Beruflich Dokumente
Kultur Dokumente
INTRODUCTION
The AN/APG-66 is the primary sensor for the F- 16
F-16 Pulse Doppler Radar air combat fighter. It is in production and operational in
the U.S. Air Force and a number of NATO countries.
(AN/APG-66) Performance It is a multimode, air-to-air, and air-to-ground radar,
whose downlook mode has high peak power and trans-
mits eight different medium PRF waveforms per dwell. It
is computer controlled, uses coherent processing, and has
a unique single channel design that is described in Sec-
MELVIN B. RINGEL, Member, IEEE tion IB-3. This paper focuses on the detection and false
DAVID H. MOONEY, Senior Member, IEEE alarm performance of this radar in its medium PRF
WILLIAM H. LONG, III downlook mode because its capabilities are most severely
Westinghouse Electric Corporation tested by the ground clutter it encounters in that mode.
General information on airborne pulse Doppler radar,
multiple PRF ranging, and pulse Doppler clutter and
range performance can be found in Hovanessian [1],
The AN/APG-66 is a digital, multimode, flre control radar that Skillman and Mooney [2], Goetz and Albright [3], Moo-
is the primary sensor for the F-16 air combat fighter. The detection ney and Skillman [4], and Ringel [5, 6].
and false alarm performance of this radar are described when it op- Fig. 1 is a simplified functional block diagram of the
erates in its medium PRF pulse Doppler downlook mode. Descrip- downlook mode. The first four blocks are the receiver
tions are included of medium PRF clutter, the AN/APG-66 signal protector, the low noise amplifier, the receiver, and the
processing, the flight tests used to obtain performance data, a com- sampled data assembly or analog to digital converter.
puter simulation of the radar, and the calibration of the simulation. Following these are the main beam clutter canceler, a 64
The detection performance presented is based on both flight tests point fast Fourier transform (FFT), the detector, the con-
stant false alarm rate (CFAR) circuitry, and the ambigu-
and the output of the flight test calibrated simulation. The false
ous range resolver.
alarm performance is based on flight tests and is accompanied by a
The data discussed were obtained from an extensive
discussion of the sources of false alarms.
flight test program undertaken during the period from
June 1977 to October 1980 and from follow-up computer
simulation studies. The remainder of this introduction
gives a summary of the main features of the ground clut-
ter seen by the downlook mode, the signal processing
used, and the computer program that was used to obtain
the simulated detection performance results. The theory
of the computer program is described in [6], but its cali-
bration by means of flight data is contained in Section
IIA of this paper. Section II presents the detection perfor-
mance flight tests and results plus the simulation results.
Section III discusses the false alarm flight tests and re-
sults.
1 2 3
(c + N)I/N
Alt RAMBIG
Range R _
Alttud Altitude
(C + N)/N
FMB FMB + PRF
Frequency
Feuny
0 Alt R- RAMBIG
Fig. 2. Clutter in the range-Doppler space of medium PRF radar.
Fig. 4. Sections through the clutter map.
150 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. AES-19, NO. 1 JANUARY 1983
TARGET: RCS = 1.3 m2 F-16: Scarn ±300 2 Bar VELR, and, if the ambiguous Doppler of the target at a
Aspect = Nose-On EL Angle = - 1.90
VELT = 340 Knots VELI = 474 Knots given PRF coincides with the ambiguous Doppler of the
1n.f
nl
ALT = 500 ft ALT = 5 kit main beam clutter, the target will "fall in the notch" and
be detected at that PRF. Thus the "visibility" of the
I I I I I
not
Upper Bar
UpperBar Clutter
Constant -Y
Model:
c 0.8 y= -12dB target, that is, the number of PRFs for which the target's
0
a Noise Limited (Simulated) ambiguous Doppler does not coincide with the ambiguous
cj 0.6 Doppler region occupied by main beam clutter, depends
0
\ Simulated on VELR.
-
a 0.4
Measuredd More generally, the specific Doppler filter that a tar-
0
get appears in when it is illuminated by a particular PRF
0. 0.2 _- depends on VELR. If we assume lead collision geometry
between the interceptor and the target, both VELR and the
0 In
I
10 12 14 16 18 20 22 24
target azimuth remain constant as the interceptor closes
Range (nmi) on the target. This implies that both the target Doppler
Fig. 7. Per-scan probability of detection versus range (upper bar). and the Doppler of main beam clutter when the radar is
looking at the target remain constant as the interceptor
Target: RCS 1.3 m2
= F-16: Scan 30", 2 Bar closes on the target. Thus a trace of the trajectory of the
Aspect Nose-On EL Angle -4.1°
target through a range Doppler map such as the one in
= =
VELT
= 340 Knots = VELI 474 Knots
1.0
=ALT 500 ft = ALT 5 kft Fig. 2, for a given PRF, would be a vertical line. The
target would start out at the ambiguous range correspond-
c 0.8 ing to its initial range, travel down the line to R = 0,
.2
0
reappear at R = RAMBIG, and continue cycling in this
8 0.6 way until it had closed to its minimum range. Depending
0 on which filter it was detected in at the given PRF, it
=
X0
0.4 would repeatedly encounter sidelobe clutter similar to one
2 or the other of the sections illustrated in Fig. 4.
01
The detection performance of a medium PRF radar in Fig. 9 also shows, however, that the combination of
clutter depends on the LOS velocity VELR of the target. the lead collision assumption with the fixed value of
This is because the Doppler of the target relative to the VELT and the varying value of TASP implies that the tar-
Doppler of main beam clutter is directly proportional to get's azimuth angle must change when the aspect angle
151
RINGEL ET AL: F-16 PULSE DOPPLER RADAR (ANIAPG-66) PERFC)RMANCE
changes. If we ignore the fact that the position of the var- and clutter parameters used are the same as those used to
ious lumps of sidelobe clutter in range-Doppler space de- obtain the results in Section IIA. Only those parameters
pends on the azimuth of the main beam, this change in affecting the radar/target geometry have been varied. The
azimuth causes no confusion. But if we wish to examine performance obtained in this way is not to be confused
the detailed effects of the sidelobe/reflection lobe gain with the example of 2 m2 target performance mentioned
pattern, the change in azimuth with aspect angle can hide in Section IIA.
some significant effects. This is because the position of Fig. 11 and Table I (which includes the flight test R85
the reflection lobe in space, relative to the position of the discussed in Section IIA) present the values of R85 com-
main beam, depends on the azimuth of the main beam puted by the program described in Section IC as a func-
(see Figs. 3 and 15). Furthermore, the Doppler of the tion of VELR for a target at azimuth angles of 0, 30, and
main beam clearly depends on its azimuth and so the po- 50 deg. The column labeled VIS gives the number of
sition of the sidelobe/reflection lobe clutter in range-Dop- PRFs with which the target was visible for that value of
pler space depends on the azimuth of the main beam. VELR. This visibility is independent of target azimuth.
Thus if VELT remains fixed while both TASP and TAZD Included with each value of R85 are VELT, the target's
are varied, the clutter environment of the range doppler true ground speed; VCL, the closing speed between target
space of each PRF changes at the same time as the posi- and interceptor; and TASP, the target's aspect angle.
tion of the target trajectory in the range doppler space of It is clear from the values presented that, for the PRF
each PRF changes. Therefore there is no assurance that set used by the AN/APG-66, R85 behaves quite reason-
this way of varying VELR will show the entire range of ably as VELR changes. This can also be seen from the
effects that the sidelobe clutter can have on range perfor- mean and standard deviation (sigma) given at the bottom
mance. The target and the worst part of the sidelobe clut- of each R85 column.
ter may, for example, "do a dance" in which neither It is interesting to note that the differences in the side-
ever encounters the other or in which they are always lobe clutter at each azimuth can cause targets with the
bumping into each other. same visibility to have significantly different detectability.
To evaluate performance as a function of VELR with This is the case for the example at VELR = 300, where
the kind of detail that does give assurance of showing the the 30 deg azimuth target is detected at 17.4 nmi, while
effects of sidelobe clutter, we therefore adopt the possibly the 0 deg and 50 deg targets are detected at 20.0 and
less intuitive but more systematic procedure illustrated in 20.2 nmi, respectively.
Fig. 10. The clutter environment in the range Doppler It is also noteworthy that, although it is generally the
space of each PRF is "fixed" by choosing, first of all, case that detection range increases with visibility, there
some fixed azimuth angle for the target. The trajectory of are cases for which a target has greater (main beam clut-
the target in the range-Doppler space of each PRF is then ter) visibility but a lower detection range than one having
moved from filter to filter by varying VELR directly, a smaller visibility. An example of this phenomenon is
starting at some minimum value and ending at some max- the pair of targets at AZ = 50, VELR = 650 (R85 -
imum value. The minimum value is chosen as the most 18.0) and AZ = 50, VELR = 450 (R85 = 20.6). The
negative (opening geometry) value of VELR that the tar- fact that the target at AZ 50 and VELR = 600 is de-
get can have and still have the interceptor close on it. The tected at 21.6 nmi shows that the decrease to 18.0 at
maximum value of VELR is chosen as the largest value VELR = 650 is not simply the result of the faster closing
that is consistent with the target's maximum true ground rate of the latter target.
speed VELT.
Target Aspect TASP
Target Velocity (Variable) VELT
\ \
152 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. AES-19, NO. I JANUARY 1983
TABLE I
Range Performance (in nautical miles) as a Function of Azimuth (in
degrees) and VELR (in knots)
350 4 350 824 18.8 423 760 34 19.4 504 655 46 19.9
400 3 400 874 18.4 465 810 31 17.1 540 705 42 17.2
450 4 450 924 19.8 508 860 28 20.3 578 755 39 20.6
500 5 500 974 20.2 553 910 25 20.2 618 805 36 20.3
550 6 550 1024 20.9 599 960 23 20.4 659 855 33 20.9
600 7 600 1074 21.0 645 1010 22 21.1 701 * 905 31 21.6 *
650 5 650 1124 20.3 692* 1060 20 18.0* 745 * 955 29 18.0 *
Mean 22.0 22.2 22.9
Sigma 2.4 2.8 3.2
Another noteworthy feature of the data is that all of it uses to prevent false alarms from other more significant
the "opening geometry" targets (negative VELR) have sources. For this reason the computed noise false alarm
significantly greater detection ranges than the "closing rate predicts very little about the false alarm performance
geometry" targets (positive VELR). This is mainly be- of the AN/APG-66 downlook mode. On the other hand,
cause the closing rate of the former targets is much the environment in which the AN/APG-66 operates is so
smaller than that of the latter targets, so there are more complex and unpredictable that it is impossible to com-
opportunities to "cumulate" the per-scan probability of pute a meaningful predicted false alarm rate that takes
detection no matter how small it is, as long as it is into account the environment. In fact, even the experi-
greater than zero. mental determination of the false alarm rate from flight
We note, finally, that the table shows those few par- data to be described in this section was exceedingly diffi-
ticular combinations of target velocity and aspect angle at cult. This was because, to begin with, it required a great
which the radar is altogether blind to the target as a result deal of "detective work" to identify the source of each
of the target's low Doppler relative to the Doppler of the target report and thereby decide whether or not it was a
main beam clutter. false alarm.
This detective work involved the comparison of var-
111. FALSE ALARM PERFORMANCE ious parameter values associated with each report with the
values expected from known phenomena. The parameter
In the literature on radar, the term "false alarm" is values were in turn derived from the data provided by the
usually applied to the effect of the crossing of a threshold AN/APG-66 instrumentation. This data included azimuth,
by noise as discussed in Marcum [8]. In the AN/APG-66 unambiguous range, signal amplitude, ambiguous range,
downlook mode, however, false alarms resulting from and Doppler for each PRF (look), and INS-derived air-
this phenomenon are virtually nonexistent because of the craft coordinates and directions, all of which were useful
adaptive CFAR, and the 3-of-8 target reporting algorithm in deducing the sources of target reports.
154 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. AES-19, NO. 1 JANUARY 1983
The reports that were counted as false alarms in the gets was plotted on a map from radar data and found to
results presented in Section IIIB were those contained in coincide with a large hangar and a large gantry structure.
the categories labeled receiver noise, sidelobe clutter dis- Perpendicular to coastline cliffs-These tests were
cretes, system instabilities, ghosts, and unknowns. This is flown at Big Sur, Calif., to evaluate the effects of large,
because such reports either were the result of no target at specular returns in the antenna main beam, which would
all, or else they give erroneous directional or range infor- aggravate any system stability problems.
mation (i.e., with a discrete). On the other hand, reports The false alarm results obtained from the flights de-
of real airborne targets, ground moving targets, and skin- scribed are shown in Table II, along with the recorded
less sidebands are considered valid reports because they data intervals and number of false alarms. The data show
give the correct range and directional information of a that the measured false alarm rate, averaged over all
real target. In the case of ground moving targets and tests, was 1.2 per minute. Further operational experience
skinless sidebands, as is seen in Section IIIC, the radar has confirmed these formal test results.
simply reports the detection of real target Doppler com-
ponents that are present in the return signal. TABLE II
False Alarm Testing Results
Number Data False
B. Flight Test False Alarm Results Test of False Interval Alarms
Alarms (Minutes) per Minute
BlipScan Runs - Edwards AFB, Cal. 165 143.9 1.15
In addition to the blip-scan runs described in the in- Panamint Valley, Cal. 7 6.0 1.17
troduction to Section II, during which false alarm data Norway Conditions - Prince Rupert, B.C. 3 11.48 0.26
were also gathered, a number of flights were made specifi- Islands - Prince Rupert, B.C. 6 9.0 0.67
Urban Area - Hill AFB, Utah 11 5.67 1.9
cally for the purpose of measuring false alarm perfor- Large Discretes - Edwards AFB, Cal. 42 17.4 Z.4
mance. These tests were designed to measure the false Coastline Cliff - Big Sur, Cal. 0 1.4 0
alarm performance under extreme conditions and are Total 234 194.85 1.2
listed below.
Flying through a valley-Three runs were made in
which the radar flew through the Panamint Valley in Cal- C. Examples of Target Reports in Each Category
ifornia at an altitude of 5000 ft. These tests were made to
observe the false alarm performance under more severe The following discussion contains some examples, ex-
clutter conditions than those found at Edwards Air Force tracted from the flight data, that contributed to the results
Base. presented in Section IIIB.
Simulated Norway conditions-Twelve minutes of
data were recorded over Prince Rupert, British Columbia, 1) Ground Moving Targets
at altitudes between 6000 and 12000 ft. The slopes, vege-
tation, temperatures, and fjords at Prince Rupert are simi- The largest of the target report categories, other than
lar to those in Norway. that of airborne targets, was the ground moving target
Islands in the sea-Another test run at Prince Rupert category. Fig. 12 illustrates a set of data of the apparent
was designed to evaluate the effects on the radar's false target "velocity" for ground movers based on their ob-
alarm performance of the abrupt land/water transitions served Doppler. These high velocities were surprising,
that occur over an area consisting of water dotted by since the radar uses a clutter notch of + 55 knots (63 mi/
small islands. h). Very few ground moving targets had been expected to
Urban area-These runs were made at Hill Air Force exceed this speed, especially in this age of 55 mi/h speed
Base near Salt Lake City, Utah, to evaluate the effects of
the very severe clutter environment of an urban area con-
taining large RCS buildings, for example, on performance
at low to medium altitudes. In addition, there were a
large number of real targets because of the close proxim-
ity of three airports. The data were obtained dunng the ap-
15 K~~~~~~~~~~~~~~~
proach and departure phases of flights at Hill Air Force Number of 10 - MBC
Occurrences
Base so that both low and medium altitude data could be -Notch-.
gathered.
Large clutter discretes-These tests were designed to 5
evaluate the radar's performance against large clutter dis-
cretes that appear in the antenna/radome sidelobes. The
radar cross section of these discretes was measured by a 0 i I I
ib i2b i. I
If
B.iI01mI~~iIi.
I r
30 40 50 eo 70 80 9o 100 1 10
clutter-mapping technique. Two structures were observed Apparent Target Velocity (Knots)
with RCS in excess of 106 m2. The location of these tar- Fig. 12. Example of histogram of low speed target velocities.
harmonics of legitimate targets rather than very high 6 3 27 Number Next to Data
- 3 21 Point = Amplitude in Quanta
speed ground moving targets or flaws in the radar. 10
8
28 - 9223 23:12:23.2
such spectra it was found that the ground movers had a -j
30 4 211
strong skin line below 55 knots which, of course, was not 32 1377
5
34 4 29
383
always occurred at twice the skin Doppler, but only for 44 20
9
one dwell time. Each point represents a visible signal at
53 23:13:53.3
54
that filter on that look, with the numbers representing Fig. 14. Dopper versus time. "tracking a truck,' tail aspect.
amplitude in quanta. The skin line is clearly visible
within the notch (along with a small inconsequential im- 2) Sidelobe Clutter Discretes
age of it due to I-Q receiver imbalance). The "2F" har-
monic is clearly visible and detectable outside of the The only false alarms resulting from detections
Doppler region of the notch even thoulgh it is much through the sidelobes that were observed were those that
weaker than the skin return. occurred when the reflection lobe illuminated discretes
whose RCS were in excess of 106 m2. Very few such dis-
cretes have been observed anywhere, so they constitute
Doppler (kHz) only a minor problem. Therefore, the single-channel side-
8 -6 -4 -2 0 2 4 6 8 lobe rejection technique handles all but the very strongest
MB Notch-el
-2 - I (-3dB) I discretes, even in the presence of a radome reflection
0
123 lumber Next to
Data Point
lob
lObe.
1100+ Amplitude
Fig. 15 is a plot of the Doppler angle of suspected
2 I
n4 81 0+
radome reflection lobe detections as a function of the azi-
E Harmonic I
71 muth of the main lobe at the time of the detection. It can
Z6
251 400+ be seen that all of the detections plotted fall in the region
08 I 84
11 300+ bounded by the azimuths of the 40 dB limits of the re-
10
I 100+ - flection lobe as a function of the azimuth of the main
12 I Image I
Skin 69 of lobe.
14 Skin Fig. 16 shows the apparent velocity of a set of reflec-
I
tion lobe discretes. Note that the Doppler is well above
Fig. 13. Example of target history witht2F harmonic. the ground moving target region.
156 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. AES-19, NO. I JANUARY 1983
(Left)
60 50
Main Beam
N
40
Elevation = -6.5'
\ "%.
30 20
..-.-
I
I-
16.
.-
201
10I
%
I-
I~
I
-
I-
II 50
Number of
Occurrences
10
VG - 300 Knots
l
(Right)
1
50
40
10
20 1
30
70
80
(Left)
RadLme
Reflection
70Lobe
Azimuth
60 (deg)
iII
10
I
I
I I
20 30 40 50 60
AA.in
main Q2..m
(deg)
L..
* *":
A,i-..*
tsamU Azlimun
I I
(Right)
Reflection
Lobe Peak
-40 dB
Limits
l
-l
5
6 -
7
8
9
n 10
E
° 12
13
14
15
16
17
IV. SUMMARY
6
-
-5
5'7
\ 1
I
0 9
-4
16 17 11 X18 12x1
z 11 - x7 X9
4 8 /11
Number Next to
~ ~ ~ we- w
Doppler Relative to MBC (kHz)
75i8
xi2
-3
X13110
121 510
-2
No Instrumentation
in This Region
x10
s 13
X1o
-1
2
0
xii
xli
8
I
%
X
Apparent
/
Skin Line
1
9 x42
~~~~42
42
\,-$~~~i14
x
1
Melvin B. Ringel (M'68) was born in Brooklyn, N.Y., on June 14, 1937. He re-
ceived the B.S. degree in mathematics from Massachusetts Institute of Technology,
Cambridge, in 1958 and the M.S. degree in mathematics from Johns Hopkins Univer-
sity, Baltimore, Md., in 1960. He also did doctoral studies in algebraic geometry at
Johns Hopkins.
From 1962 to 1968 he worked at the Carlyle Barton Laboratory (formerly the Ra-
diation Laboratory) of Johns Hopkins University. His work there included the analysis
of the vulnerability of synthetic aperture radar to ECM. He was, during the same pe-
nriod, a Lecturer in Mathematics at Johns Hopkins, Towson State College, and the Uni-
E 111_ versity of Baltimore. Since 1968 he has worked on the design and analysis of a
* - i _ number of radar systems at the Westinghouse Electric Corporation in Baltimore where
he is currently a Fellow Engineer. His work includes simulation studies of the resolu-
tion and accuracy of the E-3A radar, signal analysis studies, the design of constant
false alarm rate (CFAR) algorithms, the analysis of the clutter limited detection range
performance of the F-16 radar, and the analysis of the multipath problem for ground
based tracking radars. He holds one patent and is the author of several papers on air-
borne radar.
Mr. Ringel is a member of the Mathematical Association of America. He is also a
Go player and road runner and finished the 1978 New York Marathon only two hours
and a few minutes behind Bill Rodgers.
'4..
David H. Mooney (SM) was born in Columbia, S.C., on January 25, 1927. He re-
_ ceived the B.S.E.E. from the University of South Carolina, Columbia, in 1948 and the
"k, M.S.E.E. from the University of Pittsburgh, Pittsburgh, Pa., in 1951.
Since 1948 he has been on the technical staff of the Westinghouse Electric Corpo-
ration, with a primary field of interest in advanced airborne radar. His activities have
|_' included hardware development, systems synthesis, computer simulation, performance
analysis, and flight data analysis. He holds a number of patents in the radar field, and
is a coauthor of chapter 19 of Skolnik's Radar Handbook.
Mr. Mooney is a member of Phi Beta Kappa, Tau Beta Pi, and is a registered
Professional Engineer in the State of Maryland.
_ Ws.4i! W. Long was born in Lexington, Va., on July 2, 1940. He received the B.S.E.E.
degree from Virginia Polytechnic Institute, Blacksburg, in 1963 under the Cooperative
Engineering Program with the Virginia Electric and Power Company.
Since 1966 he has been with the Westinghouse Electric Corporation in the system's
definition and analysis area and has been involved in a number of airborne and balloon
borne radar programs. His most recent experience is with the F-16 fire control radar
system.
158 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. AES-19, NO. I JANUARY 1983