Early Advances in Radar Technology For Aircraft Detection: Donald L. Clark

CLARK
Early Advances in Radar Technology for Aircraft Detection

VOLUME 12, NUMBER 2, 2000 LINCOLN LABORATORY JOURNAL 167
Early Advances in Radar
Technology for Aircraft
Detection
Donald L. Clark
s In its early years, Lincoln Laboratory developed critical components of an air-
defense system to guard North America against the threat of intercontinental
bombers carrying nuclear weapons. Lincoln Laboratory used digital computer
technology to automate several functions of the air-defense system and improve
the quality of digitized radar data processed by the air-defense system. This
article describes some of the experimental and theoretical efforts that led to early
advances in radar technology for aircraft detection.
P
io;icr iixcoix, rui ioiiiuxxii of Lincoln
Laboratory in Lexington, Massachusetts, was
initiated in 1951 to address the problem of
defending the continental United States and Alaska
against intercontinental bombers. Researchers faced
the challenge of applying advanced technology to
achieve the following improvements in the air-de-
fense system: (1) consolidated command and control
at a central post in each air-defense sector of about
100,000 square miles, (2) provided coverage against
low-flying aircraft by supplementing the principal
long-range radars in each sector with numerous
short-range gap-filler radars, (3) automatically trans-
ferred filtered data from each radar to its central com-
mand center, and (4) improved communication be-
tween each command center and its interceptors.
(Reference 1 provides a more extensive account.)
This was an exciting era in the life of the Labora-
tory, with a talented and highly motivated staff, a can-
do spirit, and minimum administrative formality. An
Air Force unit at nearby Hanscom Field provided
substantial logistic support. Project Lincoln set up an
experimental air-defense sector called the Cape Cod
System in southeastern New England, as shown in
Figure 1. The initial long-range radar for the Cape
Cod System, the AN/FPS-3, was an L-band radar
with a nominal range of 200 miles on a high-flying
bomber. Low-flying aircraft could evade the coverage
of the AN/FPS-3 by staying below its horizon. Gap-
filler radars, assembled mostly from World War II
components, operated at S-band with a nominal
range of 32 miles. Later, the Cape Cod System was
extended to include additional long-range radars at
Montauk Point in Long Island and at West Bath,
Maine. These radars supplied only range and azimuth
coordinates. The heights of designated targets were
measured separately by a small number of height-
finder radars.
Data from all the radars were transmitted over or-
dinary leased telephone lines to a command center in
Cambridge, Massachusetts, where the Whirlwind
computer, and later the system prototype AN/FSQ-7
in Lexington, Massachusetts, processed the data in
real time to track aircraft, assist operators to perform
command functions, and guide manned interceptors.
The defense system that grew out of this effort was
called the Semi-Automatic Ground Environment
(SAGE). Developing SAGE was the major activity
during Lincoln Laboratorys early years [2].
The SAGE system also engaged Lincoln Labora-
CLARK
168 LINCOLN LABORATORY JOURNAL VOLUME 12, NUMBER 2, 2000
N
Direction center
Long-range radars (3)
Gap-filler radars (12)
Flight facilities (4)
Height-finding radars (3)
FIGURE 1. Locations of radar sites in the Cape Cod System.
tory in related air-defense problems. For example, in
parallel with developing the Cape Cod System, Lin-
coln Laboratory helped design and develop the Dis-
tant Early Warning (DEW) Line of radars across
northern Alaska, Canada, and Greenland. The need
for highly automated detection during operations in
the harsh environment of the far north complicated
the design criteria for the DEW Line radars. The early
operations of the DEW Line and the Cape Cod Sys-
tem raised fundamental questions about radar signal
processing within an integrated, digitized system.
This article reviews the theory and practice of the
principal problems in the collection, filtering, trans-
mission, and centralized processing of digitized radar
data. A significant component of the DEW Line is
described in the sidebar by Edwin L. Key entitled
The Sentinel Radar.
Automated Data Transmission
To detect low-flying aircraft, a radar cannot avoid il-
luminating the surface of the earth, which produces
clutter, or echoes, from many reflectors other than
aircraft. Distinguishing between clutter and the sig-
nals from aircraft relies on filtering out as much clut-
ter as possible. Effective filtering uses the Doppler ef-
fect and therefore requires a high degree of frequency
stability in the transmitted signal. A radar must be re-
sistant to incidental interference and intentional jam-
ming. Furthermore, the use of automated detection,
data transmission, and subsequent computer process-
ing imposes requirements on signal detection and fil-
tering different from those for use with a display for a
human operator.
Before Project Lincoln was initiated, the Air Force
CLARK
Cambridge Research Laboratory in Cambridge, Mas-
sachusetts, developed a method called Slowed-Down
Video (SDV) that serves in this discussion as a generic
model of automated detection. In SDV, range-gated
data from a scanning radar are digitized with a single
binary digit per range gate. A one value represents a
threshold crossing by the signal plus noise and a zero
value represents no crossing. Within each range gate
the ones were counted in a sliding window as the ra-
dar beam scanned a target. At typical scan rates the ra-
dar beam stayed on a target long enough for a dozen
or more pulse returns from that target. A sufficient
number of ones within a window represented a target
detection and triggered a detection signal that was
transmitted over the telephone line. Digits represent-
ing detections or nondetections for every range gate
and every beamwidth were transmitted sequentially
within the bandwidth of the telephone line. This pro-
cess yielded data at the receiving end suitable for
computer processing and generating a digitized ver-
sion of the radar display. (See also the article entitled
Radar Signal Processing, by Robert J. Purdy et al.,
in this issue.)
A form of digital signal integration can be simply
characterized for this discussion by considering a slid-
ing window observing a single range gate. A counter
totals the threshold crossings within the window. A
detection occurs when a designated number of
threshold crossings is counted. We need to know the
probability of a detection in the window as a function
of the probability of a threshold crossing in each
range gate. In mathematical terms, the sliding win-
dow is m units long and k or more threshold crossings
constitute a detection. J.V. Harrington, who analyzed
this process, used the fact that the probability of de-
tection P
d
is the sum from k to m of the well-known
binomial distribution b(m, k, p), where p is the prob-
ability of a threshold crossing in a single trial [3].
Figure 2 shows P
d
as a steep function of p for repre-
sentative values of m and k. A useful detection prob-
ability of 0.9 can be attained with those values when p
equals 0.625 or greater. Unfortunately, threshold
crossings on noise alone can also trigger false detec-
tions. If the threshold-crossing-on-noise probability p
rises much above 0.08, then the number of false
alarms can greatly outnumber true detections, over-
loading the telephone line and computer with useless
data. A technique used at the time to protect the com-
puter was simply to map out and excise any area with
too many false targets, thereby losing radar coverage
in that area. Clearly, there was a need to provide the
cleanest data possible, which translates into keeping
the false-alarm rate low. Although the tolerable false-
alarm probability P
fa
is not a sharply defined quantity,
in the following discussion we use a P
fa
value of 10
5
,
which would average less than one false target per
scan for a typical system that has a few tens of thou-
sands of range-azimuth cells per antenna scan. This
level corresponds to a threshold-crossing-on-noise
probability p of about 0.08 and is acceptable because
the system should be able to accommodate several
tens of apparent aircraft returns per scan.
Theoretical Work
Wilbur B. Davenport, Edward J. Kelly, Irving S.
Reed, William L. Root, Irwin Shapiro, and Richard P.
Wishner led the extensive theoretical work done to
establish and understand fundamental limits on radar
detection, filtering, and parameter-estimation capa-
bilities [4, 5]. They worked individually or collabo-
rated in various ways. Reed, a mathematician who
professed to have little understanding of practical
problems, circulated widely around the Laboratory,
talking with engineers about their radar problems.
FIGURE 2. Probability of detection P
d
at the output of a digi-
tal integrator versus the probability p of a threshold crossing
on a single trial. The sliding window is sixteen units long.
Eight or more threshold crossings constitute a detection. A
useful detection probability of 0.9 can be attained with those
values when p equals 0.625 or greater.
10
0
10
1
10
2
10
3
10
4
10
5
10
6
10
7
10
2
10
1
p
P
d
10
0
CLARK
ix rui iaii\ :,,cs, the United
States decided to deploy a line of
radars across northern Alaska,
Canada, and Greenland to pro-
vide early warning of a possible
USSR bomber attack on North
America. This warning system
was the Distant Early Warning
(DEW) Line. From the outset, the
logistical support of radars in such
remote locations was problematic
and expensive. Because air traffic
in these extreme northern regions
was light, however, the architects
of the DEW Line felt that con-
stant observation of radar screens
by human operators was unneces-
sary. Consequently, they planned
for the radars to be unattended
except when aircraft were actually
penetrating the warning zone.
This arrangement was achieved by
providing automatic alarms to
alert operators when aircraft de-
tection occurred. Upon such
alerts, the operators could moni-
tor the radar displays and evaluate
the observed circumstances for
potential threats. Since the opera-
tors did not need to constantly
attend the radar, they were largely
free to perform routine site duties
that would otherwise require ad-
ditional personnel.
Lincoln Laboratory designed
the experimental automatic alarm
system that was used on the modi-
fied AN/TPS-1D radar, which
later became the AN/FPS-19.
(See the article entitled Distant
Early Warning Radars: The Quest
for Automatic Signal Detection,
by F. Robert Naka and William W.
Ward, in this issue.) Because the
design was for an existing radar
with parameters not optimum for
the purpose, the results were not
entirely satisfactory. These short-
comings motivated Herbert G.
Weiss to design and advocate a
new radar that better served the re-
quirements for automatic detec-
tion. The concept was approved,
and in 1954 Lincoln Laboratory
began the development. The new
radar, called Sentinel, was com-
pleted and went on the air for test-
ing at Lexington in 1955. It incor-
porated in a unified system design
many of the innovations that are
described in the accompanying
article. Table A shows the major
parameters of the Sentinel radar.
The Sentinel radar incorpo-
rated several unusual features for
that time. These novel features
resulted from the requirement
that it be essentially unattended.
The radio-frequency (RF) power
source was a high-gain four-cavity
klystron amplifier, which em-
ployed a special modulating an-
ode to pulse the beam. The
klystron in conjunction with two
very stable oscillators provided
high-quality coherence for clutter
cancellation and velocity filtering.
Table A. Parameters of the Sentinel Radar
RF frequency 570630 MHz
Peak power 150 kW
Average power 3 kW
Pulse length 40 sec (detection)
5 sec (threat analysis)
Pulse compression
Barker 13-segment code 39 sec
compressed to 3 sec
Transmitter output tube Klystron with
62-dB gain
Pulse-repetition frequency 500 pulse/sec
Receiver noise figure <6.5 dB
Antenna aperture 45 ft 25 ft
THE S ENTI NEL RADAR
Edwin L. Key
CLARK
The crystal-controlled RF (540 to
600 MHz) stable local oscillator
(STALO) was the rock on which
Sentinels frequency stability
rested. The coherent local oscilla-
tor (COHO) was a crystal-con-
trolled 30-MHz oscillator. Its out-
put was added to the STALOs
output to provide transmitter ex-
citation (570 to 630 MHz). The
COHO signal was also used as the
phase reference for filtering the
echo signals after they had been
downshifted to an intermediate
frequency of 30 MHz plus Dop-
pler shift by mixing the RF echoes
with the STALO signal.
The pulse-repetition frequency
(PRF) was 500/sec to provide a
large unambiguous velocity range
for velocity filtering to reject re-
turns from migrating birds that
were a problem in the Arctic. The
PRF limited the unambiguous
range to 162 nm, but since the
purpose of the radars operation
was to provide a trip-wire-like
warning, this was not a concern.
Within this limited unambiguous
range Sentinel had a substantial
detection margin to allow for deg-
radation. The pulse length for
normal warning operation was 40
sec to provide high pulse energy
for detection and to completely
cover the unambiguous range
with 50 range gates. After detec-
tion the pulse length could be re-
duced to 5 sec for threat analysis.
The Sentinel radar was an early
application of pulse compression,
which had both theoretical and
practical significance.
It was recognized that the 40-
sec pulse would result in rather
large clutter power within a range-
azimuth resolution cell, but the
corresponding transmitted pulse
energy was required for detection
performance. Introducing phase-
coded compression similar to that
used in the AN/FPS-17 radar al-
lowed the radar to transmit a long
pulse while achieving range reso-
lution that corresponded to a
short transmitted pulse. (For a dis-
cussion of the AN/FPS-17 radar,
see the article entitled Radars for
the Detection and Tracking of
Ballistic Missiles, Satellites, and
Planets, by Melvin L. Stone and
Gerald P. Banner, in this issue.) A
simple Barker 13-segment phase-
reversal code was designed for the
Sentinel radar to test whether
pulse compression could reduce
clutter. The pulse-compression
waveform consisted of a 39-sec
pulse with 3-sec subpulses coded
with a particular sequence of
phase reversals, providing 13-to-1
compression. Once the clutter-re-
duction claims were verified, re-
searchers included pulse compres-
sion in the Sentinel radars design.
The Sentinel radar was ac-
quired by the Air Force, renamed
the AN/FPS-30, and manufac-
tured by Bendix. The AN/FPS-30
was deployed by the Air Force in
the extension of the DEW Line
across southern Greenland. It was
reported that the pulse compres-
sion proved to be the savior of the
system. Sometimes a large ice floe
off the coast of Greenland pro-
duced clutter returns with a 40-
sec pulse that exceeded the
subclutter-visibility capabilities of
the radar. However, the 13-to-1
compression provided enough
clutter reduction for the radar to
be able to see targets.
Edwin L. Key joined Lincoln Laboratory
in 1951 and contributed to a variety of
radar programs, including SAGE, the
DEW Line, BMEWS, and the Millstone
Hill facility. He transferred to the MITRE
corporation upon its formation in 1959.
At MITRE, he held positions of increasing
responsibility that culminated in his posi-
tion as senior vice president for research
and engineering. Although retired, he con-
tinues to consult in the areas of radar sys-
tems, radar technology, signal processing,
and communication systems.
These discussions often gave him the idea for an
analysis that he could perform. Then, in useful cross-
fertilization, he would report back on his analytical
results to the engineer who had inspired the idea. He
was one of the more prolific authors of technical re-
ports in the early days of the Laboratory. One of his
major contributions to the digital-processing com-
munity is the origination of the Reed-Solomon error-
correction algorithm, which is discussed at the end of
this article.
A significant problem under investigation during
this time involved signal integration. Our earliest
knowledge of signal integration came from Ruby
Payne-Scott [6], J.I. Marcum [7] and John V. Har-
rington [3]. Confusion existed about the relative ad-
vantages of noncoherent integration and coherent in-
CLARK
tegration. Over a period of time, with contributions
from several people, we learned that coherent integra-
tion can offer a substantial advantage, especially for
signal-to-noise ratios near unity or lower. The advan-
tage is greatest when we know precisely the frequency
of the signal being integrated. The advantage is re-
duced somewhat when the signal can have a range of
frequencies, as with Doppler frequencies from mov-
ing targets with unknown velocities.
Another problem at the time was establishing how
accurately a radar could estimate various target pa-
rameters such as range, velocity, acceleration, and azi-
muth and elevation angle. Several people within the
Laboratory, including Kelly, Roger Manasse, Reed,
and Root, and others outside the Laboratory, includ-
ing P. Swerling, addressed the problem. Their work,
and that of others, is well summarized by Swerling in
chapter 4 of Merrill I. Skolniks Radar Handbook [8].
Chaff, consisting of many small scatterers, was a
countermeasure used widely in World War II to con-
fuse enemy radars. Kelly treated extended targets like
chaff and precipitation with a mathematical model
for the radar echo as coming from a random collec-
tion of scatterers, necessarily highly idealized [9].
The characteristics of the chaff radar echo depend
upon the distribution of the scatterers, their bulk mo-
tion, and their motion relative to each other, as well as
on the characteristics of the radar pulse that illumi-
nates them. The radar echo from many scatterers has
noiselike qualities and coherent qualities. Depending
on the wind, the Doppler frequencies can be high
enough to seriously compromise Doppler-filtering
schemes. Wind shear or turbulence can generate a
spread of Doppler frequencies that further compli-
cates the job of filtering. Changing the radar fre-
quency randomly from pulse to pulse can destroy the
coherence of the echo. (A way of exploiting this fact is
briefly described later.) Other sources provide a more
detailed account of this theoretical work [913].
Looking to the future, well beyond air defense,
Shapiro studied the ability of radars to predict ballis-
tic missile trajectories. His monograph proved to be
the defining work on this subject [13]. Shapiro then
turned his attention to radar astronomy, and he has
had a distinguished scientific career at MIT and at the
Smithsonian Astrophysical Observatory.
Characterization of Clutter
The authors acquaintance with radar clutter began in
the summer of 1952 under the guidance of Robert C.
Butman, an experienced radar engineer. We had the
use of an experimental S-band radar located on the
roof of MITs tallest building. It had a good view of
the area surrounding Cambridge, which included Lo-
gan Airport and Hanscom Field. We spent many
hours observing air traffic and clutter, and experi-
menting with the Doppler-filtering moving-target in-
dicator (MTI). In principle, MTI filtered out clutter
signals from stationary and near-stationary reflectors
while passing signals from aircraft having significant
radial velocities with respect to the radar. A rough but
useful judgment of the ability of the MTI to reject
clutter could be made by comparing a plan position
indicator (PPI) display of the unfiltered radar scene
with a display produced by the MTI. Much less obvi-
ous was the radars ability to follow an aircraft
through a heavily cluttered area. At one point But-
man, taking advantage of Air Force logistic support,
arranged for a flight test to start learning something
about that ability. At the appointed time a trainer air-
craft appeared and circled our site. Butman estab-
lished contact by using a war-surplus radio. He re-
quested the pilot to fly over a heavily cluttered area
northwest of the radar to see how well the MTI could
pick the aircraft out of the clutter. When the aircraft
was lined up on the desired course he asked the pilot
to continue flying in that direction until he received
further instructions. At that point our radio emitted
clouds of smoke and died. Such was the authors in-
troduction to flight tests.
The twelve gap-filler sites of the Cape Cod System
provided an opportunity to study a variety of clutter.
An instrument called the MTI Site Evaluator was
constructed and brought to each site to observe clut-
ter in conjunction with the MTI. Unfortunately, the
name of the device initially terrified some of the site
technicians. The instrument mapped areas where air-
craft might be obscured by strong background clutter.
After the survey of all sites, the site with the largest
and most intensely obscured area was selected for a
flight test. A medium bomber, following radio in-
struction from the site, was tracked on the radar dis-
CLARK
play as it flew over what appeared to be the most se-
verely obscured area. It proved difficult to find an area
where the echo from the bomber was obscured for
more than a scan or two. This finding was consistent
with observations of aircraft targets at other sites. At
the time, we concluded that very intense clutter ech-
oes were due to specular reflections from fixed targets
of limited areal extent, such as tall buildings and wa-
ter towers. If filtered out by the MTI they were not
likely to negate the ability of the radar to track large
aircraft. The term interclutter visibility was later used
to describe this ability to track objects in areas of
strong background clutter.
Some limited observations were made with an S-
band radar on echoes from rain and chaff. Observa-
tions consisted of numerous Doppler spectra mea-
sured within a single range gate with the antenna
pointed to the region of interest. Figures 3(a) and
3(b) show sample spectra from rain and from chaff,
respectively. Both spectra were found to be highly
variable, depending upon wind conditions. A spec-
trum from a C-45 aircraft inbound is shown in Figure
3(c). Either rain or chaff could have Doppler frequen-
cies overlapping those expected from aircraft, thereby
seriously complicating the task of Doppler filtering.
These limited observations of clutter left much to
be desired. They provided only a qualitative basis for
designing filtering schemes. A proper characterization
of clutter awaited another era, when low-altitude,
low-cross-section cruise missiles were the driving con-
cern. (See the article entitled Radars for the Detec-
tion and Tracking of Cruise Missiles, by Lee O.
Upton and Lewis A. Thurman, in this issue.)
Clutter Filtering
The principal technique initially available for dealing
with radar clutter was MTI. In this scheme, the
radars phase-detected video signal was delayed by one
interpulse interval and subtracted from the next video
signal. The signal from a stationary target such as a
water tower would be effectively canceled. The signal
from a target having a significant radial velocity with
respect to the radar would be Doppler-shifted and, in
general, would not cancel. Chapter 17 of Skolniks
Radar Handbook offers a useful discussion of this type
of MTI [14]. During our experiments, the signal was
delayed as a sound wave propagating through a col-
umn of mercury driven by a piezoelectric transducer.
With careful adjustment, sharp nulls could be at-
tained on stationary targets, but they were hard to
maintain due to temperature changes and other insta-
bilities of the delay line. Effort was directed at devel-
oping better means of delaying the video signal. After
considerable trial and error, the best approach was
found to be the use of a delay line in which the sound
wave followed a folded path within a slab of fused
quartz. A cancellation null depth of 37 dB was even-
tually demonstrated, with excellent stability.
The filter response of this type of MTI had the
form of a rectified sine wave with peaks at odd mul-
tiples of half the pulse-repetition frequency (PRF)
FIGURE 3. (a) Doppler spectrum of rainstorm echoes ob-
served with an S-band radar. (b) Doppler spectrum of chaff
echoes observed with an S-band radar. (c) Doppler spec-
trum of a C-45 aircraft inbound observed with an S-band
radar.
(a)
10 500
(b)
500
(c)
10 1500
R
e
l
a
t
i
v
e

i
n
t
e
n
s
i
t
y
R
e
l
a
t
i
v
e

i
n
t
e
n
s
i
t
y
R
e
l
a
t
i
v
e

i
n
t
e
n
s
i
t
y
Frequency (Hz)
Frequency (Hz)
10
Frequency (Hz)
CLARK
and nulls (blind speeds) at multiples of the PRF. (A
blind speed occurred when the target moved radially
an integral number of half wavelengths in one
interpulse period.) Other approaches to Doppler fil-
tering were tried that offered some flexibility in shap-
ing the filter response. In one approach, coherent
video was range-gated and filtered with an analog fil-
ter for each range gate. In another approach, Thomas
C. Bazemore and Bruce Nelson saved the polarities of
the range-gated coherent video as a sequence of bi-
nary digits in a shift register. A diode or resistor array
attached to the shift register allowed periodicities that
correspond to Doppler frequencies to be detected. In
effect one had a bank of elementary digital filters in a
sliding window. This scheme was tested with a few
range gates on an S-band radar. Although it showed
promise, the digital technology available at that time
required one vacuum tube flip-flop per binary digit,
which made a full-scale implementation impractical.
John P. Perry and the author investigated another
approach to Doppler filtering called Sinufly, a tech-
nique borrowed from C.W. Sherwin at the Coordi-
nated Science Laboratory of the University of Illinois.
This technique used a storage tube similar to a cath-
ode-ray tube except that an electrostatic storage sur-
face replaced the phosphor on the face of the tube.
The electron beam, modulated by radar video, was
scanned in a raster across the storage surface, laying
down a pattern of electric charge. When the surface
was completely scanned, the unmodulated beam was
scanned in a raster at right angles to the first to read
out the stored electric charge. This arrangement had
the effect of range-gating the video and allowing the
Doppler frequencies, multiplied by a large factor, to
be read out sequentially, range gate by range gate. By
switching between two tubes all of the radar video
could be captured. With this scheme a single bank of
filters was shared among all range gates, which per-
mitted experimentation with various filtering
schemes. The system had many attractive features but
was limited by the inadequate dynamic range of the
storage tubes that we used. The tubes were laboratory
samples that were never further developed; hence we
had to abandon this scheme.
Although we explored a variety of ideas for clutter
filtering, we found that many sophisticated ap-
proaches could not be reasonably supported by the
memory technology available at the time [15]. The
best memory available to us in this era was the quartz
delay line. A demonstration system exploiting quartz
delay lines is described briefly below. A small mag-
netic-core memory, which provided some interesting
possibilities, arrived too late to be fully exploited.
Jamming and Interference
Andrew Bark and Robert Bergemann, working on ra-
dar countermeasures, demonstrated with a jammer
developed during World War II the radar problems
caused by jamming or by incidental interference. The
jammer consisted of a mechanically tunable continu-
ous-wave (CW) magnetron that radiated through a
small horn antenna in the tail of a test aircraft. Its fre-
quency could be swept rapidly back and forth across a
broad band and its signal was sufficiently strong to
penetrate the sidelobes of the radar antenna as well as
the main lobe. When swept through the radar fre-
quency band, it produced a strong signal that clut-
tered the display and reduced the radars sensitivity.
With automated detection, it overwhelmed the sys-
tem with false alarms.
A remedy, suggested by Robert H. Dicke during an
earlier study, was to limit or clip the amplitude of a
radar signal. This technique became known as the
Dicke Fix. (Actually, the idea of clipping signal ampli-
tude to reduce the effects of interference was a new
application of an old idea. Radio hams had already
been reducing interference effects in their receivers
this way.)
We explored several methods for clipping the am-
plitude of a signal. The simplest, and one of the most
effective variations, was to use a wideband intermedi-
ate-frequency (IF) amplifier with a hard limiter fol-
lowed by a narrowband filter and rectifier. A hard
limiter amplified the receiver noise to drive the lim-
iter to saturation, so that there was negligible varia-
tion of the amplitude at its output with or without
signal or interference present. The narrowband filter
responded more strongly to the radar signal than to
noise, allowing aircraft to be detected. In this context,
narrowband meant a bandwidth matched to the radar
pulse; wideband meant bandwidth several (e.g., ten)
times greater.
CLARK
A receiver using a technique like this was found by
experiment to be much more resistant to jamming
and interference than a more conventional non-limit-
ing receiver. A major frustration at the time was the
lack of an analytical model to describe the effects we
were seeing experimentally. Such a model was devel-
oped some forty years later and is described in the
next section.
Analytic Performance of a Hard-Limiting Receiver
An example using a simple analytic model of a hard-
limiting receiver can help us to understand the hard-
limiting receiver and to compare it with a conven-
tional linear receiver. In this model, a wideband filter
precedes the narrowband filter, where the ratio of
bandwidths is n. The narrowband filter is thus pre-
sented with n independent samples of noise passed by
the wideband filter, plus the signal (if present). The
noise is modeled as having random phase over 360
and a Gaussian amplitude distribution with zero
mean and unit variance. The narrowband filter, tuned
to the signal frequency, adds the n samples. The noise
adds noncoherently; the signal, having constant phase
(modulo 2), adds coherently. In the linear receiver a
threshold is set such that when the threshold is ex-
ceeded, the receiver puts out a one, otherwise, a zero.
In the hard-limiting receiver, each of the n samples is
limited to unit amplitude, with phase the same as that
of the signal plus noise of the linear receiver. We can
picture these samples as n unit vectors that add in ran-
dom-walk fashion when noise predominates, and that
line up to add in phase when signal predominates.
Again, a threshold is set to produce a one when ex-
ceeded and a zero otherwise. (For an analytical discus-
sion of related ideas see Reference 16.)
This model was implemented in a program that
generated n samples of signal plus noise, as described.
The resulting n vectors were added for each receiver,
and the amplitude of the result was compared to a
fixed threshold for each. By repeating this process
many times Monte Carlo fashion, we could estimate
the probability p of detection/false alarm per range
gate. Using the binomial formula referred to earlier
we could calculate the probability of (apparent) de-
tection P after digital integration.
Interference and/or jamming were modeled as a
probability p
I
that a noise sample would be taken
from a distribution with variance I many times that of
the noise, but otherwise similar. This probability p
I
could be chosen anywhere in the range from 0 to 1.
Figure 4 shows the behavior of the two types of re-
ceiver with fixed thresholds set to produce a probabil-
ity p of about 0.08 for a threshold crossing on noise
alone. As p
I
was increased, the false-alarm probability
P
fa
was observed for values of I that are 10 and 100
times the noise. Figure 4 shows that the hard-limiting
receiver maintained P
fa
at about 10
5
for all values of
p
I
. It is clear that the linear receiver produced intoler-
able false-alarm probabilities by the criterion de-
scribed earlier in this section for p
I
greater than about
10
2
. The curves shown are somewhat irregular, due
to the limited statistics of the Monte Carlo model.
The threshold for the linear receiver can be ad-
justed to keep its false-alarm probability nearly con-
stant as p
I
is changed. The threshold has to be raised
substantially as p
I
increases to maintain the false-
alarm probability, thereby desensitizing the receiver.
We can then observe the probability of detection with
a signal present. Figure 5 shows P
d
as a function of p
I
for a signal whose peak amplitude was the square root
of 10 times the variance of the wideband noise. The
curves shown are for I equal to 100 times the noise. It
is apparent that the linear receiver lost its ability to
FIGURE 4. Probability of false alarm P
fa
versus probability of
interference p
I
. The black curve represents data from a linear
receiver with interference 100 times the noise. The red curve
represents data from a linear receiver with interference 10
times the noise. The blue curve represents data from a hard-
limiting receiver with interference 10 times the noise. Only
the hard-limiting receiver maintains a P
fa
value of 10
5
for all
values of p
I
.
10
0
10
1
10
2
10
3
10
4
10
5
10
6
10
7
10
3
10
2
p
I
10
1
P
f
a
10
0
CLARK
detect the signal at small values of p
I
. The hard-limit-
ing receiver maintained a useful detection probability
for values of p
I
up to about 0.4, for the combination
of parameters chosen.
To summarize, in the presence of intense interfer-
ence the linear receiver either produced too many
false alarms or became desensitized. The hard-limit-
ing receiver, by contrast, maintained a constant low
false-alarm probability, together with useful sensitiv-
ity in the presence of severe interference. Franklin A.
Rodgers, who briefly led the group in which this work
was done, described the class of receivers in our ex-
periments as constant false-alarm rate (CFAR), which
was also a play on Rodgerss initials.
Performance of Actual Receivers
Figure 6 shows a practical comparison of two actual
receivers subjected to jamming. Figure 6(a) shows a
time exposure of an output display of a linear receiver
subjected to severe jamming from three airborne jam-
mers on a single plane. Figure 6(b) shows a time expo-
sure of an output display of a hard-limiting receiver
under the same conditions. The hard-limiting re-
ceiver in this case used wideband video with a zero-
crossing counter. The lines of blips represent aircraft
tracks. The results are consistent with the results of
the analytical model described above.
Some time after we had achieved the results briefly
described above, we learned that D. Griffin at Har-
vard University had made some interesting observa-
tions of the ability of bats to rely on their sonar to
navigate and capture insects. In particular, he had
tested bats in the presence of interfering noise, and
they appeared to have an astonishing capability to re-
sist the noise. Had we overlooked something that the
bats could teach us?
The Laboratory arranged to have J.J. Gerald
McCue, assisted by David A. Cahlander, work with
Griffin to follow up on his observations in more de-
tail. They set up a carefully instrumented enclosure in
which the bats could fly freely. Instrumentation in-
cluded a strobe light and high-speed movie camera to
photograph the bats in flight, a microphone and re-
cording system to record their chirps, and an adjust-
able noise generator that filled their enclosure with
continuous near-white noise to jam their sonar. Mov-
ies of flying bats were played back in slow motion.
Recorded chirps were synchronized with the movie
and slowed down sufficiently to bring the chirps
within human audible range. The movies and sound
provided a graphic and convincing way of observing
the bats. It was an enjoyable project whose propensity
to grow had to be restrained. We learned that, alas,
the bats had no more ability to resist jamming than
could be accounted for by existing models [1719].
Master-Oscillator/Power-Amplifier Transmitter
The earliest microwave radars used magnetrons to
generate the power required for the transmitter. Mag-
netrons have an honorable history as the invention
that made possible the development of microwave ra-
dar during World War II. Magnetrons operated as
self-excited oscillators whose characteristics, however,
left something to be desired. Early in the life of Lin-
coln Laboratory an arrangement was worked out with
the Physics Department at Stanford University and
with Varian Associates to supply two S-band klys-
trons, with spares, to the Laboratory. Butman and
Gordon L. Guernsey used these klystrons in an
S-band amplifier chain to generate 1 MW of peak
power and 2 kW of average power. Their driving os-
cillator was a low-power klystron whose frequency
was stabilized by a very high-Q cavity. The klystrons
were thoroughly tested, and they found extensive use
in an experimental radar at the Laboratory, where
FIGURE 5. Probability of detection P
d
at the output of a digi-
tal integrator versus the probability of interference p
I
. Inter-
ference was 100 times noise. Peak signal amplitude was the
square root of 10 times the wideband noise variance.
10
0
10
1
10
2
10
3
10
4
10
5
10
6
10
7
10
3
10
2
10
1
p
I
P
d
10
0
Hard-limiting
detector
Linear
detector
CLARK
they demonstrated stable, reliable, and highly coher-
ent operationa large improvement over the opera-
tion of the magnetron.
Somewhat later Butman and Guernsey tested a six-
cavity klystron developed by Varian Associates for
Hughes Aircraft Company. For this application, the
cavities were stagger-tuned for broad bandwidth at
the expense of gain. In the Lincoln Laboratory tests
the six cavities were synchronously tuned to maxi-
mize the gain. A stable gain of 89 dB was attained
with root-mean-square phase fluctuation under
2.5a remarkable result at that time.
These tests marked the beginning of what came to
be an extensive program in which Butman, Guernsey,
and Clarence W. Jones worked with industry to de-
velop high-power microwave components and to test
numerous high-power klystrons for a variety of appli-
cations. The development of klystrons opened up the
possibility, exploited in a later era, of using a variety of
radar waveforms on demand, such as frequency-
modulated pulses for pulse compression and short
bursts of closely spaced pulses to permit measurement
of very high Doppler frequencies.
Demonstration of an Integrated Radar System
To cap off the work described above we combined
several techniques into an integrated demonstration
system. The key idea for the system, proposed by
Martin Axelbank, was an unusual form of signal inte-
gration. He observed that the hard-limited echo from
an extended target, such as chaff or precipitation, is
decorrelated (i.e., becomes noiselike) if the radar fre-
quency jumps at random from pulse to pulse. By con-
trast, the echo from a large target, such as an aircraft,
has a component that remains steady from pulse to
pulse. Integration, necessarily noncoherent, could en-
hance the signal of an aircraft relative to that of com-
peting chaff or precipitation, thereby yielding a tech-
nique later called superclutter visibility.
The system used a two-pulse delay-line MTI, a
quartz delay-line analog integrator, and a transmitter
that randomly jumped frequency over a broad band
of frequencies after each pair of pulses. The combina-
tion of the hard-limiting receiver and frequency
jumping made the radar highly resistant to jamming.
The two-pulse MTI used a short interpulse period
that broadened the filter null around zero Doppler
frequency and moved the first blind speed out to a
high value, which provided reasonably good Doppler
filtering. The analog integrator in combination with
the frequency jumping was effective in filtering out
chaff and rain clutter. This system was as successful as
FIGURE 6. Comparison of displays from two receivers subjected to three jammers on a single B-47 at
35,000-ft altitude and 10-mi range: (a) time exposure of an output display of a linear receiver and (b) time
exposure of the output of a hard-limiting receiver. The lines of blips represent aircraft tracks.
(a) (b)
CLARK
any that we tried. We could not test it at length, how-
ever, because other radar operators nearby were dis-
tinctly unenthusiastic about the interference effects
on their radars of its frequency-jumping mode.
Error Correction
Finally, attention should be called to some work on
error correction, somewhat out of the main stream of
radar research. As pointed out above, the Cape Cod
System transmitted digitized radar data over tele-
phone lines. At that time, digitized data transmission
was a largely undeveloped art. Telephone lines that
were adequate for voice signals were often far less than
ideal for digital signals, resulting in significant errors
in the received signals. This inadequacy motivated an
investigation of how to reduce errors.
Irving S. Reed and Gustave Solomon investigated
a number of mathematical schemes with potential for
error correction. Their work culminated in publica-
tion of a fundamental and rather abstract mathemati-
cal paper in the Journal of the Society for Industrial and
Applied Mathematics [20]. Although their paper was
little noticed at the time, it contained basic ideas that
have since been developed into powerful and widely
used error-correction schemes, now known as Reed-
Solomon error-correcting codes [21]. The codes have
been used with compact discs, digital audio tape,
high-definition TV systems, and the Voyager and
Galileo spacecraft. Reed and Solomon received the
1995 IEEE Masaru Ibuka Consumer Electronics
Award for this work.
Retrospective
As the digital-computer technology needed for the
SAGE system matured, the MITRE Corporation was
set up in 1958 to oversee the implementation of the
system in the field. Related work at Lincoln Labora-
tory was attenuated. The imminent advent of inter-
continental ballistic missiles and the launch of Sput-
nik I further dampened interest in research on
air-defense techniques. The work on radar from the
Laboratorys first several years, described above, did
not have the practical impact on air-defense radars
that it might have had otherwise. One fruitful result,
however, was the education of a cadre of people at
Lincoln Laboratory in both the underlying theory of
radar detection and parameter estimation and in the
nitty-gritty of practical applications, preparing them
to take on new and greater challenges.
Acknowledgments
I am indebted to the caretakers of the Lincoln Labo-
ratory archives for help in finding early Lincoln Labo-
ratory reports and copying pertinent excerpts.
CLARK
REF ERENCES
1. A.G. Hill, J.W. Forrester, and G.E. Valley, Quarterly Progress
Report, Division 2Aircraft Control and Warning, Division
6Digital Computer, 1 June 1952. (This report was the
first in a long series of quarterly progress reports. Some infor-
mation from later reports in the series has also been used here.)
2. E.C. Freeman, ed., MIT Lincoln Laboratory: Technology in the
National Interest (Lexington, Mass., 1995), pp. 1533.
3. J.V. Harrington, An Analysis of the Target Detection of Re-
peated Signals in Noise by Binary Integration, IRE. Trans. Inf.
Theory 1 (1), 1955, pp. 19.
4. W. L. Root, Remarks, Mostly Historical, on Signal Detection
and Signal Parameter Estimation, Proc. IEEE 75 (11), 1987,
pp. 14461457
5. W.B. Davenport, Jr., and W.L. Root, An Introduction to the
Theory of Random Signals and Noise (McGraw-Hill, New York,
1958).
6. R. Payne-Scott, The Visibility of Small Echoes on PPI Dis-
plays, Proc. IRE 36 (2), 1948, pp. 180196.
7. J.I. Marcum, A Statistical Theory of Target Detection by
Pulsed Radar, and Mathematical Appendix, IRE Trans. Inf.
Theory 6 (2), 1960, pp. 59267 (originally published as
RAND Corp. Res. Mem. RM-754, 1 Dec. 1947, and RM-
753, 1 July 1948).
8. P. Swerling, MTI Radar, in Radar Handbook, M.I. Skolnik,
ed. (McGraw-Hill, New York, 1970), pp. 4.14.14.
9. E.J. Kelly, I.S. Reed, and W.L. Root, The Detection of Radar
Echoes in Noise, SIAM J. 8 (2), 1960, pp. 309341, and
SIAM J. 8 (3), 1960, pp. 481507.
10. E.J. Kelly, The Radar Measurement of Range, Velocity and
Acceleration, IRE Trans. Mil. Electron. 5 (2), 1961, pp. 5157.
11. I.S. Reed, On the Use of Laguerre Polynomials in Treating the
Envelope and Phase Components of Narrow-Band Gaussian
Noise, IRE Trans. Info. Theory 5, Sept. 1959, pp. 102105.
12. R.P. Wishner, Distribution of the Normalized Periodogram
Detector, IRE Trans. Info. Theory 8, Oct. 1962, pp. 342
349.
13. I.I. Shapiro, The Prediction of Ballistic Missile Trajectories from
Radar Observations (McGraw-Hill. New York, 1958).
14. W.W. Schrader, MTI Radar, in Radar Handbook, pp. 17.1
17.60.
15. J.P. Eckert, Jr., A Survey of Digital Computer Memory Sys-
tems, Proc. IRE. 41, Oct. 1953; reprinted in Proc. IEEE 85
(1), 1997, pp. 184197.
16. W.R. Bennett, Methods of Solving Noise Problems, Proc.
IRE 44 (5), 1956, pp. 609638.
17. D.R. Griffin, J.J.G. McCue, and A.D. Grinnell, The Resis-
tance of Bats to Jamming, J. Exp. Zool. 152, 1963, pp. 229
250.
18. D.A. Cahlander, J.J.G. McCue, and F.A. Webster, The Deter-
mination of Distance by Echolocating Bats, Nature 201, 8
Feb. 1964, pp. 544546.
19. J.J.G. McCue, Aural Pulse Compression by Bats and Hu-
mans, J. Acoust. Soc. Am. 40 (3), 1966, pp. 545548.
20. I.S. Reed and G. Solomon, Polynomial Codes over Certain
Finite Fields, SIAM J. 8 (2), 1960, pp. 300304.
21. B.A. Cipra, The Ubiquitous Reed-Solomon Codes, SIAM
News 26 (1), 1993.
CLARK
.
received a bachelors degree in
electrical engineering at the
University of Vermont in
1943. He worked in the re-
search laboratory at Stromberg
Carlson in Rochester, New
York, mainly on magnetic
recording. He earned a Ph.D.
degree in physics at the Uni-
versity of Rochester, where he
experimented with pi mesons.
He joined Project Lincoln in
1952 and enjoyed a 31-year
career at Lincoln Laboratory.
He worked mostly on various
problems of radars at the
cutting edge of technology. As
a result of the experience
described in the article he was
an early advocate for taking
into account during the speci-
fication and design phase of
military radar developments
the likely countermeasures to
which the radars might be
subjected. His group studied
test-range measurement radars,
ballistic-missile-defense radars
and other sensors in a variety
of contexts, and satellite sur-
veillance and identification
radars, exploitingwhere
possibledata from the
Laboratorys extensive field-
measurements program at
Kwajalein. His proudest and
most lasting accomplishment
was bringing a number of
outstanding people to the
Laboratory and helping to
guide their careers. He is a life
member of AAAS and IEEE, a
senior member of APS, an
emeritus member of AAPT,
and a member of Sigma Xi.

Early Advances in Radar Technology For Aircraft Detection: Donald L. Clark

Uploaded by

Copyright:

Available Formats

Early Advances in Radar Technology For Aircraft Detection: Donald L. Clark

Uploaded by

Document Information

Original Title

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

Early Advances in Radar Technology For Aircraft Detection: Donald L. Clark

Uploaded by

Copyright:

Available Formats

CLARK

Early Advances in Radar Technology for Aircraft Detection

You might also like