Proceedings of the World Congress on Engineering and Computer Science 2009 Vol I

WCECS 2009, October 20-22, 2009, San Francisco, USA

Terrestrial Free Space LDPC Coded MIMO

Optical Link

Harinder Sandhu, D. Chadha

Using error-correcting codes can increase the

Abstract – In the present paper we study a terrestrial, performance of the optical wireless links. In this work
line-of-sight free space optical (FSO) communication link using strong LDPC codes [3] are employed to enhance BER
non-repetitive MIMO scheme with On OFF Keying (OOK) performance and reduce the link outage. The LDPC
signal over turbulent propagation medium described by
Gamma-Gamma PDF channel. Due to its underlying low
codes if designed properly can give performance up to the
complexity, VBLAST-ZF (Vertical Bell Labs Layered Space Shannon’s limit.
Time Zero Forcing) detection mechanism is exploited at the This work proposes a power efficient optical link design,
receiver side to increase overall spectral efficiency and channel
based on a non-repetition MIMO system to mitigate the
capacity. The variable rate Low Density Parity Check (LDPC)
codes are employed to enhance the performance of the link.
effects of fading and turbulence, thereby increasing the
channel capacity through spatial multiplexing. Variable rate
Index Terms: FSO, LDPC, MIMO, VBLAST. LDPC codes are used to enhance the performance of the
system.
The remainder of the paper is organized as follows. In
I. INTRODUCTION section II, we review the channel model used in this work.
FSO is one of the future technologies of wireless networks Gamma-Gamma PDF channel model is used based on the
as it meets most of the quality parameters at low Kolmogrov theory. Section III deals with the basic link used
implementation cost. The future emerging wireless networks in this work, the modulation scheme and the VBLAST
over high-speed FSO based links for last-mile provisioning detection algorithm. Section IV describes the LDPC codes
applications have a bright prospect [1]. The atmospheric and in Section V the performance of the OOK modulate
optical systems are strongly influenced by the atmospheric signal with and without LDPC codes is compared and section
molecular absorption, aerosol scattering, and turbulence. The VI gives the conclusion of the work carried out.
system capacity can be increased in rich scattering
environment by using Multi Input Multi Output (MIMO)
scheme [2]. MIMO systems exhibit various significant gains, II. CHANNEL MODEL
which can be distinguished as diversity gain, multiplexing The medium of propagation or the channel dictates the
gain, and array gain. An optical MIMO system employs an performance of a communication link in the wireless domain.
array of lasers and of the photodetectors on the transmitter Therefore, to design a high-performance communication link
and receiver side, respectively. The elements of the detector for the FSO, it is of great importance to characterize the
array comprise PIN/APD detectors as desired by the system channel from the perspective of information theory. Optical
design. Total power level of the input data stream can be communication depends most significantly on channel
distributed over all the elements in a repetitive or variations due to turbulence. In this regard, various channels
non-repetitive fashion. The non-repetitive MIMO is models based on statistical concepts have been developed [4]
spectrally more efficient as compared to repetitive case but utilizing the measured link performance in various conditions
on the expense of system reliability. of turbulence, like Log Normal and Exponential which
closely approximate the results of weak and strong
turbulence conditions. However, Gamma-Gamma PDF
closely models’ experimental results over low to high
turbulence strengths and is most suitable for studying link
performance parameters for slow fading conditions. The
Manuscript submitted on 9th June 2009.
propagation through this channel model is based on
D. Chadha is with Department of Electrical Engineering, Indian Institute of Kolmogrov theory of turbulence [5, 6] in which kinetic
Technology, New Delhi-110016, India.(phone: 91-11-26591075, fax: energy from larger to smaller eddy is transferred without loss.
91-11-26591075, dchadha@ee.iitd.ac.in) Refractive index varies rapidly across eddies causing phase
Harinder Sandhu is with Department of Electrical Engineering, Indian
Institute of Technology, New Delhi-110016,
and amplitude variations to the wavefront. The strength of
turbulence is given by the Rytov variance,
σ R2 = 1.23C n2 k 7 / 6 L11 / 6 which represents the scintillation
index of an unbounded plane wave, where k = 2π / λ is
.

ISBN:978-988-17012-6-8 WCECS 2009

Proceedings of the World Congress on Engineering and Computer Science 2009 Vol I
WCECS 2009, October 20-22, 2009, San Francisco, USA

the wave number, 　 is wavelength, L is the propagation simultaneously illuminated with each data stream signal and
distance and C is the refractive index structure parameter. the space time signal processing using V-BLAST technique
2
n
[7, 8] is carried out to cancel out their interference by
successive cancellation. The individual streams are then
The irradiance of optical field is defined as the product of demapped and multiplexed into a single stream of data. The
two random processes, i.e., I = I x I y , where I x arises from impulse response of the atmospheric channel to the symbols
propagating through it is superimposed in the form a random
large scale turbulent eddies and I y from small-scale eddies.
channel matrix, which has the dimensions of the order of the
Specifically, gamma pdf is used to model both small scale system diversity and follows the PDF as discussed above.
and large scale fluctuations, leading to the so-called The total power is restricted to Po regardless of the number
gamma-gamma pdf, i.e.
(
α +β
) of transmit lasers M. The noise at the receiver comprises of
2(αβ )
2 α +β
−1)
(
the thermal noise, background noise and shot noise and
f (I ) = Kα − β ( 2 αβ I )
2
I
Γ(α )Γ( β ) (1) assumed to be equal at the N receiver-detectors. Also, it is
assumed that the thermal and background noise dominate in
Where, I is the signal intensity (I>0), α and β are parameters comparison with the shot noise, or the noise is signal
of PDF given by equations 2 and 3 below, Γ is the Gamma
independent. However, since the channel does not introduce
any nonlinearity, the overall noise component can consider to
function and K α − β is the modified Bessel function of the be normally distributed.
second kind of order 　　　.
0.49σ R In MIMO, each pair of transmitter and receiver is linked
2
α = (exp[ ] − 1)−1 through an independently and identically distributed path.
(1 + 1.11σ R )
12 / 5 5 / 6
With this formulation of symbol propagation and reception,
(2)
the decision at the receiver is hard or soft based on the
received vector Y which is N-dimensional. The vector
0.51σ R
2
contains a mixture of interfering signals that have been
β = (exp[ ])−1
(1 + 0.69σ R ) received simultaneously. The basic vector equation that
12 / 5 5 / 6
(3) describes the channel with narrowband assumption is,

Besides the gamma-gamma channel model, we assume that Y = HX + N o (4)

the spatial coherence distance of the fields at the detector is
large relative to the size of one detector, but the field is
Where H is the channel matrix impulse response
independent across detectors. We also assume the channel is
(Gamma-Gamma distributed) of M columns and N rows, X is
slow fading frequency nonselective. Hence the channel is
taken to be flat over many symbol or block of symbols. M-dimensional transmitted signal vector and noise, N o is
N-dimensional Gaussian.

III. WIRELESS FSO SYSTEM A. ON OFF KEYING

In wireless, MIMO systems are known to provide higher This is the simplest and most common form of ASK
capacity and throughput than SISO. Several schemes have which operates as a switch, using the presence of a carrier
been proposed for MIMO in the RF domain and they differ in wave to indicate a binary one and its absence to indicate a
the complexity at transmitter or the receiver or both. In an binary zero. The OOK modulation scheme is used for its easy
Optical MIMO system, as depicted in Fig. 1, input data is first implementation. The transmitter, operating at a bit rate Rb ,
de-multiplexed into M different sub streams and is then OOK
modulated independently. Different laser transmitters are emits rectangular pulses of duration 1 / Rb and intensity 2P
used to couple these streams to the channel, which is a to generate bit ‘one’, and no pulse to generate bit ‘zero’. The
gamma-gamma channel, modeled for a distance of 1 km. bandwidth required by OOK is approximately Rb = 1 / T ,
the inverse of the pulse width. Assuming matched filtering
with a maximum likelihood receiver, the bit error rate (BER)
of an OOK system is:
⎛ P ⎞
BERook = Q⎜ ⎟ (5)
⎜ N R ⎟
⎝ 0 b ⎠

Where N 0 is the power spectral density of the white

Gaussian noise and Q (x) is the Q-function.
In a terrestrial FSO link, there is a limitation of peak
power rather than the average power. Hence OOK
modulation scheme is used in this work. It is also bandwidth
efficient and can be directly implemented with any linear
block-coding scheme.
At the receiver the signal is converted back to electrical
signal which is affected by noise. The detector array is

ISBN:978-988-17012-6-8 WCECS 2009

Proceedings of the World Congress on Engineering and Computer Science 2009 Vol I
WCECS 2009, October 20-22, 2009, San Francisco, USA

over independently and identically distributed turbulence

B. VBLAST Detection Algorithm [7], [8]
channel. The simulations have been carried out over (2,2) and
MIMO detection can be most efficiently carried out using (4,4) MIMO systems. The aim of implementing different
ML (Maximum Likelihood) decoding but demands a highly parity check matrices for codewords corresponding to each
complex implementation and is therefore not cost effective as layer is to avoid lower number of ones in the columns of
it tends to cause large delay in processing. MAP (Maximum higher rate parity check matrices, as this will render these
Aposteriori Probability) using sphere decoder carries out codes unsuitable for detection by Belief Propagation
search over a smaller domain and thus reduces the algorithm. The higher rate LDPC codes have been produced
complexity of ML while providing same performance. by drawing submatrices from the parent irregular parity
Among the many sub-optimal schemes suggested, VBLAST check matrix by row elimination after converting it into a
with Zero Forcing (VBLAST-ZF) offers ML like lower triangular form. The codewords so generated are
performance at reduced complexity. having the same code lengths though the code rates are
VBLAST works on the principle that transmitters operate different. This implies that the lower layers correspond to
co-channel and symbol synchronized using same modulation higher rate codes to minimize the error propagated through
schemes for transmitting independent data sub streams. The V-BLAST detection.
channel information being known only at the receiver, there
is no power adaptation at the transmitter side, i.e. equal
power levels are radiated at each transmitter in a scaled V. SIMULATION RESULTS
manner so that the total power remains constant irrespective The simulations were performed in the Matlab
of the number of transmitters. V-BLAST is a recursive environment using 1550 nm wavelength, peak power limited
procedure in which we decode the “strongest” signal first, transmitters with OOK modulation, over low turbulence
then subtracting this strongest signal from the received strength of 0.1. BER performance was studied for (2, 2) and
signal, proceed to decode the strongest signal of the (4, 4) system using variable rate LDPC code with block
remaining transmit signals, and so on. The optimum length 200 and 1008. A fixed link length of 1 km in terrestrial
detection order in such a nulling and cancellation strategy is domain was used.
from the strongest to the weakest signal. Fig. 2 shows the block diagram of the simulator
implementation using variable rate LDPC encoders.

IV. LOW DENSITY PARITY CHECK

CODES
Low-density parity-check (LDPC) codes proposed by
Gallager in the 1960’s, and later discovered by Mackay and
Neal appears as a class of codes that can yield very good
performance near Shannon limit and are suited for
implementations that make heavy use of parallelism. They
are a class of linear block codes. These codes are constructed
with the help of a parity check matrix, which is sparse, i.e., it
contains only a few 1’s in comparison to the number of 0’s in
the rows as well as columns. The performance of LDPC code
depends on the decoding complexity, which is directly
related to the density of 1s in this matrix.

In this work, we use LDPC codes [3] over MIMO antenna

systems over a flat fading channel modeled as
Gamma-Gamma pdf, which satisfies the shallow and deep
Fig.2 Block diagram of variable rate LDPC coded
turbulence regimes. V-BLAST detection is used, which has
systems
the limitation of error propagation [7] resulting in the poor
SNR of first detected data streams. Therefore, variable rate
LDPC codes are used. This can overcome the problem by
Fig 3 shows improvement in performance of a variable
assigning LDPC codes with different code rate to each
rate LDPC coded system with 2 degrees of freedom in
transmitter. By assigning stronger LDPC codes in lower layer
MIMO over the uncoded case. The code rates employed are
reduction in error propagation can be faciliated. Different
0.665 and 0.58 for transmit antennas 1 and 2, respectively. At
parity check matrices are employed for generating these −4
codewords. a BER value of 10 an improvement of 4 dB and 9 dB
could be obtained over the uncoded channel for the code
We consider a simple detection technique, which is called lengths of 200 and 1008, respectively. Similar simulation
the QR decomposition interference suppression combined results are shown in Fig.4 for a (4, 4) variable rate system.
with interference cancellation [9], [10]. The channel matrix H For system having degree of freedom 4, code rates are 0.75,
can be factored by the QR decomposition into an 0,665, 0.58 and 0.5 for transmit antenna 1, 2, 3 and 4,
orthonormal matrix Q and an upper triangular matrix R such respectively.
that H = QR. When QR decomposition interference
suppression combined with interference cancellation is used

ISBN:978-988-17012-6-8 WCECS 2009

Proceedings of the World Congress on Engineering and Computer Science 2009 Vol I
WCECS 2009, October 20-22, 2009, San Francisco, USA

REFERENCES

[1]. A. Mahdy and J. S. Deogun, “Wireless Optical Communication: A

Survey”, in Wireless Communication and Network Conference
(WCNC), IEEE Communication Society, 2004.
[2]. Ilen Hen, “MIMO Architecture for Wireless Communication”, Intel
Technology Journal, vol. 10, Issue 2, 2006.
[3]. R.G Gallager, Low-Density Parity-Check Codes. Cambridge, MA:
M.I.T.Press, 1963.
[4]. S. Karp, R. Gagliardi, S. Moran, and L. Scotts, Optical Channels, New
York: Plenum, 1998.
[5]. L.C. Andrews, R. L. Phillips and C. Y. Hopen “Laser Beam Scintillation
with Applications” SPIE, 2001.
[6] Jaime A. Anguita, Ivan B. Djordjevic, Mark A. Neifeld and Bane V.
Vasic “Shannon Capacities and Error Correction Codes for Optical
Atmospheric Turbulent Channels” Journal of Optical Networking,
September 2005, Vol 4 No 19. pp 586-601.
[7]. G. Foschini, “Layered space-time architecture for wireless
communication in a fading environment when using multi-element
antennas”, Bell Labs Technical Journal, vol. 1, no. 2, pp. 41-59, 1996.
Fig 3 BER vs SNR for (2,2) System [8]. G. J. Foschini “Detection algorithm and initial laboratory results using
the V-BLAST space-time communication architecture”, Electronics
Letters, vol. 35, no.1, pp. 14-15, Jan. 1999.
[9] Minseok Noh, Namshik Kim, Hyuncheol Park and Hyuckjae Lee, “A
Variable Rate LDPC Coded V-BLAST System”, in Proc. IEEE
Vehicular Technology Conference (VTC), USA, Fall 2004.
[10] B.Vucetic, J. Yuan, Space-Time Coding. John Wiley, 2003.

Fig. 4 BER vs SNR for (4,4) MIMO system.

VI. CONCLUSION
The performance of non-repetitive MIMO optical link
with channel characterized by gamma-gamma PDF has been
carried out. In order to increase the robustness of the system
error correcting LDPC codes have been added. The layered
space time architecture based VBLAST-ZF detection is
chosen in this work to overcome the complexity of ML
detection. As VBLAST-ZF has a limitation in the form of
error propagation to the subsequent layers, variable rate
LDPC codes are used to overcome this. Simulation results
show significant improvement in the performance of the
system over the uncoded case. A higher order of degree of
MIMO system gives further more performance improvement
in terms of BER as compared to the uncoded case of same
order with VBLAST system. This is attributed to the
additional variability in the rates of LDPC codes that is
included in the coded case.

ISBN:978-988-17012-6-8 WCECS 2009