison of I cm RF and Ipm optical carrier characteristics.

Microwave-RF Optical Units
>. 1.0 x 10 2 1.0 x 10 6 m
v 3.0 x 1010 3.0 X 1014 Hz
tu/ 2.0 X 10- 23 2.0 X 10- 19 J
Power 1.0 x 10- 9 1.0 X 10- 9 W
rph 5.0 x 1013 5.0 X 109 Photons/s
r ph x 10- 9 5.0 X 104 5.0 Photons/ns
rph X 10- 10 5.0 X 103 0.5 PhotonS/100ps
2.3.1. Diffraction
The factor of 10,000 between RF and optical wavelength leads to a substantial differ-
ence (108 or 80 dB) in free-space (vacuum) diffraction loss estimatedin Eq. (6). Since
diffraction loss increases with the square of the link distance-wavelength product, for
equalTX and RX areas, the opticalcarriercan propagate a factorof 10,000furtherbe-
fore incurringthe same loss, a primaryreason for considering opticalcommunications
for ultra-long-haul free-space communications. The optical signal is much is more

directional and concentrated in the far field, delivering more signal to the target RX.
But it does this at the expense of more challenging PAT, in contrast with RF systems
which are more omni-directional.
The propagation losses in free-space grow as the square of the distance [see Eq.
(6)] in contrast with fiber, in which the losses are exponential. The link loss for 12%,
100%, and 1000% free-space distance multipliers are 1,6, and 20 dB, respectively. The
channel loss for some potential planetary links, relative to an Earth to geosynchronous
orbit link), are shown in Table 2 below. To highlight the difference between square-law
diffraction and exponential losses, a link from Earth to Mars connected using low loss
fiber with attenuation coefficient 0.2 dBlkm would result in a loss of 80 million dB
[104].
Table 2. Approximate maximum distance from Earth and relative diffraction loss for free-space
propagation between the Earth and geosynchronous orbit, the Moon, Mars, Jupiter, and Saturn.
Orbit Distance [km] Relative Channel Loss [dB)

Geo 40,000 0
Moon 400,000 20
Mars 400,000 ,000 80
Jupiter 1,000,000,000 88
Saturn 1,600,000,000 92
Note that while pointing, acquisition, and tracking (PAT) is outside the scope of
this paper, the TX and RX design can greatly facilitate PATfunctionality. For example,
prior to acquisition, the received SNR is generally poor due to low signal strength and
since the benefits of filtering may need to be established. Such filtering may occur in
several independent domains, e.g., spatial, temporal, spectral, and polarization; and
acquisition or alignment in each may be necessary in order to acquire the link. Since
the spatial acquisition generally poses the most fundamental challenge and time to
acquire in FSO systems , streamlining acquisition or alignment in the other domains
by providing fast clock recovery [188] recombined with stable TX polarization and
wavelength that can be quickly aligned with the RX, can greatly expedite the overall
acquisition process. PATcan also be simplified by increa sing the number of RX spatial
modes [1,2,47,189], by reducing tracking requirements and/or sensitivity to platform
jitter, although this generally comes at the cost of increased RX noise.
2.3.2. Optical Detection
In contrast with RF electronics and detectors, which can directly detect the EM field and
generally have bandwidths exceeding that of the carrier frequency, optical detectors are
generally "square law detectors" with functional bandwidth s that are a small fraction
of the carrier frequency [190]. This means output is proportional to the square of the
field, (proportional to the power intensity), with little direct dependence on optical
phase, frequency, or polarization. Optical detectors can be extremely high speed, with
bandwidths from DC to > 50 GHz. Because of optical detector bandwidth limitations,
practical detectors have a response time which is generally much slower than the
130 DavidO. Caplan
<femrosecond (l0-15S) period of the optical field, so these detectors respond to the
time average square of the field. The detectable low-frequency optical power is given
by
(26)
where the time average of (cos 2 ()) = ~. In high-speed designs, Zo is typically son
in order to match RF waveguide impedance to minimize reflections.
In view of the intensity or power dependent response of optical detectors, the most
commonly used optical modulation formats use intensity-modulated direct-detection
(1M-DO) modulations. An example on-off-keying (OaK), in which the information
is encoded by the presence (a logical "1") or absence (a logical "0") of the op-
tical signal. In order to provide access to the phase, frequency or polarity of the
electric field, more complex receivers which employ optical preprocessors are re-
quired . Preprocessors include wavelength dependent filters, interferometers, polar-
ization beam splitters, and coherent measurements enable measurement of the elec-
tric field [153,157,160,161 ,179,191-195]. Further discussion of detector technologies,
performance, and impact of noise sources is given in sections 2.3.5-2.3.7 and 4.
2.3.3. Technology Limitations
For some of the same reasons that fiber-optic communications presently dominates
long-haul high-rate guided-wave applications, namely low channel losses and wide
available bandwidth (with little or no regulatory limitations), FSO communications has
a bright future. As summarized in Table 3 the technical advantages are considerable.
For instance, there are roughly 4 orders of magnitude more bandwidth accessible via
readily available WDM and ultra-wide-band, high-gain optical amplifiers. These opti-
cal technologies enable the trade of spectral-efficiency for improved photon-efficiency
(as suggested in section 1.1), while maintaining the ability for scalable high-rate com-
munications:
FSO terminals can also leverage robust wideband fiber-optic connectivity to permit
modular construction [46] that is insensitive to electromagnetic interference (EMI) .
On the other hand, RF technologies are generally more mature and have extensive
ability to integrate . In terms of average TX power efficiency, RF transmitters have a
",3 to 6 dB advantage over amplifier-based optical transmitters, although this deficit is
mitigated by the average-power-limited properties of optical amplifiers that enable the
flexible use of variable-duty-cycle waveforms without reducing TX efficiency. Free-
space RF communication is better suited for omnidirectional applications and provide
improved reliability for links where clouds may be a limitation, However, for long-
haul-high-rate free-space applications, especially space-based links, such benefits are
overshadowed by the many advantages of optical technologies summarized above. Ex-
ample link budgets in section 2.4, Table 4 based on fundamental carrier characteristics
and in section 5.5, Table 15 based on currently available technologies illustrate these
considerations.
Laser communication transmitter and receiverdesign 131
Table 3. Comparison of RF and optical wavelength-dependent characteristics and technical lim-

itations. • Note other optical bands are available, e.g., 0.85, I and 1.3 /-lm. The table includes
referencesfor efficientopticalamplifiers at 1 /-lm [198,204), whichare generally>-1.5 to 2 times
more powerefficientthan their 1.5 /-lm counterparts.
Parameter RF Optical
Wavelength (frequency) '" 1 em, (30 GHz) 1.5/-lm (200 THz)*,
RX noise limitation [dBm/Hz] Thermal (KT =-174) Shot (hv = -159)
Bandwidthlimitations GovernmentRegulated Unregulated
«30 GHz « 200 THz
State-of-the-art amplifiers
Gain 10 dB 50 dB
Bandwidth '" 50 GHz '" 50 THz
NF 5 dB 3 dB
Transmitter Characteristics Peak power limited Average power limited
Integrability Massive Limited
Modulation Formats Sophisticated Basic
Efficiency (peak) '" 50-80% (196) '" 5-20% [172,197-200)
Efficiency (average) '" 5%-46% [196,201-203] '" 5-13% ,21 % [172,197-200)
Waveguide Characteristics
Type Coax Fiber
Bandwidth '" 20 GHz '" 50 THz
Loss 100dB/km 0.2 dBlkm
Free-space channel characteristics

Diffraction Angle 1O- 2/D 1O-6/D
Cloud penetration Good Poor
2.3.4. Average and Peak Power Limited Transmitters
RF transmitters are generally peak power limited, which means that there is an upper
limit on the peak output power they can deliver (e.g., a maximum voltage) . As a result,
in order for RF systems to deliver maximum average power (the relevant metric for
assessing communication performance), they need to operate with high >- 100% duty
cycles, which places limitations on the choice of modulation formats and waveforms.
Furthermore, in order to maintain maximum transmitted power and operate at multiple
rates, peak-power-limited systems need to alter the transmitted pulse width and shape
so that the high duty cycle is preserved . In order to maintain optimum communication
performance with this constraint, different receiver filters are necessary for each bit-
rate, increasing RX complexity.
In contrast, optical transmitters are typically average power limited, which means
that the average output power is independent of the transmitted waveform shape or
duty-cycle. This provides significant advantages over peak-power limited RF systems ,
enabling optical communications systems to use aggressive pulse shaping techniques,
and variable-duty-cycle modulation formats that can approach optimal performance
132 David O. Caplan
Arrival of Ideal
ph ot o ns fr om N(t) i(t)
Photodetector Low Pass Filter
c oherent I------I~I
s ou rce at Infin ite bandwidth h(t)

ave rage rat e No d ar k current noise
ITD
0 2 4
: ITIJ 6
I 0 0 8 10
Time. [sec]
Fig. 10. Photodetection modeledas a fast noiseless idealphotoncountingdetectorthat generates

a streamof detectioneventswhichform a discretecount N ( t) that is a function of the photonrate
r ph . detectionefficiency, and observation time-window described below. The detected impulse
streamis thenfollowed byanelectricalfilterwithimpulseresponseh( t), thatconverts thedetected
eventsto a photocurrent.
over a wide range of rates while using simple high-sensitivity multi-rate receiverde-
signs [10-15]. This subject is discussed further in sections 3 and 5.
2.3.5. Quantum Noise Limitations

WhenconsideringRFandopticalquantumcharacteristics, thegranularity of the photon
flux is much more noticeableat optical frequencies, with rate corresponding to
Rph = P/(hv) [photons/a], (27)
where P is the optical power in W,and hv is the photon energy in Joules. For a large
numberof photons per observation interval, the shot noise is relatively small relative
to the average,and the photon flux appears continuous. However, for low photon flux,
the shot noise or graininess in the interarrival times becomes apparent, as illustrated
in Fig. 10. For the case of I nW of optical power and a lOO-ps observation window
in Table I, on average, only ~ a photon is detected, which means that statistically, no
photonsare detectedin halftheobservations. Suchquantumfluctuations in thedetected
photon counts, are fundamentally part of the optical signal, and therefore impose the
ultimate limitationon optical receiverperformance. This is in sharp contrast with RF
receivers, which are typicallydominated by thermal noise in the receiver. The impact
of quantum and thermal noise on communication performance is discussed in further
detail in references [20,121,127,190,205-207].
The statisticsof photonarrivalsfora coherentstate,e.g.,a lasersourceoperatingfar
above threshold, are Poisson [121,127] -with characteristics described in Eqs. (28)-
(32) and illustratedin Fig. II . The Poissondistribution in (28) gives the probability of
detecting n photons in a time interval T given the average numberof photon arrivals
per interval is < n >= N av g :
(n-N avg)2
_ (Navg)n e- N avg e 2N avg
P[n INavg ] - , (28)
n. N-+large J21rNav g
The right-hand expression in Eq. (28) is a discrete Gaussian approximation of the

Poisson distribution which is accurate for large N avg. As shown in Eq. (29), N avg
can be interpreted as the product of the rate of detected photons rph over observation
interval T ,
'rJP
Navg = rph T = hv T = 'rJ P P B, (29)
where 'rJ is detection efficiency, P is incident optical power, hv is energy-per-photon,
and PPB is the incident number of photonslbit when Tis the bit period. For a Poisson
distribution, the mean number of photon arrivals during the observation window, m,
is equal to the variance, a 2 , as shown in Eqs. (30) and (31), which can be derived by
twice differentiating the Taylor expansion of eN [208].
~ ~ n (Navgt e- N avg
m = (n) = L- nP[n Is.; 1 = L- ( )1 = Navg, (30)
o 0 n .
ash
2 «
=\n- ())2)
n = '\"""
00
L- n
2 (Navg v:«:"
e avg
,
2
- Navg = Navg. (31)
o n.
Accordingly, the SNR (in terms of received electrical power, proportional to de-
tected photocurrent squared), given by'
_ (n)2 m2
SNR = (A,w.n2) = -2-
ash
= N avg = 'rJPPB (32)
is also equal to the mean number ofphotons received. For the case where the observation
interval is one bit duration, N avg is equal to the number of photonslbit. As the mean
number increases. the Poisson distribution becomes more symmetric and converges
towards a Gaussian, also with equal mean and variance. While not noise free. the
statistics of laser light are relatively quiet when compared to single-mode thermal
light sources that follow Bose-Einstein statistics [121] shown also in Fig. II.
From Eq. (32), the SNR for a shot-noise-limited signal can be interpreted as having
a signal with N avg photons accompanied by I noise photon with energy lu/ Joules, or
equivalently with a noise power spectral density:
NShot(V) = hv [W1Hz], (33)
the shot noise power in I Hz of observed bandwidth. For an observation bandwidth of

B Hz, the total shot noise power is
PShot(V) = hvB [W]. (34)
For an ideal noiseless photodetector, the continuous-time photocurrent i(t) gen-

erated by detecting a discrete photon stream is described by a random process with
mean (signal) and variance (noise):
1 Note that the SNR can alternatively be defined in terms of received electrical current or voltage rather than
power. This alternative definition. commonly used in photon counting applications. yields
SNR v == Mean/ v'Variance = JNa v g .
134 David O. Caplan
0 .40
N.,vQ·1
Coherent light: P[nl rT] =(rT)" e -rT
nl
r = m e an p h o t on arrival rate In photon s p er s econd
I"-"'
0.30 T = ob serv ation ti m e
N. w " = r T
'Ol
-l M ean = N .'N9
~ =
V ari an c e N .w Q
.~ 0 . 20 2Na
SNR = rvlean riance = N .N'ljI
I 0 . 10
G3lJS.sbn Fit :
me3n . 50. Y.Ir . 50
lhwnnal l c,llt
(Bo$t'-8 n~t."tt')
'-~. . ' '' . .p. lnIN. I=(_1 J...!!.L J'

ItN. l + N,
f b -50 ~"-v. Mean = N•. Var = N~
10 20 30 40 50 60 70
Counls pe r sample. n
Fig, 11, The discrete Poisson probability distribution function (PDF) gives the probability of
detecting n photons given a photon rate Rph and observation window T , or equivalently the
average number of photons in the interval N = RphT, shown for N avg = 1, 2, 5, 10, 20, and
50. Also shown are a Bose-Einstein PDF (representing the statistics of thermal light in a single
mode) with a mean of 50, and Gaussian PDF with mean and variance of 50 (m (J'2 50) . = =
i avg(t) = (i(t)) = ~~ I: P(t)h(t - T)dT [amp], (35)
a;h(t) = «(i(t) - i a v g )2) = ~~ I: P(t)h

2(t
- T)dT (36)
where h(t) is the impulse response of the filtered photocurrent with integrated area
normalized to a unit charge [20,127,208]. For example, given a photodetector with a
simple exponential electrical impulse response,
-t IT
h(t) = ~u(t)
T
+------t H(J) = 1
+J7fT
~ f ' (37)
where T is the filtertimeconstant and u(t) is the unit step function, then withconstant
incident optical power P,
.
Zavg = hv
TJq P = rphq an
d 2
a sh
TJq (q)
= hv P 2T
.
= 2qzav g B e . (38)
The far right-hand term in Eq. (38) is the well known general expression for the shot
noise power(in units of amps") associatedwith the averagephotocurrent, where Be is
the noise equivalentpower bandwidthof the electrical responseover which the noise
is observed, i.e.,
Lasercommunication transmitter and receiver design 135
B e = 1= (IH(J)1 2 / H m ax ) df. (39)
For the case of the exponential filter in Eq. (37), B = 1/(47) . We can solve for
the quantum-limited SNR for direct detection by substituting Eq. (38) into Eq. (32),
yielding
2
m ryP r ph
SNRDD = -2- = 2h B = 2B = r p h 7 eff = Neff, (40)
17s h V e e
where 7 e ff is the effective integration time of the electronic response, equal to 27 in

the example above. Note the similarities with the expressions for photon count in Eqs.
(32) and (29) above. As with photon-counting, the direct-detected photocurrent SNR is
proportional to the average number of photons Neff received per effective observation
interval. For communications, this is often the number of received photons-per-bit,
which is reduced by any losses preceding detection . Examples of the impact of discrete
photon arrivals and the resulting shot noise generated in the received photocurrent as
a function of RX bandwidth are given in Fig. 12.
In practice, it is challenging to achieve quantum-limited performance due to the
presence of other noise sources such as thermal noise (section 2.3.7) and dark current,
effects that were neglected in the analysis above. Techniques of overcoming these
limitations are discussed in sections 4 and 5.
2.3.6. Quantum-limited Direct Detection (DD)
One of the simplest optical receivers is an On-off-keyed (OaK), direct detection re-
ceiver. With knowledge of the photon statistics, the impact of shot noise on opti-
cal communication performan ce can be readily observed and quantified for the ideal
intensity-modulation TX and direct-detection (IMlDD) photon-counting RX shown in
Fig. 13. On-off-intensity-modulated-photons from the transmitter representing a log-
ical "I" and no-photons (e.g., perfect modulation extinction) representing a logical
"0" are received by a noiseless photon-counting receiver, (i.e., a Geiger Mode pho-
ton counting detector with no dark counts, after pulsing [30-32,209,210] background
noise, or other parasitics such as timing jitter and inter-symbol-interference (lSI)).
Without background noise, the optimum decision rule is: if no photons are detected,
assume that data D = "0" was sent; and if photons were detected ( D=" I" ), assume that
D = " I" was sent. For these rules, bit errors only occur if a photon is not detected, e.g.,
D = "0" when D = "I" was sent, which is an "erasure" error. The photon-counting
OaK probability of bit error is given by
PBE(OOK) = P[D = O]P[D = liD = 0] + P[D = I]P[D = OlD = 1]

= 2"I p [n = OIN] = 2"e
1 - r
p
hT
= 2"1 e-2N av g , (41)
where the number of photons during the observation interval (N) when a "I " is re-
ceived is twice the average number received (N = 2Nav g ) , which is true for equal
probabilities of transmitting "0" or "!" data. From Eq. (41), the 10- 9 BER is achieved
for N av g = 10 photonslbit and corresponds to the "quantum limit" for the ideal IM-DD
photon counting OaK, since the only limiting noise source is signal shot noise.
A similar analysis can be performed for ideal photon counting binary PPM, in
which each symbol is determined by the location of the optical pulse, with "0" corre-
sponding to photons detected in the first half of the symbol and "I " from the second
136 David O. Caplan
(a)
o 2 4 6 8
r
10
I
20 ,..----~--~--~--~---, 20 ,--- - - - --,
r ."
l
.. 1 Hz < B" 1I4~ " 2.5 Hz. ~ .. 0 .1 s ec 1,,,,) =rp" =1
(b) 10 u2; =r .,,/2't- =5
l .
SNR =2r,.,,'t =0.2
2 4 6 8 10 2
.... 3
=2 sec
3 ,..--- - - - - -,
(e)
..
C
QI
:::l
o
rpII .. 1 Hz.
.
B" 1I4~" 0.125 Hz.
_-
~
__ .
0
0
.::.
Q.
00 2 4 6 8 10 0.5
.. 2 r---~--~---~--~----, 2 r------,,---,
=
(d)
..
eQI
:;
T ..
pII
1 Hz > B" 1 /4~" 0 .025
J'"-.
Hz. ~ .. 10 sec
.. 1. 0 ] 0.05
..
u
e
e
.::.
t>-"r-,.,~~-J:'""-~ . _ •.
Q.
00 2 4 6 8 10 1 2
Time. [sec] PDF
Fig. 12. Photocurrent and shot noise. The granularity of a stream of photons and the resulting
variance in the signal is shown as a function of the effective receiver filter bandwidth B e, where
Be =1/4T for the low-pass filter described in Eq. (37). (a) A discreate Poisson impulse stream of
photons with arrival rate Tph= I photon/so (bj-td) left show the continuous-time photocurrent as
filter bandwidth Be is reduced two order s of magnitude from 25 to 0.025 Hz, and (bj-Id) right
show the corresponding probability distribution functions (PDFs) of the received photocurrent
for each filter bandwidth. The SNR improves as the bandwidth decreases or equivalently as the
filter integration time increase s. But even with significant filtering in which multiple photons are
detected within the filter time constant, the (shot) noise added by the random arrival of discrete
photons is apparent. Note that for simplicity , the units of photocurrent have been normalized per
unit charge .
half. Bit errors in this ideal photon-counting system only occur when no photons are
received . In this case, a "coin flip" is used to decide (arbitrarily) which slot is the
winner (since both slots are identically zero in the ideal noiseless case) [163,211]. The
coin flip gives the correct answer the time, yielding 4
I - r IIT/2 1 -Na v g
PBE(2 - P PM) = -2 e P = -2 e '
(42)
This is similar to the OOK result but note the factor of in the exponent that 4
comes from the fact that the number of photons during any observation interval (either
i " ici;;i"~;i;;j ;;;'----" '---'---" --"'-j

On /Off ! Pho t o-d etector ! •
Intensity >--- , <'; , !Out p ut Data (D)
Modulato r
(P e rfect ER) i I
Tx 4~_'MMMMMM_ ••• _ _ .J L~~ . __. _ _._.._. __._..__.J
"-
II
J
Input Data (D) o
0..
Time
Fig. 13. Ideal photon-counting OOK TX and RX.
"0" or "I") is the same and equal to the average number received (N = N av g ) . Thus,
the quantum-limited photon counting sensitivity for 2-PPM is N av g = 20 photons/bit
at 10- 9 HER.
Extending the binary case to M -ary PPM, in which each symbol is divided into M
positions (slots) and carries log2(M) bits per symbol, the probability of error within
a symbol is
M- l ) e -Navg ,
PoSE ( M - P P M ) = ( ~ (43)
where the coefficient term is the (uniform) probability of symbol error when no photons
are received [211], and the exponential term is the probability of detecting no photons.
Substituting Eq. (16) into Eq. (43) and accounting for the number of bits per
symbol, yields the general expression for ideal photon-counting M -PPM:
P 1 -Navg log2 M = ~ e-Navgk

BE(M -PPM) = '2 e 2 (44)
As M increases, the PPB required to achieve a particular BER is reduced by

loG2M, the number of bits per symbol, which can extend to arbitrarily low values in
theory without additional coding. However, as noted by Yamamoto and Haus [27,28],
a practical theoretical limit of ", 0.02 PPB would ultimately be imposed by the ratio of
photon to thermal noise energy. In practice, photon-counting receivers are not ideal,
having parasitic effects such as dark counts, after pulsing, timing jitter, and limited
bandwidth [1,2,30-32,209,210], which impose additional constraints on performance.
Furthermore, background noise and transmitter imperfections can further degrade re-
ceiver performance [21,212].
To illustrate this point, given a TX with imperfect modulation extinction (ER =
power off/power on) , the probability of bit-error PBS for the OOK photon -counting
expression in Eq. (41) becomes
PB E( OOK ,E R) = P[D = OlD = 1] + P[D = liD = 0]

1 1
= '2 P[n = OIN] + '2P[n > 0IN · ER]
1 1
= '2 P[n = OIN] + '2 (I-P[n = 0IN · ER])
=~ ( e-(2Navg-Navg ER) + 1- e - Navg E R ) . (45)
138 DavidO. Caplan
2 3 4 5 6 7 B 11 12
PPB, [dBI
Fig. 14. Photon-counting OOK BER curves for varying TX extinction ratio (ER).
The (l_e - NavgER) term is the modulator seepage penalty due to the N avgER
(signal-dependent) background noise photons that arrive during the "Os". As shown
in Fig. 14, the background noise resulting from the imperfect ER imposes an error
floor that is a few times higher than the ER, (i.e., a 1O- 3ER :::}rv 2 X 10- 3 error
floor). The error rate continues to worsen as the signal is increased since the optimum
threshold is a function of the signal dependent "0" noise, not zero as assumed in Eq.
(45). This illustrates a potential limitation of photon-counting using this approach in
the presence of background noise, and the need for powerful FEC[42,213] that can
generate error-free performance with input error-rates of rv 10- 2[33,214] to overcome
such limitations:
N~vg
1 -
PBE(2- PPM) = 2 e 4 Nb , (Navg « Nb). (46)
Binary PPM is similarly impacted by the presence of background-noise photons,

although an adaptable decision can be easily implemented by comparing the first-half
and second-half of the symbol. For the case where the background Ni. is much larger
than the signal, the BER expression becomes [212].
For average power limited transmitters, extending the modulation format to M-
PPM is a technique for overcoming background noise (improving SNR) without in-
creasing the average power. As shown in Fig. 15, both OOK and 8-PPM waveforms
have the same average power, the peak of the OOK waveform is ~ the background,
whereas the peak of the 8-PPM waveform is twice the background, making it easier to
distinguish (improving the SNR over the observation window), while conveying more
information (3-bits per symbol), at 1.5 times the bit-rate. A more detailed discussion
of M -PPM and other orthogonal formats is given in section 5.3.
2.3.7. Thermal Noise
There are two fundamental noise limitations in RF communications systems . One is

thermal noise in the receiver, which is due to the temperature-dependent random-
thermal -motion of charge carriers in resistive elements. This is often referred to as
P peak = MP . vg
8-PPM (I)
'C
.:: ....
C>
....
c
l6' -H-+-I'------H-rr:~----_i_+--t_If_L.
::E:
OOK
Time
Fig. 15. On-off-keyed (OaK) and 8-ary Pulse Position Modulation (8-PPM) waveforms are
shown for the same average power (Pav g ) and background noise level. For a given average
power, the peak power of the 1'.1 -PPM waveform increases with m, improving the ability to
distinguish the.signal from the noise.
Johnson-Nyquist noise after Johnson [215], who first measured thermal noise in con-
ductors, and Nyqui st [216], who derived a theoretical expression to fit Johnson's mea-
surements in 1928. The available thermal noise power spectral density delivered to a
matched load can be described by [216,217]
hv
N th (v, T) = h v /k T 1 ---+ ~ kBT [W1Hz], (47)
e B - v <lOOGHz
where kB is Boltzmann's constant (1.38 x 10- 23 J/K), T is temperature of the resister

in Kelvin, and the approximation kBT = -174 dBmlHz at room temperature. Note
that in contrast with shot noise, thermal noise is dependent on temperature, and for
practical bandwidths « rv
100 GHz), is independent of signal power or wavelength.
(See [20,156,218] for additional detail.)
The second noise source is received thermal blackbody background radiation,
which is also described by Eq. (47) for a single mode and polarization, where T in this
case is the average temperature within the field-of-view of the rece iver. This shared
expression for the two thermal noise sources arise s from the fact that the resistive
thermal noise can be viewed as the blackbody radiation emitted into a single mode.
The contributions from these two thermal noise sources can be combined in a single
effective temperature
= TRX + u« nb. (48)
The general expression for noise den sity that incorporates both thermal and shot
noise is given by [217]
hv
N(v, Teff) = Nth (v, Teff) + N Shot(V) = e hv / kB Terr - 1 +hv [W/Hz], (49)
which is plotted in Fig. 16 for a receiver operating at four effective temperatures,
6000K (sun) [47,219] , 290K (room temperature), 17K (liquid Nitrogen), and 2K (liquid
Helium) .
140 David O. Caplan
-150 r
J X;.:.;l~
O·_
l Wavelength, [m]
---==------=---=-- ---r~":'
i-
.- T = 77 K ( ~quld Nlro en
~ -180
o'" Microwave RXs limited by
thermal norse (keT)
~ -190
~
C.
(I)
... .;zOO
'o"
~
a.. .......... _...._.._........._......- .,....
.;z10
THERMAL NOISE h\' / .
expll1\'/(I<eT)}1 "
10 10 10 12 10 13 10" 10 15
Frequency, [HzI
Fig. 16. The fundamental limits of receiver noise density at variou s temper atures . At high fre-
quen cies (e.g., optical), quantum noise (shot noise) dominates. At low frequencies, (e.g., RF or
lower), temperature dependent thermal noise dominates. Adaptedfrom S.B. Alexander [201 .
Note that at 1.5 j.tm wavelengths, the shot noise dominated noise density, N sho t ,
is -159 dBmlHz, 15 dB larger than the thermal noise component at room temperature.
Expanding on an example by Alexander [20], consider the fundamental noise
limitations ofa 30-GHz (,\= 1ern) microwave and a 300- THz (,\= I j.tm) optical receiver,
both with I GHz bandwidth operating at operating in room temperature (290K), looking
into deep space with a conservative estimate of background temperature of70K [220].
Thermal noise dominates the RF receiver with a noise power of
Pth = kBTeffB = (1.38 x 10- 23)(360)(1 x 109 ) = 5 X 10- 12 W = -113 dBW,

(50)
whereas quantum noise dominates the optical receiver with a noise power of
P shot = hvB = (6.63 x 10- 34 )( 3 x 1014)(1 x 109 ) = 2 X 10- 10 W = -97 dBW.

(51)
Thus the RF receiver limited by thermal noise can be be 40 times (16 dB) more
sensitive than the shot-noise limited optical receiver, and this advantage can potentially
be increased to 200x and ",,2000x by lowering the effective temperature of the RF
receiver to 17K (using LN2) or "-'8K, respectively . Note that in order to achieve the
full limits of RX cooling, it must be looking directly into deep space (without channel
loss), where the blackbody contribution to the effective background temperature is ,,-,4K
[220,221]. It is important to take into account such factors when making comparisons
between RF and optical link designs, especially when nearly ideal superconducting
photon-counting detectors [33,34,210,222] are considered.
This illustration highlights some additional benefits of free-space optical receivers

over RF receivers-they can be much less sensitive to background noise, potentially
allowing for a wider range of use. By using, for example , a super-cooled optical
receiver including a narrow-band optical filter, thermal noise can be virtually eliminated
from the receiver, since the incoming blackbody radiation at optical frequencies is
significantly lower than that of shot noise for practical operational temperatures (i.e.,
nb nb
< 3OOK). Even when pointing directly at the sun, where =",,6000K (assuming
the telescope doesn't melt), the blackbody noise is still "" to dB lower than the shot
noise (at I ILm)2 for a single-mode receiver (single spatial-temporal and polarization
mode), degrading a shot-noise limited signal by only ",,0.5 dB (see [47,223] fordetailed
analysis of background radiation and its impact on free-space optical links). On the
other hand, for an RF receiver, this would degrade performance in excess of 13 dB.
This illustrates potential benefits of using true single-mode optical receivers. While
coupling into a single spatial mode can be more challenging, the reduced background
due to strong spatial-filtering and available low-loss matched spectral filters can enable
operation that is insensitive to background noise levels.
2.4. Example Sensitivities and Link Budget
Based on the fundamental noise limitations described above, Table 4 shows a rudimen-
tary link budget comparing Optical and RF free-space performance over a 40,000 km
link (e.g., distance to geosynchronous orbit) with the same transmit power-aperture
product.
Table 4. Comparisonof simplified FSO and RF link budgetsbasedon fundamental noise-limited

receiver sensitivities for a common I-GHz RX bandwidth (at room temperature), 0.3-m TX
antennadiameterand a 1.5-m RXantennadiameter, and 40,000 km linkdistance(Earth to GEO).
Adaptedfrom S. Alexander(20).
Item A = 1 x 10 6 m A=3xIO 2 m Units

Transmitter
I) Avg.TX Power 0 0 dBW
2) TX Losses -2 0 dB
Channel
3) Space DiffractionLoss -41 -131 dB
Receiver
4) RX Loss -2 0 dB
Powerat Comm. Detector -57 - 131 dBW

RX Sensitivityat I Gbps -97 -114 dBW
Link Margin 50 -17 dB
Given these constraints, and some practical implementation considerations, the

overall optical link has a 67-dB advantage , largely attributed to the optical carrier
2 Note. at I ILm. the kT approximation no longer holds. and blackbody rad iation need s to be calculated with
the exact expre ss ion on the left-hand side of Eq. (47) .
142 David O. Caplan
size, and corresponding spectral, temporal, and spatial advantages. Furthermore, the
lOOOx bandwidth advantage of FSO systems, which can be used to scale data rates
or improve RX sensitivity [22], is not accounted for in the basic accounting in Table
4. A more detailed link analysis which incorporates realizable modulation and coding
performance suggested in Fig. I is given in section 5.5.
It's important to note that some of the RF deficit can be improved by increasing the
RF aperture, which is practically easier due to reduced wavefront quality requirements.
Also, RF systems usually have better transmitter efficiencies, providing another "'3-
7 dB benefit, but ",60 dB is still an enormous shortfall to overcome . For instance,
if both TX and RX aperture diameters were increased by a factor of ten, this would
improve the RF link by 40 dB, still leaving a factor of 100 (20 dB) deficit. Thus, the net
power efficiency advantage combined with the fundamental ability to efficiently scale
to higher data rates, make free-space laser communications a promising technology
for future long-haul communication links.
3. Transmitter Technologies
Factors that impact optical communication transmitter performance include the fol-
lowing: modulation bandwidth and extinction ratio, waveform generation and fidelity
(control of shape and parasitic chirp), average and peak output power, output po-
larization state, wavelength, spatial profile and perhaps most important, wall-plug
(electrical-to-optical conversion) efficiency (Joules required per photon transmitted).
In contrast with transmitters for fiber-optic networks, in which the average and peak
power that can be delivered over the fiber channel are generally limited by channel
nonlinearities, the FSO link has no such constraint. While there are power limitations
within the FSO transmitter, they are several orders of magnitude higher than those
used in fiber networks.
Present lasercom transmitter designs incorporating high-power optical amplifiers,
can reliably output in excess of lOW average and", I kW peak power in a single spatial
mode with little waveform distortion or power loss due to nonlinear spectral broadening
[21,198] with potential to increase to 100 kW levels [224] . The lack of dispersion over
the FSO channel allows for more flexibility in wavelength selection and greatly reduced
need for spectral efficiency. There are '" I0 THz of optical bandwidth in 1.5 J.tm Erbium
band [92,93,126], and other bands are available for the FSO link such as 0.5 and 0.8
J.tm [81-83,98-10 I ,225],0.98 nm [226], 1.06 J.tm [1,2,223,227-230], 1.3 and 10.6 J.tm
[223]. While still subjected to practical availability and reliability constraints, the FSO
transmitter and receiver designer has considerable leeway in optimizing the overall
link efficiency,
1]link = 1]TX 1]ch 1]RX [J /bit], (52)
the Joules-transmitted per bit-received. Here, 1]Tx is the transmitter (electrical to op-
tical) efficiency [J/photon], 1]ch is the net channel loss, and 1]RX is the net receiver
efficiency [photons/bit]. While the TX contribution 1]TX is accounted for explicitly in
the efficiency with which it can generate the photons, the TX indirectly influences chan -
nel loss, which is wavelength dependent, and receiver sensitivity, which is dependent
upon modulation format and TX waveform fidelity.
3.1. Direct Modulation and Semiconductor Laser Sources
Present-day semiconductor lasers are robust, compact, efficient sources of coherent

light. For TeIcordia-qualified laser diodes [4,5], the mean-time-to-failure (MTTF) can
exceed 100 years ('" I x I06 h) [231]. A practical feature of semiconductor lasers is
that the output power can be directly modulated, at speeds approaching the relaxation
oscillation frequency [232,233], fRo , which is given by
(53)
where cln is the velocity of light in the semiconductor material, dgldN is the dif-
ferential gain, S is the photon density in the cavity, and Tph is the photon lifetime . At
output power levels of 10-25 mW, fRo can exceed e- I0 GHz, which is particularly
useful for subcarrier-multiplexing many narrow-band MHz class analog signals onto a
single laser source, a capability that is commonly used by the cable television (CATV)
industry [234,235] . However, for wide-band digital applications, direct modulation
rates are practically limited to lower values, e.g., a few GHz, by parasitic capacitance
in drive electronics or chirp-induced penalties [236].
As discussed in section I, one of the simplest forms of digital modulation is on-off-
keying (OaK), in which the logical information is imparted on the optical carrier by
switching the light on and off. While direct modulation is an easy means of applying the
intensity modulation suitable for OaK, there is a residual wavelength (or frequency)
modulation or chirp (d¢/dt) due to fast current density fluctuations and a slower
temperature dependence ofthe refractive index in the active layer which causes a shift
in the laser wavelength [233,237] . In fact, the fast change in laser frequency can be
used for directly generating FM waveforms [160,162, I77,178,180,238]. For the case
of large frequency deviations, the residual 1M due to current changes may need to
be compensated. This can be achieved via dynamic compensation in the time domain
[239] or relatively static gain/wavelength flattening in the frequency domain.
Chirp is often defined in terms of the o-parameter [240], which is the ratio of
amplitude modulation to phase modulation, where
0: = ~~ (.1 ~E) = ~~ ( IldI) ,

E dt 21 dt
(54)
and I, E and </>, are instantaneous intensity, field amplitude and phase.
The chirped-output waveforms from a direct-drive laser have a broadened spec-
trum, that can be much larger than the modulating signal spectrum, i.e., the waveforms
are not transform limited' (see, e.g., [236]). An example of a chirped waveform is
shown in the spectrogram in Fig. 17.
The spectrogram is measured using a time-resolved technique [241-244] with time
and frequency resolution of lOOpsec and 5 GHz, respectively . A direct-drive ",75 rnA
current pulse causes a ",20 GHz frequency shift in the laser center frequency, yielding
a fast wavelength tuning response ds ] di ac = - 280 MH7JmA. In addition, transient
3 The transfonn limit (or Fourier limit) , can be described as the minimum pulse duration which is possible
for a given optical spectrum of a pulse. A lransfonn-limited pulse has no chirp , a minimum time-bandwidth
product , and a spectrum that is the Fouier-Transfonn of the pulse envelope .
144 David O. Caplan
. .==========:
. . . . . . ......................... .:.:.==..:-..:;:.:;=;;::: : :""'=..============J
C• • • • • • • • • • • • . • • .•• - • • • • •• ••• • - .• •• • • • (dB)
_ _-.-.--.
.- ~_ . • . • . _• . c::§:> M e .1 S lll tH I I Il I f' II ~ lty C OIIIOIlIS
';l-40
:I:
(a) ~ . 30 ~
:- ~~ ~ ~
/r:..- ' . ,
I:::~,~ ~f~~~~~c~z~7?~~~ :il:~~i • • •

II.
~ :f ~i =>smA ··· !·/ ·..··.. ·.·E ········· · · · · · · j

100
(b)
-0.5 o 0.5
Time, [n sec]
Fig. 17, (a) Measured time-frequency spectrogram (top view) of a directly-driven DFB laser
biased at 25 rnA, about 5 rnA above the ",20 rnA threshold current. Contours represent lines of
constant power in dB, with the peak normalized to 0 dB. Due to the low bias current, the laser is
operating near threshold, yielding relatively good e- 13 dB ER (b) Measured input drive-current,
with ",400 ps pulse width and ",75 rnA peak current change . During fast current transitions,
spectral components are generated that far exceed the transform-limited waveform spectrum,
especially at the leading edge-which produces the characteristic 'h ' pattern above. This effect
grows along with timing jitter on the leading edge as the laser bias approaches the laser threshold .
spectral generation, especially during the leading edge in semiconductor lasers biased
near threshold, can be very wide band, exceeding 100 GHz . This leads to increased
dispersion penalties in fiber-optic links and can limit receiver sensitivity in nearly-
quantum-limited free-space links [245,246] . Furthermore, since chirp is a function
of modulation depth, communication links which employ direct modulation need to
balance conflicting extinction ratio and dispersion penalties .
In fiber-optic links, such considerations practically limit the extinction ratio (ER)
to rv 10 dB, which in tum can degrade RX sensitivity with rv3 dB penalties in pream-
plified RXs [see Eq. (64) in section 3.5.2], and limit modulation rates to less than
2.5 Gbps [233,247]. Since transmitter cost is generally a bigger concern than perfor-
mance in telecom applications, directly-modulated lasers have widely been used for
lower-rate « rv 2.5 Gbps) links. But in FSO links, receiver sensitivity is often a design
driver, and due to the limitations noted, conventional direct modulation is generally not
considered the preferred approach. Channel capacity and bit-error-rate (BER) perfor-
mance are degraded significantly by poor ER and chirp-induced spectral broadening,
which increases the mismatch penalty between the TX signal and RX filter, a subject
discussed in more detail in section 5.1.
3.1.1. Spectral Shaping
Direct-drive laser chirp and extinction ratio characteristics can be modified via sub-
sequent optical filtering, a process that converts the frequency-shifted laser output to
intensity modulated waveforms. When followed by an average power limited ampli-
fier, this combination can be used as a power efficient source of high-fidelity transmit
waveforms [10] despite the loss of the filter (see section 3.5.5). Reduced chirp and
Laser communication transmitterand receiverdesign 145
Unfiltered las er~

output --+r 0 Direct-
o
,-,,~dcurrent
rive
0
o
J
"f D.:~ : I pulse
~D··-V
i:::t
z 03~ ·-~~~~-.1-;r-.~...~. _-o_ -.~-...-Z/ Time, [nsec)
!J. Freq , [GHz] 0 10 .1
Fig. 18. Measured time-frequency spectrogram (oblique view) of a directly-driven DFB laser
biased at 75 rnA, with a ",20 rnA threshold current. Time and frequency resolution are approx-
imately 100 ps and 5 GHz, respectively. The response to a I-ns, ",75 -mA current pulse starting
at time=O, increases output power by 2.4 times (",4 dB) and shifts the laser center frequency by
20 GHz.
improved ",13 to 14 dB ER have been demonstrated using this approach [244,248].

More recently, direct-drive chirp-managed lasers (CMLs) [249], which incorporate pe-
riodic filters into the laser package, have demonstrated impressive 10 Gbps long-haul
fiber-communication performance without dispersion compensation [250,251] . Due to
the complex interaction of pulse shape, dispersion, nonlinearities, and power launched
over the fiber channel, these demonstrations showed better overall link performance
with poor ",8 dB ER. However, these results are contrary to the impact of ER over the
FSO channel, where good ER, generally > 30 dB is necessary to avoid performance
penalties, especially in low-duty-cycle applications.
With narrower optical filtering, however, these approaches show promise for
achieving low-chirp high-ER performance with compact power-efficient direct drive
lasers, especially when used in conjun ction with a subsequent TX amplifier (a sub-
ject discussed further in section 3.5.5). As shown in the spectrogram in Fig. 18, the
(unfiltered) '" I ns laser output has a ",4 dB ER and a ",20 GHz of spectrum. After
narrow-band optical filtering, the ER is improved to ", 30 dB and the spectrum is re-
duced to ",5GHz. As the external optical filter spectrum approaches the match to the
desired signaling waveform, nearly transform-limited waveforms can be generated.
Another limitation with direct-drive laser modulation arises due to the non-uniform
FM response that results from competing thermal and carrier modulation effects
[238,241,252-256] that can lead to waveform distortion and pattern dependent er-
rors. This can be compensated by pre-equalization of the modulation signal with a
passive network in some cases [160,177,238]. However, this solution reduces the FM
response and consequently leads to increased drive requirements. Another means of
mitigating these effects is to limit the low-frequency drive content through pattern
coding/scrambling [160,162,248] , which can be achieved by driving the laser with
low-duty-cycle waveforms such M -ary PPM or FSK modulation formats (see Fig
9). In addition , for low-duty-cycle waveforms , directly-modulated DFB lasers can be
used in cascade with an external modulator and/or narrow-band optical filters to achieve
146 David O. Caplan
time/frequency windowed direct modulation. This can be a power-efficient technique

to obtain high modulation extinction, while maintaining stable wavelength and near-
chirp-free waveforms during transmission, a subject discussed further in section 3.5.2.
3.2. Semiconductor Laser Structures
There are several semiconductor laser design structures that are commonly available
today. Fabry-Perot lasers have the simplest structure, but usually output multiple lon-
gitudinal (frequency) lasing modes, and therefore, are not well suited as signal sources
for high-performance FSO links. But they are well suited as pump laser sources for
commonly used Erbium-doped fiber amplifiers (EDFAs), Ytterbium-doped fiber am-
plifiers (YDFAs) and codoped Er:Yb EYDFAs. Here the increased spectral bandwidth
is not a liability owing to broad absorption bands in the gain media [126] and since
the effective reduction in spectral density lowers the risk of SBS penalties [110, III] .
In order to provide some spectral selectivity and stability, Fabry-Perot-based pump
lasers are often used in conjunction with narrow-band reflectors that are part of the
laser cavity, or external stabilization filters such fiber-Bragg-grating (FBG) reflectors
[257-259] to provide feedback that efficiently lock the center of the lasing spectrum
(composed of 5 to 10 frequency modes) to that of the FBG.
Semiconductor lasers with integrated distributed Bragg reflectors (DBR) have
been used to provide agile, tunable, single lasing modes in simultaneous time-division-
multiplexed and frequency-division-multiplexed (TDMlFDM) digital communications
[180,260]. In such lasers, great care needs to be taken in the control of the drive current
and chip temperature in order to provide accurate and stable wavelength control to
prevent mode-hopping. The use of such tunable sources is gaining traction in the
telecom industry for flexible reconfigurable WDM networks, and especially since a
single widely tunable laser could be used as a spare for any other channel.
For stable single-frequency operation , distributed-feedback (DFB) lasers are most
commonly used throughout the telecom industry. A representative COTS DFB laser
is shown in Fig. 19. The DFB laser has an index grating written into the gain medium
which allows for only one lasing frequency mode within the gain-bandwidth of the
laser. This typically provides a side-mode-suppression ratio (SMSR) exceeding rv45
dB, which precludes mode-hopping to other Fabry-Perot resonances. The DFB laser
linewidth (Llv) is a function of output power and can be influenced by external reflec-
tions [233,261-263]. Typically Llv <rv I MHz for well-isolated 20 mW DFB lasers,
making them well suited for DPSK communication rates exceeding rvI Gbps (see sec-
tion 5.2), since coherence length penalties are incurred as the laser linewidth exceeds
rv I % of the bit rate" [264].
4 Linewidth penalties for DPSK are negligible when the interference between the differential bits is complete.
This occurs when the coherence time of the signal is sufficiently long, i.e., T c » T d , where the laser
coherence time T c= II Llv, and Td, is the time delay between the differentially encoded bits. This ensures
that coherent interference is achieved over the entire bit duration . Typically, DPSK data are encoded using
adjacent bits, so that Td =Tb, the bit period , with bit rateR= II Tb. However, DPSK data can be differentially
encoded with a multiple -bit delay, e.g., Td = mTb, where m is a positive integer. For this general case
Iinewidth penalties are avoided when R » mLlv , and thus, multi-bit differential-encoding imposes
stricter linewidth requirements.
BUFFER
CORE e-:
Fig. 19. Typical semiconductor DFB lasers come packaged in a 14-pin butterfly package with
built-in thermo-electric cooler (TEC) and thermister for temperature control, and an internal
optical isolator for increased stability. Since most external modulators requirea polarized input,
DFBs often come with polarization maintaining (PM) output fiber. In applications where tight
wavelength stability is required, an internal etalon andmonitor diode canbe included within the
package toprovide anindependent measure of wavelength that canprovide sub-GHz wavelength
stability when combined with external feedback [265).
3.3. Laser Wavelength Control
For optimum communication performance, narrow optical receiverfilters that are well
matched to the transmit signal spectrum are used (see section 5), requiring precise
wavelength alignment of TX signal and RX filters. This can achieved via RX track-
ing of the incoming signal, but this approach is not always desirable from the system
perspective. For instance, providingopen-loop TX and RX control speeds acquisition
and enables an FSO system to coast through fades, without having to re-acquire wave-
length. In WOM applications, tight control ofTX wavelengthis valuable for both FSO
and fiber-optic links, allowing for closer channel spacing, which improves spectral
efficiency and overall link capacity. Precise open-loop control of the transmitter sig-
nal wavelength can provide additional benefits in WOM-DPSK links by facilitating
acquisition and enabling cost-effectivemulti-channel OPSK receiverdesigns [266] as
discussed further in section 5.2.
The steady-state laser wavelength is dependent on the optical path length within
the laser cavity,which is a functionof both laser temperatureand averagedrivecurrent.
At 1.55 p,m, these relations are given by
d)" ::::: 0.1 nrn = 12.5 GH z (55)

dT °C °C '
d)" ::::: 0.004 nrn = 0.5 GHz . (56)

didc rnA rnA
For a ± lOoe range with lO-bit control (corresponding to ±o.ol oe precision), the
typical OFB laser can be temperature tuned over a range of ± Inm (± 125 GHz at
1.55p,m) with ± 125 MHz resolution. Similar resolution can be obtained with current
control betterthan ±0.25 rnAas shown in Fig. 20. Notethat the wavelengthdependence
on drive current is complex (see, e.g., [233,238,253,256]), having fast components
dependent on current density and slower components that are a function of carrier and
thermal diffusion resulting from current transients.
148 David O. Caplan
1558.0 Tr:::r:::;:::::=::===.===:::==::===
- Wa velength at 22 C
= , - - - - - - - - - - - - ---,
y • 0 .0035x + 1556.8
--- Wa velength .. t 2 6 C
R 2 ·O.994
--- W a v e le n g t h a t 1 8 C
___ Wa vele n gth 3t 30 C
- L in Fit av e le n h at 3 0 C
1556.5 +------:--= ".-__.dII:~!:=::=---------__=",...t..dIl::::::==---~
1556.0 L....."f4:;:::f!~~~::::::-.- t__+_t__+_t__+_t_+_1t_+_1"+_t"+_t"+_t_+_t_+_t....... _t_+....... j

50 70 90 11 0 130 150 170 190 210 23 0 2 50
B ias Cu rrent, [rnA]
Fig. 20. Steady-state DFB laser wavelength measurements as a function of drive current for
various temperature settings. The linear fits to the data yield a current dependence of d>'jdidc=
0.0035 nmlrnA or ",440 MHz/rnA and a temperature dependence of 0.14 ml°e. Note that this
value is of larger magnitude and opposite sign of the fast wavelength dependence on current
d>.jdiac = -280 MHz/rnA observed in Figs. 17 and 18.
Over the long term,Telcordia-qualified cars continuous-wave (CW) DFB lasers

have demonstrated sub-GHz wavelength stability and repeatability over years of op-
eration in the field. Precise temperatureand current control combined with occasional
feedback to correct for potential long term drifts in laser wavelength tuning charac-
teristics s>.( i , T), could enable >- I()() MHz class wavelength tuning over the ,....,25+
year life of the laser. Such characteristicsare useful since they allow for the option of
open-loop TX wavelength alignment to RX filters, a capability which is particularly
desirablefor FSO links,especiallythoseemployingRX interferometers (e.g., DPSK-
see section 5.2), where the need for spectral acquisitionand trackingcan significantly
increase overall acquisition times.
The cost of DFB laser wavelength tunability includes two digital-to-analog con-
verters (DACs) for adjustablecurrent and temperaturecontrol, a feedback mechanism
(e.g., a wave meter) to correct for long-term drifts, and the electrical power needed to
drive the TECs to maintain a stable temperature. However, most applications require
some degreeoflaser wavelength control, especiallythoseemployingDWDM, so some
form of fixed current and temperaturecontrol is implementedanyway. The only addi-
tional cost for tunability becomes the additional DACsand the feedback mechanism.
In addition to using calibrated current and temperature to control DFB laser wave-
length, integrated etalon-based wavelength monitors are commerciallyavailable with
Telcordiaqualification [5], providingcompact, reliable, locally-resident sub-GHz res-
olution feedbackthat can be used to align the laser wavelength over a 20+ year lifetime
[231.265]. In some cases, the feedback information on laser wavelength (and power)
can be remotely located, for instance at the receive terminal,and periodically be com-
municated back to the laser control via low-bandwidth back channel [266-268] (see
section 5.2.2).
0.3 --------------------- --'--------------'-1

1550nm?OmW La;ser Diode:, Temper~ture Fixe~ at 25°C .
. . .. .. ..
0.2 ··· ·· · · ·· ·· ···t· . . ····· ··· · · · ·· :a ·· · ········ · · · ..;· · · ·· · ·· · ·· ···· i· ·· ··· ·· ·· ····· ,······ ·
~ 0.1
..,j
c:
~
:s
o
0.0
o
W
I- -0.1
-0.2
y = 0.0117x • 0.2152
R2 = 0.9988
-0.3 -t--'f-t--t-t-+--f-+-t-+--+-f-t--t-t-+--f-+-t-+--+-f-t--+-+-+-+-+-t-t-i
o 5 10 15 20 25 30
Ambient Temperatu re , [Degrees C]
Fig. 21. TEC current versus ambient temperature for several drive currents while the laser is
fixedat 2S°C. The measuredlinearfitsover the ± ISoC rangecorrespond to a TEC current slope
diJ1c : : :12 rnA/oC and d~rL~c :::
0.4 rnA/rnA. The 'zero-TEC-current' temperature is about
IS°C.
As shown Fig. 21, The TEC power can be minimized if the laser is selected to
output the desired wavelength near the 'zero-TEC-current' point, which is rv15°C in
this case. The TEC current required to change the laser temperature , or equivalently,
compensate for fluctuations in the ambient, is rv12 mN°C. Over a range of drive
currents, at 15°C ± 5°C ambient temperature, the TEC current is bounded by about
100 rnA, which for a 3-V power supply limits the power requirement to < 300 mW.
If power is not a driver in the design, TECs can be used to stabilize to temperature
differentials up to rv40°C, although this may consume several watts of power. Note
that the zero-TEC-current ambient temperature is about lOoC below the set point for
the laser in Fig. 21, due to self heating from the laser drive current. This differential is
typically in the range of 5°C to 10°C, dependent on the optical-electrical conversion
efficiency of the laser.
3.4. Cavity-Dumped and Q-Switched Lasers
Cavity-dumped and Q-switched laser sources are capable of efficiently generating

extremely high peak-power levels (e.g., tens of MW), and are especially useful for
generating periodic pulse streams, used in Laser Radar (LADAR) and Laser Imaging
Detection and Ranging (LIDAR) applications [269-273]. However, they are not partie-
150 DavidO. Caplan
To cha nnel
Data
Fig. 22. Masteroscillator poweramplifier(MaPA) transmitter, consistingof a modulated laser

source followed by a poweramplifier. For high-rate applications, the MaPA is often usesa CW
laser source and externalmodulation.
ularly well suited for high-sensitivity laser communications due to limited bandwidth
and modulation options , difficulty in pulse shaping for optimized receiver sensitivity,
and timing jitter [227], all of which tend to degrade communications performance. For
these reasons, the MOPA transmitter discussed in the next sSection is most commonly
used for high-rate-high-sensitivity communications applications.
3.5. Master Oscillator Power Amplifier (MOPA)
The Master Oscillator Power Amplifier (MOPA) transmitter shown in Fig. 22 is a

flexible and scalable approach, compatible with both TDM and WDM communications
at Mbit/s rates to tens of Gbit/s and beyond.
The MOPA design can be modular, allowing for a more flexible choice of wave-
forms, and independent design and optimization of the laser, modulator, and power
amplifier. The design is commonly used for high rate optical communications in the
telecom industry and is well suited for free-space lasercom platforms, where high-
fidelity waveforms are required .
3.5.1. Modulation
As discussed in detail in section 3.1, direct-modulation of laser current can be used

as a relatively simple method of generating a modulated optical signal at low rates. In
order to transmit at high rates (e.g., 10 or 40 Gbps) or use more sophisticated mod-
ulation formats with carefully tailored pulse shapes (e.g., M-PPM, RZ-DPSK, etc.),
external modulators that can generate nearly transform-limited chirp-free waveforms
are generally used. Two kinds of modulators are commonly used for high-rate inten-
sity modulation : Mach-Zehnder modulators (MZMs), which are interferometric based
devices ; and semiconductor-based electro-absorption modulators (EAMs) (see, e.g.,
[274-277]). EAMs are compact, offer ease of integration with other elements such
as lasers and semiconductor optical amplifiers (SOAs), and can be cost-effective for
wide-band intensity modulation transmitters. While EAMs are usually very wide band,
they can introduce parasitic chirp, and often have high-loss (5-10 dB) and poor ex-
tinction at practical drive voltages. Therefore EAMs are generally not the first choice
for use in high-sensitiv ity FSO applications.
MZMs on the other hand, are extremely versatile wide-band devices, able to mod-
ulate both intensity and phase at rates exceeding 40 Gbps. They also can be used to
generate a controlled amount of chirp dependingupon the MZM design [240], which
can be used to pre-compensate for the effects of dispersion [278-280]. In addition,
MZMs can be used to make high-rate high-fidelity pulse-carved return-to-zero (RZ)
waveforms that facilitate optimized (nearly matched) receiver design (see section S)
commonly used in series with data-modulatingMZMs. For these reasons, MZMs are
commonly used in high-performance fiber-optic and FSO links.
3.S.2. Mach-Zehnder Modulator(MZM)
The Mach-Zehnder interferometer structure (Fig. 23) named after its two inventors''
[281,282] is commonlyused in many optical devicesincluding modulators, DPSK re-
ceivers, and wavelength interleavers. ActiveMZMsare mostoften fabricated in lithium
niobate (LiN0 3) waveguides (see, e.g., [283-290]). Toimprovefabrication yields,they
are typically constructedfrom y-junction couplers so that there is usually only single
input (Ein 1 ) and output ports (Eo llt 1 )-with the secondary input and output ports be-
ing unguided waste ports. MZMsare also fabricated in semiconductormaterialssuch
as GaAs [143], InP [29 I ,292], Si [293,294], and optical polymers [29S] . For ultra-fast
applications, nonlinearinterferometric switches constructedfrom fibers, semiconduc-
tor optical amplifiers, and other nonlinear materials [296-302], have demonstrated
100 GHz class all-optical logic, wavelength conversion, and clock-recovery. Non-
interferometric switching has also been demonstrated using self-phase modulation
(SPM) and four-wave-mixing (FWM) [303-30S] nonlinear processes.
The basic MZM geometry is shown in Fig. 23, composedof two directional cou-
plers (ideally SO/SO), the first splitting an incident electric field (e.g., Eind into two
equal top and bottom propagating waves.
The two fields are then recombined in the second SO/SO splitter after experiencing
controlled phase delays <Ptop and <P bottom . The two outputs (Eout, and EoUt2) are
determinedby the coherentaddition(interference) of top and bottomfields incidenton
BS2 .The interference isa function of theiramplitudes and relativephase (11</> = </>top -
</>b otto m ) ' Chirp-free intensity modulation is achieved for balanced field amplitudes
when the MZM operates in push-pull mode, with </>to p and </>bot to m having equal
magnitude and opposite sign [240], which removes all residual phase modulation.
The MZM time-domain field impulse responses are given in Eqs. (S7) and (S8)
for the two MZM outputs,
Eoutl(t) = v'R 1R28(t + T/2) + v'T1T28(t - T/2)

1
= 2 [8(t + T/2) + 8(t - T/2)];deal' (S7)
Eout2(t) = v'R 1T28(t + T/2) - v'T1R28(t - T/2)

1
= 2 [8(t + T/2) - 8(t - T/2)];deal' (S8)
where factors of constant phase have been dropped, and the ideal impulse responses
occur when BS1 and BS2 are SO/SO splitters, i.e., R 1R2 = T 1 T2 = 1/4 and R 1 T2 =
T 1 R2 = 1/4. These correspond to the sum and difference of two delta functions
5 L. Mach and L. Zehnder inde pendently invented in 1891/1892 what has becom e known as the Mach -
Zehnder interferometer which could monitor changes in refractive index, and hence density, in co mpressible
gas flows
152 DavidO. Caplan
100% ~_V_tlJl ..----

/I - ~op - f(VloJ
I
I
Vb~om
•••••••••••••••• ~ ••• • • • 1000/0
Yo
:
BS 1(R"T1 )
I
I
~oll om - f(VbollO,J -=-
i Ein2
Fig. 23. The basic structurefor a Mach-Zehndermodulator consistsof two beam splitters 851,2
with intensity reflection and transmission coefficients R1,2 and T 1,2, respectively. Ideally. the
851,2 are 50/50 splitters, i.e., R1,2 = T1 ,2= 50%. The phase difference between the two arms:
.:1¢i = ¢itop - ¢ibottom. determines the output splitting ratio based on the constructive or
destructive interference of the top and bottom arms at 852 . For electro-optic MZMs, the phase
change is proportional to the applieddrive voltages Vt o p and Vbottorn '
separated by time difference r , In the frequency domain, the Mach-Zehnder outputs

are periodic sinusoids given in Eqs. (59) and (60):
EOut1(f) = (JR1R2ej-,r!T + JT1T2e-i1r!T) = cos(-rr!r)ideal, (59)
Eout2(f) = (JR1T2ei1r!T - JT1R2e -i1r!T) = sin(1I"!r) ideal. (60)
The sinusoidal MZM field and intensity transfer functions for EOuh(f) and
IOUtl (f) = IEOUtl (fW are shown in Fig. 24, and highlight many attractive fea-
tures. The voltage required to generate the 11" phase shift is referred to as the switching
voltage, V1r , and corresponds to the voltage required to go from an intensity transmis-
sion peak to a null. By biasing the modulator at the 50% transmission point (referred
to as the quadrature point), the MZM can generate intensity modulation (1M) wave-
forms when driven with a differential voltage LlV = ± V1r /2, as illustrated for the
superimposed OOK and PPM drive waveforms.
While it is not a requirement to operate the modulator from transmission null all the
way to the peak as shown, doing so maximizes the optical transmission and reduces the
impact of waveform distortion such as transient ringing on the input electrical signal.
This is due to the nonlinear cos\) transmission characteristics which tend to compress
fluctuations at the transmission peak and null, and steepen transitions in between.
MZMs can also be used for digital phase modulation . By driving the MZM around
the transmission null, the sign of the electric field output transitions from positive
to negative, which corresponds to an abrupt 11" phase change suitable for generating
DPSK and DQPSK waveforms. These digital characteristics offer advantages over
Laser communicationtransmitterand receiverdesign 153
.....
·1.S ,.
Fig. 24. MZM transfer function for the cost) arm versus differential drive voltagenormalizedto
V 1f, the switchingvoltage. Superimposed are digital modulation drive waveforms (drive voltage
versustime) for OOK, PPM, and DPSK formats. whichare achievedby simply varyingthe drive
bias position and amplitude.
linear phase modulation in which energy is transmitted during the phase transition (as
the modulator slews between 0 and 1fp, which degrades the SNR. In order to extract
maximum phase-modulated output power, the MZM is biased at a transmission null
and is driven between two successive transmission peaks, requiring a drive voltage of
± V"' twice that oflM, corresponding to four times the drive power. As in the 1Mcase,
digital phase modulation can be achieved using a small drive voltage around the null
to conserve power, at the expense of increased insertion loss and waveform fidelity
that is sensitive to the input drive waveform . Furthermore, for non-ideal MZMs, this
can also lead to increased chirp penalties for differential phase-modulation formats .
3.5.2.1. MZM Phase Elements
The time delay T in Eqs. (57)-(60), has components of distinct origin
T = Ts t a ti c + Tr and om + Tdrtve , (61)
which map to optical phase, ¢( >') given by
¢ ( >.) = ¢stat ic + ¢ ran do m + ¢ d r ive = T ( 21f

>'nC) [radians] . (62)
The static delay, ¢static, corresponds to a fixed difference between the two arms
due to manufacturing, and is typically many optical periods and much larger than the
other components. Thus, Tstatic determines the interferometer periodicity commonly
referred to as the free-spectral range (FSR), where
FSR = l /Ts t at ic, (63)
Modulation bandwidth is limited to a small fraction of the FSR, and therefore the
design target for Tstatic is typically near zero, which enables a wide range of operation.
For example, if Tstatic « 8 ps, the modulation bandwidth is > rv 125 GHz, or I nm at
154 DavidO. Caplan
1550nm wavelengths. Note that this time is long relative to the optical period, corre-
sponding to many hundreds of cycles of the ",5 fs carrier. For a modulator to operate
simultaneously over the entire EDFA C-band (i.e., 1535-1565nm), Tstatic « 0.2 ps,
which can be achieved in practice with precision manufacturing. This is important for
wide-band WDM applications in which a single modulator may be used to modulate
and/or shape (pulse carve) many channels (see, e.g., [93,138]) .
3.5.2.2. MZM Bias Control
The MZM random phase term, <Prandom, represents a slow drift (>'" I s to > I day time-
constant depending on the mechanism) in the relative phase between the two modulator
arms . This is a welI studied phenomenon that can be attributed to mechanisms such as
thermal gradients [306] and charge buildup [307]. Consequently, closed -loop drive bias
control is generalIy required to maintain the proper bias position while compensating
for modulator drift [180,308] . Advances in MZM designs have significantly reduced
the amount and rate of drift [309-313], but even smalI drifts can degrade extinction
ratio which is important for low-duty-cycle applications [2, 11,15,21,314] and lead to
deleterious chirp in DPSK applications [132,142]. As long as the drift rate remains
slower that the closed-loop control bandwidth, and is contained (within a few 7r), the
random phase can be compensated indefinitely with closed-loop control (see, e.g.,
[315-317]) which applies a voltage typically up to a few V" to counter the drift over
a range of several interferometer fringes. If the random-phase drift continues to move
in one direction beyond the range of the control loop, a reset is required, in which the
control loop jumps back to the center of its operating range and re-optimizes the bias
position. Such resets can be completed within >- I ms, depending on the speed of the
control system, during which the modulator is uncontrolled. With present modulator
designs, such resets are becoming less frequent, and as long as the average reset rate
is > I hour this corresponds to better than 99.9999% availability. Present modulator
designs have estimated reset-free operation times exceeding 20 years [311,313,318] .
Further, since such resets can be anticipated as the bias slowly approaches a rail, steps
can be taken to mitigate the effects of infrequent reset outages .
3.5.2.3. Extinction Ratio (ER)
Modulator extinction ratio (ER), defined as the power off-to-on ratio, impacts commu-
nication performance in many ways. It determines the initial distance between logical
'0' and 'I' power levels in OaK modulation and leads to multiple interferometric
cross talk [319], which adversely impacts both TDM [320] and WDM systems . In-
complete ER can significantly degrade the sensitivity in opticalIy preamplified links
[15,20,314,321 ,322]. For ideal optically preamplified OaK, a slight modification of
the upper-bound estimate for the SNR penalty given in [20], yields
Ll =
I
1 - 2ER
1 + 2ER + 2VER
(1 +~
3
JER) ' (64)
which is accurate to about ± 0.2 dB for all SNR levels. This corresponds to a "'0.7 dB
penalty for a 20-dB ER (i.e., ER =0.0 I) and ",2.8 dB for a 10-dB ER. In addition , poor
ER limits the dynamic range of data-rates for variable duty-cycle multi-rate optical
communications and transmitter power efficiency low-duty-cycle modulation formats
such as M-PPM [1,10,11 ,15,21,22 ,198,214,314], due to "power robbing" from the
signal at the transmitter output, (discussed further below and in section 3.5.3). Such
low-duty-cycle modulation formats are particularly attractive for use in high-sensitivity
FSO applications.
In order to have complete extinction, the fields interfering at BS2 must have equal
amplitudes, same polarization and 1f phase difference . For this reason, MZMs with
high extinction typically have polarizing elements which force the interfering waves
to be co-polari zed. For non-ideal interferometer splitting ratios, the ER is given by
2
p-1
(65)
ER= ( 2(p+ 1) ) ,
where p = R lR2/(TlT2) for EOUtl and p = R lT2 /(TlR2) for EOUt2 in Eqs. (57)-
(60). For perfect extinction p = I. A 20-dB ER results from p = 2/3, which can be
achieved with BSl = 50/50 and BS2 = 60/40 coupling ratios. For BS2 = 47/53 , a 30-dB
ER can be achieved.
Since common telecom specifications for ER requirements are only>- 13 dB, there
are few vendors that market good-ER (ER < -30 dB) MZMs, despite the fact that such
devices should be relatively easy to produce given the relations above. Nevertheless,
commercially-available MZMs typically have ER better than -20 dB (at frequency) ,
and MZMs with internal high-extinction polarizers have demonstrated modulation ER
< -45 dB individually and ER < -60 dB in cascaded operation [15]. By actively
balancing the input powers to the MZM output coupler, a commercial MZM was
configured with ER = ,....,-70 dB using a single modulator [323]. Note that for ultra-
wideband applications (e.g., >,....,30 nm), modulation bandwidth and ER may also be
limited by coupler/waveguide wavelength dependence.
For low-duty-cycle waveforms, windowed direct modulation may be used as a
power efficient technique for achieving high modulation extinction , while maintaining
stable wavelength and chirp-free waveforms during transmission as illustrated in Fig.
25. In this manner, the power required for driving the laser and external modulator can
potentially be reduced by a factor approaching the duty cycle. Such windowing could
be performed in both the frequency-domain via narrow-band optical filtering and the
time-domain, which could, for example , leverage a single pulse-carving modulator to
efficiently time-window many WDM channel s simultaneously.
3.5.2.4. Extinction Ratio Characterization and Optimization
While measurement of modulator ER is relatively straightforward for CW signals,

measurement of ER for high-frequency waveforms can be challenging. Owing to the
frequency dependence of transmitter components, the ER can vary as a function of
the bit-rate and data pattern, so CW measurements of ER are often insufficient to
accurately characterize the performance ofwide-band modulators. Consequently, high-
speed detection equipment is generally required to directly measure the ER for wide-
band waveforms using time-domain techniqus [324,325] . But these method s can have
limited accuracy and range of measurement, typically < ,....,20 dB, or impose symmetry
requirements on the waveforms being measured. Therefore they are not well suited for
measuring RZ, low-duty-cycle or high extinction ratio waveforms.
156 DavidO. Caplan
iii'
~ Direct modulation : low bandlNiddl . poor ER
and induees chirp, power efftd enl
External modulation:
ide< bandwidlh. good ER
Cascaded modulation: good ER. wide
bandwidt h, litlle chirp, power eofficie nl
tl:Z~::Z~~":":
--- Chirp_.. _. __..__..__.
._ ..... . __... __. __._._._· KL_ £..V.
Fig. 25. Illustration of cascadeddirectandexternalmodulation waveforms showingbothintensity

and wavelength changes in time. The external modulator can improve extinction, waveform and
spectralfidelity via a windowing function, butthis neednotbe squareasdepicted. Otherfunctions
such as a variety of Gaussian-like RZ waveforms (discussed at the end of section 3.5.2) can be
efficiently implemented without transmitter power penalty as long as it is followed by an APL
amplifier. In a similarmanner, spectralfiltering can be used to improve ER, spectral fidelity, and
reducepowerrequirements, by,for example,usinga passive narrow bandfilterto pass thedesired
signal >'1 and reject the unwanted >'2 .
However, with knowledge of the duty cycle" (DC), e.g., the ratio of ones to zeros
transmitted, and the ability to accurately measure the average power (P) and peak
power (Ppk), modulator ER performance can be characterized with the use ofEq. (66)
[11,314]:
ER = P/ Ppk - DC ::::} P (66)
1 - DC lim D C--+O P pk .
For average-power-limited measurements, such as of a saturated MOPA TX output,
the peak power is proportional to the average power and grows as the duty-cycle is
reduced, as long as ER « DC. This follows by solving Eq. (66) for Ppk which yields
P P
(67)
P
pk
= DC + ER(l - DC) ::::}C --tO
lim D ER '
A plot of Pp k versus duty cycle for various ER is shown in Fig. 26. As the duty
cycle approaches the ER, the peak power is reduced by half as the denominator in Eq.
(67) approaches 2. For very low duty cycles, (i.e., DC « ER) the normalized peak
output power assymtotically approaches the ER. By fitting normalized peak power
measurements to the calculated curves, the ER can be determined.
For peak-power-limited measurements, such as for the modulator output, the peak
power is fixed and equal to the average power for a 100% duty cycle. In this case,
ER = Pno r m - DC - (68)
1 - DC ::::}C --tO
lim D Pn o r m ,
where Pno r m P(DC) / P(DC) = 100% ), and the measurement of ER can be

achieved entirely with CW measurements as DC gets small relative to ER, without
6 The duty cycle ofa periodi c or re-occurring intensity waveform. s / (t) is the percentage of ' on' time during
the period T , i.e., DC= ~ JOT (s/ (t) / m axjsj (t)])dt .
ER = ·3 dB
ER = ·10 dB
ER = · 13 dB
ER = -15 dB
ER= -17dB
ER = -20 dB
-25 -20 -15 -10 -5 o

Duty Cycle , [dB]
Fig.26. Peak output power for anaverage power limited transmitter versus duty cycle forseveral
values of modulator extinction [314]. Reprinted with permission. (@2000 IEEE.)
the need for high-speed measurementcapabilities. Both the peak and average-power-
limited measurementapproachesdescribed abovehavebeen used to characterize mod-
ulator ER with good agreement, accuratelycharacterizing modulators with extinction
> 60 dB [15,214].
Modulator control and ER optimization is often accomplished by minimizing the
power-offtransmissiondirectly after the modulator [308,326]. However, for lowduty-
cycle or variable-rate average-power limited (APL) systems [II], another potentially
moresensitivemethodof minimizingthe ER is to maximize Ppk givenin Eq. (67) at the
output of the MaPA TX, since this value becomes increasingly sensitive to ER as the
duty-cycleis reduced [314]. The peak power is maximizedwhen the ER is minimized,
and this effect is exaggerated significantly at low duty-cycles, by an amount that is
approximately inversely proportional to the duty-cycle.Thus, this approach serves to
optimize ER performance by maximizing the peak signal power delivered from the
TX.
3.5.2.5. MZM Drive Power and Chirp Considerations
As shownin Table5, typical valuesof V1r range from 3 to 5 V. Since operating poweris
oftena primaryconsiderationin manyFSOapplications, lowerdrive voltage (V1r <2 V)
is desirable and provides compatibility with COTS high-frequency low-voltagedrive
electronics. For a given electro-optic coefficient, V1r can be lowered by maximizing
device interaction length (d), or increasing electric field (E) across the waveguide.
Since E = V / d, decreasing d allows the voltage to be lowered while maintaining
the same electric field . Both of these approaches tend to increase insertion loss and
capacitance, which in tum reduces device bandwidth. Traveling wave designs can be
158 DavidO. Caplan
~
~ RFdata RF: a
X •
.,-....~--'
Wavegu ide
z-cut, single drive
Fig. 27. Common LiNb03 modulatorelectrodeconfiguration andcrystalorientationwithelectric-

field lines: (left) x-cut, (center) dual-drive z-cut, (right) single-drive Z-CUl.
used to mitigate the bandwidth issue [327-329], but RF-optical velocity matching
constraints make it challenging to obtain a uniform response over the wide fractional
bandwidth (e.g., 40 KHz to 40 GHz-6 orders of magnitude) desired in broad-band
digital applications. This results in a frequency dependent V1r (f ), which can lead to
pattern dependence in the transmitted waveforms.
There are two crystal orientations commonly used to make MZMs in LiNb03,
x-cut and z-cut [283,330], shown in Fig. 27. X-cut designs have a single electrical
modulation drive input that is split within the modulator to provide the push-pull action
required for switching. If designed properly, the drive voltages arrive on each arm of
the interferometer simultaneously and impart the same phase shift (for all frequencies
of interest) , resulting in chirp-free modulation . However, since there is only a single
electrical input, there is no means of compensating with the drive waveform if there
are fabrication errors .
Z-cut modulators come in dual- and single-drive configurations. Dual-drive z-cut
MZMs can be used to adjust chirp by controlling the instantaneous imbalance of the
input drive voltages (V t op and V b ottom) . The resulting chirp parameter defined in Eq.
(3) is given by [278]
a - \!top + Vb o t tom
~-=------::-:--- (69)
- \!to p - V b ottom .
The ability to adjust the chirp can be used, for example , to compensate for mod-
ulator fabrication errors or optimize transmission through a dispersive and nonlinear
fiber-optic link [278-280,331,332] , dynamically if necessary.
For the same V1r , dual-drive z-cut modulators require half the drive power of x-
cut modulators, since each of the two drivers needs only V1r /2, which corresponds
to a net power of V; /(2Rn) for the z-cut, versus V; /2 for the x-cut, where Ra is
the modulator impedance. The dual-drive design also distributes the power load over
a larger area, reducing the heat density which can be advantageous in the carefully
controlled thermal environment of space-based platform s.
For single-drive z-cut MZMs, (Fig. 27-right) , the electric field is stronger in the
optical waveguide centered under the RF data electrode, with only >- 15% of the field
driving the secondary waveguide. This fixed asymmetry leads to a phase-imbalance and
residual chirp, with a =-0.7. A summary of the above characteristics and representative
(minimum) drive powers is given in Table 5.
Table 5. Summary of LiNb03 modulator drive and chirp characteristics for common crystal
orientations and drive configurations.
x-cut z-cut z-cut

single-drive dual-drive
Typical V" 5V 4V 4V
Drive Power V;/Ra = 0.5 W V;lRa =0.32W V;/(2Ra) = 0.16W
Ra = 50 ohms
Chirp (0:) ",,0 -0.7 - 00 < 0: < 00 (in principle)
(fixed bydesign) (fixed bydesign) -2 < 0: < 2 (usedin practice)
Layout size smaller smaller larger

Heat density higher higher lower
3.5.2.6. Pulsed Waveform Generation
Pulse carving using a MZM is a subject that has been discussed extensively in the
literature [10,13,14,25,41 ,132,138,140 ,141,159,333-336] as a means of generating
high-fidelitywaveforms that are well suited for long-haul high-ratefiberand free-space
optical links. Sinusoidal-drive techniques have also been used for ultra-high-speed
serial-parallel conversionin optical time-division multiplexed(OTDM) systems [337]
and photonic analog-to-digital converters [338]. Pulse carving reduces the influence
of imperfections in the drive electronics, such as transient ringing, pattern dependence
and inadequate bandwidth, on waveformquality. Broad-band high-fidelity electronics
are not essential, since a varietyof shaped-RZ waveforms withdesirablecharacteristics
can be generated by simple narrow-band sinusoidal waveforms with appropriate bias
and drive amplitude and frequency [132,336] . Of particular interest are Gaussian-
like waveforms that can be generated in this manner that can facilitate robust nearly-
matched optical communication performance [13,14] a subject discussed further in
section 5. A general expression for the MZM-pulse-carved power is given by
P(t) IX
2
IE(t)1 = cos
2
(~~ sin(21fft) + ~ V;) , (70)
where V" is the switching voltage, Va and f are the sinusoidal drive amplitude and
frequency, respectively, and Vb is the bias voltage, which centers the drive input at
locations summarized in Table 6. Note that the static and random phase offsets of
Eq. (II) are not shown explicitly in Eq. (19) since it is assumed that these offsets are
compensated for by bias-control.
Table 6. Summary of MZM bias points andcorresponding bias voltages.
BiasPoint
Peak nt, [nt E even]
Quadrature ~nt [rn E odd]
Offset nt ± 1/4, [m E odd]
Null rn , [nt E odd]
160 David O. Caplan
MZM Tr ansfor Function

e
o
"Bi 0.75
E
=0.65, FWHM =0.67
'~" 05 · RZ·66%: Duty cycle
. -~ RZ·SO%: Duty cycle =O.S. FWHM =0.5
~
:E 0.25
• RZ·33%: Duty cycle = 0.3S, FWHM =0.33
N RZ.S3%: Duty cyc le = 0.53. FWHM = 0.54
:E - RZ-40%: Duty cycle =0.4, FWHM =0.39
00 2 :1 4 5 6 7
Output Waveforms
~: ~1E
o 0.25 0.5
MZM Intonslty Output
0.75
Fig. 28. Pulse carving waveform generation : (top) MZM Transfer Function. (bottom left) si-
nusoidal input waveforms , and (bottom right) output waveforms for RZ-66% (CSRZ) in red,
RZ-50% in pink stars, RZ-33% (r--Gaussian) in green diamonds, RZ-53% (Raised Cosine) in
light blue triangles, and RZ-40% ("-'Gaussian) in blue circles, respectively. Note the differences
in drive amplitude, frequency, and bias position.
Depending on the drive amplitude and bias position, pulsed-waveforms of duty-

cycles ranging from 33% to 66% can be generated as shown in Figs. 28 and 29. A
summary of waveform spectral and temporal characteristics and drive sensitivities are
given in the pulse carving trade space matrix of Table 7.
The RZ-33% (Gaussian), RZ-66% (carrier-suppressed), and RZ-53% (raised co-
sine) waveforms are biased at either a transmission peak or null, and output a pulse
stream at twice the drive frequency. However, when used with DPSK or DQPSK mod-
ulation, these waveforms require accurate drive amplitude or phase balance across the
two arms of the MZM in order to avoid significant chirp-induced penalties [132,142] .
In contrast, the 50%-RZ is attractive for use with DPSK for several reasons. The
drive voltage is reduced by a factor of two, which widens the availability of driver
amplifiers and reduces the power requirements . Also, the sensitivity to drive amplitude
(Va) and bias position (Vb) is reduced, making it easier to achieve stable bias control
and long-term waveform fidelity, as well as diminished chirp penalty [132,142], since
the MZM carver does not traverse the transmission null, the point at which the output
phase is inverted. These characteristics in tum, make it easier to generate desired
transmitter waveforms and optimize receiver sensitivity [133].
A representative pulse-carving MOPA TX is shown in Fig. 30. The data source can
include an FEe encoder, which output s the encoded data and corresponding channel
clock at a rate equal to the data clock plus the coding overhead . The data formatter is
i.,
c:
2. 0.5
,j
0.;
ii: O J,-l:;....------....;,,~
.,
"0
~ -0.5
E
~ :' '"::
.5----'-~-''''''---:;'-:------:~-----:''"::---=-........--:::.--..J
-0.5 0.5 1.5
~
<0
•c
:::i
~ 0.8
~., 0.6
S 0. 4
.,
"0
~ 02
E
<5 o 1.5
z
Time . [relative)
- 33% RZ, FWHM ' !A6
iii' O, -- - ---, - - - , -- - --,- - -::;_-r-......; :---,,--- --j - - - 40% RZ, FWHM · lAll
~ _ .- · ~ RZ. FWH... · 1.J4
63% RZ, FWHM· 1.11
-11% RZ, FWHM·l.10
- Sq....-NRZ,FWHM·on
o
Frequency
Fig. 29. Pulsed RZ waveform generation using periodically driven Mach-Zehnder modulator.
(Top) Electric field; (middle) power, (bottom) power spectrum.
not needed for OOK but is needed to precode other modulation formats, such as DPSK
or PPM.
3.5.3. High Power Optical Amplifier
Efficient high-power optical transmitters are necessary in optical communications sce-

narios such as broadband distribution networks where electrical power is increasingly
becoming a valuable commodity. Efficiency also benefits large-scale WDM transmis-
sion where TX power is distributed over many channels (As), and FSO links, where
transmitter power is often a limiting design driver (see e.g., [2,21,339]). High power
optical amplifier needs have historically been met by erbium-doped fiber amplifier s
(EDFAs) operating near 1550nm [126,340-342]. When compared with RF amplifiers,
EDFAs have many desirable characteristics including high gain (> 40 dB), large av-
erage output power (> 10 W), and ultra-wide bandwidth > 10 THz . However, as
noted in Table 3 and discussed later in this section, for high-duty-cycle waveforms
162 David O. Caplan
Table 7. Pulse carving characteristics and trade space matrix highlighting drive frequency
and drive amplitude (Va) requirements, bias position, waveform duty cycle (DCpulse) , time-
bandwidth-product, and sensitivities. Adaptedfrom Stevens in [142).
I, L Lcneractensucs
Waveform J ~
T1me-Bandwtdth
___ erD[ju~
I Wave- I-~-F---- C-=- 1 ~ ~ ~ ~rDrIV~ I O;;;;:~ R~8 - -I" --

I form '3 dB 10 dB dB tv !
12 D Freq. AIT1l.
'I Dr! : e llS Sen l ltlvltlel
# .! Type ,A1a' MaIM,'''', !Ar,'Af,1 FWHM' Ie : IllRltel !lVlV1ll' Pow , Point le ,e. Orin
NRZ Mod - l.1od ·

1 Square 0.89 1.49 3.6 100% 100% 0 Sq. n. 0.5 8 Quad eret e crate-
RZ·Squa re Mod - ~1 o d·
2 lSO"~_ 0.89 _ _1.49 3.6 SO"io 50""/., .J Sq . n. 0.5 4 Ou ad elat e et ete
~.
-
RZ.J 3°k
2.- IG.... ..Ian) _ 0"49 _ _1.~_ 2L 33"10 35% 4 .8 Sin 0.5 1 16 Pea•. High High
RZ-6Cl%
4 IC~RZ1 _ 0.73 ~..L...-.]L 67% 65% -1.7 Sin 0.5 1 16 Null High High
- ;
RZ-S3%
(R. lu d ,
COline) 53"/.
5 0.63 1.62
- 2.4 54% -S.7 Sin 0.5 0.5 4 Null High ; L ow
Mod . l.1od -
-.!... RZ.s~ 0.66 , 1 ~4 ·L SO"I0 50% .J Sin 1 0.5 4 Ouad . rot. u ~h
RZ-40% Very
7 IG. use'.n) 0.66 1.57 2.6 39010 40% •7.1 Sin 1 0.:5 1 Oll• • t High L o ...
"'--------------------
: FEe Enc oded Dat a H ------ ----,:
Data MOPA Tx
1 ~~~~~l:~ :~~:~rl_ !-O~~ll'~t~:r- J
,. ----, ,----
f.!-!.~J·t:~_:
en
'"
~
E 0.5 ____ /~~\ 1:1\
./
~
! I
o
Z O +-....,...:~_L....--r-+"'-_£...--r-........:!»o---£-...,--r-....-:: _ _ £....,-4
-100 o 100 200 300

Tim e, [pse c]
Fig. 30. (Top) A pulse-carving MOPA TX with optional FEC and Data Formatting elements.
33%-RZ pulse carving requires sinusoidal drive at ~ the channel clock rate. (Bottom) Measured
10 Gbps 33%-RZ waveforms nearly indistinguishable from a superimposed 35-ps full-width-
half-max (FWHM) Gaussian fit centered at t = O.
optical amplifiers are typically less power efficient than their RF counterparts, but for
low-duty-cycle waveforms the efficiencies are more comparable.
The ability to efficiently transmit a variety of signaling waveforms at high av-
erage and peak output power levels in a single spatial and polarization mode is of
particular importance to free space applications. Single spatial mode operation is nee-
essary for efficient power delivery to a far-field FSO target. In fiber-optic networks, a
single-polarization TX simply reduces the amount of unnecessary (orthogonally po-
larized) amplified spontaneous emission (ASE) noise entering the network. In FSO
applications, single-polarization transmitters can improve TX-RX isolation and al-
low for simple polarization diplexing ofTXlRX signals within an FSO terminal [46],
without the need for active polarization tracking. Furthermore, single-polarization re-
ceivers can offer improved receiver sensitivity [20,41,77] since orthogonally polarized
amplified spontaneous emission (ASE) noise can be eliminated at both RX and TX .
The ability to maintain transmi ssion efficiency over a wide range of duty cycles,
which is facilitated by high-gain and average-power limited properties of optical am-
plifiers [11,343-346], allows for aggressive pulse shaping at the transmitter. This can
improve receiver sensitivity and provide flexible multi-rate capabilities with simplified
receiver design options [11,12,15], a subject discussed below and further in section 5.
3.5.3.1. Average Power Limited (APL) Properties
EDFAs are average-power limited (APL) amplifiers , meaning the amplifier gain and
average output power are fixed for a given average input power, independent of the
shape or duty cycle of the incoming signal. This holds whenever the power-off time
of the incoming signal is much less than upper state life time of the gain-medium
[172]. For EDFAs, which have a slow rv I ms time constant, this is true for practical
optical communications rates above rv I Mbps . Detailed experimental analysis and
simulations of amplifier transients and average power limited amplifier dynamics are
given in [126,340,347] .
For APL amplifiers, the peak to average power ratio varies inversely with the duty
cycle - as it decreases, the peak power grows while average power is maintained (see
Figs. 35 and 36). This is in sharp contrast with RF amplifiers, which are generally peak-
power limited (PPL), meaning there is a maximum (peak) voltage or output power they
can deliver. In order to maximize the average power transmitted, which is important
for power-starved communication links, a PPL TX must operate with nearly a 100%
duty-cycle-a constraint that does not apply to APL optical transmitters.
3.5.3.2. Amplifier Gain, Saturation, and Noise
The functional dependence of amplified signal output power (Poud and gain G as a
function of average signal input power (Pin) is given by
(71)
where go is the small signal gain at the signal wavelength, and Psa t is the average output
power when the amplifier is driven deep into saturation'. Note that the amplifier gain
can have a strong wavefength dependence [126, 340], but for this analysis, the gain
is assumed to be relatively constant over the narrow-band surrounding the signal. The
dependence of ampli fier gain, average output power, and noise figure on average input
power are illustrated in Fig. 31.
7 Note, that we refer to Psat as the maximumuseful amplified output power. which is differentfrom the term
Psoa~t, which is conventionally definedas the output power for which the EDFAgain has dropped by 3 dB
from the small signal gain G o [126].
164 David O. Caplan
29 90 7.0
26
23
==============~ ===~:;- =- - - - - - p;:,.- - -,- - -

ffi'20
:!:!.17 - Pout 6.0
- Gain Iii'
.5
==~i _========~=--=.: ===='
14 5.5 :!:!.
lB 11 • • NF - - - - - - - - - - - -
, •
::'= === 0;
oll 8 --------------- -----~--------------"r---- 5.0 ~
i~ ~ ==========-=_=~=J=f~~~~~~====~~~~~=~
·1 ______________-u
_ - - - - -
iisitij
___ ri~
__
ampUfler. i saturated
4
.
"'~": - - - - - - - - - - - - - ..... - - - - - - - -
_
u::
4.5 :::
'0
4.0 Z
~ -4
P5c1I.in=P sa / go ; . :
~ ·7 - - - - - - - - - - - - - - - ~ - - - - - - - - ~ .-. - - - - - - - - - - - - 3.5
·10 - -------------------~------ ~ ---------------
·13 'F--t--+-t--+-t-t-t--+---+-1-+--+--+-H-t--+-+-t-t--+---+-1-+--+--+-t--+-t--'f 3.0
-40 -30 ·20 ·10 o 10 20
P,", [dB ]
Fig. 31. Typical EDFA average power added (blue, solid line) and gain (red, dashed line) both
on the left-hand axis, and noise figure (green. dotted line) on the right-hand axis versus average
input power (Pin) . For this example, 90 = 27 dB and P sa t = 17 dBm, yielding Psa t- in = - 10
dBm .
The 3-dB gain compression point occurs when Pin = P s a t / go, which is referred to
as the saturated input power (Psa t-in). To efficiently extract maximum output power,
power amplifiers are typically run with Pin » P sat-int. In this regime, the gain
G(Pin ) is significantly compressed and the amplifier operates in a 'power added'
mode, contributing most of Psat to the total output power. For Ps« > Psat -in + IOdB,
over 90% of the available power is extracted corresponding to an efficiency penalty of
",0.5 dB. This penalty drops to < 0.1 dB for Hn > Psat- in + l5dB .
Note that as the input power exceeds P s at -in , and the amplifier gain is reduced, the
noise figure increases as a result of incomplete population inversion in the amplifying
medium [126]. However this generally is of little consequence since the SNR at the
amplifier input is typically large enough such that even a several-dB SNR degradation
is not sufficient to impact communications performance. This holds as long as
GNsp'fJch « 1, (72)
and
'"
SNR TX = N si g- in
2N > N R x-m in » 1,
3
(7 )
sp
where is the net channel loss, G is gain, SNRTX is the transmitter output SNR 8 ,
'fJch
9
N s ig-in is the input signal PPB, N R x-min is the minimum required RX SNR , and
N s p is the spontaneous emission factor, equivalent to the number of noise-photons-
per-mode at the input of the amplifier. For high-gain and a large signal input,
8 This definition of SNRTX assumes that N s ig _ i n is large enough that out-of-band ASE x ASE noise
tenus can be neglected . While this will be accurate for narrow-band or matched RX filtering, it will not be
accurate for unfiltered-broad-band RXs.
9 =
Note that for shot-no ise-Iimited direct-detected signals, SNR PPB so SNR and PPB can be used inter-
changeably. This is not the case for optically amplified signals.
NF
N sp = 2 ' (74)
where NF is the linear noise figure. For a fully inverted amplifier NF approaches the
quantum limit of 2 (3 dB) [126], corresponding to N sp = I. When the condition of Eq.
(72) is satisfied, the received TX ASE noise is small relative to the shot noise of the
signal and the minimum ASE noise added by an optically preamplified RX. If Eq. (72)
is not satisfied, communication performance then becomes dominated by the received
TX ASE [348], which limits SNRTX. However, in this case, the conditions ofEq. (73)
are generally easy to obtain, and as long as the SNRTx > NRX-min, the impact on
communication performance is negligible .
Consider, for example , a TX amplifier operating at 10 Gbps with Pin =-10 dBm,
27 dB gain, 10 dB noise figure, and P out = 17 dBm. The shot-noise-limited input
SNR during the 100 psec bit period is rv49 dB and the corresponding SNRTx =
N S i g j ( 2 Ns p ) = 39 dB. Recall that the quantum limit receiver sensitivity (at 10- 9
BER) for an uncoded optically preamplified OOK RX is rv40 PPB (16 dB SNR) .
Thus , in the regime where the SNRTX dominates, this link would have 23 dB margin.
The large SNRTX indicates that the noise added by the amplifier is still small relative
to the signal, corresponding to rv2500 ASE PPB out of 3.9 x 107 PPB transmitted
in the single-spatial-temporal-polarization mode of the signal. Subsequently passing
both the signal and ASE photons through a 60-dB attenuating channel results in a -43
dBm received signal that is nearly shot-noise limited, comprised of 38.9 signal PPB
and 0.003 ASE PPB, with an SNR of 15.9 dB. For an optically preamplified RX, as
long as the received ASE PPB « N sp of the RX, the TX ASE has a negligible impact
on communication performance . A more detailed analysis including the impact of
broad-band ASE on TX performance is given later in this section .
While we have shown above that the noise figure for a power amplifier is not a
driving design concern, it is worth noting that with careful design, power ampli fiers can
deliver high gain, efficiency, and output power. By incorporating mid-stage isolation
and out-of-band ASE filtering, multi-stage amplifiers have been designed with > 50
dB gain and near quantum -limited noise figures of rv3.1 dB, (see, e.g., [349,350]). This
class of performance can be obtained from commercially available EDFAs, with small-
signal-gain > 45 dB and NF < 3.5 dB readily available. When cascading additional
stages, the net noise figure is given by [126]
(75)
where NFi and G, represent the gain and noise figure of the i th amplifierstage. Clearly,
with high-gain in the first amplifier, the net cascaded noise figure can be dominated
by the (low) noise figure of the first. Since high-gain and low-NF can be achieved for
relatively low «rv IOOmW) pump powers [350], reduced power efficiency in the first
stage amplifier should have little impact on the overall efficiency for > I W high-power
EDFAs.
3.5.3.3. Amplifier Efficiency
The net power amplifier electrical-to-optical conversion efficiency ("lE/o) is an im-

portant design consideration, especially for space-based transmitters since it generally
consumes a significant portion of the overall power budget. For example, the design
166 David O. Caplan
target for the 5W Mars Lasercom YDFA transmitter power amplifier [1,2] transmitter
YDFA power amplifier T}E/O was rv 15% [198], requiring > 33W of electrical power,
more than 20% of the total power budget.
The main factors that contribute to T}E /O, include pump laser electrical -to-optical
efficiency (T}pump), pump coupling efficiency (T}coupling), and pump-to-amplified-
signal optical-to-optical conversion efficiency (T}%), with
P optical- out
T}E/O = P. = T}pumpT}coupli ng17%T}ov e rh ead . (76)
e lec t rica l
Here, T}o~erhead is included to account for any cooling or control that is not included
in the other terms.
Pump laser efficiency includes the electrical power required to drive the pump as
well as temperature and current control needed to reliably operate the pump at the
proper wavelength. Improvements in pump laser efficiencies in the 0.9 p,m absorp-
tion bands have occurred in several areas, including significant advancements in the
reduction of heating/cooling power required to keep the pump wavelength within the
absorption band of the gain medium . In single-mode pump lasers, FBG stabilized
designs have been used to efficiently lock the laser wavelength near the absorption
peak over an extended temperature range [257-259,351,352], with T}pump > rv40%
and> 200 mW coupled into single-mode fiber for individual single-mode sources.
Single-mode coupling techniques often incorporate polarization and/or wavelength
multiplexing to combine pump and signal. To increase single-mode pump power lev-
els, both polarization and dense WDM of stabilized pump-lasers can be used to ef-
ficiently combine multiple sources into a single-mode [353]. These techniques can
scale multi-stage single-mode EDFAs to Watt-class power levels [354]. In high-power
cladding-pumped amplifier designs [75], ER:Yb codoped gain fiber is often used to
broaden the absorption band, which can extend from 850 to 1100 nm [355-357]. This
provides greater flexibility in selection of pump wavelengths , and allows for efficient
absorption of uncooled pump wavelengths as they drift over an extended temperature
range . Codoping also permits high erbium doping concentrations in fused silica fibers
with without quenching induced degradation in gain or efficiency [358].
Cladding-pumped amplifier designs permit efficient coupling of many multi-Watt
multimode pump lasers into a double-clad gain fiber, which consists of a large mul-
timode outer-diameter waveguide with high numerical aperture (NA), and a smaller
doped-core single-mode waveguide (required for FSO application s) with lower-NA
[359]. Pump power coupled into the low-loss outer waveguide is eventually absorbed
by the gain-media within the inner single-mode core. The ability to accept output from
multimode pump lasers in the large-NA outer waveguide, significantly increases pump
coupling efficiency and available pump power levels, enabling the use of efficient high-
power (multi-Watt) sources with T}p um p = rv50% [360-362] and a variety of low-loss
pump-coupling and combining strategies [75,349,363-372].
Optical -to-optical conversion or power conversion efficiency [126], defined as the
ratio
_ P out - Hn < Apump

(77)
T/% - pPum p - A'
in B ig
is limited by the quantum-optical conversion efficiency in which a single pump-photon

at A p u m p is converted to a single signal-photon at Asi gn al. For 1.55 p,m EDFAs with
0.98 JLm pumping, "10 / 0 ::; 63%, whereas for 1.06 JLm YDFAs with 0.98 JLm pump-
ing, "10/0 ::; 90%. Optical-to-optical conversion efficiency is further reduced by other
losses in the amplifier such as incomplete pump absorption, insertion loss of filters,
isolatorsand splices, excitedstateabsorption,scattering, andother parasiticloss mech-
anisms [373].
The state-of-the-arthigh-efficiency high-power amplifierdemonstrations are based
cladding-pumped designs and include a high-reliability single-polarization EYDFA at
1550nm with output power > lOW and "IE/O = "-' 13% [199].With the combineduse
of PM componentsand internal polarizers, a 30-dBpolarization extinctionratio (PER)
was achieved. At 1064 nm, a high-efficiency YDFA design based on high-reliability
PM components but without polarizing elements demonstrated output power> 8W,
"IE/O =,,-,21 %, and "-'13dB PER [204]. Efficient laboratory-grade YDFAs with "10 /0
approaching 80% [374] and "IE/O up to 40% [200] have been reported.We note that
Nd-doped amplifiers have also been considered for use in space-based applications,
however reported efficiencies are lower and they are less tolerant to the effects of
radiation [228].
In section 1.1 , we introduced net transmitter efficiency ("ITX) , which represents
the energy required to generateTX photons. In absolute terms,this is limited by "IE/O
and TX wavelength as summarizedinTable8 for the amplifiers mentioned above.Note
that while I JLm amplifiersgenerallyhavebetter "IE/O in terms of powerefficiency due
to a smaller quantumdefect, this advantageis reduced by a comparableamount when
TX efficiency T}Tx is normalizedin terms of energy per photon as shown below. This
is a direct consequenceof Eq. (4) with photon-energy being inversely proportional to
wavelength.
Table 8. Minimum TX photon-generation efficien cy in units of Joule s/photon for demonstrated

high-efficiency power amplifiers. Note, I attoJoule (al) = 1O- 18 J.
"IE/O "ITx >

-
hv
'l E / O
= , he
A'lE / O
, [aJ/photonl
1.55 13% 1.0
1.06 21% 0.9
1.06 40% 0.5
Anotherimportantefficiency metricis the amplifierwall-plugefficiency, whichin-

cludesconversion andconditioning("IE/ E) requiredto adaptthe available powersource
from the "wall", (e.g., 120 V a e for terrestrial applications or 28 V d e for space-based
applications) to voltage levels suitable for driving the amplifier hardware, primarily
the pumps, which typically require a "-'3 V d e supply to drive lasers with bias voltage
< 2.5 V:
"Iwall = "IE/ E"IE/ O· (78)
Typical values for "IE/E are between "-'70% and 90% depending on the voltage
conversion levels, a factor that can influencethe amplifierelectrical design. In [204],
for example, the pump lasers for each stage are wired in series to maximizeelectrical
powerconversionefficiency, yieldinga YDFA with "IE/ E = ,,-,85%, and "Iwall ="-' 19%.
168 David O. Caplan
3.5.3.4. Polarization-Maintaining (PM) Fiber Amplifier Designs
As noted at the beginning of section 3.5.3, optical amplifiers that can maintain the
polarization of the optical signal are desirable for free-space applications since they
provide additional capabilities including polarization diplexing ofTX and RX signals
and improved communication performance since orthogonally polarized amplifier and
background noise can be eliminated in the receiver. They are also necessary for trans-
mitting polarization modulated signals and expand the options for implementing multi-
access FSO terminals that incorporate both wavelength and polarization multiplexing
[48].
The conventional method of maintaining a single-polarization through a fiber am-
plifier uses passive PM components and gain media. However, it is challenging to
achieve good polarization extinction ratio (PER) with this approach due to polariza-
tion cross-coupling of multiple cascaded components with limited PER. Amplifiers
using this design typically can achieve ",,12-17 dB PER as in [198,204). To maintain
good PER, polarizing elements, such as single-polarization isolators, can be introduced
to the PM optical path to remove the orthogonal polarization seepage before it becomes
excessive. As shown in [199), this approach is consistent with achieving good polar-
ization extinction (PER = ",,30 dB) without degrading electrical-to-optical conversion
efficiency. However, this approach precludes the use of polarization modulation .
Active polarization control [375-377] can be used to compensate for polarization
fluctuations in a non-PM amplifier at the expense of additional complexity and in-
creased size, weight, and power. Alternatively, passive compensation can be achieved
through a double -pass amplifier design incorporating a mid-stage Faraday rotator mir-
ror [353,354,378]. This approach can compensate for both inherent and time varying
birefringence to maintain either single or dual-polarization states with good PER unlike
the passive designs based entirely on cascaded PM components noted above.
The basic amplifier configuration is shown in Fig. 32. A single-mode PM optical
signal enters the amplifier through the input port of the PM I/O coupling element,
which could be a polarization beam splitter (PBS) in a single-polarization design or a
PM circulator for dual-polarization applications. Unlike conventional PM designs, the
input and output ports of the I/O coupler are the only ones that must be polarization
maintaining. This greatly simplifies the amplifier design since all other elements can
be non-PM. The signal makes a first pass through the gain media and filter, and is then
reflected by the Faraday rotator mirror (FRM) which reflects the amplified-filtered
signal and rotates the polarization orthogonal to that of the input. After the second
pass, the twice-amplified signal polarization is orthogonal to that of the PM input,
and is directed towards the amplifier PM output. The PER performance for the entire
amplifier is primarily a function of the PM I/O coupler PER and the 90° rotation of
the FRM, which can exceed 30 dB with commercially available components. Pump
coupling to the gain media can be achieved through a variety of single or multi-mode ,
WDM, side- or end-coupled elements mentioned in the Amplifier Efficiency Section
above and related references .
A key feature of the double-pass design is that it re-uses the same gain medium,
giving it twice the comparable small-signal gain of the equivalent single-pass design,
but with fewer components in the critical path. However, with the double-pass design,
the forward gain is equal to the backward gain and pump power that is not converted
during the forward pass can be extracted during the backward pass. When operating
PM slgn.1 1/0 Coupling Pump Gain

Input Element Coup ling Medi. r-L~;;"i;;enSitl';e region!
, FI. ~ I
! opneat F.r.d.y I
I\ Filler Mirror J
Amp lified
PM Output
Fig. 32. Schematic for double-pass polarization-maintaining optical amplifier.
30 , -- - - r- - - , -- ,= = = = = = = = = = = = = = = = 'il
15 meter first stage , 17 meters second stage ,
500 mW 980nm (co-propagating) pumps. at each stage ,
1 nm filter with 0.3 dB loss, 1 mW Input
10 15 20 25 30 35
Fib er length, 1m]
Fig. 33. Simulation of amplified signal power as a function of length and Faraday mirror reflec-
tivity in a 2-stage double-pass EDFA with forward-propagating pumps. For FRM losses between
2 and 13 dB, simulated results show < ",0.5 dB loss in output power, consistent with experimental
measurements [354].
in saturation, the average gain adjusts over the length of the double pass EDFA until
the amplified signal extracts all the available converted pump power. As a result, the
double pass design is less sensitive to the distribution of pump power, in contrast to
single pass designs, where the pump power of the final stage is critical. Furthermore,
while excess loss near the amplifier output directly reduces the output power by I dB
per dB loss, additional attenuation in the loss-insensitive region of the double-pass
has little effect on net output power, as long as the amplifier continues to operate in
saturation [354] as shown in Fig. 33.
Consequently, advantageous but lossy elements such as optical filters, additional
pump-coupling elements, WDMs, etc., can be inserted into the design in the loss
insensitive region with little power penalty at the output [379]. The optical filtering,
for example, can perform double-pass pulse-shaping and remove out-of-band ASE
that competes with the signal for gain, enabling the amplifier to saturate at lower
170 DavidO. Caplan
Optical Ampl if ier Saturation Characteristics
: E"-'~J
(II
~ .. TI. II 1111 .. ....... 111I'''''lIl
:f
0
Q;
~
0
Q.
-20 -10 0 10
Power In, [dBm]
Fig.34. Measuredoutputsaturationcharacteristics for a high-gaintwo-stage double-pass ",O.5W

EDFAwith mid-stage narrow-band optical filter [354).
input power levels. The filter also makes the amplifier more stable by increasing the
threshold of lasing at the ASE peak, thus reducing the risk of potential1y damaging Q-
switched pulses. This is illustrated in Fig. 34, for a high-gain double-pass EDFA with
mid-stage narrow-band optical filter. Without any internal isolation, the output power
is stable to better than 0.5 dB over a 30-dB dynamic range of input power and 45-dB
gain. In addition to stable PM performance, this amplifier design has demonstrated
good power efficiency, with optical-to-optical conversion efficiency of ",45% at 1.55
/-Lm [354]. Such characteristics combined with radiation tolerance of ",0.1 dB/krad
[380], are attractive for use in space-based lasercom systems [21], especial1y those
with variable-duty-cycle variable-rate capabilities discussed further in section 3.5.5
3.5.4. High-Efficiency Semiconductor Optical Amplifiers
While power-amplifier needs historically have been met by rare-earth-doped fiber am-
plifiers, alternative semiconductor-based optical amplifier technologies show promise
at Watt-class power levels [197,381]. Efficiency, single-polarization, and extened wave-
length range are some performance advantages that semiconductor optical amplifiers
(SOAs) have over fiber amplifiers. More importantly, SOAs have the potential for
improved size, weight, and power which, along with to manufacturability and in-
tegration into more comprehensive subsystems, can significantly reduce cost. Since
SOAs are electrical1y pumped, the quantum-defect and other losses associated with
optical-optical conversion are absent, so in principle, 'f/E jO could be greater than >
25%. Slab-coupled optical waveguide amplifiers (SCOWAs) are of particular interest
since they have demonstrated Watt-class (chip) output powers with large mode size
compatible with efficient coupling to single-mode fiber [197].
In comparison with EDFAs, the dynamics in SOAs, like semiconductor lasers dis-
cussed in section 3.1, are more complex and faster by over six orders of magnitude. The
slow millisecond upper-state lifetime in erbium yields attractive kHz-class high-pass
characteristics that are well suited for high-speed optical communications, enabling
EDFAs to deliver a constant average power limited (APL) output when driven by high-
frequency time-varying input signals above", 10 kHz [126]. For signals with frequency
content below the cut-off frequency, the average ampli fier gain and output power track
the input signal. Output power and efficiency are reduced for low-frequency, low-duty
Qi'
> 1 Jl
~ (I
l
(\ (1,_ ('
Qi
.:. 1\
", A
f\ r. :"
4;-
~
0
" . .. - r~ .
r. ...
;.
Q.
-I-"- "_,"",-,i'L->,""-,-"'-."'-i""4"4-"+__-"+-__'-.-''''4---'''"'-+-'''4->'!fL-'l'-'''W'...,,.o4-'''t1244 Mbps . DC- 112
° 10 TIm e, [n l ee] 16 20 2S
Fig.35. Measured peak-power-limited (PPL) variable-duty-cycle 2-PPM waveforms at the input

to a high-gain average-power-lim ited EDFA [II] .
."...
1.0
0.9 0_051 Gbps

Peak Pow er - Average Power
~ 0.8
" I
I
[DC + ER(1-DCll 0.155 Gbps
···
I 0.311 Gbps
DC - Duty Cycle 0.622 Gbps
~ 0.7 I
1.244 Gb 5
ER - Ext inction R atio - 23 dB
~ I
I
~ 0.6 I
:,
alN 0. 5 :
~ 0.4 ·!,
E 0.3
Z
11
11 : I I
~
r
0.2
~~
0.1
0.0
I" .:"
:r~
0 5 10 TIme. [ns ec) 20 25
Fig. 36. Measured variable-duty -cycle 2-PPM waveforms at the output of an average-power-
limited high-ga in EDFA [II] . Reprinted with permiss ion ofIEEE. (@1999IEEE.)
cycle waveforms, i.e., the amplifier acts peak-power-limited in this low frequency
regime. Since SOAs have a nanosecond-class upper-state lifetime, the high-pass cor-
ner frequency is shifted up towards rvI GHz, comparable to the Gbit/ sec data rates of
interest. Owing to these high-pass characteristics, maximum TX power-efficiency is
attained whenever the transmitted waveforms have little (power off) spectral content
below rvI GHz, or equivalently, the power-off time is short relative to the upper-state
lifetime [172]. For this class of amplifier, power efficiency is optimized for modulation
formats that have high rvI()()% duty-cycles, such as DPSK and FSK, or formats with
constrained power-off time such as binary-PPM at rates >rv5 Gbps or RZ-DPSKIFSK
at rates >'" I0 Gbps. Outside of these efficiency considerations, communication ex-
periments using an SOA-based MOPA TX have achieved near-quantum-limited re-
ceiver sensitivities for binary-PPM, OOK, and DPSK modulation formats , showing
that SOA-based transients have little impact on RX sensitivity [172].
172 David O.Caplan
Fig. 37. An average-power-limited MaPA TX. Net optical filtering can comprise multi-
ple pulse-shaping and mid-stage ASE rejection filters. i.e., hnet(t) = hl (t) * h2(t). or
equivalently,Hnet (J) = H 1 (J) . H2(J) .
3.5.5. ArbitraryWaveforms and Variable-Duty-Cycle Signaling
The characteristics of an average-power-limited high-gainsaturated-EDFA described

above provide a flexible and powerful platform for waveform optimization. Once a
power amplifieris driven far into saturation, the output power added by the amplifier
becomes insensitive to the average input power, making it a stable source of average
power. Higher amplifier gain is beneficial since it widens the range of input power
levels that can extract the maximumsaturated output power. This makes the TX less
sensitive to insertion loss changes in the preceding elements that can arise due to
modulation or degradation [354], and allows the use of arbitrary waveforms over a
wide range of duty-cycles withoutsacrificing transmitter powerefficiency [10].
For the constant average output power delivered by a saturated EDFA, the peak
output power is described by
p Ps a t
Ppeak = DC ==} (79)
elf goP;n » P sat DCeff
and grows as the duty cycle is reduced. Here we have introduced the term, DCeff =
DCmod DCpulse to identifythecombinedinfluenceof both modulation and pulseshape
on the effective duty cycle. The impact of variable duty cycle input waveforms (Fig.
35) on the output of a saturated high-gain EDFA is shown in Fig. 36.
Given these basic properties, an average-power-limited MOPA TX can be used to
efficiently deliver high-power high-fidelity waveforms of arbitrary type [10], within
limitationsdescribedbelow. Forinstance,a narrow-band electrical waveform s, (t) can
simplybe impartedon the CW lasersource so(t) in Fig. 37 and subsequently amplified
withoutdistortionas longas the bandwidthof the net filter H net (f) is sufficiently wide.
To generate high-fidelity optical waveforms that exceed the available electrical
bandwidth, pulse carving techniques described in section 3.5.2 can be used. In this
case, the electrical drive waveform could perform additional shaping or serve as a
lower-frequency windowing function to selectwhichpulsesare transmitted.Additional
pulse-shaping can also be performed in the opticaldomain throughsubsequentoptical
filtering, a processthat can efficiently generate shapedsignaling-waveforms withTHz
of spectral content. For example so(t), could be a windowed short-pulse laser source.
The TX output waveform STX (t) becomes
STX(t) = [Se(t)So(t)] * hnet(t) A STX(f) = [Se(f) * So(f)] Hnet(f) .

(80)
Fig. 38. Variable-duty-cycle pulse-carved MOPA-TX.
For sufficiently short optical pulses, 8 0 (t) looks like a delta function o(t), and
8TX (t) -t h net (t). These and other wide-band pulse-shaping techniques (see, e.g.,
[382]) , could, in principle, be incorporated into an APL TX with little impact on overall
TX power efficiency.
When using a MOPA configuration with a wide range of variable-duty-cycle wave-
forms such as those shown in Figs. 35 and 36, the final power amplifier stage needs
to run deeply saturated in order to efficiently deliver maximum average power. Effi-
ciency is ultimately limited by the power of the master laser, mid-stage optical filtering
and losses, and amplifier gain and noise figure. The impact on transmitter design is
illustrated in the representative TX power budget given in Table 9 for the MOPA-TX
shown in Fig. 38.
Table 9. Variable-duty-cycle, pulse-carved MOPA-TX power budget.
Low Duty Cycle High Duty Cycle

Master Laser Power (Pl aser) 15.0dBm 15.0 dBm
99:1 PM tap -D.2dB -D.2dB
Static carver loss (lLMZM-CARV) -3.5 dB -3.5 dB
Pulse-carving loss -3.0 dB -3.0 dB
99:1 PM tap -D.2dB -D.2 dB
Static data mod. 10ss(lLMZM-DATA) -3 .5 dB -3.5 dB
(VDC) Modulation loss -13.0 dB -3.0 dB
Isolator -D.6dB -0.6 dB
99:1 PM tap -0.2 dB -D.2dB
Other loss -D.8dB -0.8 dB
E = Minimum power delivered to -1O.OdBm O.OdBm
Power Amp PA (Pinrnin)
Amplifier small-signal gain (go) 50.0 dB 50.0 dB
Efficient power extraction for 30.0dBm 40.0dBm
Psat < Pinmin + go -10 dB
In this example, an amplifier with a small signal gain of 50 dBm can provide a
10-dB range in duty cycle, and deliver up to IW (30 dBm) of average power with less
than ~ dB loss in efficiency. In order to maintain maximum output power and extend
the range of duty cycles, additional gain stages along with inter-stage filtering may be
required as discussed in the following Sections. With proper design, a dynamic range
of operation in excess of 20 dB can be achieved [354].
174 David O. Caplan
3.5.5.1. Low Duty Cycle Limitations
As shown in the previous section, careful TX design is necessary to maintain TX

efficiency over a wide range of duty-cycles. Several other factors can limit the useful
range of duty cycles of a MOPA TX. These include a) limited modulation extinction,
b) buildup of ASE, and c) nonlinear impairments.
3.5.5.2. A) Limited TX Modulation Extinction
Imperfect extinction can reduce the effective transmitted signal power for low-duty-
cycle waveforms (see also section 3.5.2). As seen in Eq. (67), the peak power is a
function of the average power, effective duty cycle and extinction ratio. For the case
of an EDFA operating deep in saturation, Eq. (67) becomes [314]
Psat Psat
===> (81)
Ppcak = DCeff + ER(! - DC eff) ER«D C DC eff.
As the duty cycle becomes low, the average power seeping through the modulator
due to imperfect extinction can become significant. For example , for a I% duty cycle
signal and a modulator with a -20 dB ER, the net signal power out of the modulator
during the I% transmission will be about the same as the seepage during the remaining
99%. The subsequent average power limited amplifier will output equal power in both
signal and seepage, effectively reducing the transmitted signal power by "'-'3 dB. The
extinction power robbing penalty is given by
nc., ===> 1. (82)

1]TxER = DCeff + ER(! - DCeff) ER«D C eff
In order to extract maximum amplified signal power from a saturated EDFA the
modulation extinction must be much less than the duty cycle. For less than 0.1 dB
"power robbing" penalty due to accumulated signal seepage during the (1- DCeff) off
portion of the waveform, the ER target should be IS dB less than the duty cycle, i.e.,
ER::::; DCeff-15 dB. Thus , for I024-PPM with an 0.1% duty cycle (DCeff =- 30dB), an
ER better than -45 dB is required . A plot of ER measurements and calculated fit (from
Eqs. and associated TX power penalty for a modulator with about -52 dB extinction
is shown in Fig. 39. Using this modulator with 1024-PPM waveforms would result in
negligible < "'-'0.05 dB TX power penalty.
3.5.5.3. B) Transmitter ASE
Transmitter ASE buildup can degrade communication performance in two ways: by

limiting the transmitted SNR (SNRTX) , and by reducing the transmitter efficiency and
transmitted-signal output power. The first relates to the ratio of signal photons to ASE
photons within the signal-mode and is independent ofTX filtering, whereas the second
relates to the ratio of output signal to total ASE power (TXPowerRatiosig/AsE), and
has a strong dependence on optical filtering.
As discussed earlier in section 3.5.3, the condition s of Eq. (73) can be met by
simply providing sufficient input signal to the power amplifier. This results in sufficient
SNRTX, so that the TX ASE has little impact on communication performance. This
_ _ _ _ _ _ _ _ _ .- Ppk Colle . (dB) : ER . p.rtKt
- - - - - - - - - - - - - - - - - - '- Fit 10 P pk'P~v g : ER . - -53.0 dB 0.9

=-=-~~""
_ \- .a D.1t.1 for APL EOFA: Ppk1P3vg
0.8
-: - - - - - - - - ' - Fit to Polyg : ER . ...-51.0 d B .-.
_____ . • O.1U for PPL MZM: PavQ(DC)/Polvg(OC- ' ) . 07_ !1l.

- - - - - - - - - - - - - - - - - - '- Tx Pow . , P. n.1lty . [dB] 0.6 z:.
Iii
0..5 ~
0..
0.4 t
;:
0.3 0
0..
0 .2 ~
0.1
.6 0 · 50 ..\0 ·3 0 ·20 ·10 o

Duty Cycle. [dB)
Fig. 39. (Left axis) measurements of modulator extinction using variable duty-cycle techniques
described in section 3.5.2 . Right axis) TX power penalty [15).
TxPolI'erRati0s.,IAS£ =PS1GDo4
t P,/SF
EDF~ Pout1
!
P a..1
P_= 15 dBm Il.c.mponen.= 9 dB gol =40 dB
1~:S33dB NF1 = 5 dB
=
Psoil 20 dBm
Fig.40. MOPA transmitter configuration used for modeling the ASE buildup described by Eqs.
(83)--(88) .
also holds for a cascade of amplifiers, since the net noise figure (and N s p ) for the
cascade is dominated by the first stage.
However, high-gainmulti-stagepoweramplifiers,can generatesignificant amounts
of broad-band ASE which can reduce the TX power efficiency. With proper design of
both the TX power budget and mid-stage optical filtering the TXPowerRatios ig / ASE
can be optimized. To avoidASE power robbing, a TXPowerRatios ig / ASE > 20 dB is
desired, meaning more than 99% of the output poweris the signal and less than I% of
theTX poweris ASE.An exampleMOPA TX shownin Fig. 40 witha wideoperational
range of input powersis modeled using Eqs. (83)-(88) [126,142].
176 David O. Caplan
. PSI Gout
T x P owerRa tlosig/ ASE = P , (83)
ASE
PSIGout =G Plaser
= GRSIGin, (84)
IL componentILmodulation
Pin = PSIGin + P ASEin, (85)
PASE = (G - l )(N spmBhv ) + GPASEin · rnin(B /B in , 1) ,(86)
n; =
NF (
2
G) C;'12NF '
G -1 (87)
G(Pin, P sat) = 1 + C+ Pi~;O/ P saJ . (88)
Here , B is the optical filter bandwidth and m is the number of polarization mode s.
The se equations can be iteratively applied to solve for the cascaded amplifier output
signal to ASE power ratio, which depends directly on the signal input power, and
varies with the modulation duty cycle . For the first amplifier stage, PASEin = in
Eq. (85). The amplified signal PSI Gout and ASE power PASE from the first amplifier
°
are then used as inputs to the second amplifier stage . Variable-duty-cycle M -ary PPM
waveforms are simulated for the binary case (M = 2) through I 024-PPM with 50%-RZ
pulse carving. The corresponding modulation insertion loss, ILmodulation= l/DC = 2M
varies from 6 to 33 dB (assuming perfect ER). Representative elements contributing
to the component insertion loss (ILcomponent) are given in Table 9.
Simulation results for the two-amplifier cascade with a narrow 4
nm (62 GHz)
mid-stage filter are shown in Fig. 4 I. Over the 27-dB range of input powers, the
average output power is stable to ",0.5 dB. For M = 1024, with 10 bits of information
canied per PPM symbol , the duty- cycle and input power are the lowest, but under
these conditions, the APL amplifier generates the highe st peak power of 62.5 dBm
('" 1.8 kW). In this case, the minimum SNR Tx is ",27 dB and has more than 99% of
TX output power in the signal with TXPowerRatiosig/As E > 22 dB.
3.5.5.4. C) Nonlin ear Impairments
Optical nonlinearities are another factor that can impact low duty cycle transmission
through a saturated average-power-limited transmitter. Fast nonlinear effects such as
self-phase modulation (SPM ), four-wave mixing (FWM), and Stimulated Raman Scat-
tering (SRS) tend to shift usable signal power out of band, an effect that reduces trans-
mitter efficiency and makes an average power limited amplifier appear peak power
limited. Nonlinear effects occur when the peak power exceeds a threshold
A eff
Pth <X L ' (89)
g NL eff
in which A eff is the effective area, G N L is the nonlinear gain coefficient (a material
property), and L eff is the effective propagation length [110, III]. Standard steps to
increa se the nonlinear threshold include using larger core fiber (which reduces the
power density), minimizing amplifier lengths [383], and distributing amplified signal
energy in time to lower the effective peak power through the amplifier. A summary of
high power amplifier design option s and tradeoffs is given in Table 10.
Practically, Pth can be defined as the peak output power when greater than e- I% of
the average amplifier output is contained in the nonlinear induced spectrum. Typically,
8..•.•. • ! i : -e- T xPo werRatio .;g/ASE' [dB)

60 : ~' , , -v-- Av erage Sign al Power Out, IdBm) .
! """ ...•;;.! i _.s,- Peak Sign al Power OUl IdBm)
G ..,... : """ : - ~ - SNRT, . (dB)
--"'E}. : •.•~- . : ..D · Net G ain •
50
III
"'C
=2
40
...........
30
~. ---
..:-
2OL-_ ---=-=.:.::r::..:... -'-- -'- ---L ---' ---l
·30 ·25 ·20 ·15 -10 -5 o

Pin, [dBm]
Fig. 41. Simulated amplifier signal to ASE power ratio (TXPowerRatios ig / AS E in circles).
average signal output power (down-pointing triangles). peak signal output power (up-pointing
triangles) . TX SNR (diamonds). and net amplifier gain (squares) as a function of input power.
Pe. is about 250W to IkW in commercially available EDFAs [2 I, 198.199], although

promising research-grade EDFAs have generated 100 kW-class peak power with little
or no nonlinear effects [224] . Once beyond Pth, the nonlinear effects rapidly dominate,
For the MOPA TX modeled above, the lowest duty cycle generated a peak power of
rv 1.8 kW. which can lead to prohibitive nonlinear impairments in conventional power
amplifiers. Such effects are illustrated in Fig . 42.

Note that the signal at rv I557 nm is > 33 dB larger than the amplifier ASE which
peaks at rv I567 nm for this ampli fier. Therefore the net ASE power is less than I% of
the net transmitted power and can be neglected. As the peak power of the amplifier is
increased (by lowering the duty cycle of the input), nonlinearities appear. The combined
effects of dispersion and SPM can lead to modulation instability [I I I], which can
generate spectral sidebands around the signal. These are often the first noticeable
nonlinear effect. SPM on its own can broaden the amplified signal spectrum, although
this is not as apparent when viewing the broad band optical spectrum. As the peak
power continues to increase, the nonlinearly-induced side bands grow and take a larger
percentage of the amplifier output power. For this amplifier, SRS tends to dominate at
higher peak powers, shifting the optical spectrum to the longer wavelengths. As the
peak power is increased further the SRS-generated spectral components can extend
beyond 100 nm, stealing a significant portion of the net amplifier output power.
Another nonlinear effect is stimulated Brillouin scattering (SBS) which is a slow
nonlinearity that scatters light in the backward direction. This attenuates the forward
propagating signal, reducing the TX efficiency and can lead to catastrophic damage
in the final amplifier stage due to SBS induced Q-switching. In standard single-mode
fiber with an effective area of 50 p,m2 , an estimate of the SBS power-length threshold
178 David O. Caplan
Table 10. Design options for stable high-peak-power single-mode (spatial and temporal) wave
guide amplifiers.
rrl l~ - Opt~ Benetitl ~ I---- -~
l.argcrc orc
fi I1\; r
- 2 - 10 x Mulu mod c
opcrution ·· Poor tiber cou pling
R''l.ltl<:,·d NA
I
·· Co u pli ng Iw.; s
Scns iuvc to bend lus~
2
M iniruizc ii bcr
k nglh'lhrollgh
higher d.'pin g
- 2 - -h f\la,illallll doping
conccnu. uion-, in
luscd silicu
· Dopants m~y lead to enhanced
radiati on sensitivity for space -

based applica tions
conccnnut ion • Re duced power effic iency
Coiled
mu lri ruodc
- IlHl x Elfivicncyof
lund.uucutal modc ·· In du ced polari zation depe nde nce
Re duced p ower efficiency
.\
122--1_3X-I!
·· Rel iab ility of coi led fiber
Research grade only
~
Sdr-ililaging
wnvcuuidc
-. IlHl x T cc hn tcul, doping
leve ls ·· Coup ling effic ieucy and stability
Therma l and moda l stabilitv
13X51
1'.", .unplificd - 2 - -1 .\ Muxuuuu chirp and ·· Re searc h grade only
Red uc ed power efficiea cy
s chirped pulse
c ompres sion
Clllll p"'llsa l io n
· Pote nt ia l
distort ion
for wavefo rm
WDf\1
di ,lrihlllion l22 1
-·100 \
· Comple xity and
cost (SWAP) · Increa sed comp lex ity. cost. IIIHI
SWAP ewer sin gle channel
ft · Avai lab le
optica l
bandwidth
· FWf\I nonlinearities may lead to
data-dependent p ower efficiency
reduction, whi ch can lead to au
· FWI\I effie ieucy

redu ction
erro r 1100r w itho ut FEe'
is '"'-'20 W m within the '"'-'50 MHz Brillouin linewidth. However, in contrast with the
ultra-fastSPM and SRS nonlinearities, SBS has a narrowgain bandwidth (..1l1b < '"'-'50
MHz), and therefore, can be mitigated by broadening the signal spectrum. This can
be achieved with the use of short durationsignalingwaveforms or inducingadditional
phase modulation [110,111]. For example, by using short 100 ps pulsed waveforms
with '"'-' I0 GHz class bandwidth, this threshold can be increased by a factor of 200, to
'"'-'4 kW m. In addition to the nonlinear mitigation techniques highlighted in Table 10,
SBS can be suppressed throughtemperature gradientsin the fiber [374]. As a resultof
the temperature dependentBrillouin frequency shift, temperature segmentation of the
high-power-handling fibercan be used to disruptthe SBS phase matching, yieldingup
to 7 dB suppression [386].
The use of WDM (Fig. 43) to reduce peak power by distributing the amplified
signalenergyovermanywavelengths is particularly attractiveforuse incommunication
systems employing strong FEe. This approach directly reduces slow SBS effects by
lowering the peak power within the SBS bandwidth. However, fast SPM, FWM, and
SRS nonlinearities havemanyTHz of bandwidth, so the benefitsof thisapproachoccur
only whentheWDMsignalsdo notoverlapin timeforrelatively short,'"'-' 10m amplifier
lengths. Fortunately, as the duty cycles get lower and peak power impairments grow,
the probability of pulse overlap is reduced. Nonlinear impairments that occur due to
- - - - - -, - I ,- - - - - - ,
o ------- ---T----------------------- ---- ------

I I I
Amplified ~ Threshold for Nonlinearites :

Sinnal I
..-. -10 - :.:l. - - - - - - - ~- - - - - - - Ptil ::: A. J9NLL. 1f - 250 W Peak
aJ
~ : SRS
E
::J
-20
I
,::,
~
Co -30
...
tJ)
::J
.s- -40
::J
o
-50
-60
1500 1550 1600 1650 1700
Wavelength, [nm]
Fig. 42. Nonlinearly-induced spectrum for several peak output powers in an Er:Yb doped cladding
pump amplifier. The bottom trace shows the amplified signal spectrum from a 4-W EDFA without
any nonlinear effects . Fiber nonlineariti es can generate new spectral components which shift
usable signal power out of band, effectively making an average-power-Iimited amplifier look
peak-power-limited [21].
sporadic pulse overlap can cause a pattern-dependent performance degradation. But as

long as these events occur infrequently enough. they can be compensated for by FEe.
Consider, for example, a 10-W MOPA-TX with Hh I kW using square 1024-
="-J
PPM waveforms, and a receiver employing FEC with an error-free cutoff at a 10- 3 error
rate . Single-channel peak powers would be 10 kW. By distributing the TX over 32
"-J
wavelengths, the peak-power per channel is reduced to 320 W. But the aggregate peak
power can be higher due to multi-channel pulse overlap. Assuming equal probability
for each PPM symbol and time-alignment between the independent channels, an upper-
bound for the probability distribution of peak aggregate power is shown in Fig. 44 for
8. 16,32, and 64 WDM channels.
For the 32-channel configuration, the overlap of 3-pul ses yields a peak power of
960kW, below Hh. The overlap of 4-pulses yields a peak power of 1.28 kW, above
Pi«, but with a probability < 3 x 10- 5 , well below the FEC error cutoff. Therefore, the
WDM signal distribution over 32 channels effectively suppresses broadband nonlinear
impairments by about a factor of ten. This corresponds to the ratio of the number of
WDM channels, w , to the effective number of pulses that overlap with a probability
greater than the FEC threshold rate, lI:eff,
which is 3 in the example above. The effective
peak power then becomes
p eff
pk
= (lI:eff)
W ocu : .r: (90)
180 David O. Caplan
Power
Amp
Time
Fig. 43. M-PPMlWDM waveforms illustrating the WDM method of dimini shing peak signal
power by distributing it over many wavelengths Relative to a single channel TX of same duty
cycle , average power, and symbol time Llt, a w-channel WDM transmitter reduces the peak
power and bandwidth of each channel by a factor of w [22]
This benefit could be furthered by intentionally staggering the time alignment of the
WDM channels to reduce K eff .
10° n .",,:f:· ',.,;-7~~:---"'-----'--r==::====::==:'==:=:=:r::===:::==;-,

.. .. po P'roboblllty for • wo""n",,. ~
~ '~. "0 [ .
" 0 " po P'roboblllty for 11 wo. .l.n"".
10.2 ........•.......;
_ i ~::-_ i 1. . .. -;-r .op"akProbability for JZ wavelengths
\ ~~ '. j ..•~ .. ~
.. ... . po..... po.obob II)' for M w....I.n"".
~ - -+-~- - - - -~ - - - - -~-~ - - -~ - -
10-4 \. . ~
• I
!.. ..1.. \ ··· · -.~.·. I·······
: '.
•• •• Nonlln •• "Thr•• hold. PO",
- - FEe Thr •• hold
l--...,.:...:c=...:.:= = ---'
10-6 \. ~
'; . : ;
:
r.~ ;: i: :.- ;: ;: :
~
.
10 8 ·1 .'.. 1 '1 0
. I
i\ i :0 ~ .
o 1 2 3 4 5 6 7 8
Ppeak' [kW)
Fig.44. Peak power probab ility distr ibution due to pulse overlap in a IO-W average power 1024-
PPM multi-channel WDM MOPA TX. The pulses are assumed to be synchronized, so that the
overlap, when it occur s is complete. Note that the resulting peak powers occur at discrete levels,
depending on the integer number of pulses that overlap. Also shown are representative thresholds
for the onset of nonlinear effects (vertical dotted line) and FEC cutoff (horizontal dashed line)
[22,479] that break the chart into 4-quadrants. BER performance will be degraded for curves with
points contained within the upper-right quadrant.
The use of WDM has additional advantages . First, it increases the average input
to the TX power amp, which reduces the ampli fier gain and output ASE, and improves
the output signal-power to ASE ratio. It is also a means of overcoming electronic
bandwidth and performance limitations, enabling scaleable data-rates that can leverage
the available THz optical spectrum .
4. ReceiverTechnologies
Along with photon and power efficiency, important receiver metrics include reliability,
especially in the space environment, and performance in the presence of background
noise and atmospheric channel effects. In this section we will present the relative perfor-
mance of photon-counting, coherent-hornodyne, and preamplified receivers . Systems
based on these technologies show the most promise for next generation solutions for
ultra-long distance communication.
For a quantum-limited optical receiver, the SNR of received optical signals is
ultimately limited by the number of photons/bit, which is independent of the shape
the waveform. This leads to flexibility in the receiver to accept waveforms of arbitrary
shape. The average-power-limited properties of optical power amplifiers discussed in
section 3 allow similar freedom in waveform generation. This combination enables
the use of aggressive pulse shaping to efficiently match the TX signal waveform to
realizable RX filters. Furthermore, there is abundant optical spectrum available which
eliminates channel bandwidth limitations and enables excess spectrum to be traded for
improved sensitivity through modulation and coding .
Note that there is an important distinction between the quantum-limited SNR
defined in Eq. (32) and the optical SNR (OSNR) , which is a commonly used metric for
accessing communication performance in fiber-based WDM systems. As discussed
in section 2.3.5, the quantum-limited SNR is an absolute measure of SNR relative
to quantum-noise (hv) , equivalent to the incident number of photons/bit (for direct-
detection receivers) regardless of signal shape. This is not necessarily the case for
relative measurements of OSNR, which are often determined by measuring the signal
and out-of-band noise power within a specified bandwidth, typically 0.1 nm. The signal
power estimated using this technique can vary with pulse-shape, which can lead to dB-
class waveform-dependent OSNR errors without proper calibration. Moreover, since
the noise power can be much larger than the quantum noise, OSNR measurements of
this type are generally a poor measure of absolute receiver sensitivity-the relevant
metric for most free-space applications.
4.1. Direct Detection-PIN
The most common receiver for high-speed optical communications is the PIN-
photodiode-based detector" shown in Fig. 45. PIN detectors can be extremely wide-
band, with DC to 50+ GHz response commercially available. While PIN-based re-
ceivers are relatively simple, they are the least sensitive and require careful front-end
electronic design to achieve good high-rate performance [20,39,387].
10 The PIN or p-i-n photodiode is a semiconductor-based detector structure with an intrinsic (i) region in
between n- and p-doped region s. Relative to p-n photodiod es, PIN detectors have a thicker depletion region,
182 David O. Caplan
A
Data
Threshold
Fig. 45. Schematic of PIN-based direct -detection receiver with input waveform set ), electrical
amplification with load resistance RL ' gain G and noise figure FRF' postdetction filtering with
impulse response hRF(t) and effective electrical bandwidth Be [see Eq. (39)], prior to sample
and thresholding circuitry.
In section 2.3.5, we introduced the subject of shot-noise in the ideal detection

process [(see Eqs. (32) and (35)-(40)] . In practice other noise sources such as dark
current i« and thermal noise contribute to the SNR, with
(91)
a;h = 2q(is + id)Be , [amps"] , (92)
where a;h is the shot noise resulting from signal and dark current, and a;h is the
thermal noise generated in the load resister RL [216] scaled by the amplifier noise
figure FRF [20, Ill]. Incorporating these noise sources into the direct-detection SNR
expression in Eq. (40) we obtain
(93)
where P, is the incident signal power, and i s = R A / W P, is the signal photocurrent.

The detector responsivity R A / W = nq [h» = I mAlmWat 1.55 11m, for detection
efficiency fJ =rvO.8, causing a rv I dB reduction in detected power. When a;h » a;h'
Eq. (93) converges to the quantum-limited SNR in Eq. (40), as expected.
Amplifier noise is often given in terms of thermal noise current density ith
[pA/H z l / 2], which incorporates both thermal noise and amplifier excess noise. This
results in a current noise variance a;h = i;hB [amps''], and corresponds to an effec-
tive resistance R s« = 4kBT/i;h ' While thermal noise is a white noise source, RF
amplifiers can add frequency dependent 1/f noise, which further degrades the SNR at
frequencies below rv I00 MHz [20]. Relative to the dominant thermal noise, the few
nanoamp dark current typical of detectors at 1.55 11m can be neglected in Eq. (93).
To relate the signal and noise terms to OOK communication performance, we
introduce the Q-factor, which can be expressed as
RA /WPs (1- ER)

J2q(R A / w Ps(1 + ER) + id)B e + 2J(4kBT/RL)FRFB e '
(94)
which improves detection efficiency and reduces capacitance. resulting in wider bandwidth-making them
well suited for communications applicati ons.
where i 1,0 and 001,0 are mean signal photocurrent and noise terms for the ' I' and '0'
symbols, respectively, and ER = io/i 1 is the modulation extinction ratio (see section
3.5.2), ideally equal to O. With optimized thresholding, the bit-error-rate (BER) can be
estimated by the Q-function given in Eqs. (8) and (9), with the Q-factor above as the
argument. A 10- 9 BER is achieved for Q = 6.
To reduce the impact of thermal noise, the load resistance can be increased using
transimpedance amplifiers (TIA s). However, due to residual capacitance this limit s
the bandwidth of the system to about (21rRq-1 [20]. Commercial l-kf? TIAs are
available with Be = 7 GHz suitable for 10 Gbps communications. Typical thermal
1j
noise current for this class of devices is i th ="-'7 [pAlHz 2 ] , which corresponds to
Reff = 320 .0, or equivalently FRF ="-'5 dB. For these parameters, the 10- 9 BER
can be achieved with -2ldBm power incident on the detector, which corresponds to
"-'6200 PPB .
Another means of improving the performance of direct -detection RXs limited by
thermal noise is to use low-duty-cycle return -to-zero (RZ) impulsive coding. As noted
by Personick [39] in 1973, and more recently, demonstrated by Boivin et al. [343],
such techniques can lead to several dB enhancement of RX sensitivity. Discussion
of the transmitter considerations of using such waveforms is given in section 3.5.5,
and we elaborate on the use of this class of waveforms in section 5.1 for preamplified
receivers. For further details on optimizing the direct-detection RX, we refer the reader
to thorough analysis by Winzer and Calmar [345].
We note that while PIN-based RX performance is far from ideal, it provides a basis
for understanding the benefits of techniques used to achieve near quantum-limited RX
sensitivities discussed in the following sections.
4.2. Direct Detection Avalanche-Photodiode (APD)
APDs can improve performance over PIN detectors by providing internal gain within
the detection process, which multiplies the photocurrent by an average factor of M .
However as with any amplification process it is not noiseless. The resulting direct-
detection SNR expression becomes [20,121,236]
·2
t av g
SNRDD = 2
OO i + OO t2h 2qFAPD(i av g + id)B e + (4kBT/RL)FRFBe/M2'
(95)
where F APD is the excess noise factor associated with variations in M. The thermal
noise term is reduced by M 2 relative to the PIN detector, whereas the shot-noise term
is increased by FA PD, which generally grows with M. Typical values of M range from
"-'3 to 100, corresponding to FA P D values of 2 to 10, although these values are vary
considerably depending on wavelength, data rate, and technology. At rates of 10 Gbps ,
M ="-'3 to 10, and FAPD =,,-,4, providing about a IO-dB benefit over PIN-receivers,
with 10- 9 BER achieved for -29 dBm or "-'I000 PPB receiver sensitivity (see, e.g.,
[388,389]). At lower rates-less than I Gbp s, APDs have demonstrated performance
in the 100 to 200 PPB regime [390-392]. However, PIN-based RXs offers superior
bandwidth and dynamic range ofoperation, and are less temperamental, offering robust
performance over a wide range of environmental conditions.
184 David O. Caplan
4.3. Direct Detection-Photon Counting
Qualitatively, photon-counting detectors can be viewed as an extension of APD-type

detectors with infinite gain in which, a digital output signal is generated for each
detected photon. Due to the binary nature of the detection output , noise in the detection
process appears in the form of dark counts or varying detection efficiency, and the
detector is limited to counting at most one photon per reset time. As a result, the
detector is blinded after a detection event, missing all incident photons until reset,
an effect which is referred to as blocking loss [34,35]. Detector arrays can be used
to extend the dynamic range of detected signal to multiple counts per-interval and
mitigate blocking loss, although this comes at the expense of increased dark count
rate.
Photon -counting receiver architectures have been proposed [1,26,43,169,170,393]
and realized at Mbit/s [44] and <Gbit/s [33,34] data rates with the best demonstrated
coded RX sensitivities near I photonlbit (PPB), with potential for improvement to
multiple-bits/photon sensitivities and greatly simplified processing due to the digital
nature of the counting process.
However, these photon-counting RXs are presently limited to power-starved links
with little or no background noise and < "'Gbit/sec rates due to dark-count and reset-
time constraints, which preclude their use with the sun in the field of view. Rates are
also limited by the need of near-Shannon-limited-capacity coding [42,394] needed to
avoid performance degradation due to dark counts, blocking losses, and background
noise. These powerful FEe codes place additional burden on the RX electronics in
terms of SWAP and bandwidth constraints[45]. Also, the suitability of the key detector
technologies, Geiger-mode avalanche photodidode (GM-APD) arrays [32] and super-
conducting single photon counting detectors SSPDs [222] has not yet been established
in the space environment. Si-based GM-APDs, for example, have shown sensitivity
to radiation [395], and SSPDs require cooling to cryogenic temperatures, requiring a
significant overhead in SWAP.
For ground -based receivers, however, where reliability and receiver SWAP are not
a driving limitation, these technologies offer significant potential. Unlike coherent and
preamplified RXs, these photon-counting detectors can receive multiple spatial modes
and can efficiently collect signals distorted by the atmospheric channel without the need
for wavefront correction. Furthermore, the net detection area can be efficiently scaled
in distributed telescope arrays without the need to build large and costly telescopes
[396].
The scalable-multimode-detection capability of photon-counting RXs comes with
the expense of increased sensitivity to background noise, which is enhanced by the
broad, several-hundred-nanometer spectral response of the detectors. However, the
background can be mitigated with the use of single-polarization narrow-band low-loss
high-rejection spectral filtering. Such filtering can be implemented using a cascade
of loss-loss volume-Bragg-grating [397] and etalon filters with ",0.5 -I dB insertion
loss and noise-equivalent bandwidth of 0.0 I to 0.1 nm over a spectral range exceeding
500 nm .[398]. Thus, many of the key elements needed for high-performance photon-
counting receivers have been reduced to practice [44]. As the detector technologies
mature, they may become the technology of choice for future ground and space-based
receiver solutions.
P,(I)
A
Data
Threshold
Optical PLL
Fig. 46. Schematic of balanced coherent homodyne PSK receiver.
4.4. Coherent Homodyne Receivers
Coherent receivers are another means of boosting the shot noise relative to the thermal
noise though coherent gain in the detection process . Homodyne detection of binary-
PSK provides among the best theoretical RX sensitivity and is spectrally efficient.
Coherent mode-matching requirements combined with matched electrical filtering also
provide superior rejection of background noise over multi-mode receivers, which could
enable, for example, improve communication performance with the sun in the field of
view as noted in section 2.3.7. Information is carried in the phase <Pm(t ) of the optical
carrier and demodulated using a balanced receiver shown in Fig. 46.
The signal is received in one port of a salsa
coupler and a local oscillator (LO)
laser of the same frequency is injected into the other input port. The signal and LO
mix at the two balanced square-law detectors, and generate a difference photocurrent
that is given by [20,156,236]
.
Llt(t)= ((1Jl-'T]2) (PLO ~ cos(21rLljt+Ll<l>+ <I>m(t» ) hv
r. ) +2'T]avgyPsPLO
-2-+2 q
--+
2'T]avg ~
7)1 =7)2 =7) - - v PsPLO cos (Ll<l>m(t» [amps], (96)
hv
f 8=ho
where 1]av g = (1]1 + 1]2) /2 is the average detection (ideally equal to one), !if and
!i<p are the frequency and phase differences between LO and signal, that are set to
zero with feedback from an optical phase-locked loop [153,156,166, 99], and <Pm (t)
contains the (0 or 7l") phase modulation . The resulting SNR is given by [20, 156, 236]
(97)
where the shot noise term is due to the LO power and R avg = 1]avgq /hv. The local
oscillator noise term alo is the due to excess relative intensity noise (RIN) in the LO,
which vanishes when the two detectors have the same detection efficiency [400,401] .
In this case, the shot noise power can become the dominant noise source by increasing
PLO. In the limit when a;h » a;h:
186 DavidO. Caplan
SNRp SK = 2'T/~~~s = 4SNRoo , (98)
four times the SNR of ideal direct-detection. Even in the shot-noise limit, the SNR is
proportional to the detection efficiency 'T/a vg, meaningany detection losses, including
spatial or polarization mode-mismatch, directly impact RX sensitivity.
However, in practicethe high-sensitivity potentialof opticalPSK has not yet been
realized, in part due to challengingcomponent and laser linewidth requirements, the
need for high-detection efficiencies, signal-LO polarization mismatch, and difficulties
associated with phase-locking the local-oscillator [156,399]. In the Gbitls regime the
best reported uncoded PSK performance is ",,35 PPB at ",,6 Gbitls and ",,80 PPB at
",,8 Gbps [158],providing little performance benefitoveropticallypreamplified DPSK
[24,25,133], which is WDM scalable and easier to implement (see Differential Phase
Shift Keying (DPSK) in section 5.2). At lower rates < I Gbps, PSK performance is
somewhatbetter, with uncoded RX sensitivities of 16 and 20 PPB at 4 and 565 Mbps,
respectively [157,195].
Some of the reduction in PSK sensitivity at high data rates can be understood
through the expression for SNR degradation in Eqs. (99) and (I (0). To the extent that
the shot noise does not overwhelm all other sourcesof noise in the receiver, the SNR
is degraded by
( 2)
.1, = 10 log 10 1 + -2-
crth
crs h
= 10 log 10 (1 + -..- ) ,
teq
t avg
(99)
. 2k BT
t eq = - - [amps], (100)
qRL
where i e q is the equivalentphotocurrent needed to make cr;h = cr;h ' For RL =50n at
290K (room temperature), i e q = I rnA which corresponds to ""I mW.
In order for shot noise to dominate thermal noise, large photocurrent or load
resistanceis required. For a maximumphotocurrent of ""I rnA and RL =50n the shot
noise and thermal noise are equal and the SNR is halved. This penaltycan be reduced
by boosting RL, at the expensiveof limitingthe bandwidthdue to residualcapacitance
in the TIA. Furthermore, as noted in the Direct detection-PIN section, the effective
resistance in wide band TIAs is often limited by excess noise to ",,300 n, which for a
""I rnA current reduces the RX sensitivity by > II 2 dB.
Thus, the design of the front-end receiver electronics in coherent receivers re-
main critical [402]. At high data rates, achieving near-quantum-limited performance
with coherent-homodyne PSK receivers is challengedby a combination of detection-
efficiency, photocurrent, and transimpedance-gain limitations, whereas at lower data
rates, laser-linewidth, RIN, and IIf noise impose additional constraints [20]. In the
following sections,we discuss the use of low-noise high-gain wide-bandwidth pream-
plification as a means of overcoming thermal noise and bandwidth limitations in the
detection process,and reducingthe impact of the front-end design on receiverperfor-
mance.
4.5. Optically Preamplified Direct Detection
Preamplified RXs can directly leveragethe field-tested heritage of telecom-type 1.55

Jlm technologies that are compatible with operation in the space environment and
A
s(t) Data
S(t)
Threshold
Fig. 47. Schematic of an optically preamplified receiver with input signal s(t), optical and RF
postdetction filters ho(t) and hRF(t), respectively, optional polarizer, and sample and thresh-
olding circuitry.
have demonstrated the best sensitivities at high data-rates (>",-,Gbit/s) of 25-30 PPB
for uncoded DPSK and 7-10 PPB with coding [24,25,133]. With the use of M-ary
orthogonal modulation formats, RX sensitivities can approach the 1-2 PPB regime
albeit with substantial bandwidth expansion [15,22]. In this case, the use of hybrid
modulation formats including frequency, position, and polarization modulation along
with WDM rate scaling can be used to access the many THz available in EDFAs at
1.55 JLm and/or YDFAs at 1.06 JLm, and overcome electronic bandwidth limitations.
We address these modulation possibilities further in section 5.3.
In sections 3.5.3 and 3.5.5, we introduced some of the high-level characteristics of
EDFA performance from the transmitter perspective . Here we provide additional detail
on the impact of amplifier noise on SNR and receiver sensitivity. The noise processes in
EDFAs and impact on receiver performance has been discussed extensively in the lit-
erature e.g., [20,36-38,76,77,126,336,340,403-406] . The basic optically preamplified
receiver setup is shown in Fig. 47.
The signal field s(t ) with power Fs(t) = Is(tW is amplified by a factor of G
which adds GN s p noise photons per mode with noise power spectral density N ASE
=mhvGN sp [WlHz]. Here, m is the number of polarization modes, and N s p is the
spontaneous emission factor of the ampli fier (NF =2Ns p ) . Out-of-band amplifier noise
(ASE) is removed by optical filter H0 (f), with field impulse response li; (t). The mean
detected signal current is given by
(101)
where B o and B; are the noise equivalent bandwidths of the filter Ho(f) and signal
S(f) spectra respectively, and
y(t) = s(t) * ho(t) = it s(t - T)ho(T)dT (102)
is the filtered signal field incident on the detector. Due to the square-law detection
process, like-polarized signal and noise components mix, generating S x ASE and
ASE x ASE noise terms in addition to shot-noi se a;h
and thermal noise a;h'
After the
postdetection filter noise-terms are given by [344,345]
188 David O. Caplan
!
00 00
2
alsE xAsE = d71hRF(t - TI}! dT2hRP(t - T2)2Nl sE IRa (71 - T2)1 , (103)
I:
-00 - 00
a~XASE(t) =; Re{ dT1hRF(t - TI}
X I: dT21 hRF(t - T2)2NASEY*(TI}Ro (71 - T2) Y(T2)} , (104)
a ;h= 2q(R A/ WGPs + mGNspBo + id)B e, (105)
a;h = (4kBT/RL)FRFBe, (106)
where R; (t) = h o(t) * b; (t) is the filter autocorrelation, hRF (t) is the RF (electrical)
filter impulse response, and Be is the noise equivalent bandwidth of RF filter given
in Eq. (39). For square optical filter Ho(f) and Ps(t) constant, Eqs. (103) and (104)
reduce to [20, 37, 126]
2 2 2 2 2 P,
aSxASE ~ 4R A/ WGPsNASEBo = 4q NspTJ G hIJBo, (107)
alsE xAsE ~ mRl/wNlsEBe(2Bo - Be) ~ m (NspTJqGBo)2, (108)
and the expression for SNR becomes
(P sR A / W G)2
SNRPreamp = 2 2 2 2
a sh + a th + a SxASE + a ASExASE
(p !1!1.)2
= ,,2 ,,2 s hI' 2 P B' (109)
~ + ~ +m(NspTJqBo) +4q2NspTJ2~~
where we have assumed Bs = rvBe, a subject discussed further in section 5.1.2.

For large gain, optical bandwidth, and bit rate (e.g., G > rv 40 dB, Be > I GHz, and
Rb > IGbps), the a;h and a;h terms are overwhelmed by the a~ x ASE and aiSE x AS E
=
terms . Form = I and N sp I, when
Ps Bo
= (110)
hIJ 4 '
a~XASE and aiSEXASE terms become equal [20]. Normalizing Eq. (110) by the bit
rate this condition becomes
~ = PPB = B o ~! . (I II)
hIJRb 4Rb 4
For these filtering condit ions (B; ~ Be ~ Rb), this implies that whenever the number
of photons/bit » 1/4, a~ XASE » a iS E XASE ' In this case, Eq. (109) becomes
(112)
and we approach ~ shot-noise limited direct-detection performance as expected due

to the 3-dB noise figure of the optical amplifier. Note the performance in this limit
is not impacted by thermal noise or detection efficiency. For high-sensitivity applica-

tions, G >rv40 dB is often necessary depending on the data rate and amplifier design,
sometimes requiring more than two gain-stages and photocurrent > rv I mA to over-
come all other sources of noise. This makes high-current-capable high-speed 50 n
photodetectors attractive [407,408] . Wideband transimpedance amplifiers can reduce
this power requirement somewhat, but scaling the EDFA output power to the necessary
levels is relatively straightforward and it is often challenging to obtain the combina-
tion of bandwidth, waveform fidelity, and dynamic-range of operation of the filtered
EDFA-PIN photodetector combination with the addition of RF amplifiers . The satu-
rating characteristics of the EDFA combined with adjustable pump power in the final
gain stage can be useful for performing automatic gain control when operating over a
wide range of power-levels or data rates.
The addition of a polarizer eliminates half the ASE (and other orthogonally polar-
ized background [47,223]) , leading to theoretical sensitivity improvements, which are
typically about 0.5 dB in practice [41]. This require s some form of polarization control
in the TX, RX, or both. Unlike coherent RXs, for preamplified RXs this is an option,
and we note that control of polarization can increase sensitivity and extend modulation
options as discussed in section 5.3.
Like coherent receivers, single-polarization optically preamplified receivers when
implemented with matched optical filtering provide true-single-mode background re-
jection (see section 2.3.7), and therefore can perform well with the sun in the field of
view (as long as coupling and tracking elements continue to function) . In fact, since
preamplified RX SNR is already degraded by ASE noise, the change in SNR due to
solar background is less than that for coherent receivers .
The combination of low-noise, high-gain and bandwidth available in preamplified
RXs relaxes many of the downstream component requirements in the receiver. Since
EDFA gain and power in the preamplifier are inexpensive, they are a cost-effective
means of achieving near quantum-limited receiver sensitivity over a wide range of data
rates and modulation formats. Furthermore, rare-earth-doped EDFA and YDFA tech-
nologies have been shown to be robust in the space environment [198,204,380,409],
making it the present technology of choice for future space-based applications. Meth-
ods of optimizing preamplified RX performance are discussed in the next section .
5. Performance and Implementation Considerations
For average-power-limited optical transmitters operating over the free-space channel,

without dispersion or nonlinear impairments, the signaling waveform can be chosen
arbitrarily without sacrificing transmitted power (subject to constraints discussed in
section 3.5.5). This enable s the selection of robust waveforms that are well suited
for optimum communications performance and facilitates all-optical matching of the
transmitter to the receiver [10]. With an abundance of available optical spectrum,
waveforms can be chosen to provide nearly matched performance at multiple rates,
diminished lSI, with little or no power penalty at the transmitter end. This approach is
a paradigm shift from conventional techniques based on peak-power and bandwidth-
limited RF designs , where there is always a significant trade between the choice of
waveforms, filtering, and the power transmitted .
190 DavidO. Caplan
Transmitter Receiver
>-- t-i""-u rl Channel
Fig.48. Schematic of an average-power-Iimited MaPA TX withoutput fieldwaveform STX (t)

separatedfroma preamplified RX by an attenuating channel.A band-pass opticalfilteris followed
by a polarizer, square-law detector, and postdetection sampling. The electrical responseof the
receiveris assumedto be relatively wideband, so that RF filtering is not explicitly included.
In the remainderof this section,we will combinewell knownconceptsof matched

filtering with the use of pulse shaping in the transmitter and optical filtering in the
receiver to achieve robust high-sensitivity performance in optically preamplified re-
ceivers.
5.1. Waveform and Filtering Considerations
For high-rate and high-sensitivity applications, the MOPA TX and optically prearn-
plified RX setup shown in Fig. 48 have demonstrated the best performance. It is well
knownthat matchedoptical filtering yields the best performance for a signal corrupted
by AWGN noise [156,410]. The SNR relative to the optimal matched filtercondition
is given by [18,120,411]
(113)
where STX(t) is the optical signal (field) waveform and ho(t) is the optical receiver
filter (field) impulse response. The resulting SNR (.1,) is maximized whenever
S(t) = h(-t) ~ S(f) = H*(f) . (114)
The ratio in Eq. (113) represents the filtered signal waveform (power incident on the
detector)as a function of time, normalizedby the productof signaland filterwaveform
energies. The best SNR is achieved by sampling this waveform at its maximum.
The matching condition does not specify the waveform; rather it specifies the
optimumrelationship betweenthe signal STX (t) and filter ho(t), that occurs whenthe
filterimpulseresponseis the time-reverse of the signal waveform. For waveforms that
are symmetric in time, Le., s(t) = s(-t), the matching conditions reduce to
FT
~ (115)
meaning the signal and filter impulse response have the same shape in both the time
and frequency domains.
Laser communication transmitter andreceiver design 191
Early efforts towards optimizing preamplified RX sensitivity for high-rate op-

tical systems focused on matching square non-retum-to-zero (NRZ) ·signal wave-
forms[410,412] or optimizing performance with available Fabry-Perot (FP) optical
filters .[128,404,4 13,4 14]. The NRZ waveform is particularly difficult to match as we
show later in this section, and is prone to inter-symbol interference (lSI) penalties since
the NRZ waveform energy is uniformly distributed between the symbol boundaries.
This is especially problematic when received with the asymmetric exponentially de-
caying response of the FP filter, which requires a large optical bandwidth, B; = 3.7 Rb
and tight postdetection filtering in order to avoid significant lSI penalties [413]. The
optimized FP bandwidth in this case is more than seven times larger than the best FP
filter without lSI (see Table II).
Matched optical filtering is desirable since it minimizes ASE or other out-of-band
background optical noise, such as black-body radiation (see section 2.3.7) or inter-
channel-interference (lei) from adjacent WDM wavelengths, prior to conversion to
the electrical domain . In this regard, all-optical matched filtering is desirable, from
the standpoint of achieving optimum sensitivity [36,156] and a secondary benefit of
improved spectral efficiency, since it enables tighter channel spacing [13].
The use of pulse-shaping in average-power-limited optical transmit ters [10] im-
proves RX sensitivity for two primary reasons . First it can reduce lSI penalties and
second , it can facilitate the matching of the transmit waveform to realizable optical fil-
ters. This has led to high-sensitivity demonstrations (--",2 dB from theory) of multi-rate
variable-duty-cycle PPM [11,12], and RZ-OOK and RZ-DPSK [129] using FP optical
filters.
Fortunately, there are many good waveforms that are easily generated (see section
3.5) and a wide selection of available filters such as single and multi-pass Fabry-Perot
and dielectric filters, arrayed-waveguide (AWG), diffraction [415], and fiber Bragg
(FBG) grating filters [133,416], and integrated waveguide filters [417]. Many of these
can be customized to achieve a desired transfer function [418].
For the TX and RX in Fig. 48, the optimum signal-lo-filter pulse-width-ratio {J,
defined as the ratio of signal FWHM to filter-impulse-response FWHM power wave-
forms, is calculated for a variety of waveform and filter type comb inations summarized
in the Matching Matrix in Table II. For each waveform, the time-bandwidth product
(L1tI L1fI) is given for 3, 10, and 20 dB signal power levels. The SNR loss (L1,) is
given in dB for the optimized (3, without post-detection filtering. Note that lSI effects
are assumed to be negligible, which is valid if these pulsed waveform s are of suffi-
ciently low duty-cycle . No electrical filtering is included in this optimization, implying
that the receiving electronics is relatively wide-band relative to the optical filter.
Also listed is the ~ dB (3-margin, which is the percent change from the optimum
(3 that can be tolerated with less than ~ dB reduction in SNR . This is an important
measure of the SNR or matching robustness for the signal-filter pair and the sensitivity
to deviations from the optimum . In this regard, the matched square waveform is a poor
performer, a subject discus sed further in section 5.1.2.
The example below illustrates how to use the matching matrix in Table II .
Determine the optimum RX filter bandwidth for a IO-Gbps data stream
formed by RZ-33% pulse carver.
From its name and as shown in Fig. 29, the FWHM of the RZ-33 % is 33% of
the period. For 10 Gbps, this would correspond to a 33-ps FWHM . Using a
192 DavidO. Caplan
Table 11. Matchingmatrixof 14signalingwaveform types (powerpulse shapes)and a selection

of four receiverfilters. The SNR loss, .el, is calculatedfor the optimum {3 To/Ts, defined as =
the ratio of signalFWHM to filter-impulse-responsepwjjjq (power) waveforms. Also listed for
each filter type is the ~ dB {3 margin. which is the percent deviation in {3 with < ~ dB SNR
change.
Waveform
Type 3dB
,
Ga ussia n.
exp(.~2/a2)
0.44 1.41 2.9 0.5 ! 0.60 ! 50 2.3 I 95 0.3 1.0 10 0.0 1 i 65
!
! ! !
l-~;'~;~Il) 0.49 1.54 2.1 0.5 ! ! 0.60 55 2.2 100 0.3 1.1 10 0.0 1.1 60
Rl .40% I I
0.55 1.51 2.6 0.4 ! 0.10 ! 45 2.1 80 0.3 1.1 10 0.0 1.1 65
(-Ga ussian)
i j
Rl·SO% 0.66 1.62 4.0 0.3 ! 0.80 ! 35 2.9 90 0.3 1.3 70 0.1 50
Rl-S3 %
(Raised Cos)
0.63 1.62 2.4 0.3 ! 0.10 !~ 2.5 ! 110 0.3 1.2 15 0. 1 1.3
Rl.66·1oICSRl) 0.13 1.62 2.9 0.2 I ! 0.80 40 3.1 1 85 0.3 1.5 55 0.2 1.3 66
Soch '( lItl
(Solllol1)
0.31 1.34 3.6 0.1 ! 0.60 ! 50 2 ! 100 0.3 0.9 75 0.0 0.9 60
Ga ussia n'
expf.•'/s'l'
0.68 1.58 3.4 0.3 i! 0.80 i 35
I
2.9 90 0.3 1.5 55 0.1 1.4 SO
Gaussia n'
exp(_t'/s2)4
0.80 1.56 4.0 0.1 ! ! 0.90 25 0.9 3.5 10 0.4 1.7 SO 0.2 1.1 45
Squar e
,.c'lllt)
0.89 1.49 3.6 0 i ! 1 10 0.9 3.5 80 0.6 1.8 45 0.5 j
I
1.1 50
FpC
1,'J6I'., pl·2I1tl
0.40 1.51 2.1 0.5 i 0.59 i 50 1.5
I- - ;-
2.5
- ;;--- I-
90 0.1
-+-
1.0
- ;-
80 0.1 !
- I'- --+-
1 ! 65
-+- -l
FP'
(,'m'., pl·Wt]
0.39 1.51 1.5 0.5 ! 0.25 i 50 1.6 2.4 i 100 0.8 1.0 80 0.2 i Ii 10
FP'
(1I' expf·2titl
0.35 1.51 4.6 ! !
0.6 8.33
- -! - -! - -
50 1.8
!-
2.2 ! 120
- ;-
1.0 1.0
- -. - - - -
80 0.3 i Ii 10
- ,--, -
FP
0.p(.2.kl 0.11 1.11 1.3 0.9 !; 0.01 !
;
80 2.1 ! ;
1 ! 130
i
1.8 0.4 105 1.0 i 105
0.4 ,
matched Gaussian filter. the optimum filter impulse response ({3 = 1.1) has a
FWHM of ",30 ps. From the time-bandwidth product. this corresponds to a
14.7 GHz filter. which provides nearly optimum SNR performance (",0 dB
SNR loss) and can accommodate up to a ",60% bandwidth mismatch with
only 0.5 dB added penalty.
The relaxed tolerances in this example can be used to assist in manufacturing yields
and reduce component cost. The flexibility in filter bandwidth permits, for example,
widening the filter to mitigate lSI or narrowing the filter to dimini sh ICI with little or
no SNR penalty.
5.1.1. Symmetric Filtering
Pulse-shaping at the transmitter can be exploited to implement near optimum com-

munications waveforms with non-ideal components (modulators, drivers, and receiver
filters). Consider the TX in Fig. 37 with a wide-band optical source generating short
(impulse-like) pulses in time followed by a high-gain APL amplifier with optical fil-
tering. Passive optical filtering can be used to perform TX pulse shaping such that
STX(J) ~ Hnet(J). A symmetrically filtered RX has the same net optical filtering,
H net (I), and if the filter time-response is also symmetric, match filtered performance
can be achieved . For filters with an asymmetric response , such as the Fabry-Perot,
with multi-pass or cascaded filtering, the response can be made progressively more
symmetric (and Gaussian-like). In this manner, near optimum communications per-
formance can achieved by cascading readily available filters which would otherwise
yield sub-optimal performance when used individually [10]. Mathematically, this is
equivalent to convolving a function with itself, which qualitatively tends to spread a
function making it more symmetric .
The performance benefits of using symmetric TX and RX filtering for cascaded
asymmetrical filters can be observed for the case of the commonly used Fabry-Perot
filter (FP n ) listed in Table 11. Without additional filtering, a first-order FP incurs
a 2.7-dB SNR loss. For the second order (Fp 2 ) , which can be implemented by a
double-pass through a single FP, the SNR falls about I dB from optimum. A three-
pass configuration yields a mere ""0.6 dB deviation , and as expected, departure from
the optimum continues to diminish as additional iterative passes are made, and the
time domain signal becomes increasingly symmetric. Note that along with increased
symmetry, the time and frequency domain characteristics of FP n start to resemble
those of a Gaussian, the subject of the next section .
5.1.2. Gaussian Waveforms and Matched Optical Filtering
Combining symmetricfiltering with the use of matched Gaussian -like waveforms [14],
nearly optimal performance can be achieved with relaxed filter tolerances [13,133].
Gaussian waveforms are relatively easy to generate (see Pulsed Waveform Generation
in section 3.5.2) and Gaussian filters can be realized with filter technologies such as
diffraction gratings [13,415] , AWGs [419,420], apodized FBGs [133,421], and cas-
caded filters discussed in the previous section . The Gaussian probability distribution
function is known as a minimum-uncertainty-packet in quantum-mechanics, providing
the minimum variance in two conjugate variables (e.g., position and momentum, or ar-
rival time and energy of a particle). As a signaling waveform, Gaussian waveforms have
a narrow time-bandwidth product, fJ./I fJ.tI =0.44 (at the 3-dB power point), making
them well suited for both TDM and WDM communications. The Gaussian function
is part of a class of function s known as "Self-characteristic functions" [422], which
have a Fourier transform of the same form, i.e., a Gaussian time-domain waveform
also has a Gaussian spectrum. Another well known optical waveform which shares this
property is the soliton, which is Gaussian-like and also has a narrow time-bandwidth
product (see Table II). The soliton waveform is considered for use in ultra-high-speed
fiber-optic links since the waveform maintains its shape due to a balance of disper-
sion and nonlinearities [111,236,423,424] . In this regard, fiber-based communications
using solitons is similar to communication over the dispersionless free-space channel.
5.1.2.1. Relaxed Filter Tolerances
In contrast with self-characteristic minimum-uncertainty waveforms, the square-NRZ

signaling waveform, which is completely contained in the time-domain, has a fre-
quency domain spectrum with infinite extent, making it a poor choice for use in dense
WDM systems . Furthermore, as noted earlier, the square waveform is susceptible to
inter-symbol interference (lSI) penaltie s and requires tight filter tolerances to avoid
194 David O. Caplan
' -" ,- (a)

't--
(e)
M
U
FHWM ' - l
0 .5 l
" \\, --'--
fWH"'1·-'
-;:- #' -,
~ 0 .:' '-~~
do -2 ·1 0 1 2
· ~I D· DE~~lj
(b)
~2 ·1 - 0- 1 2
Time. (Linear]
Fig. 49. Signal Si(t) = Is(t)1 2 in solid line and filter impulse response hi(t) = Ihf(tW in
dashed line for Gaussian and square waveforms in (a) and (b) above. The pulse width ratio j3 is
defined as the signal FWHM to filter-impulse-response FWHM. For the waveforms in (a) and
(b), j3 =2. (c) Shows the reduction in SNR as a function of j3 for matched Gaussian and square
filters.
SNR penaltiesas illustrated for the case of Gaussianand square matched filters in Fig.
49.
In Fig.49(a), a widerGaussiansignal waveform (solid) and filterimpulse response
(dashed) are shown.The full-width-half-max pulse width ratio for the signal and filter
response is defined as {3, where the matched filter occurs for (3 = I. Similarly,a wide
square-signal-waveform and filter-response areshownin Fig.49(b). Forboth waveform
types illustrated,(3 = 2. ExcludinglSI effects,theSNR penalty (,1,) is calculatedusing
Eq. (113) and plotted for Gaussian and Square waveform types in Fig. 49(c). While
both Gaussian and square waveforms have no penalty for the matched condition ({3
=0), the square degrades much more quickly from waveform-filter mismatch. For a
60% mismatch ({3 = 1.6), the Gaussian incurs ",0.5 dB, whereas the square waveform
4
degrades by ",2 dB. For SNR loss < dB the square filter bandwidth must be within
10%of the signal bandwidth,effectively six x more or sensitiveto mismatchthan the
4
Gaussian at the dB point. For {3 = 2, the square waveform SNR degrades by 2 dB
more than the Gaussian, illustrating that the Gaussian waveform is more robust and
easier to match than the square.
In an attempt to capture both SNR magnitude and its matching tolerance for a
particular waveform combination in a single term, we introduce Ll,eff , defined as the
average SNR for {3optimum < {3 < 1.6, i.e.,
,1, ({3) d{3

A. 1 1. 6 !3o p t im u m
.Bop t m m m
(116)
.w'eff = -~------
0 .6{3optimum
,1, is listed in Table 12and displayed in Fig. 50 for severalsignaling waveforms and
filter types, sorted by Ll,eff, and shows Gaussian-like properties and better effective
SNR.
Laser communication transmitter and receiver desig n 195
Table 12. Effective Matching-matrix. Effective SNR loss, L.\')'eff resulting from the overlap of
several signaling waveform types (power pulse shapes) with a selection of RX optical filters.
sorted by L.\')'eff from Eq. ( 116). lSI effects are neglected.
Signal I ~cel ver Filter, hl(t)

---------------- --
Waveform, stCt) II Gauss I, _FP' T.FP' I~~~~.sI21[~~~~ JI.Gaus
141
I Square Ii .. FP
Gaussian 0.2 0.3 0.3 0.3 0.4 0.5 0.8 1.1
RZ-33% 0.2 0.4 0.4 0.2 0.5 0.4 0.7 1.1
- - -
RZ40% 0.2 0.4 0.4 0.4 0.5 0.4 0.7 1.1
--- - - - - - - I - -- - - -
Soliton 0.2 0.4 ~4_ 0.4 0.5 0.5 0.8 1
RZ-50%
---
0.3 0.4 0.4 0.4 0.5
- 0.5
- - - 0.7 - - 1.1
-
RZ-53% 0.3
---
0.4
---
0.4
-
0.2
-
0.5
- - 0.4
--
0.6 1.1 _
RZ~6% 0.3 0.6
- --
0.6 1.10.4 0.6
- 0.5 0.6
SuperGauss(21
-- - - 0.4
-- --- --- --- --- --- -- -
0.3
-- - --- ._--
0.4 0.3 0.5
--- --- ---
0.5
-
0.7
- - - -1
-
FP(' I 0.3 0.7 0.7 0.4 0.8 0.5 0.8 1.6
FP(31
--
0.3
- --- --
0.7 0.7
- - - - --- --- --- ---
0.4 0.9 0.6 0.8 1.7
- - - - -- - - - - - - - - - - - - - - - - - -
Fpl21 0.4 0.9 0.9 0.5 1.1 0.6 0.7 1.8
- - - - - - - - - - -- - - - - - - - - - - - -
SuperGauss(41 0.5 0.5 0.5 0.4 0.7 0.5 0.7 1.2
Square
-0.7
-- - - - - 0.7
- - 0.7
- - - - - - - 1.2
- - 1 - -1 -
- - -
FP
- - - - 0.7
1.0
-- -- - - - - - 0.9
1.6
-- -- 1.6
- --- ---
1.0 1.9 1.0 1 2.7
-'- - -:5- 3-0

11_ ·
-~.-~.~~.~ . ~.~~:~:~·-r~~·~ ·-~~~··~-~·~·~I
_. _ ~._ .. - •
. • . . . • 20·25 .
0 1 5-2 0
0 1 0-1.5
. 0 5-1 0 • .•. · •
.' 0 0 0-0 5
Fig. 50. Effective SNR loss for a variety of signal and filter shapes sorted L.\')'eff . The more
symmetric, Gaussian-like waveforms yie ld better SNR performance, wherea s the sharp-edged
square and asymmetric FP filters incur the bigge st losses .
5.1.2.2. Reduced Sensitivity to Timing Jitter
The SNR analysis in section 5.1 assumes that the received waveform is sampled at its
maximum . However, deviations from the optimum sample point due to noise in the
clock-recovery processes [236,419,425] or other sources of timing-jitter [426] reduce
196 David O. Caplan
Square
Pulse:
GaUSSian~
a
Slt l
t
*
a
hlt l
t
= y.(tl ~.
T 2T t
Pulse: T
*
~ -
h(l) i /"'0.. _
y~
...) . . ---:c
T --10'.•.••.••.
Fig. 51. Convolution of signal and matched filter field-impulse-response h(t) for square (top)
and Gaussian waveforms (bottom) of the same pulse-width. The filtered signal field y( t) is the
solid line and signal power (e.g., photocurrent after square law detection) is in the dotted line
(13).
·0.5 0
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Normali zed Ti ming deviation, L\t/t
bll
Fig. 52. SNR sensitivity to timing jitter for matched Gaussian-like and square-NRZ waveforms
(excluding lSI) (13).
the SNR at the thresholding circuitry which degrades RX performance. Such effects
are dependent on waveform shape as shown in Figs. 51 and 52, where, square s(t)
and Gaussian g(t) field waveforms are received by their respective matched filters
h(t) . The output from the matched optical filter y(t) = s(t) * h(t), with detected
photocurrent i( t) = 1Jly( t W,
indicated by the dotted-lines in Fig. 51.
The convolution of the square waveform and its matched filter yields the triangular
waveform, which is subsequently narrowed by the square-law detection processes.
The best SNR is achieved by sampling photocurrent at the peak. In this case, the peak
is a relatively narrow target, which tends to magnify the any SNR penalties due to
fluctuations in the sample time .
For the Gaussian waveform, the matched-filtered output is broadened 41% by the
convolution process, but the waveform remains Gaus sian. After square-law detection,
the received photocurrent still has a Gaussian shape, but the pulse width is reduced by
y'2, so that it has the same pulse width and shape as the incident signal field. Note
that the peak sampling point for the Gaussian is a much broader target than that of the
square. The resulting SNR degradation (.1,) as a function of time deviation from the
optimum sampling point is shown in Fig. 52.
The sensitivity of matched-Gaussian-like waveforms to timing errors and sam-

pling duration is significantly better than that of matched-square waveforms as shown
above. This can reduce the impact of timing-errors from a variety of sources (e.g.,
Gordon-Haus timing-jitter [426]), and relax tolerances for sampling speed, accuracy,
and stability of clock-recovery and detection hardware. This is especially important
for M -ary PPM and high-rate communications which are more sensitive to the effects
of timing jitter. For example , at 40 Gbps the timing precision required for less than
0.1 dB penalty is "" I ps for 33% RZ waveforms . Despite being three-times wider, the
square-NRZ waveform is over three-times more sensitive than 33% RZ to timing jitter,
requiring ..1t < ",,300 fs for < 0.1 dB penalty. For only ..1t = "" I ps SNR is reduced
by ""0.35 dB, excluding additional lSI penalties .
5.1.2.3. Combined Optical and Postdetection Filtering
For systems where electronic bandwidth is a limitation, both optical and RF postdetec-
tion filtering can be adjusted for better overall RX performance. Practically, obtaining
RF bandwidths wide relative to the data rate is not always an option, but SNR loss
(..1')') can be avoided to a large extent, by increasing the optical filter bandwidth ac-
cordingly. As noted earlier, a similar trade between optical and RF bandwidths can
be used to compensate when suitable narrow (matched) optical filtering is unavailable
[345,346,404].
The SNR penalty as a function of optical and electrical bandwidths [344,345,404-
406,427,428] has been evaluated for a variety of waveforms and filter shapes, often
numerically and for the case ofoptical filters with bandwidths wide relative to the signal
spectrum. Below, we evaluate Gaussian-like signal-waveforms and Gaussian-optical-
and postdetection-RX filters for which the impact of deviation s from the matched
condition on SNR can be estimated analytically [429].
The SNR trade space for optical- and RF- filter time-respon se pulse widths (To
and TRF) relative to the optical signal pulse width (T s ) is shown in Fig. 53. We use the
signal-pulsewidth for normalization rather than the bit-duration (Tbit) since matching
is independent of data rate. The bit period, for instance, can be adjusted to vary the
data rate or lengthened to reduce the effects of lSI if needed .
As expected, for wide RF filtering (TRF / T s « 1), optimum performance is
achieved for the matched condition, when f3 = t.Jr, =1, but extends over a large
sweetspot surrounding the dashed cente rline (TR F :::::; 3(Ts - To)) . As noted earlier, the
Gaussian waveform relaxes the optical filter tolerances needed for nearly-matched per-
formance, i.e., <""0. 1 to 0.2 dB filtering penalty. Similar benefits are seen for Gaussian
electrical filters.
If the Ts/rbit ratio is sufficiently short, lSI penalties can be made negligibly small,
and the analysis in Fig. 52 is broadly applicable. Otherwise, uncorrected lSI penalties
start to degrade performance for longer To and TRF (upper right quadrant), compress-
ing the SNR contours towards shorter pulse widths (lower left quadrant) as TsfTb it
increases . Staying to the lower-left of the centerline. avoids lSI penalties as shown
for the displayed RZ33, RZ40, and RZ50 data points. The SNR-optimized pulsewidth
coordinates for Gaussian filter and RZ signals shown in Fig. 53 are linked to the data
rate and bit duration is in Table 13.
For all three waveform types, optical and electrical bandwidths can be chosen for
near-quantum-limited performance well within the O.I-dB SNR sweetspot in Fig. 53,
198 David O. Caplan
Fig.53. Estimated SNR impact (Ll,) on a preamplified receiver with combined optical and RF
Gaussian filtering and low photon/bit Gau ssian signal s. Contours represent lines of constant SNR
(in dB) as a function of both RF and optical filter-respon se-pulse-widths relative to the signal
power pulse width T s . lSI effects are not included in this calculation since they are dependent
on both waveform shape and duty-cycle. However, lSI can be neglected for small To and TRF
(the lower-right quadrant of the filter space) and for low-duty-cycle waveforms , i.e., T s «
Tbit .
Detailed parameters for RZ33 , RZ40 , and RZ50 points are given in Table 13. Simulations based
on SNR analy sis by J. Shapiro [429).
Table 13. Relative pulse width and bandwidth for data points shown in the optimized SNR region
of Fig. 53. The bit duration Tb = IIR = I for all entrie s. Optical parameters are in terms of FWHM
intensity.
__ ~pl!c!!~i911~~' I_ ___()p~ ~
_S{~;.:~ l~~b)I~L~.~;~_~~;~~ ~bn ~1~t.J~~~~[~';-f~!~~ i~~~~t.:~~fRF-r :~ ,

RZ33 0.33 0.49 1.5 0.21 0.65 0.44 2.1 0.30 0.90 0.22 0.7
RZ33 0.33 0.49 1.5 0.25 0.75 0.44 1.8 0.20 0.60 0.22 1.1
RZ33 f-.Qd.~.. ..Q..:l.!1.. .L~. . Jl~~~._ JM?. _Q~4L ...LL ..QJ.lL .QJQ.. J).:..~L ..~.L
RZ40 0.4 0.55 1.4 0 .26 0.65 0.44 1.7 0.32 0.80 0.22 0.7
RZ40 0.4 0.55 1.4 0 .30 0.75 0.44 1.5 0.20 0.50 0.22 1.1
.B.~4~ QA_- __Q..? L __.1.L _Jl_·lL _Q&9_ _ Q~¥-.. ...LL ..QJA.. ..lL~i.
f--. _~L~L .1&..
RZ50 0.5 0.66 1.3 0 .35 0.70 0.44 1.3 0.30 0.60 0.22 0.7
RZ50 0.5 0.66 1.3 0 .38 0.75 0.44 1.2 0.23 0.45 0.22 1.0
RZ50 0.5 0.66 1.3 0 .41 0.82 0.44 1.1 0.12 0.24 0.22 1.8
with realizable optical filter bandwidths rv I to 2 times the data rate (R) and RF filter
bandwidths 0.7 to 2.2 R. When limited RF bandwidth drives the design, the optical
filter can be chosen to reduce the RF bandwidth to < 0.7 R with little or no SNR penalty.
The breadth of matching tolerance is also noticeable . For an optical filter bandwidth
of about 1.3R, nearly optimal performance can be achieved for all three waveforms
with RF bandwidth between Rand 2R.
We note that with nearly matched optical filters, a single-polarization preamplified

RX achieves true single-mode spatio-tcmporal filtering, which maximizes rejection to
background noise, but also modifies the ASE noise statistics [122-125] . While accurate
analysis of single-mode Bose-Einstein and noncentral-negative-binomial distributions
for '0' and 'I' ASE statistics, respectively, may influence theoretical predictions of
optimum threshold and receiver sensitivity, they are unlikely to impact the results
above. Near-quantum-limited demonstrations using this class of Gaussian waveforms
and filters for OOK [13] and DPSK [133] modulation lend additional support to the
analysis above.
5.1.3. Optimized Multi-Rate Transceivers
In many free-space applications, the flexibility to operate over a wide range of data
rates is desirable since it provides a straightforward means of adjusting the transmit-
ted signal-to-noise ratio, providing increased margin as needed. This can be used to
accommodate varying link losses, due to changes in link distance or atmospheric chan-
nel effects, or compensate for performance degradation and extend the useful lifetime
of the system. Especially in space-based systems, such capabilities should not incur
increased complexity, SWAP, or reduction in RX sensitivity.
For systems that employ APL amplifiers (see section 3.5.3), multi-rate communica-
tions with nearly matched filtering can be efficiently implemented[ 11,15]. As discussed
in the sections 5.1.1 and 5.1.2, the transmitter pulse shape can be adjusted so that it is
well matched to the receiver without sacrificing transmitted energy-per-bit. This can
be extended to multiple bit-rates simply by lowering the duty-cycle or average rep-
etition rate [10-12,15], causing the peak output power to increase while the average
saturated output power remains constant (see Figs. 35 and 36). Lowering the bit-rate
in this manner while maintaining the same pulse shape and width (i.e., reducing the
duty cycle), increasing the transmitted energy-per-bit thereby improving the error rate
while maintaining optimum performance.
This variable-duty-cycle multi-rate approach has been demonstrated for binary-
PPM at 51 to 1244 Mbps [11,12] with performance ",2 dB from quantum-limited
theory at all rates. Subsequent M-PPM demonstrations at 2.5 Gslotsls [15] similarly
yielded near-quantum-limited multi-rate performance but with improved sensitivities
(see Fig. 61), spanning a range exceeding 20 dB with only a 12-dB change in data
rate. Such features are particularly useful for providing bandwidth on demand or fall-
back modes for communications over a noisy or uncertain channel since SNR can be
improved by simply lowering the duty cycle (bit-rate) without additional penalties or
hardware . It is particularly well suited for free space laser communications since it
allows for graceful degradation without significant cost or complexity. Effects that can
limit the range of practical duty cycles are discussed in section 3.5.5.
Variable duty cycle techniques can also be used to simplify multi-rate DPSK
receivers [10,134] discussed in the next section . With a reconfigurable DI (see section
5.2.4) that can adjust the delay to accommodate the bit rate, a single filter design can be
used to achieve nearly-quantum-limited receiver sensitivity at all rates. Alternatively,
for harmonically related data rates, a fixed DI sized to the lowest rate can be used [134]
along with appropriate differential precoding . In this case, the same DI can be used
to simultaneously demodulate multiple-rate WDM-DPSK signals (see section 5.2.3),
providing both rate-flexibility and WDM scalability [136].
200 DavidO. Caplan
Fig. 54. Typical opticallypreamplified DPSK receiver.
5.2. Differential Phase Shift Keying (DPSK)
Optical differential-phase-shift-keying (DPSK) modulation was introduced in section

2.2.1 as a means of improving sensitivity over commonly used OOK. DPSK theo-
retically can offer among the best combined spectral and photon efficiency without
requiring a coherent local-oscillator-based receiver. This has generated considerable
attention by both the FSO community and the telecom industry, and led to the most
sensitive high-rate demonstrations [24,25,41 ,128-131,133,172]. However the sensi-
tivity benefits of DPSKcome at the cost of increasedcomplexity over OOK, requiring
a phase modulator and differential precoding in the TX, and an optical delay-line
interferometer(DI) and balanceddetection in the RX as shown Fig. 54. The remain-
ing DPSK RX elements includingsubsequent electrical amplification and filtering are
similar to OOK RXs.
In practice, the complexities associatedwithDPSK makeit challengingto achieve
the potential3-dB sensitivity advantage over OOK.The most significant performance
penalties result from residual chirp in the transmitterwaveforms [132,142] discussed
section3.5.2, and alignmentof DI and the signalwavelength [144,146,430], discussed
in the section 5.2.2. Smaller degradations in DPSK performance arise from non-ideal
interferometerdelay [135,146,31], extinction ratio, and detector imbalance in ampli-
tude and timing [132,145]. The reader is directed to an excellenttutorialon long-haul
fiber-based DPSK systemsby Gnauckand Winzerin [132]and other references in this
section for further details and analysisof DPSK performance.
Despitethese possibledegradations, DPSKreceiversensitivities closelyapproach-
ing quantum-limited theory can be achieved [133]. However, it is importantto realize
that while implementation penalties may individually be considered negligible, the
combinedimpact of severalof these effects could be prohibitive. Unless the improve-
ment in RX sensitivity can be realizedwith long-termreliability, DPSKprovideslittle
value for free-space applications, in contrast with long-haul fiber-based DPSK appli-
cations, which benefit from the reduction of nonlinear impairments that result from
lower peak power.
In the DPSKreceiver, the DI enablesa comparison of theoptical phaseby splitting
the opticalsignal field and recombining the two polarization-aligned components with
a relativetime-delay rd. At the output coupler, the two overlapping field components
interfere constructively or destructively depending on the relative phase of the differ-
entiallyencodedbits, withthe resultingoutputpowerproportional to the productof the
two fields, being directed to the first or secondoutput port. For complete interference,
the incidentfields needto havethesameamplitude, anda delaythat is a positive integer
multiple n of the bit-period 7bit [134,135], i.e.,
(117)
where FSR is the DI free spectral range. Since the data rate Rb = IIThit , it is ideally
equal to an integer multiple of the FSR, resulting in a delay-error .dT = lTd - Thit I =
0. The associated delay-error or rate-to-FSR mismatchpenalty is limited to less "-'0.3
dB for .dT < 0.05Tbit [146,431]. However, for larger errors, the SNR degradation is
waveform and filterdependent [135]. For receivers optimized for 33%-RZ waveforms,
10% and 14% delay-errors lead to ,,-,0.5 dB and "-'I dB SNR penalties, respectively.
While manufacturers typically specify a "-'I % tolerance for the DI delay accuracy, in
practice, largerdelay-errors can result from channel-rate changes in deployedsystems
to accommodate different FEC overhead [132] or to simplify WDM-DPSK receiver
implementation while maintaining compliancewith existing wavelength and rate stan-
dards [149,151,432,433].
For stability, size, and performance reasons, r« is typically chosen to equal only
one bit period.This relaxessignalcoherencerequirements[264], minimizesfrequency-
alignment penalties, and simplifiesTX precoding, whichcan be implemented with an
OR gate to logicallycombine DATA and CLOCKinputsto driveaT-flip-flop[140,434-
436], causing the transmitted phase to change wheneverthe DATA is a ' I'. However,
for multi-rate and multi-channel DPSK applicationsdiscussed later, it can be desirable
to use a multiple-bit delay to accommodate simplified implementations that provide
rate-flexibility and scalability with penalty-free performance that is compatible with
existing channel-rate and channel-spacingstandards [135, 136].
The DI outputs are received with balanced detection, which generates a signal
.di(t) that is proportional to the power difference between the two output arms of
the DI. This can be implementedby directly subtracting the photocurrentoutput from
discrete [128]or integratedbalancedphotodetectors [25,408,437], or usingsubsequent
differentialelectronics to generate the differencesignal [25,130,159], all of which are
commercially available with bandwidths exceeding 40 GHz. Balanced detection pro-
vides an implementation benefitover singledetector RXs since the resulting photocur-
rent deviation between "0" and " I" symbols is twice that of single-detector receivers,
making it easier to directly overcome the noise threshold in the subsequent decision
circuitry without additional RF amplifiers.
5.2.1. DPSK WavelengthAlignment Considerations
Of the additional RX elements needed for DPSK, the delay-line interferometeroften

imposesthe mostdemandingrequirements. For good performancethe DI is preferably
polarization insensitiveand the two arms must be stable to a small fraction of a wave-
length. This requires careful, thermo-mechanical packagingand/or active stabilization
[432,438-440], adding to DI size, weight, power, and cost. The DI is most commonly
implementedwith an asymmetric Mach-Zehnder design (see section 3.5.2 for detailed
description), although MichelsonII [432,441,442] and Sagnac-interferometer-based
11 Albert Abraham Michelson, 1852-1931 . invented and built what has become known as the Michelson-
interferometer, used in accurat e measurements of length and the speed of light. Conducted the Michelson-
Morley experiment (1887) with E. W. Morley, which attempted to detect the expect ed difference in the
speed of light caused by the motion of the earth . Michelson was awarded the Nobel Prize in physics in
1907 .
202 David O. Caplan
Fig.55. Periodic transfer function (fringes) of a DI: sin2 0 and cos 2 0 outputs ; and calculated
(.1')' calc) and measured (.1')'ex p d SNR penalties at 10- 9 BER as a function of carrier frequency
offset .1/. normalized by the DI FSR (assuming polarity correction) (443).
designs [440] with an equivalent transfer function [see Eqs. (57)-(60)] may be used
as well (see section 5.2,4 on Reconfigurable DPSK Demodulators below).
In contrast to the Mach-Zehnder interferometer-based modulators discussed in
section 3.5.2, where a short sub-picosecond static time delay is desirable, the rela-
tively long delay of at least one bit-period needed for DPSK demodulation causes an
asymmetry that makes stabilization more difficult. Small changes in DI temperature,
for example, even if uniform, generally lead to a phase difference between the arms
that shift the spectral alignment of the interferometer, resulting in significant SNR
degradation .
The ideal intensity transfer functions of the two interferometer output arms are
periodic and complementary, given by
I1 -- IE1 12 -- cos, 2 (.E.L L!.<I» '

7TF S R +""2
(118)
I 2 -- IE2 12 -
- • 2
sin (.E.L L!.<j,) '
7TFSR +""2
where L1¢is a measure of the relative optical phase between the internal interferometer
arms and is typically controlled to maximize the signal interference in one or both
of the output arms. When the bias L1¢ = 0, the L1f term represents the frequency
deviation from optimum alignment of the incoming signal to the interferometer. The
SNR degradation associated with L1f (frequency offset error) forms periodic SNR
fringes at even multiples of FSR/2 as shown in Fig. 55. The received signal is reduced
by an amount proportional to the difference of signal intensities at the output arms.
The received signal is maximum for L1f =0 and goes to zero asL1f approaches odd
multiples of FSR/4, when the outputs of the two arms are equal. Beyond this point, the
received data are inverted, an effect that can be compensated with polarity correction,
until the next maximum is reached which occur an even multiple FSR/2.
The net DPSK wavelength offset SNR penalty includes both signal reduction
and increased noise that is numerically analyzed in [132,144,430,431] . A simplified
estimate of the offset penalty can be obtained in closed-form following the approach
in [116,120,404]. Using the Marcum Q-function defined as
f xexp[-~(x2+a2)]10(ax)dx,
00
QMarcum(a,b) = (119)
b
where 10 is the modified Bessel function of the first kind of zero order, the BER and
SNR estimate ('/'calc) are given by [135]
"" [1 - QMarcurn ( J2/'th cos( ;~~) , J2/'th sin( ;~~) ) ]

BERcalc = 0.5 ,(120)
+QM arcurn ( J2/'th sin( ;~~ ) , J2/'th cos( ;~~ ))
/'calc ~ -In(2BERcalc), (121)
.:1/' ~ 10 log 10 (/'calc) , (122)

/'th
where /'th is the ideal SNR (or PPB) at the input to the preamplified RX (without
frequency offset) and .:1/' is the effective SNR penalty in dB. This straightforward
estimate for the offset SNR penalty provides excellent agreement with the measured
data at 40 Gbit/s shown in Fig. 55 and in [135] at 2.5 Gbit/s, and is consistent with
measurements and caJcuations in [144,146,431] . For l.:1fIFSRI <",,4%, .:1/' is less
than ~ dB at the 10- 3 BER, which is an error-free threshold point for commonly used
FEe [24,25,41,115,444]. Note that for a fixed frequency offset, this alignment penalty
is reduced as the FSR gets larger, an effect that makes it more challenging to avoid
penalties at lower data rates. For 40 Gbit/s DPSK channels received with a 40-GHz
FSR interferometer, wavelength alignment (l.:1fl) needs to be within ±1.6 GHz in
order to limit .:1/' < ~ dB. At 2.5 Gbit/s, the ~ dB wavelength tolerance is reduced to
± 100 MHz. But this can be achieved with commercial sources, as noted in section 3.3.
With appropriate temperature and wavelength control, laser DFB laser wavelength can
be stable for extended periods with less than 15 MHz standard deviation, consistent
with 2.5 Gbit/s DPSK communication performance with < 0.1 dB SNR deviation
[135].
5.2.2. Interferometer Stabilization
Given the significant SNR penalties for wavelength offset errors, the need for stable
signal and DI wavelength alignment is clear. Passive athermal DI designs have been
constructed to align to the 50 GHz lTV grid with .:1f less than ±0.8 GHz [432,433]
over a 0-7OC o temperature range. Assuming an otherwise perfectly aligned incom-
ing signal wavelength, this would lead to only ",,0.1 dB temperature induced penalty
at 40 Gbit/s. However, at 10 Gbit/s or lower data rates, even this level of stability
would lead to prohibitive SNR penalties. Moreover, the TX signal wavelengths are
not necessarily stable in free-space applications, an effect that discourages the use
of passively-stabilized DIs. For example , space-based links between Geosynchronous
and Low-Earth orbits (GEO-LEO) can experience Doppler shifts up to ",,9 GHz (for
1.55 flm signal wavelengths), that would preclude the use of an athermal interferometer
without some form of TX signal compensation [135] for Doppler or other long-term
wavelength shifts.
For single channel RXs, the preferred means of wavelength-offset compensation
is local tracking at the RX for reasons of simplicity, speed, and stability of control. This
204 DavidO. Caplan
In ut
r~::~,
Feedback
telemetry
to Tx or
network
Wave
meter
C Control
Processor
Fig. 56. Examplesetup for a pilot-stabilizedDPSKreceiver.
is typically implemented using synchronous phase-locking techniques that introduce a

small dither on L\i s in order to determine the sign of the phase error, and maximizing
the peak baseband (RF) signal power accordingly [399]. The dither may be applied
to one of the or arms [438] or TX signal wavelength [439,445,446] [,14> and L\f,
respectively in Eq. (118)]. The amplitude of the dither can be reduced to sufficiently
low levels so that active control can be achieved with little or no RX sensitivity penalty
[133].
An alternative approach to interferometer stabilization makes use of pilot tones
to stabilize the or [135], which can provide dither-free open-loop or control without
an RX signal, and the flexibility to tune the or to optimize performance when an
RX signal is present [267,268] . Such capabilities are particularly useful during the
spatial acquisition phase in a free-space link. With absolute DI wavelength control, for
instance, the DJ can be tuned to the proper wavelength in advance of the RX signal,
thereby expediting the acquisition process and eliminating the need to re-align the or
after dropouts or fades. This also removes the polarity ambiguity, in which the sign of
the data is unknown when using the standard phase-locking techniques.
An example of a pilot-based stabilization setup is shown in Fig. 56. The pilot
signal, which could be generated by a low-power DFB laser, is injected into the or
through an optical tap in the reverse direction. As discussed in section 3.3 and earlier
in this section, the pilot wavelength Ap can be calibrated via temperature and current
settings with a (short term) stability < ", 30 MHz [135], or through other methods,
such as a feedback from a built in temperature controlled etalon [231] or an external
wavelength reference such as a wavemeter, with sub-GHz long-term stability.
The pilot tone outputs from the or (Ipi and I p 2) can be detected by low-speed
photodetectors and the resulting photocurrents are processed to measure the normalized
L\ipilot or contrast ratio given by
(123)
Here B = 7f L\ip IFSR + L\4>ctrl +L\4>error is the net phase argument, L\ip is
the relative pilot carrier frequency, which is either known or measured, L\4>ctrl is
an interferometer phase control term, and L\4>error is a random interferometer phase
term, which can be constant or have time and temperature dependencies that need to be
compensated for. Fig. 57 shows a plot of D and normalized I p i and I p 2 as a function
of the L\f
Lasercommunication transmitterand receiverdesign 205
Fig. 57. The contrast ratio D and normalized pilot outputs Ipl and I p 2 as a function of £J.f
normalized by the DI FSR. Arrowsabovethe Ipl and I p 2 fringe peaks indicatethe location and
polarity of signal wavelengths that the DI can accept without penalty when the DI is locked to
the pilot wavelength at Ap at £J.fIFSR = 1.
The contrast ratio ranges from + I to - I and has a period equal to the DI FSR.
Thus, with the pilot capabilities shown in Fig. 56, the FSR and other important DI
parameters such as extinction ratio can be measured by sweeping the pilot carrier
frequency with fixed phase terms .d</>ctrl and .d</>error, a capability that would also
be useful for performing automated diagnostics in the field. Such capabilities could
be enhanced with the use of multiple pilot tones (distinguishable by dither frequency
and or wavelength), in which one pilot is used for stabilization, and the other for
characterization.
By choosing a target contrast ratio, e.g., D t arge t = 0, .d</>ctrl can be adjusted to
make the error between the measured and target contrast equal to zero (i.e., D error =
Dmeasured - - Dta rg ed using standard feedback control techniques. This locks the
DI to the pilot wavelength so that it is aligned to accept signal wavelengths at
FSR
As(m) = Ap ± (2m + 1) -4- ' (124)
where m is an integer and polarity correction is needed to invert the data when m
is odd. This provides flexibility in placement of the pilot wavelength . For instance , it
could be placed outside of the communication band to conserve spectrum and prevent
potential in-band interference. While the pilot tone in the example above can be used
as a CW single wavelength source, dithers in either wavelength and/or amplitude can
be introduced to allow synchronous control/detection methods .
In practice, a pilot-stabilized DPSK RX can be used in the following manner. Prior
to acquisition , the pilot tone can be used to align the interferometer to the anticipated
RX wavelength . This eliminates the need for scanning the interferometer to search
for an incoming wavelength as long as the anticipated and received wavelength are
relatively close, e.g., correct to within about ± 10% of an FSR. Once the signal is
received, with incoming power above a predefined threshold Prnin, the interferometer
control can switch to tracking the signal using standard approaches. Therefore, the
control system only needs to track on the signal, if at all, when the signal is relatively
strong, which reduces the dynamic range of operation required for signal tracking.
206 DavidO. Caplan
While the signal is strong, any misalignment betweenthe pilot wavelength needed for
correctinterferometeralignmentto the signalcan beeliminatedby havingthe pilottone
align to the interferometer. In this manner, a locally resident pilot tone can effectively
track an incoming signal that can be slowly drifting due to poor control, aging, or
Doppler shifts.in moving platforms. Should the signal drop out, the interferometer
control can switch back to the well aligned pilot control, which can remain stably
alignedto thecorrectwavelength untilthe incomingsignalexceedsPmin . This provides
seamless interferometer performance during fades, and reduces any interferometer
induced acquisition delay [268].
5.2.3. Multi-Wavelength DPSK ReceiverOptions
For high-rate 100+channelWDM-DPSK systems(e.g., [138-140,167,447]), the size,

weight, power, and costs associated with reliably maintaining a stable delay-line in-
terferometerfor each channel can be substantial. Miyamotoand coauthors [149,151]
demonstrated concurrentPSK to ASKconversion of 43 Gbit/s WDM-DPSK channels
to WDM-duobinary channels on the 100GHz ITU grid using a 50-GHzfree-spectral-
range(FSR)DI.Whilewellsuitedforapplications wherespectralefficiency isrequired,
this simplified DPSKRX incurssensitivity penalties of3 dB due to single-ended DPSK
reception and another ,.",0.5 dB due to the >- 14% mismatch between the 50-GHz DI
FSR and the 42.8 GHz data rate when used with66% RZ waveforms. As notedearlier,
the mismatch penalty is waveform dependent; with r- 14% delay error, the penalty
increases to ,.", I dB for 33% RZ waveforms further reducing the sensitivity benefit of
DPSK.
Simplified multi-channel 'OPSK' receivers havealso been implemented with peri-
odic narrow band optical filtering and similarduobinary signalsover dispersive chan-
nels [148,150,416,448]. While dispersion tolerant, these single-ended demodulators
also incur sensitivity penalties in excess of 3 dB when compared to balanced OPSK
receivers.
However, multi-channel WOM-OPSK reception can be achieved using a single
01 with near-quantum-limited performance and be compatiblewithexistingstandards
without waveform-dependent penalties [135].
By leveraging the DI's periodictransferfunction shownin Fig. 55, andconstraining
the received 'wavelength spacing (,1vc h), multipleWDM-DPSK channels can be si-
multaneously demodulatedusinga singleinterferometer. Unlikesingle-channel DPSK
receiversin which the DI can track the incoming signal, in the multi-channel configu-
ration, independent channels must have the same wavelength periodicity as the DI in
order to avoid significantSNR penalties. For WDM-PDSK signals coming from the
same TX, this periodicity can be configured locally at the TX. In this case, the RX
DI can also track locally on any of the received signals in order to align to all of the
incomingWDM channels. However, local tracking at the RX is precluded for multi-
access applications, where independent TXs send WOM-DPSK signals to a common
RX. In this case, proper wavelength alignment can be achieved with either absolute
A-control at the TX and RX, or with (slow) feedback from the RX via network-level
maintenance and control of the TXs. Transmitter-centric control could also be used to
compensate for aperiodicities that can occur in the DI [449]. As noted in the previous
section,absolute A-control of the DIcan be achieved via stable thermal design [432]or
with pilottone stabilization, whichcould be a locally-resident calibratedlaser. Another
.
,,. - _. - - - - - - _. - - - - - - - - - - - - - - - - - - - - ---------------------..
,,
,
1.., .J..,. .... 1... :
NFB U ...n
Fig. 58. A Multi-channel DPSKRX with feedback and feedforward alignment capabilities that
can include a pilot tone (Ap) or master-channel, wavemeter, channel power and bit-error rate
(BER), and interactive communication with the network (NFB) [266].
option is to use a master-reference channel that the or, and all other signal and pilot
AS can align to with feedback [268], providing the capability for sub-GHz (perhaps
sub 100 MHz) wavelength alignment capability across the network.
WDM-DPSK channel separation can be achieved via optical demultiplexing after
the or [149,151 ,266,450,451] as shown in Fig. 58. Since balanced detection is needed
to achieve high-sensitivity, this WDM-DPSK RX requires one extra WDM filter, but
the WDM cost is small relative to the potential cost of tens of interferometers and
associated stabilization hardware.
In order to avoid SNR degradation when using a multi-channel DPSK receiver,
three conditions must be satisfied . First, the channel rate (R) must be nearly equal to
an integer multiple n of the FSR to satisfy Eq. (117) and minimi ze delay-error. Second,
.:1vc h must be within a small frequency offset (.:1f) from an integer multiple m, of
FSRl2 (with polarity correction), i.e.,
FSR R
.:1vc h = m - - ± .:1f = m- ± .:1f. (125)
2 2n
The third condition requires that transmitted channels each be aligned to target
SNR fringe peaks so that .:1f -+ O.
For wavelengths on the lTV grid (e.g., 100 GHz channel separation), standard
SONET rates of2.5, 10, and 40 Gbps for instance, are compatible with these conditions
since the channel rates (or half-channel rates) are integer submultiples (11m) of the
lTV channel spacing.
However, when commonly used 7%-overhead G.709 compliant FEe is used with
standard SONET rates , bringing a IO-Gbp s SONET data rate to a 1O.7-Gbps coded
channel rate, it is impossible for all of the lTV grid-based WDM channels to align with
the periodicity of the interferometer and satisfy the conditions in Eq. (125) above. In
this case , SNR penalties can be avoided by either: a) adjusting the channel spacing
to be a multiple of half the channel rate, abandoning the lTV grid if necessary, or b)
adjusting RJ2 to be an even factor of the channel spacing, abandoning, for example,
SONET or G.709 standards if necessary.
208 DavidO. Caplan
X, (10 .7Gbps)
Fig. 59. WDM-DPSK channel spacingfor 10.7 Gbps rate signals aligned near the 501100 GHz
ITU grid and to a 10.7-GHz FSR delay-lineinterferometer. Arrowscenteredon each wavelength
indicate location and polarityof the receiveddata.
If conformingwithexistingstandardsis the priority, Eq. (125) maynot be satisfied

exactly, but the associatedperformance penaltiesand deviation from the standardscan
be constrainedto acceptablelevels. Forexample, 10.7 Gbitlschannelson the lOO-GHz
lTV grid can be demodulated without performance penalty by a 10.7-GHz FSR DI,
whichcan accept optical center frequencies every "-'5.3 GHz with polaritycorrection.
While most of the DI fringeswill not align exactly to the lOO-GHz lTV grid, none will
misalignby more than 5.3/2 or ,,-,2.7 GHz (FSR/4), with the averagedeviation of only
"-'1.4 GHz (FSR/8) as shown in Fig. 59. This deviation can be further reduced by a
factorof n by demodulating with an n-bit DI, albeit with tighterwavelength alignment
requirements as shown in Fig. 55 and boundedin Eq. (125) [266].
It is important to note that there are additional cost and performance benefits as-
sociatedwith the increasein alignmentrequirements neededfor multi-channel WDM-
DSPK RXs. For instance, spectral efficiency and RX sensitivity could benefit from
reduced wavelength uncertainty throughout the network, which in tum, may allow
reduced filter bandwidth margins(needed to accommodate wavelength uncertainties)
and improved performance. Moreover, as noted in section 3.3, the additional cost of
providing infrequentfeedback to optimize wavelength alignmentand communication
performance of relatively stable tunable devices can be smaller than the expense of
developing fieldable hardwarewith 20+ year stability.
Expanding on the example above, with proper differential precoding, the same
1O.7-GHz DI could also be used to simultaneously demodulate 43 Gbitls channels
with negligible delay error. Thus, for near-harmonically related data rates, a single
interferometercan be used to simultaneously demodulate multiple-rate WDM-DPSK
signals [136].Tofurtherdiminishdeviation fromthelTV gridand to provideadditional
rate/alignment flexibility, a 2.68 GHz DI could be used to simultaneously demodulate
2.5,2.67, 10.7,40,42.7, and 43.02 Gbps SONETand G.709-compliant WDM-DPSK
signals within 700MHz of the lTV grid and with less than 7% delay-error.
For high-channel count WDM-DPSK links where spectral efficiency is more im-
portant,tightchannelspacingis desirablebutoftencomeswitha decreasein RX sensi-
tivitydue to inter-channel interference resulting from limitedWDMchannelisolation.
Forsinglepolarization RXs,thiscross-talkpenaltycan be reducedby polarization mul-
tiplexingodd and even channels [452].Another methodof suppressing cross-talkcan
be seen in Fig. 55, for WDM channelsseparatedby exactly an odd multipleof FSR/4.
For this spacing,the signal intensities on the DI outputarms are equal regardless of the
data, and the resultingsignal photocurrent Lli( t) goes to zero. This property could be
Spare input
Fig. 60. Areconfigurable polarization-independent delay-line interferometer. basedona double-

pass Michelson interferometer design using Faraday-rotator mirrors (FMs) to compensate for
residual polarization rotation in the arms of the interferometer. Variable-delays and switching
elements canbeincluded toadjust the delay orFSR, and variable-optical-attenuators(VOAs) can
adjust therelative power inthe two arms tooptimize theDI-extinction [267,268).
used to benefitdense WDM systems, by staggering the channel spacing so that delay-
line interferometer improvesrejection and reduces potential cross talk penalties. This
is especially helpful for adjacent channels where WDM rejection is the poorest. For
the multi-channel receivers described above, this could be implemented by sending
odd and even WDM-DPSKchannels to two separate WDM-DPSK RXs, staggered by
FSR/4. Moreover, when n-bit DIs are used for demodulation(yielding FSR = Rb/n),
the channel spacing can be adjusted by integer submultiples of the data rate, provid-
ing n-times finer wavelength adjustment for optimizing the trade between photon and
spectral efficiency of the system.
5.2.4. Reconfigurable DPSK Demodulators

The ability to simultaneously demodulate many WDM-DPSK signals with a single
DI' becomes even more useful if the DI can be reconfiguredto optimize performance
or accommodateother data rates. For example, a DPSK receiver could be adjusted to
accept different (non-harmonically related) rates or FEe upgrades without delay-error
penalties. Even in single-channel DPSK RX designs, it may be preferable to have one
reconfigurable DI and associated control that can operate at several rates, rather than
a fixed DI for each rate.
A polarization independent extension of the double-pass amplifier design [354]
discussed in section 3.5.3 is a Michelson-based interferometer that incorporates Fara-
day rotator mirrors in both arms of the interferometeras shown in Fig. 60.
This geometry compensates for residual polarization rotation over a broad-range
of wavelengths, making it well suited for WDM-DPSK applications, and the lack
of polarization dependence enables useful non-polarization-maintaining elements to
be incorporated into the arms of the interferometer without introducing polarization
dependent losses (PDL) or degradation in DI performance. For example, a standard
single-mode variable-delay-line with > ",200 ps delay could be used to continuously
tune a reconfigurable DI to accept DPSK signals at any rate greater than ", 2.5 Gbit/s,
and provides a built-in means of adjusting the FSR to compensate for rate-FSR mis-
match. The tuning and phase-control range in this case benefits by a factor of two
from the double-pass geometry. Alternatively, I x N switching elements can be used
210 David O. Caplan
to vary the DI-delay by discrete steps to extend the DI tuning range further. When used
in combination with a variable-delay. the continuous tuning provides additional rate
flexibility and relaxes path-length tolerances.
Other elements. such as a variable-optical-attenuator(YOA)can be used to com-
pensate for relative changes in insertion loss between the two paths. which. in com-
bination with pilot-based characterization. can be used to measure and optimize DI-
extinction ratio. By including the ability to terminate the reflection from the arms of
the interferometer. either by adjusting the YOA or switching to an angle-terminated
port. the splitting ratio of the 50/50 coupler and the output intensities I s 1 and I s 2 and
corresponding photocurrents can be measured and balanced.
Pilot tones can be injected at the spare input port in this geometry to stabilize
and characterize the DI (see Fig. 58 and section 5.2.2. This. in combination with
variable-delay adjustment enables autocalibration and in-situ feedback and optimiza-
tion capabilities. For example. the DI-delaycan be adjusted to minimizereceivedBER.
and pilot-based stabilization/characterization can provide independent measure of the
DI-FSR (channel-rate) and the incoming signal wavelength. This information can be
used to track/manage wavelengthsand relativeDoppler shifts throughout the network.
5.3. Hybrid Modulation Formats
In order to optimize overall system design. performance. and cost. we can consider
hybrids of the orthogonal modulation formats described in section 2.2.1. which can be
used to access the manyTHz of availableoptical spectrum. Selectionof I -FSK frequen-
cies. p-PPM positions. and L-polarization states distributed over w-WDM channels
can be used to overcome practical limitations such as electrical bandwidth. nonlinear
impairments. and availablechannel bandwidth as highlightedin Eqs. (126)-( 129).The
number of bits/symbol in such a multi-dimensional hybrid format is given by
(126)
where Me« is the effectivenumber of orthogonal symbols in the constellation. In order

to maintain orthogonality. the maximum number of polarization states L m a x = 2. In
the Shannon-sense. the bandwidth expansion factor (as illustrated in Fig. I) is given
by
F= ~= log2 (pI L) k
(127)
r pI L (1 + 1]FEc) Meff (1 + 1]FEc)
which can be used to improve RX sensitivity and photon efficiency. Here r is spectral
efficiency. and the (I +1]FEC ) term accounts for additional FEe overhead such that
the channel rate. Rb (I +1]FEC ). For a given data bit rate R b• the combination of
the number of wavelengths and symbol constellation size can be used to reduce the
required electrical bandwidth per channel:
B _ Rb (1 + 1]FEC)p
E - wk '
(128)
which benefits from the use of orthogonal bases (polarization and frequency) and
independent WDM channels. The net optical bandwidth required is
B _ 2Rb (1 + 1]FEC) pi
0- k ' (129)
where the factor of 2 is a conservative estimate for the minimum wavelength spacing.
The corresponding effectivepeakTX power is given by
p eff = (Keff) Fp (130)

pk W DCpulse '
where Keff is the effectivenumberof overlapping symbolsdefined Eq. (90) in section

3.5.5A.
From these equations, the impact of adjusting the hybrid-modulation parameters
(P, t, L, and w) can readily evaluated, to optimize the lasercomsystem. For example,
given a maximum electronic bandwidth limitation BF!ax, the maximum achievable
data rate can be determined from Eq. (128),
ma x
Rmax _ B E Wk = Swk : - -
-r-t-
(131)
b - (1 +1]FEC)p (1 + 1]FEC) p ,
where S is the bandwidth-limited p-PPM slot rate. From Eq. (131) it is easy to see that
Rb'ax can be increaseddirectlythroughthe numberofWDM channelsor by increasing
the net bits per symbol k, by expanding the constellation size with supplementary
frequency or polarization bases. The formerlinearlyincreases Rb'ax whilemaintaining
the same spectralefficiency and sensitivity per channel, whereas the latter can be used
to augment Rb'ax while improving RX sensitivity [see Eqs. (126) and (17)-(20)] at
the expense of spectral efficiency.
5.4. Demonstrated Communication Performance
Throughout the paper, we have discussed methods of optimizing receiver sensitivity

through use of variousTX and RX designs, modulation and coding. An extension of
Alexander's compilation of representative high-sensitivity lasercom demonstrations
is given in Table 14 and Fig. 61 [20]. Receiver sensitivity at 10- 9 BER is listed as
a function of data-rate for various modulation formats, waveforms, and RX designs.
These demonstrations bridge the gap between theory and what can be achieved given
realistic design and technology constraints, and support many of the conclusions and
design guidelines presented here. As laser communication technologies and TXIRX
designs have matured, data rates haveincreasedand sensitivities haveimproved, mov-
ing towards the lower-right quadrant of the Fig. 61, with the current state-of-the-art
rapidly approaching the fundamental quantum and Shannon limits of communication
performance.
As we can see in Fig. 61 , at high rates, optically preamplified receivers have
demonstrated the best sensitivities,despite inferior theoretical sensitivities relative to
coherent-homodyne and photon-counting RXs.This is largely due to the nearly ideal
characteristics of wideband high-gain average-power-limited optical amplifiers dis-
cussed throughout this paper that enable the preamplified RX structure to leverage
the strengths and mitigate the deficiencies in essential RX electronics. As a result,
preamplified systems can often leverage ongoing developments in wideband tele-
com optical and high-speed electronic technologies, proving a cost-effective means
of advancing the state-of-the-art. Whencombinedwith the maturityand sophistication
212 David O. Caplan
Table 14. List of high-sensitivity optical receiver demonstrations shown in Fig. 61.
Rate, Sensitivity, Modulation! Coding,

# [Mbitls] [avg. PPB] Waveform RX am Ref.
1 65* I 32-PPM Phot. Cnt. SSPD SCPPM Turbo. 100% [34]
2 14 I 64-PPM Phot. Cnt. GM-APD SCPPM Turbo, 100% [44)
3 781 2 32-PPM SSPD Phot. Cnt. SCPPM Turbo, 100% [34]
4 73 4 256-PPM Opt. Preamp RS255/239,7% (15)
5 2 5 PSK Homodyne Conv. Cd. [157]
6 10000 7 DSPK-RZ Opt. Preamp Turbo Block, 25% [24]
7 781 8 32-PPM Phot. Cnt. SSPD SCPPM Turbo, 100% [33]
8 40000 8.7 DSPK-RZ Opt. Preamp E-RS255/239.7% [25]
9 78 9 256-PPM Opt. Preamp no (15)
10 136 10 128-PPM Opt. Preamp no [15]
11 10000 II DSPK-RZ Opt. Preamp E-RS255/239,7% [41]
12233 10 64-PPM Opt. Preamp no [15)
13 389 12 32-PPM Opt. Preamp no [15)
144 16 PSK Homodyne no [I57)
16 10000 16 OOK-RZ Opt. Preamp E-RS255/239,7% [41)
17 565 20 PSK Homodyne no (195)
18781 20 32-PPM Phot. Cnt. SSPD SCPPM Turbo, 100% [33)
21 2500 25 DPSK-RZ Opt. Preamp no [133)
22 10000 -30 DSPK-RZ Opt. Preamp no [129)
23 1244 31 4-PPM Opt. Preamp no [15)
25 10000 36 DSPK Opt. Preamp no [129)
26400 36 PSK Homodyne no [155)
27 110 37 4-FSK Heterodyne no [160)
28 1000 37 OaK Opt. Preamp 4/5 Conv. Cd., 25% [454)
29 10000 38 DSPK Opt. Preamp no [128)
3042700 38 DPSK Opt. Preamp no (159)
31 5000 43 OOK-RZ Opt. Preamp no [13)
32 1250 45 OOK-RZ Opt. Preamp no (455)
33400 45 DPSK Heterodyne no [456)
34 12500 45 DSPK-RZ Opt. Preamp no [24]
35 42700 45 DSPK-RZ Opt. Preamp no (130)
36 1250 46 FSK-RZ Opt. Preamp no [455)
38 10000 57 OOK-RZ Opt. Preamp no (406)
39 10000 60 OOK-RZ Opt. Preamp no [129]
40 1244 61 2-PPM Opt. Preamp no (15)
41 3000 62 DPSK Opt. Preamp no (438)
* Note that #1 is the #3 32-PPM demonstration at 10 Gslotsls with a 12x repeat code . This lowers photon flux
and reduces blocking losses by 3 dB , and lowers the data rate to 65 Mbps. It also emulates the performance of
a 12-detcctor array operating at 781 Mbps .
Table 14. (Continued) List of high-sensitivity optical receiver demonstrations shown in Fig. 61.
Rate, Sensitivity, Modulation! Coding,

# [Mbitls] [avg. PPB] Waveform RX om Ref.
42 51 65 2-PPM Opt. Preamp no [11,12]
43 155 65 2-PPM Opt. Preamp no [11,12]
44 311 65 2-PPM Opt. Preamp no [11,12]
45622 65 2-PPM Opt. Preamp no [11,12]
46 1244 65 2-PPM Opt. Preamp no [11,12]
472500 66 CPFSK Heterodyne no [194,457]
48 60 68 4-PPM APD-FET no [391]
49 120 71 4-PPM APD-FET no [391]
50 4000 72 PSK Homodyne no [458]
51 200 74 FSK Heterodyne no [179]
52 560 78 DPSK Heterodyne no [459]
5330 80 4-PPM APD-FET no [391]
54 7800 81 PSK Homodyne no [158]
55 10000 81 OOK Opt. Preamp no [129]
57 10000 88 Duobinary Opt. Preamp no [152]
586000 89 CPFSK Heterodyne no [461]
595000 93 PSK Homodyne no [166]
60 10000 100 PSK Homodyne no [462]
61 15 108 4-PPM APD-FET no [391]
62325 110 OOK APD-trans-Z no [390]
63 10000 112 OOK Opt. Preamp no [463]
65 3000 116 DPSK Opt. Preamp no [465]
67 5000 135 OOK Opt. Preamp no [466]
68 2500 137 OOK Opt. Preamp no [316]
69 10000 147 OOK Opt. Preamp no [467]
70622 152 OOK Opt. Preamp no [468]
71 622 155 OOK APD-HEMPT no [469]
72 1244 156 OOK Opt. Preamp no [470]
734000 175 OOK Heterodyne no [471]
744000 191 FSK Heterodyne no [471]
75 10000 193 OOK Opt. Preamp no [316]
77 20000 270 OOK Opt. Preamp no [472]
78 10000 297 PSK Homodyne no [165]
79 2000 705 OOK APD-FET no [473]
80 1000 1162 OOK APD-FET no [474]
81 20000 1600 OOK Opt. Preamp no [475]
82 10000 3000 OOK-RZ APD-FET no [388]
214 David O. Caplan
Fig.61. Summary of high-sensitivity optical communication demonstrations showing reported

sensitivities as a function of data rate . Modulation type is indicated by marker type, for instance
PPM is represented by circles . Receiver type is indicated by color : Red indicates coherent (dark
red =homodyne and pink =heterodyn e), dark blue indicates an optically preamplified, and light-
blue indicates a photon-counting RX. Coded demonstration results are shaded . Also shown are
the quantum-limited sensitivities for the various modulation formats (uncoded), M -PPM theory
for 2.5 Gslots/s, Shannon -limited sensitivities , and lines of constant power (dashed diagonal).
See also Robinson et al. [214], 1.5 PPB sensitivity at 100 kbps. Adaptedfrom S. Alexander [20).
L = 400,000 km
Spac e loss: D J
11_'" ( JID4'1" '
Fig.62. Illustration of bidirectional Earth-Moonlasercom link (22).
of these technologies, performance is more robust and much closer to the theoreti-
cal limits over a wide range of data rates from the Mbitls regime to tens of Gbitls
[13,15,24,25,41,130,159]. Furthermore, the data rate for high-speed preamplified re-
ceivers can be scaled using standard WDM techniques without additional performance
penalty, providing a significant performance advantage over coherent- and photon -
counting-based receiver designs.
Recent advancement in photon-counting detector technologies combined with
powerful FEe has led to the most sensitive demonstrations in Fig. 61. These demonstra-
tions include detector characteristics such as detection-efficiency and internal detector
noise sources such as dark-counts, after-pul sing, and blocking loss (due to limited
detector bandwidth). Note that these demonstrations do not include external back-
ground noise, which is an important consideration that can vary widely with channel
conditions [108, 109,476]. The multi-mode detection area of these detectors that is ben-
eficial for mitigating channel turbulence and simplifying spatial acquisition, tracking,
and photon-collection, combined with a broad spectral acceptance, make background
noise a practical limitation in many photon-counting applications. Thus, for photon-
counting receivers limited by background noise, optical filtering along with associated
insertion losses must be included in the evaluation of RX performance. However af-
ter incorporating the additional "'0.5 to I dB losses for narrow-band optical filtering
[398,477 ,478], photon-counting receivers sill show great promise for use in power-
starved links, especially ground-based receivers , where atmo spheric effects may be
considerable and size, weight, power, and reliability are not necessarily critical design
drivers .
5.5. Applications: to the Moon and Beyond
We conclude this paper with discus sion of a long-haul application, which illustrates
many of the benefits of FSO . Renewed interest in manned missions to the moon and
beyond would require robust bidirectional high-rate links to support the human in-
frastructure, telemetry, science data, diagnostics, remote monitoring and control, and
web-based connectivity. Here we present four notional lasercom link budgets for the
",400,000 km Earth-Moon link (Fig . 62), based on mature technologies that are readily
available today and incorporate many of the concepts discussed earlier in the paper.
The budgets in Table 15 use the following communication parameters:
Modulation format M -PPM

216 David O. Caplan
Slot Rate: S = 10 Gslot/s

Coding gain: 5 dB
Waveform: 50% RZ
Wavelength: A = 1550 nm
Aperture size: DTx = DTx = 0.2 m -+ 7Jspac e = 86 dB.
The budgets also include the following estimates of implementation losses:
Transmitter optics : 7]Tx = -3 dB

Receiver optics: 7JRX = -3 dB
Pointing and tracking: 7JPAT = -2 dB
Receiver dB from theoretical: 7JQL = -2 dB.
Raybon and co-authors presented an impressive all optical 400 km regenerating

optical loop experiment at 40 Gbit/s, with each loop consisting of four EDFA- and
Raman-amplified 100 km spans [138,303]. They postulated that such technologies
could be used to span a distance equivalent to the Earth- Moon link, but would require
rv I000 regenerators to overcome the rv I00,000 dB fiber loss. Long-haul FSO has clear
benefits over such guided-wave links, since the net space-loss for 0.2 m TX and RX
apertures is rv86dB, a savings of over 99,000 dB. This advantage combined with more
energy-efficient modulation and coding, and reduced data-rates enables the link to be
closed with Watt-class transmitters.
Table 15. Lunar lasercom link budgets for various configurations of PPM constellation size (M) ,
number of channel s (w) , and aperture diameter. Common to each of the four configurations is
the slot rate S = 10 Gslot/s, coding overhead, wavelength ()., = 1550 nm), link distance, and
assumptions for implementation penalties . These include losses in the TX (1)TXOpt = -3 dB)
and RX (1]RXOpt = -3 dB) optics , point ing and tracking (1]PAT = -2 dB) and RX deviation
from theoretical performance (1]QL = - 2 dB) . Adapted from [22).
J(eiP-arameteisJor:.Earth .... Moon:linO [JJi1iiS] 't#~ t#2~ f#3~ ~~

Co mm u nica tion Sys tem Pa rame te rs :
1 ccnetene ucn . Ize (M) 1024 128 1024 256
2 Number 01WDM channel. (w ) 1 1 8 120
3 Data Rate/cha nnel . R.. ' S'log, (M)11M 'l l+ nft cll GBltI.ec 0.065 0.365 0.065 0.208
Tra nsmitt e r:
'7 Ave rage Tx Powerlchann el (P h ) dBW .J .J ·12.0 ·7.8
5 Peak Tx Power/channel ' P,;M/lw Pul. eoc) dBW 30 21 21
6 Net Tx Power · w ' Ph dBW ·3 ·3 ·3
~
13
Receive r:
7 Approx. Uncoded Rx .ens ltlvlty M·PPM. IPPBoo.) dB·PPB 6.0 7.6 6.0 7.0
8 Net Rx . en. ltivity @ Data Rate (5 . , ) dBW .107.8 ·98.7 ·107.8 ·101.7
9 Rec eived Powerlch ann el at EDFAInput (P... ) dBW ·96.9 ·96.9 ·105.9 .101.7
Pe rlo rma nc e
10 Implemented Receiver Sensitivity dB·PPB 3.0 .6 3.0 4.0
11 Net Band width Required ' 2'S ' w' (1+nft clI125E9 nm 0.24 0.24 1.92 28.8
12 Net Rate : Rnet » Re,,'w G Bit/sec 0.065 0.365 0.521 25.0
13 Margin dB 10.9 1.8 1.8 0.0
A representative TX and RX pair suitable for long-term use in space [21,198,380,409]

that can achieve the performance in the budget above is a Master-Oscillator-Power
Amplifier (MOPA) transmitter with an optically preamplified receiver [15]. Quasi-
Gaussian 50%-RZ pulse-shaping in the transmitter is used to ensure a robust match
between the TX-waveforms and the Gaussian optical filter in the RX and minimize
the impact of timing jitter [14], with demonstrated performance within I dB of the
quantum limit [13,133]. Furthermore, as discussed in sections 3.5.5 and 5.1.3, fixing
the pulse shape for all M enables flexible multi-ratecapability with a single receiver
optimized for all rates [11,12,15].
In Table 15, configuration #1 highlights a low-rate65-Mbit/s link achieved with
'" II dB marginusinga 0.5 W transmitter, 0.2mTX and RX apertures, and 1024-PPM.
Of course, the margin could be traded to reduce aperture size or net output power.
Alternatively, by simply changing the PPM alphabet size to M =512, 256, or 128
as in configuration #2, the excess margin could be used to increase the data rate up
to 365 Mbit/s with ",2 dB margin, providingthe option for bandwidth-on-demand or
fall-back modes of operations as needed.
Anotheroption wouldbe to includeadditionalWDMchannelswith0.2nmspacing
as in configuration #3, which can provide up to 521 Mbit/s with ",2 dB margin.
Relative to #2, the increase in net data rate is enabled by the 1.6 dB improvement in
RX sensitivity due to the larger alphabet size. Although more hardware is required
to support the additional WDM channels, #3 provides similar flexibility in net data
throughput as #1 and #2 with additional redundancy. In addition, having multiple
channels share a common power amplifier effectively reduces the peak TX power
avoiding the nonlinearimpairments that can limit M [21]. Such effects start to impact
communication performance at 24 to 30 dBW peak levels as discussed in section
3.5.5C,andcould necessitate a changeof design parametersin#1. Forexample,hybrid-
orthogonal constellations including polarization and frequency bases can be used to
maintain M eff and reduce the peak power.
By increasingaverageTX power to 13 dBW (20 W), and WDM channel count to
120in configuration #4, a link with up to ", 25 Gbit/s throughputcan be achieved. This
requires",29nm of bandwidth, whicheasily falls withinthe 35-nmEDFAC-band, and
correspondsto only 0.007 bit/slHzspectralefficiency. However, the resultantimprove-
ment in receiver sensitivity to 2.5 PPB enables a respectable distance-rateproduct of
'" 10 16 km-bit/s- comparable to the best fiber-based demonstrations [93,138,139,141],
without the need for r- 1000 regenerators, emphasizingthe differencebetween square-
and exponentially-dependent channel loss. For #4, the overall link efficiency (using
TlE/ O = 13% [199]) is
pne t
Tllink = TX = 6 [nJ / bit] (132)
TlE/oR ne t
for the 400,000 km lunar link.

We conclude that established technologies developed for the telecom industry
can be leveraged to implementscalable photon-efficient optical communication capa-
bilities today. These example lunar link budgets, based on an average-power-limited
MOPA TX anda near-quantum-Iimited M-orthogonal optically-preamplified RX,show
the potentialfor high-capacity long-haul free-spaceoptical linksat varietyof data rates
from the Mbit/sregimeto tensof Gbit/s, rates that cannot be practically achieved using
RF technologies. And, as photon-counting technologies mature,the sensitivity of opti-
cal receivers could potentially improveanother 5-10 dB. These capabilitiesare likely
to becomean essential part of the space-basedcommunications infrastructure, provid-
ing cost-effective support for deep-space exploration and future manned missions to
the Moon, Mars, and beyond.
218 David O. Caplan
Acknowledgments
For most of the past decade , I've had the pleasure to work in the challenging and
innovative atmosphere at MIT Lincoln Laboratory, perhaps best described by Don
McClellan, the former head of the Communications Division who said, "If the laws of
physics will permit it, we can build it."
Many thanks to Vincent Chan, Roy Bondurant, Kristin Rauschenbach, and Eric
Swanson for providing me with the initial opportunity to work in this stimulating
environment, where I have been able to contribute to the field of optical communica-
tions and bring many new ideas to reality, alongside an extremely talented group of
present and former colleagues from the MIT community. These include Walid Atia,
Richard Barron, David Bold, Don Boroson, Larry Candell, Steve Constantine, Jerry
Chen, Eric Dauler, Bill Freking, Jeff Gottschalk, Matt Grein, Farhad Hakimi, Katie
Hall, Scott Hamilton, Scott Henion, Paul Juodawlkis, Alan Kachelmyer, John Kauf-
mann, Andrew Kerman, Sumanth Kaushik, Bill Keicher, Mark Kuzenetzov, Farzana
Khatri, Emily Kintzer, Sarah Klein, Jeff Livas, Jeff Mendenhall, John Moores, Bob
Murphy, Dan Murphy, Ron Parenti, Wayne Phoel, Al Pillsbury, Jinendra Ranka, Jeff
Shapiro, Neal Spellmeyer, Peter Schulz, Todd Ulmer, Simon Verghese, Matt Willis,
Tim Yarnell, and many others in the community.
It has been a privilege to work closely with people of the caliber of John Car-
ney, Bryan Robinson , Fred Walther, and especially Mark Stevens. As friends and
colleagues with tremendous breadth and depth of experience with optics, communica-
tions, and making things happen, they have been a catalyst for many of the new ideas
that are reflected in this paper. Along with other members of the Optical Communica-
tion Technology Group , they have contributed significantly to the content and initial
demonstration of many of the concepts presented.
To my wife Rose and my family lowe the most gratitude. Without their continuing
support and encouragement, this would not have been possible .
Lasercommunication transmitterand receiverdesign 219
Symbols
0: Chirp parameter
.1, SNR penalty
Keff Effective numberof overlappingWDM pulses
>. Wavelength = ch». [m]
T/atm Atmospheric loss
T/eh Net channel loss =T/TxOptT/atmT/PATgT/spaeeT/RXOpt
T/link Overall link efficiency =T/TXT/ehT/RxTX-Joules/RX-bit, [Jlbit]
T/PAT Pointing and trackinglosses
T/RX Net Receiverefficiency =PPBQLGeode[Photonslbit]
T/RXOpt Receiving optics (telescope) losses
T/spaee Space diffraction loss =[11" DTD R/(4AL )]2
T/Tx Net Transmitterefficiency (ElO) [TX-Joules/Photon]
T/TXER TransmitterER penalty
T/TXOpt Transmitting optics (telescope) losses
v Optical frequency =c] >. [Hz]
Tbit Bit or symbolduration [s]
To 3 dB pulsewidth of optical filter impulseresp. [s]
r, 3 dB pulsewidth of signal (power) [s]
B Bandwidth[Hz]
B; Optical bandwidth[Hz]
c Speed of light in vacuumc = AV =3.0 X 108 [m/s]
DTX Diameterof transmitaperture [m]
DRX Diameterof receiveaperture [m]
DCeff Effective duty cyle DCeff =DCmod DCpul se
DCmod Modulation duty cycle
DCpul se Pulse shape duty cycle
e Electroncharge 1.601 x 10- 19 [C]
Ie RF filter3-dB bandwidth (cutoff frequency) [Hz]
I -FSK 1= # of FSK frequencies in a hybrid symbol
go Optical amplifiersmall-signal gain
G Gain
G eode FEC coding gain
h Planck's constant 6.626 x 10- 34 [J/s]
ke Boltzman's constant 1.379 x 10- 23 [WICK-Hz]
k numberof bits-per M -ary symbol k = log2(M)
keff Effective numberof bits-per M -ary symbol
L-PolSK L = # of PoSLK polarizations in a hybrid symbol
M Symbol constellation size, as in M -ary PPM
M= r ;k = )Og2 (M) = # of bits/symbol
Navg Average photoncount per interval [photons]
Nb Background-noise photoncount [photons]
N o /2 Noise power spectral density [W/Hz]
NRX-min Minimum required RX SNR
N shot Shot noise power spectral density = hv [WlHz]
N sp Spontaneous emission factor [photons/mode]
Nth Thermal noise power spectral density = kBT [W/Hz]
220 DavidO. Caplan
Pin-sat Amplifier input saturation power Pin- sat = Psat/go

Psat Amplifier saturated output power
P sh ot Shot noise power spectral density = lu/ B [W]
Po. Thermal noise power = ke T B [W]
PPBQL Uncoded quantum-limited RX sensitivity [photonslbit]
p-PPM p = # of PPM positions in a hybrid symbol
r Spectral efficiency [(bits/s)/Hz]
Rph Photon flux rate P / (hv) [photons/s]
Rb Bit rate, Rs = 1/7bit[bits/s]
R A / W Detector responsivity, R A / W = nq]hv [mA/mW]
V" Switching voltage
w Number ofWDM channels
z, Impedance [l = [VIA]
Acronyms and Abbreviations
ADC Analog-to-digital converter

APL Average power limited
ARTEMIS Advanced Relay Technology Mission Satellite
ASK Amplitude shift keying, 2-ASK the same as OOK
AWGN Additive white Gaussian noise
BER Bit-error rate
CML Chirp-managed laser
CW Continuous wave
COTS Commercial off-the-self
DAC Digital -to-analog converter
DC Duty cycle
dB Decibel = 1O*log10 (Powerl Pref)
dBm 10 loglO(PowerlI mW)
dBW 10 loglO(PowerlI W)
DI Delay-line interferometer
DPSK Differential phase shift keying
EDFA Erbium-doped fiber amplifier
EYDFA Erbium-Ytterbium (codped) fiber amplifier
EM Electromagnetic
EMI Electromagnetic interference
ER Extinction ratio = PowerofriPoweron
FBG Fiber Bragg grating
FEC Forward error correction (coding)
AT Failure in time (#failures per billion hours of operation)
FM Frequency modulation
FP Fabry-Perot
FP n nth order Fabry -Perot
FSK Frequency-shift keying
FSO Free- space optical
FWHM Full width half max
FWM Four-wave mixing
Gbps Gigabit per second

GEO Geostationary earth orbit
GeoLiTE Geosynchronous Lightweight Technology Experiment
GOLD Ground -to-Orbit Lasercom Demonstration
1M Intensity modulation
IM-DD Intensity-modulated direct-detection
ICI Inter-channel-interference
lSI Inter-symbol-interference
ITU International Telecommunications Union
ITU-R Radiocommunication Standard
ITU-T Telecommunication Standard
ITU-grid Standard 100 GHz WDM channel spacing
LEO Low earth orbit
Mbps Megabit per second
MLCD Mars Laser Communication Demonstration
MZI Mach-Zehnder interferometer
MZM Mach-Zehnder Modulator
MTTF Mean time to failure
NA Numerical aperture, NA =nsin(B)
NF Noise Figure
mCETS Optical Inter-orbit Communications Engineering Test Satellite
OOK On-off keying (binary-ASK)
PAT Pointing acquisition and tracking
PM Polarization maintaining
PPB Number of photonslbit
PPM Pulse position modulation
PPL Peak power limited
PRBS Pseudo-random bit sequence
PSD Power spectral density, [dBmlHz]
P[alb] Probability of a given b
PER Polarization extinction ratio
RF Radio frequency, generic term for electrical
RX Receiver
SCOWA Slab-coupled optical waveguide amplifier
SCM Subcarrier multiplexing
SBS Stimulated Brillouin scattering
SILEX Semiconductor Intersatellite Link Experiment
SPM Self-phase modulation
SPOT-4 System Probatoire/Satellite Pour d'Observation de la Terra 4 satellite
SNR Signal-to-noise-ratio
SRS Stimulated Raman scattering
TDM Time division multiplexing
TEC Thermo-electric cooler
TIA Transimpedance amplifier
TX Transmitter
VDC Variable duty cycle
VBG Volume Bragg grating
WDM Wavelength division multiplexing or multiplexer
YDFA Ytterbium-doped fiber amplifier
222 David O. Caplan
References
I. B.L. Edwards, et. al., "Overview of the Mars laser communications demonstration project,"
in American Institute of Aeronautics and Astronautics, Space 2003 Conference & Exposi-
tion,2oo3.
2. S.A. Townes, et. al., "The Mars laser communication demonstration," presented at IEEE
Aerospace Conf., 2004.
3. E.A. Swanson and R.S. Bondurant, "Fiber-based free-spa ce optical system," in US Pat.
5,062,150 ,1991.
4. "Reliability assurance practices for optoelectronic device s in interoffice application s," Bell-
core TR-NWT-000468 Issue I, Dec. 1991.
5. "Generic reliability assurance requirements for optoelectronic devices used in telecommu -
nications equipment," Telcordia GR-468-CORE no. 2, Dec. 2002.
6. "Generic requirements for fiber optic branching components," Bellcore GR-1209-CORE,
Issue I, Nov. 1994.
7. "Generic requirements for optical fiber amplifiers," GR-1312 -CORE, Issue 2, Dec. 1996.
8. "Generic reliability assurance requirements for fiber optic branching components," GR-
1221-CORE , Issue I, Dec. 1994.
9. "Test methods and procedures for microelectronics," Military Standard MIL-STD -883C,
Aug. 1983.
10. D.O. Caplan, M.L. Stevens, and D.M. Boroson, "Variable-rate communication system with
optimal filtering," in US Pat. 6,694,104, 2004, (filed 1998).
II. D.O. Caplan, M.L. Stevens, D.M. Boroson, and J.E. Kaufmann, "A multi-rate optical com-
munications architecture with high sensitivity," in LEOS, 1999.
12. M.L. Stevens, D.M. Boroson, and D.O. Caplan, "A novel variable-rate pulse-position mod-
ulation system with near quantum limited performance," in LEOS, 1999.
13. D.O. Caplan and WA. Atia, "A quantum-limited optically-matched communication link,"
in Optical Fiber Conference (OFC) , 2001.
14. D.O. Caplan and W.A. Atia, "Methods of achieving optimal communications performance,"
in US Pat. 7,181,097, 2007.
IS. D.O. Cap,lan, B.S. Robinson, R.J. Murphy, and M.L. Stevens, "Demonstration of2.5-Gslotls
optically-preamplified M-PPM with 4 photons/bit receiver sensitivity," in Optical Fiber
Conference (OFC): Paper PDP23, 2005.
16. C.E. Shannon, "A mathematical theory of communication," Bell Syst. Technol. J. 27, 379-
423, 623-656 , 1948.
17. S. Haykin, Digital Communications (John Wiley & Sons, Inc., 1988).
18. J.G. Proakis and M. Salehi , Communication Systems Engineering (Prentice-Hall, Inc., 1994).
19. Peebles, Peyton Z. Jr., Digital Communications System s (Englewood Cliffs, NJ: Prentice-
Hall, 1987).
20. S.B. Alexander, Optical communication receiver design (Bellingham, Washington, USA:
SPIE Optical Engineering Press, 1997).
21. D.O. Caplan , "High-performance free-space laser communications and future trends," in
Optical Amplifiers and Their Applications (OAA'05) Topical Meeting, Budapest, Hungary,
2005.
22. D.O. Caplan , B.S. Robinson , M.L. Stevens, D.M. Boroson, and S.A. Hamilton, " High-Rate
Photon-Efficient Laser Communications with Near Single Photonlbit Receiver Sensitivi-
ties," in Optical Fiber Conference (OFC) , 2006.
23. D.M. Boroson, "Optical Communications, A Compendium of Signal Formats , Receiver
Architectures, Analysis Mathematic s, and Performance Comparisons," 2005.
24. T. Mizuochi, et. al., "Forward error correction based on block turbo code with 3-bit soft
decision for 10-Gb/s optical communication systems," IEEE Sel. Top. Quantum Electron .
10,376-386,2004.
25. N.W. Spellmeyer, J.e. Gottschalk, D.O. Caplan, and M.L. Stevens, "High-sensitivity 40-
Gb/s RZ-DPSK with forward error correction," IEEE Photon. Technol. Lett. 16, 1579-1581
(2004) .
26. J.R. Pierce, "Optical Channels : Practical Limits with Photon Counting," IEEE Trans . Com-
mun. COM-26, 1819-1821, 1978.
27. Y.Yamamoto and H.A. Haus, "Preparation, measurement and informat ion capacity of optical
quantum states," Rev. Mod. Phys. 58, 1001-1020 , 1986.
28. H.A. Haus, "Limits on communication using photon s," in MIT-EECS Colloquium Series,
1997.
29. G.N. Gol'tsman, O. Okunev, G. Chulkova, A.L.A. Semenov, K. Smirnov, B. Voronov, A.
Dsardanov, C. Williams, and R. Sobolewski , "Picosecond superconducting single -photon
optical detector," Appl. Phys. Lett. 79, 705-707, 2001.
30. K.A. McIntosh, et. al., "lnGaAsPnnP avalanche photodiodes for photon counting at 1.06
/-Lm," Appl. Phys. Lett. 81 (2002).
31. J. Zhang, W. Slysz, A. Verevkin, O. Okunev, G. Chulkova, A. Korneev, A. Lipatov, G.N.
Gol'tsman, and R. Sobolew ski, "Response time characterization of NbN superconducting
single-photon detectors," IEEE Trans. App. Supercond. 13, 180-183,2003.
32. K.A. McInto sh, et. al., "Arrays of III-V semiconductor Geiger-mode avalanche photodi-
odes," in LEOS, 2003.
33. B.S. Robinson, A.J. Kerman, E.A. Dauler, RJ. Barron, D.O. Caplan, M.L. Stevens , JJ.
Carney, S.A. Hamilton, J.K .W. Yang, and K.K. Berggren, "78 1-Mbitls Photon-Counting
Optical Communications Using Superconducting NbN-Nanowire Detectors," Opt. Lett. 31,
444- 446 (2006).
34. B.S. Robinson ,AJ. Kerman, E.A. Dauler, RJ. Barron, D.O. Caplan, M.L. Stevens, 1.1. Car-
ney, S.A. Hamilton , J.K.W. Yang, and K.K. Berggren, "High-Data-Rate Photon-Counting
Optical Communications Using a NbN Nanowire Superconducting Detector," in Conference
on Lasers and Electro-Optics (CLEO), 2006.
35. E.A. Dauler, B.S. Robinson , AJ. Kerman, V. Anant, RJ. Barron, K. Berggren, D.O. Caplan,
1.1. Carney, SA Hamilton , K.M. Rosfjord , M.L. Stevens, and J.K.W. Yang, "1.25-Gbitls
photon -counting optical communications using a twoelement superconducting nanowire
single photon detector," in Proc. SP1E, (Advanced Photon Counting Techniques) , 2006.
36. P.S. Henry, "Error-rate performance of optical amplifiers," in Optical Fiber Conference
(OFC), 1989.
37. N.A. Olsson , "Lightwave Systems With Optical Amplifiers," J. Lightwave Technol. 7, 1071-
082 (1989) .
38. P.A. Humblet and M. Azizoglu, "On bit error rate of lightwave systems with optical ampli-
fiers," J. Lightwave Technol. 9 (1991) .
39. S.D. Personick, "Receiver design for digital fiber optic communication systems, I & II,"
Bell Syst. Tech. J. 52, 843-886, 1973.
224 David O. Caplan
40 . S. Vanstone and P.C. v. Oorschot, An introduction to Error Correcting Codes with Applica-
tions (Kluwer Academic Publishers, 1989).
41. D.O. Caplan, J.e. Gottschalk, RJ . Murphy, N.W. Spellmeyer, M.L. Steven s, and A.M.D.
Beling, "Performance of high-rate high-sensitivity optical communications with forward er-
ror correction coding," in Conference on Lasers and Electro-Optics (CLEO) : Paper CPDD9,
2004 .
42. B.E. Moision and J. Hamkins, "Coded modulation for the deep-space optical channel : seri-
ally concatenated pulse-position modulation," 42-161 , 2005 .
43. D.M. Boroson, C.C. Chen , and B.L. Edwards, "The Mars laser communication demon-
stration project: truly ultralong-haul optical transport," in Optical Fiber Conference (OFC) ,
2005.
44. P.I. Hopman, P.w. Boettcher, L.M. Candell, J.B. Glettler, R. Shoup, and G. Zogbi , "An
End-to-End Demonstration of a Receiver Array Based Free-Space Photon Counting Com-
munications Link ," in SPIE, (Free-Space Laser Communications VI), vol. 6304, 2006.
45. R. Shoup, "Hardware implementation of a high-throughput 64-PPM serial concatenated
turbo decoder," in SPIE, (Optical Information Systems IV), vol. 6311, 2006.
46. E.A. Swanson and R.S. Bondurant, "Using fiber optics to simplify flee-space lasercom
system s," in Proc . SPIE , (Free -Space Laser Communication Technologies II), 1218, 70-82
(1990) .
47. R.M. Gagliardi and S. Karp, Optical Communication, 2nd Ed. (New York,: John Wiley &
Sons, Inc., 1995).
48. F. G. Walther, J.M. Roth, W.E. Keicher, A.E. DeCew, and "Wavelength division and polar-
ization division multiple access free space optical terminal using a single aperture," in US
Pat. Appl. 20040081466, 2004 .
49. A.G. Bell and S. Tainter, "Photophone transmitter," in US Pat. 235,496,1880.
50. A.G. Bell, "On the Production and Reproduction of Sound by Light," Am. J. Sci., 3rd Series,
XX, 305-324 (\880).
51. A.G. Bell and S. Tainter, "Photophonic receiver," in US Pat. 241,909,1881.
52 . A.G. Bell, C.A. Bell, and S. Tainter, "Transmitting and recording sounds by radiant energy,"
in US Pat. 324,213,1886.
53. D. Killinger, "Free space optics for laser communication through the air," in Optics &
Photonics News, 2002.
54. Hertz, "Heinrich Rudolf Hertz discovered GHz radio waves in 1887," 1887.
55. A.G. Bell, "Improvement in Telegraphy," in US Pat. 174,465, 1876.
56 . T.A. Edison, "Telephones or Speaking Telegraphs," in US Pat. 203,018, 1878.
57. G. Marcon i, "A system of telegraphy using Hertzian waves," in British Pat. 12039, (filed
June 2, 1896, demonstrated Sept. 2, 1896), 1896.
58 . M. Loomis , "Improvement in Telegraphing," in US Pat. 129,971, 1872 (first US wireless
telegraphy patent) .
59. R.A. Fessenden, "Wireless Telegraphy," in US Pat. 706,737,1902, (filed 1901).
60. A.L. Schawlow and e.H. Townes., "Infrared and Optical Masers ," Phys. Rev.H2, 1940-
1949 (\ 958) .
61. T.H. Maiman, "Stimulated optical radiation in ruby," Nature 187,493 (\960).
62. T.H. Maiman , "Ruby Laser Systems," in US Pat. 3,353,115, 1967, (filed 1963).
63. A.lavan, Bennett, W.R.lr., and D.R. Herriott , "Population Inversion and Continuous Optic al
Maser Oscillation in a Gas Discharge Containing a He-Ne Mixture," Phys. Rev. Lett. 6, 106-
110,1961.
64. "Re search groups at GE, IBM, and Lincoln Laboratory at MIT demonstrated semiconductor
lasers using gallium arsen ide (GaA s) in 1962. (Courtesy W. Keicher)," 1962.
65. C.H. Gooch , Gallium Arsen ide Lasers (New York: lohn Wiley & Sons, 1969).
66. C.1. Koester and E. Snitzer, "Amplification in a Fiber Laser," Appl. Opt. 3, 1182-86 (1964) .
67. E. Snitzer, "Means for producing and amplifying optical energy," in US Pat. 3,729 ,690,
(filed 1969, 1961), 1973.
68. Z.I. Alferov, V.M. Andreev, D.Z. Garbuzov, Y.V. Zhilyaev, E.P. Morozov , E.L. Portnoi , and
V.G. Trofim, "Inve stigation of the influence of the AIAs-GaAs heterostructure parameters on
the laser threshold current and the realization of continuous emis sion at room temperature,"
Sov. Phys . Semicond. 4,1573-1575 (1971) .
69. CK. Kao and G.A. Hockham, "Dielectric-Fiber Surface Waveguides for Optical Frequen-
cies," Proc . lEE 133,1151-58 (1966) .
70. R.D. Maurer and P.c. Schultz , "Fused silica optical waveguide ," in US Pat. 3,659 ,915, 1972,
(filed 1970).
71. D.B. Keck and P.c. Schultz, "Method of producing optical waveguide fibers," in US Pat.
3,711,262, 1973, (filed 1970).
72. C.K . Kao and T.W. Davies, "Spectroscopic studies of ultra low loss optical glasses ," 1. Sci.
Instrum. (1968) .
73. EP. Kapron, D.B. Keck, and R.D. Mauer, "Radiation losses in glass optical waveguides,"
Appl. Phys. Lett. 17, (1970) .
74. E. Snitzer, "Cylindrical dielectric waveguide modes," 1. Opt. Soc. Am. (1961) .
75. E. Snitzer, H. Po, R.P. Tumminelli, and E Hakimi, "Optical fiber lasers and amplifiers," in
US Pat. 4,815 ,079, (filed 1987), 1989.
76. R.J. Mears, L. Reekie, I.M. Jauncey, and D.N. Payne, "Low-noise erbim-doped fibre am-
plifier operating at 1.54 fJom," Electron . Lett. 23, 1026 (1987) .
77. E. Desurvire, 1.R. Simpson , and P.c. Becker, "High-gain erbium-doped traveling wave fiber
amplifier," Opt. Lett. 12, 888 (1987).
78. "TAT-8, the first trans-Atlantic fiber cable (2 x 280 Mb/s, 1.3 fJom), was AT&T' s 8th transat-
lantic telephone cable, in operation from 1988, initially carry ing 40,000 telephone circu its
between USA and France .."
79. "TAT- IO, AT&T's 10th transatlantic telephone cable, in operation from 1992, initially car-
rying 2 x 565Mb/s between USA and German y, using 1.5 fJom technology."
80. Y. Arimoto, M. Toyoshima , M. Toyoda, T. Takahashi, M. Shikatani, and K. Araki , "Pre-
liminary result on laser communication experiment using Engineering Test Satellite-VI
(ETS-VI)," in Proc . SPIE 2381,1995.
81. K.E. Wilson, 1.R. Lesh, K. Araki, and Y. Arimoto, "Preliminary result s of the Ground/Orbiter
Lasercom Demonstration experiment between Table Mountain and the ETS- VI satellite,"
in Proc. SPIE (Free-Space Laser Communication Technologies VIII), 1996.
82. K.E. Wilson, et. al., "Results from Phase- I and Phase-2 GOLD experiments," Feb. 15 1997.
83. K.E. Wilson and l .R. Lesh, "Overview of the Ground-to-Orbit Lasercom Demon stration
(GOLD)," in Proc. SPIE (Free-Space Laser Communication Technologies IX), 1997.
84. "TAT- I2I13 (1996) first used optical amplification (EDFA) and ring topology."
226 David O. Caplan
85. A.R. Chraplyvy, A.H. Gnauck, R.W. Tkach, J.L. Zyskind, J.W. Sulhoff, AJ. Lucero, Y.
Sun, R.M. Jopson, F. Forghieri , R.M. Derosier, C. Wolf, and A.R. McCormick, " I-ThIs
transmission experiment," Photonics Technol. Lett. 8,1264-1266 (1996).
86. A.H. Gnauck, et. al., "One Terabitls Transmi ssion Experiment," presented at Optical Fiber
Conference (OFC), 1996.
87. T. Morioka, et. aI., " 100 Gb/s x 10 Channel OTDMlWDM Transmission Using a Single
Supercontinuum WDM Source," presented at Optical Fiber Conference (OFC), 1996.
88. H. Onaka, et. al., "1.1 ThIs WDM Transmission over a 150 km 1.3 mm Zero-Dispersion
Single-Mode Fiber," presented at Optical Fiber Conference (OFC), 1996.
89. KMI Corporation, Nashua , NH, USA .
90. S: Bigo, A. Berta ina, Y. Frignac, S. Borne, L. Lorey, D. Hamoir, D. Bayart, J.-P. Hamaide ,
W. Idler, E. Lach, B. Franz, G. Veith, P. Sillard, L. Fleury, P. Guenot, and P. Nouchi , "5 .12
Tbitls (128 x 40 Gbitl s WDM) transmission over 3 x 100 km ofTeraLight fiber," in Proc.
Eur. Conf. Optical Communications (ECOC), paper PDI .2, Munich, Germany, 2000.
91. W. Idler, S. Bigo, Y. Frignac, B. Franz, and G. Veith, "Vestigial side-band demultiplexing for
ultra-high capacity (0.64 bitlslH z) of 128 x 40 Gbitl s channels," in Optical Fiber Conference
(OFC) ,2001.
92 . K. Fukuchi, T. Kasamatsu, M. Morie, R. Ohhira, T. Ito, K. Sekiya, D. Ogasahara, and T. Ono,
"10,92 Tbl s (273 x 40 Gb/s) triple band/ultra dense WDM optical repeatered transmi ssion
experiment," presented at Optical Fiber Conference (OFC) , 200 I.
93. S. Bigo, Y. Frignac, G. Charlet, W. Idler, S. Borne, H. Gross, R. Dischler, W. Poehlmann, P.
Tran, C. Simonneau, D. Bayart, G. Veith, A. Jourdan, and J.-P. Hamaide , " 10.2 Thitls (256
x 42.7 Gbitls PDMlWDM) transmission over 100 km TeraLight fiber with 1.28 bitls/Hz
spectral efficiency," in Optical Fiber Conference (OFC), paper PD25, 2001.
94. Y. Frignac, G. Charlet, W. Idler, R. Dischler, P. Tran, S. Lanne, S. Borne, C. Martinelli , G.
Veith, A. Jourdan , J.-P. Hamaide, and S. Bigo, "Transmission of 256 wavelength-division
and polarization-division multiplexed channels at 42.7 Gb/s (10.2 ThIs capacity) over 3 x
100 km of TeraLight fiber," in Optical Fiber Conference (OFC), 2002.
95. "Press Release : NRO GeoLITE Satellite Succe ssfully Launched". http ://cartome.org/geolite
.htm, 200 1.
96. "Press Release : Delta Launches GeoLITE Satellite for U.S. NRO."
http://www.spaceandtech.comldigestlflash200 IIflash200 1-038.shtml Andrews Space &
Technology, 200 I.
97. "Press Release: NRO Awarded the David Packard Excellence in Acquisition Award".
http://www.nro.govlPre ssReleases/prueI62.html. 2002.
98. T.T. Nielsen and G. Oppenhaeuser, "In-orbit test result of an operational optical intersatell ite
link between ARTEMIS and SPOT4 , SILEX," in Proc . SPIE, (Free-Space Laser Commu -
nication Technolog ies XIV), 2002.
99 . T.T. Nielsen, G. Oppenhaeuser, B. Laurent, and G. Planche , "In-orbit test results of the optical
intersatellite link, SILEX . A milestone in satellite communication," in 53rd International
Astronautical Congre ss, IAC-02-M .2.0 I, 2002.
100. H.P. Lutz , "Optical Communications in Space-Twenty Years of ESA Effort," in ESA Bul-
letin ( http://esapub.esrin.esa.itlbulletinlbullet911b911utz.htm), 1,25-31 (1997) .
101. B.I. Edelson, J.N . Pelton, c.w. Bostian , W.T. Brandon, V.W.S. Chan, E.P. Hager, N.R. Helm,
R.D. Jennings, R.K. Kwan, C.E. Mahle , E. F. Miller, and L. Riley, "Satellite Communica-
tions Systems And Technology," in http ://www.wtec .orglloyolalsatcoml ,Loyola College in
Maryland, 1993.
102. M.Reyes,S. Chueca,A.Alonso, T. Viera, and Z. Sodnik,"Analysisof thepreliminary optical

links betweenARTEMIS and the Optical GroundStation," in Proc. SPIE 4821, 2003.
103. A. Alonso, M. Reyes, and Z. Sodnik, "Performance of satellite-to-ground communications
link betweenARTEMIS and the Optical Ground Station," in Proc. SPIE 5572, 2004.
104. D.M. Boroson, e.e. Chen, and B.L. Edwards, "The Mars laser communications demon-
strationproject: truly ultralong-haul optical transport," in OpticalFiber Conference (OFC),
2005.
105. M. Toyoshima, "SpecialReport: Trendsof research and development of optical spacecom-
munications technology," SpaceJapan Review 12- I, No. 44, 2005.
106. E. Hecht, Optics, 2nd Ed. (Addison Wesley, 1987).
107. A. Biswas, K.E. Wilson, S. Piazzolla, J.P.Wu,and W.H. Farr,"Deep-space opticalcommu-
nications link availability and data volume," in Proc. SPIE 5338, 175(2004).
108. F.I. Khatri, D.M. Boroson,D.V. Murphy, andJ. Sharma, "Linkanalysis of Mars-Earth optical
communications system," in Proc. SPIE 5338, 143(2004).
109. F.I. Khatri and A. Biswas, "Signal and Background Levelsfor the Mars Lasers Communi-
cations Demonstration (MLCD)," in IEEE LEOS SummerTopical Meetings, 2005.
110. R.W. Boyd, Nonlinear Optics (NewYork: Academic Press, Inc., 2003).
III . G.P. Agrawal, Nonlinear Fiber Optics, 2 n d ed. (NewYork: Academic Press, Inc., 1995).
112. W.E. Webband J.T. Marino, Jr., "Threshold detection in an on-off binary communications
channel with atmospheric scintillation," Appl. Opt. 14, 1413-1417(1975).
113. W.e. Brown,"OptimumThresholds for Optical On-Off Keying Receivers Operating in the
Turbulent Atmosphere," in Proc. SPIE, (Free-Space Laser Communication Technologies
IX), vol. 2290, 254-261 (1997).
114. H. Haunstein, R. Schlenk, and K. Sticht, "Control of Combined Electrical Feed-Forward
and Decision FeedbackEqualizationby Conditional ErrorCountsfromFEC in the Presence
of PMD," in Optical FiberConference(OFC), 2003.
115. D.M. Castagnozzi, "Digital signal processing and electronic equalization (EE) of lSI," in
Optical Fiber Conference (OFC), 2004.
116. M. Schwartz, W.R. Bennett, and S. Stein, Communication Systems and Techniques (New
York: IEEE Press, 1996).
117. S.D. Personick, P. Balaban, J. Bobsin, and P. Kumar, "A Detailed Comparison of Four
Approaches to the Calculation of the Sensitivity of Optical Fiber System Receivers," IEEE
Trans. Comm. 25,541-548 (1977).
118. R.G. Smith and S.D. Personick, Semi condu ctor device s for optical communication (New
York: Springer-Verlag, 1982), vol. 39.
119. S.O. Rice, "Mathematical analysis of random noise," Bell Syst. Technol. J. 24, 46--156
(1945).
120. S. SteinandJ.1. Jones, Modern communication principles (NewYork: McGraw-Hili, 1967).
121. Saleh, B.E.A. and M.e. Teich, Fundamentals of Photonics (NewYork: Wiley, 1991).
122. T. Li and M.e. Teich, "Bit-Error Rate For A Lightwave Communication System Fibre
Amplifier Incorporating An Erbium-doped Fibre Amplifier," Electron. Lett. 27, 598-599
(1991).
123. T. Li and M.e. Teich,"Photon Point Process for Traveling-Wave Laser Amplifiers," IEEE
J. QuantumElectron. 29, 2568-2578 (1993).
228 DavidO. Caplan
124. W.S. Wong, H.A. Haus, L.A. Jiang, P.B. Hansen, and M. Margalit, "Photon statistics of
amplifiedspontaneousemissionnoisein a IO-Gbitys opticallypreamplified direct-detection
receiver," Opt. Lett. 23, 1832-834 (1998).
125. W.S. Wong, J.D. Moores,J. Korn,and H.A. Haus,"Photon statisticsofNRZ signalsin high-
bit-rate optically pre-amplifieddirect detectionreceiver," Optical Fiber Conference(OFC),
1999.
126. E. Desurvire, Erbium-dopedfiberamplifiers (NewYork: John Wiley& Sons, 1994).
127. H.A. Haus, Electromagnetic Noise and Quantum OpticalMeasurements (Springer-Verlag,
2000).
128. J.C. Livas, "High sensitivity optically preamplified 10 Gb/s receivers," in Optical Fiber
Conference(OFC), Paper PD4, 1996.
129. W.A. Atia and R.S. Bondurant, "Demonstration of return-to-zero signaling in both OOK
and DPSK formats to improvereceiversensitivity in an optically preamplified receiver," in
LEOS, 1999.
130. A.H. Gnauck, S. Chandrasekhar, J. Leuthold, and L. Stulz, "Demonstration of 42.7-Gb/s
DPSK receiver with 45 photonslbit sensitivity," IEEE Photon. Technol. Lett. 15,99-101
(2003).
131 . J.H. Sinsky, A. Adamiecki,A.H. Gnauck,C.A. Burrus,J. Leuthold,O. Wohlgemuth, and A.
Umbach,"A 42.7-Gb/s IntegratedBalancedOptical Front End with RecordSensitivity," in
Optical Fiber Conference(OFC): Paper PD39-1, 2003.
132. A.H. Gnauckand PJ . Winzer, "Optical phase-shift-keyed transmission," J. Lightwave Tech-
nol. 23, 115-130 (2005).
133. D.O. Caplan,M.L. Stevens,J.1. Carney,and R.1. Murphy, "Demonstration of OpticalDPSK
Communication with 25 PhotonslBit Sensitivity," in Conference on Lasers and Electro-
Optics (CLEO), 2006.
134. J.R. Minch, D. J. Townsend, and D.R. Gervais, "Rate Adjustable NRZ-DPSKModulation
Scheme with a Fixed Interferometer," in IEEE LEOS, 2005.
135. D.O. Caplan, M.L. Stevens, and U . Carney, "A High-Sensitivity Multi-Channel Single-
InterferometerDPSK Receiver," Opt. Express 14,10984-10989 (2006).
136. D.O. Caplan, M.L. Stevens,and J.1.Carney,"High-Sensitivity Demodulation of Multiple-
Data-RateWDM-DPSKSignals usinga SingleInterferometer," inOptical FiberConference
(OFC),2007.
137. K. Yonenaga and K. Hagimoto,"IO-Gitls x four-channel WDM transmission experiment
over 2400-km DSF using optical DPSK direct detection scheme,"in Optical Fiber Confer-
ence (OFC), 1997.
138. A.H. Gnauck, G. Raybon, S. Chandrasekhar, J. Leuthold, C. Doerr, L. Stulz, A. Agarwal,
S. Banerjee, D. Grosz, S. Hunsche,A. Kung,A. Marhelyuk, D. Maywar, M. Movassaghi,
X. Liu, C. Xu, X. Wei,and D.M. Gill, "2.5 ThIs (64 x 42.7 Gb/s) transmission over 40 x
100 km NZDSF using RZ-DPSKformat and all-Raman-amplified spans," in Optical Fiber
Conference(OFC), 2002.
139. J.-X. Cai, D.G. Foursa,C.R. Davidson, Y. Cai, G. Domagala,H. u, L. Liu, W.Patterson,A.
Pilipetskii,M. Nissov, and N. Bergano,"A DWDMDemonstrationof3 .73 ThIs over 11,000
km using 373 RZ-DPSK Channels at 10 Gb/s," in Optical Fiber Conference (OFC): Paper
PD22-1,2003.
140. C. Rasmussen, T. Fjelde,J. Bennike,F. Liu, S. Dey,B. Mikkelsen, P. Mamyshev, P.Serbe, P.
vander Wagt,Y. Akasaka, D. Harris,D. Gapontsev, V. Ivshin, and P. Reeves-Hall, "DWDM
40G transmission over trans-Pacific distance (10 000 km) using CSRZ-OPSK, enhanced
FEC and all-Raman amplified 100 km Ultrawave fiber spans," in Optical Fiber Conference
(OFC) : Paper POI8-1 , 2003.
141. B. Zhu, L. E. Nelson, S. Stulz ,A .H. Gnauck, C. Doerr, 1. Leuthold, L. Gruner-Nielsen, M.O.
Pedersen, 1. Kim, R.L. Lingle, Jr., Y. Emori, Y. Ohki, N. Tsukiji, A. Oguri, and S. Namiki,
"6.4 Tbitls (160x42.7 Gb/s) transmission with 0.8 bitls/Hz spectral efficiency over 32 x 100
km of fiber using CSRZ-DPSK format," in Optical Fiber Conference (OFC) : Paper PD 19,
2003.
142. M.L. Stevens, "Transmitter pulse issues," private communication, 2004.
143. R.A. Griffin, "Integrated DQPSK Transmitters," in Optical Fiber Conference (OFC), 2005.
144. H. Kim and P.1. Winzer, "Robustness to laser frequency offset in direct-detection DPSK and
DQPSK systems," 1. Lightwave Technol. 21, 1887-1891 (2003).
145. G. Bosco and P. Poggiolini, "On the joint effect of receiver impairments on direct-detection
DQPSK systems," 1. Lightwave Technol. 24, 1323-1333 (2006) .
146. P.1. Winzer and H. Kim, "Degradations in balanced DPSK receivers," IEEE Photon . Technol.
Lett. 15, 1282-1284 (2003) .
147. I. Morita and N. Yoshikane, "Merits of DQPSK for Ultrahigh Capacity Transmission," in
LEOS, 2005.
148. A. Royset and D.R. Hjelme, "Novel dispersion tolerant optical duobinary transmitter using
phase modulator and Bragg grating filter," in ECOC, 1998.
149. Y. Miyamoto, H. Masuda, A. Hirano, S. Kuwahara, Y. Kisaka, H. Kawakami, M. Tomizawa,
Y. Tada, and S. Aozasa , "S-band WDM coherent transmission of 40 x 43-Gbitls CS-RZ
DPSK signals over 400 km DSF using hybrid GS-TDFAs/Raman amplifiers," Electron . Lett.
39,1569-1570 (2002) .
150. O. Penninckx, H. Bissessur, P. Brindel, E. Gohin, and E Bakhti, "Optical differential phase
shift keying (DPSK) direct detection considered as a duobinary signal," in ECOC, 2001.
151. Y. Miyamoto, A. Hirano, S. Kuwahara , M. Tomizawa, and Y. Tada, "Novel modulation and
detection for bandwidth-reduced RZ formats using duobinary-mode splitting in wideband
PSKIASK conversion," 1. Lightwave Technol. 20, 2067-2078 (2002) .
152. L. Moller, C. Xie, R. Ryf, L. Xiang, and X. Wei, "10 Gb/s duobinary receiver with a record
sensitivity of 88 photons per bit," presented at Optical Fiber Conference (OFC), 2004 .
153. L.G. Kazovsky and D.A. Atlas., "PSK synchronous heterodyne and homodyne experiments
using optical phase-locked loops," presented at Optical Fiber Conference (OFC), 1990.
154. L.G. Kazovsky and D.A. Atlas, "A I320-nm Experimental Optical Phase-Locked Loop:
Performance Investigation and PSK Homodyne Experiments at 140 Mb/s and 2 Gb/s,' 1.
Lightwave Technol. 8, 1414-1425 (1990).
155. ET. Herzog, "An optical phase locked loop for coherent space communications," Swiss
Federal Institute of Technology, 2006.
156. L.G. Kazovsky, S. Benedetto, and A. Willner, Optical Fiber Communication Systems (Nor-
wood, MA: Artech House , Inc., 1996).
157. B. Wandernoth, "5 photon/bit low complexity 2 Mbitls PSK transmission breadboard ex-
periment with homodyne receiver applying synchronization bits and convolutional coding,"
Proc. Eur. Conf. Opt. Commun. (ECOC), I, 59-62 (1994).
158. R. Lange and B. Smutny, " Highly-Coherent Optical Terminal Design Status and outlook,"
in LEOS, 2005 .
230 DavidO. Caplan
159. J.H. Sinsky, A. Adamiecki, A.H. Gnauck, C.A. Burrus, J. Leuthold, and O. Wohlgemuth,
"RZ-DPSK Transmission Using a 42.7-Gb/s Integrated BalancedOptical Front End With
RecordSensitivity," J. Lightwave Technol. 22, (2004).
160. S.B.Alexander, R.Barry, D.M.Castagnozzi,V.W.S.Chan,D.M. Hodsdon, L.L.Jeromin,J.E.
Kaufmann, D.M. Materna, RJ. Parr, M.L. Stevens, and D.W. White, "4-ary FSK coherent
optical communication system," Electron. Lett. 26,1346-1348 (1990).
161 . M.L. Stevens, D.M. Boroson, and J.E. Kaufmann, "A near-optimum discriminator demod-
ulatorfor binary FSK with widetone spacing," IEEEMicrowave and GuidedWave Lett.3,
227-229 (1993).
162. R. Noe, et. al., "Optical FSK transmission with pattern independent 119photoelectronslbit
receiversensitivity with endless polarization control," Electron. Lett. 25, 757-758 (1989).
163. B.S. Robinson "Semiconductor-based all-optical switching for optical time-division multi-
plexednetworks," Thesis, MIT,2003.
164. J.M. Kahn,"1 Gbit/sPSK homodyne transmissionsystemusingphase-lockedsemiconduc-
tor lasers," IEEE Photon. Technol. Lett. 1,340-342 (1989).
165. S. Norimatsu, K. Iwashita,and K. Noguchi, "10 Gbit/s optical PSKhomodyne transmission
experiments using externalcavity DFB LDs," Electron. Lett. 26, 648-649 (1990).
166. S. Norimatsu, K. Iwashita, and K. Sato, "PSK optical homodyne detection using external
cavity laser diodes in Costas loop," IEEE Photon. Technol. Lett. 2, 374-376 (1990).
167. B. Zhu, L. Leng, L. E. Nelson,L. Gruner-Nielsen,Y. Qian, J. Bromage, S. Stulz, S. Kado,
Y. Emori,S. Namiki,P.Gaarde,A. Judy,B. Palsdottir,and R.L. Lingle,Jr., "3.2 This (80 x
42.7 Gb/s) transmission over 20 x IOOkm of nonzerodispersionfiberwithsimultaneous C
+ L-banddispersioncompensation," in Optical Fiber Conference (OFC), 2002.
168. J.M. Ross, S. I. Green, and J. Brand, "Short-pulse optical communication experiments,"
Proc. IEEE 58, (1970).
169. J.R. Lesh, J. Katz, H.H. Tan, and D. Zwillinger, "2.5 bit/detected photon demonstration
program: description, analysis, and phase I results," Jet Propulsion Laboratory, Pasadena,
CA 42-66, Dec. 1981.
170. J.R. Lesh, "Capacity Limitof the Noiseless, Energy-Efficient OpticalPPMChannel," IEEE
Trans. Comm., 31, 546-548 (1983).
171. AJ. Phillips, R.A. Cryan, and J.M. Senior, "An optically preamplified intersatellite PPM
receiver ~mploying maximum likelihood detection," IEEE Photon. Technol. Lett. 8, 691-
693,1996.
172. D.O.Caplan, P.W.Juodawlkis, J.J. Plant,andM.L.Stevens, "Performance of high-sensitivity
OOK, PPM,and DPSKcommunications using high-powerslab-coupled optical waveguide
amplifier(SCOWA) based transmitters," in Optical FiberConference(OFC), 2006.
173. C. Chen and C.S. Gardner, "Performance of PLL Synchronized OpticalPPM Communica-
tion Systems," IEEETrans. Comm. COM-34, 988-994 (1986).
174. F. Nekoogar, Ultra-Wideband Communi cations: Fundamentals and Applications (Upper
Saddle River: PrenticeHall, 2005).
175. M.L. Stevens, "Estimation of M -PPMSpectrafor Pseudo-Random BitSequences." private
communication, 2006.
176. B.S. Robinson, S.A. Hamilton, and E.P. Ippen,"Demultiplexing of 80 Gbit/s pulse-position
modulated data withan ultrafastnonlinear interferometer," IEEEPhoton. Technol. Lett. 14,
2002.
177. R.S. Vodhanel, J.L. Gimlett, N.K. Cheung, and S. Tsuji, "FSK Heterodyne Transmis sion
Experiments at 560 Mbit/s and 1 Gbit/s," J. Lightwave Technol. LT·5, 461-468 (1987) .
178. A.R. Chraplyvy, R.W. Tkach, A.H. Gnauck, and R.M. Derosier, "8Gbit/s FSK modulation
of DFB lasers with optical demodulation," Electron. Lett., 25, 319-321 (1989).
179. B. Glance, et. aI., " Densely spaced FDM optical coherent system with near quantum-limited
sensitivity and computer controlled random access channel selection ," in Optical Fiber
Conference (OFC) : Paper PDII , 1989.
180. M.L. Stevens, B. R. Hemenway, and S.B.Alexander, "Simultaneous TDMlFDM digital opti-
cal transmission with polarization-diver sity heterodyne detection," Microwave Symposium
Digest, 1,171-173 (1994) .
181. R. Gross and R. Olshan sky, "Multichannel Coherent FSK Experiments Using Subcarrier
Multiplexing Techniques," J. Lightwave Technol. 8, 406-415 (1990) .
182. H. Gross, M. Schmidt , R. Olshansky, and V. Lanzisera, "Coherent Transmission of 60 FM-
SCM Video Channels," Photonics Technol. Lett. 2, 288-290 (1990) .
183. S. Bendetto , R. Gaudino , and P. Poggiolini, "Direct detection of optical digital transm ission
based on polarization shift keying modulation ," IEEE Sel. Areas Commun., 13, 531-542
(1995) .
184. E. Hu, K. Wong, M. Marhic, L.G. Kazovsky, K. Shimizu , and N. Nikuchi , "4-Level Direct-
Detection Polarization Shift-Keying (DD-PoISK) System with Phase Modulators," in Op-
tical Fiber Conference (OFC), 2003.
185. M.M. Matalgah and R.M. Radaydeh, "Hybrid Frequency-Polarization Shift-Keying Mod-
ulation for Optical Transmission," J. Lightwave Technol. 23, 1152-1162 (2005) .
186. M. Nazarathy and E. Simony, "Generalized Stokes Parameters-Shift Keying: A New Per-
spective on Optimal Detection Over Electrical and Optical Vector Incoherent Channels,"
IEEE Trans. Comm . 54, 499-509 (2006) .
187. M. Nazarathy and E. Simony, "Stoke s Space Optimal Detection of Multidifferential Phase
and Polarization Shift Keying Modulation," J. Lightwave Technol. 24, 1978-1988 (2006) .
188. R. Zhang and G.S. La Rue, "Clock and data recovery circuit s with fast acquisition and low
jitter," presented at IEEE Workshop on Microelectronics and Electron Devices, 2004.
189. N.W. Spellmeyer, "Communications performance of a multimode EDFA," IEEE Photon .
Techno]. Lett. 12, 1337-1339 (2000) .
190. R.H. Kingston, Detection of Optical and Infrared Radiation (New York: Springer-Verlag,
1978).
191. B. Glance, "Polarization independent coherent optical receiver," J. Lightwave Technol. 5,
274-276 (1987) .
192. B.S. Glance , K. Pollock , C.A. Burrus, B.L. Kasper, G. Einstein, and L.w. Stulz , "WDM
coherent optical star network," J. Lightwave Technol. 6, 67-72 (1988).
193. B.S. Glance and M. Kavehrad, "Pol arization-insensitive frequency-shift-keying optical het-
erodyne receiver using discriminator demodul ation," J. Lightwave Technol. 6, 1386-1394
(1988) .
194. N. Ohkawa, T. Sugie, and Y. Hayashi, "A highly sensitive balanced receiver for 2.5 Gbls
het-erodyne detection systems," IEEE Photon. Technol. Lett. 3, 375-377 (1991) .
195. B. Wandernoth, "20 photonlbit 565 Mbit/s PSK homodyne receiver using synchronization
bits," Electron. Lett. 28, (1992) .
232 DavidO. Caplan
196. EH. Raab, P. Asbeck, S. Cripps, P.B. Kenington, Z.B. Popovic, N. Pothecary, J.E Sevic,
and N.O. Sokal, "RF and Microwave PowerAmplifier and TransmitterTechnologies-Part
I," High FrequencyElectronics, 22-36 (2003).
197. P.W.Juodawlkis,J.J. Plant,R.K. Huang,L.J.Missaggia, andJ.P. Donnelly, "High-power1.5-
/spl mu/m InGaAsP-InP slab-coupled optical waveguide amplifier," IEEE Photon. Technol.
Lett. 17, (2005).
198. N.W.Spellmeyer, D.O.Caplan,and M.L. Stevens, "Design of a 5-WattPPMtransmitterfor
the Mars Laser Communications Demonstration," in LEOS, 2005.
199. P. Wysocki, T.Wood,A. Grant, D. Holcomb, K. Chang,M. Santo, L. Braun,and G. Johnson,
"High Reliability 49 dB Gain, 13W PM Fiber Amplifierat 1550 nm with 30 dB PER and
Record Efficiency," in Optical Fiber Conference(OFC), paper PDP17, 2006.
200. L. Goldberg,J.P.Koplow, and D.A.V. Kliner, "Highly efficient4-WYb-doped fiberamplifier
pumped by a broad-stripelaser diode," Opt. Lett. 24, 673-675 (1999).
201. A.N. Curren, J.A. Dayton, Jr., R.W. Palmer, KJ . Long, D.A. Force, C.E. Weeder, Z.A.
Zachar, and w.L. Harvey, "The Cassini mission Ka-bandTWT," in International Electron
DevicesMeeting, 1994.
202. D. Morabito, S. Butman, and S. Shambayati, "The Mars Global Surveyor Ka-Band Link
Experiment(MGS/KaBLE-II)," in Telecommunications and Mission Operations Progress
Report42-137, Jet Propulsion Laboratory, 1999.
203. I. Haque,"Ka-BandTravelingWaveTubeAmplifier,' in INDTechnology andSciencyNews,
Jet Propulsion Laboratory, 2002, pp. 11-14: http://tmot.jpl.nasa.govlProgram_Overview_
Information/lNDJ>rogram..NewslIssue 15.pdf.
204. N.W. Spellmeyer, D.O. Caplan, B.S. Robinson , D. Sandberg, M.L. Stevens, M.M. Willis,
D.V. Gapontsev, N.S. Platonov, and A. Yusim, "A High-Efficiency Ytterbium-Doped Fiber
AmplifierDesignedfor Interplanetary Laser Communications," in OpticalFiberConference
(OFC),2oo7 .
205. R. Loudon, The QuantumTheoryofLight (NewYork: Oxford University Press, Inc., 2000).
206. c.w. Gardiner and P. Zoller, QuantumNoise (NewYork: Springer Verlag, 2000).
207. J.W. Goodman,Statistical Optics (NewYork: John Wiley & Sons, Inc., 2000).
208. A. Papoulis, Probability, RandomVariables, and Stochastic Processes (NewYork: McGraw-
Hill, Inc., 1984).
209. R. Sobolewski, et. aI., "Ultrafast superconducting single-photonoptical detectors and their
applications," IEEE Trans. Appl Superconduct13, 1151-1157 (2003).
210. J.K.W. Yang, et. al., "Fabrication development for nanowire GHz-counting-rate single-
photon detectors," IEEE Trans. Appl, Superconduct., 2005.
211. B.S. Robinson, "Private communication," 2005.
212. J.H. Shapiro, "Imaging and Optical Communication throughAtmospheric Turbulence," in
Laser Beam Propagation in the Atmosphere,J.w. Strohbehn,Ed. (Berlin: Springer-Verlag,
1978).
213. RJ. Barron, "Binary shaping for low-duty-cycle communications," in International Sym-
posium on Information Theory (ISIT), 2004.
214. B.S. Robinson, D.O. Caplan, M.L. Stevens, RJ. Barron, E.A. Dauler, and S.A. Hamilton,
"1.5-photons/bitPhoton-CountingOpticalCommunications UsingGeiger-Mode Avalanche
Photodiodes,' in IEEE LEOS SummerTopical Meetings, 2005.
215. J.B.Johnson,"Thermalagitationof electricityinconductors," Phys. Rev. 32, 97-109 (1928).
Laser communication tran smitter and receiver design 233
216 . H. Nyquist, "Thermal agitation of electric charge in conductors," Phys. Rev. 32 , 110-113
(1928) .
217. B. M. Oliver, "Thermal and quantum noise," presented at Proc . IEEE, 1965.
218. A. Yariv, Optical Electronics in Modem Communications, 5th Ed. (New York: Oxford Un.
Press, 1997).
219 . S.O. Lambert and W.L. Casey, Laser Communications in Space (Boston: Artech House ,
1995).
220. ITU-R Recommendation PI.372-6: Radio Noise, 1994.
221. K. P. Phillips, "An overview of propagation factor s influencing the design of mobile satellite
communication systems," Electron . & Commun. Eng . J., (1997) .
222. K. Rosfjord, J. Yang, E. Dauler, A. Kerman , V. Anant , B. Voronov, O. Gol'tsman, and K.
Berggren, "Nanowire single-photon detector with an integr ated optical cavity and anti-
reflection coating," Opt. Expre ss, 14,527-534 (2006) .
223 . W.R. Leeb, "Degradation of signal to noise ratio in optical free space data link s due to
background illumination," Appl. Opt. 28, 3443-3449 (1989).
224. ED. Teodoro, J.P. Koplow, and S.W. Moore, "Diffraction limited , 3OO-kW peak power pulse s
from a coiled multimode fiber amplifier," Opt. Lett ., (2002) .
225 . V.W.S. Chan , "Space coherent optical communication systems-An introduction," J. Light-
wave Technol. 5, 633-637 (1987) .
226 . S.B. Alexander, E.S. Kintzer, and J.e. Livas, "A Gbps, I Watt Free-space Coherent Optical
Communication System," in LEOS, 1992.
227 . A. Biswas, H. Hemmati, and J.R. Lesh, " High data-rate laser transmitters for free space
laser communications," in Photonics West, 1999.
228 . E. Rochat, R. Dandliker, K. Haroud , R.H. Czichy , U. Roth, D. Costantini, and R. Holzner,
"Fiber Amplifiers for Coherent Space Communication," IEEE ScI. Top. Quantum Electron.,
7,64-81 (2001) .
229. R. Lange and B. Smutny, "BPSK Laser Communication Termin als to be verified in space,"
in Milcom, 2004 , pp. 441-444.
230 . R. Lange and B. Smutny , "Optical inter-satellite links based on homodyne BPSK mod-
ulation : Heritage, status and outlook," in Proc . SPIE, (Free -Space Laser Communication
Technologies XVII) , 2005 .
23 I. T. Shinagawa, "Detailed investigation on reliability of wavelength-monitor-integrated fixed
and tunable DFB laser diode modules," J. Lightwave Technol. 23, 1126-1136 (2005).
232. T. Ikegami and Y. Suematsu, "Resonance-like characteristics of the direct modulation of a
junction laser," presented at Proc. IEEE, 1967.
233. L.A. Coldren and S.W. Corzine, DiodeLasers and Photonic Integrated Circuits (John Wiley
& Sons , Inc., 1995).
234. D.A. Ackerman, "Laser diodes for CATV," in IEEE LEOS , 1995.
235. M.R. Phillips, et. al., "112 channel split-band WDM lightwave CATV system," IEEE Photon .
Technol. Lett. 4, 790-792 (1992) .
236. G.P. Agrawal, Fiber-Optic CommunicationSystems (New York: John Wiley & Sons , 1992).
237 . J. Piprek and J.E. Bowers, "Analog modulation of semiconductor lasers," in RF Photoni c
Technologyin Optical FiberLinks, W.S. e. Ch ang, Ed. (Cambridge Univ. Press, 2002) .
234 David O. Caplan
238. R.S. Vodhanel, A. F. Elrefaie, M. Z. Iqbal, R. E. Wagner, J.L. Gimlett , and S. Tsuji, "Per-
formance of directly modulated DFB lasers in to-Gb/s ASK, FSK, and DPSK lightwave
systems," J. Lightwave Techno!. 8,1379-1386 (1990) .
239. J. Zhang, N. Chi, P. V. Holm-Nielsen, C. Pueucheret, and P. Jeppesen, "An Optical FSK
Transmitter Based on an Integrated DFB Laser-EA Modulator and Its Application in Optical
Labeling ," IEEE Photon . Techno!. Lett. 15,984-986 (2003).
240. F. Koyama and K. Iga, "Frequency chirping in external modulators ," J. Lightwave Techno!.
6, 87-93 (1988).
241. M. Kuznetsov, J. Stone, and L.W. Stulz, "Time- and frequency-re solved measurements of
frequency-modulation and switching of a tunable semiconductor laser," Appl. Phys. Lett.
59,2492-2494 (1991) .
242. M. Kuznetsov and D.O. Caplan, "Time-frequency analysis of optical communication signals
and the effects of second and third order dispersion," in Conference on Lasers and Electro-
Optics (CLEO), 2000.
243. R.A. Linke, "Modulation Induced Transient Chirping in Single Frequency Lasers," IEEE J.
Quantum Electron. QE·2I , 593-597 (1985) .
244. P.A. Morton, G.E. Shtengel, L.D. Tzeng, R.D. Yadvish, T. Tanbun-Ek, and R.A. Logan,
"38.5 km error free transmission at 10 Gbitls in standard fibre using a low chirp, spectrally
filtered, directly modulated 1.55 JLm DFB laser," Electron. Lett. 33, 310-311 (1997).
245. D.R. Hjelme and A. Royset, "RZ versus NRZ in space communication system using direct
current modulated transmitter and optically pre-amplifed receiver with ultra-narrow optical
filter," in LEOS, San Francisco, 1999.
246. M.M. Strasser, P.J. Winzer, M. Pfennigbauer, and W.R. Leeb, "Significance of Chirp-
Parameter for Direct Detection Free-Space Laser Communication," in SPIE, 200 I.
247. P. Corvini and T. Koch, "Computer simulation of high-bit-rate optical fiber transmission
using single-frequency lasers," J. Lightwave Techno!. 5, 1591-1595 (1987).
248. D. Mahgerefteh, P.S. Cho, J. Goldhar, and H. I. Mandelberg , "Penalty -free propagation
over 600 km of nondispersionshifted fiber at 2.5 Gb/s using a directly laser modulated
transmitter," in Conference on Lasers and Electro-Optics (CLEO) , 1999.
249. Y. Matsui, D. Mahgerefteh, X. Zheng, C. Liao, Z.E Fan, K. McCallion, and P. Tayebati,
"Chirp-Managed Directly Modulated Laser (CML)," Photon . Techno!. Lett. 18, 385-386
(2006) .
250. S. Chandrasekbar, C.R. Doerr, L.L. Buhl, Y. Matsui, D. Mahgerefteh, X. Zheng, K. Mc-
Callion, Z. Fan, and P. Tayebati, "Repeaterless Transmission With Negative Penalty Over
285 km at to Gb/s Using a Chirp Managed Laser," Photon. Techno!. Lett. 17,2454-2457
(2005) .
251. S. Chandrasekbar, A.H. Gnauck, G. Raybon, L.L. Buhl, D. Mahgerefteh, X. Zheng, Y.
Matsui, K. McCallion, Z. Fan, and P. Tayebati, "Chirp-Managed Laser and MLSE-RX
Enables Transmission Over 1200 km at 1550 nm in a DWDM Environment in NZDSF
at 10 Gb/s Without Any Optical Dispersion Compensation" Photonics Techno!. Lett. 18,
1560-1562 (2006) .
252. M. Ito and T. Kimura, "Stationary and transient thermal properties of semiconductor laser
diodes ," IEEE J. Quantum Electron . 17,787-795 (1981).
253. H. Shalom , A. Zadok, M. Tur, P.J. Legg, W.D. Cornwell, and I. Andonovic , "On the Various
Time Constants of Wavelength Changes of a DFB Laser Under Direct Modulation ," IEEE
J. Quantum Electron. 34, 1816-1822 (1998).
254. A. Ma, J.C. Cartledge, and H.E. Lassen , "Performance implications of the thermal-induced
frequency drift in fast wavelength switched systems with heterodyne detection," J. Light-
wave Technol. 14, 1090-1096 (1996) .
255. D.O. Caplan, G.S. Kanter, and P. Kumar, "Characterization of dynamic optical nonlinearities
by continuous time-resolved Z-Scan," Opt. Lett, 21, 1342-1344 (1996) .
256. A. A. Saavedra, R. Passy, and J.P. von der Weid, "Thermal drift in wavelength-switching
DFB and DBR lasers," Electron. Lett. 33, 780-781 (1997) .
257. CR. Giles, T. Erdogan, and V. Mizrahi , "Simultaneous wavelength-stabilization of980-nm
pump lasers," IEEE Photonics Technol. Lett. 6, (1994).
258. S. Mohrdiek, T. Plisk, and C. Harder, "Coolerless operation of980 nm pump modules," in
Optical Fiber Conference (OFC), 200 I.
259. J.-L. Archambault and S.G. Grubb, "Fiber Gratings in Lasers and Amplifiers," J. Lightwave
Technol. 15, 1378-1390 (1997) .
260. B. R. Hemenway and M.L. Stevens, "Simultaneous TDMlFDM Using Rapidly-tunable
Transmitters and Receivers For Multi-access Optical Networks:' presented at IEEElLEOS
Integrated Optoelectronics Proceedings, 1994.
261. C.H. Henry, "Theory of the Iinewidth of semiconductor lasers," IEEE J. Quantum Electron.
QE·18, 259-264 (1982) .
262. K. Kojima, K. Kyuma, and T. Nakayama , "Analysis of the spectrallinewidth of distributed
feedback laser diodes," J. Lightwave Technol. 3,1048-1055 (1985).
263. H. Nakano, S. Sasaki, S. Tsuji, N. Chinone , and M. Maeda, "Comparison of optical reflection
tolerance between conventional and ?/4-shifted DFB lasers in a 2.4 Gbit/s system:' Electron .
Lett.4, 1049-1051 (1988) .
264. G. Jacobsen, "Performance of DPSK and CPFSK Systems with Significant Post-Detection
Filtering," J. Lightwave Technol. 11, (1993) .
265. H. Nasu, T. Mukaihara, T. Takagi, M. Oike, T. Nomura , and A. Kasukawa, "25-GHz-spacing
wavelength-monitor integrated DFB laser module for DWDM applications:' Photon . Tech-
nol. Lett. 15,293-295 (2003) .
266. D.O. Caplan, "Multi-channel DPSK Receiver: ' in US Pat. Appl. 11/022,344,2004.
267. D.O. Caplan, "Polarization independent optical interferometers," in US Pat. Appl ., 2004 .
268. D.O. Caplan, "Reconfigurablc Polarization Independent Interferometers and Methods of
Stabilization:' in US Pat. Appl. 11/318,255,2005.
269. J.B. Abshire , et. aI., "The Geoscience Laser Altimeter System (GLAS) for the ICESat
mission," presented at Conference on Lasers and Electro-Optics (CLEO), 2000.
270. M. Albota , et. al. , "Three-dimensional imaging laser radars with Geiger-mode avalanche
photodiode arrays ," MIT Lincoln Lab. J. 13,351-370 (2002).
271. W.T. Roberts, "Cavity-dumped communication laser design," 42-152, Feb. 2003.
272. X. Sun, et. al., "Cloud and aerosol lidar channel design and performance of the Geoscience
Laser Altimeter System on the ICESat mission ," presented at Conference on Lasers and
Electro-Optics (CLEO), 2004.
273. X. Sun, et. al., "Design and performance measurement of the mercury laser altimeter:' in
Conference on Lasers and Electro-Optics (CLEO), 2004.
274. M. Aoki, M. Suzuki, H. Sano, T. Kawano, T. Ido, T. Taniwatari, K. Uomi, and A. Takai,
"InGaAs/InGaAsP MQW electroabsorption modulator integrated with a DFB laser fabri-
cated by band-gap energy control selective area MOCVD:' IEEE J. Quantum Electron . 29,
2088-2096 (1993) .
236 David O. Caplan
275. W.S.c. Chang , "Multiple quantum well electroabsorption modulators for RFphotonic links,"
in RF Photonic Technology in Optical Fiber Links, W.S.c. Chang, Ed. (Cambridge Univ,
Press, 2002) .
276 . G. Raybon, U. Koren, M.G. Young, B.I. Miller, M. Chien , T.H. Wood, and H. M. Presby,
"Low chirp transmission at 5.0 Gbitls using an integrated DBR laser-modulator transmitter,"
Electron . Lett. 30, 1330-1331 (1994) .
277. Y. Akage, K. Kawano , S. Oku , R. Iga, H. Okamoto, Y. Miyamoto, and H. Takeuchi, "Wide
bandwidth of over 50 GHz travelling-wave electrode electroabsorption modulator integrated
DFB lasers," Electron . Lett. 37, 299-300 (2001) .
278 . A.H. Gnauck, S.K. Korotky, J.J. Veselka, J. Nagel, C.T. Kemmerer, W. J. Minford , and D.
T. Moser, "Dispersion penalty reduction using an optical modulator with adjustable chirp,"
IEEE Photon . Technol. Lett. 3, 916-918 (1991) .
279. A.H . Gnauck , S.K. Korotky, and J.E. Zucker, "Tunable chirp , lightwave modulator for
disper sion compensation ," in US Pat. 5,303,079,. USA, 1992.
280. R.M. Jopson and A.H. Gnauck, "Dispersion compensation for optical fiber systems," IEEE
Communi. Mag., 33, 96-102 (1995) .
281. L. Zehnder, "Ein neuer Interferenzrefractor,' Z. Instrkde 11,275-285 (1891).
282. L. Mach, "Uber einer Interferenzrefractor," Z. Instrkde 12,89-93 (1892) .
283. R.C. Alferness, "Waveguide Electrooptic Modulators,"IEEE Trans. Microwave Theory
Technol. ,82, 1121-1137 (1982) .
284 . R.c. Alferness, "Corrections to "Waveguide Electrooptic Modulators"," IEEE Trans. Mi-
crowa ve Theory Technol. 83, 315 (1983) .
285 . R.C. Alferne ss, L.L. Buhl, J.L. Jackel, S. P. Lyman, and V. Ramaswamy, " Fabrication
method for LiNb03 and LiTa03 integrated optics devices," in U.S. Pat. 4,439,265, USA,
1984.
286. L. Thylen, "Integrated optics in LiNb03 : recent developments in devices for telecommuni-
cation," J. Lightwave Technol. 6, 847- 861 (1988).
287. K. Noguchi, O. Mitomi, and H. Miyazawa, "Low-voltage and broadband Ti:LiNb03 mod-
ulators operating in the millimeter wavelength region ," in Optical Fiber Conference (OFC) ,
1996.
288 . K. Noguchi , O. Mitomi , and H. Miyazawa, "Millimeter-wave Ti:LiNb03 optical modula-
tors," J. Lightwave Technol. 16,615-619 (1998) .
289. G.E. Betts, "LiNb03 external modulators and their use in high performance analog links,"
in RF Photonic Technology in Optical Fiber Links, W.S. C. Chang, Ed. (Cambridge Univ.
Press, 2002) .
290 . M.M. Howerton and W.K. Burns, "Broadband traveling wave modulators in LiNb03," in RF
Photon ic Technology in Optical Fiber Links, W.S. C. Chang , Ed. (Cambridge Univ. Press,
2002) .
291. O. Leclerc , et. al., "40 Gbitl s polarization-independent, push-pull InP Mach-Zehnder mod-
ulator for all-optical regeneration," presented at Optical Fiber Conference (OFC), PD35,
1999.
292 . O. Leclerc , B. Dany, D. Rouvillain, P. Brindel , E. Desurvire, C. Duchet , A. Shen, F. Devaux,
E. Coquelin, M. Goix, S. Bouchoule, L. Fleury, and P. Nouchi, "Simultaneously regenerated
4x40 Gbitls dense WDM transmission over 10000 km using single 40 GHz InP Mach-
Zehnder modulator," Electron . Lett. 36, 1574-1575 (2000) .
293. B. Li, et. al. , "SiGe/Si Mach-Zehnder Interferometer Modulator based on the Plasma Dis-
persion Effect ," Appl. Phys. Lett. 74, (1999) .
294. L. Liao, D. Samara-Rubio, M. Morse, A. Liu, and D. Hodge, "H igh speed silicon Mach-
Zehnder modulator," Opt. Express, 13, (2005).
295. G. L. Li and P.K.L. Yu, "Optical intensity modulators for digital and analog applications,"
J. Lightwave Technol. 21, 20 I {}-2030 (2003) .
296. J.P. Sokoloff, P.R. Prucnal, I. Glesk, and M. Kane, "A terahertz optical asymmetric demul-
tiplexer (TOAD)," IEEE Photon . Technol. Lett.5, (1993) .
297. T. Durhuus, C. Joergensen, B. Mikkelsen, and K.E. Stubkjaer, "Penalty free all-optical
wavelength conversion by SOA's in Mach-Zehnder configuration," in ECOC'93. Montreux,
1993.
298. B. Mikkelsen, et. al., "20 Gbitls polarisation insensitive wavelength conversion in semicon-
ductor optical amplifiers," in ECOC'93. Montreux, 1993.
299 . N. Patel, et. al., "40-Gb/s demultiplexing using an ultrafast nonlinear interferometer (UNI),"
300 . N.S. Patel, K. L. Hall, and K.A. Rauschenbach, "Interferometric all optical switches for
ultrafast signal processing," Appl. Opt. 37, 2831-2842 (1998).
301. C. Janz, F. Poingt, F. Pommereau, W. Grieshaber, F. Gaborit, D. Leclerc, I. Guillemot, and
M. Renaud, "All-active Dual Order Mode (DOMO) Mach-Zehnder wavelength converter
for 10 Gb/s operation," Electron. Lett. 35, 1862 (1999) .
302 . B.S. Robinson , SA Hamilton, and E.P. Ippen , "Demultiplexing of 80 Gbitls pulse-position
modulated data with an ultrafast nonlinear interferometer," IEEE Photon . Techno] . Lett. 14,
2002.
303. G. Raybon, Y. Su, J. Leuthold, R.-J. Essiambre, T. Her, C. Joergensen, P. Steinvurzel, and K.
D. K. Feder, "40 Gbitls pseudo-linear transmission over one million kilometers," in Optical
Fiber Conference (OFC): Paper FD I 0-1, 2002.
304. J. Leuthold, G. Raybon, Y. Su, R. Essiambre, S. Cabot, J. Jaques, and M. Kauer, "40 Gbitls
transmis sion and cascaded all-optical wavelength conversion over 1000000 km," Electron.
Lett. 38, 89{}-892 (2002).
305 . Y. Su, G. Raybon, R.-J. Essiambre , and T.-H. Her, "All-optical 2R regeneration of 40-Gb/s
signal impaired by intrachannel four-wave mixing," Photonics Technol. Lett. 15, 35{}-352
(2003) .
306. J. Nayyer and H. Nagata, "Suppression of thermal drifts of high speed Ti:LiNb03 optical
modulators," IEEE Photon . Technol. Lett. 6, 952-955 (1994) .
307. S.K. Korotky and J.1. Veselka, "An RC network analysis of long term Ti:LiNb03 bias
stability," J. Lightwave Technol. 14,2687-2697 (1996) .
308 . A. Waksberg and J. Wood, "An Automatic Optical Bias Control for Laser Modulators," Rev.
Sci. Instrum., 43,1271-1273 (1972).
309. C.T. Mueller and J.G. Coffer, "Temperature-dependent bias drift in proton-exchanged
lithium niobate Mach-Zehnder modulators," in Conference on Lasers and Electro-Optics
(CLEO), 1999.
310 . H. Nagata, "DC drift failure rate estimated on 10 Gb/s x-cut lithium niobate modulators,"
IEEE Photon. Techno!. Lett. 12, 1477-1479 (2000) .
311. H. Nagata , N. Papasavvas, and D.R. Maack, " Bias stability of OC48 x-cut Iithium-niobate
optical modulators: four years of biased aging test results," Photon . Technol . Lett. 15,42-44
(2003) .
238 DavidO. Caplan
312. H. Nagata,G. D. Feke, Y. Li, and W.R. Bosenberg,"DC drift ofZ-cut LiNb03 modulators,"
IEEE Photon. Technol.Lett. 16, 1655-1657 (2004).
313. H. Nagata, Y. Li, D.R. Maack, and W.R. Bosenberg, "Reliability Estimation From Zero-
Failure LiNb0 3 Modulator Bias Drift Data," IEEE Photon. Technol. Lett. 16, 1477-1479
(2004).
314. D.O. Caplan, "A technique for measuring and optimizing modulator extinction ratio," in
Conferenceon Lasers and Electro-Optics (CLEO), 2000.
315. N. Kuwata, H. Nishimoto,T. Horimatsu, and T. Touge,"Automaticbias control circuit for
Mach-Zehndermodulator," presentedat Nat.Meet.Inst. Electon.Comm. Eng., Japan, 1990.
316. A.H. Gnauck and C.R. Giles, "2.5 and 10 Gb/s transmission experimentsusing a 137pho-
tonlbit erbium-fiberpreamplifier receiver," IEEE Photon.Technol.Lett. 4, 80--82 (1992).
317. Q.Jiang and M. Kavehrad, "ASubcarrier-Multiplexed CoherentFSKSystemUsinga Mach-
ZehnderModulatorwithAutomatic Bias Control,"IEEE Photon. Technol. Lett. 5, 941-943
(1993).
318. H. Nagata,Y. Li, K. R. Voisine, and W.R. Bosenberg, "Reliability of Nonhermetic Bias-Free
LiNb03 Modulators," IEEE Photon. Technol.Lett. 16, (2004).
319. L. Fenghai,CJ. Rasmussen,and RJ .S. Pedersen,"Experimental verificationof a newmodel
describingthe influence of incompletesignalextinction ratio on the sensitivitydegradation
due to multiple interferometric crosstalk," IEEE Photon. Technol. Lett. 11, 137(1999).
320. Z. Li,Y. He, B.E Jorgensen, and RJ. Pedersen, "Extinction ratioeffectfor high-speedoptical
fiber transmissions," presentedat Int. Conf. on Comm. Tech. Proc. (ICCT '98), 1998.
321. M. Pauer and PJ . Winzer, "Impact of Extinction Ratio on Return-to-Zero Coding Gain in
Optical Noise Limited Receivers," IEEE Photon. Technol. Lett. 15, 879 - 881 (2003).
322. H. Kim and A.H. Gnauck, "Chirp characteristics of dual-drive mach-zehnder modulator
with a finitede extinctionratio," IEEE PhotonicsTech. Lett., 14, (2002).
323. T. Kawanishi, T. Sakamoto, M. Tsuchiya, and M. Izutsu, "70 dB extinction-ratio LiNb03
optical intensitymodulatorfor two-tonelightwave generation," in OpticalFiberConference
(OFC), 2006.
324. "Measuring extinction ratio of optical transmitters," Hewlett Packard App. Note, 1550-8,
1998.
325. P.O. Andersson and K. Akermark,"Accurate-optical extinction ratio measurement," IEEE
Photon. Technol.Lett. 6, 1356-1358 (1994).
326. C.R. Yang, W.-Y. Hwang, H. Park, H.H. Hong, and S.G. Han, "Off-level sampling method
for bias stabilisationof electro-opticMach-Zehnder modulator," Electron.Lett. 35, 590--591
(1999).
327. R.C.Alferness,S.K. Korotky, and E. Marcatili,"Velocity-matching techniques for integrated
optic traveling wave switch/modulators," IEEE J. Quantum Electron. 20, 301-309 (1984).
328. R.C. Alferness," Travelingwave,electroopticdeviceswith effectivevelocity matching" in
US Pat. 4,448,479, 1984.
329. o. Mitomi, K. Noguchi, and H. Miyazawa, "Broadband and low driving-voltage LiNb03,
optical modulators," lEE Proc. Optoelectron. 145,360--364(1998).
330. R.C. Alferness, "Guided-wavedevices for opticalcommunication," IEEEJ. QuantumElec-
tron. 17,946-959 (1981).
331. N. Henmi, T. Saito, and T. Ishida, "N. Henmi,T. Saito, and T. Ishida,"Prechirp techniqueas
a linear dispersion compensation for ultrahigh-speed long-span intensitymodulation direct
detection optical communication systems," J. Lightwave Technol. 12, 1706-1719 (1994).
332. S.K. Kim, O. Mizuhara, Y.K. Park, L.A. Tzeng , Y.S. Kim, and J. Jeong , "Theoretical and
experimental studies of 10 Gb/s transmission performance using 1.55 lIm LiNb03 - based
transmitters using adjustable extinction ratio and chirp," J. Lightwave Technol. 17, 1320-
1325 (1999) .
333. JJ. Veselka, S.K. Korotky, P. V. Mamyshev, A.H. Gnauck , G. Raybon, and N. M. Froberg,
"A Soliton Transmitter Using a CW Laser and an NRZ Driven Mach-Zehnder Modulator,"
334. N.M. Froberg, A.H. Gnauck, G. Raybon , and U . Veselka, "Method and Apparatus for
Generating Data Encoded Pulses in Return -to-Zero Format," in US Pat. 5,625 ,722, 1997.
335. B. Zhu, L. Leng, A.H. Gnauck, M.O. Pedersen, D. Peckham, L.E. Nelson, S. Stulz, S. Kado,
L. Gruner-Nielsen, R.L. Lingle, Jr., S. Knudsen, J. Leuthold, C. Doerr, S. Chandrasekhar,
G. Baynham, P. Gaarde, Y. Ernori, and S. Namiki, 'Transmission of 3.2 ThIs (80 x 42.7
Gb/s) over 5200 km of UltraWave™ fiber with 100-km dispersion-managed spans using
RZ-DPSK format," in ECOC : Paper PD4.2, 2002.
336. PJ . Winzer, "Optical transmitters, receivers , and noise," in WileyEncyclopedia ofTelecom-
munications, J.G . Proakis, Ed. (New York: Wiley, 2002), pp. 1824-1840.
337. M.L. Dennis, W L Kaechele, WK. Burns, T. F. Carruthers, and LN. Duling, "Photonic
Serial-Parallel Conversion of High-Speed OTDM Data," IEEE Photon . Technol. Lett. 12,
1561-1563 (2000) .
338. R.e. Williamson, J.L. Wasserman, G.E. Betts, and J.e. Twichell, "Sinusoidal Drives for
Optical Time Demultiplexers," IEEE Trans. Microwave Theory and Technol. 49, 1945-
1949 (2001) .
339. V.WS. Chan , "Optical satellite networks," J. Lightwave Technol. 21, 2811-2827 (2003) .
340. MJ.F. Digonnet, Rare Earth Doped FiberLasers and Amplifiers (New York: Marcel Dekker,
1993).
341. CR. Giles and T. Li, "Optical amplifiers transform long-distance lightwave telecommuni-
cation s," Proc. IEEE, 84, 870-883 (1996) .
342. J.-M.P. Delavaux and J.A. Nagel, " Multi-Stage Erbium-Doped Fiber Amplifier Designs," J.
Lightwave Technol. 13, 703-720 (1995) .
343. L. Boivin, M.C. Nuss, J. Shah, D.A.B. Miller, and H.A. Haus, "Receiver sensitivity im-
provement by impulsive coding," Photon . Technol. Lett. 9, 684-686 (1997) .
344. L. Boivin and GJ. Pendock , "Rece iver sensitivity for optically amplified RZ signals with
arbitrary duty cycle," presented at Optic . Amplifiers and their Applications (OAA'99), 1999.
345. PJ . Winzer and A. Kalmar, "Sensitivity enhancem ent of optical receivers by impulsive
coding ," J. Lightwave Technol. 8, 171-177 (1999) .
346. W.R. Leeb, PJ . Winzer, and M. Pauer, "The potential of return-to-zero coding in optically
amplified lasercom systems," in LEOS, 1999, pp. 224-225.
347. M. Pauer, PJ . Winzer, and WR. Leeb, "Booster EDFAs in RZ-coded links: Arc they average-
power limited"," in Proc. SPIE, (Free-Spa ce Laser Communication Technologies XIII), vol.
4272, 118-127, San Jose, USA, 200 I.
348. PJ . Winzer, A. Kalmar, and WR. Leeb, "Role of amplied spontaneous emission in optical
free-space communication links with optical amplication - impact on isolation and data
transmission ; utilization for pointing, acquisition , and tracking," in Proc. SPIE (Free-Space
Laser Communication Technologies XI), vol. 3615, 104-114, San Jose , CA, USA, 1999.
349. DJ. DiGiovanni and C.R. Giles, "Multistage optical amplifiers" in US Pat. 5115338, 1992.
240 David O. Caplan
350. R.I. Laming , M.N. Zervas, and D.N. Payne, "Erbium-Doped Fiber Amplifier with 54 dB
Gain and 3.1 dB Noise Figure," IEEE Photon . Technol. Lett. 4, 1345-1347 (1992) .
351 . T. Pliska, S. Mohrdiek, and C. Harder, "Power stabilisation of uncooled 980 nm pump laser
modules from IOto 100°C," Electron . Lett. 37, 33-34 (2001).
352 . B. Schmidt; S. Pawlik, N. Matuschek, J. Muller, T. Pliska, J. Troger, N. Lichtenstein, A.
Wittmann, S. Mohrdiek, B. Sverdlov, and C. Harder, "980 nm single mode modules yielding
700 mW fiber coupled pump power," in Optical Fiber Conference (OFC), 2002 .
353 . F. Hakimi, E.S. Kintzer, and R.S. Bondurant, "High-power single-polarization EDFA with
wavelength mulitplexed pumps ," in Conference on Lasers and Electro-Optics (CLEO), 1998.
354 . D.O. Caplan and F. Hakimi , "A high-power high-gain single -polarization EDFA," in Con-
ference on Lasers and Electro-Optics (CLEO), 2000 .
355 . C. Lester, A. Bjarklev, T. Rasmussen, and P. G. Dinesen , "Modeling of Yb3+ -Sensitized
Er3+ -Doped Silica Waveguide Amplifiers," J. Lightwave Technol. 13,740-743 (1995).
356 . Z. J. Chen , J.D. Minelly, andY. Gu , "Compact low cost Er3+/Yb 3+ co-doped fibre amplifiers
pumped by 827 nm laser diode," Electron. Lett. 32, 1812-1813 (1996) .
357 . M. Karasek, "Optimum Design of Er3+ _Yb3+ Codoped Fiber s for Large-Signal High-
Pump-Power Applications," IEEE J. Quantum Electron . 33, 1699-1705 (1997).
358 . J. Nilsson , P. Scheer, and B. Jaskorzynska, "Modeling and Optimization of Short Yb3+ _
Sensitized Er3+ -Doped Fiber Amplifiers," Photon. Technol. Lett . 6, 383-385 (1994).
359 . A. Galvanauskas and B. Samson, "High Fiber," in SPIE 's oemagazine, 2004 , pp. 15-17.
360. I. Berishev, A. Komissarov, N. Moshegov, P. Trubenko, L. Wright, A. Berezin, S. Todorov,
and A. Ovtchinnikov, "AIGalnAs/GaAs record high-power conversion efficiency and record
high-brightness coolerless 915-nm multimode pump s," in SPIE, 2005.
361 . V. Gapontsev, I. Berishev, G. Ellis, A. Komissarov, N. Moshegov, O. Raisky, P. Trubenko,
V. Ackermann, E. Shcherbakov, J. Ste ineke, and A. Ovtchinnikov, "High-efficiency 970-nm
multimode pumps " in SPIE , 2005 .
362 . V. Gapontsev, I. Ber ishev, G. Ellis, A. Komis sarov, N. Moshegov, A. Ovtchinnikov, O.
Raisky, P. Trubenko, V. Ackermann, and E. Shcherbakov, "9xx nm single emitter pumps for
multi-kW systems," in SPIE, 2006 .
363 . DJ. Rip in and L. Goldberg, " High efficiency side-coupling of light into optic al fibres using
imbedded v-grooves," Electron . Lett. 31, 2204-2205 (1995) .
364. L. Goldberg, B. Cole, and E. Snitzer, "V-groo ve side-pumped 1.5 Jim fibre amplifier,"
Electron. Lett. 33, 2127-2129 (1997) .
365 . L. Goldberg and J. Koplow, "Compact, side-pumped 25 dBm Er/Yb co-doped double
cladding fibre amplifier," Electron. Lett . 34, 2027-2028 (1998) .
366. DJ. DiGiovanni and A.M . Vengsarkar, "Article comprising a cladding-pumped optical fiber
laser," in US Pat. 5,708 ,669, 1998.
367 . DJ. DiGiovanni and AJ . Stentz, "Tapered fiber bundles for coupling light into and out of
cladding-pumped fiber device s," in US Pat. 5,864 ,644 , 1999.
368 . V.P. Gapontsev and I. Samartsev, "Coupling arrangement between a multi-mode light source
and an optical fiberthrough an intermediate optical fiber length," in US Pat. 5,999 ,673, 1999.
369 . R.P. Espindola, I. Ryazan sky, AJ. Stentz, K. L. Walker, and P.F. Wysocki, "Multi-stage
optical fiber amplifier having high conversion efficiency," in US Pat. 6,104 ,733 , 2000 .
370 . F. Hakimi and H. Hakimi , " New side coupling method for double-clad fiber amplifiers," in
371. L. Goldberg and M. Le Flohic, "Optical fiber amplifiers and lasers and optical pumping
device therefor" in US Pat. 6,608,951, 2003.
372. A.B. Grudinin , D.N. Payne, W. Paul, LJ.A. Nilsson, M.N. Zervas, M. Ibsen, and M.K.
Durkin, " Multi-fibre arrangements for high power fibre lasers and amplifiers" in US Pat.
6,826,335, 2004.
373. Y.Jeong, J. Sahu, D.B.S. Soh, C.A. Codem ark, and J. Nilsson , "High-power, tunable, single-
frequency , single-mode erbium :ytterbium codoped large core fiber master-oscillator power
amplifier source," Opt. Lett. 30, 2997 (2005) .
374. Y. Jeong, J. Nilsson , J. Sahu, D.B.S. Soh, C. Alegria, P. Dupriez, C.A. Codemark, and
D.N. Payne, "Single-frequency, single-mode, plane-polarized ytterbium-doped fiber master
oscillator power amplifier source with 264 W of output power," Opt. Lett. 30, 459-461
(2005) .
375. N.G. Walker and G.R. Walker, "Polarization Control for Coherent Communications," 1.
Lightwave Technol . 8, 438-458 (1990) .
376. F. Heismann and M.S. Whalen, "Fast Automatic Polarization Control System," Photonics
Technol, Lett. 4, 503-505 (1992) .
377. P. Oswald and C.K. Madsen, "Deterministic Analysis of Endless Tuning of Polarization
Controllers," J. Lightwave Technol. 24, 2932-2939 (2006) .
378. Duling, LN. III and R.D. Esman, "Single-polarisation fibre amplifier," Electron . Lett. 28,
1126-1128 (1992) .
379. D.O. Caplan, "Method and apparatus for stabilizing a high-gain, high-power single polar-
ization EDFA," in US Pat. 6,831,779, 2004.
380. F. Hakimi, D.O. Caplan, H. Hakimi, and A.L. Tuffli, "Radiation effects on a two-stage
double-pass single-polarization erbium fiber amplifier," in Conference on Lasers and
Electro-Optics (CLEO) , 2002.
381. K. Morito and S. Tanaka, "Record High Saturation Power (+22 dBm) and Low Noise
Figure (5.7 dB) Polarization-Insensitive SOA Module," in Optical Amplifiers and Their
Applications (OAA) Topical Meeting , paper TuC2, 2005.
382. Z.Jiang, D.E. Leaird, and A.M. Weiner, "Optical Arbitrary Waveform Generation and Char-
acterization Using Spectral Line-by-Line Control ," J. Lightwave Technol. 24, 2487-2494
(2006) .
383. J. Nilsson , Y.Jeong, C. Alegria, R. Selvas , J. Sahu , R. Williams, K. Furusawa, W. Clarkson,
D. Hanna, D. Richardson , T. Monro , D. Payne, K. Yla-Jarkko, S. Alam, and A. Grudinin,
"Beyond I kW with Fiber Lasers and Amplifiers," in Optical Fiber Conference (OFC) , 2003.
384. J. Limpert, A. Liem, H. Zellmer, and A. Tuennermann, "Continuous wave ultrahigh bright-
ness fiber laser systems ," in IEEE Photonics West, 2003 .
385. LT. McKinnie,J.E. Koroshetz , W.S. Pelouch, D.O. Smith , J.R. Unternahrer, S.w. Henderson,
and M. Wright, "Self-imaging waveguide Nd:YAG laser with 58% slope efficiency," in
386. M.D. Mermelstein, A.D. Yablon, and C. Headley, "Suppression of Stimulated Brillouin
Scattering in an Er-Yb Fiber Amplifier Utilizing Temperature-Segmentation," in OAA, Paper
TuD3, Budape st, Hungary, 2005.
387. R.G. Smith and S.D. Personick, "Receiver Design," in Semicondu ctor Devices f or Optical
Communication , H. Kressel, Ed. (New York: Springer-Verlag, 1980).
388. Y. Miyamoto, Y. Hagimoto, and T. Kagawa, "A 10 Gb/s high sensitivity optical receiver
using an InGaAs-InAIAs superlattice APD at 1.3 flm/I .5 flm ," IEEE Photon . Technol , Lett.
3,372-374 (1991) .
242 David O. Caplan
389. T.Y. Yun, M.S. Park, 1.H. Han, I. Watanabe, and K. Makita, "IO-Gigabit-per-Second High-
Sensitivity and Wide-Dynamic-Range APD-HEMT Optic al Receiver," Photonics Technol.
Lett. 8, 1232-1234 (1996) .
390 . T.V. Muoi, "Extremely sensitive direct detection receiver for laser communications," in
391 . A. MacGregor and B. Dion, "39 Photons/bit direct detection receiver at 810 nm, BER= 1E - 6 ,
60 Mb/s , QPPM ," Proc . SPIE (Free-Space Laser Communication Technologies III), 1417,
(1991) .
392 . H. Matsuda, A. Miura, H. Irie, S. Tanaka , K. Ito, S. Fujisaki , T. Toyonaka , H. Takahashi , H.
Chiba, S. Irikura, R. Takeyari, and T. Harada , "High-sensitivity and wide-dynamic-range 10
Gbitls APD/preamplifier optical receiver module ," Electron . Lett. 38, 650-651 (2002) .
393 . 1.R. Lesh, "Power Efficient Communications for Space Applications," in International
Telemetering Conference, 1982.
394 . B.E. Moision and 1. Hamkins, "Deep-Space Optical Communications Downlink Budget:
Modulation and Coding ," IPL IPN Progress Report 42-154 , 2003 .
395 . X. Sun, et. al., "Space-qualified silicon ava1anche-photodiode single-photon-counting mod-
ules," 1. Mod. Optics 51, 1333-1350 (2004) .
396. D.M. Boroson, R.S. Bondurant, and D.V. Murphy, "LDORA:A Novel Laser Communication
Receiver Array Architecture," Proc . SPIE 5338, 16-28 (2004).
397. O.M. Efirnov, L.B . Glebov, L.N. Glebova , K.e. Richardson, and V.1. Srnirnov, "High-
Efficiency Bragg gratings in photothemorefractive glass," Appl . Opt. 38, 619-627 (1999).
398. D.O. Caplan, "Spectral Filtering," in MarsLaser Communication Demonstration (MLCD)
ReceiverFinalReport (MIT Lincoln Laboratory, 2005) .
399 . F. Herzog , K. Kudielka, D. Erni, and W Bachtold, "Optical Phase Lock ing by Local Oscil-
lator Phase Dithering," IEEE 1. Quantum Electron . 42, 973-985 (2006) .
400 . H.P. Yuen and V.W.S. Chan, "Noise in Homodyne and Heterodyne Detection," Opt. Lett . 8,
177-179 (1983) .
401. G.L. Abbas, V.W.S. Chan, and T.K. Yee, "A Dual-Detector Optical Heterodyne Receiver
for Local Oscillator Noise Suppression," IEEE 1. Lightwave Technol. LT-3 (5),1110-1122
(October 1985).
402 . S.B. Alexander, "Design of Wide-Band Optical Heterodyne Balanced Mixer Receivers," 1.
Lightwave Technol , LT-5, 523-537 (1987).
403 . L.G . Kazovsky and O.K. Tonguz, "Sensitivity of Direct-Detection Lightwave Receivers
Using Optical Preamplifiers," Photon. Technol. Lett. 3, 53-55 (1991) .
404 . S.R. Chinn , D.M. Boroson, and I .e. Livas, "Sensitivity of optically preamplified DPSK
receivers with Fabry-Perot filters," 1. Lightwave Technol. 14, (1996) .
405. P.J. Winzer, M. Pfennigbauer, M.M. Strasser, and WR. Leeb, "Optimum filter bandwidths
for optically preamplified NRZ receivers," 1. Lightwave Technol. 19,1263-1273 (2001).
406 . M. Pfennigbauer, M.M. Stras ser, M. Pauer, and P.J. Winzer, "Depe ndence of Optically
Preamplified Receiver Sensitivity on Optical and Electr ical Filter Bandwidths-Measurement
and Simulat ion," Photon . Technol. Lett. 14,831-833 (200 2).
407. L. Y. Lin, M.e. Wu, and T. Itoh, "Figure of merit for high-power, high-speed photodetectors,'
in Optical Fiber Conference (OFC), 1997.
408 . A.M.D. Beling, D. Schmidt, H.-G. Bach, G.G. Mekonnen , R. Ziegler, V. Eisner, M. Stoll-
berg, G. Jacumeit, E. Gottw ald, and 1.-1. Weiske, "High power 1550 nm twin -photodetector
modules with 45 GHz bandwidth based on InP," in Optical Fiber Conference (OFC) , 2002 .
409 . T.S. Rose, D. Gunn , and G.c. Valley, "Gamma and proton radiation effects in erbium-doped
fiber amplifiers: active and passive measurements," J. Lightwave Technol. 19, 1918-1923
(2001) .
410. P.A. Humblet, "Design of optical matched filters," in Globecom '91 , 1991.
411. H.L. Van Trees, Detection, estimation, and modulation theory, Part J (New York: Wiley,
1968).
412 . H. Geiger, M. Ibsen, and R.1. Laming, "Optimum receivers with fiber gratings," in Optical
Fiber Conference (OFC), 1998.
413 . S.R. Chinn, "Error-rate performance of optical amplifiers with Fabry-Perot filters," Electron.
Lett. 31, 756-757 (1995) .
414. R.C. Steele and G.R. Walker, "High-sensitivity FSK signal detection with an erbium-doped
fiber preamplifier and Fabry-Perot etalon demodulation," IEEE Photon . Technol . Lett. 2,
753-755 (1990).
415 . J.D. Berger, F. IIkov, D. King, A. Tselikov, and D. Anthon, "Widely tunable, narrow optical
bandpass Gaussian filter using a silicon microactuator," in Optical Fiber Conference (OFC),
2003.
416. A. D'Errico, R. Proietti, N. Calabretta, L. Giorgi , G. Contestabile, and E. Ciaramella,
"WDM-DPSK Detection by Means of Frequency-Periodic Gaussian Narrow Filtering "
in Optical Fiber Conference (OFC) , 2006.
417. B.E. Little, et. al., "Very high-order microring resonator filters for WDM applications,"
IEEE Photon . Tech. Lett. 16,2263-2265 (2004) .
418. c.K. Madsen and J.H. Zhao, Optical filter design and analysis (New York: John Wiley &
Sons, Inc., 1999).
419. H. Takahashi, K. Oda, H. Toba, and Y. Inoue , 'Transmission characteri stics of arrayed-
waveguide N x N wavelength multiplexer," J. Lightwave Technol. 13,447-455 (1995) .
420 . K. Takada, M. Abe, T. Shibata, and K. Okamoto, "A 25-GHz-Spaced 1080-Channel Tan-
dem Multi/Demultiplexer Covering the So, Co, and L-Bands Using an Arrayed-Waveguide
Grating With Gaussian Passbands as a Primary Filter," Photon . Techno!. Lett. 14,648-650
(2002).
421. 1. Littler, M. Rochette, and B. Eggleton, "Adj ustable bandwidth dispersionless bandpas s
FBG optical filter," Opt. Express 13,3397-3407 (2005).
422 . A. Nosratinia , "Self-characteristic Distributions ," J. Franklin Institute 36, 1219-1224 (1999) .
423 . A. Hasegawa and F. Tappert, "Transmission of stationary nonlinear optical pulses in disper-
sive dielectric fibers. 1.Anomalou s dispersion," Appl. Phys. Lett. 23, 142-144 (1973) .
424. L.F. Mollenauer, "Ultra-long distance soliton transmission: putting fiber nonlinearity to
work," in LEOS, 1993.
425 . J. O'Reilly, J. da Rocha, and K. Schumacher, "Optical Fiber Direct Detect ion Receivers
Optimally Tolerant to Jitter," IEEE Trans . Commun . 34,1141-1147 (1986).
426. J.P. Gordon and H.A. Haus, "Random walk of coherently amplified solitons in optical fiber
transmission ," Opt. Lett. 11,665-667 (1986) .
427. G. Bosco, R. Gaudino, and P. Poggiolini , "An exact analysis of RZ versus NRZ sensitivity
in ASE noise limited optical systems ," in ECOC, 2001.
428. G. Bosco, A. Carena, V. Curri, R. Gaudino, and P. Poggiolini, "On the Use of NRZ, RZ,
and CSRZ Modulation at 40 Gb/s With Narrow DWDM Channel Spacing," J. Lightwave
Techno!. 20,1694-1704 (2002) .
244 David O. Caplan
429 . J.H . Shapiro , "Signal-to-Noise Ratio Analysis for a Preamplified Direct-Detection Receiver
with Pre- and Post-Detection Matched Filters," private communication, 2005.
430. G. Bosco and P. Poggiolini, "The Effect of Receiver Imperfection s on the Performance of
Direct-Detection Optical Systems using DPSK Modulation," in Optical Fiber Conference
(OFC), 2003.
431 . G. Bosco and P. Poggiolini, "The Impact of Receiver Imperfection s on the Performance of
Optical Direct-Detection DPSK," J. Lightwave Technol. 23, 842-848 (2005) .
432 . J. Hsieh, A. Chiayu , V. Chien, X. Liu, A. Gnauck , and X. Wei, "Athermal Demodulator for
42.7-Gb/s DPSK Signals ," in ECOC, 2005.
433 . X. Liu, A.H. Gnauck, X. Wei, J. Hsieh, C. Ai, and V. Chien, "Athermal optical demodulator
for OC-768 DPSK and RZ-DPSK signals ," Photon. Technol. Lett. 17,2610-2612 (2005) .
434. M.L. Stevens, "A High-speed DPSK encoder," private communic ation, 1998.
435. W. Kaiser, T. Wuth, M. Wichers, and W. Rosenkranz, " Reduced complexity optical duobi-
nary 10-Gb/s transmitter setup resulting in an increased transmission distance ," Photon.
Techno!. Lett. 13, 884-886 (200 I).
436 . 1. Kang, C. Xie, C. Dorrer, and A. Gnauck, "Implementations of alternate-polarization
differential-phase-shift -keying transmis sion," Electron . Lett. 40, 333-335 (2004).
437 . C. Schramm, H.-G. Bach, A.M.D. Beling, G. Jacumeit, S. Ferber, R. Ludwig, R. Ziegler, G.
G. Mekonnen , R. Kunkel, D. Schmidt , W. Schlaak, and G. Unterborsch, "High-bandwidth
balanced photoreceiver suitable for40-gb/s RZ-DPSK modulation formats," IEEE Sel. Top-
ics Quantum Electron . 11,127-134 (2005) .
438. E.A. Swanson, J.C. Livas, and R.S. Bondurant, " High sensitivity optically preamplified
direct detection DPSK receiver with active delay-line stabilization," IEEE Photon. Technol.
Lett. 6, 263-265 (1994).
439. D.G. Heflinger , J.S. Bauch, and T. E. Humes, "Apparatus and method for tuning an optical
interferometer," in US Pat. 6,396,605 , 2002.
440 . F. Seguin and F. Gonthier, "Tuneable AII-Fiber@ Delay-Line Interferometer for DPSK
Demodulation," in Optical Fiber Conference (OFC) , 2005.
441. D.O. Caplan, "Polarization Independent Interferometer designs," MIT Lincoln Laboratory,
private correspondence, 1999.
442. D.A. Rockwell, D. H. Matsuoka, and C. L. Schulz, "Differential Phase Shift Keyed Demod-
ulator System," in US Pat. 6,834,146 B2, 2004.
443 . M.L. Stevens, S. Constantine, and D.O. Caplan, "Measured and calculated DPSK SNR
change due to frequency offset error at 40 Gbit/s,' private communication, 2006.
444 . PJ . Winzer, F. Fidler, MJ. Matthews , L.E. Nelson, HJ. Thiele, J.H. Sinsky , S. Chan-
drasekbar, M. Winter, D.M. Castagnozzi, L.w. Stulz, and L.L. Buhl, " 1O-Gb/s Upgrade
of Bidirectional CWDM Systems Using Electronic Equalization and FEC," J. Lightwave
Technol. 23, 203-210 (2005) .
445. S.K. Nielsen, B.F. Skipper, and J.P. Vailladsen, " Universal AFC for use in optical DPSK
systems," Electron. Lett. 29,1445-1446 (1993) .
446 . K. Kudielka and W. Klaus, "Optical homodyne PSK receiver: Phase synchroni zation by
maximi zing base-band signal power," in LEOS, 1999.
447 . A.H. Gnauck, G. Raybon , S. Chandrasekhar,J. Leuthold, C. Doerr, L. Stulz, and E. Burrows,
"25 x 40-Gb/s copolarized DPSK transmission over 12 x 100-km NZDF with 50-GHz
channel spacing," IEEE Photon . Technol. Lett. 15, 467-469 (2003).
448 . H. Bissessur, G . Charlet, E. Gohin , e. Simonneau, L. Pierre, and W. Idler, " 1.6 Thit/s (40 x
40 Gbit/s) DPSK transmission over 3 x 100 km of TeraLight fibre with direct detection,"
Electron . Lett. 39, 192-193 (2003) .
449 . T. Mizuno, M. Oguma, T. Kitoh, Y. Inoue , and T. Takahashi, "Mach-Zehnder Interferometer
Exactly Aligned With ITU Grid Frequencies," Photon . Technol. Lett. 18, 325-327 (2006) .
450 . T. Hoshida and H. Onaka, "Method and system for demultiplexing non-intensity modulated
wavelength division multiplexed (WDM) signals ," in U.S. Pat. 7,035,543,2006 (filed 200 I).
451. PJ . Winzer, "Optical Receiver for Wavelength-Division-Multiplexed Signals," in US Pat.
Appl. 2004/0258423 AI , 2004.
452. B. Zhu, L. E. Nelson, S. Stultz , A.H. Gnauck, e. Doerr, J. Leuthold, L. Gruner-Nielsen, M.O.
Pedersen, and R.L. Lingle , Jr., "High Spectral Density Long-Haul 40-Gb/s Transmission
Using CSRZ-DPSK Format," J. Lightwave Technol. 22, 208-214 (2004) .
453. D.A. Atlas and L.G. Kazovsky, "An optical PSK homodyne transmis sion experiment using
1320 nm diode-pumped Nd:YAG lasers," IEEE Photon . Technol. Lett. 2, 367-370 (1990).
454 . D.M. Castagnozzi, J.e. Livas, E.A. Bucher, L.L. Jeromin, and J.W. Miller , "Performance
of a I Gbit/s optically preampl ified communication system with error correcting coding,"
Electron . Lett. 30, 65-66 (1994).
455 . D.O. Caplan , S. Constantine, and M.L. Stevens, "Near-quantum-limited OOK and binary-
FSK receiver sensitivity using a Filtered Direct -Drive pulse-carved MOPA transmiter and
an optically preamplified receiver at 1.25 Gbit/s," private communication, 2006.
456 . R.A. Linke, et. al., "Coherent lightwave transmission over 150km fiber length s at 400 Mbit/s
and I Gbit/s data rates using phase modulation," Electron. Lett. 22, 30-31 (1985) .
457. T. Imai , T. N. Ohkawa, Y. Ichihashi, T. Sugie , and T. Ito, "Over 300 km CPFSK transmission
experiment using 67 photon/bit sensitivity receiver at 2.5 Gbit/s,' Electron . Lett. 26, 357-358
(1990) .
458 . J.M. Kahn, A.H. Gnauck, U . Veselka, S.K. Korotky, and B.L. Kasper, "4-Gb/s PSK homo-
dyne transmission system using phase -locked semiconductor lasers," IEEE Photon . Technol.
Lett. 2, 285-287 (1990).
459 . T. Chikama, et. al., "Modulation and demodulation techniques in optical heterodyne PSK
transmissi on systems," 1. Lightwave Technol. 8, 309-321 (1990) .
460 . I. Hardcastle, T. Large, F. Davis, and A. Hadjifotiou, "High performance 140 Mbit/s FSK
coherent system," Electron . Lett. 26, 1523-1525 (1990) .
461. T. Naito, T. Chi kama, and G. Ishikawa, "Optimum system parameters for multigigabit
CPFSK optical heterodyne detection systems," J. Lightwave Technol. 12, 1835-1841 (1994) .
462 . S. Norimatsu, H. Mawatari, Y. Yoshikuni, O. Ishida, and K. Iwashita, "IO Gbit/s optical
BPSK homodyne detection experiment with solitary DFB laser diodes," Electron. Lett. 31,
125-127 (1995).
463 . R.I. Laming, A.H. Gnauck , e.R. Giles , M.N. Zerva s, and D.N. Payne, " High-sensitivity
two-stage erbium-doped fiber preamplifier at IO Gb/s ,' IEEE Photon . Technol. Lett. 4,
1348-1350 (1992) .
464 . E. Meissner, "116 photons/bit in a 565 Mbit/s optical DPSK heterodyne transmission ex-
periment," Electron . Lett. 25, 281-282 (1989) .
465 . TJ. Paul, E.A. Swanson , J.e. Livas, R.S. Bondurant, and RJ . Magliocco, "3 Gbit/s optically
preamplified direct detection DPSK receiver with 116 photon/bit sensitivity," Electron. Lett.
29,614-615 (1993) .
246 David O. Caplan
466. Y.K. Park, J.-M. P. Delavaux, Mizuhara, L.D. Tzend, T.V. Nguyen, M.L. Kao, P.D. Yeates,
S.W. Granlund, and S.1., "5 Gbitls Optical Preamplifier Receiver with 135 Photonslbit
Usable Receiver Sensitivity," in Optical Fiber Conference (OFC), paper TuD4, 1993.
467. T. Saito, Y. Sunohara, K. Fukagai, S. Ishikawa, N. Henmi, S. Fujita, and Y. Aoki, "High
Receiver Sensitivity at 10 Gb/s Using an Er-Doped Fiber Preamplifier Pumped with a 0.98
J.Lm Laser Diode " IEEE Photon . Technol. Lett. 3, 551-553 (1991) .
468 . P.P. Smyth , R. Wyatt, A. Fidler, P. Eardley, A. Sayles, and S. Craig-Ryan, "152 photon s per
bit detection at 622 Mbitls to 2.5 Gbitls using an erbium fibre preamplifier," Electron . Lett.
26,1604-1605 (1990).
469 . L.D. Tzeng, R.E. Frahm, and W. Asous, "A high-performance optical receiver for 622 Mb/s
direct-detection systems," Photon . Technol. Lett. 2, 759-761 (1990) .
470 . K. Karman, et. al., "High-sensitivity receiver optical preamplifiers ," IEEE Photon . Technol,
Lett. 4,272-275 (1992) .
471. A.H. Gnauck , K.e. Reichmann , J.M. Kahn, S.K. Korotky, U . Veselka, and T.L. Koch, "4-
Gb/s heterodyne transmission experiments using ASK, FSK and DPSK modulation," IEEE
Photon . Technol. Lett. 2, 908-910 (1990) .
472 . T. Kataoka, Y. Miyamoto, K. Hagimoto , and K. Noguchi , "20 Gbitls long distance trans-
mission using a 270 photonlbit optic al preamplifier receiver," Electron . Lett. 30, 716-716
(1994) .
473 . M. Shikada, et. al., " 1.5 m high bit rate long span transmi ssion experiments employing a
high power DFB-DC-PBH laser diode," in European Conference on OpticalCommunication
(Istituto Internazionale delle Communicazioni : Genva, 1985).
474. J.e. Campbell, et. al., "High performance avalanche photodiode with separate absorption
grading and multiplication regions," Electron. Lett. 19,818-820 (1983) .
475 . K. Hagimoto, et. al., "Twenty-Gbit/s signal transmission using a simple high-sensitivity
optical receiver;' in Optical Fiber Conference (OFC), 1992, paper Tu13.
476 . V. Vilnrotter, C.-W. Lau, M. Srinivasan, K. Andrews, and R. Mukai , "Optical Array Receiver
for Communication Through Atmospheric Turbulence," J. Lightwave Technol. 23, 1664-
1675 (2005) ..
477 . D.O. Caplan and U . Carney, "Angstrom class narrow band filters at 1060-1080 nm," MIT
Lincoln Laboratory, internal memorandum, 2003.
478 . D.O. Caplan, "Spectral Filtering " in Mars Laser Communication Demonstration(MLCD)
PreliminaryDesign Review (PDR) (MIT Lincoln Laboratory and NASA, 2005) .
479. D.O. Caplan and B.S. Robinson , "WDM Mitigation of Nonlinear Impa irments in Low-
Duty-Cycle M-PPM Free-Space Optical Transmitters," submitted to OFC2008 .
© 2004SpringerScience + BusinessMediaInc.
001 : I0.1007/s10297-005-()()33-5
Originally published in J. Opt. Fiber. Commun. Rep. 1,355-379 (2004)
Free-space laser communications with adaptive optics:

Atmospheric compensation experiments
Thomas Weyrauch' and Mikhail A. vorontsov'v'
1 IntelligentOptics Laboratory, Institutefor Systems Research,University of Maryland,A.V.

Williams Bldg., Mail Stop 1103,College Park, Maryland20742
Email: weyrauch@umd.edu
2 Computational and Information Sciences Directorate, U.S. Army ResearchLaboratory, 2800
PowderMill Road,Adelphi,Maryland 20783
Email: mvorontsov@arl.army.mil
Abstract
Refractive index inhomogeneities of the turbulent air cause wave-front distortions of

optical waves propagating through the atmosphere, leading to such effects as beam
spreading, beam wander, and intensity fluctuations (scintillations) . These distortions
are responsible for severe signal fading in free-space optical communications systems
and therefore compromise link reliability. Wave-front distortions can be mitigated, in
principle, with adaptive optics, i.e., real-time wave-front control, reducing the likeliness
of signal fading. However, adaptive optics technology, currently primarily used in
astronomical imaging, needs to be adapted to the requirements of free-space optical
communication systems and their specific challenges.
In this chapter we discuss a non-conventional adaptive optics approach that has
certain advantages with respect to its incorporation into free-space optical commu-
nication terminals. The technique does not require wave-front measurements, which
are difficult under the strong scintillation conditions typical for communication sce-
narios, but is based on the direct optimization of a performance quality metric, e.g.,
the communication signal strength, with a stochastic parallel gradient descent (SPGD)
algorithm .
We describe an experimental adaptive optics system that consists of a beam-
steering and a higher-resolution wave-front correction unit with a I 32-actuator MEMS
piston-type deformable mirror controlled by a VLSI system implementing the SPGD
algorithm. The system optimizes the optical signal that could be coupled into a single-
mode fiber after propagating along a 2.3-km near-horizontal atmospheric path. We
248 ThomasWeyrauch and Mikhail A. Vorontsov
investigate characteristics of the performance metric under different atmospheric con-

ditions and evaluate the effect of the adaptive system. Experiments performed un-
der strong scintillation conditions with beam-steering only as well as with higher-
resolution wave-front control demonstrate the mitigation of wave-front distortions and
the reduction of signal fading.
1. Introduction
Although the term "free-space optical wave propagation" in its strict sense is associated
with propagation in vacuum, in most applications of the so-called "free-space optical
(FSO) communication" technology wave propagation occurs in the atmosphere, which
constitutes a media with random optical inhomogeneities caused by turbulence . The
wave-front distortions arising from these inhomogeneities may severely impact the
performance of FSO communication systems resulting in communication link deteri-
oration, i.e., an increase of the number of errors in the received signal. The bit error rate
(BER) - the major characteristic of communication system performance - depends on
both short-term errors resulting from electronic-circuit related noise and the turbulence -
induced long-term random breaks (up to tens of milliseconds) in communication data
traffic known as atmospheric signal fading. The short-term data losses can be recov-
ered (at least partially) using various data coding techniques developed for wire and
fiber-based communication systems [I]. In contrast, atmospheric-turbulence-induced
deep signal fading represents a unique and significantly more challenging problem,
which cannot be solved using conventional data coding techniques without sacrificing
communication system efficiency, i.e., data throughput rate.
The major incentive for incorporation of adaptive optics (AO) technology into
atmospheric laser communication systems is the active prevention of long-term data
loss rather than data recovery. This could be accomplished by introducing aberrations
to the communication system's optical antenna that compensate or at least mitigate
atmospheric-turbulence-induced wave-front phase distortions . The challenge is the
real-time control of the active/adaptive optical elements (wave-front correctors) incor-
porated in the receiver or transmitter optics in order to make communication signal
fading less likely. Atmospheric turbulence mitigation with adaptive optics technique
will hopefully result in a communication system performance comparable to that of
true free-space propagation. In this sense, the term "free-space laser communication"
being applied to atmospheric systems represents an advanced (optimistic) view of
future developments in atmospheric laser communication technology.
In order to achieve this goal, adaptive optics has to be adapted for the specific re-
quirements of free-space laser communications. AO technology was developed initially
for atmospheric compensation in astronomical observatory sites, whose location are
selected specifically in view of low turbulence-induced distortions . Novel approaches
for AO may be more successful in meeting the challenges of free-space laser communi -
cation systems in near-ground scenarios characterized by stronger turbulence effects.
This chapter is devoted to the experimental evaluation of such a non-conventional adap-
tive optics technique and is organized as follows. In Sec. 2 we discuss architectures
for the incorporation of adaptive optics in FSO communication systems from a more
general point of view. In Sec. 3 an experimental setup for evaluation of adaptive optics
performance for FSO communications is described and results of the characterization
2 Adaptive Optics Architectures for Free-Space Optical Communication Systems 249
(a)
Optical . . .
signal
(b)
Low-pass
filter
Optical . . . . ~_-..-Co m m .
signal signal
Fig. 1. Schematics of adaptive optics approache s for FSO communication. (a) Conventional
adaptive optics uses a wave-front sensor (WFS) and wave-front reconstruction for control of the
wave-front corrector. (b) Wave-front distortion compensation based on blind optim ization of a
system performance metric may use the received signal strength (determined from the commu-
nication signal after low-pass filtering) for feedback
of the atmospheric propagation path used in the experiments are presented. In Sec.
4 low-order wave-front control with a beam-steering system is analyzed. The high-
order wave-front control system as well as results from atmospheric compensation
experiments are described in Sec. 5.
2. Adaptive Optics Architectures for Free-Space Optical Communication

Systems
The conventional adaptive optics approach for atmospheric compensation is based on

wave-front sensing and reconstruction [2,3]. In order to apply this technique for FSO
communication systems, a part of the received beam has to be directed to a wave-front
sensor. The wave-front is reconstructed from the measured data and used to calculate
the control signals for the actuators of the wave-front corrector (Fig. Ia). While con-
ventional AO has been successfully implemented in a number of (mostly astronomical)
systems, in these applications atmospheric turbulence induced wave-front distortions
occur in the near field of the (receiver) telescope's aperture, and intensity fluctuations
(scintillations) can be considered as relatively weak [2,3]. The situation is different
for laser communication applications, where distributed turbulence is affecting the
beam along the entire propagation path [4,5]. Wave diffraction from continuously
distributed refractive index inhomogeneities results in corruption of both phase and
intensity. Strong intensity scintillations in the receiver aperture make wave-front mea-
surements difficult, if not impossible, mostly because of the occurrence of branch
points in the optical field's phase that represents a challenge for phase reconstruction
techniques [6,7] .
In order to avoid wave-front measurements that are undesired in strong scintillation

conditions, control of the wave-front correctors in AO systems can be performed by
blind (or model-free) optimization of a system performance metric, e.g., the Strehl
ratio or the received signal strength (Fig. lb) . Although this concept of wave-front
control without wave-front measurements has been known since the early stages of
AO technology development [8-10], it was largely disregarded because estimation
techniques for the gradient of the metric (cost function) in respect to the control signals
imposed limitations for the control bandwidth [11,12] .
This situation has changed, mainly because of the development of several novel
technologies: new efficient control algorithms such as stochastic parallel gradient de-
scent (SPGO) [13-15], their implementation using parallel processing hardware based
on VLSI micro-electronics [16], and the emergence of high-bandwidth wave-front
phase controllers, e.g., deformable mirrors based on micro-electromechanical systems
(MEMS) [17]. For free-space optical communication there can be additional advan-
tage of using wavefront control by optimization of a performance metric. Information
about the received power level is often available in communication terminals (e.g. it
can be obtained by low-pass filtering of the communication signal) and may be used
as metric for the AO controller.
Adaptive optics based on model-free optimization can be used in different ways
in FSO communication terminals . In general, one can distinguish between adaptive
receiver, adaptive transmitter and adaptive transceiver architectures. The most straight-
forward approach is the adaptive receiver (Fig. 2a). Here, the distortions of the received
beam are compensated by the adaptive optics (AO) allowing for a better focusing of
the optical signal onto a small receiver area. The received signal strength is determined
by the receiver sensor (Rx) and provided as metric signal, J, to the controller, which
evaluates this feedback signal and applies the control signals, {Uj}, to the adaptive
optics.
Compensating only the lowest-order aberrations, i.e., wave-front tip and tilt, which
stabilizes the beam centroid location in the focal plane, may be sufficient if the beam
is focused directly on a detector with a large enough photosensitive area. However,
many FSO communication scenarios for enhanced signal throughput require the cou-
pling of the received beam into an optical fiber (often a single-mode fiber) and use
of wavelength-division multiplexed (WOM) beams and standard fiber-optic WOM
technology for combining or separating different wavelengths . Since wave-front aber-
rations beyond tip and tilt errors sincerely impact the fiber-coupling efficiency, high-
resolution phase compensation is required in addition to beam steering in order to
avoid (mitigate) signal fading.
The potential of the adaptive receiver configuration for FSO communication sys-
tems is limited, because it allows only to compensate distortion s of light waves that
already entered the receiver aperture. In order to maximize the irradiance at a re-
ceiver aperture, an adaptive transmitter system at the other end of the propagation
path is required, which mitigates atmospheric turbulence-induced beam spreading by
pre-compensation of the transmitted wave.
An adaptive transmitter may be realized in different ways. The implementation
shown in Fig. 2b is specifically designed for model-free-optimization AO [18]. The
adaptive transmitter uses feedback from its FSO communication counterpart, i.e., the
signal strength, J2, received by the remote receiver (RX2). The disadvantage of this
approach is the need for a means to transmit metric data from RX2 to the adaptive trans-
2 Adaptive Optics Architectures for Free-Space Optical Communication Systems 251
(b)
Comm .
t
R X2
• Optical• • signal
signal
Tx
__
~~_7-_.........J --
.. .J.z
Comm .
signa l
Tx
(d) Tx
Comm.
signal
Fig. 2. Simplified schematics of SPGD adaptive optics architectures in FSO communication. (a)
An adaptive receiver improves focusing the received beam on the receiver (Rx) using the received
signal strength (J) as feedback signal and reduces thus the risk of data loss in the communication
signal (dashed lines indicate waves without adaptive optics). (b) An adaptive transmitter reduces
turbulence-induced beam spreading of the beam originating from the transmitter (Tx) . It uses the
signal strength received by the remote receiver (RX2) as feedback signal (h) . (c) Received and
transmitted optical signals share the same adaptive optical elements in an adaptive transceiver.
Feedback is provided from the local receiver (Rx) and/or the remote transceiver (TxRx2). (d)
The use of fiber optics components reduces the complexity of the free-space optical setup in an
adaptive transceiver.
mitter's AO controller. Metric data could be sent multiplexed (in time or wavelength
division) with other information over the optical transmission channel or transmitted
by a radio frequency signal [19]. The feedback signal's inherent latency due to the
propagation time constitutes a basic limitation for this approach, and very long dis-
tances between transmitter and receiver are not feasible with a reasonable update rate
for the control voltages.
The second approach for realization of an adaptive transmitter with model-free
optimization strategy uses the beam from a beacon at the remote FSO communication
terminal to obtain information about the wave-front distortions during propagation
between the terminals . Control of the adaptive optics is performed by maximization
of the beam quality (e.g., the Strehl ratio) of the received light from the beacon, and
the transmitted laser beam is sent in the opposite direction through the same adaptive
optics elements . This control scheme, similar to a conventional AO transmitter, relies
on the assumption that wave-front correction for the received beam is simultaneously
the optimal pre-compensation for the outgoing beam. This may not necessarily be
the case considering the effects of long, horizontal atmospheric propagation and the
related scintillation effects typical for many laser communication applications .
In FSO communication systems, the optical signal from the remote terminal
may serve as beacon, so that the configuration corresponds actually to an adap-
tive transceiver that uses the received signal strength, J, as performance metric and
where received and transmitted optical communication signals (beams) are counter-
propagating through the same adaptive optical system (Fig. 2c). Feedback for the
adaptive optics controller could be provided by the signal strength (metric J) from the
local receiver, from the remote receiver (metric J2), or any combination thereof (with
the need for experimental determination of the rule for the optimal combination of the
performance metrics [19]).
Mutual misalignment of transmitter and receiver is a particular problem related
with the adaptive transceivers setup shown in Fig. 2c. It can be avoided if the FSO
communication terminals link the counter-propagating optical signals directly from
fiber to fiber. Separation of the counter-propagating received and transmitted waves
can then be performed in the fiber domain, e.g., using optical circulators or WDM
technology. A schematic of the corresponding setup is shown in Fig. 2d.
3. Experimental System Arrangement and Components
Experimental evaluation of the performance of an adaptive optics system based on

blind optimization of received signal strength was performed on a 2.3-km atmospheric
propagation path that linked free-space optical communication transceivers located on
the top of a 73-m -high water tower and an AO transceiver system in the laboratory.
A detailed description of the propagation path as well as atmospheric characterization
capabilities of the used atmospheric laser optics testbed can be found in [20].
A schematic of the wave propagation geometry is shown in Fig. 3. The first laser
communication transceiver (T1 ) on the tower uses a 15-mW laser diode at 785 nm and
a 5-cm-diameter collimating lens. A diffuser placed between the laser diode and the
collimating lens provides a beam divergence of about 2 mrad. The second transmitter
(T2) is collimating light emerging from a multi-mode fiber with 2oo)lm core diameter
using a 60 mm-diameter lens. The light from a IOO-mW laser diode (808 nm) is coupled
into the other end of the 100-m-Iong fiber. The laser beam divergence is 2.8 mrad that
results in a 6.5-m beam footprint at the receiver plane.
At the laboratory site, a 45° mirror mounted on the laboratory's roof directs the
light from the transceivers toward an optical table, where the adaptive optics system
is installed. A telescope (F/8 refractor, focal length F, = 1.2 m) was used to reduce
the received beam diameter from 15em at the pupil P to 12.5 mm. For mitigation of
atmospheric turbulence effects, we used an AO system comprising two subsystems
with three active mirrors (depicted as transmissive elements in Fig. 3). The beam
tracking system uses the slow tracking mirror, TM, to correct for large-scale pointing
errors induced, e.g., by thermal expansion of the laboratory building and the water
tower. The fast beam steering mirror, FSM, compensates tip and tilt errors induced
by atmospheric turbulence and jitter from building vibrations. The MEMS piston-
type deformable mirror, pDM, in the high-resolution adaptive optics ()lAO) system
mitigates the higher-order wave-front phase distortions from atmospheric turbulence.
3 Experimental System Arrangemen t and Components 253
E
::?
l
Performance
\---.:--i. metric
sensor
Fig. 3. Schematic of the experimental setup with lasercom transcei ver module s (T l, T2) on the
water tower, the transceiver telescope with D p = 15 cm apertur e, and the beam steering system
with the tracking mirror TM and the fast steering mirror FSM. Higher-order wave-front distor-
tions were co mpe nsated with the MEMS de formable mirror ILDM, driven by the SPGD controller.
(The active mirror s of the AO subsyste ms are depi cted as transmissive elements to simplify the
schematic.) The fast digital camera FDC was used for recording irradiance fluctuations (scintil-
lations) in the image plane -pr of the entrance pup il
The tracking mirror is placed in an image plane of the transceivers telescope 's pupil P .
With subsequent re-imaging onto the fast beam steering mirror and the micro-machined
mirror by two optical relays, all active mirrors are in planes conjugate to the pupil P .
To evaluate the irradiance fluctuation s (scintillations) for waves from the beacon
T2, we imaged the pupil plane of the receiver telescope using a fast framing camera and
recorded the temporal evolution of the irradiance distribution at about SOO frames/s o
Then the probability distribution of irradiance values, the scintillati on index as well as
the irradiance fluctuation s for individual pixels and their temporal spectra were calcu-
lated from the recorded data (929 frame s). Figure 4a presents the probability density
funct ion, p(I), for inten sity scintillations for three different turbulenc e strengths cor-
respondin g to Rytov variances, a"k, of 0.2, I, and 10, respectively. The a"k
values were
calculated from the path-averaged refra ctive index structure con stant, C~, measured
durin g the experiments, according to a"k = 1.23C~e /6 £ 11 / 6 (k = wave number, L
=propagation distance). These values cover the range from relatively weak turbulence
(a"k = 0. 2) to the scintillation saturation regime [21] (a"k = 10). The irradiance
values, I , were normal ized by their respective average, (1), in order to allow for an
easier comparison of the curves. The insets show single frames, I (r), for each cond i-
tion (r = (x , y) is a pixel loca tion vector in the pupil plane). The scintillation indices,
254 Thomas Weyrauch and Mikhail A. Vorontsov
l.4 .fFn";r"r-'-------;-TI~
p
1.2
1.0
0.8
0.6
0.4
0.2
0.0.J:==~~=~=___..........-~.--,.j
0.1 1/< / > 10
~00~
(J~ =0'2 b 0.06 e
~ <~>
o ~~
150 0 200 400 600 800 1000
::~ ::Lr- ~..,._-_-~~!! I! -i i l.l 10

250 0 200 400 600 800 1000 w,
/ (J~ = 10 d
200 2
150
100
50
o+-'--=.''-'''';-:-:-'--=-T'''-'---'-'-r-'-........--'-I.I-'---'-I.'''--i 10 100 ( HZ)

o 200 400 600 800 t (ms )l000
Fig. 4. Irradiance distributions in the receiver's pupil plane (P): (a) Probability density function
p(I) of measured irradiance values I for three different atmospheric conditions (C~ = 1 X
10- 15 m- 2 / 3 , 5 x 10- 15 m- 2 / 3 , and5 x 10- 14 m- 2 / 3 ), which correspond to Rytov variances
(T~ = 0.2, I, and 10, respectively. Insets : sample short-exposure irradiance distributions for each
condition (labelled by the (T~ value; photo area corresponds to 11 x 11 em? in P). (b-d) Temporal
evolution of the irradiance at a sample pixel for (T~ = 0.2, I, and 10 (b.c, and d, respectively).
(e) Average temporal spectra (S[(I») of the irradiance fluctuations in individual pixels . (f) The
contents spectra W[(f) = f(s[(f») reveal the main spectral contributions on the logarithmic
frequency scale
o-J = (/2) / (/)2 -1, were 0.09, 0.31, and 1.06 for this particularset of measurements.
These values are smaller than the respective Rytov variance values ((J'~ = 0.2, I , and
I0) even for the weak fluctuation regime, which might be explainedby the partial loss
of coherenceof the lightat thetransceiveropticalantenna (T2) after IOO-m propagation
in a multi-mode fiber.
The temporalevolutioncurves, I (t), of the irradiance in an individual samplepixel
are shown in Figs. 4b-d for (J'~ = 0.2, I, and 10,respectively. To obtain the temporal
spectra of the irradiance fluctuations we used Fouriertransformations of the evolution
curve /(t) for all pixel locations r within the receiverpupil and calculatedthe average
spectra (s[(I)); the resultsare shown in Fig. 4e. The curves were normalized by their
4 Compensation of Low-Order Distortions 255
d.c. spectral component SI(f = 0). The main spectral contributions, which can be
identified in the contents spectra, WI(f) = ! (SI(f) ), shown in Fig. 4f, are in the
range of a few Hz to more than 400 Hz (the maximum frequency detectable with the
used camera). The crosswind speed v measured at the laboratory site was less than
I mls for the measurements corresponding to CJk = 0.2 and I, while for CJk = 10
crosswind was considerably stronger with v ~ 6 mis, which contributes to the shift of
the maximum in the contents spectrum, WI(f).
4. Compensation of Low-Order Distortions
Conventional (nonadaptive) laser communication systems typically use a divergent

beam for transmission so that a sufficiently large beam footprint at the receiver end of
the communication link eliminates the need for precision tracking . However, only a
small portion of the light emitted by the transmitter is actually entering the communi-
cation receiver aperture in this case. The available laser power could be utilized much
more efficiently with a narrow beam using active beam steering in a FSO commu-
nication terminal in adaptive transmitter configuration. To evaluate the potential , we
performed atmospheric compensation experiments with low-order (tip/tilt) adaptive
optics compensation, i.e., using only the tracking and fast beam steering subsystem
(BS) of the setup.
4.1. Tracking and FastBeam Steering System
The beam tracking and steering controllers share a single sensor fordriving both mirrors
as shown in Fig. 5 (dashed box). The received wave is reflected sequentially from the
tracking and beam steering mirrors (TM and FSM, respectively), which are optically
interfaced by an optical relay (lenses L 3 and L 4 ) . To provide a feedback signal , the
wave is split after reflection from the fast steering mirror by beam splitter BSl, and
a portion of the beam is focused by lens L 5 onto the 2-dimensional linear position
sensitive detector (PSD) ; its output voltages, Vx and Vy , correspond to the horizontal
and vertical beam-centroid displacement from the sensor's center point, respectively.
The fast steering mirror, FSM, is actuated using a piezoelectric tip/tilt platform; its
angular control range corre sponds to ±50 urad in the pupil plane P . Control of both
x and y actuators of TM and FSM are performed independently; in the following
only the x control channel is described. The error signal e~n) = V~n) - Vx,o, i.e.,
the deviation of the PSD output at iteration n from the desired value Vx,o, is used to
calculate the control signal U~n) using the discretized proportional-integral-derivative
(PID) algorithm
The error integral S~n) is calculated iteratively as S~n) = Qe~n) + (1 _ Q)S~n-l).

The coefficients k p , k), and ko, which determine the magnitude of the proportional ,
integral and derivative terms in (I), respectively, as well as Q were determined by
optimizing the response to a step-wise input signal. To suppress mirror ringing due to
the fast steering mirror's mechanical resonance at about 2.4 kHz, the output voltages
Tracking and fast beam steering (8 S ) subsystem
Fig. 5. Schematics of the AO setup . BS (dashed box) : beam steering subsystem with active
mirrors for slow tracking and fast beam steering . !lAO(dotted box) : higher resolution wave-front
correction subsystem with MEMS mirror and SPOD controller. The pupil plane P is imaged by
the transceiver telescope (L1. L2) onto the tracking mirror (TM). and subsequently by an optical
relay (L a. L4) onto the fast steering mirror (FSM) . Part of the light is coupled out by beam splitter
BS 1 and focused by lens L5 onto the position sensitive detector. PSD. to provide feedback signals
Vx , Vy to the fast steering controller. The tracking controller monitors the FSM control voltages
iU«, U y ) and applies correction signals (U x , Uy ) to TM if required. The optical relay (L6, L7)
connects the subsystems BS and !lAO imaging the pupil onto the MEMS deformable mirror
J-LDM. After reflection from J-LDM and beam splitter BS2, a part of the received light is coupled
into the single-mode fiber SM 1 in focal plane :P. The optical power in the fiber, measured by
the photomultiplier (PM), is the feedback signal J for the SPOD controller. Another part of the
received light is transmitted by BS2 and BSa and focused by onto the camera CCD in focal plane
F'" (far field intensity distribution). For adaptive transmission, the light emerging from SM2 is
collimated by L9 and propagates after reflection from beam splitter BSa in opposite direction
through the AO system's optical train
Ue and Uy were low-pass filtered with I-kHz cutoff frequency. The iteration rate for
the discretized PID controller was up to 12 kHz.
The tracking controller continuously monitors the voltage Ux applied to the fast
steering mirror, FSM, and actuates the tracking mirror,TM, with control signal Ux
when the control signal value U'; is outside of a tolerance range Ux,o ± L1Uc around
the center of the dynamic range, Ux,o. The angular tracking range of TM corresponds
to rv 10 mrad, and the maximum angular speed to rv 50 urad/s in the pupil plane P.
The closed-loop performance of the fast steering system was determined in ex-
periments without atmospheric propagation but laboratory-generated tilt aberrations
with sinusoidal oscillating amplitude . A comparison with open-loop behavior revealed
~
0.4 on
Po
0.3
•
0.2
1+--60 f1rad----+l
0.1 +- Feedbackon
Feedbackoff
-
~-- -- ---- --- ----- - -
-- _.. - - ...
- '"
0.0
-40 -20 o
0.2: o , - n - - - - - - - - - - - - - - - - . . ,
Sa \
,
0.1& \
,
'. Feedback off
,,
0.10- ,,
,
,
,
0.05- "
Feedback on -,••.• ,-~~J .

. .....
O.oo-t-=::=;::;:;;:;:::::::::::::::::;::;:;;;::::~~::::;::::;:;:::;:;;;;:=l
0.1 10 100 f (Hz) 1000
Fig. 6. (a) Probability density function Po for the tilt angle Ox in the receiver pupil plane P
after 2.3 km atmospheric propagation with and without tip/tilt correction (feedback on and off,
respectively). The left inset shows the trajectory of tip/tilt angles acquired during two seconds
under open-loop conditions; the right inset is the corresponding measurement with closed feed-
back loop. (b) Temporal spectra soU) for the tilt angle Ox without tip/tilt correction (feedback
off) and for the residual tilt angle with beam-steering system operation (feedback on)
better than 90% suppression of the tilt aberrationamplitudeat frequencies f S; 60 Hz

and better than 50% for f S; 330 Hz.
4.2. Compensation ofAtmospheric Wave-Front Tilt Distortions
Forevaluation of the fastbeamsteeringsystemperformance in atmosphericconditions,

we compared the statistics of the measured tip and tilt wave-front angles after 2.3 km
propagation with the feedback off and on. Figure 6a shows the probability density
functions Po forthetiltangle (}x forbothcases.Experiments demonstrated the reduction
of the root-mean-square value (}rms of the total deviation angle () = (();, + ()~) 1/2 from
(}r~s = 28 wad (beam steeringsystem off) to (}r°'::s = 2 wad with closed-loopcontrol
(thesevaluesrefer to the angle in the pupil plane P, see Fig. 5). The left insetin Fig. 6a
showsthe trajectoryof the beamcentroid's angularcoordinates«(}x, (}y) during a period
of two seconds while the tip/tiltcontrol was off (the side lengthof the box corresponds
to 60 urad in the pupil plane P). For comparison, the trajectory of «(}x, (}y), recorded
during two seconds with closed tip/tilt control loop, is presented on the same scale in
the right inset of Fig. 6a.
Figure 6b shows the temporal spectra of the wave-front tilt angle for both cases.
Distortions of the tilt angle were significant only below 100Hz; the fast beam steering
system compensated these distortions effectively. The spectrum for the wave-front
tilts in the absence of feedback control also reveals distortions at discrete frequencies
(e.g. at 23 Hz) that were caused by vibrations of the laboratory building. As seen from
the results in Fig. 6b, this jitter was efficiently mitigated by the tip/tilt compensation
system .
4.3. Laser Communication with Tip/Tilt Control
For exploration of the influence of low-order (tip/tilt) adaptive beam steering on the
performance of a laser communication system, we built a communication setup com-
bining nonadaptive transceivers with an adaptive communication transmitter as shown
in Fig. 7a. The bit pattern generator of a bit error rate (HER) performance analyzer
modulated the laser source of the adaptive transmitter A, which transmitted the optical
signal to the communication transceiver B on the water tower. The communication
signal was retransmitted by transceiver B toward the laboratory building and was re-
ceived by the communication receiver C, which was connected to the input of the HER
analyzer to determine the overall HER of the closed-loop communication link.
Transceiver B was also used as the beacon for the adaptive transmitter A. It has
concentric transmitter and receiver apertures (diameters of 5 cm and !Ocm, respec-
tively; see T; in Fig. 3). Note that a diffuser was placed between the laser diode and
the transmitter's collimating lens to achieve a beam divergence of about 2 rnrad. The
optical field from the beacon (dashed arrow in Fig. 7a) entered theAO system telescope
at the laboratory (adaptive transmitter A in Fig. 7a). The power received by the AO
telescope was in the order of 0.5 J.1W.
The adaptive transmitter A used the received optical wave to determine and com-
pensate the wave-front phase tilts introduced by atmospheric turbulence and beam
jitter. A laser beam was transmitted through the compensated beam propagation train
as shown in Fig. 7b. For this the light from a laser diode, LD, was coupled into a single
mode fiber, SMF, and the end of this fiber was placed in a plane conjugate to the beam
positioning sensor, PSD. Thus the transmitted wave-front was pre-compensated by
removing the wave-front tilts that resulted from the beam propagation from the beacon
to the optical receiver system. The transmitter laser diode, LD, was modulated with an
125 Mbit/s bit stream.
The large footprint of the divergent beam received from transceiver B allowed
the placement of the communication receiver C at a distance h ~ 2 m from the
AO transmitter A, thus providing weak correlation of atmospheric phase aberrations
between transmitted and received waves (space diversity) and avoiding backscatter
enhancement effects that accompany counter-propagating waves [22,23] . Note that in
the adaptive communication transmitter system no diffuser was used, and the trans-
mitted power was intentionally reduced to only 70 J.1W in order to obtain a variation
of the atmospheric-turbulence-induced HER that was large enough for a reasonable
measurement duration. The beam footprint at the water tower site was less than 20 ern.
The influence of beam steering on the HER was investigated by sequentially turn-
ing on and off the fast beam steering feedback control in intervals of 60 seconds and
Pattern
generato
...
BER
l xl0- 2
l xl0-3 --- •_.•-

• .I
Beam Steering: 0 on
••
• ~
•
• of! I
•
•
C
l x l 0~
0 • • •
l x10 -5
0 •• •
l x10~ 00
• O .
l x10 - 7 0
r5C)
1x10 -8 0 ~ 0
0 00
0
1x10-9 c:fL-
0 120 240 360 480 t (5) 600
Fig. 7. (a ) Schem atic of the setup for bit error rate (BER) measurements for a closed-loop free-
space optical communication link with a single adaptive transmitter. The bit pattern generator
modulates the light transmitted by the adaptive transmitter A . The transceiver B on the water
tower receives and retransmits the optical bit stream, which is received by the nonadaptive receiver
C and sent to the bit error rate analyzer. The telescopes for A and C are separated by distance
h = 2 m to introduce spatial diversity. (b) Schemat ic of the adaptive transmitter with the tracking
mirror T M and the fast beam steering mirror F S. It uses the light from transceiver B as beacon
for tip/tilt correction. (c) Bit error rates measured during a 600 s period with the beam steering
turned on and off alternatively in intervals of 60 s each. The horizontal lines indicate the average
BER for each 60 s interval
simultaneously monitoring the BER analyzer. Figure 7c shows the rates for bit error s
accumulated during intervals of ten seconds within a total measurement time of ten
minutes. No errors were observed in the cases where no symbol is shown (the smallest
possible non-zero value for the BER is 8 x 10- 10 , which corresponds to a single error
within ten seconds for 125 MBit/s). The horizontal lines show the average values for
the 60-second interval. As can be seen from Fig. 7c, the BER in an interval without
wave-front tilts compen sation was at least three order s of magnitude higher than in the
respective preced ing period with active beam steering. Since both free-space optical
links (the commercial configuration with diffuser [link path A -+ C J and the configu-
ration with the wave-front tip/tilt compen sation [link path B -+ AD contribute to the
260 Thomas Weyrauchand MikhailA. Vorontsov
BER, but compensation was only performed over the propagation path B -+ A, one
can expect an even higher improvement if compensation occurred in both directions.
s. SPGD High-Resolution Wave-Front Control
The high-resolution adaptive system used for the atmospheric compensation experi-
ments is based on analog VLSI controller chips that implement a stochastic parallel
gradient descent (SPGO) algorithm for optimization of a system performance metric
in combination with a PC as supervisory controller and a MEMS deformable mirror
as wave-front corrector. Details about the algorithm and the VLSI architecture can
be found in [16,24]. An analysis of the compatibility of the SPGO approach with
various MEMS deformable mirrors and an experimental evaluation of the efficiency
of SPGO control with a low-resolution AO system were presented in [25]. First ex-
periments for wave-front distortion compensation using an AO system similar to the
one described here were performed using artificially generated turbulence [26]. In the
following, we discuss wavefront control using the SPGO adaptive system operating on
a 2.3 km atmospheric propagation path with respect to free-space laser communication
applications.
5.1. SPGD Adaptive Optics System
A 140-element piston-type micro-machined mirror (pDM) [17,27] is used for the

MEMS adaptive optics (JlAO) system. The mirror is placed in an image plane of the
AO receiver system pupil P (see Fig. 5). An optical relay system (lenses £6 and £7)
reduces the beam size to match the 3.3 mm side length of the MEMS mirror. The
control of the pDM is based on the optimization of a signal proportional to the optical
power received through the single-mode fiber SM 1 (system performance metric J). The
VLSI controller chips generate at each iteration step, n, a set of statistically independent
control parameter perturbations, {oujn)}, in parallel for all N control voltages, ujn)
(j = 1, . . . ,N; N = 132 -only the 132 mirror elements that were within the circular
aperture of the beam were used). The perturbations have all the same absolute values
c: but pseudo-random signs (Bernoulli distribution, ou = ±c:) with a probability of
0.5 for either value. After application of the control voltage perturbations to the pDM,
the system measures a perturbed metric value J~n), then the signs of all perturbation
voltages are inverted and the corresponding value J~n ) is measured, i.e.,
(n) _ J( (n)
J± - u1
± uu
J: (n) (n) ± UU
J: (n) (n) ± uU
J: (n»)
(2)
1, · · ·, u j j " ' " UN N .
The control voltages are then updated according to
ujn+1) = ujn) +/ [J~n) _ J~n)] sign(Oujn») . (3)
The update coefficient, / , was controlled by a supervisory control loop [25] using
/ = /0/ (J + C) in order to allow for faster corrections when the metric value J is
low (the constant C > 0 limits / for very low J values). The response bandwidth of
the MEMS deformable mirror limited the maximum achievable iteration rate, which
was about 7,000 s -1 .
5 SPGDHigh-Resolution Wave-Front Control 261
5.2. Temporal Behavior of the Received Power in the Tip/Tilt-Compensated

Receiver System
Since the power of the received optical signal is used by the SPGD AO system as feed-
back signal J, the study of its characteristics after compensation of tip and tilt errors
under real atmospheric propagation conditions is important for the understanding of
the system's challenges and performance. In experiments, we recorded the temporal
dependence of the metric, J, under different atmospheric conditions and measured
simultaneously the path-averaged refractive index structure constant, C~. As an exam-
ple, Figure 8a shows two different time series of the received power signal J(t), which
were recorded while no feedback control was applied to the mirror pDM. Both beam
tracking and fast steering control systems were operating during the experiments. The
data presented in Fig. 8a were acquired on different days, but the average refractive
index structure constant C~ = 4 X 10- 14 m - 2 / 3 was identical during both experi-
ments. A weather station at the laboratory site recorded no wind during measurement
of curve I while the crosswind speed was v = 3.5 m/s during measurement of curve 2.
Both cases are characterized by strong intensity scintillations (scintillation saturation
regime with the Rytov variance (j~ = 1.23C~e/6 £11 /6 =8 for both cases) . Despite
similar C~ values the normalized variances of metric values (jJ = (J 2)j(J)2_1
were different: (jJ ,1 = 0.92 and (jJ ,2 = 0.57 for curves I and 2, respectively.
The distinct difference in the temporal behavior is also visible in the average
temporal power spectra, (SJ(J)), shown in Fig. 8b. To obtain the (SJ(J)) curves,
we calculated the power spectra, SJ(J), for 1000 recorded time series, J(t), with
about I s duration each by fast Fourier transformation and averaged these spectra .
The average spectra are normalized to their d.c. component, (SJ (0)). Some spectrum
features are better visible in the contents spectra, wJ(J) = !(SJ(J)) , shown in Fig.
8c. Apparent parasitic signals at ! :::::: 0.8 kHz and! :::::: 2.4 kHz were the result of
incomplete suppression of the ringing of the fast steering mirror, FSM. Atmospheric
distortions contribute to the signal mostly in the frequency range < 500 Hz, which is
very similar to the scintillation spectra presented in Figs. 4e,f. The main contributions
to the distortions were below 300 Hz.
An important characteristic of the metric J, i.e., the received power, in respect
to free-space laser communication is the probability density distribution (PDF) for
the metric values, p( J), because the probability of low power values determines the
likelihood of fading and thus the bit error ratio [22, 28]. Hence, we analyzed the
tip/tilt-compensated PDFs of the metric for a broad range of turbulence conditions.
Fig. 9 shows three representative examples corresponding to different metric variances
(jJ = (J2) j (J)2 - 1. In general, the variance of the received power observed in our
experiments was between (jJ = 0.15 and (jJ = 1.1. In Fig. 9, the J values were
normalized by their respective average value, (J), in order to allow for a comparison
of the curve shapes. The curves p( J) for (jJ = 1.03 and 0.48 have apparently a
Gaussian distribution-like shape on the logarithmic J-scale, i.e., they are similar to
a log-normal distribution, while p(J) for (jJ = 0.16 is significantly different. The
curves in Fig. 9 show that even for weak fluctuations (as for (jJ = 0.16) there is a
considerable probability for the occurrence of low metric values, i.e., deep fading in
terms of laser communication, which could be (at least partially) prevented by higher-
resolution adaptive optics .
1.0
J
0.8
0.6
0.4
0.2
0.0
1.0
~I~
J
0.8
0.6
0.4
0.2
0.0
0 200 400 600 800 1000
t (ms)
b
io'
<SJ>
,
-,
,, 2
lcr' 1 ,
lcr'
'--
10 100 1000 (Hz)
3
C
wJ
' .'
'. #"" ~
..-
0.3
10 100 1000 (Hz)
Fig. 8. (a) Two samples of the metric (received power) evolution J(t ) after 2.3 km atmospheric
propagation; C~ values were identical for the measurements butthe wind speed was considerably
higherfor measurement2 (beam steering was on in both cases). (b)The corresponding (averaged)
powerspectra (sJ (J ») and (c) the contents spectra wJ (J) = f (sJ (J ») reveal the difference in
the the main spectral components (f = frequency). Note the frequency shift of the maxima of
W J (J) by a factor of 5
5.3. Wave-Front Control with the SPGD AO System
For evaluation of the SPGD AO system performance we compared the average values
of the metric (received power)with and withoutoperationof the system. Since the AO
system was not able to cancel out the strong fluctuations (cf. Fig. 8a) of the received
5 SPGD High-Resolution Wave-Front Control 263
1.2- . - - - - - - - - - - - - - - - - - - ,
P cr} =1.03 ,", cr} =0.16
1.0
">;: " " !.~'1"",i..
0.8
2 ,
,r
I \
,
. . •
crJ = 0.48 ,
0.6
~
,
,
0.4 ,
,, .""
0.2 ,,
\
.... ~~-:'. • .1- ··
0.0 I--~~~~,........,-~~~~.....,.-~~~
0.01 0.1 J/<J>
Fig. 9. Three sample probability density funct ions p( J) for the received power J demon strating
the range of the variance l1J
= (/ 2)/ (1)2 - 1 observed under various atmospheric conditions.
J values were normalized by their corre sponding mean value (J )
Phase: 1 2 3 4 5 Phase: 1
Tilts: off on on on off Tilts: off
llAo : off off on off off IlAO; off
5 5
J Trial 1 S., J Trial 2
4 "tl 4
Ol
c
3 .,
"S;
III
3
0
0 2 3 4 t (5) 5
Fig. 10. Method for evaluation of the performance of the AD system: Metric (received power)
values J were recorded for a trial with five phases of different operation conditions forthe adaptive
systems (as indicated) and subsequently saved before the next trial was started. Data evaluation
(e.g. calculation of average metric evolution curves) was performed on data sets of 1000 or more
recorded trials
power completely (due to the received wave's intensity scintillations and residual wave-
front errors), a rather long integration time was required in order to obtain an acceptable
error figure for the average power, (J) . Monitoring the atmospheric conditions of
our propagation path with a scintillometer showed that atmospheric conditions could
change quite rapidly and correlation could be lost within minutes. Thus , it was not
feasible to record J( t) for a longer time when the AO system was off for obtaining the
uncompensated power average, (Joff ), and compare it with the value (Jon) , recorded
earlier or later with the AO system working . A separation between the effects from the
AO system and from the changes in atmospheri c condition s would not be possible .
Since correlation between turbulence conditions is higher within short temporal
distances, we performed experiments where we switched among five different adaptive
system operation conditions in short identical time intervals (usually one second or
less) and repeated these trials many (typically WOO) times. The procedure is depicted
in Fig. 10. In phase I we observed the metric J with the feedback control off for
264 Thoma s Weyrauch and Mikhail A. Vorontsov
1. 0 , . - - - - - - - - - - - - - - - - - - ,
P ,," ~' .
\
0.8 "\ . .'....

P1(J) " :, I ,
[no adaptatlon] ,: 'v,t, ..
0.6 ~ : p,(J) [wave-front " I
I tift compensation I '. "
only] ~,'.,
I
0.4 I
P3(J) [beamsteering I
and SPGDAO '
0.2 compensationl~
0.01 0.1 J
Fig. 11. Probability density functions Pi (J) of received power values J for three different oper-
ation conditions of the AO system : no adaptation (i = 1, BS and !JAO off), beam steering only
(i = 2, BS on,!JAO oft), and beam steer ing together with high-resolution SPGD AO control
(i = 3, BS and !JAO on) . Deep fading (occurrence of low J values) is strongly reduced with
wave-front correction
both the fast beam steering and the SPGD AO systems ; in phase 2 the beam steering
(wave-front tilts compensation) system was on while the SPGD AO control remained
off. In phase 3 both adaptive subsystems were operating. In phase 4 the ~O system
was off, and finally in phase 5 no adaptive correction was used (both subsystems were
off).
In each phase , a fixed number of iterations (typically several thousand) were per-
formed. The values of the metric J measured for all iteration steps were saved after
one trial (phases I through 5) was completed and before the next trial was started.
By implementing this measurement scheme we were able to compare only data that
were measured during close time intervals and thus minimized the influence of long-
term fluctuations of atmospheric conditions , while short-term fluctuation effects were
mitigated by averaging a large number of trials.
The influence of both the beam steering system (BS) and the SPGD AO system
(~O) on the probability density function, p(J), of the received power values is shown
in Pig. I I. Curves Pi(J) were calculated for each of the phases i = 1 to 3 by process-
ing the recorded metric data after the whole set of trial measurements was finished.
Data points within the first O. I seconds of each trial phase were neglected so as to
exclude the influence of the transition processes. The PDP curve P2 (J) corresponding
to wave-front tilt compensation is significantly narrower than the PDP PI (J) without
compensation and its maximum value is shifted to larger values of the metric. Just
wave-front tip/tilt compen sation alone allowed reduction of the received power vari-
ance O"} = (J2)j(J)2 - 1 from 0.53 (no compensation) to 0.19 (wave-front tip/tilt
compensation). When the SPGD AO system was turned on in addition to the beam
steering system (PDP P3(J) , the probability of low values of the received signal,
i.e., the fading responsible for bit errors in laser communication systems, was fur-
ther reduced and the average (J) almost doubled. The average C~ value during this
experiment was 3.6 x 10- 15 ill - 1/3 corresponding to O"'k = 0.73.
1.0 a
<J>
0.8
0.6
0.4
0.2
....... jlAO on
.. .......... :
~
jlAO off ..
0.0~==='===~==~==:2==~~
1.0
2
crJ
0.8
0.6
o 200 400 600 800 1000

t (ms)
Fig.12. Mitigation ofatmospheric-turbulence induced wave-front distortions with the SPGD AO
system (~O). (a) The evolution curve demonstrates the improvement of the average received
power (J (t) ) after ~O is switched onat t = 0 and itsdecrease after feedback is switched offat
t ~ 550ms. The beam steering system is always onduring the experiment. (b)Thecurve shows
thecorresponding improvement of the normalized metric variance o-}(t) = (J 2) / (J}2 - 1
Figure 12a shows the averaged metric evolution curve, (J (t) ), for an experiment
with C~ = 2.1 X 10- 15 m- 2 / 3 (Rytov variance 11'~ = 0.42). Since we were inter-
ested particularly in the performance of the SpaD AO system, a shorter trial scheme
comprising only phases 3 and 4 of the measurement procedure shown in Fig. 10 was
used, i.e., the beam steering system was continuously operating. The SPGD AO sys-
tem (~O) was on for 550 ms and then turned off during the consecutive 550 ms. The
average curve (J (t) ) was calculated from 1000 trials. As can be seen from Fig. 12a,
operation of the SPGD AO system improved the averaged metric value (J) by a factor
of 2.3. Metric fluctuations were significantly lowered by the ~O system as shown
in Fig. 12b, where the normalized metric variance 11'}(t) = (J (t )2)j( J (t ))2 - 1 is
plotted.
The dynamics of the average metric (J (t )) during the optimization process after
switching the SPGD AO system on at t = 0 characterize the adaptation speed (and
hence the closed-loop compensation bandwidth) of the system for compensation of
distortions from atmospheric turbulence. In contrast, the decay of (J (t) ) after ~O is
switched off at to = 550 ms reveals information about the temporal correlation and
hence the dynamics of atmospheric-turbulence-induced phase distortions. Wave-front
correction for the data shown in Fig. 12a was possible because the SPGD adaptation
process convergencewas faster than the characteristic time scale of turbulence induced
distortions changes (80% of the metric improvement during the optimization process
200
150
100
50
250
200
150
Mode-field
diameter of
single-mode
100 fiber
SPGD AO system
(~O)off
50 "-" , ,
oi==;;::;:::::::;:::;.......,.. . . . . .~~...--,-~-r-...--,-~~ ::=;:::::;:::::j

-25 -20 -15 -10 -5 o 5 10 15 20 25
X (um)
Fig. 13. (a, b) Long-exposure distributions of the irradiance I(r) in the focal plane F" (in Fig.
5) and the corresponding 3-D plots recorded while the spaD AO system (~O) was off and on
(a and b, respectively). Wave-front tilts control with the beam steering system (BS) was on in
both cases. (c) The corresponding 2-D beam profiles were re-scaled to the dimen sions in the focal
plane F' of the receiver fiber SM 1 . The mode field diameter of SM 1 (5.5 11m) is indicated as
circle
was accomplished within 20 ms, while it took 50 ms to loose 80% of the improvement
after the JlAO system was turned off).
The long-exposure irradiance distribution (IF" (r)) in the focal plane:F" recorded
under the same atmospheric conditions with the camera CCD (see Fig. 5) for both
operation conditions, SPOD AO system off and on, are shown in Fig. l3a,b along with
the respective 3-dimensional plots of the focal plane irradiance. Beam steering was
on for both conditions; we discuss here only the improvement of the beam in addition
to beam steering . The irradiance pattern (!:r-" (r)) was obtained by averaging 400
frames recorded during one minute. The corresponding beam profiles are depicted in
Fig. l3c. Similar to the metric values in Fig. l2a, the maximum intensity was about
2.3 times higher with the SPGD AO system operating than with beam steering only.
However, the intensity profile curve obtained even with higher-order wave-front control
(i.e., during flAO operation) was still about two times wider than the diffraction limit,
which corresponds approximately to the mode-field diameter (5.5 urn) of the single-
mode fiber SM t (indicated in Fig. l3c as dashed circle). Tests showed that this can not
be attributed to the optical system's aberrations, but is due to residual distortions from
atmospheric turbulence.
5.4. SPGD Adaptive Transceiver System
In the atmospheric compen sation experiments described in Sec. 5.3, the SPGO AO
system was used in an adapti ve receiver configuration (cf. Fig. 2a). However, the
experimental setup allowed also to investigate an adaptive transceiver as shown in Fig.
2c. In order to transmit light from the laboratory to the water tower through the adaptive
optics system, we coupled the light from laser diode LD (see Fig. 5) into single-mode
fiber SM2 and positioned the fiber end within focal plane F '" in a position conjugate
to the receiver fiber SM 1 in plane F'. After collimating the beam with lens L9, the
beam was coupled by beam splitter BS3 with opposite propagation direction into the
light path of the received beam .
The transmitter of the optical antenna T2 (Fig. I) was used as beacon, and its
receiver unit determined the magnitude of the optical signal received on the water
tower, i.e., the remote metric, h. Customized transceiver electronics digitized h data
so that they could be transmitted optically to the laboratory site, where they were
received by a second transceiver unit, which provided the Ja values to the supervisory
computer of the SPGO system . Although data transmission was performed at a rate of
nearly 10,000 s -1, the latency from the signal coding electronics was too large to use
h data as feedback signal for the SPGO controller. Hence , it was not possible to use
this setup as an adaptive transceiver that uses a combination of the local and the remote
metric (J and J 2 , respect ively), or even as an adapt ive transmitter (Fig. 2b), which
relies solely on feedback from J2. Experiments were thus performed maximiz ing the
local metric J while recording the simultaneously measured data for J2.
Results for the averaged evolution curves of both metrics J and J 2 for an exper-
iment with C~ = 2.1 X 10- 15 m - 2 / 3 «1~ = 0.42) are shown in Fig. 14. Again the
trial consisted only of phases 3 and 4 (cf. Fig. 10), i.e., the SPGO AO system was
periodically turned on and off. Indeed both metrics (J (t)) and (h (t )) were improved
when the IJAO system was working, i.e., the SPGO adapti ve transmitter configuration
was capable of mitigating beam spreading and increasing the irradiance at the receiver
aperture on the water tower. Note that this improvement in J 2 was in addition to beam
steering; the tilt compen sation subsystem was always on).
Although there is a simultaneous improvement of metrics J and h when the IJAO
system is working, the system's influence on J2 is much less than on J . This can be
explained by the anisoplanatism of the setup. Transmitter and receiver apertures of
the lasercom antenna T2 on the water tower (both 6 ern in diameter) are mounted with
a distance (between centers) of about 8 cm corresponding to an angle of 341lrad as
seen from the AO system in the laboratory. This is larger than the isoplanatic angle,
e iso , which can be estimated to e iso = [ 1. 0gk2C~ L8 /3 r 3 /5 ~ 131lrad for the at-
mospheric conditions during the experiments (assuming a constant C~ value along
the propagation path). Since e iso correspond s to a displacement of 3 cm at the wa-
ter tower, wave-front distortion compensation for the transmitted beam can not be as
successful as under isoplanat ic conditions. This example illustrates that a transceiver
with an aperture shared by transmitter and receiver is important for the efficiency of
atmospheric compen sation with adaptive optics in free-space laser communications.
2. 6 . . . . . . - - - - - - - - - - . - - - - - - - - - - - ,
<J> ,'\ !
2.4
<J 2 > A ,./ \\"'V'.,.JII"'\;t'V~~
./V I
2.2 I I
i I
2.0 ,i
, , ~ J (local receiver)
1.8 ,,, !: /
I '~
1.6 ,
,
i:
,, I:,
1.4 ,,, ,
.......,....\
1.2 ~•••\,.••I" ..
...... ;'-\ ...
,,;,,,...,.,,,,,,,,,
IJAO on
o 200 400 600
Fig. 14.Resultsfroman experimentusingan adaptive transceiver configuration: (J (t») describes

the evolution of the local metric, J (powerof the beam receivedat the laboratory with the AO
setup),whichservedas feedbacksignalfortheSPGDcontroller. TheSPGDAOsystem(~O) was
turnedon at t = 0 and off at t = 550 ms while beamsteering(BSS) was continuously working.
The curve of the remote receiver metric, Ja. i.e., the signal strength of the beam received by
transceiver T2 at the water tower, demonstrates simultaneous improvement of incoming and
outgoing beam.
6. Summary and Conclusion
Wave-front distortions induced by atmospheric turbulence can severely impact perfor-

mance of free-space optical communications causing signal fading on a millisecond
time scale. In principle, these aberrations can be mitigated by adaptive optics (AO),
but conventional techniques with wave-front sensing and reconstruction face very
challenging problems and perform poorly under conditions of strong scintillations
caused by distributed turbulence along the propagation path. Using a laboratory-based
non-conventional, SPGD based adaptive optics system, we demonstrated wave-front
control for free-space optical communication scenarios under adverse conditions (scin-
tillation index o} ~ 1) for optical wave propagation over a 2.3 km near-horizontal
atmospheric path. The AO setup consisted of a beam steering system for tip/tilt control
and the higher-resolution AO system that omitted wave-front measurements by using a
stochastic parallel gradient descent (SPGD) algorithm for optimization of the received
signal strength used as performance metric. Thus, SPGD AO offers an attractive al-
ternative that can operate under strong scintillation conditions. Because of its system
architecture, incorporation ofSPGD adaptive optics in free-space laser communication
terminals is relatively simple.
In the future, there are several ways to improve the performance of SPGD AO
systems, i.e., primarily to increase convergence speed. First, there is the prospect of
faster wave-front correctors [17]. Second, the SPGD system performance can be in-
creased by the introduction of an additional low-resolution deformable mirror with
an independent control loop for compensation of low-order but high-amplitude dis-
tortions (as defocus or astigmatism) in the AO system [29]. Third , advanced SPGD
References 269
controllers may implement "turbulence-friendly " perturbation statistics, which provide

much faster convergence than Bernoulli statistics [14]. New SPGD control hardware
must be developed before the advantage of such improved perturbation statistics can
be demonstrated on an atmospheric propagation path.
Acknowledgment
The authors are grateful to their colleagues Gary Carhart and Matthew Banta at the
Army Research Laboratory's Intelligent Optics Laboratory for their important contri-
butions to the development of the atmospheric laser optics testbed used in our experi-
ments. We also thank Thomas Bifano for providing the MEMS deformable mirror and
Marc Cohen for assistance with the AdOpt VLSI controller. Part of the experimental
work was performed while T. Weyrauch held a National Research Council Research
Associateship at the Army Research Laboratory . This research was funded by the U.S.
Joint Technology Office under contract JTO-02-602-l8, the Battelle Memorial Insti-
tute under contract DAAD0599D70l4, and the National Science Foundation under
contract ECS-OO 10026.
References
I. I. S. Reed and X. Chen , Error-Control Codingfor DataNetworks, Vol. 508 of The KluwerIn-
ternationalSeries in Engineering and ComputerScience (Kluwer Academic, Boston, 1999).
2. J. W. Hardy, Adaptive Opticsfor AstronomicalTelescopes, Vol. 16 of OxfordSeries in Optical
and ImagingSciences (Oxford University Press, UK, 1998).
3. F. Roddier, Adaptive Optics in Astronomy (Cambridge University Press, UK, 1999).
4. C. A. Primmerman, T. R. Price, R. A. Humphreys, B. G. Zollars, H. T. Barclay, and J. Her-
rmann, "Atmospheric-compensation experiments in strong-scintillation conditions," Appl.
Opt. 34, 2081-2088 (1995).
5. B. M. Levine, E. A. Martin sen, A. Wirth, A. Jankev ics, M. Toledo-Quinones, F. Landers, and
T. L. Bruno, "Hor izontal line-of-sight turbulence over near-ground paths and implications
for adaptive optic s correction in laser communications," Appl. Opt. 37,4553-4560 (1998).
6. N. B. Baranova, A. V. Marnaev, N. F. Pilipet sky, V.V. Shkunov , and B. Y. Zel'dovich, "Wave-
front dislocations: topological limitations for adaptive systems with phase conjugation," J.
Opt. Soc. Am . 73,525-528 (1983).
7. D. L. Fried, "Branch point problem in adaptive optics," J. Opt. Soc. Am. A 15,2759-2768
(1998).
8. A. Buffington, F. S. Crawford, R. A. Muller, A. J. Schwemin, and R. G. Smits, "Correction of
atmospheric distortion with an image-sharpening telescope ," J. Opt. Soc. Am. 67, 298-303
( 1977).
9. S. L. McCall , T. R. Brown, and A. Passner, "Improved optical stellar image using a real-time
phase-correction system: initial result s," Astrophysical Journal 211,463-468 (1977).
10. J. W. Hardy, "Active optics : a new technology for the control of light," Proc. IEEE 66,
651-697 (1978).
II . T. R. 0 ' Meara , "The multi-dither principle in adaptive optics," J. Opt. Soc. Am. 67, 30~315
(1977).
270 Thomas Weyrauch and MikhailA. Vorontsov
12. J. E. Pearson and S. Hansen, "Experimental studies of a deformable-mirroradaptiveoptical

system," J. Opt. Soc. Am. 67, 325-333 (1977).
13. M. A. Vorontsov, G. W. Carhart, and J. C. Ricklin, "Adaptive phase-distortion correction
based on parallel gradient-descentoptimization," Opt. Lett. 22,907-909 (1997).
14. M. A. Vorontsov and V. P.Sivokon,"Stochastic parallel-gradient-descent techniquefor high-
resolutionwave-frontphase-distortioncorrection," J. Opt. Soc. Am. A 15,2745-2758 (1998).
15. J. C. Spall, Introduction to StochasticSearch and Optimization (Wiley,Hoboken,NJ, 2003).
16. R. T. Edward, M. Cohen, G. Cauwenberghs, M. A. Vorontsov, and G. W. Carhart, "Ana-
log VLSI parallel stochastic optimization for adaptive optics," in Learning on Silicon, G.
Cauwenberghsand M. A. Bayoumi, eds., (KluwerAcademic, Boston, 1999),Chap. 16, pp.
359-382.
17. T. G. Bifano, J. A. Perreault, P.A. Bierden, and C. E. Dimas, "Micromachined deformable
mirrors for adaptive optics," In High-Resolution Wavefront Control: Methods, Devices, and
Applications IV, J. D. Gonglewski, M. A. Vorontsov, M. T. Gruneisen, S. R. Restaino, and
R. K. Tyson,eds., Proc. SPIE 4825, 10--13 (2002).
18. M. A. Vorontsov, G. W. Carhart, J. W. Gowens II, and J. C. Ricklin, "Adaptive correction
of wave-front phase distortions in a free-space laser communication: system and method,"
Patent pending.
19. M. A. Vorontsov, G. W. Carhart, L. A. Beresnev, and J. W. Gowens II, "Adaptive WDM
free-space laser communication system: low-order aberration compensation experiments,"
presented at the International Symposium on OpticalScience and Technology, Denver, Col-
orado, Conference 5550 "Free-Space Laser CommunicationsIV," 2-6 August 2004.
20. M. A. Vorontsov, G. W. Carhart, M. Banta, T. Weyrauch, J. Gowens II, and J. C. Carrano,
"Atmosphericlaseroptics testbed(A_LOT): atmosphericpropagationcharacterization,beam
control and imaging results," In Advanced Wavefront Control: Methods, Devices, and Ap-
plications, J. D. Gonglewski, M. A. Vorontsov, and M. T. Gruneisen, eds., Proc. SPIE 5162,
37-48 (2003).
21. M. E. Gravecha,A. S. Gurvich, S. S. Kashkarov, and V. L. V. Pokasov,"Similarity relations
and their experimental verification for strong intensity fluctuations of laser radiation," in
Laser Beam Propagation in the Atmosphere,J. Strohbehn, ed., (Springer, NewYork, 1978).
22. L. C. Andrews, R. L. Phillips, and C. Y. Hopen, Laser Beam Scintillation with Applications
(SPIE Press, Bellingham,WA, 2001).
23. Y. A. Kravtsov, "New effects in wave propagation and scattering in random media (a mini
review)," Appl. Opt. 32, 2681-2691 (1993).
24. M. A. Vorontsov, G. W. Carhart, M. Cohen, and G. Cauwenberghs, "Adaptive optics based
on analog parallel stochasticoptimization: analysisand experimentaldemonstration," J. Opt.
Soc. Am. A 17, 1440-1453 (2000).
25. T. Weyrauch, M. A. Vorontsov, T. G. Bifano,J. A. Hammer,M. Cohen,and G. Cauwenberghs,
"Microscale adaptiveoptics: wave-frontcontrol with a fL-mirror array and a VLSI stochastic
gradient descent controller," Appl. Opt. 40,4243-4253 (200I).
26. T. Weyrauch and M. A. Vorontsov, "Dynamic wave-front distortion compensation with a
134-control-channelsubmillisecondadaptive system," Opt. Lett. 27,751-753 (2002).
27. T. G. Bifano,J. Perrault, R. Krishnamoorthy Mali, and M. N. Horenstein,"Microelectrome-
chanical deformable mirrors," IEEE J. Sel. Top. Quantum Electron. 5, 83-89 (1999).
28. J. C. Ricklin and F. M. Davidson, "Atmospheric optical communication with a Gaussian
Schell beam," J. Opt. Soc. Am. A 20, 856-866 (2003).
References 271
29. T. Weyrauch and M. A. Vorontsov, "Atmospheric compen sation with a multiconjugate

(piston-MEMS/modal OM) adaptive system," In Target-in-the-Loop: Atmospheric Track-
ing. Imaging and Compensation, M. T. Valley and M. A. Vorontsov, eds., Proc . SPIE 5552,
73-84 (2004).
© 2005 SpringerScience-BusinessMediaInc.
DOl: 1O.1007/sI0297-004-0025-x
Originally published in J. Opt. FiberCommun. Rep. 2, 56-85 (2005)
Optical networks, last mile access and applications
E. Leitgeb, M. Gebhart, and U. Bimbacher
Department of Communications and Wave Propagation

Graz Technical University, A-80 10 Graz, Inffeldgasse 12, Austria
Email :leitgeb@inw .tu-graz .ac .at
Abstract. Free Space Optical (FSO) links can be used to setup FSO communication
networks or to supplement radio and optical fiber networks . Hence, it is the broadband
wireless solution for closing the "last mile" connectivity gap throughout metropoli-
tan networks . Optical wireless fits well into dense urban areas and is ideally suited
for urban applications. This paper gives an overview of free-space laser communi-
cations. Different network architectures will be described and investigated regarding
reliability . The usage of "Optical Repeaters ", Point-to-Point and Point-to-Multipoint
solutions will be explained for setting up different network architectures. After having
explained the different networking topologies and technologies, FSO applications will
be discussed in section 2, including terrestrial applications for short and long ranges,
and space applications. Terrestrial applications for short ranges cover the links between
buildings on campus or different buildings of a company, which can be established with
low-cost technology. For using FSO for long-range applications, more sophisticated
systems have to be used. Hence, different techniques regarding emitted optical power,
beam divergence, number of beams and tracking will be examined. Space applica-
tions have to be divided into FSO links through the troposphere, for example up- and
downlinks between the Earth and satellites, and FSO links above the troposphere (e.g.,
optical inter-satellite links). The difference is that links through the troposphere are
mainly influenced by weather conditions similar but not equal to terrestrial FSO links .
Satellite orbits are above the atmosphere and therefore, optical inter-satellite links are
not influenced by weather conditions. In section 3 the use of optical wireless for the
last mile will be investigated and described in more detail. Therefore important design
criteria for connecting the user to the "backbone" by FSO techniques will be covered,
e.g., line of sight, network topology, reliability and availability . The advantages and
disadvantages of different FSO technologies, as well as the backbone technology are
discussed in this respect. Furthermore, the last mile access using FSO will be investi-
gated for different environment areas (e.g., urban, rural, mountain) and climate zones.
The availability of the FSO link is mainly determined by the local atmospheric condi-
274 E. Leitgeb, M . Gebhart, and U. Bimbacher
tions and distance and will beexamined forthelast mile. Results ofvarious studies will
complete these investigations. Finally, anexample forrealizing a FSO network for the
last mile will be shown. In this example FSO transmitters with light emitting diodes
(LED) instead of laser diodes will be described. By using LEDs,problems with laser-
and eye safety are minimized. Some multimedia applications (like video-conferences,
live TV-transmissions, etc.) will illustrate the range of applications forFSO last mile
networks.
1. Optical Networks
Free Space Optics (FSO) links can be used to setup a complete FSO network or as a
supplement to conventional radio links and fiberoptics.
Most FSO links are point-to-pointlinks betweenone transmittingand one receiv-
ing station. However, current research [1,6,9] is also investigating point-to-multipoint
configurations. By using light sources like LEOs with a wide beam angle and/or mul-
tiple transmitter optics, the output power within the same laser class can be increased
multiple times, and several terminal stations can be connected to central station.
1.1. Types of FSO Systems for Different Network Architectures
DifferentFSOsystemsfor variousapplicationshavebeendevelopedby researchgroups

and industry. In order to describe the different technologiesimplementedin available
systems, we simplify the concepts to some important elements. Such a simplified
scheme for a typical FSO unit for data transmission is shown in Fig. I. On the basis of
this concept, three types of system design can be distinguished,each as a compromise
regardingreliableoperationand installationcosts forcertainapplicationsanddistances.
E ltc tronic s R'tc.imt Op tics Regime A tmo Jphe ric Link
Fig. 1. Simplified scheme of the elements of a FSO unit.
For the firsttype of systems, availablestandardcomponentsare combined to build

cost-effective solutions. Generally, the value of the beam divergence is high (up to
60 mrad) and allows an easy alignment for the user without the need for a telescope.
All elements including the optics do not require as high precision as compared to the
other two approaches, allowing higher tolerancesat fabrication, and allowing the use
of simple mounts at installation. Owing to large divergence the requirements for a
stable underground are not very high, allowinga quick installation. Suitable distances
for high availability operation are limited up to 300 m because of the wide beam angle
(and depending on local climate). Owing to laser safety regulations the use of light
sources with higher beam divergence allow more output power in the same safety class,
which improves the link budget for short distance applications.
All Optical Network
OGS
Fig, 2. All Optical Network .
Systems of the second type take advantage of a collimated beam as much as

possible without active alignment, which leads to a beam divergence of about 2 to 6
mrad. The implementation of fiber-optic technology for coupling directly from fiber
to FSO is a suitable concept. To achieve the same intensity at the receiver in longer
distance, the concept requires precise optical components being carefully adjusted, and
telescopes or adjustment procedures are needed for the installation. A stable mount on
stable ground is required. These first two types of FSO systems are mainly used for
short range terrestrial applications and especially for last mile access . For increasing the
intensity at the receiver and to overcome atmospheric turbulences (scintillation) caused
by variation of the refraction index, more than one transmitter and / or receiverunits
are used in one housing . Systems with multiple transmitter/receiver units are called
Multibeam- or Multlink-FSO systems and are found in both system types .
The third category contains the sophisticated solutions, including strategies to
extend the distance for reliable operation as much as possible. The beam divergence
reaches values less than I mrad leading to low geometrical losses, concentrating all
incoming light on a small area around the receiver. To compensate building sway and
deviation, automatic tracking of the beam is implemented, adaptive optics may help
to compensate atmospheric fluctuation s or allow higher optical output over a large
area. These systems are in use for long distances and the most accurate solutions with
acquisition and tracking for space applications. For space applications not only direct
detection for the receiver is used, but also coherent detection.
276 E. Leitgeb,M. Gebhart, and U. Birnbacher
Using FSO systems from all three categories, various communication links or
networks can be realized, including terrestrial FSO (between buildings, hospitals,
campus), horizontal and slant paths, air-borne and space communications (between
aircraft, unmanned aerial vehicles (UAY), high altitude platforms (HAP) and satel-
lites. By connecting FSO links and networks to the Backbone realized with optical
fibers, the FSO networks and links can be seen as parts of a global all optical network
around the world (Fig. 2). A very important issue is the reliability and the availability
of FSO links and FSO networks. In order to estimate the reliability of a network, the
reliability of each single component has to be taken into account. For exampled, the
reliability of one FSO terminal depends on the reliability of all electrical and opti-
cal components built into the terminal, including the connection and interface to the
network (Fig. 3). The reliability of the FSO link is determined by the reliability of
the transmitting FSO terminal (unit I at location A) and the receiving FSO terminal
(unit 2 at location B) and the quality of the optical path in-between . The reliability and
availability of the optical path is mainly influenced by the local weather conditions,
with fog being the most limiting factor. If not only a single FSO link but a network
of optical wireless connections is used, the overall reliability is calculated by taking
into account all the links in the network and their specific arrangement to each other
(parallel or serial). Hence, different network architectures have different availabilities
and reliabilities .
Reliability of Reliability of
the FSO- Reliab ility the FSO -
System 1 of the System 2
FSO-li nk
Re li ability of R eli abili ty of
Reliability of Reliabi lity of
th e electri cal th e el ectri cal
t h e interface the interfac e
and optic al an d optical
an d connection an d conn ecti on
comp one nts ma inly influen ced by comp on ents
the local weather
Fig. 3. Reliability of FSO links.
1.2. Architectures of FSO Networks (Point-to-Point and Point-to-Multipoint

Configurations)
In the following section different architectures are described and compared regarding
the overall reliability.
1.2.1. Optical Wireless in Ring Architecture

In Fig. 4 a FSO network in ring architecture is shown. In the given example, the
distances between the buildings are up to 500 m. In the minimum configuration two
optical receiver/transmitter units are installed on the top of each building. Optical
Repeaters have to be used, if there is no line of sight between transmitter and receiver.
In the event of a broken link or for example a link failure between buildings I and
2, the indirect connection can be used. Thereby, the information is sent in the other
Legend .. Fr ee Sp ace
Optica l - Link
• O pt ical Tr ansceiver Unit

• Opt ica l Repeater
Fig. 4. FSO network with ring architecture.
direction of the ring network passing buildings I, 6, 5, 4, 3, and 2. Hence, using a

ring architecture, a partial security against failure can be achieved. The installation of
additional, redundant links increases the availability and the security against failure.
1.2.2. Optical Wireless in Star Architecture
In Fig. 5 a test installation at the University of Technology Graz in Austria (TU Graz)
is shown, using a star architecture. The coverage area of this FSO network is about
300 m in diameter. An optical multipoint unit is mounted at the roof of a building.
Five user terminals are permanently connected by their optical transceiver units to the
optical multipoint hub station . The five user terminals are located at the surrounding
buildings and offer a connection via the FSO link to the optical multipoint unit. In
this configuration the optical multipoint unit is realized with five FSO Point-to-Point
units, each of them directed to one user FSO terminal. The optical multipoint unit is
interconnected by switches with the backbone network ofTU Graz.
The advantage of this configuration is the shorter distance between any two FSG
units. In general, the optical multipoint unit is located in the center of the area, mounted
on the tallest building. But this architecture has the disadvantage of a single point of
278 E. Leitgeb, M. Gebhart, and U. Bimbacher
Building 6
Street
Legend ... ... Free Space

Optical - Link
• Optical Transceiver Unit
Optical Muhi-point Unit
Fig. S. FSO network with star architecture.
failure. If the optical multipoint unit fails, a system breakdown of the whole installation
is caused . To improve the reliability of this architecture, a redundant Multipoint Unit
would have to be installed . The second optical multipoint unit can also be mounted on
moveable platforms, e.g., a van, for increasing the flexibility and decreasing the setup
time. However, for the installation of multipoint units on cars, FSO systems with "auto
tracking" are preferable.
1.2.3. Optical Wireless in Meshed Architecture

For high reliability, the optimum network architecture is a meshed network . Meshed
networks combine the benefits of the above described architectures, because different
connections are possible. An example of a meshed FSO network is given in Figs. 19
and 20 in section 3.
1.3. Connecting to the Backbone
The access network resides between the individual subscribers and the network op-
erator's backbone network . fiber often does not reach paying customers. Only 3% of
worldwide businesses are on fiber, and 75% are within a mile from fiber. Nowadays
the backboneis realizedwith fiberand so the connection with FSO offers manyadvan-
tages. The FSO systems work protocol transparent like a fiber link. Hence, the same
networking technologies used in fibers can be used over FSO. It is possible to couple
from FSO systems directly into the optical domain or with optical/electrical conver-
sion and regeneration to the network-fiber. Also coupling into a Wavelength Division
Multiplex (WDM) is possible, by connectingdifferentoptical wirelesssystems with a
wavelength division multiplexing unit to the WDM-fiber network.
In different architectures (section 1.2) and configurations, the FSO unit can be
connected to satellites, directional radio links, (mobile) telephone networks, or fiber
optics. In Fig. 6 a connection to a fiber- based backbone realized with a Point-to-
Multipointarchitecture is shown. The optical multipoint unitis connectedwitha switch
or router to the backbonenetwork(thicksolid line in Fig. 6). The users in the buildings
I, 2, 3, and 4 are linked with their FSO terminalunits to the central optical multipoint
unit.
2. FSO Applications
This section covers terrestrial applications for short and long ranges and space ap-
plications. Terrestrial applications and FSO links trough the troposphere are mainly
influenced by weatherconditions. Therefore,some importantcharacteristics of the at-
mospherehave to be discussedbefore describingthe FSO applications in more detail.
The lowestpartofthe atmosphere upto 10kmabovetheEarth's surfaceiscalledthe
troposphere or the weathersphere.It has a varying refraction indexwhichis dependent
on the height above the Earth's surface. Normally the refraction index decrease with
the height, but at inversion situationsthere is different relationship.
Atmospheric conditionsdegradelaser communications throughthe atmospherein
two ways. First, the atmosphere acts as a variableattenuatorbetween the transmitting
and receiving terminals. Second, a free space laser link is subjected to scintillations.
Attenuation is caused by the weatherconditionsalong the transmission path. Gen-
erally, there is low atmospheric attenuation during clear days and high attenuation
during foggy days. Raindoes not influence opticaltransmissions heavily, becauserain-
drops have the size of a few millimeters and are large compared to laser wavelengths
(1.5 microns) and thus cause minimal scattering of the laser energy. Furthermore, water
has minimal absorption at a 1550-nm laser wavelength. Therefore, it is not surprising
that the optical transmission is not heavily impacted by rain (only about 3 dBlkm).
Similarly it is not surprising to find out that optical transmission is impacted dramati-
cally by heavy fog(30 dBlkm). This is becausethe fogaerosolshavea comparablesize
as the used wavelengths, causing much scattering of the laser energy as the fog gets
thicker. Absorption effects can be subdivided into absorption and scattering effects.
Absorption is caused by many differentspeciesof gas in the atmosphere, the dominant
one being water vapor, which is in the wavelength region used for wireless optical
links. There are two types of scattering, Rayleigh and Mie scattering (see section 3).
The amount of scattering depends on the particle size distribution and the density of
particles. Generally, any optical wavelength could be used for FSO, but because of
the atmospheric conditionsand due to the laser safety regulations, 1550nm is the best
280 E. Leitgeb, M. Gebhart, and U. Birnbacher
Building 4
Legend
Optical Fiber
.. .. Free Space
Optical- Link
o Optical Mutti-point Unit
Fig. 6. Optical Wireless Access with Point-to-Multipoint Architecture.
suitable wavelength. The losses due to Mie-scattering in haze or light fog are smaller
at longer wavelengths (1550 nm) than at shorter ones (850 nm).
The second major influence on FSO transmission is scintillation. which is caused
by small-scale fluctuations in the refraction index of the atmosphere. Its primary effect
is signal fading due to phase changes in the wave front arriving at the receiver. Unless
the receiver has a very high dynamic range or the aperture is large enough to average
out the scintillation spots, this can have an extremely detrimental effect on the signal.
As described above, a foggy day is very bad for free space communication due to the
high attenuation. On the other hand, it can be observed that scintillations are very low
on a foggy day, because of low fluctuations in the atmosphere.
2.1. Short Range Aapplications and Last Mile Access
FSOshort-range applications willbe usedup to I km (maximum 2 km) for connecting

buildings or establishing a connection to the backbone.
Within the range of up to 2 km distance, it makes sense to use FSO systems of
type I or type 2 (see section 1.1). Thesesystems will be cost-efficient and they have a
high availability and reliability. For short ranges, "auto tracking" is not necessary. In
[ I] the ranges for developed type I FSO systems are 100, 300, and 800 m (Figs. 7A
and B). In Fig. 8 an example for a type 2 system is shown.
Fig. 7, Point-to-Point system of type I for ranges of A) 100 and B) 800 m (with 8 transmitter
units).
FSO
LlvIDS
Fig. 8. Installation of a GoC Multilink 155F in combination with LMDS.

282 E. Leitgeb,M. Gebhart, and U. Birnbacher
In [2,6] various FSO systems have been evaluated in regard to different weather
conditions. For terrestrial applications it is important to know the influence of the path
through the atmosphere, which can be either simulated and modelled or measured
over long time periods . The atmospheric path for transmission of a collimated beam
of light may be seen as an information channel, to which information is added and on
which information gets lost (Fig. 9). A variety of effects can be described and several
theoretical models are known, but not all of them are relevant to this technology.
Once more it is a question of technology used, distance and application to find out
relevant impacts on FSO data transmission . Before coming to quantitative results from
experiments, a qualitative introduction to some relevant problems is given in this
section .
Dsta.Networ k i
ii FSO-Unit
lnforrnation Channel
l(IU . or""lh.Ol ....... . <:Il
. , , ,•• • •• .:0.'1( ••
Fig. 9. Schemeof data transmission and impactson the channel.
Ambient light that passes the optics and reaches the detector element, which usu-
ally is a PIN-photodiode or an avalanche-photodiode (APD), causes additional current
that leads to additional white Gaussian noise in the receiver. Depending on the semicon-
ductor materials and construction principle of the detector there are additional effects
coming up. Flicker noise produces irregular disturbances which, if we consider data
transmission, affects throughput of longer data packets most. At the same time the noise
level raises, the impedance of the detector element decreases, which in most receiver
principles leads to a level reduction of the received information signal, sometimes also
referred as "bum out" effect. The dominant source of ambient light for outdoor FSO
systems is the sun. Therefore the sunlight and the possible positions of the sun have to
be considered for practical applications.
An intensity of about 1340 W/m 2 reaches the Earth, also known as the solar
constant, which is reduced to about WOO W/m 2 global radiation on ground under
clear sky conditions due to wavelength dependent absorption in the atmosphere. Over
90% comes as direct sunlight, the rest looses directivity and is scattered over the sky
contributing to background illumination . The sun is seen from the Earth under an
angle of about 9.31 mrad (diameter) with a relative movement of 0.0728 mrad/s. The
so-called windows of transmittance of the atmosphere include the two most important
wavelength regions, leading to specific intensities of about 0.5 to 0.9 W/m 2 and per
nanometer (W/m 2nm) in the 850-nm spectral region and about 0.17 W/m 2nm in the
1550-nm region for direct sunlight on ground . All these figures are estimations for
clear sky conditions being subject to atmospheric influences, leading to variations.
In general, the impact of sunlight on FSO depends on the angular and the spectral
Optical networks, last mile accessand applications 283
sens itivity of the receiver and the possibility of direct sunlight at the receiver, referring
to orientation and mount of the installed system .
Attenuation is the most critical factor for longer FSO links. The contribution of
the free atmosphere is comparatively small at the most commonly used wavelengths
around 850 and 1550 nm. Values for specific attenuation between 0.2 dBlkm under
very clear atmospheric conditions up to about 10 dBlkm due to dust in urban regions
are reasonable.
Far more critical is the impact of the weather situation. It has been reported, that
specific attenuation may temporarily raise to more than 300 dBlkm in heavy fog, even
though these are very rare occasions depending on local climate (and within 3 years of
measurement have never been observed in Graz) . The situation can be explained by a
theory describing the interaction of electromagnetic waves and particles , which can be
simplified depending on the ratio of particle size and wavelength, to geometrical op-
tics for interaction with comparatively large particles causing wavelength -independent
absorption, and to Rayleigh-scattering for comparatively small particles . The most crit-
ical condition appears when both are approximately in the same order, such as optical
wavelengths in the order of I J.Lm and haze or fog that consists of water droplets
at diameters from about I to 15 J.Lm. This can be characterized by Mie's theory of
scattering.
However, many factors depending on the propertie s of the particles are not known,
therefore a different approach leads to a well known deterministic formula (I) based
on visibility, which can be used in practice . Attenuation caused by scattering can be
estimated by
as ~ ~ (5~5) O.195- S (I)
In this equation as is the specific attenuation in decibel per kilometer, S is the visibility
for human eyes (sight) in kilometers and A is the wavelength of transmitted light in
nanometers. Usually, the range of visibility is defined as a path of 2% transmission
in air (corresponding to about 17 dB attenuation at 555 nm wavelength), the contrast
resolution of the human eye. Records of visibility are available at airports or meteoro-
logical stations and may be used as one reference to estimate the probability of fog for
the local climate of a certain location. The impact of rain is less critical because of the
larger particle size in the order of 0.1 to about 5 mm diameter, which has more effect
on longer wavelengths such as millimeter waves at several tens of GHz . Falling snow
simply absorbs the light by the irregular shapes of particles in the size of about 2 up
to 25 mrn, leading to a varying attenuation depending on the relation of particle and
receiver optic s area. For the same visibility range the conditions for FSO are worse than
fog, although site diversity realized by using multiple transmitter optic s can improve
the quality of the link.
Light transmission in media follows the principle of Fermat, according to which the
way of light from one point to another follows the shortest optical path length, which
depends on geometrical distance and optical properties of the medium, given by the
fraction index n. Consequently the fraction index has an impact on light propagation.
For free air at sea level n is in the order of 1.0003, but depends on many factors, e.g.,
the wavelength used for transmission, the temperature, the atmospheric pressure and
the humidity of the air.
Different sheets of air being crossed by light lead to beam deviation (refraction of
light) due to a temperature or pressure gradient, an effect which leads to results similar
to misalignment or building sway, causing a reduction of received power which can

get critical for very small divergence.
Short-time fluctuations of n, so-called turbulence cells, lead to irregular changing
intensity in the beam, wave-front distortion (changing the angle of received light and
causing aspot oflight dancing around the focal point) and changing deviation, causing
a varying input power at the receiver, also known as scintillation. The largest gradients
and therefore strongest effects are observed near hot surfaces like streets, rooftops or
side walls, or over outlets of air conditions or in winter heaters.
The influence of the turbulence cells is not interesting for short ranges, but very
important for long distances (section 2.2). Type I and 2 systems will be cost effective
and they will have a high availability and reliability at the last mile area.
2.1.1. FSO in Combination with Satellite and Wireless LAN

As an example for short-range applications, the use of FSO for connecting a satellite-
Earth station to the backbone or to a local area network is described. This is a example
for a temporary installation and shows the flexibility of FSO systems.
The satellite videoconferencing system was used at a tele-medicine conference in
Vienna (Fig. 10), but it is also well suited for any conference, for tele-teaching and for
military issues. In Fig. II we can see the video presentation of a doctor in Graz and
experts at the conference in Vienna. In the lecture room in Vienna Wireless LAN was
available. In order to connect the Wireless LAN with the mobile satellite-Earth station
outside the UNO-City building, a FSO link was used.
2.2. Long Range Applications
For communication links, which run over distances above 2 krn, it is important to use
systems of type 3 technology. An example of a commercially available system is shown
in Fig. 12.
Another important installation is the Wallberg experiment [4) in Germany at the
German Aerospace Centre (DLR). DLR has performed an optical free-space data trans-
mission experiment along with the European Aeronautic Defence and Space Company
(EADS), Germany and Contraves Space AG, Switzerland. The scope of this experi-
ment is to verify the tracking capabilities of the OPTEL 02, a space-qualified optical
terminal, and to demonstrate optical high data rate transmission through the atmo-
sphere. Two laser diode transmitters at 980 nm, each mounted on a tripod for static
pointing and laterally separated by about 4 m, were placed on a mountain top in the
German Alps at a height of 1620 m. Either a pseudo-noise pattern or video data can
be transmitted. The OPTEL 02, performing acquisition and tracking, is situated at the
DLR site in Oberpfaffenhofen near Munich at 620 m. An APD receiver front end is
connected to an additional 75-mm telescope in order to receive data up to 270 Mbps.
The optical path length between both terminals is 61 km.
At Wallberg a building with two rooms with one window each, physically separated
by a distance of four meters, is at their disposal. Despite facing towards a popular foot
path, there are no problems with laser safety. The two separated windows allow the
use of a two-transmitter concept to reduce scintillation effects. Figure 13 shows a map
of the southern Munich area showing Wallberg and Oberpfaffenhofen including an
altitude profile for the optical link.
Optical networks, last mile access and application s 285
Fig. 10. Mobile satellite-Earth station with FSO Connection demonstrated at the UNO-City
Vienna.
Tele-Consultancy over Satellite

Combination of different wireless technolog ies
indoor relay statio n
Consu ltants outdoo r equipment Hospaal J Con ference

Locario nA Lo cat ion B
Fig. 11. Video conference connection between Graz (Location A) and Vienna (Location B) for
tele-medicine.
Fig, 12. DT-50/DT-IF156 CanoBeam.
Munich
- .J
Fig. 13. Locationand scenarioof the Wallberg experiment.
As can easily be seen in the profile, there is a steep slope from the Wallberg
mountain top to Lake Tegemsee on the plain. This gives a good reason to place the
transmitter on top of the mountain, as little atmospheric disturbance can be expected
there. The last few kilometers of the link are close to ground . North of Stamberger
lake, there is a marshland, which might cause severe beam disturbance .
The described link is a terrestrial FSO path for very long distance. Hence, the
beam divergence must be smaller than I mrad, resulting in low geometrical losses and
concentrating all incoming light on a small area around the receiver. To compensate
building sway and deviation, automatic tracking of the beam is necessary. Adaptive
optic helps to compensate atmospheric fluctuations. The path characteristic of the
Wallberg experiment is similar to space applications (section 2.3).
Optical networks, last mile accessand applications 287
2.3. Space Applications (Aircraft and Satellites)
The main advantages of FSO links compared to microwave links are small high gain
antennas (telescopes), light terminals, highest possible data rates at low signal power,
no interference with other transmission systems and tap-proof operation using coherent
methods . Possible applications for optical free-space links are inter-satellite links (lSL)
in satellite networks, links for deep space missions, links between unmanned aerial
vehicles (DAV) and high altitude platforms (HAP) and data links from geostationary
satellites (GEO) to earth ground stations.
Space applications are divided into FSO links in the troposphere (for example, up-
and down-links between the Earth and satellite) and FSO links above the troposphere
(e.g., optical inter satellite links). Links through the troposphere are mainly influenced
by the weather conditions similar but not equal to terrestrial FSO links. Optical inter
satellite links are not influenced by weather conditions, because satellite orbits are
above the atmosphere.
While for terrestrial optical wireless the most effort is given to compensate for
the local weather conditions, for space applications, the acquisition and tracking [7] is
the greatest challenge, because satellites and aerial vehicles have relative movements .
Hence, we can see that for these applications only the most expensive systems of type
3 can be used.
Regarding FSO links through the troposphere, there is furthermore a difference
depending on the communication direction. On the up-link , the FSO beam from the
Earth to the satellite, the divergence of the optical beam is increasing quickly , because
the refraction-index is decreasing with height and therefore the optical ray will bend
away from the normal. Hence, the light beam gets wider on its first part of the path
and is affected by the troposphere.
On the down-link, the FSO beam from the satellite to the Earth, the narrow beam
transmitted from the FSO unit onboard the satellite is influenced by the troposphere
on the last 10 to 15 km of its path. Because of the character and distribution of the
refraction index in the troposphere, the light is coming from the medium with lower
index and is going to the medium with higher index. Hence, the ray bends towards the
normal. These effects result in a larger beam diameter for the received up-link beam
compared to the received beam of the down-link, which has a smaller diameter.
An interesting European experiment is shown in Fig. 14, the world first data trans-
mission between satellites using laser light which took place at 22nd November 200 I.
A data link between satellites was established using a laser beam as signal carrier.
On board ESA's Artemis satellite is the SILEX [14,15] system . This system provides
an optical data transmission link with the CNES Earth observation satellite SPOT 4,
which is orbiting the earth at an altitude of 832 krn, while Artemis was temporarily in
a parking orbit at 31,000 km. Through the laser data link, images taken by SPOT 4
can be transmitted in real-time to the image process ing center in Toulouse, France, via
Artemis, thus drastically reducing the time between taking the picture and its delivery
to the Earth. This is possible whenever the two satellites are in line of sight. Without
the Artemis relay the images are stored on board in SPOT 4' s memory and dumped to
the ground stations at a later time.
The experiment performed consisted in establishing the link four times: in the
course offour successive SPOT 4 orbits , the SILEX terminal on board Artemis activated
its optical beacon to scan the area where SPOT was expected to be. When contact was
288 E. Leitgeb, M. Gebhart,and U. Birnbacher
Fig. 14. SILEX Experiment.
made, SPOT 4 responded by sending its own laser beam to Artemis . On receiving the
SPOT 4 beam, Artemis stopped scanning and the optical link was maintained for a
pre-programmed period lasting from 4 to 20 minutes.
During the period when the two satellites were "communicating", test data were
transmitted from SPOT 4 to the ground via Artemis at a rate of 50 Mbps. The extremely
high accuracy of the data stream was confirmed at ESA's test station in Redu (Belgium)
and the SPOT 4 receive station in Toulouse.
The main challenge in establishing an optical link between satellites is to point a
very narrow beam with extreme accuracy to illuminate the partner spacecraft flying
at a speed of 7000 mls. This experiment was performed under worst-case condit ions
since Artemis was not in its nominal geostationary position but in a lower parking
orbit, circling the Earth every 19 hours. The experiment was preceded by a series of
tests a week earlier, during which a link was established between Artemis and ESA's
optical ground station in Tenerife. Those tests demonstrated the correct operation of
the SILEX terminal and paved the way for the subsequent steps.
After moving Artemis to its final geostationary orbit at 36,000 km the operational
phase has started and the link between the two satellites is established at least 5 times
a day. The SILEX system consists of two terminals : one on board Artemis, the other
on SPOT 4.
For optical wireless space applications there have been developed coherent receiver
instead of the direct detection . The optical free-space links have to cover large distances
in space applications (inter-satellite links, inter orbit links, connections to deep space
probes) , the main criterion for selecting an appropriate transmission scheme is power
efficiency or the sensitivity of the receiver. Hereby coherent methods are superior to
direct detection methods , which are commonly used in terrestrial communications. In
particular the PSK homodyne receiver shows a very high sensitivity. The advantages
of a homodyne receiver against a direct detection concept are a higher sensitivity
and less problems with background light. One key element of the homodyne receiver
is the synchronisation of the local oscillator laser with the incoming signal using a
rather complex optical phase locked loop (OPLL) . If the optical link goes through the
atmosphere the atmosphere will cause wavefront distortions and fading of the incoming
light. These effects make the realisation of an OPLL very complicated and also reduce
the error performance in an direct detection system . So special techniques like array
detection, channel coding or diversity are studied [4].
3. Last Mile
Now the use of optical wireless for the last mile will be investigated and described
in more detail. Therefore line of sight, reliability and availability, different FSO tech-
niques for the last mile and multimedia applications are described. Results of various
studies will complete these investigations. Finally, an example for realizing a FSO
network for the last mile will be shown.
3.1. Line of Sight
Optical wave propagation through the atmosphere requires a free line of sight from the
location of the transmitting terminal to the receiving terminal. Even though receivers
for infrared transmission are more sensitive compared to the human eye in the visible
light region, the influence on the radiation is very similar. Especially for networking
applications, where high availability is essential, the weather influence is a key factor.
Visibility data, collected over several years either by free eye estimation or, more
accurately, with a transmissionmeter instrument at meteorological stations or at airports
as runway visibility range (RVR) can be used to calculate availability in free-space optic
network planning . Atmospheric transmission can be described by the Beer-Lambert
law:
T = P d = e --yd (2)
Po
The transmission T for an optical wavelength means the relation of the optical
power Pd at the end of an atmospheric path of distance d to the optical power Po
which was originally sent (Fig. IS). According to the Beer-Lambert law transmission
can be expressed by the extinction coefficient "( and the atmospheric path distance d.
The extinction coefficient "( is constituted by processes of absorption a and scattering
f3 by particles in the atmosphere. In detail, these particles can be the molecules of the
atmospheric gas constituents and the larger aerosol particles:
,,((,X) = a('x) + f3(,X) (3)

= am('x) + aa('x) + f3m('x) + f3a('x) .
For the wavelengths used in FSO, especially 850 and 1550 nm, atmospheric ab-
sorption can be neglected. Scattering by molecules f3m (,X), which can be described
by the Rayleigh scattering model, only get important for optical wavelengths smaller
than 400 nm. The main important process that remains is the scattering of light by
aerosol particles f3a('x) in fog. It can be described by the Mie scattering model for
single interactions between particles and waves.
By definition the visibility range V is the atmospheric path distance for a trans-
mission of 2%. This is expressed in the Koschmieder equation:
290 E. Leitgeb, M. Gebhart.and U. Birnbacher
~+
.·.+ifo+.+•.
...
.. • •
+ • • • • +.+ •
••• •
to.to.
. ..
+.+ •
.. +
.
..
-..
d
Fig. 15.Transmission according to the Beer-Lambert law.
In ( ~ ) In (0.t2 ) =3.912
V=--= - -. (4)
,550 nm , 550 nm , 550 nm
Sometimes a slightly different definition is used in the literature, defining the

visibility range at 5% transmission. This lead s to a factor of 2.996 instead of 3.912 .
Reforming this equation allows to calculate the extinction coefficient for 550 nm, at
the center of the visible light range .
To adapt visibility data for infrared wavelength transmission, the wavelength de-
pendency of the scattering process has to be taken into account. A first relation based
on empirical measurement data was proposed by Kruse [5]:
(5)
The exponent q in Eq. (5) depends on the visibility distance range . To include also
low visibilities in den se fog, the original Kruse equation was modified by Kim [3, 5]
to
1.6 for V > 50 km
1.3 for 6 km < V < 50 km
q= 0.16V + 0.34 for 1 km < V < 6 km (6)
1 o
V - 0.5 for 0.5 km < V < 1 km
for V < 0.5 km.
In general it is problematic to find a relation which allows the recalculation of
visibility data from 550 nm to longer wavelengths. Different types of fog can cause
different attenuation for longer wavelengths at the same visibility range , actually de-
pending on the particle size distribution and the particle density according to the Mie
scattering model. But visibility data does not include information about the particles,
even rain and snow are included. Measurement data indicates that longer wavelengths
are less attenuated in haze and light fog, while there is no wavelength dependency in
dense fog. New theoretical calculations from Alnaboulsi for advection fog and convec-
tion fog indicate that shorter wavelengths are less attenuated at very short visibilities,
which is the most critical case for high available FSO.
Hence, even if visibility data should not be over- interpreted, it is a useful in-
formation for availability prediction. Probably the best compromise is to assume no
Optical networks, lastmileaccess and applications 291
wavelength dependency. Especially for short-range high availability applications this

seems to be the best approach, except for 10 J.L technology. It is convenient for technical
applications to calculate in decibels, so the extinction coefficient can be recalculated
to attenuation due to scattering :
(7)
Equation (7) considers attenuation aSCAT (in dB) over the full atmospheric path
distance d, while Eq. (8) considers the specific attenuation aS CAT ,SPEC (in dBlkm)
for a transmission threshold of 2%:
10 10 log U) "-J 17 dB
aS CAT ,SPEC = In(lO),(,X) = V = V- . (8)
3.2. Reliability and Availability
To introduce this topic, which is of high relevance for the practical use of FSO systems
in communication networks, the meaning of the expressions shall be explained.
System reliability R(T) is the probability that the system works correctly during
the time period T under defined environmental conditions .
System availability A(t) is the probability that the system works correctly at the
time t.
Because the FSO link just offers the physical layer for data transmission, the
conditions for correct operation of the system usually are defined by a maximum
tolerable bit error rate (BER) for the specified data rate (e.g., BER = 10- 9 for 100
Mbps). The bit error rate increases, if too much or not enough optical power is received .
In this sense, FSO systems usually are very secure by construction, if they can avoid
overload and if they simply switch off the connection to the network, when the received
optical power gets too low because of atmospheric attenuation , before too many bit
errors are produced. This threshold is also referred as the receiver sensitivity limit
PRS . Especially short-range systems can be described in this way and behave like a
wired connection, offering the specified performance over the full distance. For long
distance FSO connections using more collimated and more coherent light, fluctuations
in the received power get increasingly important. This makes it difficult to set the right
sensitivity limit and could also cause performance degradation by burst errors , which
can be avoided by channel coding and error correction techniques .
Related are safety issues , which do not actually refer to data transmission, but
in general to conditions for operation. To satisfy laser safety regulations according to
lEe/EN 60 825, especially to meet laser class I which certifies safety under all possible
conditions over long time, or the classes 1M, 2 (visible light) or 3A is important to
allow the operation without further restrictions . The main concern of laser safety is
the human eye, which focuses light coming through the iris aperture of max. 7 mm
diameter to the retina , so light intensity is actually measured . In the same eye safety
laser class , more power is allowed for longer wavelengths up to 1400 nm because of
transmittance and focus characteristics of the eye. Hence, 1550 nm technology has an
advantage against 850 nm here. More power is also allowed for light source s with a
higher divergence angle compared to point sources, because only point sources can be
concentrated in a verysmallspot on the retina.For this reason, lightsourceslike LEDs,

optics using diffusers, or multipletransmitteroptics havean advantage. In addition to
laser safety also the regulations for mounting the systems must be met, too.
The availability of an installedFSO link mainlydepends on the powerlink budget
and the local climate conditions, causing increased attenuation over periods of time.
The properties of an FSO system excluding the effects of the atmospheric path and
distance can be summarizedin the system power factor PSYS (in dBm) to
PSYS = PTX + GTx + ARX - l:)aTX + cnx) (9)
= PTX + 10 log (21r(1_4:os a/2)) + 20Iog(RA) - l:)aTx + cnx ).
In Eq. (9) the optical output power in decibels over one mill watts (dBm) is rep-
resented by PTx, the geometrical transmittergain in decibels is GTX (depending on
the full beam divergence angle a), the geometrical gain at the receiverdue to optics
aperture radius RA is A RX and losses in the optics of transmitter and receiver are
representedas aTX and aRX (dB). This allows calculatingthe powerreceived PRX in
distance d by
PRX = PSYS - DL (10)
= PSYS - 20 log ( 12~) .

Equation(10)includesthedecibelvalue D L of thegeometrical distance d in meters
andgivesthereceivedpower PRX withoutanyatmospheric influenceor pointinglosses.
The specific link margin MSPEC for an installation can be found with the receiver
sensitivity limit PRS for proper operation
1000mM
M SPEC = --d- (II)
1000m
= -d-[PsyS - D L - PRS] .
The basicconditionforsystemavailability A is thatthespecific linkmargin exceeds

the specificattenuation, which can be expressed by
L: aSCAT,SPEC
A= {~ for MSPEC 2:
for MSPEC < L: aSCAT,SPEC (12)
at the time t.
The appearance of fog can only be statistically described, usually as the steady-
state availability, which is approximated by measurements over long periods of time.
As an example, a one-year runway visibility range data set from Graz, Austria, is
shown in Fig. 16. In this data, point 3 represents the probability for a visibility of less
than 1500m, equal to a specific attenuation of 11.4dBlkm, which was approximately
5% of one year's time. The probability for visibilities less than 200 m indicated in
point I was 0.01%, which meansthat a FSO installation wouldreach 99.99% average
availability during this year if it has more than 85 dBlkm specific margin. Such link
availabilities and even more are possible if only the link distance does not exaggerate
a certain value.
o 500 1000 l500m

Visb ility d irtAn" On)
Fig. 16. Full year visibilitydata profilefor the city of Graz, Austria.
Based upon this climate example, it is possible to calculate availability contours

for specifie FSO systemscharacterized by their system power factor. Figure 17shows
this for two types of systems. Typeone is a cost efficientgeneral purpose type with a
comparatively large beam divergence and receiver acceptance angle, allowing quick
installation and easy mount.Type three is a sophisticated system with small beam di-
vergence andautomaticbeamtrackingspecializedfor longerdistances.Typicalsystem
properties may be between these two cases.
Type I !'So System

M(O) =62dB
Beamdiv. • 42 mred I , -; «
100 200 300 400 00 600 700 800 900 1000.P 3 00m
Type 3 FSO system
Link di stance
~O~-,,~O_~, M~"
99. 994 % 99.95 % 0 99.5 % 0 99.0 % 0 95 % 0 no cmnection
Fig. 17. Calculated availability contours for two FSO systems for the visibility data set of Fig.
16.
A practical measurement with a commercial FSO system was performed in [2,6]

over the same period of time as used in Fig. 16.The system used 850 nm wavelength
anda beamdivergence of2 mradforan installedlinkdistanceof2. 7 km. Multiplebeam
technology was implemented to reducescintillation effectsin theopticaldomainand to
increasesystem reliability. The system power figurewas 96 dBm and the specificlink
margin approximately 7 dBlkm. Figure 18showstheseasonalanddiurnaldependencies
of the unavailability of the link.
A clear dependence on the season can be seen, as it is typical for Central Europe
and NorthAmerica, for example. The majorcause of unavailability was fog, especially
convection or radiationfog, whichappearsat theendof thedayandduringnight,caused
by ground cooling due to radiation, and is resolvedby sunlight during the day.This is
the cause for the daytime dependency.
The performance of the system was very good; at 155 Mbpsdata rate the monthly
averaged bit error rates (BER) reached values between 10- 11 and 10- 13 . A one-year
follow-up trial [II] of the same optical link combined with 40 GHz technology with
approximately 2.5 dBlkm specific margin did result in 99.926% reliability for the
lClI'S
:J5,am i
31l1'S ~
:!5m'1 ~
~
3Ill'S .!;
,
~
ji
Fig. 18. Seasonal and diurnal dependency of measured unavailability of a 850-nm FSO system
at 7 dB/km specificmargin in Graz,Austria.
combination of optical and millimeter wave, a so-called hybrid wireless link, while
the availability of the millimeter wave link was 97.989% and for the FSO system
96.813% during this year. A short range FSO system with 70 dB/km specific margin
did show an availability higher than 99.96% in the winter season during this time.
For practical applications this means that broadband and high available wireless
links can either be reached over short distances with optical links only, or over longer
distancesby thecombinationof veryhighfrequency andopticallinks. Ifthe application
is not time critical or does not require high availability (like, for example, remote data
storage),additional FSOlinksoverseveral kilometerdistancescan increasethe network
throughputdramatically.
Reliability refers to the inner propertiesof the system in first place, which can be
expressed by the mean time-to-failure (MITF) , or the FIT rate actually. 1 AT means
the probability for a deviceto fail in 109 hoursof operation. In FSO systemsespecially
the transmitting elements have a limits of lifetime, dependingon the conditions under
which they are operated. In general, the output power reduces gradually, and finally
the device will fail. Major manufacturers qualify their productsand providestatistical
data for the performance over time. VCSELs and LEOs havea typical lifetimeof 105
to 108 hours of operation. Providers of fiberoptic networks state that typically 1-2 of
1000 media converters can fail within a year. Operatingthe devices below the output
power limit and avoiding extreme temperature variations can improve the lifetime a
lot.
3.3. Different FSO Techniques for the Last Mile
In sections 1and 2 we haveseenthedifferenttypesof FSOsystems, theircharacteristics

and their advantages. For the last mile FSO systems with light emittingdiodes (LED)
as source are a well suited solution,becausethey are cheap and data rates in the range
Optical networks. last mile access and applications 295
of tens of Mbps are sufficient. Using LEOs has the advantage that problem s with laser-
and eye safety are minimized . Low-cost FSO systems using LEOs instead of laser
diodes have been developed by [1,9] for data rates of 10 and 100 Mbps .
The developed systems combine available standard components to realize cost-
effective solutions. The beam divergence usually covers values from about 8 to 60
rnrad, allowing an easy alignment for the user without the need of a telescope. All
elements including the optics do not require a precision as high as needed for other
approaches. allowing higher tolerances at production . and the use of simple mounts .
Owing to large divergence. the requirements for a stable underground are not very
high allowing a quick installation. Suitable distances for high availability operation
are limited by the wide beam angle and depending on local climate to up to 300 m.
Owing to Laser Safety Regulations the use of source s with larger emitting area allows
more output power in the same safety class, which improves the link budget.
3.3.1, Small FSO System for 100 m
The developed FSO system for a range of 100 m consists of two main parts, one
transmitter (LED) and one receiver (photo-PIN-diode) . The LED and the photo-PIN-
diode are mounted in the focus of a cheap plastic lens encased by a tube of aluminium
alloy. The transmitter and receiver units are mounted on a printed circuit board in a
housing ofa video camera (Fig. 7A). On a single printed circuit board the electronic part
of the FSO system is located . The receiver electronic includes the photo-PIN-diode.
the amplifier, and the data-interface. For the transmitter electronic, the data-interface,
the driver amplifier and the LED are necessary. The beam divergence of this system is
about two degrees.
Currently, solutions are available for 10 and 100 Mbps. The data-interface converts
the signal from the electrical domain into fiber or RJ-45 (Ethernet or Fast-Ethernet).
By using VCSELs instead of LEOs and three transmitting units instead of one.
the range of the system can be increased . The measured voltage of the receiver unit
is shown on an analogue voltage display (integrated at the back of the system) . The
installation of this FSO system is very easy, because of the beam divergence of two
degrees and the above mentioned voltage display.
3.3.2. FSO System for 300 m
The developed FSO system for a range of 300 m is mounted in a plastic housing (Fig.
7B). The FSO unit consists of 8 transmitters (LED) and one receiver (photo-PIN-
diode) . The LEOs are also mounted in the focus of a cheap plastic lens encased by a
tube of Aluminium alloy.
The received light is focused by a large lens to the photo-PIN-diode. The electronic
part on the printed circuit board is similar to the small system for the lOO-m range.
The data-interface converts the signal from the electrical domain into fiber or RJ-45
(Ethernet). By using VCSELs instead of LEOs, the range ofthe system can be increased
up to 800 m.
3.4. FSO network/or a Small City and Multimedia Applications
The use of FSO is shown for selected scenarios and some multimedia applications (like
video-conferences, live TV-transmissions, etc.) will illustrate the range of applications
for FSO last mile networks.
3.4.1. Multimedia Applications
In this section some examples for temporary and nomadic use of FSO are described .
Nomadic Use of FSO at the Folk-Festival "Aufsteirern" A real service demon-

stration for nomadic use of optical wireless networks took place at the folk-festival
"Aufsteirern 2002" in the historical town of Graz. Folk dancers and brass bands of the
whole country of Styria came together, playing at at different places and streets. Free
Space Optic links were used to transmit live video pictures (with a data rate of 10 Mbps)
from two locations ("Landhaushof' and "Tummelplatz") to the central enquiry kiosk
located at "Herrengasse". Thereby, the visitors of the festival could inform themselves
what was going on at the other locations. In Fig. 19 the setup for this demonstration is
shown in the city map of Graz . The transmission from the location Tummelplatz was
realized with the use of an optical repeater, because there was no line of sight between
Tummelplatz and the enquiry kiosk. For this purpose two systems (for 300 m range)
have been interconnected. The transmission from the Landhaushof was implemented
with a small FSO (l00 m) system .
Fig. 19. Locations Landhaushof and Tummelplatz shown in the city map of Graz (Event Auf-
steirern) .
Hybrid Wireless Networks-High Availability with Combined Optical/Microwave

Links The availability of FSO links is limited by weather patterns like fog and heavy
snowfall. Microwave-based communication links operating at high frequencies (40-43
GHz) offer comparabledata rates and need line-of-sight as well. Link availability for
microwave systems is limited by heavy rain. Combining FSO links with microwave
links within a hybrid FSO/microwave communication network has the advantage of
added redundancy and higherlinkavailability. Measurements overa periodof one year
show a combined availability of99,93
Discussing the individual advantages and disadvantages of FSO and radio com-
munication technologies, the idea was born to combine optical and microwave links
for introducing redundancy and to achieve higher availability. Therefore a measure-
ment scenario has been set up in Graz, Austria to compare the link availability of a
microwave and a FSO link between the same locationsinfluenced by the same weather
conditions. The distance of this redundant link is 2.7 km and the two locations are in
lineof sight.Graz is locatedin an area wherevariousweatherconditionsappear and the
climatic zone is very problematic for optical and microwave communication systems.
Microwave and FSO communication links (Fig. 8) have roughly the same proper-
ties regarding offered data rates and flexibility of setup, but operate under different
conditions, with their benefits and challenges.
The measurement result is drawn in Fig. 20 for each month. The dotted line indi-
cates the optical link. Obviously the optics is highly influenced by fog and snow fall,
which can be seen in December. In contrast the microwave link does not have exces-
sive losses during this period,and the hybrid reaches nearly 100%. A similar situation
appearsin January and February. During this time snow fall, fog and rain occur, which
can be seen in the low availability of both links. Nevertheless, the hybridhas very high
availability. This can be seen as proof, that the systems havenearlyorthogonalbehav-
ior.From Marchto July bothsystems haveapproximately the average availability. The
next interesting section is the summerwith heavythunderstorms. In the observedyear
this happened in August, where both systems do not have excessive bad availability,
but the hybrids availability decreases to 99.28%. The reason for this behavior is the
loss of the optical link during heavy thunderstorms. A month later nearly the same
single availability is obtained, but the hybrid has again a high availability.
The big advantageof the hybridis the fact, that systemswitha single systemavail-
abilityof97% can be improved to 99.926%,whichis identical witha gainenlargement
of transmitting power of 12dB in the microwave link case.
3.4.2. FSO Network for a Small City
Using FSO technology a small network, offering various applications can be setup.
The architecture and possible applications are described in the following.
For high reliability, the optimum network architecture is a meshed network, be-
cause it combines the advantages of the above described architectures. Differentcon-
nections are possible, of which two examples are described. In Fig. 21 an expanded
version of a ring architecture is shown. The installation of additionally links (thediago-
nal dashedlines in Fig. 21) increasesthe reliability and the securityagainst failure. For
example, a connectionfailurein building2 (in the worstcase a total system breakdown
of all FSO links of building 2) does not cause a disconnection of the other buildings.
An information flow from building I to building 3 can be realized in different forms.
The information can be sent in the other direction(counter-clockwise) of the ring net-
work (buildings I, 6, 5, 4, 3), as shown in Fig. 3. A further routing possibility is from
100,0000
98,0000
--J----{
·····w
[J
- ",.........., '"'""'""1J-:-'I(~:~~~-~
/-.~~~ . c ~ .~:-;
~<-TD ~ ~ ,\[J~.?
96,0000
. .
..
"
I
0
94,0000
92,0000
0 . .. '"]
I
.. 0---
I
90,0000 , ,
,
88,0000
-:-Microwave 408Hz overall availability: 97.989 , I
86,0000 ,
, I
84,0000
82,0000
~ LJ .Freespace optics overall availability: 96.813
-L--Hybrid overall availability: 99.926
,
. ,
I
80,0000 ' I i
~" ~'),. ~'),. ~'),. ~'),.. ~'),. ~'),. ~'),. ~'),. ~'),. ~'),. ~'),. ~'),. ~":>
;t.. .. e- ~ . .. ~ . ~. '8-' ~. ,so. ~. ~. ~. ~. ;t.. . ~.
<;)0 ~flj «0 ~flj ~ ~ ~'5 '$ ,,?-'5 00 0 ~o <;)0 ~flj
Fig. 20. Availability of the separated linksand the hybrid.
building lover g 4 to 3 or from building lover 6 to 3. Hence, more security against

failure can be achieved and a network with high reliability can be realized.
In Fig. 22 a mixture of a ring and a star network is shown. In regard to reliability,
shorter distances are advantageous, because the optical multipoint unit is located in
the center of the area. In comparison to Fig. 21 the optical multipoint unit is used
as additional active optical repeater. If there is more than one optical multipoint unit
connected to the backbone a partial security against failure can be achieved.
Both examples of meshed FSO networks have shown advantages in regard to ring
or star configurations. Meshed networks combine the benefits of the above described
architectures .
In each configuration, the central FSO unit can be connected to satellites, direc-
tional radio links, (mobile) telephone networks, or fiber optics . In Fig. 5 a connection to
a fiber-based backbone realized with a Point-to-Multipoint architecture is shown. The
optical multipoint unit is connected with a switch or router to the backbone network .
The clients in the buildings 1,2,3 and 4 are linked with their FSO units to the Optical
Multipoint. This solution is similar to the FSO network installed at the Department of
Communications and Wave Propagation at the TV Graz [I] .
A wireless optical access technology allows quick deployment like a RF cellular
network, combined with the good performance of an optical technology . The most
critical aspect is the weather dependency and the reliability of the links, but as shown
in the previous section, this is not a problem if only the point-to-point distance is not
too long. FSO technology can offer high data rates very cost effective, and allows
implementing the same networking protocols as fiber optic links.
If the network is not just offering transmission capabilities at the physical layer, but
is used by a network operator to offer services to the customers, higher layer protocols
have to be considered. Nowadays, most applications can already be offered of the
Internet, hence all-IP networks seem to be the choice of future access networks .
Building 5
Legend ... Free Space

Optical - Link
.. - -~ Redundant FSO - Link

• 0 pticaI Repeater
Fig. 21. Optical wireless (meshed architecture, solution A).
In general, a typical access network consists of a central office, the network in-
frastructure and the customer premise's equipment. At the central office, servers are
needed to offer the services to be provided, for example, a mail server, servers for
Web-service or externaldata storage. Access to the Internet requires an Internet router
and a firewall. Furthermore, a router/switch is needed in order to connect the incom-
ing links from the users to the backbone network. Additionally, software is needed to
manage and control the network, for example software tools for user administration
and billing. A typical structure for the central station is shown in Fig. 23.
Regarding the data rates of the network links to the end users, the applications
and services have to be considered and will be in the range of I to 100 Mbps. If for
example, TV-broadcasting services are offered, the link bandwidth has to allow for
the simultaneous transmission of up to three TV-program streams, this means three
data streamsbetween2 and 7 Mbps,dependingon the imagecontent and compression
format.
Building 5
Free Space
Legend ... Optical - Link
• Optical Repeater
OpticaI MuIti-point Unit
Fig, 22. Optical wireles s (meshed architecture, solution B).
For the deployment network the implementation of a broadband technology is

essential. To make the network attractive, it should not only satisfy today's needs, but
leave room for future expansion. On the other hand, the affordable costs per user are
very limited, for TV and 10 Mbps networking services the break-even point which
customers were willing to pay was around 70 Euro per month in 2002.
For best performance, the solution of choice for an access network is optical
technology, especially if the network provider is an emerging company and does not
yet have infrastructure like the Telecom companies . Thereby networks can be built
using FSO links only, or fiber and FSO links can be combined in order to satisfy the
bandwidth demand of broadband application s like TV broadcasting .
4. Summary
Optical wireless is an excellent broadband solution for connecting end users to the
backbone (last mile access). This technology should be seen as supplement to conven-
Optical networks, last mile acces s and applications 301
Internet POTS / ISDN

public phone network
I
ISP
Inte~t-Rou ter web-server Vol P~Gateway

Flrewall l NAT el)lli il-Server (lefephone)
1// ..,Video-Server
/
/
- ---<-- IP Core Network (Gb E ) -"'-:/,.-------~'---_;r_---
/
/ 7
/
/
/
I IIII ~
Iilll 11111
11111 11111 ~
Management Router I SWItch
Workstation 100BaseFX Ports
Fig. 23. Central office network structure.
tional radio links and fiber Optics. The use of low-cost FSO systems for short distances
makes this technology interesting for private users.
At the moment the main work in this field is to increase reliability and availability.
Those two parameters of the FSO link are mainly determined by the local atmospheric
conditions. Good reliability and availability can be achieved by using the FSO for
short distances, by calculating enough link-budget and by using the optimal network
architecture for each FSO application .
The optimal solution for FSO configurations is a meshed architecture. This network
architecture combines shorter distances and high reliability, because of the location of
the optical multipoint unit in the center of the network . For increa sing the reliability
and availability it is also necessary to perform field-tests with FSO systems regard-
ing the local atmospheric conditions. Models for propagation and for predicting link
availability in different climate zones could improve the installation of this technology .
The combination of FSO and microwave links is also a further possibility for in-
creasing reliability and availability, because terrestrial FSO is most effected by fog,
whereas the microwave propagation is mainly influenced by rain. Within parallel stud-
ies, wireless hybrid (optical /microwave) links have been evaluated at the Department
of Communications and Wave Propagation [I , II] . First results show a reliability of
99.9991 % for hybrid systems .
References
I . E. Leitgeb, J. Bregen zer, M. Gebhart, P. Fasser, A. Merdonig, "Free-space optics : Broadband

wireless supplement to fiber network s", Proceedings SPIE, Vol. 4975 -07 (2003) , Janner
2003, San Jose , USA
2. E. Leitgeb, M. Gebhart, P. Fasser, J. Bregenzer, J. Tanczos, "Impact of atmospheric effects

in free-space optics transmission sytems", Proceedings SPIE, Vol. 4976 -28 (2003) , Janner
2003, San Jose, USA
3. I. Kim, B. McArthur, E. Korevaar, "Comparison of laser beam prop. at 785 nm and 1550
nm in fog and haze for optical wireless communications", Proceedings SPIE , 4214, 26-37
(2001)
4. F. David, D. Giggenbach et aI., "Preliminary results of a 61 km Ground-to-Ground Optical
IMIDO Data Tramsmission Experiment ", Proceedings SPIE, Vol. 4635 (2002)
5. H. Sizun, M. Alnaboulsi , O. Veyrunes, O. Bouchet, "Free-space optical communication
links, Bibliographical study and experimentation", European workshop on integrated radio-
communication systems , Angers, France (5/2002)
6. E. Leitgeb , M. Gebhart, P. Fasser, "Reliability of Free Space Laser Communications-
Investigations at the TU Graz", Proceedings of the 8th Annual WCA Technical Symposium,
14-16 Jan. 2002, San Jose, CA, USA
7. P. J. Winzer, W. R. Leeb, "Space-borne optical communications-a challenging reality",
Proceedings of the 15th Annual IEEE /LEOS-Meeting Nov. 2002, Glasgow
8. G. S. Mecherle, "Active Pointing for Terrestrial FSO" , Proceedings of the 15th Annual
IEEE/LEOS-Meeting Nov. 2002, Glasgow
9. E. Leitgeb, J. Bregenzer, P. Fasser, M. Gebhart, "FSO -Extension to fiber-Networks for the
' Last Mile' ", Proceedings of the 15th Annual IEEE/LEOS-Meeting Nov. 2002, Glasgow
10. H. Willebrand, B. S. Ghuman , FSO, ISBN 0-672-32248-X (SAMS , 2001)
I I. W. Kogler, P. Schrotter, U. Bimba cher, E. Leitgeb, O. Koudelka, "Hybrid Wireless Net-
works-High Availability with combined Optical/Microwave links", Proceedings of the
Conference of Telecommunications and Mobile Computing (TCMC03) April 2002, Oraz
12. E. Kube, "Nachrichteniibertragung mit Lichtstrahlen in der Atmosphare", VEB VerlagTech-
nik, Nachrichtentechnik Heft 19, Seite 201-207 (1969)
13. J. R. Barry, "Wireless Infrared Communications", ISBN 0-7923-9476-3 (Kluwer Academic
Publisher s, 1994)
14. Webpage ESA (Dec. 22nd 2003)
http://www.esa.intlexportlesaCPIESASOBZ84UC-index_0.html
15. Webpage OGS (Dec. 22nd 2003) http://www.iac .es/gabinete/oteide/ogs/20gs.html and
http://www.iac.es/gabinete/iacnoticias/I-2ool/47.pdf
© 2007 SpringerScience + Business Media Inc.
DOl: 1O.1007/sI0 297-006-0077-1
Originally published in 1. Opt. Fiber. Commun. Rep. 4.363--405 (2007)
Communication techniques and coding for atmospheric

turbulence channels
Xiaoming Zhu! and Joseph M. Kahn 2
1 QUALCOMM Inc., 675 Campbell Technology Parkway

Suite 200, Campbell, CA 95008
E-mail : xzhu@qualcomm.com
2 Dep artm ent of Electrical Engineering, Stanford Universit y
372 Packard Building, 350 Serra Mall
Stanford , CA 94305-9515
E-m ail: j mk@ e e . stanford. e du
Abstract. In free -space optica l co mmunication link s, atmo spheri c turbulence cau ses
fluctuations in both the inten sity and the phase of the recei ved light signal, imp airing
link performance. In this paper , we describe severa l communication techniques to
mit igate turbulence-induced intensity fluctuation s, i.e., signal fading . The se techniques
are applicable in the regime in which the receiv er aperture is smaller than the correlation
length of the fading, and the observation interval is shorter than the correlation time of
the fading. We assume that the receive r has no knowledge of the instantane ous fading
state . The techniques we con sider are based on the stati stical properties of fading, as
functions of both temporal and spatial coordinates. Our approache s can be divided into
two categories : temporal domain techniques and spatial domain techniques.
In the spatial domain techniques, one must employ at least two receivers to collect
the signal light at different positi ons or from different spatial angle s. Spatial diver sity
recepti on with mult iple recei vers ca n be used to overcome turbulence-induced fading.
When it is not possible to place the receivers sufficiently far apart, the fading at dif-
ferent receivers is correl ated , redu cing the diversity gain . We descr ibe a ML dete ction
techn ique to reduce the diver sity gai n pen alty ca used by such fadin g correl ation.
In the temporal domain techniques, one empl oys a single receiver. When the re-
ceiver know s onl y the margin al statistics of the fading , a symbol-by-sy mbol ML dete c-
tor ca n be used to optimize perform ance. When the receiver also knows the temp oral
correlation of the fadin g, maximum -likelihood seque nce detection (MLS D) can be em-
ployed, y ieldi ng a further perform ance impro vement, but at the cos t of very high co m-
plexity. We descri be two reduced- compl exity impleme ntations of the MLSD, which
make use of a single-s tep Markov chain model for the fading co rrelation in co njunc-
304 XiaomingZhu and Joseph M. Kahn
tion with per-survivorprocessing. Next, we also investigate the performance of using

error-control codingand pilotsymbol-assisted detectionschemesthroughatmospheric
turbulence channels.
1. Introduction
Recently, free-space optical communication has attracted considerable attention for

a variety of applications [1-8] . Because of the complexity associated with phase or
frequency modulation, current free-space optical communication systems typically
use intensity modulation with direct detection (IMIDO). However, in practice, the
performance of free-space optical communication systems can be degraded by many
effects, such as fog, obstruction of the line-of-sight path, atmospheric turbulence and
the nonideal characteristics of optical transmitters and receivers. In this chapter, we
focus on communication techniques and coding schemes to counter the degradation
caused by atmospheric turbulence in IMIDO links.
Atmospheric turbulence can degrade the performance of free-space optical links,
particularly over ranges of the order of I km or longer. Inhomogeneities in the temper-
ature and pressure of the atmosphere lead to variations of the refractive index along
the transmission path. These index inhomogeneities can deteriorate the quality of the
image formed at the focal plane of an imaging receiver, and can cause fluctuations in
both the intensity and the phase of the received signal. These fluctuations can lead to
an increase in the link error probability, limiting the performance of communication
systems . Aerosol scattering effects caused by rain, snow and fog can also degrade the
performance of free-space optical communication systems [9, 10j, but are not treated
in this paper.
Atmospheric turbulence has been studied extensively, and various theoretical mod-
els have been proposed to describe turbulence-induced image degradation and intensity
fluctuations (i.e., signal fading) [11-17] . Two useful parameters describing turbulence-
induced fading are do, the correlation length of intensity fluctuations and TO, the cor-
relation time of intensity fluctuations. When the receiver aperture Do can be made
larger than the correlation length do, then turbulence-induced fading can be reduced
substantially by aperture averaging [17]. However it is not always possible to satisfy
the condition Do > do. In this paper, we will introduce communication techniques
to mitigate intensity fading in IM/OO links, in the regime where Do < do. At the bit
rates of interest in most free-space optical systems, the receiver observation interval To
during each bit interval is smaller than the turbulence correlation time TO. Throughout
this paper, we will assume that Do < do and To < TO.
The techniques we consider in this paper are based on the statistical properties of
turbulence-induced signal intensity fading, as functions of both temporal and spatial
coordinates. Our approaches can be divided into two categories : temporal domain
techniques and spatial domain techniques .
In the spatial domain techniques, one must employ at least two receivers to collect
the signal light at different positions or from different spatial angles [6]. To maximize
the diversity reception gain, the multiple receivers should be placed as far apart as
possible , so that the turbulence-induced fading is uncorrelated at the various receivers.
In practice, however, it may not always be possible to place the receivers sufficiently
far apart. Hence, in this chapter, making use of the spatial correlation of turbulence-
induced fading, we derive the optimal ML detection scheme for correlated spatial
diversity reception [6).
In the temporal domain techniques, one employs a single receiver. If the receiver
has knowledge of the marginal fading distribution, but knows neither the temporal
fading correlation nor the instantaneous fading state, a maximum-likelihood (ML)
symbol-by-symbol detection technique can be used. If the receiver further knows the
joint temporal fading distribution, but not the instantaneous fading state, the receiver
can use a ML sequence detection (MLSD) technique [7)[8). Error-control coding [18)
and pilot-symbol assisted detection schemes [19) can also be applied in free-space
communication systems to mitigate turbulence-induced fading.
We note that the statistical properties of turbulence -induced signal intensity fading,
as functions ofboth temporal and spatial coordinates, can also be applied to study space-
time codes [19). In this paper, however, we will not cover the design of space -time
codes and coded modulation schemes for free-space optical communications.
The remainder of this paper is organized as follows. In section 2, we review the
theories used to model atmospheric turbulence, and use them to derive the spatial and
temporal coherence properties of the optical field in weak turbulence channels. We
then use the coherence properties obtained to derive the joint probability distribution
of the light intensity received on turbulence channels, characterizing the correlation
of signal fading over both space and time. In section 3 we first use the marginal
distribution of fading to derive a ML symbol-by-symbol detector for systems using
on-off keying (OaK). We then use the joint temporal distribution of fading to derive
a MLSD for OaK. In section 4, we introduce the spatial-diversity reception scheme,
where the instantaneous fading state at each receiver is unknown . With knowledge
of the joint spatial distribution of fading, ML detection can help to mitigate the di-
versity reception gain penalty caused by correlation among multiple receivers . We
also numerically demonstrate that in the dual-receiver case, this ML detection scheme
has better performance than conventional equal-gain combining (EOC). Then, we de-
scribe the temporal domain techniques in section 5. As shown in section 2, the high
computational complexity of the MLSD makes it impractical for most applications.
Hence, in section 5, we first describe two reduced-complexity implementations of the
MLSD , which make use of a single-step Markov chain model for the fading correla-
tion in conjunction with per-survivor processing . We then introduce the pilot-symbol
assisted detection (PSAD) scheme . We experimentally demon strate the effectiveness
of these two temporal domain techniques in a 500-m terrestrial link using OaK, where
MLSD and PSAD yield signal-to-noise ratio gains of 2.4 and 1.9 dB, respectively. In
section 6 we study coding for free-space optical communication systems and derive
an error-probability bound for coded OaK free-space optical communication through
atmospheric turbulence channels. In section 7, we summarize the conclusions of this
paper.
2. Modeling of Optical Communication through Atmospheric Turbulence
In this section, we first review the theories used to model atmospheric turbulence. We
then use these to derive the spatial and temporal coherence properties of the optical
signal in weak-turbulence channels. Finally, we present the joint spatial and temporal
distribution of turbulence-induced fading.
306 Xiaoming Zhu and Joseph M. Kahn
2./. Modeling ofAtmospheric Turbulence
Atmospheric turbulence can be physically described by Kolmogorov theory [11-14].

The energy of large eddies is redistributed without loss to eddies of decreasing size
until finally dissipated by viscosity. The size of turbulence eddies normally ranges
from a few millimeters to a few meters, denoted as the inner scale Lo and the outer
scale Lo, respectively. We can express the refractive index as n(r, t) = no + n1 (r, t),
where no is the average index and n1 is the fluctuation component induced by spatial
variations of temperature and pressure in the atmosphere. The correlation function of
n1 is defined as
(I)
r
Setting tt = tz in Eq. (1), we obtain n 1 (rr . r 2), which describes the spatial
coherence of the refractive index. To study the spatial coherence of the refractive
index, many models have been proposed, which assume exponential, Gaussian or other
solvable function forms for r n 1 (rr , r2) . We define the wavenumber spectrum Pn (k)
to be the spatial Fourier transform of r n 1 (r1 ' r2) . A widely used model with good
accuracy was proposed by Kolmogorov, which assumes the wavenumber spectrum to
be
(2)
Here, C n is the wavenumber spectrum structure parameter, which is altitude-dependent.
Hufnagel and Stanley gave a simple model for C; [13] :
C~(z) = K OZ - 1 3
/ exp( -z/zo) , (3)
where Ki, is parameter describing the strength of the turbulence and zc is effective
height of the turbulent atmosphere. For atmospheric channels near the ground (z <
18.5 rn), Cn can vary from 10- 13 m- 2/ 3 for strong turbulence to 10- 17 m- 2/ 3 for
weak turbulence, with 10- 15 m- 2/ 3 often quoted as a typical "average" value [12].
Other models and recent measurements of the vertical profile of C n can be found in
[14,15] .
2.2. Spatial and Temporal Coherence of Optical Signals through Turbulence
To describe spatial coherence of optical waves, the so-called mutual coherence function
(MCF) is widely used [12]:
(4)
where u(P, t) is the complex optical field and r is the space vector at P . Setting t1 = t 2
in Eq. (4), we obtain the spatial MCF r(r1 ' r2) . The Rytov method is frequently used
to expand the optical field u(r) :
uo(r) = A(r) exp[i4>(r)] = uo(r) exp(P), (5)
where uo(r) is the field amplitude without air turbulence :
uo(r) = Ao(r) exp[i4>(r)] . (6)

Communication techniquesand codingfor atmospheric turbulence channels 307
The exponent of the perturbation factor is
<PI = 10g[A(r)/Ao(r)] + i[¢(r) - ¢o(r )] = X + is , (7)

where X is the log-amplitude fluctuation and S is the phase fluctuation . We assume X
and S be homogeneous, isotropic and independent Gaussian random variables . This
assumption is valid for long propagation distances through turbulence.
In order to characterize turbulence-induced fluctuations of the log-amplitude X,
we use the log-amplitude covariance function
Since the random disturbance is Gaussian-distributed under the assumption of weak

turbulence, we can use the Rytov method to derive the normalized log-amplitude
covariance function for two positions in a receiving plane perpendicular to the direction
of propagation [17]:
(9)
where d12 is the distance between Hand P2. We define the correlation length of
intensity fluctuations, do, such that bx (do) = e- 2 . When the propagation path length
L satisfies the condition lo < y)L < £0, where>' is the wavelength and lo and £0
are inner and outer length scales, respectively, do can be approximated by [13]:
do ~ ill. (10)
In most free-space optical communication systems with visible or infrared lasers and
with propagation distance of a few hundred meters to a few kilometers, Eq. (10) is
valid. We note that aerosol forward scattering can further degrade the coherence of the
optical field and thus affect the correlation length. In this paper, however, we focus
only on atmospheric turbulence effects .
Atmosphere turbulence also varies with time and leads to intensity fluctuations that
are temporally correlated. Modeling the movement of atmospheric eddies is extremely
difficult, and a simplified "frozen air" model is normally employed, which assumes
that a collection of eddies will remain frozen in relation to one another, while the entire
collection is translated along some direction by the wind. Taylor's frozen-in hypothe sis
can be expres sed as [14]:
(II)
where v is the velocity of the wind, which has an average u and a fluctuation v f . If
v f is negligible and u is transverse to the direction of light propagation, then temporal
correlation becomes analogous to spatial correlation; in particular, the correlation time
is TO = do/ u. Assuming a narrow beam propagating over a long distance, the refractive
index fluctuations along the direction of propagation will be well-averaged, and will
be weaker than those along the direction transverse to propagation. Therefore we
need only consider the component of the wind velocity vector perpendicular to the
propagation direction Ul-. The turbulence correlation time is therefore
do
TO= - . (12)
U l-
200 300 400 500 600 700 800 900 1000

Propagation Distance L (m)
Fig. 1. Standard deviation of the log-amplitude fluctuation versus propagation distance for a
plane wave.
2.3. Probability Distributions ofTurbulence-Induced Intensity Fading
As discussed previously, when the propagation distance is long, log-amplitude fluctu-

ations can become significant. In this section, we will derive the statistical properties
of the log-amplitude fluctuations, which we refer to as "intensity fading" or simply
"fading". The marginal distribution of fading is derived in section 2.3.1, while the joint
spatial and temporal distribution of fading are derived in section 2.3.2.
2.3.1. Marginal Distribution of Fading
In this section, we derive the marginal distribution of fading at a single point in space
at a single instant in time . The marginal distribution is used in symbol-by-symbol ML
detection , which is discussed in section 3.1.
For propagation distances less than a few kilometers, variations of the log-
amplitude are typically much smaller than variations of the phase. Over longer prop-
agation distances, where turbulence becomes more severe, the variation of the log-
amplitude can become comparable to that of the phase. Based on the atmosphere
turbulence model adopted here and assuming weak turbulence , we can obtain approx-
imate analytic expressions for the covariance of the log-amplitude fluctuation X of
plane and spherical waves [14]:
7/61L
2
(TXlplane -
_
0.56 (T27f ) Cn(x)(L - x) 5/6dx ,
0
2
(13)
2 _
(Txl spherical - 0.56
(27f)7/6
T t'
Jo Cn( x)
2 (X) 5/ 6
L (L - x )
5/6
dx . (14)
Figure I shows the standard deviation of the log-amplitude fluctuation (Tx for a
plane wave, computed using Eq. (13), as a function of the propagation distance L. In
Fig. I, we again assume a wavelength of 529 nrn, and assume C~(z) to be constant.
Communication techniques and codingfor atmospheric turbulence channels 309
Fig. I shows that for propagation distances of a kilometer, a x varies from 10- 2 to I
for different values of C;.
Consider the propagation of light through a large number of elements of the atmo-
sphere, each causing an independent , identically distributed phase delay and scattering.
By the Central Limit Theorem, the marginal distribution of the log-amplitude is Gaus-
sian:
1 { (X - E[X] 2} (15)
fx(X) = (27rai )1/2 exp - 2ai .
The light intensity I is related to the log-amplitude X by
I = 10 exp(2X - 2E[X]) , (16)
where E[X] is the ensemble average of log-amplitude X.

From Eqs. (15) and (16), the average light intensity is
E[I] = E[Io exp(2X - 2E[X])] = 10 exp(2ai) . (17)
Hence, the marginal distribution of light intensity fading induced by turbulence is

log-normal:
f (I) = ~ 1 {_ [10(1) - 10(1oW} (18)

I 21 (27rai )l /2 exp 8ai .
2.3.2. Joint Spatial and Temporal Distributions of Fading
In this section, we derive the joint spatial and temporal distributions of fading. The joint
spatial distribution describes the fading at multiple points in space at a single instant
of time, and is used in section 4 in evaluating the performance of spatial diversity
reception . The joint temporal distribution describes the fading at a single point in space
at multiple instants of time. This distribution is the basis for the MLSD introduced in
section 3.2.
We assume that the log-amplitude at n receivers is described by a joint Gaus-
sian distribution. From Eq. (9), the auto-covariance matrix of the log-amplitude at n
receivers in a plane transverse to the direction of propagation is given by
(19)
where d i j is the distance between points i and j in the receiver plane. Based on "frozen-
in" model, and ignoring wind velocity fluctuations , Eq. (19) can also be modified to
describe temporal fading correlation by making the substitution
d ij = Ii - j !T u.L , (20)
where T is the time interval between observation s. In all that follows, we assume a
communication system using OaK, in which case, T is the bit interval. We denote the
covariance matrix of a string of n bits as
310 Xiaoming Zhu andJosephM. Kahn
(T~ (T~bx (~ do) 2 bX [(n-l)T

•• • (Tx -T-O- d0 j
c~=
(T~bx(~do) (T~ •••
2 b
(Tx X
(n-2)T d
-T-O- 0
(T2 b [(n-l)T d ] (T2 b [(n -2)T d ] (T~

X X TO 0 X X TO 0 n xn
(21)
From the previousdiscussion about spatial and temporal correlation (see section
3.2), the correlation time TO equals dO/u.l...
The joint distribution of intensity for a sequenceof transmitted On bits is
Similarly, when an On bit is transmitted, the joint distribution of intensity at nre-

ceivers is given by
... (~)]Cxl ~ ...

xexp
{ -k[ln(~) (~) ] } .
[ In (~)
(23)
3. Maximum-Likelihood Detection of On-Off Keying in Thrbulence Channels
In most practical IMiDD free-space optical communication systems usingOOK, the

receiver signal-tonoise ratio (SNR) is limited by shot noise caused by ambient light
muchstrongerthanthedesiredsignaland/or by thermal noisein theelectronicsfollow-
ing the photodetector. In eithercase, the noisecan usually be modeledto highaccuracy
as additive, white Gaussiannoise that is statistically independent of the desiredsignal.
Let T denotethe bit intervalof the OOKsystem,and assumethat the receiverintegrates
the received photocurrent for an interval To ~ T during each bit interval. At the end
of the integration interval, the resultingelectrical signal can be expressedas
r e = T/(Is + h) + n, (24)
where Is is the received signal light intensity and h is the ambient light intensity.
Both of thesequantitiescan be assumedto be constantduring the integration time. The
optical-to-electrical conversion coefficient is given by
(25)
where I is thequantumefficiency of the photodetector, e is theelectroncharge, A is the

signal wavelength, his Plank's constant,c is the speed of light.The additive noise n is
Communication techniques andcoding foratmospheric turbulence channels 311
white and Gaussian, and has zero mean and covariance N12, independent of whether
the received bit is Off or On.
In this section, we will describe symbol-by-symbol ML detection and compute its
error probability in the absence of error-correction coding. We will then describe the
MLSD.
3.1. Symbol-by-Symbol Maximum-Likelihood Detection
We assume that the receiver has knowledge of the marginal distribution of the
turbulence-inducedfading,but has no knowledgeof thechannel's instantaneousfading.
After subtraction of the ambient light bias "lIb, the signal r = r e - ryh is described by
the following conditional densities when the transmitted bit is Off or On, respectively:
P( rIOff) = 1 exp
V1rN (r- N
2
)
' (26)
P( rIOn) = I: P(rIOn ,X)fx(X)dX
1 [ (r - ryIoe 2X - 2E[X ] )2] dX,

f
oo
= c;;:rfx(X) exp - N (27)
- 00 V1rN
The optimal maximum a posteriori (MAP) symbol-by-symbol detector decodes
the bit s as [21,22]:
s = argrnaxP(rls)P(s) , (28)
where P(s) is the probability that a On bit or Off bit is transmitted. P(rl s) is the
conditional distribution that if a bit s(On or Off) is transmitted,a signal level r will be
received. If On and Off bits are equally probable, or if their a priori probabilities are
unknown, the symbol-by-symbol ML detector decodes the bit s as
s= argmax .Ptr ]s). (29)

s
The likelihood function is

_ P(rIOn) _ J OO [(r _ ryIoe2X- 2E[Xl)2 - r 2 ]
'\(r) - P(rIOff) - - 00 fx(X) exp N dX. (30)
In Fig. 2, we see that the likelihood ratio increases monotonically with r for
o :::;r :::; 1, so that ML detection can be implemented by simply thresholding the
received signal. In Fig. 3, we plot the optimal ML threshold for 0 :::; r :::; 1 versus the
log-amplitude standard deviation a x . We see that as a x increases, the ML threshold
decreases toward zero, because turbulence-induced fading increases the fluctuation
of the On-state signal level, while leaving the fluctuation of the Off-state signal level
unchanged. In Fig. 3, we also see that as the additive Gaussian noise covariance NI2
increases, the fluctuations of the Off and On state becomes more closely equal, and
the threshold increases toward 112.
The bit-error ratio (BER) of OOK can be computed as
Pb = P(Off)P(Bit Error lOff) + P(On)P(Bit ErrorDn) , (31)

10-4 '-----_~ _ _"'-_~_'___'_'__ _"___ __'

o 0.1 0.2 0.3 0.4 0.5 0.6
Nonnalized Signal Amplitude -E'
I
n [J
Fig. 2. Likelihood ratio versus normalized received signal amplitudefor differentvaluesof the
log-amplitude standarddeviation.
0.5 r ; ; ; : c - - - , - - - - - r - - - - - , - - - - - , - - - - ,
!]E[~ =1
0.2
o 0.1 0.2 0.3 0.4 0.5

Log-Amplitude standard deviation "X
Fig. 3. Normalized thresholdfor maximum-likelihood symbol-by-symbol detection versus log-

amplitudestandard deviation for different valuesof the noisecovariance.
where P(Bit ErrorlOft) and P(Bit Errorlfhi) denote the bit-error probabilities when
the transmitted bit is Off and On, respect ively. Without considering intersymbol inter-
ference, which can be ignored when the bit rate is not high and multipath effects are
not pronounced, we have
P(Bit ErrorlOff) = r JA(r»l

p(rIOff)dr, (32)
P(Bit Errorlf)n) = r p(rIOn)dr. (33)

J A( r) <l
3.2. Maximum-Likelihood Sequence Detection
The MLSD exploits the temporal correlation of turbulence-induced fading, and is

thus expected to outperform the symbol-by-symbol ML detector. For a sequence of
ntransmitted bits, the MLSD computes the likelihood ratio of each of the 2npossible
bit sequences s = [S1S2 . . . SnJ and chooses
s= argmax P(rls)
.§
=
r
arg~ax lx fx(X) exp - ~'
[ n (r _1JS!OeX i- 2ElXi]) 2]
'N dX . (34)
i
Here, each s, can take the value Off or On, so that S E {O, I} :
(35)
The complexity of MLSD is proportional to n 2n , because it requires computing an n-

dimensional integral for each of'Z" bit sequences . This complexity is excessive for most
applications. In section 5.1, we describe two reduced-complexity implementations
of the MLSD, which make use of a single-step Markov chain model for the fading
correlation in conjunction with per-survivor processing.
4. Spatial DiversityReception
Spatial diversity reception, which has been well-studied for application at radio and
microwave frequencies , has the potential to mitigate the degradation caused by atmo-
spheric turbulence [6,21-25]. Spatial diversity reception in free-space optical com-
munication has been proposed and studied for both near-earth links [6,23] and in-
terplanetary links [25]. Ibrahim [24] has studied the performance of spatial-diversity
optical reception on turbulence channels, assuming that turbulence-induced fading is
uncorrelated at each of the optical receivers. In order for this assumption to hold true,
the spacing between receivers should exceed the fading correlation length in the plane
of the receivers. It may be difficult to satisfy this assumption in practice, for various
reasons. Available space may not permit sufficient receiver spacing. In power-limited
links, which often employ well-collimated beams, the receiver spacing required for
uncorrelated fading may exceed the beam diameter.
4./. Maximum-Likelihood Diversity Detection on Turbulence Channels
In the case of spatial diversity reception with nreceivers, the received signal is described
by an n-component vector r Taking account of correlation between the receivers, Eqs.
(26) and (27) are modified as
314 Xiaoming Zhu andJoseph M. Kahn
P(rIOff) = exp [ - L ~.2] IT J 7rN,

n
i=1 '
1
.'
n
1
(36)
I 2X- 2ElXI)2]
P(rIOn) =
1
X
[L
fx(X) exp -
n
i= 1
(
ri - TJ oeN ' ,
'
IT J 1
n
1 7rN,
dX ,
(37)
where N;l2 is the noise covariance of the ith receiver and
(38)
Here, ex is the covariance matrix of the log-amplitudes in the n receivers, as in

Eq. (19). The likelihood function is
On
On the ML detector employs the decision rule A(r) ~ 1. Since the log-amplitude
Off
follows a joint log-normal distribution, calculation of the likelihood function in Eq.
(39) involves multi-dimensional integration. We emphasize that this decision rule has
been derived under the assumption that the receiving party knows the fading correlation
but not the instantaneous fading state.
The BER of the ML receiver is given by
Pb = P(Off)P(Bit Error Off) + P(On)P(Bit Error On), (40)
where P(Bit Error.Dff) and P(Bit Errorlfm) denote the bit-error probabilities when
the transmitted bit is Off and On, respectively. Without considering intersymbol inter-
ference, which can be ignored when the bit rate is not high and multipath effects are
not pronounced, we have
P(Bit Error Off) = r

} A(r >1
p(rIOff)dr (41)
and
P(Bit Error On) = r
} A(r <1
p(rIOn)dr. (42)
To evaluate the optimal ML diversity detection scheme, we compare it with the

conventional equal -gain combining (EGC) scheme [21]. In the EGC scheme, we as-
sume that the receiving party has knowledge of the marginal distribution of the fading
at each receiver, but has no knowledge of the fading correlation or the instantaneous
fading state. For each individual receiver output, we can find an optimum threshold
Ti . The EGC detector then adds together the n receiver outputs with equal gains and
compares the sum to the threshold :
Ma ximum-
Deooded
Likelihood
Bits
Detection
(a)
Deooded
Bits
(b)
Fig. 4. Dual-branch reception on atmospheric turbulence channels with correlated turbulence-

induced fading: (a) maximum-likelihood detection, (b) equal-gain combining with threshold
detection.
n
Tth = L7;. (43)
; =1
The error probability of EGC is
H = i fx(X)P(Bit ErrorIX)dX, (44)
where
P(Bit ErrorlX) = P(Off)Q ( YiN)

TIl ( e 2XI- 2E[Xd + e2X2- 2E[X2J) - T. ]
+ P(On)Q [
./ 0 V2 t h .(45)
2N
4.2. Numerical Simulationfor Dual Receivers
In thissection,wepresentnumerical simulations of the performance of spatial-diversity

detection for the dual-receiver case, which is illustratedin Fig. 4. As described above,
theMLreceiver[Fig.4(a)] hasfullknowledge of theturbulence-induced fadingcorrela-
tion matrix ex , whilethe EGCreceiver[Fig.4(b)] has knowledge only of the marginal
fadingdistributions at the individual receivers. Weassumethat E[h] = E[h] = E[I]
and N 1 = N2 = N , anddefinetheelectricalsignal-to-noise ratio SNR =(1]E[I])2 I N .
We have used the expressions given in the previous section to numerically compute
the average bit-error probabilities.
In Fig. 5(a) we plot the simulation results, assuming E[X] = 0 and a x = 0.1,
varying the normalized correlation Pd = bx(d I 2 ) from 0 to 0.9. In Fig. 5(b), we
presentcorrespondingresultsassuming ax = 0.25. ComparingFigs. 5(a) and (b), we
rt: 10- 3
f 10-4 (a) E!X] = 0, ax = 0.1

1l
e
a.
10- 5 ~ ML,Pd=O
- 0 - ML, Pd = 0.3
W
g 10-6 . ·0· · ML, Pd = 0.9
- EGC,Pd=O
~ w-7 - - - EGC, Pd = 0.3
.. . .. EGC, Pd = 0.9
- 0 - Single Receive
10-9 ' - - _ ' - - - _ ' - - - _ ' - - - _ L . . . ----JL...----Jl...----Jl...--"-l
4 6 8 10 12 14 16 18 20
ElectricalSignal-to-Noise Ratio (~E If ])2/N (dB)
(b) E!X] = 0, ax = 0.25

~ ML,Pd=O
- 0 - ML, Pd= 0.3
.. 0 .. ML, Pd= 0.9
- EGC,Pd=O
- - - EGC, Pd= 0.3
.. - .. EGC, Pd= 0.9
- 0 - SingleReceive
Fig. 5. BER of dual-branch receiver versus average electrical signal-to-noise ratio using
maximum-likelihood detection (lines with circles) and equal-gain combining (lines without sym-
bols) for different values of Pd, the normalized correlation between the two receivers . The line
with squares represents the BER using a single receiver. The turbulence-induced fading has stan-
dard deviation a x = 0 .1 in (a) and a x = 0.25 in (b). In both (a) and (b), this fading has mean
E[X] =0.
see that turbulence-induced fading causes a greater degradation of the BER when the
standard deviation o x is larger. Diversity reception with two receivers can improve
the performance as compared to a single receiver. With two receivers, ML detection
achieves better performance than EGC for a given SNR. The advantage of ML over
EGC is more pronounced when the correlation Pd between the two receivers is high.
It is also more pronounced when the SNR is high, so that errors are caused mainly by
turbulence-induced fading, as opposed to noise.
4.3. Summary
For IMIDO free-space optical communication in the presence of turbulence-induced

fading, spatial diversity with multiple receivers can mitigate this fading, improving
system performance. When the instantaneous fading state is not known, the conven-
tional spatial-diversity detector must employ EGC and simple threshold detection, but
the correlation of fading at different receivers can reduce the diversity gain achieved
by EGC. In this section , we have derived the ML decision rule under the assumption
that the fading correlation properties are known, but the instantaneous fading state is
not known. We have performed numerical simulations for the dual-receiver case. Our
results show that the ML scheme provides better performance than EGC, particularly
when the fading at different receivers is highly correlated and/or when the average
SNR is high.
5. Temporal Domain Techniques
5.1. Markov Chain Model in Maximum-Likelihood Sequence Detection through

Turbulence
The optimal maximum-likelihood sequence detection (MLSD) requires complicated

multidimensional integration, and its computational complexity is exponential in the
length of the transmitted bit sequence . To simplify implementation of the MLSD , we
propose a single-step Markov chain (SMC) model for the fading temporal correlation,
and we use the SMC model to derive an approximate higher-order distribution of bit
errors , as wel1 as two reduced-complexity MLSD algorithms based on sub-optimal
per-survivor processing (PSP). We use simulations to investigate the accuracy of the
SMC model and the effectiveness of the suboptimal MLSD techniques .
5.1.1 . Joint Temporal Distribution for Turbulence Induced Fading
In a free-space optical communication system using OOK, we assume a n-bit sequence

S = [SIS2 ' " sn] is transmitted . We define the index subset of On-state symbols
SOn = {n i E {I, 2, .. . , n}, sn i = 1 } ~l ' We also have the index subset of Off-
state symbols So« = {/ j E {1 ,2, .. . ,n}, Slj = l}j~;n . Ignoring intersymbol
interference (lSI) , the receiver would only receive signal light when the On-state is
transmitted. The joint distribution of the signal intensity of On-state symbols is
In (~) ] }
X exp { -Hln e;: )In e;:)] (C~"r' [
In (~) , (46)
where the ith On-state symbol intensity
I n i = 10 exp(2X n i - 2E[X]) = 10 exp(2x n ;) . (47)
Here, x n i = X n i - E[X] can be modeled as a Gaussian random variable with zero

mean and covariance 0'1-.
For a string of bits, the covariance matrix of On-state bits is
ci: =
nXn
(48)
where T is the bit interval.
One can also show that the joint distribution of x = [x n1, Xn2, .. . , x nm [7]:
(49)
5.1.2. Single-Step MC Model for Fading Correlations
Here, we consider the SMC model describing the correlation of fading at a sequence
of equally spaced times. It is straightforward to extend the treatment to spatial cor-
relation provided that the receivers are equally spaced along a line perpendicular to
the direction of propagation. Let X n denote the log-amplitude at time n, and let X~-l
denote [Xl , . . . ,x n- I]. Assuming the turbulence-induced fading is a single-step MC,
we have
P(XnIX~ -l) = P( xnlxn-I) . (50)
Define or = bx (doT/To). If xf follows ajoint Gaussian distribution, the conditional

distribution of X n given Xl is
(51)
From Eq. (51), we see that the exponent of correlation has dropped to I , implying that
the correlation in the SMC model is stronger than in the exact correlation model.
Since the SMC model only takes into consideration the probability distribution
of the most adjacent On-bit that contains information on the instantaneous fading
state, it is applicable to modeling spatial correlation only in the special case that the
multiple receivers are aligned along a line perpendicular to the direction of propagation.
However since the number ofreceivers is typically small, the computational complexity
of ML detection with diversity reception will typically be reasonable, as we have shown
in [6,7].
5.1.3. Burst Error Distribution for Symbol-by-Symbol Detection
To test the validity of the SMC model, we compute the burst error probability for
symbol-by-symbol detection of OOK. Assume that a sequence of On bits is transmitted.
The probability of having m consecutive erasures is
where Tth denotes the receiver decision threshold . Using the chain rule, Eq. (52) can
be written as
Communication techniques and coding for atmospheric turbulence channel s 319
10- 2 ~--.---,-----,----.-----r---..
~
:>
co
5 10-6
t:
W - - Pr=0.1
iii - 0 - Pr=0.3
E 10-8 _Pr=0.5
- 0 - Pr=0.7
--+- Pr=0.9
- - - Pr=0.1, SMC
- -0 - Pr = 0.3, SMC
- ... - Pr= 0.5, SMC
- -a - Pr= 0.7, SMC
- -+ - Pr= 0.9, SMC
2 3 4 5 6 7
Numberof Consecutive Errors, m
Fig. 6. Distribution of consecutive bit errors and its upper bound derived using the single-step
Markov chain model. For simplicity, additive white Gaussian noise is ignored here.
POn = Pon(ro > rth)POn(rl < rthlro > rth)

X . . . Pon(rm+llro > rth; r . < rth, Vi ,l :::; i:::; n) , (53)
which is upper-bounded by
POn = Pon(ro > rth, rl < rth)Po n(r2 < r thlr l < rth
x .. . Po n(rm+ l > rth\r m < r th
= Pon(ro > r th,rl < r th) [Po n(r2 < r thlrl < rth] m-l
x Pon(rm+l > rtllJr m < rth) . (54)
To simplify the calculation, let us first ignore AWGN and focus on errors caused by
turbulence-induced fading. In the absence of AWGN, we have
Pon(ri < rth, r n < rth)

Pon(rl < rth)
J~~ f(Xl) J~~ f(xnll xI)dxndxl
(55)
J~ ~f(xddxl
where Xth can be calculated by rth:
(56)
Note that to compute eqs. (54) and (55), we need only perform two-dimensional inte-
gration, independent of the number of bits in the sequence.
In Fig. 6, we present the distribution of consecutive bit errors for several different
correlation parameters, comparing the SMC model to the exact correlation model.
Figure 6 has been computed assuming "lIo = 1, E[X] = 0, ax = 1, and a decision

threshold
1
r th = 10 exp(2E[X]) = O.l.
As we wouldexpect,the probability of a burstof length m > I increasesas we increase
the correlation parameter PT. We see that the SMC model yields a fairly tight upper
bound on the exact burst-error probability, validating its use in modelingturbulence-
induced fading correlation. Note that as the correlation PT increases, the probability
of a single bit error (m = 1) decreases,because it becomes morelikely that a burstof
more than one bit error will occur.
Considering AWGN, (55) is modified to
Tth
Pan(r < r th) = j- 00 P(rIOn)dr, (57)
where
(59)
Note that a four-dimensional integration is requiredto compute Eq. (58), independent

of the length of the sequence.
In the above,we have focused on misseddetectionof On bits. When a sequenceof
Off bits is transmitted, the probability of m consecutive false alarms (falselydetecting
a sequence of m On bits) is simply
(60)
where
Po« = roo P(rIOff)dr. (61)
i:
Becausethe correlationof atmosphere turbulence only affectsthe detectionof On bits,
a burst of m missed On bits is much more likely than a burst of m false alarms.
It is obvious that
Pan(rl < r th,r n > rth) = Pan( rl < r th) - Pan(rl < r th,r n < rtl.) . (63)
We can modify Eqs. (54)-(61) to derive the higher-order BER distribution for a
sequenceof bits withsymbol-bysymboldetection. SinceonlyOn-statebits are affected
by atmosphericturbulence, basedon the SMC model,the On-stateBER woulddepend
on the joint probability distribution of the most recent preceding On-state bit and the
current On-bit.
5.1.4. Sub-Optimal Per-Survivor Processing for MLSD
The MLSD, as expressed in Eq. (51), is optimal for detecting a sequence of bits that
is i.i.d. and uniform on the set {Off, On}. Detecting a sequence of nbits requires a
complexity of order n2 n , becau se it requires computing an n-dimensional integral
for each of 2n bit sequences. To reduce the complexity, we consider PSP, which was
proposed by Polydoros to extend the Viterbi algorithm to uncertain environments [26].
The key idea is to use the received intensity of recently detected On bits to reduce
uncertainty about the state of the turbulence-induced fading . If we consider AWGN and
use the SMC model for the fading temporal correlation, then knowing the correlation
between two consecutive On bits is sufficient to perform MLSD . Even under the SMC
assumption, however, the likelihood function in Eq. (34) cannot be decoupled into a
sum of per-branch metrics, which is requ ired for a reduced-complexity implementation.
We will modify the metric function to allow us to implement a sub-optimal MLSD
using PSP. This sub-optimal MLSD can decode an n-bit sequence with a complexity
of order n 2 , as compared to the optimal MLSD described in Section 3.2, which has a
complexity of order n2 n .
Assume a transmitted n-bit sequence s = [S1 S2 ' " s-]. Define the index subset
of On-state symbols SOn = {n, E {1,2 , ... , n },Sn i = 1}~1 with n, = I,i.e.,
assume the first bit is On. We also have the index subset of Off-state symbols See =
{I j E {I , 2, .. . , n} , S lj = O}j~;n. The exact likelihood function in Eq. (34) is
(64)
where XO n = {x n il ni E SOn }~1' Based on the SMC model, we have
~'
n- m
L(8) '" exp [- L

i = 1
it
] f!r x.. )f(x", Ix",) .. . f( x"m IX"mo')
x On
Ii E SOff
m
(,", - "lo e"' , )' ]
x exp [- L
i = 1
s; , dxo n . (65)
n i E SOn
In order to decouple Eq. (65), we modify the weighted integration of Eq. (65) by
322 Xiaoming Zhu andJosephM. Kahn
m-l
x II
i = 1
(66)
n i E SOn
We modify Eq. (66) by decoupling the integration as
m-l
II
i = 1
X ni E SOn
m-l
(67)
II
i = 1
ni E SOn
since the first term in the denominator is identical in the likelihood function for all
codewords. We can write the modified likelihood function of 8 as
m -l
i
II
= 1
ni E SOn
£(8) = - -m-
-l
--------------
II
i = 1
ni E S On
(68)
From Eq. (68), we can define the metric function of the kth branch (k > I):
BMk(Off) = exp ( - ~:) , (70)
where k = ni and ni-l denotes the position of the most recent On-state bit.
In terms of the branch metrics, the MLSD can be expressed as
m n- m
8 = argmax
s
II
i =2
BMni(On) IIi = 1
BMli(Off) . (71)
n i E SOn I i E SOff
Communication techniques and coding foratmospheric turbulence channels 323
Fig.7. Viterbi algorithm for maximum-likelihood sequence detection with turbulence-induced

fading. Solid lines denote Offbits and dashed lines denote On bits. The branch metric is marked
oneach branch. The decoded bitsequence is indicated at the bottom.
Since the MLSD considers path metrics that are the product of branch metrics,
it can be implemented using the Viterbi algorithm, with a complexity of the order of
n 2 /2. Note that computation of each branch metric of the form (69) requires only a
two-dimensional integration, independentof n . In Eq. (69), we see that computing the
branch metric for an On bit requires information obtained during the most recently
transmitted On bit, so we can only choose a survivor path when the previous bit is
known to be On; otherwise, we must keep track of the amplitude of the most recently
receivedOn bit, and must also keep track of all survivorpaths whose last bit is Off.An
example of the asymmetric PSP is shown in Fig. 7. The number on each branch is the
branch metriccomputed using Eqs. (69) and (70). The firstand sixth bits correspond to
the On state. We see that in this asymmetric PSP, we can only eliminate non-survivor
paths when the most recent bit corresponds to the On state. In order to reduce the
complexityof this algorithm, we can add an On bit at the beginningand ending of each
n-bit sequence, as we havedone in this example. Wecan simply employ the starting On
bit of the next sequenceas the ending On bit of the previoussequence. The complexity
of such an algorithm would be of order n 2 12, and only two-dimensional is required.
However, we need extra memory to keep track of the survivor path information. Also,
some bit overhead is required for implementation of this algorithm.
The algorithmdescribed above still requiresa largecomputationalload to perform
the two-dimensional integration. To reduce this complexity, we can estimate X n using
rn-l . Since
s = arg~ax Ix f(X) exp [ - t (ri - TJ%~oe2Xi )2] dx
= arg~ax Ix f(xd4>(Sl, r l , xd g [f( xd x i-d4>(Si, ri , Xi)] dx , (72)
(r; - TJSi10 e2Xi)2]

4>(Si, ri, Xi) = exp [- N ' (73)
i
Replacing f(Xilx i-d with f(Xiri-d , we can define the metric function for
branch i:
BMi (On ) =
f f(Xl)4>(Sl = l,rl,xddxl, i = 1
(74)
{ IXi f(Xilri-l)4>(Si = 1,ri, xi)dxi ,
Xl
i
•
» 1
BMi(Off) = 1,
f(X ilri-d4>(Si = 0, ri, Xi)dxi. (75)
In terms of the branch metrics, the sub-optimal MLSD can be expressed similar to Eq.
(74) with branch mteric functions in Eqs. (74) and (75).
To demonstrate the effectiveness of the sub-optimal PSP algorithm, we present the
simulation results in Fig. 8. In this simulation, we assume there is at most one error
between two correctly decoded On bits, and we assume C7x = 0.25 , 10 = I. In Fig.
8(a), we assume E[X] = 0 and C7x = 0.1. We plot the BER versus average electrical
SNR, given by SNR = TJE[1]) 2 IN. We consider the two choices of branch metric
discussed above; Method I uses Eqs. (69) and (70), while Method 2 uses Eqs. (74)
and (75). The temporal correlation coefficient PT is chosen to be 0.15 and 0.95. In Fig.
8, we see that both Methods 1 and 2 can achieve much better bit-error performance
than the symbol-by-symbol decoding scheme. Method 2 is subject to a penalty of a
few dB compared to Method I, but avoids the two-dimensional integration required
by Method I.
From the discussions above, we see the SMC modelcan help to greatly simplify the
implementation of MLSD with the sub-optimal PSP algorithm , leading to a significant
improvement in bit-error performance.
5.2. Pilot-Symbol Assisted Detectionfor Correlated Turbulent Free-Space Optical

Channels
Pilot-symbol assisted detection (PSAD) has been employed in applications at radio

and microwave frequencies to mitigate the effects of channel fading in situations where
channel side information (SI), i.e., the instantaneous fading state, is not known to the
receiver [27,28]. In PSAD, the transmitter periodically inserts known symbols, which
comprises a reference for receiver to derive the correlated channel fading. PSAD can
also be incorporated with maximum likelihood sequence detection (MLSD) to truncate
the survivor path of information bit string [8].
In this section, we first introduce the pilot-symbol assisted maximum likelihood
(PSA-ML) detection as well as another simpler PS-assisted detection scheme with
variable threshold (PSA-VT). We then present the numerical simulations for these
detection techniques and compare them with the normal ML symbol-by-symbol de-
tection.
Communication techniques and coding for atmospheric turbulence channel s 325
(a) E!X] = 0, ax = 0.1

- Method1,Pr=0.15
- - - Method 1, Pr= 0.95
--0- Method 2, Pr= 0.15
- 0 - Method 2, Pr = 0.95
. - - _. Symbol-by-Symbol
10 12 14 16 18 20
ElectricalSignal·to-Noise Ratio (~E [I ])2/N (dB)
(b) E!X] = 0, ax = 0.25

- Method1,Pr=0.15
- - - Method 1, Pr = 0.95
--0- Method 2, Pr= 0.15
- 0 - Method 2, Pr = 0.95
Symbol· by-Symbol
10-4'-----_---'-_ _ --'-_ _ '-----_---'-_ _--"-_---'
8 10 12 14 16 18 20
ElectricalSignal-lo-NoiseRatio (~E [I ])2/N (dB)
Fig. 8. BER of different decoding schemes versus average electrical signal-to-noise ratio with
turbulence-induced fading. The dotted line represents the BER using a symbol-by -symbol de-
coding scheme. The solid lines consider a PSP algorithm based on the SMC model using branch
metric functions (69) and (70), while the dashed lines represent a similar algorithm using Eqs.
(74) and (75).
5.2.1. Pilot-Symbol Assisted Maximum-Likelihood Detection
In PSAD, we periodically insert On-state symbols in the information bit string. The
composite symbols are transmitted in the usual way over the channel. The resulting
frame structure is shown in Fig. 9. At the receiver, we can decode the information bit by
considering the joint distribution of the turbulence induced fading at the information
bit and the adjacent pilot symbols . Although PSAD will lead to delay in the receiver
because of the need to store the whole frame before decoding, the pilot symbols (PS)
provide the receiver with information on the state of turbulence-induced fading, and
can thus help mitigate the effects of fading. In terms of the branch rnetrics, the sub-
optimal MLSD can be expressed similar to Eq. (71) with branch metric functions in
Eqs. (74) and (75).
~~~~ •
EEEEJ ~l • ~ • ----
14 M ~I
one frame
P: PilotSymbol
D: DataSymbol
Fig. 9. Pilot-symbol assisted detection frame structure.
In Fig. 9, the composite frame is of length M, the first symbol of each frame is the
On-state PS followed by M - 1 information symbols . In detecting the information bits
in one frame, we can refer to the PS in the current frame and in the next frame. Assume
r; is the received photo-current signal of the ith information bit in the frame, ro and
r M are the received signal of the PS of the current frame and the next frame. The joint
probability distribution of r = [ro , ri , r M] conditioned on the ith (1 :::; i :::; M - 1)
information bit Si are
2
1
p(rlsi = 0) = (1rN) 3/2 exp -
[
iv j -<Xl<Xl j -<Xl<Xl fX(XO,XM)
r ]
(rj - 'T]!o e2Xr2ErXj))2]

x exp -
[ L N dXodXM , (76)
j=O,M
1
p(rlsi = 1) = (1rN)3 /2
j -<Xl
<Xl j -<Xl
<Xl j -<Xl
<Xl fx(XO ,Xi ,XM)
(r - 'T]!o e2Xr 2EIXj))2]
x exp [ - j=~M J N dXodXidXM . (77)
The PDFs oflog-amplitude f(Xo , XM) and tix. , Xi, XM) are shown as in Eq. (35).
The likelihood ratio is
1
The ML detector employs the decision rule A(r) ~ 1. We emphasize that this deci-
o
sion rule has been derived under the assumption that the receiver knows the fading
correlation but not the instantaneous fading state.
The BER of the ith information bit in the M -bit frame is given by
Pi = P(Si = O)P(BitErrorls i = 0) + P(Si = l)P(BitErrorl si = 1), (79)
where P(Bit Error s, = 0) and P(Bit Error s, = I) denote the bit-error probabilitie s
when the transmitted ith information bit is 0 and I, respectively. Without considering
intersymbol interference, which can be ignored when the bit rate is not high and
multipath effects are not pronounced, we have
P(BitErrorlsi = 0) = { p(rl si = O)dr (80)

} A(r >1
and
P(BitErrorlsi = 1) = { p(rlsi = l)dr. (81)
} A(r < 1
The average BER of a frame is

M-1
Pb= M~l Lh (82)

i=1
5.2.2. Pilot-Symbol Assisted Symbol-by-Symbol Detection with Variable

Threshold
The PSA-ML detection scheme described above employs the joint PDF of turbulence-
induced fading. This scheme requires very lengthy computations to perform multi-
dimensional integration. Here, we propose a simpler PS-assisted detection scheme
with variable threshold set with the aid of received pilot symbols . We refer to this
scheme as PS-assisted variable-threshold (PSA-VT) detection .
Since [Xo, X i, XM] is jointly Gaussian-di stributed, E[Xi IXo, XM] is an affine
function of the log-amplitude at two adjacent PS positions, Xo and XM :
E[XiIXo, XM] = a?X o + af'l XM, (83)
a? ] _ [ 1 bx ( ~: dO)] -1 [ bx ( ~: dO)] (84)

[ af'l - bx (~oT do) 1 bx [(M:ai)Tdo] .
We simply assume that the received PS signals are noise free. Therefore we can set the
variable threshold for the ith (\ ~ i ~ M - 1) information bit in the frame to be
(85)
Using PSA- VT. the BER of the ith information bit in an M -bit frame is given by
Pi = P(Si = O)P(BitErrorlsi = 0) + P( Si = l)P(BitErrorl si = 1), (86)
P(BitErrorlsi = 0) = l. r
i > Ti p(rl si = O)dr, (87)
P(BitErrorlsi = 1) = l. r
i < Ti p(rl si = l)dr. (88)
The expression for the average HER is the same as Eq. (82).
328 Xiaoming Zhu andJoseph M. Kahn
5.2.3. Numerical Simulation
In this section, we present numerical simulations of the detection schemes we pro-

posed in sections 5.2.1 and 5.2.2. We assume the turbulence-induced log-amplitude
fluctuation has zero mean, covariance ax = 0.1 and TITo = 0.001. The BER of
ML symbol-by-symbol detection versus the average electrical signal-to-noise ratio
SNR = (1710)2 IN is shown in Fig. 10 by a dotted line. Recall that the ML symbol-
bysymbol detector has knowledge only of the marginal distribution of fading, but not
the instantaneous fading. If the ML symbol-by-symbol detector had knowledge of the
instantaneous fading, it could achieve better performance; this idealized case is plotted
in Fig. 10 using a dash-dot line. SI on the instantaneous fading state is typically not
available to the receiver. If we assume the receiver has full knowledge of the joint tem-
poral distribution of the correlated fading, the PS-assisted detection schemes discussed
in Sections 5.2.1 and 5.2.2 can be applied . In our simulation, we use the PSAD frame
structure shown in Fig. 9. In Fig. 10, we plot the BER of the ith (i = M 12) symbol in
an M -bit frame with PSA-ML detection and PS-assisted variable threshold (PSA- VT)
detection, respectively. In Fig. lO(a), we plot the BER versus the average electrical
SNR of each channel bit and in Fig. lO(b), we plot the BER versus the average elec-
trical SNR of each information bit. We assume the frame size M to be 10, 100, and
1000, respectively. In Fig. I I(a) and (b), we plot the average BER of an M-bit frame
with PSA- VT detection versus the average electrical SNR of each channel bit and of
each information bit respectively, varying the frame size M to be 10, 100, and 1000.
In Figs. 10and II, we observe that when the receiver has knowledge of the temporal
correlation of fading, but does not have SI on the instantaneous fading state, PS-
assisted detection techniques can give better BER performance than ML symbol-by-
symbol detection . However with larger frame sizes, the correlation between the PS and
information symbols will become weaker. This will make the PS-assisted detection
less effective. In our simulation, we observe that in order to ensure that PS-assisted
detection achieves a gain of at least 0.5 dB over ML symbol-by-symbol detection, we
require M < dr IT.
In peak power-limited systems, for a fixed transmission bit rate over the channel,
using smaller M will improve BER performance, as shown in Figs. lO(a) and II(a).
Another benefit of choosing smaller M is that it will reduce the delay incurred during
detection. Choosing smaller M, however, does entail a loss of information throughput.
In average power-limited systems, the inserted PS consume additional power and
decrease the SNR of bits transmitted on the channel. As we can see in Figs. lO(b) and
II(b), the BER performance versus average SNR per information bit is very close for
M = 10 and M = 100. If we choose M even smaller, however, the penalty due to the
additional power required to transmit the PS will surpass the PSAD gain. Therefore,
in average power-limited systems, we should choose M to be too small.
We also note that PSA-ML can achieve better BER performance than PSA-VT, as
seen in Fig. 10. PSA-VT does not require multi-dimensional integration and is much
faster and simpler to implement than PSA-ML detection, however.
5.3. Experimental Demonstration on Temporal Domain Techniques
We have performed transmission experiments to demonstrate the effectiveness of the

MLSD and PSAD techniques in mitigating scintillation noise [29]. Because our proof-
(a)
'.----~--~--~---,---~--~---.
0
10
..;- ·2
g10
~ -3
10
"2
c,
5 10. 4 -&- PSA-M L, M = 10
t: - PSA-ML,M = 100
W
-+- PSA-M L, M = 1000
-e - PSA-VT, M = 10
---- PSA-VT, M = 100
·6 -+ - PSA-VT, M = 1000
10 - Symbol-by-S ymbol w /o SI
_ .- S ~mbol-b~-S ;'mbol wi SI
10.7 "l-_ _ ~ _ _~_ _~_ _~_ __'__ _~_ __..J
6 8 10 12 14 16 18 20
Average SNR Per Channel Bit: y = (n,/o)2/N o (dB)
(b)
10 0
10. 1
..;- 10 ·2
~
:.0 10 -3
"""-e'"
...0 10 · 4
t: -&- PSA -ML, M = 10
w - PSA-ML,M = 100
.!. 10 . 5 -+- PSA-M L, M = 1000
iii -e- PSA-VT , M = 10
-- -- PSA-VT, M = 100
10 · 6 -+- PSA-V T M = 1000
- Symbol- by-S ymb ol w /o SI
_ .- Symbol-by-Symbol wi SI
10 '76l---~8--~10--~1-2---1~4---1~6---1~8-----.J20
Average SNR Per Information Bit: y = (I - I/M)( n,/o)2/N o (dB)
Fig. 10. BER of ith symbol in an M -bit frame versus (a) average electrical SNR of each channel
bit and (b) average electrical SNR of eachinformation bit using PSA-ML and PSA variable
thre shold (VT) . Wechoose different values of frame size M (10, 100, and 1000, respectively).
Here i = M /2. The covariance oftog-amplitude due to turbulence-induced fading is a X 0 .1. =
The ratio of bit-interval T versus fading coherencetime d T is T / dT = 0 .001. The BER of ML
symbol-by-symbol dete ction with or without instantaneous channel fad ing side-information (SI)
are also shown in the figure for comparison.
of-concept system employs a personal computer (PC) for data acquisition and decod -
ing, the bit rate is limited to 3 kb/s. Using appropriate special-purpose hardware, these
detection techniques can be implemented at the bit rates of interest in most applications
(Mb/s to ObIs).
Our experimental system is shown in Fig. 12. At the transmitter, we employ a
675 nm, 0,95 mW (Class II) laser diode module with an output beam divergence of
(a)
10° r---~-~--~--~--~--~------,
-&- PSA-VT, M = 10
- PSA-VT, M = 100
-+- PSA-VT , M = 1000
- Symbol-by-Symbol wlo SI
_ .- Svmbol-bv-Svmbol wi SI
8 10 12 14 16 18
Average SNR Per Channel Bit: y = (T]/ o)2IN o (dB)
(b)
10°
10.1
._ ._ ._._._._._._._._._<,
Cl; 10.2
g
:E 10. 3
e'"
.D
~
g 10. 4
~
.!.
iii 10.
5 -&- PSA-VT, M = 10
- PSA-VT. M 100 =
-+- PSA -VT, M = 1000
10.6 - Symbol-by-Symbol wlo SI
_.- Svmbol-bv-Symbol wi SI
10.7 '----'---~'---~--~--~--~-----'
6 8 10 12 14 16 18 20
Average SNR Per Information Bit : y = (I - IIM)( T]/o)2IN o (dB)
Fig. 11. Average HERof an M -bit frameversus(a) average electricalSNRof eachchannelbitand

(b) averageelectricalSNRof each information bit usingPSA-VTdetection. Wechoosethe frame
size M to be 10. 100, and 1000. respectively. The covariance of log-amplitude fluctuation due
to turbulence-induced fading is a x = 0.1. The ratio of bit-interval T versus fading coherence
time dT is T/dT = 0.001. The HER of ML symbol-by-symbol detection with or without
instantaneous channel fading side-information (SI) are also shown in the figure for comparison.
0.35 mrad. Wemodulatethe injectioncurrent to achieveOOK with non-return-to-zero

pulses, The receiveruses a telescopewith an entranceaperturediameter of D = 8 em.
An optical bandpass filterwith 10nm bandwidth(-3 dB) is used to minimize ambient
light noise. A 1.I-mm 2 p-i-n photodiodeis coupled to a transimpedance preamplifier
and a second-orderBessellowpass filterhavinga 2-kHz bandwidth(-3 dB).The back-
to-back receiversensitivity is -69 dBm at a BER of 10- 4 • Received electrical signals
Communication techniques and codingfor atmospheric turbulence channels 331
[--------- :::::::::::::::::::::::::11
r-c. 0 r>; T~escope :, :,
Trans~
[···~;~=---:: U9.0~' : r:~CK ~ PSAD ~i
i Rec.
lD ~~ u~
ii
Bits : or I I Bits
, . TZ Sarreler MLSD
i..:r~~.':'. _.i QqPOo .

~=c : _ :f~~~ ._....~~~!~.~~~: j i
~~ .!
Fig. 12. Experimental 3 kb/s free-space opticallinkusingpilot-symbolassisted detection (PSAD)

or maximum-likelihood sequence detection (MLSD) to mitigate turbulence-induced scintillation
noise.
are sampled using a PC-interfaced data acquisition card, and the MLSD and PSAD
algorithms are implemented in Lab VIEW software.
Optical signals are transmitted over a 500-m outdoor path between Cory Hall and
Doe Library on the U.c. Berkeley Campus. Because of the relatively large photodiode
size, turbulence-induced image degradation has negligible impact overthe transmission
range employed. The intensity correlation length can be estimated as >-.L, where >-. is
the transmission wavelength and L is the transmission range [17], yielding do ~ 1.8
cm. Although the receiver aperture, D = 8 em, is somewhat larger than this estimate
of do, aperture averaging does not completely eliminate scintillation noise, and the
observed standard deviation of the log-amplitude is a x ~ 1.6 . Based on experimental
measurements, we estimate the intensity correlation time to be TO ~ 35 ms, which
is much longer than the receiver observation interval To ~ 0.0 I ms. Hence , both
MLSD and PSAD are expected to be effective in mitigating scintillation noise . We
have implemented MLSD using the SMC model and the Method 2 described in [8],
and have implemented PSAD using PSA- VT with a frame size of M = 12 bits.
Our experimental results are shown in Fig. 13, which shows the BER for On-state
bits versus the average received electrical signal-to-noise ratio (SNR) for various de-
tection techniques. The measured results for ML symbol-by-symbol detection, PSAD
and MLSD are indicated by squares, triangles and circles , respectively . Using mea-
sured values of a x and TO, we have used the theories described in [7,8,19] to compute
the dashed, dot-dashed, and solid lines, respectively. We obtain excellent agreement
between experimental and theoretical results with no adjustable parameters. As com-
pared to symbol-by-symbol detection, we observe experimentally that PSAD gives an
SNR gain of about 1.9 dB, while MLSD yields an SNR gain of about 2.4 dB (both at
10- 3 error probability). Even larger gains are expected for longer propagation paths
with larger values of the log-amplitude standard deviation a x .
10-1 =----r---,----r-----.---.---,-----,---=
s
:0 10-2
III
.D
e
0.
...
g 10-3
W ..... ...
..!. ..•..
iii
Q) • Symbol-by-symbol (expt.) ...........:
iii ... PSAD (expt.)
Ci5
C: 10-4 • MLSD (expt.)
o ------. Symbol-by-symbol (calc.)
-.-.-.-. PSAD (calc.)
-- MLSD (calc .)
10-5 l - _ - L_ _-'-_-.l_ _--I-_ _l - - _ - - - L_ _.J......_-l
16 17 18 19 20 21 22 23 24
Electrical Signal-to-Noise Ratio (1lE[/o,J)2/cri (dB)
Fig. 13. BER for On- state bits versus average received electrical signal-tonoise ratio in a 500-m
outdoor transmission experiment. PSAD: pilot symbol-assisted detection. MLSD: maximum-
likelihood sequence detection.
6. Performance Bounds for Coded Free-Space Optical Communication

Through Atmospheric Thrbulence
Error-control coding can be applied to improve the error performance of free-space

optical communication links over atmospheric turbulence channels [18]. The theoret-
ical error performance of coded systems over time-varying channels has been under
research for many years [22,30-32] . In most wireless communication system s, the
channel is not memory less. The error performance of such continuous fading channels
with memory often requires lengthy computer simulations. Analytical performance
bounds have been widely used to study the error performance of communication sys-
tems. In this section, we find an expression for the pair-wise codeword-error probability
and upper-bound the codeword-error probability Pblock by
(89)
where So is the set of all codewords and POj is the probability that codeword OJ
is transmitted, The pairwise error probability P( OJ,Ok) is the probability that when
codeword OJ is transmitted, the decoder favors selection of an incorrect codeword
Ok over OJ. With the knowledge of the weight enumerating function (WEF) , we can
further simplify the calculation of Eq. (89) and extend it to accurately estimate the
error bounds of constituent codes where the number of codewords is infinite, such as
convolutional codes and turbo codes [32,33]. Many error bounds have been introduced
for radio-frequency channels , which often can be well-modeled as Rayleigh or Rician
fading channels [30,31].
In this section we will derive an error performance bound for coded OOK free-
space optical communication through atmospheric turbulence channels, where the
fading is described by ajoint log-normal distribution.
6.1. Pairwise Codeword-Error Probability Bound
In this analysis we assume the turbulence -induced fading to be piecewise-constant

during each bit interval and known to the receiver, i.e., receiver has perfect state infor-
mation (SI) about the channel. The receiver utilizes ML soft decoding .
. . k k k
Consider two n-bit codewords Gj = [c{ , dz , , c~ l and Ci; = [cl, C2 , . . . , cn].
Define the symmetric difference of their On-state symbol index subsets, i.e., the set
consists of all those points that belong to one or the other of the two sets but not to
both:
Sda. .c; = { ni {E 1,2 , ., ., n }, c-
. i ...I- k }mj ,k
n · r Cn , (90)
t 1. i =l
and define its complement set:
c. ic; _ { . . _ k }n-mj ,k
Se - n t E {1, 2, .. . ,n },d,. - cn · . (91)
t i= l t
The energy over the symmetric difference set of two codewords can be defined as
Mj .k
L
i = 1
(92)
ni E S~j 'Ck
Owing to the linearity of the code, the pairwise error probability between Gjand
Gk can be approximated by the pairwise error probability between a codeword whose
index subset of On-state bits is the same as the symmetric difference set Sfj ,Ck and
the all-zero codeword:
E[Pe(Gj , Gk ) ] ~ E [Q ( VE~~~k )]
= r f(X)Q (VEC
t;
j,C
2No
k
) dX . (93)
Since Q( v'x) :::; ~ exp ( ~x ), an upper bound on the pairwise error probability is
(94)
Defining the average signal to noise ratio (SNR) 'Y = ( Tf~~) 2 we can express Eq. (94)
as
334 XiaomingZhu and JosephM. Kahn
Under the assumptionof weak turbulence (a x «I), we can approximatethe upper

bound (95) as
(96)
and
< 1
E[Pe(Cj ,Co)] rv "2 exp
v».»
(--4-) g vI
mj,k 1
+4"/Ai exp
[mj'k
~
"/2 Ui2 Ai ]
2(1 +4,,/Ai) ,
(97)
where Ai is the ith eigenvalue of the covariance matrix C~k and Ui is the sum of the
elements in the corresponding eigenvector.
To verify the accuracy of the approximate upper bound (97), we calculate the pair-
wise error probabilityof two codewords whose difference set is Sfj ,Ck = {I , 3, 4, 6}
versus the average SNR, choosing T / TO = 0.04 . The approximate upper bounds for
log-amplitude variances a x = 0.05, 0.15, and 0.25 are indicated by the solid lines in
Fig. 14.The pairwise codeword-errorprobabilitycalculated using Eq. (93) is indicated
by the dashed lines in Fig. 14 for comparison. We can see that the upper bound (97) is
accurate under the assumption of small a x in the weak-turbulenceregime.
To demonstrate the limits of applicability of Eq. (97), we also consider larger
values of ax in Fig. 15. In Fig. 15, we see when ax increases from 0.25 to 0.65,
which should obviously result in higher error probability, the approximate pairwise
error probability computed using Eq. (97) decreases instead. This is because terms of
higher order in the log-amplitude,which we ignored in deriving Eq. (96), are no longer
negligible when a x is large, even though the weak-turbulenceapproximation made in
Eq. (5) is still valid. Therefore, Eq. (97) is no longer valid for large ax .
Since many free-space optical communication systems operate in the weak-
turbulence regime, the approximate upper bound (97) should be widely applicable
to estimate the pairwise error probability for long block codes and constituent codes,
such as convolutional codes and turbo codes, which can be useful in optimizing the
design of codes for free-space optical turbulence channels.
Communicationtechniques and coding for atmospheric turbulencechannels 335
10°
10.2
~
:E
'"....
.D
0
10.4
0...
....
0
t::
"'-l
"0
....
10-6
- Bound °X=0.05
0
..,
~
-e- Bound o X=0.15
"0 8 ~ Bound 0x=0.25
0 10.
U
.., --_. Exact 0x=o .05
. '"
~
-0- Exact 0x=O·15
' <;j 10·10 -8- Exact 0x=O·25
0...
10.12
0 2 4 6 10 12 14
Average SNR: y = (TJlo)2/No (dB)
Fig. 14. Pairwisecodeword-errorprobabilityversus averageSNR usingexact integration(dashed

lines) and using approximate upper bounds (solid lines) for various log-amplitude variances
(ax = 0.05, 0.15, 0.25).
6.2. Error-Probability Bounds for Various Coding Schemes

Using the approximate pairwise codeword-error upper bound that we have derived,
we can compute upper bounds on the error probability for various coding schemes .
In this section, we restrict ourselves to considering linear codes, and assume that all
codewords are selected with equal probability. We derive upper bounds on the BER of
block codes, convolutional codes and turbo codes.
6.2.1. Block Codes
For a binary linear (n, k) block code with a set of codewords {Co, C1,. . ., C2 k -I}'
where Co denotes the all-zero codeword, the average block-error probability with
maximum likelihood (ML) decoding is:
2k - 1
E[Pblock] ::; L E[Pe(Cj ,Co)], (98)

i= 1
and the average BER upper bound is

2k _ l
E[H] ::; ~ L BjE[Pe(Cj, Co)], (99)

i =l
0
10
-I
10
&'
:.0
'"8
.D
10
·2 -------------9 _
c..
...
g
u.l -3
'E 10
0
~
'"
"='
0
- Bound (1x=O.05
U 10-4 -e- Bound (1x=O.25
.~'" -B- Bound (1x=O.65
'eo Exact (1x=O.05

e,
10.5 --e-' Exact (1 x=O.25
-B-- Exact (1 x=O.65
10-6
0 2 4 6 8 10
Average SNR: y = (1l10)2/No (dB)
Fig. 15. Pairwise codeword-error probability versus average SNR using exact integration (dashed
lines) and using approximate upper bound s (solid lines) for larger values of the log-amplitude
variances (ax = 0.05, 0.25, 0.65) .
where B, is the Hamming weight of the information sequence corresponding to code-

word c..
lt is straightforward to apply Eq. (97) in Eqs. (98) and (99) to obtain performance
bounds for block codes.
6.3. Convolutional Codes
Consider a rate R = kin linear convolutional code, where Co is the all-zero codeword
and {Cj, j > O} is the set of nonzero codewords whose initial state is the all-zero
state which first re-rnerge with the all-zero state at their final state. With ML decoding,
the average BER can be upper-bounded by
1
kL
00
E[Pb] ~ BjE[Pe(Cj, Co)J . (100)

j=l
To estimate Eq. (100), we have to reorder the pairwise error element s by sorting
the codewords according to their Hamming weights. Define W H (C i , Cj ) to be the
Hamming distance between two codeword s. Let Sw be the subset of codewords with
Hamming weight w: Sw = {CjklwH(Cj k, Co) = w} ~~l' where n w is the number
of codewords in Sw. The upper bound (100) can be expressed as
where Wfree is the free distance of the code, which is the minimum Hamming weight of
any codeword except Co. We can truncate the sum in Eq. (10 I) by ignoring negligible
terms having large Hamming weights. When the SNR is high, such an approximation
is quite accurate and we can approximate the average BER by
(102)
Applying Eq. (97) to Eq. (102), we obtain the approximate BER :
rv 1
n ] =
E [orb - exp (')'Wfr
--- ee )
2k 4
X
n~e
L.-
{BJ"1 Wfr
ITee 1
exp
[Wfreeu )2 ]}
L ,')'," )..JI(
.(103)
jl
l)
1=1 i=1 )1 + 4')'>,{1 i=1 2(1 + 4')'>'i
Gil E s Wfre e
We can also apply Eq. (97) to Eq. (10 I) to find an upper bound to the BER :
E[Pb] ;S 2
1k
w~ree { exp ( - ')'W~ree)
x L
n
w
[
BJi ITW 1 exp
(W
L >.jl(
i ')'u
2(1+ 4')'>'i l )
jl)2)]}
i
" . (104)
1=1 i =1 )1+4')'>'i 1 i=1
Gil E s.;
In Eqs . (103) and (104), >'i l is the ith eigenvalue of the covariance matrix C and x ui
is the sum of the elements in the corresponding eigenvector. To simplify Eq. (104) , we
can define the fading-induced degradation factors :
W 1
Q w = max IT V+ 4')'Ail"
Gil E S w, 19 $ n w i =1
(105)
1
and
>'il (,),ui l ) 2
(Jw = max
G j l ES w ,19$nw i =1
L
W
2(1 + 4')'>'i l)
". (106)
Then the upper bound (104) can be more simply expressed using Eqs. (105) and (106) :
(X)
E[H];S 2~ L BWQwexp(- ')': +(Jw')'2) , (107)

W=Wfre e
where B W is the sum of all BJi for which codeword C j l has Hamming weight w. Note
that BWcan be obtained using the transfer function of the code [32,33]. The upper
bound (107) can also be applied to long block codes for which Eq. (99) would involve
a prohibitive amount of computation.
6.3.1. Turbo Codes
Turbo codes offer excellent performance in a variety of applications, including optical

communications, and have aroused significant interest in the coding community. Turbo
codes are parallel concatenated convolutional codes (PCCC) in which the encoder is
formed by two or more constituent systematic recursive convolutional encoders joined
through an interleaver [32-35]. The information sequence is divided into blocks of
length equal to the interleaver length K . The input to the first encoder is the orig-
inal information bit sequence, and the inputs to the other encoders are interleaved
versions of the information block. The encoded sequence consists of the information
sequence and the parity check bits from all encoders. Many decoding schemes and
error-performance analyses for turbo codes have been documented in the literature .
An abstract uniform interleaver approach has been widely used to derive the average
of the upper bounds obtainable for the whole class of deterministic interleavers [34].
We start with the definition of the input-redundancy weight enumerating function
(IRWEF) for systematic convolutional code :
(108)
W,z
where i and j denote the initial and final states of the codewords. A~/ denotes the
number of codewords generated by an input information word having Hamming weight
w and having parity check bits of weight z, so that the overall Hamming weight of the
systematic codeword is w + z . We can also define the conditional WEF:
AW(w
'tt) ,
Z) == WWAC(Z)
't,J
= W W"""
L...,; AW'.zZZ.
'I.,)
(109)
Z
Making use of the properties of a uniform interleaver [34], we obtain the aver-
age conditional WEF of all possible turbo codes with respect to the whole class of
interleavers :
A C1(Z)A C2(Z)
A:;',j,(w,Z) ~ WW c.i (~)" , (110)
where A fj (Z) and A~~(Z) are the conditional WEFs of two encoders , respectively,
and K is the interleaver length. It has been shown when K is sufficiently large, we can
accurately approximate the error performance with the paths which diverge from the
Off-states of both constituent encoders and re-merge into Off-states after K steps [34],
whichhaveWEFAoo,oo(WZ) = L:w, z A;;'o~oWw Z Z.A tight bound for the pairwise
codeword error in correlated turbulence-induced fading channel requires knowledge
of the positions of differing symbols, as we discussed above. For simplicity, we can
loosen the bound by making the pessimistic assumption that all d = w + z differing
symbols are adjacent. Therefore the average BER is upper-bounded as
E[FtIS ~ ~A~6;'oP.(w + z.) ~ ~ t. [f. A~6~ow p . (d)] , (111)
where Pe( d) is the pairwise codeword-error bound of the all zero codeword and the
codeword with d adjacent On-state bits. Let us define
Communication techniques and coding for atmospheric turbulencechannels 339
•
10
-2
10
"':lc"
0
a:l
...
"0.0. 10
-4
;:l
~
:0
ee
-6
.0 10
2
e,
g .g
w 10
.!.
as --- -- T=O.OO 1 to
"eco -I. ----. T=OA to
10 - T=I to
<"
:>
KT=IOOOt 0
_Il
10
4 6 8 10 12 14 16
Average SNR : y = (l]lo)2/N o (dB)
Fig. 16. The simulated BER (lines with circles) and the approximateupperbound (lines without
circles) for (7. 4) Hamming codes versus average SNR. The log-amplitude variance CTx = 0.2
and the ratio of the adjacentcodewordbit interval T to the coherencetime of turbulence-induced
fading (T /TO) is chosen to be 0.001. 0.4. and I. respectively. Wealso show the simulationresults
where a block interleaveris used with KT /TO = 1000.
(112)
which allows us to express Eq. (III) as
E[Pb ] :::; L AdPe(d). (113)

d
We can simplify Eq. (113) by ignoring the negligible higher-order terms with larger
Hamming weights as discussed in section 6.2.2.
6.4. Numerical Simulation Results
In this section, we use the approximate error-probability upper bounds derived above
to numerically evaluate the performance of some practical codes .
We first study a Hamming (7, 4) code. In Fig. 16, we plot the approximate BER
upper bound (lines without circles) versus average SNR with log-amplitude variance
(J x =0.2 . T / TO takes on the values 0.00 I, 0.4, I, and 1000, respectively. We also
present the BER estimated using Monte-Carlo simulation (lines with circles) for com-
parison . Fig. 16 shows that when T /Toincreases, the average BER for (7, 4) Hamming
340 Xiaoming Zhu and JosephM. Kahn
Input ~I
Fig. 17. Rate1/3convolutional encoder with3-stageshift registers.
codes will decrease. Therefore, we can implement interleaving to compensate for the
coding-gain penalty due to the memory of channel. Consider the normal block in-
terleaver of degree K [22], where the codewords are interleaved so that the adjacent
bits of coded bit sequence are transmitted at intervals of KT. Figure 16 shows that
interleaving will improve the bit-error performance when KT / TO is large. However
when KT > TO, further increase of the interleaver depth will not significantly improve
the bit-error performance. As shown in Fig . 16, when T /TO increases from I to 1000,
the corresponding average upper bounds to the BER are very close. Therefore it is not
necessary to further increase the interleaver depth when KT > TO.
For convolutional codes , we consider the example of a rate-I13 code , whose en-
coder diagram is shown in Fig. 17. We choose the log-amplitude variance a x = 0.2.
T [t» takes on the values 0.001, I , and 1000, respectively. In Fig. 18, the approximate
BER versus average SNR, computed using Eqs. (103) (lines with squares), (104) (pure
lines), and (107) (lines with crosses) respectively. The simulated BER is also shown
(lines with circles) in Fig . 18 for comparison. We see that Eq. (107) yields a very
accurate estimate of the approxi mate upper bound (104) . Also, when the SNR is high,
the higher Hamming weight terms in Eq. (104) are negligible, and the approximate
BER (103) merges with the approximate upper bound (104) . Interleaving can help to
improve the system performance for convolutional codes as well. Similar to the block
coding case , there is not much difference in error performance when T / TO increases
from I to 1000. as shown in Fig. 18. We conclude that for convolutional codes , the
interleaver depth K is sufficient when it satisfies KT > TO.
Finally, we present simulations for a rate-I13 turbo-coded system whose encoder
structure is shown in Fig. 19. The approximate BER upper bound is calculated using Eq.
(113) with log-amplitude variance a x = 0.2 and T [ r« = 0.001 . The interleaver length
K of the uniform interleaver takes on the values 100, 1000 and 10000, respectively.
In Fig. 20, we plot the approximate BER upper bound (lines without circles) and the
simulated BER (lines with circles) versus the average SNR for comparison. Comparing
the simulation results shown in Fig. 20 to those of the rate-1I3 convolutional code in
Fig. 18, we see turbo coding can achieve better BER performance through atmospheric
turbulence-induced fading channels if the interleaver length is sufficiently long. In Fig.
20, we see that the error-probability performance of turbo codes continues to improve
with increasing interleaver length even when KT > TO, especially at low average
Communication techniques and coding for atmospheric turbulen ce channels 341
o
10
1=
10'" '-- '---- '---- -'-- -'- ...L\-_ - ' \- - - '
4 6 8 10 12 14 16
Average SNR: y = ( 'l lo)2/No (dB)
Fig. 18. Average BER for a three-stage, rate-l/3 convo lutional code versus average SNR with
log-amplitude variance o x = 0.2 and for TITo = 0.001 and I. The estimates are provided
by Eqs. ( 103) (line s with squares), (104) (lines with triangles), and ( 107) (lines with crosses)
respecti vely. The simulated average BER (lines with circles) is also shown for compariso n. .
SNR [34]. Thi s favors the use of longer interlea vers with turbo codes. However, system
complexity and dela ys in coding and decoding will limit the length of interlea vers in
practical systems using turbo codes .
6.5. Summary
Error-control code s can help to mitigate turbulence-induced signal fading in wire-

less optic al communication through atmo sphere turbulence. To study the efficiency of
various coding schemes , performance bound analysis has been used for its simplicity.
In this section, we first derived an upper bound on the pairwise codeword-error
probability for correlated atmo spheric turbulence channels. To avoid complicated mul-
tidimensional integrati on, we have also derived an approximation for this upper bound
under the assumption of weak turbulence. The accura cy and the limits of applicability
of this appro ximation have been demonstrated using numerical simulations. We then
applied this approximate upper bound to derive the error performance bound s and their
approximations for various cod ing schemes through atmospheri c turbulence channels,
includin g block coding, convolutional coding and turbo codin g. The analytical upper
bound s were then applied to compa re the performance of a few specific example code s.
The effect of varying the interleaver depth was studied through numerical evaluation
of the perfo rmanc e bound s.
x h-------------.x
Interleaver
length=K
'-------y/
1.-----_y2
Fig. 19. Encoderstructureof a rate-1I3 turbocode with uniforminterleaver of length K.
Conclusions
In this paper. we have discussed free-space optical links using intensity modulation
with direct detection (IMlDD). We have described several communication techniques
to mitigate turbulence-induced log-amplitude fluctuations (i.e., signal fading) in the
regime in which the imaging receiver's diameter Do is smaller than the turbulence
coherence length do, and the observation interval To is smaller than the correlation
time TO. We assume that the receiver has no knowledge of the instantaneous fading
state. We have described several classes of communication techniques:
When a single receiver has knowledge only of the marginal statistics of fading, a
symbol-by-symbol ML detector can be used to improve detection performance .
When a single receiver has knowledge of the joint temporal statistics of the fad-
ing, maximum-likelihood sequence detection (MLSD) can beemployed, yielding
a further performance improvement, at the cost of very high complexity. We in-
troduced a single-step Markov chain model for the fading correlation, and used it
to derive a low-complexity, sub-optimal MLSD based on per-survivor processing.
Spatial diversity reception with multiple receivers can also help to mitigate
turbulence-induced fadings. Using the joint spatial statistics of fading at multiple
receivers, we have derived the maximum-likelihood (ML) decision rule for spatial
diversity reception . ML spatial diversity is applicable when the fading correlation
properties are known, but the instantaneous fading state is known.
When a single receiver is employed, pilot symbol-a ssisted detection and error-
control codes can also be used to mitigate the impact of turbulence-induced signal
fading.
Communication techniques and coding for atmospheric turbulencechannels 343
o
10
K=I00
K=IOOO
K=IOOOO
":~::.::.:.::.---- ...
........... .
. ..
· 12
10 '---------'---------'-----'-----'------'-----~
o 2 468 10 12
Average SNR: y = ('1lo)2/No (dB)
Fig. 20. The simulated BER (lines with circles) and the approximate upper bound (lines without
circles) of a rate-1/3 turbo code versus average SNR with log-amplitudevariance a X = 0.2 and
T f ro = 0.001. A uniform interleavcr with different interleaver depths (K = 100, 1000, 1000)
is considered.
Acknowledgment
This researchhas been supportedby the DARPA STABProgramundercontract number

DAAHOI-OO-C0089 and the DARPA MTO MEMS Program under Contract Number
DABT63-98-I-OO18.
References
I. P.p. Szajowski, G. Nykolak, JJ. Auborn, H.M. Presby and G.E. Tourgee, "High-Power
Optical Amplifier-Enabled I550-nm TerrestrialFree-Space Optical Data-LinkOperating at
10 Gb/s", Proc. of IEEE Mil. Commun. Conf., Atlantic City, NJ, Oct. 3 l-Nov, 3,1999, pp.
687--689.
2. V.S. Hsu, "MEMS Corner Cube Retroreflectors for Free-space Optical Communications",
Masters Thesis, EECS Department, U.c. Berkeley, October 1999.
3. J.M. Kahn, R.H. Katz and K SJ. Pister, "Mobile Networking for Smart Dust", Proc. of
ACM/IEEE Intl. Conf. on Mobile Computing and Networking(MobiCom 99), Seattle, WA,
August 17-19, 1999.
4. Chu, P.B., Lo, N.R., Berg, E., Pister, K.SJ, "Optical Communication Using Micro Corner
Cuber Reflectors", Proc. of IEEE MEMS Workshop, Nagoya, Japan, 26-30 Jan 1997, pp.
350-5 .
5. L. Zhou, K.SJ. Pister and J.M. Kahn, "Assembled Corner-Cube Retroreflector Quadruplet",
to be presented at 15th IEEE IntI. Conf. on MicroElectroMechanical Systems, Las Vegas,
NV, January 2024, 2002.
6. Xiaoming Zhu, J.M. Kahn, "Maximum-Likelihood Spatial-Diversity Reception on Cor-
related Turbulent Free-Space Optical Channel s", IEEE Conf. on Global Commun., San
Francisco, CA, November 27-December 1,2000.
7. X. Zhu and J.M. Kahn, "Free-Space Optical Communication through Atmospheric Turbu-
lence Channels", IEEE Trans. Commun . 50 (8),1293-1300 (Augu st 2002).
8. X. Zhu andJ.M. Kahn, "Markov Chain Model in Maximum-Likelihood Sequence Detection
for Free-Space Optical Communication through Atmospheric Turbulence Channels", IEEE
Trans. Commun . 51 (3),509-516 (March 2003).
9. D. Sadot, A. Melamed and N.S. Kopeika, "Effects of Aerosol Forward Scatter on the Long-
and Short-Exposure Atmospheric Coherence Diameter", Waves in Random Media 4 (4),
487-98 (Oct. 1994).
10. S. Arnon and N.S. Kopeika, "Effect of Particulates on Performance of Optical Communi-
cation in Space and an Adaptive Method to Minimize Such Effects", App. Opt. 33 (21),
4930-37 (20 July 1994).
11. Max Born and Emil Wolf, Principles of Optics, Sixth edition (Pergamon Press, Canada ,
1980).
12. Joseph W. Goodman , Statistical Optics (John Wiley & Sons, Inc., 1985).
13. Akira Ishimaru, Wave Propagation and Scattering in Random Media , Vol. 1-2 (Academic
Press, Inc., 1978).
14. S. Karp, R. Gagliardi S.E. Moran and L.B. Stotts, Optical Channels (Plenum Press, 1988).
15. N.S. Kopeika, A. Zilberman and Y. Sorani, "Measured Profiles of Aerosols and Turbulence
for Elevations of 2-20 km and Consequences on Widening of Laser Beams", Optical Pulse
and Beam Propagation III, Proc. SPIE, Vol. 4271, 200 I.
16. A. Zilberman, N.S. Kopeida andY. Sorani, "Laser Bean Widening as a Function of Elevation
in the Atmosphere for Horizontal Propagation", Laser Weapons Technology II, Proc. SPIE,
Vol. 4376, 2001.
17. L.c. Andrews and R.L. Phillips, " Laser Beam Propagation Through Random Media", SPIE
Optical Engineering Press, 1998.
18. X. Zhu and J.M. Kahn, "Performance Bounds for Coded Free-Space Optical Communica-
tions through Atmospheric Turbulence Channels", IEEE Trans. Commun.51 (8), 1233-1239
(August 2003).
19. Shane M. Haas and Jeffrey H. Shapiro, "Capacity of Wireless Optical Communic ations" ,
IEEE J. Se!. Areas Commun . 21 (8), 1346-1357 (October 2003).
20. X. Zhu and J.M. Kahn, "Pilot-Symbol Assisted Modulat ion for Correlated Turbulent Free-
Space Optical Channels", SPIE's 46th Annual Meeting , The International Symposium on
Optical Science and Technology, San Diego, Californ ia, July 29-August 3, 2001.
21. Gordon L. Stuber, Principles of Mobile Communication (Kluwer Academic Publishers,
1996).
22. John G. Proakis , Digital Communication, Third Edition (McGraw-Hili , Inc., 1995).
23. M. Srinivasan and V. Vilnrotter, "Avalanche Photodiode Arrays for Optical Communication
Receivers", NASA TMO Progress Report 42-144 , Feb. 15,2001.
24. Ibrahim, M.M., Ibrahim, A.M., "Performance analysis of optical receivers with space di-
versity reception", lEE Proc. Commun ., 143 (6), 369-372 (Dec. 1996).
25. D.M. Boroson, "Interplanetary Deep-Space Optical Communication Systems and Tech-
niques" , Proc. of IEEE Lasers and Electro-Optics Society Annual Meeting, Tucson, AZ,
October 26-30,2003.
26. R. Raheli, A. Polydoros and C-K. Tzou, "Per-survivor processing : A general approach
to MLSE in uncertain enviroments", IEEE Trans. Commun. 43, 354-364 (Feb.lMar.lApr.
1995).
27. J.K. Cavers, "An Analysis of Pilot Symbol Assisted Modulation for Rayleigh Fading Chan-
nels", IEEE Trans. Commun . 40 (4),686-93 (Nov. 1991).
28. S.B. Bulumulla, S.A. Kassam, S.S. Venkatesh, "Pilot symbol assisted diversity reception for
a fading channel", Proceedings of the 1998 IEEE International Conference on Acoustics,
Speech and Signal Processing, Seattle, WA, USA, Vol. 6, pp.34 17-20, May 1998.
29. X. Zhu, J. M. Kahn and J. Wang, "Mitigation of Turbulence-Induced Scintillation Noise in
Free-Space Optical Links using Temporal-Domain Detection Techniques", IEEE Photon .
Technol. Lett. 15 (4), 623-625 (April 2003)
30. Francois Gagnon and David Haccoun , "Bounds on the Error Performance of Coding for
Nonindependent Rician-Fading Channels", IEEE Trans. Commun . 40 (2), 351-360 (Feb.
1992).
31. Gideon Kaplan and Shlomo Shamai, "Achievable Performance over the Correlated Rician
Channel", IEEE Trans. Commun . 42 (11),2967-78 (Nov. 1994).
32. Tolga M. Duman and Masoud Salehi, "Performance Bounds for Turbo-Coded Modulation
Systems", IEEE Trans. Commun . 47 (4), 511-21 (Apr. 1999).
33. Kamran Kiasaleh, "Turbo-Coded Optical PPM Communication Systems", J. Lightwave
Technol. 16 (I), 18-26 (Jan. 1998).
34. Sergio Benedetto and Guido Montorsi, "Unveiling Turbo Codes: Some Results on Parallel
Concatenated Coding Schemes", IEEE Trans . Commun . 42 (2), 409-428 (March 1996)
35. Eric K. Hall and Stephen G. Wilson, "Design and Analysis of Turbo Codes on Rayleigh
Fading Channels", IEEE J. Sel. Areas Commun . 16 (2),160-74 (Feb. 1998).
© 2005 Springer Science + BusinessMedia Inc.
DOl: 1O.1007/sI0297-005-0057-x
Originally published in J. Opt. Fiber. Commun. Rep. 2,558-602 (2005)
Optical communications in the mid-wave IR spectral band
Narasimha S. Prasad
NASA Langley Research Center. 468ILEOB

5 North Dryden Street, B 1202
Hampton, VA 23681-9403
Email: n , s . prasad@larc . nasa. gov
Abstract. The mid-wave IR (MWIR) spectral band extending from 3 to 5 microns

is considered to be a low loss atmospheric window. The MWIR wavelengths are eye
safe and are attractive for several free-space applications including remote sensing of
chemical and biological species, hard target imaging, range finding, target illumination,
and free-space communications. Due to the nature of light-matter interaction charac-
teristics. MWIR wavelength based systems can provide unique advantages over other
spectral bands for these applications. The MWIR wavelengths are found to effectively
penetrate natural and anthropogenic obscurants. Consequently, MWIR systems offer
increased range performance at reduced power levels. Free-space, line-of-sight opti-
cal communication links for terrestrial as well as space based platforms using MWIR
wavelength s can be designed to operate under low visibility conditions. Combined
with high-bandwidth. eye-safe , covert and jam proof features, a MWIR wavelength
based optical communication link could playa vital role in hostile environments.
A free-space optical communication link basically consists of a transmitter, a re-
ceiver and a scheme for directing the beam towards a target. Coherent radiation in
the MWIR spectral band can be generated using various types of lasers and nonlin-
ear optical devices. Traditional modulation techniques are applicable to these optical
sources. Novel detector and other subcomponent technologies with enhanced charac -
teristics for a MWIR based system arc advancing. Depending on the transmitter beam
characteristics, atmospheric conditions may adversely influence the beam propagation
and thereby increasing the bit error rate . For satisfactory transmission over a given
range , the influence of atmosphere on beam propagation has to be analyzed. In this
chapter, salient features of atmospheric modeling required for wavelength selection
and performance prediction is presented. Potential optical sources and detectors for
building a practical MWIR communication link are surveyed . As an illustration, the
design configuration and experimental results of a recently demonstrated free-space,
obscurant penetrating optical data communication link suitable for battlefield appli-
348 Narasimha S. Prasad
cations is discussed. In this case, the MWIR wavelength was derived using an all
solid-state, compact, optical parametric oscillator device. With this device, weapon
codes pertainingto small and large weapon platforms were transmitted over a range
of 5 km. Furthermore, imagetransmission through lightfog, accomplished using this
hardware, is also presented.
Advances in sourceand detector technologies are contributing to the development
of cost effective systems compatible with various platforms requirements. In com-
ing years, MWIR wavelengths are anticipated to playa vital role in various human
endeavors.
1. Introduction
The mid-wave IR (MWIR) spectral band is generally defined over 3-5 J-Lm. In this
spectral band, extraneous thermal and solar background emissions are relatively low.
Furthermore, there are several spectral sub-bands with relatively low atmospheric at-
tenuation in the MWIR region. Hence, the MWIR spectral band is considered to be
a low loss atmospheric window and hence is very attractive for several free-space
applications. MWIR wavelengths can be generated using lasers and nonlinear opti-
cal devices. The MWIR wavelengths are useful in probing various states of matter by
quantifying the effects on wavelength, amplitude, phase and polarization due to various
optical phenomenon including reflection, refraction, diffraction and interference. The
MWIR window plays a significant role in applications involving transmission of opti-
cal wavelengths through atmosphere over long ranges. Such applications include but
not limited to remote sensing of chemical and biological species, hard target imaging,
range finding, illumination, and free-space communications.
In general, for a given application, the characteristics of the atmospheric medium,
optical transmitter, remote target and receiver characteristics have to be evaluated. In
free-space based applications, the effects of atmospheric components on propagating
optical radiation parameters have to be ascertained to choose operational wavelengths.
Unwanted background emissions due to various sources and reduction in transmitted
beam intensity have to be minimized . In many practical applications, the influence on
optical amplitude or intensity over a given range is of primary importance. The opti-
cal radiation undergoes spectrally dependent attenuation due to various atmospheric
processes . The extent of attenuation depends upon a variety of atmospheric weather
conditions.
The use of an appropriate wavelength or wavelengths for obtaining maximum range
performance accompanied by detection schemes that permit precision amplitude mea-
surements provide valuable information. The spectral attenuation characteristics are
exploited in the atmospheric measurements of chemical or biological species . For, e.g.,
in Light Detection And Ranging (lidar) measurements, depending on the operational
scheme, the relative attenuation ofjudiciously selected two or more wavelengths with
respect to the molecular resonance features provide useful information about the con-
centration of desired species . In free-space communication applications , wavelengths
that undergo minimum attenuation under desired weather conditions to provide maxi-
mum range performance with sufficient signal-to-noi se ratio (SNR) are of importance.
Target characteristics in addition to spectral attenuation have to be considered for hard
target imaging and range finding applications. In any case, the atmospheric modeling
Optical communicationsin the mid-wave IR spectral band 349
has to be carried out to achieve optimum system performance and is discussed in the
following section.
2. Atmospheric Modeling
Absorption and scattering processes attenuates the propagating optical radiation .

Molecular absorptions are attributed to electronic, vibrational, or rotational transi-
tions, which manifest as sharp lines that are localized in wavelength. Water vapor and
carbon dioxide are two prominent contributors for atmospheric absorption. On the
contrast, broad featureless absorption bands are referred to as continuum absorptions
and are caused by the wings of vibrational and vibrational rotational absorption bands
or to collision-induced absorption bands. Backscattering off airborne particulates such
as dust is often referred to as aerosol scattering, which also contributes to atmospheric
attenuation. The primary source of dust is the terrestrial soil carried by wind. How-
ever, the aerosol particulates could be hygroscopic in nature that generally exists in
the vicinity of water. Hygroscopic aerosols are typically tars and resins created by the
oxidation of vegetation extruded aromatic hydrocarbons or sea salts that are injected
into the atmosphere at ocean surfaces. The mean size of hygroscopic particles in-
creases with increasing humidity. Scattering efficiency increases with particulate size.
The atmo spheric attenuation due to aerosol scattering is inversely proportional to the
wavelength or in other words it becomes weaker as wavelengths increase.
Atmospheric modeling allows the selection of appropriate wavelengths for a given
application. Atmospheric transmission can be modeled using FASCOD3 (Fast At-
mospheric Signature CODe) and FASCOD3p . Both simulation codes were developed
underthe sponsorship of the U.S. Air Force Geophysics Laboratory (AFGL) (Hanscom
AFB) and represent the culmination of several decades of research . FASCOD3 is the
newest version of the transmission code and was released in 1996. The codes provide
an "exact" or line-by-line radiance and transmission model that calculates the effect
of the Earth's atmosphere on the emis sion and propagation of electromagnetic radi-
ation. The codes are anchored by atmospheric measurements to the extent allowed
by available databases. Transmission losses due to molecular absorptions, continuum
absorptions, and aerosol scattering can be calculated either separately or aggregately
by FASCOD [I]. Either version ofFASCOD uses HITRAN molecular spectro scopic
databases to generate high-resolution spectral transmittance data. Each version of HI-
TRAN contains line-by-line absorption data, including absorption cross-sections, for
32 (HITRAN'92) or 38 (HITRAN '96) most prevalent molecular species . The database s
also include aerosol data. FASCOD3p interrogates the HITRAN database and then ap-
plies empirical line-broadening algorithms upon the HITRAN molecular absorption
bands. These line-broadening algorithms take into account many simulation parame-
ters such as, the simulation path, the model atmosphere, visibility, rain, aerosol models
etc. The resolution of FASCOD is many orders of magnitude lower than LOWTRAN
and is predominately a function of the resolution of the HITRAN databa se.
The atmospheric transmission can be calculated using Beer's Law and is given by
T(R) ~ exp ( - ! a(r) dr) , (I)

Atmospheric T ranu n_ion
1.0, - - -- - - - - -- - - - - - - -- - -----.
09 t - - - - - - - - - - -- - - -- - -- - - - - l
o.sr-- - -- - - - - - - -- - - -- - --rr+MhH-iI--i
0.7r-- - - - - - - - - - - - - - - - ...,...-
.§ 0.6+-- - - - -- - -- - - - - - - - -M-
.! O.0t--- - - -- - - -- - - - - - r H I I I -
L 4 t-- - - - - - .-- - - - - - - - ..,..jj.-!II-
0 .3 t - -- - -- - -ti-:,------- - -- j- +lII-Jill--Jl!I-
0.2 +------~
0.11rt---,-- - - -l--1h
3 .05 3 .1 3.15 32 3.25 3.3 3 35 3.4 3.45 3.5

W.,~in.nUcl'OJUl
Fig. l. Atmospheric transmission from 3 to 3.5 /lm .
AtJnDsphs.it T ranlrniam n
."
•.
' .7
.8
·1 .5
•
...! ••
OJ
ez
OJ
3 .55 3£ 3.65 3.7 3.75 3.8 3.95 3 .9 3.95

Wao,·eImc;;th in microns
Fig. 2. Atmospheric transmission from 3.5 to 4 /lm .
where a(r) is the extinctio n coefficient profile along the path [O,R] and R is the range
from the transmitte r to the receiver. The extinction coefficient consists of a combination
of aeroso l and molecular extinctions and can be modeled using AFGL's FASCODE
coupled with the HITRAN data base. Figures I to 4 illustrate atmosp heric modeling
using FASCOD3p in the MWIR spectra l band. It is seen that several sub-bands with
relatively high atmospheric transmission over 3.4 to 4 J.Lm exists.
Besides FASCODE, Line-by-Li ne Radiative Transfer Model or LBLRTM is
anothe r code available to comp ute atmospheric transmis sions. The information on
LBLRTM can be accessed from www.rtweb.aer.com.This code was developed
Optical commun ication s in the mid-wave IR spectral band 351
Atm.upheric Transmission
0 .9
0 .8
0.7
~
'~
,6 0.6
T"~
'i 0.5
~ 04 r"\
0.3
0.2
0.1
,
o
,l lj~
4 4.05 4 .1 4.1:5 4 .2 425 4,3 <1 .35 44 4.45 4 .5
Wav t"~ inmi:rons
Fig. 3. Atmospheric transmission from 4 to 4.5 J.Lm.
Atn~sprerif. Transnrssien
lb ,-- - - - - - -- - - - - - - - -- - - - -----,
,. -1--- - - - - - - - - - - - - - - - - -- ---1
.. -I---------------------~
" -I---------------------~
,~ "' - I - - -- -h-IJIo-,H ---.----M-I I - - - - - -- - - - - - ---1

.~ 05 +-- - -
~ .. +-- -iI+-
e
01
".'15 <.•
W:nrelength in mirrons
Fig. 4. Atmospheric transmission from 4.5 to 5 J.Lm .
by Atmospheric and Environmental Research Inc. (AER) with the support of Depart-
ment of Energy Developed on the legacy ofFASCOD. The LBLRTM uses the HITRAN
database and is considered to be efficient and highly flexible than FASCOD.
Plots of the one-way atmospheric transmission at a representative wavelength of
3.8 J.lm wavelength versus range, calculated using FASCOD3p using a Mid-Latitude
Summer model with rural haze (5 and 23 km) visibility, no rain, and at sea level is
shown in Fig. 5. The extinction profiles (extinction coefficient versus height) for 3, 4,
and 5 J.lrn wavelengths are shown in Fig. 6.
O. U
0.'
-<. ------- --
,
........
0."
<. ---------
<.
"'-~
s.:
0.6 i
[. .. ... .I,I "' <,
_ ~ 1 "" :1 '( ' .
~
0.$
o 0:
F:'lIl" ill bn
Fig. 5. One-way transmission at 3.8 J.Lm for 5 and 23 km visibilities .
Atm osphoric Extin ction . MLS rural 23 km ylslb liity

100 ,- - - , - - - - , . . -- - - - r - - - - r - ;= ====c;]
- J,.m
•••. 4 I1J11
90 ... _. ~ "'"
so
10 <:
"
-.
60 '\
' .
.:: 80
..e
-e 40
30
20
10
1
10 10..
E xtinction (not. Jkm)
Fig.fi, Extinct ion profiles at 3, 4, and 5 J.Lm.
2.1. Atmospheric Turbulence
Laser based optical sources generate Gaussian intensity beams. The TEM oo spatial
mode is normally selected for applications involving beam propagation in free-space .
Opticalcommunications in the mid-wave IR spectralband 353
However, the Gaussian beam spreads and breaks up into beamlets as it propagates
through the atmosphere due to turbulence . Atmospheric turbulence is the result of
erratic air movements arising due to wind and convection processes driven by temper-
ature differences. The effect of atmospheric turbulence has to be carefully considered
when estimating the system performance. [2,3] The two parameters that characterize
the effects of turbulence are Rytov variance, which describes scintillation, and co-
herence length, which affects beam spreading and determine s the size of hot spots in
the detector plane. Salient features of these parameters are discussed in the following
sections.
Simulation of beam propagation through turbulence can be done numerically using
the Fresnel approximation to the Huygens-Fresnel integral. [3-5] Two-dimensional fast
Fourier transforms (FFfs) allows efficient implementation of Fresnel approximation
through numerical means. To incorporate the effects of turbulence into the simulation,
the propagation is done in multiple steps and turbulence is added after each step. The
turbulence is simulated using phase screen s. Phase screens are random realization s of
phase noise that the beam encounters as it propagates. The phase screens are generated
by assuming that the phase noise follows a specified atmospheric power spectral density
(PSD) . Spatially correlated Gaussian noise with the correct statistics must be created in
order to properly generate phase screens to simulate propagation through turbulence .
Simulations can be carried out using several different types of phase screens , namely,
Kolmogorov, Tatarskii, Von Karman, and Hill phase screens .
In order to simulate the propagation of beam through atmospheric turbulence, a
turbulence spectrum is selected to model the atmo sphere. Then, random phase screens
are generated and the phase values of the numerically sampled beam are changed by
those values to simulate the turbulence . It can be shown that these phase screens distort
the beam significantly over a path of few kilometers. Both truncated as well as untrun-
cated Gaussian beams can be used for modeling. However, the numeric propagation
of an untruncated Gaussian beam is very accurate, but the numeric propagation of a
truncated Gaussian beam introduces slight errors in the field magnitude .
The beam distortion due to atmospheric turbulence simulated using Hill's phase
screens for collimated untruncated Gaussian beams is illustrated in Fig. 7. For this
simulation, a wavelength of 3.5 {lm, transmission range of 2 krn, and a C~ value of
IE - 13 m- 2 / 3 are assumed. It can be shown that beam distortion increases with
turbulence .
2.1.1. Rytov Variance
The Rytov variance is a measure of the strength of scintillation. The scintillation is con-
sidered weak when the Rytov amplitude variance is less than 0.3. In the weak regime,
where log-normal statistics apply, the normalized intensity variance [var/mean"] or
scintillation index is approximately four times the log-amplitude variance (o} rv 4
(1~), where A = exp[xl is the amplitude, X is the log-amplitude and I = A 2 is the
intensity. For a Kolmogorov spectrum and a spherical wave model, the log-amplitude
variance is given by [I]
f C~(r)
R
(1~ = O.56e/
6
w(r) dr, (2)
o
..
; ;
l. "
•I '
.... .., _N
• " n .. 04 U .., _N
I II 11 I) ,t II
Fig. 7. The effect of atmospheric turbulence on beam propagation. Top left: The spatial beam
profileof size 2 em diameterat the transmitteraperture. Topright:The anticipatedbeamprofileat
a range of2 km in vacuum (under no turbulence). Bottom: The beam distortion due to turbulence.
where C~ (r) is the refractive index structure constant profile along the path, rand k=
27r/>... The weighting function, w(r), is, for a spherical wave, independent of the path
direction because it is symmetric about the middle of the path and is given by
w(r) = [r(l- r /R)] 5/ 6 . (3)
For a constant C~ and a spherical wave, Eq. (3) becomes
a~ = 0.1235eI6C~Rll /6 . (4)
Figure 8 shows plots of the Rytov variance versus range parametric in C~. Values of
Rytov variance below 0.3 is usually indicate weak turbulence effects.
2.1.2. Coherence Length

The coherence length is a measure of the transverse dimension over which the electric
field is coherent. It is the exp(-I) point of the electric field autocorrelation function .
For a spherical wave and a Kolmogorov spectrum, the coherence length is given by
(5)
Opticalcommunications in the mid-wave IR spectral band 355
••
,.o
'!:. . ....
'" - - C. '= l E. 14 m
'E~
- - C. -= l E. 13 m
- - C. ' = l E. 12 m '
'E'" -fL-~- -.-~---r-_---!:===r===;=,=;;===;==:d.1
R.... ge j .. km
Fig. 8. Plotsof Rytovvariance as a function of range for different values of constant C~ . When
the Rytov variance is belowthe dashed line at 0.3, the turbulence is considered weak,
The weighting function depends upon the path direction. For transmitter to receiver
paths a reverse weighting function, W r (r) is used :
(6)
This weighting function weights C~ near the receiver more heavily than near the trans-
mitter. The reverse path weighting is used to estimate the coherence length affecting
the transmitted beam diameter at the receiver plane . A forward path weighting that
weights C~ near the transmitter more heavily than near a target is given by
(7)
Equation (7) is used to estimate the coherence diameter in a target plane . For a free
space communication system (transmitter to receiver paths), the reverse weighting
function is appropriate. For a constant C~ and a spherical wave, however, Eq . (5)
simplifies to
3] - 3 /5
Po = [2 .91k2C~R8 (8)
Figure 9 shows plots of the coherence diameter (twice the coherence length or 2 x po)
versu s range for difference values of constant C~ .
2.2. Beam Diameter
Taking into account near and far field effects as well as transmitter focus (F), and
refractive turbulence, a Gaussian beam with initial beam diameter Db(O) has a beam
diameter at range R given by
356 Narasimha SoPrasad
- - c =IE -1 4m
roo .
- - c =I E-13 m
~
IE , ,
II - - c = I E- 12m
.
.5
:
\~
III
'0
IE
...
------
oil!
III
II \ ~-------------
.."
'~-------..._---~----------
III
III
""
u•
0 0
'
o 2 II
Rahge i h km
Fig. 9. Plots of the coherence diameter (2xpo) as a function of range for different values of
constantC;;.
(9)
The first term in the square brackets of Eq. (9) determines the near field effects
and effect of focus on the beam diameter. The next term describes the far field effects
and the last term describes the effects of coherence length (po) on the beam diameter.
Thus, for a collimated (F = 00) Gaussian beam, Eq. (9) becomes
D (R) = 2[(Db(0))2 R
2>.2 (_4_ 2-)] 1/2
(10)
b 2 + 1l'2 D~(O) + P6
When the coherence diameter (2po) becomes similar in size to the initial beam diameter,
the beam diameter starts to increase more quickly than it would for a diffraction limited
system where Po = O.
Figure 10 shows plots of beam diameter versus range for collimated (F = 00)
Gaussian beams with I cm initial beam diameter propagating through various amounts
of refractive turbulence. The beam diameter versus range for a diffraction limited
collimated Gaussian beam is also plotted in each panel. Note that, for a refractive
index structure constant of C~ = 1 x 10 -1, when the coherence diameter (2po)
becomes larger than the initial beam diameter of I cm at a range of approximately
4 km the area of the beam is approximately doubled (i.e., the beam diameter is J2
larger) compared to that of the diffraction limited system. If the transmitted beam is
focused at each range, the equation for the beam diameter becomes
(II)
!l00
- - C. '=lE -14m
- - C.' =l E- 13 m '
E ..00
."
...
- - C.' =l E- 12 m '
Q 3 ....
......:.
C
.
E 21>..
E
.
.. tOO
III
o 2 ..
Range in km
Fig. 10. Beam diameter versus range for collimated Gaussian beam under turbulence ,
Note there is very little difference between a transmitted beam that is focused at each
range and a collimated transmitted beam (focused at infinity). This is due to the fact
that the Rayleigh range is around 50 m for a l-cm beam diameter and around 0.5 m
for a I-mm beam diameter. Thus, most ranges in these cases are in the far field where
the beam diameter will be affected more by diffraction and refractive turbulence than
by focusing at the aperture .
2.3. Direct Detection SNR
Estimate s of signal to noise ratio (SNR) for each range can be found by using noise
models for a particular detector (e.g. an avalanche photodiode) and combining these
with the calculated received power Pr(R) in the form
SNR (R) - [lRPr (R) M]2 (12)

- [(lRPr (R) + lRH g + [db) M2 F + Ids]2qB n + NsoB n '
where lRis the detector responsivity given by
lR = 'f)qq (13)
hv
and q = 1.602 X 10- 19 1 is the electron charge, h = 6,626 X 10- 34 1s is Planck's
constant, and v is the frequency of radiation .
The received power at range R, incident on an aperture of diameter Dr can be
estimated by
P(R) r = Px'f)T(R) 'f)o(R), (14)
where P« is the power of the transmitted beam, T( R) is the atmospheric transmission
at range R, 'f) is the combined transmitter/receiver optical efficiency, and no is a range
dependent overlap (between detector area and beam area) efficiency given by
358 NarasimhaS. Prasad
(15)
When the detector diameter, Dr , is significantly smaller than the beam diameter at
range R(D r « Db), the argument of the exponential in Eq. (15) becomes much
smaller than unity. Thus the approximation
exp ( - x) ~ 1- x for x« 1 (16)
can be used to find an approximation for received power incident on the detector:
(17)
where
R2Qx = A xr = 7rD~ (R) (18)
4
is the area defined by the diameter of the transmitted beam at range R.
2.4. Large Scale and Small Scale Turbulence-Spherical Waves
The scintillation index, defined by (pp. 236-237 of [I]):
2( R) _ (12 (r,R)) -1- r 4 (r,r,r,r,R) -1 (19)

aj r, - 2 -
(1 (r, R)) n ( r, r, R) rdr, r, R) ,
where 1 (r, R) is the irradiance at point r in the field at range R, r 4 (ri , r2, r3, r4, R)
is the fourth-order coherence function of the field at R, and n
(rr, r2, R) is the mutual
coherence function of that field. The irradiance at a given range can be modeled as a
product of large scale (x) and small scale (y) irradiance fluctuations written as
1 = xy, (20)
where the large scale and small scale irradiance fluctuations have gamma distributions
given by
px ( X )
-
-
o(ox)( n -l )
r(o) exp ( -ax ) x >0
(3( (3 ) (13- 1)
(21)
py (y) = ~((3) exp (-(3y) y>0
The overall irradiance will therefore have a gamma-gamma PDF, written as
(22)
The a and (3 parameters in the distribution(s) can be identified with large-scale and
small-scale scintillation as
2 1 1 1 2 2 22
ai = ~ + 73 + a(3 = a x + a y + a xa y, (23)
where a; describes the large scale scintillation and a~ describes the small-scale scin-
tillation. Reference [2] presents the process for finding a form for the scintillation index
dependent only on the Rytov variance. The result is
a/
2
= exp [ 0 . 4
9B 5 7 /6 + 0.51B 5 5/6 ] -1, (24)
(1 + 0.56B62/5) (1 + 0.69B62/5)
where
(25)
is a strength of turbulence parameter approximately equal to 4 times the Rytov variance
defined in Eq. (2).
Contributions to scintillation from small-scale fluctuations are related to turbulent
J
cells that have asize less than the smaller ofeither the first Fresnel zone, LFz = Rfk,
or the coherence radius, po. Contributions to scintillation from large-scale fluctuations,
however, are related to turbulent cells that have a size greater than the larger of either
the Fresnel zone or the "scattering disk", where the scattering disk is defined as
R
LSD = kpo ' (26)
The plots of all three of these parameters (coherence length, scattering disk, and Fresnel
zone) . help determine the contributions of different types of cells to either large scale
or small scale fluctuations for a given refractive turbulence parameter. Thus, turbulent
cells that are smaller than the coherence length will contribute most to the small-scale
scintillation at these ranges . If the scattering disk is greater than the Fresnel zone, it
indicates that contributions to large-scale scintillations will come from cells that are
larger than the scattering disk. The Fresnel zone is the limiting factor in determining
which turbulence cells contribute to large-scale scattering. One can see that large-scale
scintillation has a greater influence (more cell sizes will contribute) in situations with
smaller refractive turbulence structure constants . We see these effects when the mean
SNR for small aperture diameters is smaller for low refractive turbulence than it is for
higher refractive turbulence .
2.5. Bit Error Rate (BER) and Minimum SNR
The Bit Error Rate (BER) is a parameter that is often used to characterize the perfor-
mance of a communic ation system . In this section, the relationship between BER and
receiver SNR for different receiver aperture diameters and different refractive index
structures, C; is discussed . The main conclusion to draw from this section is that the
mean SNR at a given range and given C;is highly dependent on the receiver aperture
diameter and the scintillation index. The scintillation index for each range in tum de-
pends on C;
and receiver aperture diameter. Hence, care must be taken in choosing
the correct combination for the particular range(s) of interest. In general, however,
increasing the receiver aperture diameter will improve mean SNR unless the refractive
turbulence is high (e.g., C; = 1 x 10- 12 ) in which case the receiver SNR is low at
most ranges and so the mean SNR saturates with increasing receiver aperture diameter.
The bit error rate (BER) or probability of error in an on-off keying diffraction
limited communication system is given by
SER \'S . SNA (dB)

10'
10~
10-'
II:
w
<0
10
-l ;
\
10""
10-1::
0 5 10 15
SNR (dB)
Fig. 11. Plot of theoretical bit error rate (BER) vs. SNR.
BER = ~erfc (S~~R)) . (27)
For a maximum required BER of I x 10- 9 , the minimum required SNR is about 21
dB as can be seen in Fig. II .
When refractiveturbulenceis presentin thecommunicationsystem, the probability
of error must be averaged over the probability distribution function of the signal, 8 .
Thus the BER becomes (pp. 236-237 of [2]):
1
BER = -2
/00 PI (sjerfc ( (SNR) 8)
v'2 1/' 2 ds . (28)
2 2 (zs)
o
where the probabilitydensity function, PI (8), is givenby a gamma-gammadistribution:
(o+~) (O !~ )
PI (8) = r
2 (a,8)
r
(a) (,8) (is)
2 [8]
(is)
-1
K o- {3
(2 ~)
af38
(iJ 8> O. (29)
and KvO is a modified Bessel function of the second kind. The a and f3 parameters
for the gamma distributions are found using:
0 .4 9{36
~ =exp
(I+O .18d2+0 . 56{3~2 /5) 7/6
(30)
1 _ 0 .5 1{36 (I+O . 69{3~2 /5r 5/6 ,
73 - exp (I+0 .90d2+0.62d2 {3~ 2/ 5 )
where
B6 = 0.494e /6 C~ R ll/6 (31)
is a strength of turbulenceparameterapproximatelyequal to4 times the Rytov variance,
and the term d is dependent on the aperture diameter Dr and range, R, according to
d = JkD;' (32)
4R '
The parameters 0: and (3 can be thought of as representing the effective number of large-
scale cells and small-scale cells, respectively. The mean SNR, (SNR), is determined
by
(SNR) = SNR (33)
~ + a; (Dr) SNR'
where SNR is the receiver SNR defined previously and Prd / P; is a power reduction
factor consisting of the ratio between the received power in a diffraction limited system
and the received power when optical turbulence is present. This ratio is given by
Prd ( 12 / 5 2R )
Psr = 1 + 1.63a x kw2 (R) , (34)
where
2
w2 (R) = Wo
2
+ -R2>..2- ( 2"
1 + 2"
1) (35)
1r Wo Po
is square of the beam radius at the receiver. The scintillation index ai, in Eq. (33) is
given, for a spherical wave, by
2 [ 0.49(35
a[(D) = exp 7 /6
(1 + 0.18d + 0.49(36 2 / 5 )
2
+
(
0.51(35 1 + 0.69(36
2 5
/ )-5/6] -1 (36)
1 + 0.90d 2 + 0.47d 2 (362 / 5 .
In summary, the (SNR) of a received signal is dependent on the scintillation index

of that signal which in tum is dependent on the refractive index structure constant,
C;, the receiver aperture diameter, and the range to the receiver. A smaller refractive
index structure constant (C; = 1 x 10 - 14 ) along the path will result in breakup
of the beam into larger section s that may miss a small receiver aperture altogether.
A larger refractive index structure constant (C; = 1 X 10- 12 ) , however, will break
the beam up into smaller sections, increasing the probability that there will be signal
on the detector at a given time, thus increasing the (SNR) and decreasing the BER.
If the aperture diameter is increased , however, the probability that the large beam
sections (from C; = 1 X 10 - 14 along the path) increases and the (SNR) becomes
larger than that for a beam that is more broken up (from C; = 1 X 10 - 12 along
the path) . Another way to look at this is to notice that the scintillation index for a
beam propagating through strong refractive turbulence (C; = 1 x 10- 12 ) is fairly
insensitive to an increase in receiver aperture diameter whereas the scintillation index
for a beam propagating through weaker refractive turbulence (C; = 1 x 10- 14 ) is
highly sensitive to increases in receiver aperture diameter.
The C; and hence the scintillation index can be measured using a scintillometer.
In a typical arrangement, the scintillometer consists of a transmitter and a receiver.
These two units are placed at a convenient distance say I km at a site where the C; has
to be measured . Figure 12 shows a typical measurement in which the refractive index
structure constant, C;was measured using a scintillometer at I Hz rate. In this figure,
the C; level remains between 1 x 1013 and 2 x 1013 m- 2 / :1 during the times shown.
SCintilla tio n Ind e ., May 2 9 . 2003
~<:
~
~
;§ 0 .5
oLL.. ......... -'- --L_--J
5.84 5 .B6 5 .BB

T ime (s)
Fig. 12. The C; measurements using a scintillometer is shown. The computedscintillation index
using Eq. (19) for a time interval (.1T) of8 seconds is also plotted.
3. The MWIR Optical Sources
Following the selection of an appropriate candidate wavelength or wavelength win-

dows, the hardware availability has to be carefully considered. The MWIR wavelengths
are considered as eyesafe . [7] The source characteristics have to be blended into the
choice of an optimum operational wavelength . Often, this could be an iterative process.
An optical transmitter generally consists of an optical source, drive electronics, and
an output telescope or beam shaping optics. Sometimes scanners could be a part of
the transmitter unit. Lasers and nonlinear optical devices such as optical parametric
oscillators can be utilized for the generation of MWIR wavelengths . Due to the physi-
cal nature of operation, lasers generate discrete wavelengths in the MWIR region with
very limited tunability. Some of the considerations that go into selection process of an
optical source include overall output power (average or peak), CW or pulsed mode of
operation, spectral width, reliability, size, weight, operational elegance, environmental
compatibility and cost. In the case of pulsed operation, PRF, pulsewidth and pulse en-
ergy parameters have to be looked into. In this section, a brief description of different
types of MWIR sources is presented.
The most common MWIR source is the Helium Neon laser. The Helium Neon gas
laser operating at 3.39 J1m in CW mode is commercially available from several ven-
dors. However, Helium Neon gas lasers can only provide limited output powers (tens
of milliwatts) and furthermore they are bulky and less versatile. These lasers can be
externally modulated using acousto-optic or electro-optic modulators . For pulsed oper-
ational requirements, theses lasers arc of limited use. Hence these lasers are generally
used in laboratory experiments. Another laser type is XeHe that generates wavelengths
over 2 to 4 J1m band with a prominent emission at 3.506 J1m . However, so far XeHe
lasers have not been practical as far as applications are considered.
3.1. Semiconductor Based Lasers
Semiconductor-based lasers are also utilized to generate the MWIR wavelengths . In

this category, there are three classes of devices namely, lead salt lasers, III-V strained
Opticalcommunications in the mid-wave IR spectralband 363
quantum-well lasers, and quantum cascade lasers . Small size, wavelength diversity and
ease of modulation are primary advantages. Disadvantages of these devices typically
include cryogenic operation, low duty cycles, and low efficiencies. However, the tech-
nology of these devices is slowly maturing that minimize these disadvantages. Each
of these classes of devices is discussed in greater detail in the following sections.
3.1.1. Lead-Salt Lasers
Lead salt lasers are based on IV-VI semiconductor materials. They are so called be-
cause majority of these compounds contain lead . These lasers operate at wavelengths
ranging from 3.3 to 29 /Lm. Devices operate exclusively at sub-ambient temperatures.
Typical operation is at cryogenic temperatures with average output powers of tens of
milliwatts. [8-10] Limited pulsed operation of a MWIR device has been observed at
room temperature. However, robust operation of devices in the 3-10 /Lm wavelength
range is restricted to liquid nitrogen temperatures. Average output powers can be as
high as tens of milliwatts. However, device efficiencies are well under I %. Efficien-
cies are limited by low internal quantum efficiencies and high material losses . Room
temperature lead salt laser research is in the exploratory stage.
3.1.2. III-V Strained Quantum-Well Lasers
Quantum well lasers (QWL) are commonly used in the Short -Wave InfraRed (SWIR)
telecommunications windows at 1.31 and 1.55 usc: They are less developed and less
robust at MWIR wavelengths with longer wavelength CW operation limited to cryo-
genic temperatures. The longest reported emission wavelength of a III-V QWL is 4.5
ILm. [II] The device lased in pulsed mode at temperatures up to 144 K and in CW
mode at temperatures up to 105 K. With electrical pumping, cw output powers of 30
mW were produced at 70 K. Electrical-to-optical efficiencies were less than I%. The
development of MWIR QWLs is difficult because the valence-band offset and refrac-
tive index separation between the active and barrier layers must be made smaller for
operation at longer wavelengths. Growing devices with appropriate characteristics is
difficult. This limitation may be due to non-uniform carrier distribution in the quantum
wells at elevated temperatures.
3.1.3. Quantum Cascade Lasers
Quantum cascade lasers (QCLs) are fundamentally different from diode lasers . [12-
13] Diode lasers generate light when conduction band electrons and valance band
holes radiatively recombine across a band gap that exist s because of the presence
of discrete energy bands in the semiconductor. QCLs generate light when electrons
(only) make transitions between conduction band states (no valance band) arising
from size quantization in semiconductor heterostructures. QCL devices are created
using molecular beam epitaxy and band structure engineering. QCL are waveguide
devices with core and cladding regions . Continuous-wave (CW) operation of QCLs
is limited to cryogenic temperatures. Pulsed devices have been demonstrated at room
temperature. These devices output nsec duration pulses at kHz to MHz repetition
frequencies. QCLs operate better at lower temperatures, since (I) laser gain decreases
State of the art: aCL performances
AlJ l l. ....p lh.'li c w indnv....

---- -e-e-
• 0 \' j lnr
• P" ''''<!
•
•
0 \'
p" ["" I
lf (.;.,,\,
· Got'll Mid -II< ClW<'J~ .

• Tl'T ITtI.'f U promi-..lng
~::Ntul ~ ~1)1 j"';-

,u f""'t-.-,.. lo.l:r
,......
.....I l . . U.. ifoo-.<... ...l
TL:\
N ,'UO-.. ... U'"
., ; "'J.....L. ,'T\.. .....-e.
\\' a vd en · th un
Fig. 13. The state-of-art performance of QCL devices (Courtesy of Alpes lasers).
l N2...cW·OFQ.1o.X lN1'·CW·Of B-tO-X
~
..~
__ ... i
u ,;"
~ , '
(. _
.rr.-
"_ 1 ..
_
JIOOnO '. "h...,
, '
CJIt
~- _
_
... -
11 .. _
.. . . ~
_ .. _
......
,.'
j.
- ...
- ....
a.
• •
IQto ..
'h• ...,
~.:,,.
~ _ ll
~
....-"'two _- . _. ,.)!<S:
~lol \.l.oh
"1Wl
.,
• Q.1 • • llJo
<"nfI I IAI
I. I': U
.; l ."
::, .. 1M :I N ",. II:": I n. :ur.
w...·-............ l<'WlJ
~7i';..,..I!N, __'..m.'
"'.'.H,..... ...'...
·1 '''l f ''' ' '' 6U .~ ~
1'", l' ... · 4 . 1.... "' ~
.. ..
Fig. 14. The MWIR QCL laser marketed by Alpes lasers.
with increasing temperature due to an increase in the width of the gain spectrum, (2)
the laser's upper state lifetime also decreases with increasing temperature due to the
presence of larger numbers of optical phonons at elevated temperatures, and (3) the
thermal population of the laser's ground state increases with increasing temperature.
QCLs tend to have high thresholds (compared to diode lasers) due to high waveguide
losses. The origin of these losses is not well understood. Typical electrical-to-optical
efficiencies for QCLs are approximately 0.1%.
Currently,Alpes Lasers fromSwitzerland (www .alpeslasers .ch) is a leading
manufacturer of QCL lasers. Pulsed DFB QCL on Peltier cooler have been commer-
cialized for wavelengths in the range 4.3 to 16.5 J-Lm as illustrated in Figs. 13 and
14.
3.1.4. Solid-State Lasers
There are a variety of solid-state laser options in the MWIR spectral region. Available
and potential sources in both the MWIR spectral region are reviewed below.
There are few demonstrated solid-state lasers in the MWIR . None are tunable .
Reported systems include a cryogenic Er:YLF laser at 3.41 J-Lm, [14] a room temper-
ature Er:ZBLAN fiber laser at 3.5J-Lm,[15] a Ho:ZBLAN fiber laser at 3.9 J-Lm [16]
and cascade laser emission in Ho:YLF at 2.9, 3.4, and 3.9 J-Lm [17] These systems and
several potential alternatives are reviewed below.
The Er:YLF work was done at 77 K and the laser had a 2% slope efficiency. The
Er:ZBLAN laser was demonstrated at both reduced and room temperatures. Room
temperature output powers were a couple of milliwatts for pump powers, at 488 nm,
approaching I W. The probability of adapting this fiber laser in a hardened package is
low. Problems include the lack of a convenient demon strated pump source; the fragility
of ZBLAN fiber; and difficulties in stripping, cleaving , and generally working with
ZBLAN fiber. Additionally, the Er:ZBLAN laser is expected to offer lower overall op-
tical efficiency than some of the nonlinear frequency conversion techniques discussed
in the next section.
The Ho lasers are cascade devices that function by lasing multiple transitions
simultaneously. The YLF work reported stimulated emission but laser energies were
too low to measure . The fiber work is more intriguing but has only been demonstrated
at cryogenic temperatures (below 150 K). Laser power at 77 K was limited to I I mW
for 880 mW of launched pump power. The laser was pumped by either a dye laser at
640 nm or a Ti:Sapphire laser at 890 nm. In our opinion , these lasers have limited use
at room temperature due to low fluorescence quantum efficiencies.
A promising possibility in this wavelength range is to lase a Co doped isomorph
of Cr:ZnSe. Two interesting possibilities are Co:ZnSe and Co:ZnTe. The former has
a broad emission extending from 2.5 to 4.0 J-Lm and the latter from 2.7 to beyond 4.2
J-Lm. The materials also possess broad absorption bands in the 1.4-2 .0 J-Lm wavelength
range . Room temperature fluorescence quantum efficiencies are approximately 25% in
Co:ZnSe and 8% in Co:ZnTe. These efficiencies substantially increase as temperatures
are reduced to 200 K.
3.1.5. Chemical Lasers
In the chemical laser category, DF and HF based lasers are available. [9] The DF
laser provides wavelengths over a range of 3.5 to 4.2 whereas the HF laser provides
wavelengths in from 2.6 to 3.3 J-Lm. Large size and toxic nature are major limitations
of these laser systems .
3.2. Nonlinear Frequency Converters
Three-wave mixing (TWM) , also known as the optical parametric interaction, is an

alternate process to generate MWIR wavelengths through light-matter interaction. [18-
20] The TWM process basically requires an intense laser pump source and a nonlinear
medium. In this process, two propagating monochromatic waves with frequencies WI
and W2 mutually interact to generate new frequency through sum (sum frequency gen-
eration or SFG) or difference (difference frequency generation or DFG) processes . The
relation between polarization density and the electric field vectors describes light and
medium interaction. Three wave mixing (TWM) is possible through the second order
nonlinearity, PNL = 2dE 2 ; d = 2nd order nonlinear coefficient and E is the electric
field vector. Optical parametric amplification (OPA), Optical parametric oscillation

(OPO) and frequency up-conversion are examples of TWM. In principle, all three
processes can be used in the generation of MWIR wavelengths. The second harmonic
generation is a special case of SFG. In the case of second harmonic generation, two
pump photons of frequency W o interact inside a suitable nonlinear medium to generate
a photon of frequency 2wo • In the wavelength domain, two longer wavelength photons
are combined to generate a shorter wavelength photon. However, second harmonic
generation devices are not attractive for generating MWIR because of the lack of suit-
able small and compact Long- Wave InfraRed (LWIR) laser pump sources. To generate
4 J.Lm wavelength we need a laser pump source operating at 8 J.Lm. One possibility
is to use C02 gas laser generating wavelengths in the vicinity of 10.6 J.Lm. However,
these lasers are bulky and impractical for many applications. In the case of frequency
mixing, two pump beams usually of different frequency have to interact inside a non-
linear medium to obtain either the sum or difference of the input frequencies. Again,
similar to second harmonic generation sum frequency is not practically viable due
to source limitations. However, difference frequency mixing (DFM) is more viable
using readily available short wavelength sources. The main disadvantage is that this
technique requires two pump sources and hence becomes less practical. The optical
parametric oscillator (OPO) is the opposite of SFG and is considered most versatile
and practical means to generate MWIR wavelengths among the above three types of
nonlinear interaction and hence is described in greater detail in the following section.
An OPO generates broadly tunable coherent optical radiation over a wide wave-
length range via second order nonlinear process called three-wave-mixing in certain
non-centrosymmetric crystals. Near-UV to LWIR wavelengths can be generated using
OPOs. In an OPO, the intense pump photon generates two higher wavelength photons
when frequency and phase matching conditions are simultaneously sati sfied. These
conditions are W3 = WI + W2 and k 3 = k i +k 2 ; w = frequency, k = momentum vector,
and 1,2 and 3 correspond to pump, signal and idler wavelengths, and can be satisfied in
certain anisotropic crystals. Figure 15 shows the schematic of an OPO . For nonlinear
interaction to occur, an OPO requires a strong pump source whose intensity is above
a certain oscillation threshold value. Intense pump radiation can be achieved from a
suitable Q-switched laser.
The OPOs can be normally pumped using pulsed solid-state laser sources. Gener-
ally, the Q-switched pulsed lasers provide sufficient intensities to initiate parametric
interaction in many OPO materials to overcome the oscillation threshold. Besides
pulsed laser pumping, OPOs can also be pumped using lasers operating in the CW
mode. [21] OPOs can also be directly diode pumped provided high power diodes
and/or materials with high nonlinear figure of merit (D eff) are available. [22] Nor-
mally, indirect diode pumping of OPO are adopted due to limitations of pump laser
diode power in single emitter configuration, power supply constraints, overall size, and
cost considerations. With the advancement in laser diode technology, direct pumping
using laser diodes could become attractive. The advantage of diode pumping is that the
diode can be modulated and the OPO output will be similarly modulated. Additionally,
diode pumping enables the construction of small and compact transmitters.
Above threshold, the pump radiation generates two wavelengths called signal and
idler when phasematching conditions are satisfied. The signal wavelength is less than or
equal to idler wavelength. The phase matching conditions are derived from fundamental
energy and momentum conservation laws . For phase matching to occur, the crystal
~[ ,
===:> I dl..
OPO ===:> SI~ll.1

Cry_a! ] ===:> R"<l dll:lol Pump
Inpne R . SOlI :.10 1
Ou tpu t
l 'Oll pl... Coup1ft
Fig. 15. The schematic of an OPO.
has to be cut at a phasematching angle to obtain desired wavelengths under normal

operating conditions. Without proper phasematching a laser beam propagating through
a nonlinear crystal will continue to transfer energy back and forth between the pump
beam and the signal/idler beam pair. The result is that no net energy conversion takes
place.
The OPOs are intensity driven devices . Although the OPO output is at two wave-
lengths, it offers zero quantum defect. Furthermore, the device gain is independent of
tuning. The pump intensity has to exceed the oscillation threshold . Large D effectiv e , and
doubly resonant oscillation (ORO) reduces the oscillation threshold and increases con-
version efficiency. The crystal selection depends on its transparency range, nonlinear
coefficient, (higher the better), phasematching capability, availability in appreciable
sizes, optical quality, damage threshold (higher the better), commercial availability
and cost. The input and output couplers are achieved by having dielectric or trichroic
coatings of desired bandwidth and transmittance on suitable mirror substrates. Broad
tunability allows an arbitrary wavelength to be accessed .
In bulk crystals , a unique orientation of a birefringent crystal can be chosen for
phasematching purposes. The birefringent phase matching can be further sub-classified
as either Critical or Non-Critical Phase Matching (CPM or NCPM) . The CPM tech-
nique offers large wavelength tunability through angle rotation . However, for achieving
optimal OPO efficiency, the limitation s imposed by low angular acceptance and large
walkoff parameters have to be carefully considered. The NPCM offers high angular
acceptance and very low walkoffwith limited wavelength tunability [23]. Phase match-
ing can be either Type I or Type II depending on the polarization of three beams. With
reference to the frequency matching condition, W p =W s +W i, a particular combination
of the polarizations of pump, signal and idler wavelengths are allowed . In the case
of a negative uniaxial crystal, Type I phasematching refers to e = 0 + 0 and Type
II phase matching refers to e = e + 0 or 0 + e, where "e" is the extra-ordinary ray
and "0" is the ordinary ray. This is reversed in the case of a positive uniaxial crystal.
Wavelength tuning is achieved by varying refractive index via crystal rotation (angle
tuning), temperature, or electro-optic means.
Alternatively, a material can be engineered so that light is re-phased every coher-
ence length in the crystal. The former technique is referred to as birefringent phase-
matching and the later technique is known as Quasi-Phase Matching (QPM). [24,25]
QPM based materials were selected instead of bulk materials since they can provide
higher nonlinear coefficient, D eff than corresponding bulk materials. Other advan-
tages of QPM materials include zero walkoff, and high angular acceptance. The QPM
material is fabricated from wafer of a bulk material by " periodically poling" the ferro-
electric domain s. Effectively, this changes the sign of the nonlinear coefficient, D eff,
'!.lr +-+-+-+-+-+-
Wk
RoDdoal
Pump Plan>
-
""-
Fig. 16. Schematic of a QPMmaterial.
for every coherence length and is illustrated in Fig. 16. The alternating dark and light
regions indicate a change in the sign of the nonlinear coefficient Pump light propa-
gates through the crystal perpendicular to the poled regions. The domain (also known
as grating) width determines the idler and signal wavelength for a given pump wave-
length. The QPM materials are engineered to provide relatively high Deff convenient
for TWM operation which otherwise is difficult to achieve in bulk materials. The output
wavelengths can be conveniently tuned either by varying the OPO crystal temperature
or by mechanical translation of the crystal over varying grating periods. The former
one provides fine tuning of wavelengths where as the latter allows for course change
in wavelengths . A commonly used QPM material to generate near IR and MWIR
wavelengths is the periodically poled lithium niobate (PPLN) material. Other recently
developed materials include periodically poled KTA and periodically poled RTA.
The anisotropic OPO crystal is located inside a resonator. For OPO operation,
idler and/or signal and/or pump can be resonated . An OPO can be operated as a Singly
Resonant Oscillator (SRO) or a Doubly Resonant Oscillator (DRO. In a SRO, either
signal or idler is resonant In a DRO both the signal and idler wavelengths are resonant
In general, the output coupling could vary from 0% to close to 100%. A DRO with
100% output coupling at either wavelength is generally called an SRO. The optimal
oscillator design is determined by the application. SROs have higher spectral stability
than DROs because DROs suffer from mode "cluster hopping" . This phenomenon is
manifested by frequency hops from one pair of resonant (signal and idler) frequencies
to another spectrally close pair. The underlying cause is that a cavity length that is
optimized for the signal frequency is unlikely to be simultaneously optimized at the
idler frequency. The result is small frequency hops as the OPO tries to find the optimum
(highest gain) axial mode pairs to operate at For the same reason, SROs are easier
to wavelength tune than DROs and require mirrors with less coating constraints. As
such, SRO mirrors cost less. The main advantage of a DRO is that it will have a lower
power threshold than an SRO. The pump radiation itself can be resonated in both
cases . If all three wavelengths are resonated, the OPO is known as triply resonant
oscillator (TRO). TRO configuration, in general, offers lower oscillation threshold but
present practical difficulties in the fabrication of resonator mirrors. Besides having an
OPO resonator arrangement external to the laser resonator, it can be operated in an
intracavity configuration by placing the OPO crystal inside the laser resonator. [26]
OPOs operating in the MWIR spectral region have been widely demonstrated. [27-
28] The OPOs with various cavity configurations can be developed to generate CW or
pulsed output In one configuration, the lasing and OPO interaction was accomplished
inside a single nonlinear medium namely Nd:MgO:LiNb03. [29] Additionally, a vari-

ety of nonlinear gain media may be used depending on the pump wavelength chosen
and desired device attributes . Practical pump sources include Nd:YAG laser (1.064
Mm), Er:YAG laser (1.54 Mm), and Tm:Ho:YLF (2.06 Mm) and their designs are dis-
cussed in [30]. The potential nonlinear media for generating mid-wave IR (MWIR),
3-5 Mm wavelengths are LiNb03, periodically poled LiNb03 (PPLN), RTA and KTA.
For output at 4 Mm OPOs are typically pumped around 1 Mm due to the availability
of commercial Nd:YAG laser pump sources . In pulsed mode, these systems typically
output nanosecond duration pulses due to the short pulse duration of the Nd:YAGpump
lasers. Finally, OPOs can outperform state-of-the-art QCL devices in peak power, room
temperature operation, wavelength agility, size and cost.
3.2.1. Performance Modelingof OPOs

To start with, the performance of an OPO has to be modeled. The mathematical re-
lations for the estimation of various OPO parameters are described in the literature
[31]. The performance modeling is an iterative process based on the selection of a
suitable pump source wavelength and the OPO material characteristics. For perfor-
mance modeling, the OPO material, pump wavelength, OPO material characteristics
including length and dispersion characteristics and resonator parameters have to be
determined. In the beginning, phase matching curves (or tuning curves) are generated
for a given pump wavelength by using the phase matching conditions, Snell's law and
corresponding Sellmeyer's equations for the refractive indices for a given OPO mate-
rial. Based on the D eff, Type I or Type II operation is selected . Note that larger the
D eff, lower the oscillation threshold and hence the pump intensity requirement. The
crystal cut angle is then determined for nominal output wavelength operation. Based
on walkoff considerations, optimum OPO crystal length is determined. Most of the
time, the selection of an OPO crystal and its dimensions are dictated by its commer-
cial availability, procurement time and cost. The pump beam is modematched to the
OPO resonator mode. The mirror closest to the laser pump is designated as the input
coupler and the farthest mirror as the output coupler. The OPO resonator mirrors are
designed based on the SRO or DRO configuration. In the case of a SRO OPO, the
input coupler has to be AR coated at the pump wavelength and highly reflective at the
signal and idler wavelengths . Also, the output coupler has to be highly reflective at the
resonant wavelength . Based on the source and OPO material characteristics and the
transmission characteristics of resonator mirrors, the oscillation threshold intensity is
computed. The total conversion efficiency then depends on the pump intensity above
threshold. The OPO design details can be found in [20] and [32]. In the following
subsection, salient features of PPLN OPO for MWIR wavelength generation design
are discussed .
3.2.2. PPLN OPOs

PPLN is a commonly used material for MWIR OPO operation. The major advantages of
this material include : (I) offers good optical transmission from approximately 0.35 to
4.2 J.Lm wavelengths, (2) provides a relatively high nonlinear figure-of-merit (Deff '" 17
pm/V), (3) offers low optical loss at most wavelengths shorter than 4 Mm, and (4)
allows phasematching with visible to near-IR wavelengths where robust laser sources
II ~. '1 • •• • ••
T._•• .•
tI • • , . I.
~I·_I
III
Fig. 17. The dependence ofgrating period ontemperature atpump wavelength of 1064 nm. The
grating period selected for operation at 3.82 JLm is shown.
are available. As discussed above, the PPLN material is fabricated by "periodically

poling" the ferroelectric domainsof a bulk waferofLiNb03. Up to I mm thick wafers
and crystal lengths up to 5 em are commercially available. Severalexamplesof PPLN
OPOsbeing usedto generate4 Jlm light havebeenreportedin the literature. [21.24,29]
In the following section, PPLN OPO for generating wavelengths around 3.82 Jlm is
discussed. Note that 3.82 Jlm is one of several low loss sub-bands.
PPLN OPO Design Considerations. To obtain a given MWIR wavelength, proper

grating width for a given operational temperature is required. It is determined using
the following equation.
np ns ni 1
Ap = As + Ai + A' (37)
where n p , n s , and n i are the refractive indices correspondingto pumpwavelength, Ap ,

signalwavelength, As, and idlerwavelength, Ai. Figure 17illustratesthe dependence of
grating widths on wavelength and temperature. It can be seen that for obtaining 3823
nm at 50°C (323 K) PPLN with 28.9 Jlm is required. If a slightly different grating
period is used then operational temperature has to be adjustedaccordingly. Figure 18
shows the tuning of idler and signal wavelengths as the temperature is varied for the
PPLN crystal of grating spacing equal to 28.9 Jlm. Due to weak dependence of idler
and signal wavelengths on temperature, the operational wavelength can be kept stable
within the tolerance range of 9 nm or 0.009 Jlm using temperature controllers.
3.2.3. Oscillation Threshold Calculations
The oscillation thresholdcalculationsare widely described in literature. [33-35] Two

cases, unfocused(or unoptirnized) and focusedGaussian pump beams are considered
for modeling. For un-optimized pumpbeam,theoscillationthresholdwascomputedto
be 757 kW/cm2 . The input parameters used werea PPLNcrystaloflength 5 em, Deff=
17 pm/V and a pump radius of 100 Jlm withTEM mode profile. This is for the case of
single pass pump beam. If the pumpbeam is double passed, the thresholddecreases to
489 KW/cm2 • [30] When the pump beam is optimized with a focusing parameterof
Optical communications in themid-wave IRspectral band 371
•••
•••
.... )12~.3
i
,;
in•
Je_ ;::'
DUI,. T~ .. ,er.tMre
t
Idle, Wavel ••,th
.! ,..
; 21.
; zs_
-: 24_
i.
.!!l u•
21_
I PPLH Cryshl
Gratin, Peri •• I: 2I .t u 1ft
I
.
..... ... .. ... L- - .. ... .. ..
_ 1• •
1.7 .. 3 .... Silna' Wav.tantUt
:; 11_ -,
,
•• •• MI
T, til per.t.n, K
Fig. 18. Temperature tuning of PPLN crystal of 28.9 mm grating period.
1.26, the threshold drops to about 33 kW/cm2 . The threshold calculationsgive us an

approximate pumpintensityrequiredto start seeingan OPOoutput.The peak intensity
ofa 15 pJ/pulse at 10ns when focused to a lOO-p.m spot radiusis 9.55 MW/cm 2 • This
peak intensity is nearly 12.5 times the single pass, un-optimized pump beam. When
the pump beam is optimallyfocused, then this figure goes up even further. The rule of
thumb is that the conversion efficiency can be maximized if the thresholdis more than
five times the threshold.
The interaction length is proportional to the crystal length. Since PPLN crystals
have no walkoff, longer the interaction length, higher is the conversion efficiency.
However, OPO parameters have to be optimized to minimize back conversion. To
achieve maximumconversion efficiency, the pump beam with a TEMoo mode profile
has to be properly injectedinto the OPO cavity. The pump beam spot has to be located
in the center of PPLN crystal. Furthermore, the signal and idler radiations have to be
contained inside the crystal. It is also clear that powerthresholdincreasesas the pump
beam waist diameter increases. However, smaller beam waists result in greater beam
divergence in the PPLN gain medium. The pump beam optimization has to be carried
out to avoid beam clipping by the PPLN crystal. In this case, the Rayleigh range has
to be greater than the PPLN crystal thickness. The pump beam spot radius of about
100 p.m at the center of the PPLN crystal of length 5 em will be appropriate. At 10
p.J/pulse, a IOO-p.m spot radius has intensitygreater than 2 MW/cm 2 . Greater than 100
p.m spot radius at the output coupler of the pump laser can be obtained for the plano-
concave cavity configuration of length 25 mm and output coupler radius of curvature
of 50 mm. A 2-W diode pumped, passively Q-switched Nd:YAG laser with 7.5 %
total optical slope efficiency operating at 10 kHz PRF would provide 15 p.J/pulse.
Assuming 25% pump depletion, > I p.J/pulse at idler wavelengths can be obtained.
These are very conservative figures and a better performance can be anticipated by
minimizing coupling losses.
The output beam from the laser resonator can be mode-matched to the PPLN
crystal usingsuitablebeam shapingarrangement. The PPLNcrystal has to be enclosed
inside an oven for temperature control. For partial ORO operation, the specifications
of the OPO resonator can be as follows : A concave-concave type resonator with 50
mm radius of curvature mirrors and a 5-cm length x I-cm width x 500-/Lm thick
PPLN crystal. The input coupler mirror will have an AR coating at pump wavelength
and high reflectance coating at signal and idler wavelengths. The output coupler will
totally reflect the pump and signal wavelengths, and completely transmit the idler
(3.82 /Lm) beam. The partial ORO configuration can be chosen to enhance the OPO
conversion efficiency. For this OPO cavity, the signal beam that resonate s inside will
also be completely contained within the 500-/Lm-thick PPLN crystal.
4. The MWIR Detectors
Pyroelectric detectors and PbSe detectors are available foruse from 3 to 5 /Lm. These de-
tectors allow room temperature operation. [36] Pyroelectric detectors generate charge,
and, hence, a signal, in response to heating. They can be operated in either voltage or
current mode. Advantages include (I) extremely broad spectral response (0.0001-1000
/Lm), (3) wide operating temperature ranges (-55 to 125°C) . [37] PbSe detectors have
peak sensitivity at 3.8 /Lm. They are photoconductive devices and operate by exciting
photoelectrons across a semiconductor band gap into a conduction band. Advantages
of this type of detector include (I) slightly higher specific detectivity, or D* values,
than the pyroelectric detector, (2) shock resistance since they are photoconductive
devices, (3) wide operating temperature ranges (-20 to SO°C) . Both detectors can
be equipped with a wide array of optical bandpass filters. [38] Both the PbSe and
pyroelectric detectors are inexpensive, can be ruggedized . Typical specifications of
PbSe and pyroelectric detectors are given in Table I. New detector technologies in the
MWIR spectral region, however, offer substantially better performance. Examples of
these new detectors include (I) "Dember effect" or "non-equilibrium" detectors, (2)
Quantum Well Infrared Photodetectors or QWIPS , (3) microbolometers, (4) resonant
enhanced cavity detectors, and (5) Schottky barrier PdSi, PtSi, and PtGexSh- x de-
tectors. A disadvantage of some of these technologies, however, is that performance
specifications are classified and/or not commercially available.
Table 1. Atmospheric transmission from 3.5 to 4 JLm.
Type Active Area Peak Sensitivity Sensitivity Detectivity (D*) Rise Time
Model Number (mm) Wavelength (JLm) (cmH z1/ 2/W )
PbSe P791-02 3x3 3.8 70 JLY I x 109 2 to 5 u see
Pyroelectric P3782 ¢2 flat 1500 Y/w 1.7 x 108 100 msec
4.1. Dember Effect Detectors
These devices are based on Zn-doped HgCdTe semiconductor materials with free car-
rier concentrations below their (nominal) equilibrium values (non-equilibrium mode).
[39] Typical D* value of these devices at room temperature is about I x 1011 em
HZ 1 / 2 JW in the MWIR spectral region. This is a significant performance enhancement
over the PbSe and pyroelectric detectors previously discussed, and use ofOember effect
detectors would substantially reduce the required laser transmitter power. The Oember
CdZnTe hemi spherical

or hwerhemisphericallen s
w re + meta l ization p ath
Si substrate
epol.xi.1HgCdZnTe detector
Fig. 19. Schematic diagram of a commercially available HgCdTe non-equilibrium or "Dember

effect" detector.
P:an lN'Rr ~riIl<'a do ",

.!!.!L-
;,~~.
Puk Dnt ctJ.vuy. emHzO ',w >3· 10' DEm:TOR
"
RtliponsavrtyY IW 24 .~ -
Rt:fl 0~ TUll ~ r5 < 10 r---"
Ruuta rr, n
Active AU'a. rr.m J

3l
1
1
i
I
~ .
N cue Du)Sllty.nYIHzO ' ;:±:
.
0.8
FOV, deg:ees 42 ........!L
Fig. 20. The detector specifications supplied by the manufacturer and its dimensions.
effect detectors are attractivedue to severalreasons. They include(a) ambient temper-

ature operation, (b) no bias requirement. (c) short response time, (d) no flicker noise,
(e) wide dynamic range and (f) commercial availability at relatively low cost. These
detectors are available for operation in the spectral range of 2-12 /lm and individual
detectorsare designedtoexhibitpeakperformanceat a specificwavelength. Demberef-
fectdetectorsare commercially available fromVIGOSystemsLtd. (www.vigo .com)
from Polandand are distributedby BostonElectronics, Inc. in USA. Figure 19showsa
schematicdiagramof a Dembereffectdetector. The HgCdZnTe photodetectoris seated
on a Si substrate. A CdZnTehyperhemisphericallensis formedover the detectorusing
ultrasonicand classical mechanical manufacturing techniques. The monolithic design
permits the construction of rugged and mechanically stable devices. Additionally, by
properly coating the spherical lens surface the device can be (I) shielded from un-
wanted backgroundradiation and (2) antireflection coated for operationat the desired
wavelengths.
ForMWIRwavelength detection. thePDI-5detectoranda matchingtransimpedance
amplifier. Model 481-20PEM are availablefrom VigoSystems.
Figure 20 shows the vendor measuredspecifications and the detector dimensions.
The procureddetectorhasa specific detectivity, D* of > 3E9 cmHz O.5/W. The PDL-5
is a photovoltaic heterostructure based on epitaxial variable band gap Hg1 - xCD xTe
structures grown using Isothermal Vapor Phase Epitaxy. The photovoltage signal is
due to both bulk photodiffusion effect (the Dembereffect)and photovoltage at hetero-
junctions.Actually, in the short wavelength (~3.4p,m) devices, the responseis mainly

due to the photovoltaic response at heterojunctions, the contribution of the volume
photovoltage is small. The devices are monolithically optically immersed to CdZnTe
lens (n = 2.7, radius of curvature 0.8 mm, position of active layer-aplanatic plane
(r + r In .distance from surface of the lens). Coating of the devices can increase the
quantumefficiency by ~20%. Due to optical immersion the physical active area is n 4
times (~50) times smaller than the apparentoptical area. In consequence, this reduces
electrical capacitance by the same factor in comparison with conventional device of
the same opticalarea. This aspect is very importantfor frequency response! The main
issue is the lowresistanceof activedevice,comparableor evenlowerthan the parasitic
series resistance of the device. The large series resistance is due to resistance of the
p-type layer. Actually, the device has distributed parameters and is complexto analyze
its performance. The result of the parasitic series resistance is a low current respon-
sivity and apparent quantum efficiency. Simply, the series resistance reduces current
responsivity, even if the deviceswouldhave 100% percentof quantumefficiency. This
is specific feature of photovoltaic devicesoperatingat long wavelength and high tem-
peratures. One can bias this detector device.The device exhibits soft breakdown due
to heating. One has to be careful to preventexcess thermal breakdown. Starting with
a low bias current (few {LA) and increasing it to the highest photoresponse one will
notice an increaseof photocurrent and, more pronounced, photovoltage. The decrease
of photoresponse means heating of the device, so it is better to work with bias cur-
rent below the current of the maximum response. One can expect significant drop of
photovoltages on parasitic impedances, which will reduce possible gains. The noise
of the non biased photovoltaic device with no or weak illumination is the Johnson-
Nyquist noise [V n = (4kTRL1J) 1/2] . When biased one can sec increased current
and low frequency noise. In the future, detectors providing a quantum efficiency of
about 80-90% and frequency response up to several hundreds of MHz are possibleto
fabricate with the following techniques: (a) reduction of series resistance by the use of
heterostructure with heavily doped n+ layer as the first deposited layer on substrate.
This will significantly improve the measured quantum efficiency and response time
(b) use of AR coating on immersion lens.
Current Dember effect detectors are within a factor of 6 of the theoretical Back-
ground LImited Performance (BLIP) for 21l" steradian field-of-view and 300 K sky
temperature. .It is expected that BLIP limited performance at 4 p,m will be achieved
within the next few years with improvements in semiconductor materials processing.
It is anticipatedthat Dember EffectDetectorswith D*= I x 1011 em Hzo.5 tw, within
a factor of 100of those availableat I p,m, will be available in the near future. Higher
sensitivity detectors mean that lower transmitterpowers are required.
s. Data Communications in the Mid-IR
Usingthe emergingQCL devicetechnology, data communication throughdirect mod-

ulation can be carried out. [40] Due to limited wavelengths and output power (less
than 25 mW) available, the QCL lasers are suitable for very limited free-space range
applications. These diodes would be attractive provided low loss optical fibers in the
MWIRbecomeavailable. However, anotherapproachis to use the OPOdevices. These
OPO devices are capable of providing very high peak powers in the pulsed mode. In
Optical communications inthe mid-wave IRspectral band 375
the following section, the performance of a compact,eyesafe, all-solid-state, mid-wave

IR (MWIR) transceiver for data communication through low visibility conditions is
discussed.
The transceiverwas developed for Multiple Integrated Laser EngagementSystem
(MILES) application. [41-42] The MWIR wavelengths are derived using a passively
Q-switched Nd:YAG laser pumped periodically poled lithium niobate based OPO.
MILES weapon code transmission for small and heavy weapon platforms have been
demonstrated through dense theatrical fog. With less than 3 pJ/pulse at ",4 j,tm and
a room temperature Dember effect IR detector, greater than 5 km range has been
successfully demonstrated [43] and is discussed in the following subsection.
5.1. Weapon Code Transmission in MWIR
Multiple Integrated Laser Engagement System (MILES)is a tacticalengagementsim-

ulator that facilitates force-on-force training in realistic battlefield scenarios. MILES
is a free-spacecommunication transceiverthat generates weaponcodes that is charac-
teristic of the weaponplatform,ammunition features and user. It transmitsan encoded
message simulating weapon firing characteristics, round dispersion patterns, hit or
miss probability with range information. The receiverdecodes the message following
which furtheraction is initiated. MILES is a laser communications system. It operates
by transmitting and receiving encoded"words" that identifythe shooterand the amount
of damage inflicted on the receiver(hit or miss). MILES transmitters are designed to
mimic the actual firing rates and effective ranges of the weapons they simulate. The
MILES has been proven to increase the survivability, combat readiness, and fighting
effectiveness of military forces. The current MILES, however, suffers range limita-
tions under certain reduced visibility conditions. These conditions include inclement
weather (rain, snow, fog), dust clouds, and camouflage smokes. This range limitation
reduces training effectiveness, can lead to training postponement and hence higher
trainingcosts. For the warrior to effectively train to the capabilityof his weapon it will
be necessary for futureMILESto penetrateboth naturaland anthropogenic obscurants
as effectively as a FUR.
The current MILES and MILES 2000 trainers operate at 904 nm. The 904-nm
MILES that is currently being widely deployed providesexceptional performancein
clear atmospheric conditionsonly. The weaponcode sequencein a 904 nm wavelength
MILESis generatedusinga GaAs laserdiode. A Si photodiodebasedreceiverdecodes
the received code sequence. Besides eyesafety concerns, the device does not operate
in the presenceof obscurants. Advantages of the system includesmall size and weight,
environmental ruggedness, demonstrated effectiveness, and low componentcost. The
principal disadvantage is worse transmission than a thermal imager through some
weatherconditions and obscurantclouds. This limitation can result in reducedtraining
effectiveness, canceled training activities, misses and false alarms, and troop frustra-
tion. Although in the past, MILES operating at 1.5-,2.1 - and 1O.6-j,tm wavelengths
have been developed to address the transmission shortcomings, they do not provide
optimal performance. The 1.5-and 2.1-j,tm wavelengths do not penetrateobscurants.
The 1O.6-j,tm system paired with relatively inefficient pyroelectric detectors are bulky
and requires high prime power.
MILESoperatingin the mid-wave IR (MWIR)spectralregionfor small and heavy
weapon platforms has been investigated by CoherentTechnologies, Inc. (CTI). It was
shown that the MWIR MILES has the potential to provide eyesafe and optimal per-
formance in the presence of battlefield and natural obscurants . The MWIR MILES
obscurant penetration capability has been validated using a theatrical fog generator at
the Table Mountain test site located near Boulder. The MWIR wavelength is generated
using an all-solid-state device technology using a PPLN OPO. Field tests using com-
pact MWIR MILES have produced range performance greater than 5 km. Although
specifically developed for MILES, the proposed approach is, in general, applicable for
data communications in the MWIR region. In this section, wavelength selection and
obscurant penetration analysis, transceiver design approach, and field performance of
a MWIR MILES transceiver in the presence of obscurants are discussed .
5.1.1. Wavelength Selection

A variety of atmospheric conditions including clear days, rain, snow, and fog were
modeled using FASCOD3p. Furthermore, fog oil based battlefield obscurants consist-
ing of airborne particulates that reduce visibility were also considered for modeling.
The FASCOD3p parameters were chosen to mimic conditions in locales where MILES
training is conducted. The data required for simulation was collected from respective
locations through STRICOMIUS Army. The DoD obscurant handbook [44] describes
typical CL products for battlefield use of fog oil smokes. Manmade battlefield obscu-
rants behave differently because they have been engineered to greatly reduce visibility
at visible wavelengths . Typically, the particles that compose manmade obscurants have
wavelength dependent absorptions associated with them. An example of such an ob-
scurant - that is commonly used in training is fog oil. The DoD obscurant handbook
describes typical CL products for battlefield use of fog oil smokes . With enough power
any wavelength will penetrate the atmosphere regardless of weather or obscurant. How-
ever, from eyesafety, power consumption, and platform compatibility considerations,
MWIR wavelength offers significant advantages. The 4-t-tm wavelength offers several
significant advantages over the other wavelengths considered. Based on modeling, the
4-t-tm window was selected because it offers high transmission and can be readily
accessed using "practical" all solid-state laser hardware . Furthermore, longer wave-
lengths offer relatively reduced scintillation effects. Additionally, the 4-t-tmwavelength
certain atmospheric conditions this wavelength offers advantages over either the 900-
nm or 1.55-t-tm wavelengths. The FUR sight is unaffected by the smoke. The 905-nm
and 1.55-t-tmwavelengths are severely attenuated by the smoke and the 3.8-t-tm wave-
length is somewhat attenuated. Despite the reduction in visibility at 3.8 t-tm due to the
smoke, the 3.8-t-tm and the 10-t-tm wavelengths are attractive for transmission through
obscurants.
5.1.2. Transceiver DesignApproach

The critical specifications for small weapon platform such as M16 rifle include range
equal to 500 m, beam divergences of 0.7 and 1.5 mrad for hit and near miss cases,
and a single 9V battery operation satisfying 200 rounds. For heavy weapon platforms
such as M IA2 tanks, the range is 3.75 km, and beam divergences for hit and near miss
cases are 1.2 and 1.7 mrad, respectively. Ample power is available in tank platforms .
One attractive spectral region is around 3.8 t-tm wavelength . There are several low
atmospheric loss sub-bands available for use and are illustrated in Fig. 21. The 3.8-t-tm
Optical communications inthemid-wave IR spectral band 377
.
··
.~ I .•
y: • .•
• I .Y
: ...
U
-= .. ,
·e
£ . ...
E ' .J
... ' .2
Fig.21. One-way transmission inthe 4-mm region expanded around 3.82 mm wavelength.
wavelength region is farther away from C-H absorption bands. From obscurant pen-
etration analysis based on atmospheric modeling, fog oil characteristics and source
availability, operational wavelength selected for MWIR MILES was 3.82 J-lm with a
tolerance width of 9 nm. To obtain 3.82 J-lm radiation, the optical parameter oscillator
(OPO) based approach was selected.OPOs are viable candidatesfor a next-generation
MILES transmitter due to their broad tunability in the MWIR and potential high effi-
ciency.As mentioned earlier, OPOs can be directly or indirectly laser diode-pumped.
The advantage of either pumping scheme is that the laser diode can be modulated to
output the appropriate MILES message code and the OPO output will be similarly
modulated. Additionally, diode pumping enables the construction of small and com-
pact transmitters.The indirectdiode pumpingbased on Nd:YAG pump laser approach
provides short pulse, high peak power, and extended range performance that is not
practical with direct diode pumping.
The all-solid-stateMWIRMILEStransmitterconsistedof a passivelyQ-switched,
Nd:YAG pump laser coupled to a periodically poled lithium niobate (PPLN) OPO.
The passive Q-switching was achieved using the Cr4+:YAG material. PPLN is an en-
gineered quasi-phase-matched (QPM) nonlinear optical material that can efficiently
generate mid-IR wavelengths. The advances in the QPM technology have permitted
smallerand moreefficientdevicesto be built.A 5-cm-IongPPLNcrystal waslocatedin-
side a temperature-controlled oven. The PPLNOPO resonator was a concave-concave
type and was operated in a singly resonant oscillator mode with signal resonance. The
PPLN OPO was temperature tuned from 3.6 to 4 J-lm. Initially, photorefractive dam-
age of the PPLN material was a concern. The effects of photorefractive damage were
evaluatedby operating the PPLN OPO at 50°C for 10days with 8 hoursof continuous
operation each day. However, no visible damage to the PPLNcrystal was observed.A
miniature telescope was used for output beam shaping of the idler wavelength.
The transmitterdriveelectronicsconsisted of laserdiode driver, encoder and signal
conditioningelectronics. The pump laser cavity was optimized for a free-running pulse
repetition frequency of 10kH z. Typical laser pulsewidths were around 6.5 ns. Short
pumppulsewidths are advantageous since an OPO requires intensepump pulseswhose
MCCJ7 We N. ra- a...'OC'TI'llllt B&MCloc:kR•• · 48 kHz

$ub-hamccUc Uted . 3 ldf z
1 Wcrd - lOI.oc1eSt.t.ttf
I Word. Milts Code. SPID
Wcri Bit WIf'dt - 10
MIlH Code Sl.ilH t w.ic:h1: . 6
SPlD W, ic.ht- 4
1 WcrdTunt . 11Tunt Slob
1 Tunt Slot- 16 Bw
Btit Cb=k KlIl:. · 48kHz
Su'b-hum:>rw: UHd. 3 »:H:
I .Locic Jt.d t' Ut pI..aced intunl ,btJ W lnbi.Nl. 6, ~ or IOori)'. I WotdTunt • 3 61 M
2 . Hotmon ~ 2 k:C:u: stdR1 i unytime slot.
3 . A .,.w wordwl1U"'..... 1ocic,ta_ 1 in'biJI I oU m l Mo tunI slob ud . I TimeSlot · D 3.D llf
a
Jop: . td, in 'oin I of the3-d tUr. , lot I 8Ul- JJ .8 3~,
Fig. 22. A summaryof MCC97 MILES code format(MCC97 manualobtainedfrom STRICOM

website).
intensity has to be above the oscillation threshold. The laser diode was operated in the
pulsed mode. The laser drive current pulsewidth was adjusted such that it generates
only one laser pulse. Typical current pulsewidths were less than 120 J.Ls. A laser pulse
was generated within 100 J.Ls after the onset of laser drive current pulse and this interval
can be adjusted by varying the laser cavity parameters.
The encoder pulses trigger the laser diode driver current. Basically, the encoder
drives the laser diode electronics to generate a desired MILES code. The MILES code
format had to be compatible with MCC97 regulations. For details of MCC97 code,
readers are referred to the STRICOM website. A summary of MCC97 is given in Fig.
22. For generating MCC 97 compatible MILES weapon codes, an encoder developed
by Lockheed Martin Corporation was used. Figure 23 shows the internal view of a
battery operated MILES encoder device. The encoder unit can be set to function as a
decoder and vice versa. The encoder device was compliant with MCC97 requirements.
A toggle switch is used to change or set the MILES code parameters while a fire
control button to is used to transmit the desired code. The LCD panel displays the
MILES code parameters. The encoder generates", I J.Ls width pulses corresponding
to a I bit that acts as the trigger pulse to the laser diode current driver. Similarly, the
decoder requires pulses of I J.LS width corresponding to I bit for decoding the code.
The received sequence from the preamplifier is inverted TTL logic to be compatible
with the decoder. The receiver electronics detects and condit ions the received code
for decoding purposes . The receiver electronics consisted of a MWIR detector, signal
conditioning electronics, and a decoder.
A room temperature "Dember" effect based IR detector with a peak response at
4 l.Lm was used for the MWIR MILES experiments . These IR detectors, as discussed
in Section 4.1, are fabricated using Zn:HgCdTe based compound whose principle of
operation is based on volume photovoltaic effect. The detector specifications for our
experiments include : active area of I mrrr', specific detectivity of3 x 109 cm HzO.5 rw,
voltage responsivity of24 vrw and response time less than 10 ns.A CdZnTe lens above
the active area provided a 42° field-of-view. Measured characteristics of this detector
include: dark current of Id ",0.4 rnA, noise equal to - 70.82 dBm (8.28E -11 W), and
a quantum efficiency of ",20% at 3.39 J.Lm. With AR coatings, the quantum efficiency
of the detector can be substantially enhanced.
Fig. 23. The MILES encoder and decoder unit.
Fig. 24. The MWIR MILES brassboard transmitter unit.
The brassboard unit used to transmit heavy weapon codes is shown in Fig. 24. The
breadboard transmitter satisfied the range requirements of a small as well as heavy
weapon platforms and validates obscurant penetrating property of the selected MWIR
wavelength . Table I illustrates the performance of the MWIR MILES transmitter.
The transmitter electronics shown in Fig. 25 comprises of an encoder interface,
laser diode driver, OPO heater controller, TEe controller and display interfaces. The
battery operated unit consists of a low noise, high bandwidth and high gain preamplifier,
automatic noise thresholding circuit, an one shot pulse digitizer circuit, and a pulse
stretcher. The pulse stretcher circuit stretches received '"'-'6.5 ns idler pulses up to I
usee duration to be compatible with the MILES decoder. The output is inverting and
follows TTL logic. The 3 dB bandwidth was 26.4 MHz. The receiver circuit shown in
Fig. 25 can be reduced to less than I square inch.
Table 2. Demonstrated performance characterisitics of the compact,MWIR MILES receiver.

Laser Diode Power =2W
Max. CW LaserPower = 840 mW
PassiveAbsorber = Cr:YAG, T = 90%
Max. Q-switch Average Power = 320 mW
PRF Range = 4-20 ns
Pulsewidth Range =4-66 kHz
Max. Laser Pulse Energy = 16J.lJ/pulse
Spatial Mode =TEMoo
Max. Idler Pulse Energy = 2 J.lJ/pulse
Max. PRF Jitter = ± 2.5 J.lS
Max. Amplitude Jitter = < 5%
Max. Pulsewidth Jitter '" 2 ns
Pulse Mode Operation = Highlystable
Size = 12 x 2 x 1.31 inches
Fig. 25. Transmitterelectronics(left) and receiverelectronics (right).
The fog was generated using a fog oil based commercially available theatrical
smoke generator. The fog fluid was completely water soluble at 70°F, with a vapor
pressure of 8.4 mmHg. The unit generated concentrated heavy smoke output with
pleasant mild odor.
Finally, with MILES devices, eye safety is an issue. Based on the calculations
that were carried out using ANSI Z 136-1 1993, it was apparent that any of the longer
wavelength options remains eye safe at transmit power levels substantially higher than
those permissible at 905 nm. For instance, the MPElpulse increases from 2.5 ergs/cnr'
to 3.5 x 104 ergs/em" when the wavelength is increased from 905 nm to 3.82 JLm.
Hence, the MWIR MILES was eyesafe for all practical distances.
5.1.3. Experimental Results
The output characteristics of the PPLN OPO is greatly influenced by the Nd:YAG
pump laser characteristics. Figure 26 shows a typical pulse profile and PRF character-
istics of the passively Q-switched Nd:YAG pump laser. In this case, the pulsewidth is
6.5 ns and the PRF is around 13 kHz. While the PRF MWIR OPO remains the same,
the pulsewidth of the OPO will be lesser than that of the pump laser due to parametric
Optical communications inthe mid-wave IRspectral band 381
\
\ 2
j 1
T
M &.59 ns kt 153.8 Kit et 75.671 us !& 13.215 ~ Hz
Fig. 26. Left: Typical pump laser pulse profile. Right: PRF characteristics of the pump laser.
conversion process and is illustrated in Fig. 27. Also shown in Fig. 27 is the jitter
characteristics of the passively Q-switched Nd:YAG laser. The typical amplitude and
PRFjitter characteristics of the pump laser is also shownin Fig. 27. For data commu-
nicationapplication under consideration, the PRFjitter has to be substantially smaller
than the duration of bin in each time slot as illustrated in Fig. 27. The PRF jitter of
a passively Q-switched laser is larger than that in the case of an actively Q-switched
laser. The timingjitter has to be controlled for data communication applicationssince
it will effectively limits the data rate. Althoughin this discussion, the applicationof a
passively Q-switchedlaser for communications is illustrated, an active Q-switch such
as an electro-optic Q-switch based laser can be utilized to provide PRF jitter of less
than lOOns for large data rate applications.
As mentioned earlier, the MILES transmitter operational wavelength was set by
fine tuningthe temperature of the PPLNOPO. Figure28 illustrates a sequenceof output
wavelengths measured using a monochromator as the PPLN crystal wavelength is
set at four different temperature values. Followingseveral measurements, the MWIR
wavelength tuning rate of the PPLN OPO was determined to be "-'1.37 nml°e. To
obtain 3.82 tun, the operating temperature was set at ,,-,50°e. The measured FWHM
of the idler pulse was ± 2 nm. The heatercontroller kept the OPO temperature within
± 0.1°e.
Up to 30 mW of average power with good spatial quality was achieved at idler
wavelengths. At 10 kHz, this average power corresponds to 3 pJ/pulse. Using this
device, a totalOPOconversion efficiency greaterthan53% has beendemonstrated.The
field experiments using MWIR MILESwere conductedat a laser test range locatedon
theTableMountain near Boulder, Colorado. The free-space range available for testing
was 2.5 km. Typical idler pulse energy used for our small weapon platform related
experiments was0.5 pJ/pulse.The MILEStransmitterand receiverwere locatedinside
a 30-ft-Iong shelter. A front surface coated mirror was placed at a distance of 250 m
from the MILES transmitterto reflectthe idler beam back into the shelter.
In order to transmit the MILES weapon code, pulses corresponding to a " I" bit
were generated. An electronics interface to the encoder was constructed to generate a
TTL triggerpulseforeach" I" bit.The triggerpulsestartsthe laserdiodecurrent.Based
- - - - - - - - - _.
t,·. ". "I

I''''
t Vbtl fI- Pu~ '
li t . I I ••
PO SigJ Al-
t i! ---=
I.
-
- '- - ~
~- :~ r---
jf
- . - - -- - _I
l IIi 1
l
s.se ns l&. 153.l MH?
Fig. 27. Right: Pulse compression due to OPO interaction. Pulse profiles beforeand after OPO
interaction are shown. Left: Amplitude and PRF jitter characteristics of the pump laser. Pulse
amplitudejitter is less than ± 5% (peak-to-valley). The PRFjitter of the pump laser was around
± 2.6J.ls.
xtt.n: '1;'w1, l<:toDh . r~ ""I I C xtt.n: ~dot\Oh .T · S ;' (
~l,
l~ O .
".!
A
s-, ~
~
"f "i
!
]
!
!, "·r
U:f
j
J. J. lO
r
.~ :S.lll )81 ' l 81. "Z> 3a:l6
lll'O
WlYd..,~. CJ UI.) 'y,' Il'1UtJ~, (rIlb.)
)1I1. C; ....'YId 'h DJ'-T · jl (1C
".
.r-.
'.lI lf' .....»"" .., ,! ,h ,.~
"°1 •! "O!
~
uo
A
~
; 1: 0
} :(1
') :11
j !
l: O
~
.1
x• ~ ":1
J.
J. sc
2,,,
"Bm
J'~~ !~Y.I l ' l-l U Zl ~ )2
'M 1GB ~ lt1 5tl:l l :uH
·....wdo¢. (rtll. ) ",' N+)Q¢ ,O'tn)
Fig. 28. The MWIR wavelength tuningby varying the PPLNcrystal temperature.
on the resonator dynamics, the time required to generate a laser pulse was determined .
As soon as the laser pulse is generated, the laser diode current was shut off. In this way,
Fig. 29 illustrates the generation of a" I " bit. Whenever the trigger pulse corresponding
to a "I" bit occurs, the laser diode current is turned on for a set period to allow the
generation of laser pulse. Due to 4 J.tm detector characteristics, OPO pulse is broadened
by several ns and is also shown in Fig. 29. However, the OPO pulse broadening by
several nanoseconds had no significant effect on decoding process.
The required transmission range was confirmed by calculating the time elapsed
between the instant the OPO pulse leaves the transmitter aperture and the moment
it returns back at the receiver using an oscilloscope. A retro-reflector was located at
~
I Ie'"
l (l . a ~'
I
~
.~ -9. 1}1'"
'2 1
I
~ •
/ 1
I I I I 1\ /
I I I I 11 '\
I
Fig. 29. Left: A weapon code bit generation. Top trace: Trigger pulse to the laser diode driver
from an encoder interface. Middle trace: Current pulse to the laser diode. Bottom trace: The OPO
pulse. The onset of OPO pulse can be adjusted by varying laser cavity parameters. Right: The
detected OPO pulse is broadened to '" 25-30 ns due to the 4 tLm detector characteristics.
,
4
- I-
!-- 1-
I
q
I t
---ll
I
1
, '
I
!
I
I
I "'r
I I-
~
1t :L I -'
t
j
,l J l4
I I I I.
M 16 . 796 J.lS ~59. 539 (I t"
Fig. 30. Left: Range measurements. Top trace: Return OPO pulse. Bottom trace: Transmitted
OPO pulse. Right. Checking for bit drop outs. TopTrace: Trigger pulses from the encoder. Center
Trace: Drive current pulses to the laser diode. Bottom Trace: The OPO pulse train.
about 2.5 km away from the transmitter. As shown in Fig. 30, the elapsed time was
found to be Llt = 16.796 ms. Using the relation, Range = cLlt, where c, the speed of
light, equal to 3 x 108 mis, the transmission range was 5.039 km. Following range
finding measurements, bit drop outs measurements under clear weather conditions and
mild atmospheric turbulence. These measurements were done to check whether the
transmitter, retro-reflector, and receiver optics were aligned by sending a continuou s
train of pulses that correspond to a "I " bit. In this case, continuous trigger pulses were
transmitted to make sure that for each encoder trigger pulse, a corresponding OPO
pulse is emitted.
Following free space transmission experiments, the shelter was filled with dense
fog. The visibility was less than I ft. The path of the idler beam inside the fog filled
~. . - ,~I-1; -
_L ..
,
!-... I f- - -
· ·1- I l-
'1
I , I
I I II I
I
, Uil 3
~
I I t , ~
f
T
I "- - - -II
I,
..
I I
..
1 I
.... ....
I
f--
-·1·
I
I T
I
!at .0 ~s !£, 1<. 2 KHz

3 DC 0.94 V
I L JL JI. pW$ 5.0 ns
H~ H~I'-O-- OR 10. 0 ns ~pu 0
Fig. 31. Oscilloscope traces of the MI6 weapon code sequence transmission. Top trace is the
transmitted code sequence, middle trace represents OPO pulses, and the bottom trace is the
decoded code sequence.
shelter was 60 ft. Under dense fog conditions, successful M16 weapon code trans-
mission was accomplished. The transmission results captured on an oscilloscope are
illustrated in Fig. 31. The top trace is the input weapon code sequence. The middle
trace corresponds to the output idler pulses. The pulses are very short and are not
properly captured due to inadequate sampling. The bottom trace corresponds to the
received code sequence . The MWIR MILES performance was compared with that of
a 904-nm MILES. The 904-nm MILES totally stopped working at a distance of 30 ft
range in the presence of very light fog.
Figure 32 shows the experimental arrangement used for testing heavy weapon
codes . The code corresponding to M I05 weapon was selected for our tests. Since the
one-way range available at the test site was 2.5 km, a set of gold-coated retroref'lectors
was placed at a distance of 2.5 km from the transmitter location. The smoke was
released along its path.
Figure 33 illustrates the results of heavy weapon code transmission. The received
sequence was reconditioned to be inverted TTL logic such that it was compatible with
the decoder requirements
In summary, data communication at MWIR wavelengths for obscurant penetra-
tion is discussed in this section. The MWIR transceiver, that provided high peak power
and short pulses, was designed for MILES applications. Weapon codes corresponding
to small and heavy weapon platforms were successfully transmitted though fog. The
MWIR MILES transmitter concept, in general, is applicable to all types of MILES
Fig. 32. The MWIR MILES transmitter setup. The MWIR detector and the collectingoptics are
located on the top of transmitter. A visible CCDcameraon the right is used for bore sighting.
platforms. In the following subsection, a bit map image transmission at MWIR wave-
lengths using this device is presented. Test images consisting of 50 x 40 pixels and
100 x 80 pixels were successfully transmitted through free space .
5.2. Image Transmission in MWIR using an OPO
There is a need for free-space, short range, compact, secure, eye-safe, obscurant pen-
etrating and cost effective image transmission systems for defense applications. Such
systems can playa significant role in battlefield communications. The MWIR MILES
transmitter scheme described in the previous section is capable of providing multi-
ple functions . Besides weapon code transmission, other functions demonstrated us-
ing MWIR MILES include: Direct Detection LADAR, Target illumination, aimpoint-
ing, Monostatic Reflectance Distribution Function (MRDF) measurements, Bistatic
Reflectance Distribution Function (BRDF) measurements and Image Transmission
through obscurants. The MWIR MILES device can be easily adapted for Combat ID
applications and flying-spot active imaging through obscurants. In this section. the
feasibility of image transmission using the MWIR MILES transceiver through low
visibility conditions is discussed. [45]
The MWIR MILES transceiver, consisting of a passively Q- switched, short-pulsed
Nd:YAG laser pumping a PPLN based OPO and a Dember effect detector was utilized
for image transmission experiments. The MWIR MILES transceiver generates pulse
position waveform s. The optical data link consisting of transmitter drive electronics,
pulse conditioning electronics and a computer generating pulses compatible with the
24oo-baud rate RS232 receiver was utilized. Data formatting and RS232 transmission
and reception were achieved using a computer. Data formatting transformed an arbi-
trary image file format compatible with the basic operation of pump laser. Images were
transmitted at a date rate of 2400 kbits/ sec with 16 bits/pixel. Test images consisting of
50 x 40 pixels and 100 x 80 pixels were transmitted through free-space filled with light
fog (visibility ,,-,5 ft) up to 120 ft. Fog was generated using a theatrical fog generator.
I I
~
I I I I: I I
, ... ...
I ·· 1. .. \ 1
I, I I . i
I
.il I. I II I I I I ~
I I I
I I II I I I !IIII 3
.. . . .... .. .. .. . . .. .. .. .. . . . . . . .. .... . . . .
- --
~t 134. 22 ~~ :ii. t . 450, !.It
S 2 DC 3. 5 V
[
Fig. 33. Oscilloscope traces of the M I05 weapon code sequence transmission. Top trace is the
received code sequence (before decoding), middle trace is the transmitted code sequence, and the
bottom trace is the inverted decoded code sequence.
Besides OPOs, the proposed concept can be extended to optical parametric amplifiers,
Raman lasers and other nonlinear optical devices to achieve multi-functionality.
The elements of the data link are illustrated in Fig. 34. The data link consisted of
a computer, a level shifter and pulse generator, the laser drive electronic s, the MILES
transmitter, a Dember effect detector, the pre-amplifier and threshold detection cir-
cuit, and a pulse generator and level shifter. The Q- switch timing of the passively
Q-switched pump laser was electronically adjustable. The externally generated trigger
pulse to the laser diode driver starts the laser diode current pulse. The diode current
pulsewidth is variable and can be set to a predetermined value. As soon as the Q-
switched laser pulse is emitted, the diode current pulse is shut off to eliminate spurious
pulses. In our case, the laser pulse emission was typically adjusted to occur after 80
J.Ls from the start of laser diode pulse. The Q-switch laser pulse onset was adjusted by
varying the laser cavity parameters and the Q-switch crystal transmission.
A computer was used for data formatting and RS232 transmission and reception .
The level shifter and pulse generator between the computer and the laser drive elec-
tronics was required to generate pulses compatible with the laser drive electronics and
the basic laser operation. The level shifter and pulse generator between the thresh-
old detection circuit and the computer was required to "stretch" the detected pulses
to a length compatible with the 2400 baud rate RS232 receiver. Data formatting was
Fig. 34. The schematic of imaging scheme .
required to make the arbitrary image file format compatible with the basic operation
of the laser. For RS232 computer serial data transmission the output will remain at a
high voltage level for a "I " data bit level and remain at that level until a "0" data bit
is encountered or until the data byte is transmitted. The MILES transmitter can only
respond to distinct drive pulses. Driving the laser electronics with a high level for an
extended period will only result in multiple erroneous pulses being sent out of the
laser (after the first pulse the laser emits pulses at a high PRF). This is the result of
the build-up time of the laser varying in the short term such that a specific drive level
to create exact correspondence of output pulses to high level drive voltage duration.
Subsequently, to prevent the occurrence of concatenated" I"s (high level laser drive
voltage) in the data output, "O"s are inserted in the data for every other bit. For RS232
data format there is a I transmitted as a "Stop" bit with every word transmitted, which
is appended to the eight bit of data. To accommodate the stop bit the "O"s are inserted
in the even bit location, the LSB being bit I.
For data framing error mitigation, a synchronization bit pattern is inserted in the
data and the output data stream is reformatted to conform to the synchronization
rule. The error sources are from undetected pulses or false alarms in the data that may
occur during transmission . For data reception the program detects this synchronization
pattern and attempt to restructure the data to maintain the synchronization logic. After
synchronizing the formatting is translated and the image is returned to its original
format, including any residual transmission errors.
Figure 35 illustrates transmitted and received images through light fog over 120
ft at 3.82 fLm. For a typical case of operation, the transmitter PRF and pulse length
were 10 kHz and IOn s. The images consisted of low resolution (50 x 40 pixels) and
high resolution (100 x 80 pixels). The data rate selected was 2400 bits/sec with 16
bits/pixel. No bit errors were detected in this demonstration. The dimensions of the
room used to contain the fog limited the transmission range. The interface electronics
and the laser PRF limited the data rate.
However, energy scaling and increasing the transmitter PRF combined with im-
provements in electronics hardware can alleviate these deficiencies to a certain extent.
For improved timing control and higher PRF, the passive Q-switch can be replaced
by an electro-optic or an acousto-optic Q-switch at the expense of higher power con-
sumption and relatively comple x electronic drivers.
Co) (b)
«) (d)
Fig. 35. Image transmission through light fog (visibility rv 5 ft) using the MILES unit. Data
rate is 2400 kbits/sec with 16 bits/pixel. (a) Input low-resolution image (50 x 40 pixels), (b)
Received low-resolution image, (c) Input high-resolution image (100 x 80 pixels), (d) Received
high-resolution image.
Data link was designed to be highly compatible with common computer interface .
Testing demonstrated transmission of image files of over 4 kilobytes with no errors in
a moderately smoke filled path up to 120 ft at a 2400-Hz baud rate. The system proved
the capability to transfer bit-mapped images through moderate smoke/fog and could
have been modified for other image or data formats easily. Initially, MWIR MILES
was designed to be battery operated and hence passive Q-switch scheme was selected .
Such a unit with image transmission capability will be compatible with small weapon
platforms.
In summary, Image transmission using passively Q-switched laser pumped OPOs
is feasible. Increasing the transmitter PRF can increase the data rate. As such, this
concept can be extended to any optical wavelength and to high-power diode-pumped
pulsed laser based transmitters. Such transmitters can be utilized to pump OPOs, optical
parametric amplifiers, frequency up-converters, Raman lasers, and other nonlinear
optical devices to achieve multi-functionality.
6. Summary and Conclusions
The MWIR spectral region is considered to be a low loss atmospheric window. As

such, it is playing a considerable role in lidar for hydrocarbon detection [46,47] and
ladar applications . With the emergence of novel source and detector technologies, it
will be contributing significantly to the development of systems whose functionality
depends upon atmospheric propagation. In the above discussion , salient features of
atmospheric effects on beam propagation , atmospheric modeling, MWIR sources with
emphasis to OPOs, and MWIR detectors are presented. Furthermore, the free-space
data communication and image transmission in the MWIR region using OPOs are
discussed . In conclusion, the MWIR spectral region offers significant advantages for
lidar, ladar, free-space optical communication, illumination, and hard target imaging
applications for military and civilian use.
References
I. M.E. Thomas, D.O. Duncan, Atmospheric transmission, in Atmospheric Propagation of

Radiat ion, vol. 2 of The Infrared & Electro-Optical Systems Handbook, edited by F. G.
Smith, pp. 1-156, 1993.
2. L. Andrews, R. Phillips, Laser Beam Propagation through Random Media (SPIE Optical
Engineering Press , Bellingham, WA, 1998).
3. L. Andrews et. al., Laser beam scintillation with applications (SPIE Optical Engineering
Press, Bellingham, WA, 2001) .
4. J. Goodman, Introduction to Fourier Optics (McGraw-Hili, New York, 1996).
5. W. Cole s et al., Simulation of wave propagation in three-dimensional random media, Appl.
Opt., 34 (12), 2089-2101 (1995) .
6. R. Frehlich, Simulation of laser propagation in a turbulent atmo sphere , Appl. Opt ., 39 (3),
393-397 (2000).
7. American National Standards for Safe Use of Lasers, ANSI ZI36.1-2000, publi shed by
American National Standards Institute, New York, 2000 .
8. M. Tacke, New developments and applications of tunable IR lead salt lasers , Infrared Phys .
Technol. , 36, 447-463 (1995) .
9. J. Hecht, The Laser Handbook (McGraw-Hili, New York, 1992).
10. Improved lead- salt lasers set power record , in Laser Focu s World, vol. 33, p. 9, January
1997.
11. H.K. Choi , G.w. Turner, H.Q. Le, InAsSb/InAIAs strained quantum-well lasers emitting at
4.5 J.Lm, Appl. Phys. Lett., 66, 3543-3545 (1995) .
12. J. Faist, F. Capasso, D.L. Sivco, C. Sirtori, A.L. Hutchin son, A.Y. Cho , Quantum cascade
laser, Science, 264, 553-556 (1994).
13. J. Faist, F. Capasso, C. Sirtori, D.L. Sivco, J.N . Baillargeon, A.L. Hutchinson, S.N.G. Chu ,
A.Y. Cho, High power mid-infrared (). '" 5 J.Lm) quantum cascade lasers operating above
room temperature, Appl. Phy s. Lett ., 68, 3680-3682 (1996) .
14. J.F. Pinto, G.H. Rosenblatt, L. Esterowitz, Cont inuous-wave laser action in Er 3+ :YLF at
3.41 11m, Electron . Lett., 30, 1596-1598 (1994) .
15. H. Tobben, Room temperature cw fibre laser at 3.5J.tm in Er3+ -doped ZBLAN glass, Elec-
tron. Lett., 28,1361-1362 (1992).
16. J. Schneider, C. Carbonnier, U.B. Unrau, Continuous wave fiber laser operation at a wave-
length of 3.9 micrometers, OSA TOPS on Advanced Solid-State Lasers, 1,333-334 (1996) .
17. R.C. Eckardt, L. Esterowitz, I. D.Abella, Multiwavelength mid-IR laser emission in Ho:YLF,
paper FM5, Digest for Conference on Lasers and Electro-Optics, p. 160, 1982.
18. R.L. Byer, Optical parametric oscillators, in Quantum Electronics : A Treatise, edited by H.
Rabin and C.L. Tang, pp. 587-701,1975.
19. A. Yariv and P. Yeh, Optical Waves in Crystals (John Wiley & Sons, New York, 1984).
20. V.G. Dmitriev et. al., Handbook ofNonlinear Optical Crystals (Springer- Verlag, New York,
1991).
21. WR. Bosenberg, A. Drobshoff, J.1.Alexander, L.E. Myers , R. L. Byer, Continuous-wave
singly resonant optical parametric oscillator based on periodically poled LiNb03 , Opt. Lett.,
21,713-715 (1996).
22. M. Scheidt, B. Beier, R. Knappe, K.-J. Boller, R. Wallenstein, Diode-la ser-pumpedcontinuous-
wave KTP optical parametric oscillator, J. Opt. Soc. Am. B, 12,2087-2094 (1995).
23. H. Komine, J.M. Fukumoto, WH. Long, Jr., E.A. Stappaerts , Noncritically phase matched
mid-infrared generation in AgGaSe2, IEEE J. Select. Top. Quantum Electron., 1,44-49
(1995) .
24. L.E. Myers, R.C. Eckardt, M.M. Fejer, R.L. Byer, WR. Bosenberg, J.W. Pierce, Quasi-
phase-matched optical parametric oscillators in periodically poled LiNb03, J. Opt. Soc.
Am. B,. 12,2102-2116 (1995) .
25. L.E. Myers, R.C. Eckardt , M.M. Fejer, R.L. Byer, WR. Bosenberg, Multigrating quasi-
phase-matched optical parametric oscillator in period ically poled LiNb03, Opt. Lett., 21,
591-593 (1996) .
26. R. Lavi, A. Englander, R. Lallouz , Highly efficient low-threshold tunable all-solid-state
intracavity optical parametric oscillator in the mid infrared, Opt. Leu ., 21, 8OD-802 (1996) .
27. H. Plaessmann, A. Drobshoff, W.R. Bosenberg, Long-pulse , amplitude-modulated optical
parametric oscillator, Appl. Opt., 35, 5964-5966 (1996) .
28. M. Scheidt, B. Beier, R. Knappe, K.-J. Boller, R.Wallenstein, Diode-laser-pumped continuou s-
wave KTP optical parametric oscillator, J. Opt. Soc. Am. B, 12,2087-2094 (1995) .
29. A.R. Geiger, H. Hemmati, WH. Farr, N.S. Prasad, A Directly Diode Pumped Optical Para-
metric Oscillator, Opt. Len ., 21, 3 (1996).
30. W Koechner, Solid-state Laser Engineering, 4th ed. (Springer, New York, 1996).
31. Selected papers on Optical Parametric Oscillators and Amplifiers and Their Applications ,
edited by Jeffrey H. Hunt, SPIE Milestone Series, Vol. MS 140, Bellingham, WA, 1997.
32. SNLO nonlinear optics code available from A.V. Smith Sandia National Laboratories, Al-
buquerque, NM 87185-1423 .
33. S.L. Brosnan and RL Byer, Optical parametric oscillator threshold and linewidth studies,
IEEE J. Quantum Electron ., QE-15, 415-431 (1979) .
34. S.D. Boyd and D.A. Kleiman, Parametric interaction offocused Gaussian Light beams, J.
Appl. Phys., 19 (8), 3597-3639 (1968) .
35. S. Guha et.al., The effects of focusing on Parametric Oscillation , IEEE J. Quantum Electron .,
QE-18 (5), 907-912 (1982).
36. K. Kaufmann, Detectors cover the spectrum of instrument applications, in Laser Focus
World, vol. 30, pp. 99-105,1994.
37. Eltec Instruments, Inc., Passive Infrared Technology, High Megohm Resistors, Hybrid Elec-
tronics Product Catalog, Daytona Beach, FL.
38. Hamamatsu Corp., Optosemiconductors Condensed Catalog, Bridgewater, NJ.
39. J. Piotrowski, W. Gawron, Z. Djuric, New generation of near-room-temperature photode-
tectors, Opt. Eng., 33,1413-1421 (1994) .
40. S. Blaser, D. Hofstetter, M. Beck, J. Faist, Free-space optical data link using Peltier-cooled
quantum cascade laser, Electron . Lett., 37,12 (2001).
41. Eye-Safe Multiple Integrated Laser Engagement System, Phase I Final Report CTI-TR-
9708.
42. All Solid-State Mid-Wave Infrared Multiple Integrated Laser Engagement System (MILES) ,
Phase II Final report CTI-TR-2001 -27.
43 . Narasimha S. Prasad, Duane D. Smith, James R. Magee, Data communication in mid-IR
using a solid-state laser-pumped optical parametric oscillator, Proc. SPIE, 4821, 214-224
(Dec. 2002).
44. Quantitative Description of Obscuration Factors for Electro-Optical and Millimeter Wave
Systems, DoD-HDBK-178 (ER), 1986.
45. Narasimha S. Prasad, Pat Kratovil, James R. Magee, Image transmission in mid-IR using
a solid-state laser pumped optical parametric oscillator, Proc. SPIE, 4635, 272-277 (April
2002).
46 . R.M. Measures, Lasing Remote Sensing (John Wiley & Sons, New York, 1984).
47. Narasimha S. Prasad, Allen R. Geiger, Remote Sensing of Propane and Methane Using a
Differential Absorption Lidar by topographic reflection, Opt. Eng., 35, 4 (1996).
© 2005SpringerScience-BusinessMedia Inc.
001 : 1O.l007/sI0297-005-0052-2
Quantum cascade laser-based free space optical communications
R. Martini and E.A. Whittaker
Department of Physics and Engineering Physics

Stevens Institute of Technology
Hoboken , NJ 07030
Email:rmartini@stevens .edu
and ewhi t tak@stevens . edu
Abstract. The advent of the quantum cascade laser (QCL) with emission wavelengths
available in the infrared range from 3 JIm through more than 100 JIm opens up the
possibility of exploiting infrared atmospheric transparency windows for free space op-
tical communications. In an effort to establi sh the efficacy of using directly modulated
QCLs for free space communications we have conducted a series of investigations
that demon strate the potential advantages of this technology. In these experiments we
first establish that the QCL has very high modulation bandwidth. We then implement
a practical free space communications link that under conditions of atmospheric fog,
dust and other obscurants offers significant transparency advantage when compared
with near infrared wavelength sources presently used in commercial free space optical
communications links .
1. Introduction
Free space optical communications links have recently received attention in the re-
search and commercial development communities because of their potential to serve
applications requiring high bandwidth with relatively high security to eavesdropping
but without the complexity of installing optical fiber. Commercially available systems
have largely been based on the adaptation of telecommunications components in the
near-infrared spectral region. However, the free space (atmospheric) optical channel
has characteristics that are quite different from fiber-based channels. Molecular absorp-
tion, light scattering from suspended particulates, fog, rain and even snow all conspire
to make a more complex and less controlled transmission medium for free space opti-
cal transmission systems. The choice of source wavelength is thus a critical parameter
394 R. Martiniand E.A. Whittaker
in the design of such systems and it is natural to explore whether a significant advan-
tage may be gained by employing longer wavelength systems. However, existing long
wavelength laser systems do not enjoy the advantages of the typical semiconductor
system employed in telecommunication system such as small size, potential for high
bandwidth direct current modulation and high quantum efficiency. A possible solution
for this niche has recently become available in the form of the quantum cascade laser
(QCL) emitting in the middle infrared (8-10 j.tm). QCL's have established themselves
as versatile semiconductor light sources for the mid-infrared (IR) spectrum and beyond
(). '" 3.5 - 24 j.tm) and are already widely used for spectroscopic applications [I].
We will discuss first the high-speed modulation properties of the QCL, address
then a comparison of FSO experiments between a QCL and NIR laser based link in
controlled low visibility experiments and finally present a recent FSO application of a
QCL based link used for transmission of satellite TV data streams.
2. High Frequency Analog and Digital Modulation
Based on a unipolar lasing mechanism, the QCL possesses unique high-frequency

characteristics with theoretical bandwidths in excess of several hundred Gigahertz
[2]. Several publications address this issue for applications like gain-switching [3],
mode-locking [4], and high-speed modulation without relaxation oscillations [5]. We
report in the following from measurements on the high-speed bandwidth of QCLs as
well as successful high-speed data transmission as reported in [6]. These experiments
were performed using experimental QCLs provided by Lucent Technologies, grown
by molecular beam epitaxy in the GaInAslAlInAs material system, and based on the
"3-well vertical" design of the active region [I] . The results presented herein refer
to 1.25-mm long, 4.5-j.tm-wide deep-etched ridge lasers with an emission wavelength
near 8.1 j.tm.To optimize their high-frequency characteristics, the lasers were packaged
and processed using a chalcogenide lateral waveguide as described in [3].
The inset of Fig. I shows the experimental setup used to measure the high-
frequency modulation response of the specified QCL. The device was biased above
its cw threshold, and modulated by the output of a synthesized signal generator with
frequencies ranging from 0.1 to 10 GHz at a level of 0 dBm. The light output was
transmitted over a distance of I m and detected with a GaAslAIGaAs quantum-well
infrared photodetector (QWIP) [9], packaged for high-speed operation. The resulting
photocurrent was amplified and fed into a microwave spectrum analyzer, where the
modulation amplitude was measured. The main part of Fig. I shows a typical fre-
quency response of this device for a de current of 300 rnA and a laser temperature
of 20 K. The data were normalized to the frequency response of the detector [9] to
reflect the modulation response of the QCLs. Aside from a low-frequency shoulder
around 2 GHz, which we ascribe to residual parasitic effects , the modulation response
remains relatively flat up to roughly 7 GHz, which is adequate for high speed broad-
band transmission. As discussed in [4], the observed modulation response is still fully
parasitics-limited, and nearly independent of the dc current.
To demonstrate the high-speed data transmistion capability of the QCL, the device
was modulated with a non-return-to-zero (NRZ) pseudo-random bit stream (PRBS)
from a bit error rate (BER) test system. This test pattern is the most common test
system to monitor the performance of standard network systems. The quality of the
Quantum cascade laser-based free space optical communications 395
'E -55
CO
::2.
ro -60
c -=
- so""
C> -=
U5 ~= \~f
~t:1
-65
-70'----'---'"--'"--'-..................---'--....................-'-................
0.1 1 10
Frequency [GHz]
Fig. 1. High-frequency modulation response of the I.25-mm-long quantum cascade laser at a
temperature of 20 K. The insetshows schematically the experimental setup.
communication link is thereby measured using an eye-diagram test as well as in the

bit error rate (BER) measurement. In an eye-diagram measurement the clarity of a
decision process between a logical "0" and "I " is determined as "openness" of the
eye, which is created by averaging over multiple transmitted bit patterns, with the
recordingoscilloscopesynchronized with the transmission clock rate. Figure 2 shows
a typical eye diagram observed using the same setup as mentioned above for a 23 1 _
I bit long pattern (amplitude2 Y, corresponding to 10 dBm rf power)at a data rate of
2.5 Gbit/s for a heat-sinktemperature of 20 K and a de current of 250 mAoIn general,
a clear and open eye is observed for heat sink temperatures up to 85 K, demonstrating
that the receiving systemcan distinguishbetweena logical"0" or" I". However, some
deviation in the average "I " and "0" levels can be observed, which we attribute to
heating and cooling of the laser due to preceding long "on" and "off' l-times. This
effect can be prevented using a return-to-zero (RZ) modulation, which is especially
suitable at higher data rates, and at operating temperatures of the laser close to its
maximumcw temperature.
We also performed BER measurements on the QCL data transmission system.
In these tests the average percentageof incorrectly transmitted bits ("0" identified as
"I" and vice versa) within a standard transmission session is evaluated. Typical high
performance networksystems must achievea minimum 10- 9 BER, which allows for
error-freedatacommunications utilizingenhancedforward errorcorrectionalgorithms.
Naturallypreferableare lowerBER as theyenhancethe stabilityof the communication
link. Our QCL transmission link showed error-free data transmission with a BER <
10-12 observedfortemperaturesupto 85Kandovertheentirerangeofdc currenttested
(175 to 300 mA).Toquantifythe powermargins for theseerror-freedata transmission,
a variable beam attenuator was included in the beam path. The threshold value for
10- 9 BER transmission was reached around 3 dB attenuation, which corresponds to
a received modulation power of 500 JLW (-3 dBm). Similar results were obtained
396 R. Martini and E.A. Whittaker
300
.......... 200
> 100
E
..........
CO 0
c
.2> -100
(f)
-200
-300
0.0 0.2 0.4 0.6 0.8 1.0
Time [ns]
Fig. 2. Observedtransmittereye diagramof the QC laser at 20 K.
for attenuation of the electrical modulation amplitude. This relatively high value in
comparison to fiber optic communication systems (1.3 or 1.55 /-Lm) is attributed to
the lower quantum efficiency and higher noise equivalent power (-20 dBm) of the
QWIP detector as well as to the previously discussed deviation in detected digital level.
Nonetheless, these measurements prove the ability of a QCL based FSO link to serve
in high bandwidth digital networks.
2.1. Stability for NIR-Laser and a QCL-Based FSO Link Under Strong Scattering
Aside from the modulation speed, other parameters have to be taken into account for
the design of a QCL-based FSO link. Design improvements in the basic technology are
resulting in steady improvements in optical power, room temperature operation and
maximum wavelength of QCLs but whether these improvements will make a QCL-
based long-wavelength free space links a logical choice for commercial development is
still not clear. Additional factors that must be considered include receiver performance
as well as the specific application environment.
A first step in accessing these parameters was made in recent experiments [9],
which analyzed the transmission characteristics of three commonly used telecommu-
nications wavelengths and the QCL using standard commercial photodetectors without
evaluating the performance of the receivers themselves.
3. Experimental Apparatus
To better understand the wavelength dependence of a free space link we conducted a

systematic study to compare systems using three wavelengths used by commercially
available free space links (0.85, 1.3 and 1.55 /-Lm) with a quantum cascade laser based
link operating at 8 /-Lm. The measurements were performed within a wind tunnel facility
New optical device technologies for ultrafast OTDM systems 397
Windtunnel ->
Lock-In amp lifie rs
Fig. 3. Arrangement oflaser source, detector and wind tunnel for the transmission experiments.
The optical path was approximately 65ft.
that can simulate specificweatherconditionsand allow for detailedcomparisonof the

four wavelengths. The 76-foot-longPacific NorthwestNational Laboratory's (PNNL)
AerosolWindTunnel Research Facilityallows us to inject particles with well defined
size distributions into the beam path and then to measure power transmission and the
bit-error-rate for each of the four wavelength lasers.
Figure 3 shows schematically the setup used for these measurements. The PNNL
wind tunnel was modified by adding optical windows at either end of the 76-foot-
long test leg of the tunnel.The three NIR wavelengths passed through two BK7 glass
windows and the QCL beam was directed through an Amtir-I window. The four dif-
ferent optical links were sent through the 65-foot-long wind tunnel and detected by
independentphotodetectors (Thorlabs PDA-400) as well as a liquid-N, cooled MCT
detector. All lasers were operated in cw-mode above threshold and were electrically
modulated around 25 kHz. The modulation amplitude of the detected radiation was
extracted using lock-in detection. All four transmitted signals were recorded using an
electronicstrip chart recorderimplementedusing LabView® on a personalcomputer.
As the intensity variedover several ordersof magnitude, the scale settings on the lock-
in amplifiers were changed several times. Raw data were recordedand then processed
by factoring in the scale changes and referencing transmitted intensity to initial in-
tensity to produce the smoothly varying plots shown in this report. We also analyzed
the transmittedintensitiesto generate an approximate measureof signal-to-noise ratio
(SNR) by calculating the rms fluctuation in intensity divided into the rms magnitude
of the intensity.
The wind tunnel was equipped with an injection apparatus for water fog, oil fog
and dust particles.A sampling apparatus was used to monitorthe density and in some
cases, the particle mass/size distributionof scatterersin the tunnel. During a particular
measurement, the air velocitywould be set at a fixed value,typically a few meters per
second. The particle injection apparatus would then be activated. The particle density
then gradually increased with time over the course of 30-60 minutes and the laser
intensities were recorded.The particle measuring apparatusgenerally showeda linear
increasein particledensityas a function of time and thus the stripchart recordsof laser
398 R.Martini and E.A. Whittaker
100
Scatter: dust
10
""
~
z;
s
~ 0.1
1000
§
Scatter: oildrops
8 100
10
10
Size[J.lIlll
Fig.4. Typical particle size distributions generated in the wind tunnel foroil (lower) and dust
(upper) particles. The different curves represent different measuring time with increasing con-
centrations of scatter particles.
intensityas a function of time are also uncalibrated measurements oflaser intensityas a

functionof particledensity.In a typicaldata run, we wouldactivatethe particleinjector
and run the measurementuntil the transmittedintensityof the most severelyattenuated
beam dropped below our detection limits. We then turned off the particle injector and
simultaneously activateda ventilator, allowingthe particledensity to diminish and the
transmittedintensities to return to their unattenuatedlevels. Such a data run would last
approximatelyone hour.
The particle size distributiondepends on the type of particle and particle density.
A laser scattering apparatus measured the size distribution of the oil vapor and dust
particles. Some typical plots of size distributions for these two types of particles are
shown in Fig. 4. For dust we obtained a broadband size distribution peaking at '" 1.1
jLm and ranging from 800 nm to 5.5 jLm, whereas the oil vapor distribution peaked
around 850 nm with a range from 600 nm to beyond 6 jLm. With increasing density
the size distribution seemed to be constant for dust particle, whereas for oil vapor the
smaller particle concentration rises first strongly, with a reverting effect for higher
concentration.
In the next two figures, the data obtained for scattering from water fog and oil
fog are shown. In the left-hand side of Fig. 5 the water fog results show a dramatic
decreasein intensity for the three NIR wavelengths, whereasthe right-hand side shows
the extracted SNR ratio separately for each of the communication links separately. At
the highest concentration(at 15:52), the 1.3-and 1.55-jLm beams experiencedover 40
dB losses and the 0.85 jLm somewhat less.
However, the 8-jLm beamexperiencedonly 3 dB of attenuation.The corresponding
SNR charts show a similar result with nearlyconstant SNR for the MIR link and drops
New optical device technologiesfor ultrafastOTDM systems 399
0
3O' ~ 1
20: I
10 . - e,m]
~ =:~
-10
CD
:£
(J)
-20 ",·1
B
c
.Q 10·
-~
Q)
.I
~,-~- -- '-
,/
\.../
•
0 -30
.~ .~ 30 i~'55 ,m .
'E
(J)
Z 20.
I
c
-40 ~ 10:
~ $ 1
~ ---1
30. - It l hm
-50
20~ "------ /"
10 \/
-60 -' - 15:3O-Hfoo

15:15 15:30 15:45 16:00 16:15
Time
Time
Fig. S. Transmission losses for the four different wavelength links are shown on the right as a
function time during which the watervaporconcentrationincreased. Independentmeasurements
show that the concentration of the particles increased linearly with time so the horizontal axis
also can be interpreted as scattering particle concentration. The left side shows the resultant
signal-to-noise ratio as a functionof water vaporconcentration.
in the signal down to (and below) 10 dB for the NIR links. Figure 6 shows the data
set for oil fog. The oil fog differs from the water vapor in that the O.85-p,m beam was
attenuated more than the 1.3- and 1.55-p,m beams but as with the water vapor the 8-
p,m beam shows almost no loss under circum stances that completely attenuate the NIR
beams. The difference between the water and oil fog most likely reflects differences in
particle size distribution . Unfortunately, the laser scattering apparatus that we used to
measure the particle distribution is unable to measure the water fog size distribution
directly so a direct test of this assumption could not be made.
However, the slope of the transmission data obtained during the particle density
increase phase allowed us to make a summary comparison of the attenuation efficacy
of each of the scattering media as a function of the transmission wavelength . To de-
termine the scattering efficiency for the different wavelengths the transmission losses
along the length of the link were calculated assuming a scattering concentration of
1000 scatterers/crrr'. The results of this analysis are shown in Fig. 7. Although there
are slight variations in scattering efficiency for the three NIR wavelengths, in all cases
the 8 p,m beam is almost unaffected by scattering conditions (0.06-0.4 dB), whereas
the NIR beams experience much stronger attenuation ranging from ",2.5 dB for dust
to 10 dB for water, having at least one order of magnitude stronger losses. The case for
water droplets is thereby even more interesting, as it reveals again the different wave-
0
:,- _._-
:f 8 ,m
~:I
ilJ -10
~
en
~
.Q
-20
c .~ 10
ill 40 r'-
0
.~ '
'E
-30 ~ 30 1 -
en
c i 201
~ -40 Ul
10
~. .... .
40 .
30 Im rm
-50 r.
20.
10·
I . . --,
830 nm
-60 13:45
L_ 14:00
. .- 14:15
14:00 14:15
Tirre
lime
Fig. 6. Transmission losses for the four different wavelength links are shown on the right as a
function of increasing oil vapor concentration. The left side shows the resultant signal-to-noise
ratio as a function of oil vapor concentration.
.length dependent behavior compared to dust or oil scattering: Here the link operating
at 850 nm experienced less losses then the other NIR links-whereas for the other
scattering system the losses seemed to be qualitatively stronger for longer wavelength.
We attribute this behavior to a significantly different particle size distribution for the
created water fog.
In an additional set of experiments we evaluated the influence of the scattering
environment on digital communication properties by measuring bit error rates (BER)
and eye diagrams for each of the four wavelengths while stepwise increasing the parti-
cle concentration. For this purpose the sinusoidal modulation was replaced by a PRBS
data stream at a data rate of 34 MBitls (corresponding to a DS-3 data transmission
rate) and the detectors of the NIR wavelength links were replaced by Thorlabs D400FC
detector with I GHz bandwidth (the MCT detector for the MIR link provided enough
bandwidth for the communication measurements). The digital data were then ampli-
fied corresponding to automatic gain compensation in standardized communication
systems and analyzed with a Hewlett-Packard 3784A BERT. For further evaluation
corresponding eye diagrams were recorded in parallel using a Tektronix storage oscil-
loscope.
For the NIR communication links the signal quality decreased drastically with
scatter particle density, resulting in noisier eye diagrams. Utilizing high gain (factor of
125) and based on the relatively high link margin (typically SNR > 30 dB in the start)
Newoptical device technologies for ultrafast OTDM systems 401
10
~9~
~~nm
~31 .3pm
1.55pm
0.1
Water Oil Dust

Scattermaterial
Fig.7. Summaryanalysis of thedata shownin Figs. 3-5 . Byextracting the slopeof the attenuation
curve during the scattering density increase phase we obtain a measureof scatteringefficacy for
each of the wavelengths and each of the scattering types.
error-free data transmission could still be ensured until higher scattering densities,
but typically could not be sustained at the highest densities. The upper part of Fig. 8
shows typical eye diagrams obtained for the 1.3-p,m link under conditions of low oil
concentration (0.57 rug/em") and roughly 20x higher oil concentration (II mg/cm")
with 125 x amplification. The open eye at the low concentration has completely closed
at the high concentration, and the signal has highly degraded from 1.5 V to 200 mV.
On the other hand, data transmission in the MIR-link only slightly affected and no
amplification was needed for the tested environments. The comparable eye diagrams
for the 8 p,m link, shown in the lower part of Fig. 8, shows almost no degradation
and we still obtained error free transmission (BER > 10- 13 ) at the highest oil vapor
concentration. We obtained similar results for each of the other scattering materials .
Within these experiments the performance of free space optical communication
links operating at three near-infrared wavelengths (0.85, 1.3 and 1.55 p,m) using com-
mercially available telecommunications diode lasers and at 8 p,musing an experimental
quantum cascade laser produced at Lucent's Bell Laboratories were compared under
low visibility conditions. To simulate a variety of atmospheric conditions, the four
links were transmitted through the 65-foot-long measurement leg of the Pacific North-
west National Laboratory's (PNNL) Aerosol Wind Tunnel Research Facility. The wind
tunnel was selectively injected with water vapor, oil vapor, and dust particles . We then
measured the transmitted intensity of each of the links as the scattering density in-
creased. We also recorded bit error rate and eye diagrams using a bit error rate tester
operating at 34 MHz. The results show that each of the NIR wavelengths become
strongly attenuated at the highest concentration of scattering particles whereas the
Low oil concentration (0.67 mglcm3) High oil concentrat ion (11 mglc m 3 )
00
x
~
8
E
c .. .,
~
,g ;;- ~
...5 '".
'0
0>
.0
~ . , .o
.
~ -0 .4
<f>
0
E
.
l!!
Cl
..
'0
,.,
0>
W ·2.0
20 ·'0 '0 20
'"
1.0 1.0
X
J;
E
c:
0
0
0
00
.2
al ;;- ;;-
c '::" 00 ~ 0 .0
~0
~
'"
<f>
.
'"
E
eCl
.!l!
..
'0
W
,.,
0>
., 0 ·1. 0
20 10
Time (n_j
.' 0
Time (ns)
20
'"
Fig. 8. The top row shows eye diagrams for the 1.33-Jlm laser link operating at a data rate of 34
MBitls (D5-3) under conditions of low concentration (left) and high concentration. The bottom
row shows eye diagrams for the 8 J.Lm quantum cascade laser link operating at a data rate of 34
MBitls (D5-3) under corresponding conditions oflow concentration (left) and high concentration.
8-J.lm QCL does not experience significant attenuation . Likewise, the NIR links are
unable to sustain error free transmission at the highest attenuation levels but the QCL
link remains-error free under the same particle density conditions.
3.1. Satellite 1VTransmission Using a QCL-Based FSO Link
The first two parts of this chapter addressed the two dominant requirements for QCL
based FSO links: high modulation bandwidth and high transmissivity in scattering
environments. In this final section we report on the application of a QCL-FSO link
for high-bandwidth satellite TV-data transmission as described in [12]. The upper part
of Fig. 9 shows the optical setup of the transmission link. The same QCL's described
in the modulation experiment were employed in this setup. The QCL beam was col-
limated using an f/3 ZnSe-lens and then transmitted over an open-air 100m path to a
retroreflector, mounted on another building of Bell Laboratories in Murray Hill, New
Jersey. The reflected light was collected using an f /9 telescope with an aperture of
76 mm and focused onto a high-speed liquid nitrogen cooled MCT detector (Sagem
HgCdTe 0 II). To compare the effect of the longer wavelength on the link quality and
New optical device technologies for ultrafa st aTDM systems 403
b) ,J Spectrum .Analyzer
.~ -rpi QC-Iaser I- Satellite Set- Top Box
MCT-del.
DC - Source DC - Voltmeter
~ TV M onno r)
Fig. 9. Schem atics of the optical and electrical setups of the transmission links.
stability, a second beam was included in the path, originating from a 0.85 J.Lm diode
laser ( 10 mW output power) and detected with a standard Si-detector. To ensure an
identical beam path and easy adjustment the optics for the outgoing beam and for the
detection were rigidly connected to the telescope.
The lower part of Fig. 9 shows the electrical setup of the QC-Iaser and of the
detection system. The signal was received from a satellite dish using a LNB (low noise
block down converter)-module. This high-frequency signal (750 MHz-I 045 GHz) was
combined with a dc current to drive the QC-Iaser continuously above its threshold . The
modulated laser radiation was transmitted over the total distance of ",,200 m before
it was detected. The de component was split off with a bias-Tee and used as monitor
of the received laser intensity. The high- frequency part was amplified and fed into a
spectrum analyzer as well as into a standard satellite set-top-box connected to a TV
monitor.
Under typical QC laser operating conditions (500 rnA dc current at a temperature
of 25 K) the link could be run continuously and stably for at least five hours. Owing
to the beamsplitter and the multiple optical elements in the outgoing beam path only
7.5 mW of the initial 25-mW output power were actually used for the transmission.
About 10% of the original intensity (0.75 mW) could still be detected under good
weather conditions after the transmission and the collecting telescope optics . The
simultaneously transmitted beam with a wavelength of 0.85 J.Lm had comparable losses,
which are therefore attributed to beam spreading and losses in the optical elements.
A typical example of the transmitted data stream is shown in the upper part of Fig.
10. The modulation in the frequency region from 900 MHz to 1045 GHz contains the
digitally encoded information (QPSK-code: quadrature phase shift keying) consisting
of around 800 television channels and 100 radio channel s. The dashed line in Fig. 10
-70 -70
-75 -75
E
Q5 -80 .9)
'--'
Q)
>
Q) ·85
(1J
5> ·QO - - Tra nsmitted

\.. -00
(j)
·Q5 - - Received -95
-100'---..........- - - - - - - ' - - - - " ' - - - - - ' - - - ' ·100

10 1.2 1.4
Frequ ency (GHz)
Fig. 10. Signal at the output of the LNB (upper curve, dashed) shown together with the signal
after transmission over the free-space link (solid line).
represents the original signal received from the LNB, whereas the solid line shows
a typical signal detected after the QC laser-link. Owing to the limited bandwidth of
the detector, the channels in the higher frequency region were detected with a 10dB
higher loss, reducing the number of actually decodable channels to around 650. The
link power margin is 7 dB corresponding to a received power of 0.125 mW, below
which the receiver becomes unstable.
To evaluate the advantage of the longer wavelength relative to the collinearly
propagating near-infrared beam (0.85 t-tm) the intensity of the latter was monitored
in parallel. For typical weather conditions including clear air, strong rain as well as
a thunderstorm, no differences in sensitivity were observed . Nevertheless, a strongly
pronounced deviation was observed during a dense fog situation, with nearly zero
visibility. Figure II shows the temporal evolution of the detected de intensities for
both laser links, starting at a very dense fog situation in the early morning of 8/14/0 I.
As the fog lifted slowly around 3: 15 AM. the QC-laser link regained transmission
much quicker than the near-IR link. The QC-Iaser link had reached transmission of
nearly 70% of its optimal value, when the intensity of the near-IR link was still below
the detection limit. As a result around 4:00 AM, the mid-IR television link became
stable again, whereas the near-IR link was still unstable almost for another hour. The
superior performance of the QC-Iaser link compared to the near-IR link can readily
be understood from the wavelength dependence of Rayleigh- and Mie-scattering and
relates well to the aforementioned windtunnel measurement.
Following this measurement the conclusion has to be drawn, that reliable free
space optical communications should be possible using the QCL under conditions that
would make communications impossible using shorter wavelengths. As detector and
New optical device technologies for ultrafast OTDM systems 405
mid-IR
link ....... near-IR
link
03:21 03:50 04:19 04:48

Time
Fig. 11. Comparison of the received intensities of the mid-IR (.\ = 8.1 jlm) and the near-IR
(.\ = 0.85 jlm) link as a function of time. Fog came up around 2AM and progressively dissipated
during the measurement.
laser technology continue to improve in the mid-infrared spectral range, this result is
likely to be a driver for commercial development of such links.
Acknowledgments
The work reported in this chapter represents the re sults of collaborative efforts with
the Bell Laboratories-Lucent Technologies Quantum Cascade Laser group (F. Ca-
passo, C. Gmachl, R. Paiella, H.Y. Wang, D. L. Sivco, IN. Baillargeon, A.Y. Cho), the
remote sensing group at Pacific Northwest National Laboratory (W.w. Harper, Y-F.
Su, J.F. Shultz) , and H. C. Liu from National Research Council of Canada and we
gratefully acknowledge the contributions of these collaborators. Portions of this work
were supported by the United States Department of Energy under contract DE-FG03-
99NVl3656 and by the US Army/DARPA under contract DAADI9-00-C-0096. R.
Martini acknowledges the support of Bell Laboratories-Lucent Technologies during
the early portions of this research .
References
I. Killinger. D., Free Space Optics for Laser Communication Through the Air. Optics and
Photonics News, October 2002, pp. 36-42.
2. Capasso, E , Gmachl, c, Paiella, R., Tredicucci, A., Hutchinson, A.L., Sivco, D.L., Bail-
largeon, J.N., Cho, A.Y., and Liu, H.C., New Frontiers in Quantum Cascade Lasers and
Applications, IEEE J. Select. Top. Quant. Electron., 6, 931-947 (2000) and references
therein
3. Capasso,E et aI., IEEE J. Quant. Electron., 38 (6), 511-532 (2002).

4. Faist, J., Capasso, E, Sirtori, C; Sivco, D.L., and Cho, A.Y., Quantum Cascade Lasers,
ChapterVIII,in lntersubbandTransition in QuantumWells: Physicsand DeviceApplication
ll, edited by Liu, H.e. and Capasso, E (Academic Press, London, 2000), pp. 29-40.
5. Paiella,R., Capasso,E, Gmachl, c., Bethea,C.G., Sivco, D.L., Baillargeon, J.N., Hutchin-
son, A.L., Cho, A.Y., and Liu, H.e., Generation and Detection of High-Speed Pulses of
Mid-Infrared RadiationwithIntersubband Semiconductor Lasersand Detectors, IEEEPho-
ton. Technol. Lett ., 12, 780- 782 (2000).
6. Paiella,R., Capasso,E, Gmachl,C; Sivco,D.L., Baillargeon, J.N., Hutchinson, A.L., Cho,
A.Y., and Liu,H.C.,Self-mode-locking inquantumcascadelaserswithgiantultrafastoptical
nonlinearities, Science, 290,1739-1742 (2000).
7. Paiella, R., Martini, R., Capasso, E, Gmachl, Wang, H.Y., Sivco, D.L., Baillargeon, J.N.,
Cho,A.Y, Whittaker, E.A., andLiu, H.C., High-frequency modulation withouttherelaxation
oscillation resonance in quantumcascade lasers, Appl. Phys. Lett ., 19,2526-2528 (2001).
8. Szajowski, P.E, Nykolak, G.,Auburn,JJ., Presby, H.M.,Tourgee, G.E., Korevaar, E., Schus-
ter, J., and Kim, 1.1.,: 2.4 km Free-Space Optical Communication 1550 nm Transmission
Link Operating at 2.5 Gb/s-Experimental Results in Optical Wireless Communications,
1998, edited by Korevaar, EJ., Proc. SPIE 3552, pp. 29-40.
9. Martini, R., Gmachl,c., Falciglia, J., Curti,EG., Bethea c.o., Capasso,E, Whittaker, E.A.,
Paiella, R., Tredicucci, A., Sivco, D.L., and Cho, A.Y., High-speed modulation and free-
space optical audio/video transmission using quantum cascade lasers, Electron. Lett., 37,
111-112 (2001).
10. Blaser, S., Hofstetter, D., Beck, M., and Faist,J., Free-space opticaldata link usingPeltier-
cooled quantumcascade laser,Electron.Lett., 37, 778-780 (200 I).
II. Liu, H.C., Lin, J., Buchanan, M., and Wasilewski, Z.R.,: High-Frequency Quantum-Well
InfraredPhotodetectors Measured by Microwave-Rectification Technique, IEEEJ. Quantum
Electron., 32, 1024-1028 (1996).
12. Martini,R., Gmachl,C., Falciglia, J., Curti,EG., Betheae.G ., Capasso,E, Whittaker, E.A.,
Paiella, R., Tredicucci, A., Sivco, D.L., and Cho, A.Y., High-speed modulation and free-
space optical audio/video transmission using quantum cascade lasers, Electron. Lett., 37,
111-112 (2001).
13. Blaser, S., Hofstetter, D., Beck,M., and Faist,J., Free-space opticaldata link usingPeltier-
cooled quantumcascade laser,Electron. Lett., 37, 778-780 (2001).
14. Martini, R., Gmachl, Paiella, R., Capasso, E, Whittaker, E.A., Liu, H.C., Hwang, H.Y.,
Sivco, D.L., Baillargeon, J.N., and Cho, A.Y., High-speed digital data transmission using
mid-infrared QuantumCascade lasers, Electron. Lett., 37, 1290-1291 (200 I).
© 2007 Springer Science + Business Media Inc.
DOl: 1O.IOO7/sI0297-()06-0078-0
Originally published in 1. Opt. Fiber. Commun. Rep.4,214-224 (2007)
All-weather long-wavelength infrared free spaceoptical

communications
D.P. Hutchinson and R.K. Richards
Engineering Science and Technology Division

Oak Ridge National Laboratory
Bethel Valley Road, Oak Ridge, TN 37831
Abstract. Long-wavelength infraredradiationpossessessuperiortransmissionthrough

commonatmosphericproblemssuch as fog, clouds, and smokethan the shorter wave-
length laser sources in use today. Recent improvements in LWIR laser and modulator
design makespossible reliableoptical replacementsfor radio and microwave commu-
nicationslinks in manyapplications. This paper describescomponentsand techniques
developedfor high-speed, full-duplex all-weather infraredcommunications systems.
1. Introduction
Long-wavelength infraredradiationpossesses better all-weathertransmission than the

shorter wavelength laser sourcesin use today. The superiortransmission throughcom-
monatmospheric problemssuchas fog, clouds, and smokecoupledwithimprovements
in LWIRlaser and modulatordesign makes possible reliable optical replacements for
radio and microwave communications links in many applications Another advantage
of LWIR laser radiation is the inherent eye safety of this wavelength region. To take
advantage of the transmission characteristics of infrared radiation in the atmosphere,
we havedeveloped components fora high-speed, full-duplexall-weatherinfraredcom-
munications link for ranges up to 12 km. In supportof this program,we havedesigned
andconstructeda compact,RF-driven waveguide C02 laseranda dielectric-waveguide
Stark modulator. The modulatoris based on the Stark shift of NH2D (deuteratedam-
monia). The laser is an RF-driven waveguide design.
408 D.P. Hutchinson and R.K. Richards
2. Atmospheric Transmission: The Case for LWIR
Transmission of optical radiation through the atmosphere is reduced by molecular ab-

sorption and aerosol extinction (absorption + scattering). Absorption and scattering
are highly wavelength dependent phenomenon. For this reason, selecting a laser wave-
length may be the most important variable when considering a given communications
application. This section on atmospheric transmission includes four subsections that
we feel are the most important aspects to consider for an infrared communications
and/or lidar application on the battlefield . The first subsection, molecular absorption,
uses the well known HITRAN [1] database to show the atmospheric transmission su-
periority of the 13C1602-isotope wavelengths (rv 10.8 p,m) over the standard 12C1602
laser wavelengths (rv 10.6 p,m). The remainder of the subsections deal with aerosols.
Aerosols, particles suspended in the atmosphere, broadly encompass everything from
microscopic dust to macroscopic fog droplets. A baseline continental aerosol extinction
model is presented, then smoke and fog are covered individually. Aerosol extinction
models for C02 wavelengths are plotted in comparison to visible, and when possible,
1.55 p,m wavelengths . The aerosol extinction coefficients available for the standard
12C02 ' wavelengths are assumed to be representative of the 13C02 isotope wave-
lengths as well. Owing to the numerous possible combinations of variables within
each subsection one or two representative graphs will be provided and trends and
exceptions will be noted in the paragraphs.
2.1. Molecular Absorption Model
The molecular absorption study was conducted with the computer program HITRAN-
PC [2]. The transmission calculations, shown in Figs. I and 2, were preformed with the
1976 U.S. Standard for partial pressures of the constituent gases in air. The roundtrip
path length used in the calculations is 12 km. The calculations used a standard pressure
of 1 atm and temperature of 296°K. Figure 1 shows the atmospheric transmission
around the standard 12C1602 laser wavelengths . The most intense P (20) line at 10.59
p,m only provides a 40% transmission. Other nearby laser lines are also shown in Fig.
1, and a similar reduction in radiation intensity is also experienced due to the natural
abundance of 12C1602 in the atmosphere. Figure 2 shows a very promising alternative,
the 13C1602-isotope. The three specific laser lines shown in Fig. 2 are the R(20), R( 18),
and R(l6) at the wavelengths of 10.768, 10.784, and 10.80 p,m, respectively. Each of
these laser lines provides more than 80% transmission over a total distance of 12 km
at the given conditions.
In addition, a few of the other R-branch transitions of this isotope also provide
over 95% transmission.
2.2. Smoke Extinction Model
The smoke extinction model employed in this subsection is built around optical theory
of small particles that group together to form larger particles . For the sake of clarity, the
small particles will be referred to as spherules and the larger particles will be referred
to as clusters . Each of these clusters has a fractal dimension, which describes how well
packed the individual spherules are within the cluster. A minimum fractal dimension of
Transm issio n vs wa velengt h f or 12 Km path tor 12 C0 2

T - 293 K
10.591
0.8
0.6 : ... 1......
0.4 ... .
,I
'!
0' '· fI - I
10.61
,
···· i··············· l h···················1
10.63 10.6 5
Wa vele ng th {micro ns}
Fig. 1. Atmospheric transmission for a 12-km range .
Trans mission \IS Hum id it y for 12 K m path tor 1Ie 0 2

T = 293 K
1.0 = ;;::---;""""'::o<r = - - --="....-- - - - -,
0 .2 .
~0'="
.75=----'---~':::----""----,-'..,---JW-----l
10 .79 10 .83 10.87
Wavele ngth lmlt rons)
Fig. 2. Transmission of three lines for the Carbon-I 3 isotope of carbon dioxide.
I refers to an array of just barely touching spherules, and a maximum fractal dimension
of 3 refers to a totally compact cluster with no voids. A fractal dimension, D, of 1.78
has been accepted for smoke [3]. The average number of individual spherules in each
cluster is given by [4]:
N = (R /a)D, (I)
where R is the average cluster radius and a is the average spherule radius.
The absorption cross -sections were calculated for vector waves [4]:
au = 47fNka 3Im[(M 2 - 1)/(M 2 + 2)], (2)

c
o 1.0 f':
, :===:::==-'-----:------:----,
'iii
·s...... 0.8 ~\; j..==~.~ ~~==t===- .
c 0.6 \< :..... . j .
l!
I- '. , L-_--,--_----'
0 .4 - - - - - -'~ :: .. ~ _..-_. --
(ij
···· ··············:l>:::~-"'.<~~L.' ··'·" '·"" " 'r " '·

C
+l
o
u
l!
0.2
: : ---.:-~- -_.
.
IL
Density of Smoke (etustersrem")
Fig. 3. Comparison of transmission through smoke.
c 1.0 0'--. : :
o
'iii
...
·s... 0.8
-, -......-......'
· - _:\;~· · -- · ·~~·~-.:J· · ---·· ·· -- ····· -- · ·- · ---I
" ,____________
,....
'
......
,
= ~o!5,:~cr~:ss
----,
......J
c
l!
0.6 ··········,,<....·····r·.. · ·· · · · ·~··· f· · ..·
I- '. : ~~
(ij 0.4 · ······ · ·· · · · · ··· · :>"-~~· ·f ····· ··· · ·· · · · · · · · · · · · · ·· ·i · · ~
c ~.~", '
:8u 0.2 ······ ·r·· ·:~:-.~ , ·--=.:~:.: :.::."4=~~=··~·······

l!
IL OL.-------'----------'-------C.J
o ~ ~ n
Fog Density (mglcm3)
Fig. 4. Fractional transmission of 1.5 and 10 /Lm light through 100 m of fog.
where k = 21r / A and M is the complex index of refraction at wavelength A.

The transmission of radiation is calculated:
(3)
where a is the extinction cross-section, p is the density, and L is the total path length.
The average spherule radius is around 0.01 J.Lm [4]. For smoke the absorption cross-
section is much larger than the scattering cross-section, thus a a can be taken as a good
estimate of the extinction cross -section a.
Figure 3 shows the transmission of 1O-J.Lm, 1.5-J.Lm, and 500-nm radiation versus
smoke density. The total path length is 1000 m. All calculations are for a cluster size of
R =0.1 J.Lm. The indice s of refraction used for smoke particles for 10-J.Lm and 500-nm
were MlO = 3 + i and Mo. 5 = 1.75 + .3i, respectively [4]. Since Berry and Percival
did not provide an index at 1.5 J.Lm for soot particles one was calculated by using the
same value as for visible light, i.e., M 1.5 ~ M O.5 = 1.75 + .3i.
Other sources report smoke particle size within the range of 0.005-0.15 J.Lm range
[4]. This reduction in cluster size would correspondingly yield much longer path
Fig. S. Air-cooled, RF-driven C02 Laser.
lengths (for all wavelengths), but the wavelength trends shown in Fig. 3 will stay the
same. With this in mind, Fig. 3 is best used as a relative comparison of the superiority
of IO-j,tm to visible and 1.5-j,tm radiation.
2.3. Fog Extinction Model
The estimate for extinction of 10- and 1.5-j,tm radiation by fog is shown in Fig. 4.
These extinction coefficients were obtained from graphical data presented in Rensch
and Long [5]. The extinction coefficients for 10- and 1.5-j,tm radiation were 0.55 and 2.0
(dB/km)/(mg/m3), respectively. The extinction coefficient chosen above from Rensch
and Long for 1O-j,tm seems to agree within 10% with the experimentally measured
coefficients in Chimelis [3,6]. The average water droplet radius for the fog is I j,tm and
the path length is 100m.
3. Component Development for LWIR Communications
3.1. Compact RF-Driven Laser
One of the drawbacks to more widespread applications for long wavelength infrared
communications systems is an inexpensive, compact source of 10-j,tm radiation. We
have developed a compact CW RF-driven, air-cooled, sealed-off waveguide C02 laser
featuring a power level of over I W. The hollow Ab03 ceramic waveguide has an
i.d. of 2A-mm and provides a gain length of 20 ern. Our waveguide laser is based on
a design reported by Walsh [7]. A photograph of the laser is shown in Fig. 5. This
laser produces a power level of approximately 1.6 W using 12C02 and 0.8 W using
13C02, both in the EHll dielectric waveguide mode. With a grating installed, our
laser produces 004 W on the I0.59-j,tmline of 12C02 . The output mode structure of the
laser, measured with a thermally sensitive liquid crystal film, appears to be a TEM oo
Gaussian mode. Frequency measurements indicate that the laser operates in a single
transverse and single longitudinal mode.
The gas mixture in the laser is He:C02:N2 :CO in the ratio of 65:18:15 :2, re-
spectively. This is a commercial mixture purchased for a pulsed C02 laser and the
composition has not yet been optimized for our CW waveguide laser. Walsh used a
mixture of He:C02:N 2:Xe in the mixture ratio 77: 10:10:3, respectively.
One end of the laser cavity consists of a 0.5-in.-diameter, 3-m-radius concave ZnSe
95 % reflective output coupler attached to the waveguide through a brass bellows. The
ceramic waveguide is attached with epoxy to a brass fitting soldered to the bellows.
Epoxy is also used to attach a ZnSe Brewster window is to the other end of the
ceramic tube, which was ground to the Brewster's angle. A 150 l/mm flat master
grating mounted on a piezoelectric actuator in an adjustable mirror mount forms the
other end of the cavity. A I-in.-diameterinvar rod machined flat on opposing sides serve
as a temperature stable mounting surface for the insulating support for the electrodes
and the mirror mounts. The optimum operating pressure of the sealed-off laser is 60-65
torr. The laser is driven by a 58.5-MHz RF amplifier at a power level of approximately
50 W. Machined aluminum heatsinks are mechanically clamped to each side of the
waveguide serve as RF electrodes and provide cooling for the tube. The two electrodes
are shaped to conform to the round dielectric waveguide. A thin coating of heatsink
compound applied to the electrodes during assembly improves thermal contact with the
dielectric waveguide and enhances cooling . One side of the electrode is grounded and
an air wound autotransformer couples RF power to the other electrode . The inductance
of the 4-tum autotransformer resonates with the capacitance formed by the heatsinks
attached to the waveguide at a frequency of58.5 MHz. Another air-core inductor is used
to connect the 50-D coaxial cable from the RF power supply to a tap (approximately
I-tum from the grounded end) on the autotransformer to efficiently couple power to
the laser. The inductance of the coupling coil is adjusted by compressing or expanding
the coil to optimize the impedance match to the 50-D cable. Both coils are constructed
from #12 AWG copper wire.
3.2. Stark-Effect Modulator
Stark-effect modulation occurs when an electric field is applied to a gas molecule

that has a substantial polarization . The applied electric field effectively changes the
energy spacing of the molecular levels changing the optical frequency or wavelength
that is absorbed by the gas. Also, the energy spacing is very small compared to the
energy of the optical photon interacting with the gas. The modulator is filled with
approximately 2 or more torr of partially deuterated ammonia (NH2D), which has a
molecular absorption resonance near the 10.59-jlm wavelength line of emission of a
CO 2 laser. The frequency difference between the absorption resonance and the laser
line is reported to be approximately 2l89-MHz [8] from the laser line. The dotted
curve in Fig. 6 depicts the transmission of a 30-cm cell containing 2-torr of NH 2D
in the absence of an applied electric field. The laser light (located at zero difference
frequency on the scale) is not strongly absorbed by the cell. The solid curve in the left
graph in Fig. 6 shows the transmission of the same cell in the presence of an applied
electric field of 300 V/mm. The electric field causes this absorption line to split into
nine Stark components classified by the designation M = 0, ±1 , ±2 , ±3, and ±4.
The M = 0 component is roughly in the center of the plot at a difference frequency of
-I - I
380V lmm
1.0
i'
0.8 0.8
c
0
'iii
III
0.6 : .. ... 0.6 --- -~ -- ,-
- -- -- -'- - - --
'EIII \ I
c
co OA 0.4
__.I.,
t: l,
0.2 0.2 __ __J, . .
::\',
0
0 3 4 5 234
Frequency Shift (GHz) Frequency Shift (GHz)
Fig. 6. The application ofanelectric field todeuterated ammonia causes the absorption peak to
split into nine components.
2189 MHz and the M = ±4 componentsare at the extreme left and right ends of the
structure. As the electric field is increasedthe M = ±4 absorption component moves
closer to zero difference frequency until the peak of the absorption is in coincidence
with the laser line corresponding to an electric field of approximately 380 VImm. This
situationis depictedin left graph shownin Fig. 6. As the appliedelectric field is varied
from 350 to 380 V/mm, the transmission of the 30-cm-longcell varies from 90% to
75%. If the length of the cell is increased, the modulation amplitudecan exceed 70%.
If we impresseda steady state or de electric field on the electrodes half-way between
350and 380VImm or 365VImm and applieda sinusoidal alternatingor ac voltagewith
an amplitudeof 30-V peak-to-peak, a sinusoidal modulation of the laser beam would
occur. Digital modulation can be achieved by setting the steady field at 350 VImm and
applying a 30-V digital signal to switch the transmission of the cell between the 75%
and 90% states.
3.3. Dielectric Waveguide Modulator
Normally Stark modulators are constructedfrom two parallelelectrodes separated by

a distance much smaller than the width to minimize the variation of the electric field
across the laser beam. The electrodes are placed inside of a dielectric tube, typically
glass or ceramic, held at a pressureappropriatefor the gas used for modulation, typi-
cally a few torr.A typicalspacing wouldbe on the order of 2-mm with a widthof 20 or
more mm and a length of more than 200 mm. One of the problems with this design is
that normalexpansion of the laser beam throughsuch a structurecauses a loss of laser
light due to vignettingof the beam by the electrodes. To reduce this vignetting loss,
we have designed a Stark modulator using a hollow glass dielectric tube to confine
the optical beam with minimum loss. The laser beam is focused into the proper size
by a lens to form a match to the EH1 1 waveguide mode of the dielectric tube. A lens
following the waveguide collimates the beam for propagation to a detector or to other
optical components. The electrodes are external to the dielectric tube that serves to
Le xO-n
r-
10.0
Gl0-55 '.JO- veg lJid e
0.90
AllJMinlJM Etec tr ocre s

2.2
Fig. 7. End viewof O.9-mm i.d. glass modulator. All dimensions are in millimeters.
confine the optical beam and provide a means to operate the modulator at the reduced
pressures required for operation.
To test this concept we constructed a O.9-mm i.d. x 20-cm dielectric waveguide
modulator. The waveguide material is borosilicate glass and is suspended between
two l-cm-wide by 20-cm-Iong flat aluminum electrodes. Lexan inserts are used to fill
the gap between the electrodes outside of the waveguide. These Lexan inserts support
the waveguide, align the electrodes and flatten and enhance the electric field in the
waveguide . A sketch of the cross-section of the modulator is shown in Fig. 7. The ends
of the waveguide we are sealed with ZnSe Brewster windows .
One problem that arises with this external electrode design is charge build-up
from stray ions and electrons in the insulating waveguide. When an electric field is
applied to the electrodes, the stray free charge in the waveguide migrates to the sides of
waveguide nearest the electrodes with electrons attracted to the most positive electrode
and ions attracted to the most negative electrode. This charge migration acts to cancel
the electric field inside the waveguide thereby canceling modulation . We measured the
time required for this canceling charge to build-up and found that it is on the order of
or less than I sec. Because the charge build-up requires a finite time to occur, we have
developed a biasing method that negates the effect of the stray canceling charge.
3.4. AC Biasing Method
The Stark components are created when an electric field is applied to the molecule.
For the deuterated ammonia line that is in near coincidence with the 1O.59-p,m C02
laser line, the selection rules for absorption require that the electric field be applied
perpendicular to the electric field of the optical beam. The polarity of the electric field
is not important. For instance, if the electric field is applied with the positive electrode
on the right and the negative electrode on the left or vice-versa, the absorption structure
is not changed. The absorption for plus and minus M components are the same. The
absorption is proportional only to the magnitude of the electric field, not direction, as
long as the proper polarization relationship between the applied field and the field of
"""bod A·C Bw Vohe to P....
Fig. 8. The applied ac bias on the left produces the equivalent to a steady "de" bias as shown on
the right of the diagram .
lJr
1
I
I
I
." ,' lnn
/:.~ IU~
: L...... ,.
~ .1;
I
Fig. 9. The polarity of the data waveform must match the polarity of the bias waveform to
properly modulate the laser beam.
the laser beam is maintained . Therefore a circuit has been developed that generates
a low frequency (on the order of 100 to 1000 Hz) square wave that is applied to the
electrodes. The waveform produced by the Hac bias" circuit is shown in Fig. 8. The
polarity of the data waveform must match the polarity of the bias waveform to properly
modulate the laser beam. This relationship is shown in Fig. 9. In addition the logic
circuits buffer and temporally store the data and transmit the data waveform during the
constant portion of the bias signal. High voltage amplifiers such as the Apex PA-94
have slew rates of approximately 1000 tpsec. Assuming a bias voltage level of 500 V
and a bias frequency of 100 Hz, the switching transient will cause a duty cycle loss of
only 0.05%. Using this type of biasing and data amplifiers, the waveform transmitted
by the modulator should resemble the waveform shown in Fig. 10. The bandwidth of
the modulator is limited only by the homogeneous linewidth of the deuterated ammonia
gas. Using a filling pressure of 8-torr, a data rate of over 300-Mbits/sec is available.
4. Experimental Results
Using 0.9-mm i.d. x 20-cm dielectric waveguide modulator design discussed above
and shown in Fig, 7, we measured the transmission of the modulator at pressures of 8, II
o
Time
Fig. 10. The total waveform produced by the ac bias circuit combined with the data circuit
appears identical to that produced using a de bias voltage.
0.4 r-..,....__-,--,-----,~.--_.____,_----,-.,____._____,_____,_-,_..,....____,
....j ! 4 ~0 V6LTS ····· 60 0 ~ O L:T S · f · · · · · · ! j j. + .
":::::)::::::j::::::r::::::C -.- ":::::r::::r::::r:::::::::::;::::::r::::r:::::~:::::·
......~ ~ ; ; ~ .;. ; .
.......; ;.
.
.;
.
.;
. .;. .
.:::::!::::::l::::::L:::::t:: ::: :::::: :::::: -- : :~~~: C1 = 32 MH zltorr

......... ................. .... 11 torr
14 torr
o · · · · ; · · · · · · ;· · · · · j · · · · · ·~ · ·
o 500 1000 1500
APPLIED VOLTAG E
Fig. 11. The measured transmission of the 20-cm-long cell shows excellent agreement with
theory.
and 14 torr as a function of the applied voltage. An hac bias" waveform was generated
by applying a square wave drive voltage from a signal generator to an iron-core step-
up transformer (a 110 VAC to 6.3 VAC filament transformer operated in reverse). The
bias was scanned from 0 to 1100 V peak and the transmitted signal was measured
with a pyroelectric power meter. The hac bias" frequency was 3000 Hz. The resulting
modulator transmission as a function of applied voltage is shown in Fig. II . The open
circles are the measured data and the solid curves are the calculated transmission
based on a theoretical model taking into account the cell length, pressure, e-field
enhancement and deuterated ammonia fraction. The M = 3 and M = 4 absorption
All-weather long -wavelength infrared free space optical communications 417
peaks are clearly visible. The peak modulation index for this cell is approximately
30%. Using this informati on and verification of the model, we have designed and are
constructing a l-rnm i.d. x 36-cm modulator for a prototype free-space optical link
for high-speed digital transmission experiments.
Reference s
I. 1996 HITRAN Molecular Line Database, U.S. Air Force .

2. HITRAN-PC, versio n 2.5 I, distrib uted by Ont ar Corporation.
3. M. V. Berry and l.C, Percival, "O ptics of fractal clu sters such as smoke," Opt ica Acta 33, 5
(1986).
4. P. Chyl ek, V. Ram aswamy, R. Chen g, and R.G . Pinni ck, "Optical properties and mas s
concentration of carbonaceo us smokes," Appl . Opt. 20, 17 ( 1981).
5. D.B. Rensch and R.K . Long, "Co mparative studies of extinction and backscattering by
aeroso ls, fog , and rain at 10.6 j.L and 0.63 j.L," Appl . Opt. 9, 7 (1970 ).
6. V. Chimelis, "Extinction of C02 laser radiation by fog and rain," Appl. Opt. 21, 18 ( 1982) .
7. C.J. Walsh, "An rf excited circular waveguide C02 laser," Rev. Sci. Instrum . 61, 9 (Sept.
1990 ).
8. A.R. John ston and R D.S. Mel ville, "St ark-E ffect Modul ation of a C0 2 Laser by NH2D ,"
Appl, Phys. Lett . 19, 12 ( 197 1).

Free-Space Laser Communications

Uploaded by

Copyright:

Available Formats

Free-Space Laser Communications

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Free-Space Laser Communications

Uploaded by

Copyright:

Available Formats

OPTICAL AND FIBER

Editorial Board Masat aka Nakasawa

Principles and Advances

ISBN-13: 978-0-387-28652-5 e-ISBN-13: 978-0-387-28677-8

© 2008 Springer Science+Business Media, LLC

Printed on acid-free paper.

Free-space laser communications, also referred to as optical communica-

Chapter 4 provides an overview of state-of-the-art, as well as future optical

in controlled low-visibility experiments is discussed. A recent application of

Dr. UHa Birnbacher Dr. Stephen Hammel

Dr. David Caplan Dr. Donald P. Hutchinson

Dr. Erich Leitgeb Dr. Jennifer C. Ricklin

Atmospheric channel effects on free-space laser communication 9

Free-space laser communication performance in the atmospheric channel 57

Laser communication transmitter and receiver design 109

Free-space laser communications with adaptive optics :

Optical networks, last mile access and applications 273

Communication techniques and coding for atmospheric turbulence channels 303

Optical communications in the mid-wave IR spectral band 347

Quantum cascade laser-based free space optical communications 393

All-weather long-wavelength infrared free space optical communications 407

Recent rapid progress in information and communication technologies has exceeded

/./. Brief History

Some typical scenarios where FSO lasercom can be used are

1.3. Advantages and Challenges

High data rates

Int ers atellite

Merroplitnn link Sate llite-

Fig.T, Free-Space LaserCommunications Architectures for ground-to-ground, ground-to-space,

1.5. Basics ofOperation

The object of the communication system is to transfer data or information from a

2. Understanding Free-Space Laser Communications Systems Performance

2. J. Bit Error Rate

BER = Po /2 + Pl/2 , (I)

2.2. Fundamental Limit of Light Detection

Np = -In(BER) = -In(1O- 9 ) ~ 20 photons. (4)

2.3. BERfor a Random Stochastic Communication Channel

When the communication channel is a turbulent atmosphere with random variations

(BER )Turbulence > (BER )NoTurbulence. (6)

2.4. Wavelength Selection Criteria

The choiceof the transmitting laserwavelength willdependuponthe atmospheric prop-

2.5. Adaptive Optics for Free-Space Laser Communications

Refractive-index inhomogeneities of the turbulent atmosphere cause wave-front dis-

2.6. Coding for Atmospheric Channel

I. http://www.eee.strath .ac.uklug-infoIl9984/comohl .pdf

Atmospheric channel effects on free-space laser communication

Jennifer C. Ricklin 1, Stephen M. Hammel", Frank D. Eaton", and Svetlana L.

1 Defense Advanced Research Projects Agency, Advanced Technology Office

increasing the signal gain. Molecular absorption may be minimized by appropriate

2. Beam Extinction Due to Atmospheric Aerosols and Molecules

Extinction is defined as a reduction in the intensity of a propagating laser beam.There

Extinction describes the attenuation of a laser beam as it passes through a medium

dl; = -Iv{3ext,vds, (I)

f3ext,v = {3sca,v + {3ab s, v , (2)

{3a bs ,v =L (3abs ,v (i). (3)

Tv(hl, h2) = {3ext,v (s ) d s . (4)

2.2. Molecular Extinction

Molecular absorption is a primary factor in beam attenuation. At wavelengths above I

2.3. Molecular Transmittance Codes

2.4. Aerosol Extinction