15

Predicted Performance of
Phased-Array Radars

15.1 INTRODUCTION
The performance evaluation of phased-array radars is addressed in this chapter.
Topics covered include:

• Functional performance:
– Target detection:
Clear environments
Clutter environments
Jamming environments
– Tracking:
Ballistic missile targets
Air targets
– Interference suppression performance:
Sidelobe canceller performance
Open-loop nulling performance
Adaptive array performance
Doppler processing performance
Time and frequency domain excision performance

301
302 Chapter 15

– Clutter cancellation performance:

Land clutter
Sea clutter
Rain and other weather clutter
– Hardware subsystems:
Range sidelobes
Channel-to-channel alignment
Amplitude and phase errors
Phase noise evaluation
Wideband operation versus subarray size.

Two basic types of performance prediction are discussed: (1) functional perfor-
mance, that is, how well a specified radar performs its functions (e.g., search,
track) and (2) mission-level performance, that is, how well the radar performs its
mission (e.g., air defense, missile defense, early warning).
The following sections address the specific aspects of functional and mission-
level performance prediction.

15.2 FUNCTIONAL PERFORMANCE

15.2.1 Target Detection
Ultimately, predicting target detection performance entails calculating the prob-
ability of detection ( P D ) for a specified target, range, radar cross section (RCS),
and waveform for a defined probability of false alarm ( P FA ). The detailed ap-
proach to accomplishing this objective will vary with the particular type of tar-
get model, detection rule, waveform parameters, and signal-to-noise ratio
(SNR). The theoretical bases for this performance prediction are provided in
Chapter 2 and, for example, [12].
Consider a ± 45-degree azimuth horizon search fence for ballistic missile ac-
quisition that covers from 500 km out to 1,500 km. The missile RCS follows a
Swerling I fluctuation model, exhibits a mean RCS of 0 dBsm, and develops
signal-to-noise ratios of 10, 15, and 20 dB depending on radar sizing. If the
search is designed to provide four statistically independent looks at a missile
while in the fence, then the cumulative probability of detection can be calcu-
lated. An alternative design that must operate in narrowband sidelobe noise
 Predicted Performance of Phased-Array Radars 303

jamming uses frequency hopping to avoid the jammer. Assume the frequency
hops are sufficiently diverse in frequency to decorrelate the target from look to
look. This has the result of transforming the Swerling I fluctuation model into a
Swerling II model. The predicted performance of the two alternative designs is
to be calculated. Table 15.1 displays the results.
Table 15.1 Horizon Search Fence with Swerling I and II Targets

Swerling I Swerling II

Single-Pulse Cumulative Probability Single-Pulse Cumulative Probability

Probability of of Detection for Probability of of Detection for
SNR (dB) Detection 4 Independent Looks Detection 4 Independent Looks

10 0.285 0.738 0.091 0.318

15 0.655 0.986 0.401 0.871

20 0.872 1.000 0.735 0.995

Comparing the cumulative probabilities of detection for the two Swerling

models indicates that if a cumulative probability of at least 0.96 is required, then
with no jamming (i.e., Swerling I) an SNR of 15 dB provides margin, whereas for
the jamming case with hopping (i.e., Swerling II), closer to 20 dB SNR is re-
quired. One interpretation of these results is that the jammer degrades search
performance by nearly 5 dB (or alternatively, an approximate 5 dB larger radar is
required for equivalent performance). A similar analysis could be performed for
non-coherent integration, with the results compared to those for binary integra-
tion case above.
Table 15.2 considers an alternative design with sidelobe cancellers that pro-
vide an average of 15 dB of jammer cancellation that achieves signal-to-
interference ratios (SIRs) of 9, 13.5, and 18 dB. As can be seen, by incorporating
sidelobe cancellers into the radar system design, a radar that in the clear pro-
duces 15 dB SNR, will achieve 13.5 dB SIR and meets the cumulative probability
of detection requirement of 0.96 (with no margin).
Now consider the same radar operating in a severe clutter environment such
that the signal-to-clutter ratios (SCRs) are –23 dB, –18 dB, and –13 dB, respec-
tively. If a three-pulse moving target indicator (MTI) canceller provides clutter
mitigation adequate to produce 8.5 dB, 13 dB, and 17.5 dB, respectively, then Ta-
ble 15.3 contains the corresponding performance predictions.
304 Chapter 15

Table 15.2 Horizon Search Fence with Swerling I Targets and Sidelobe Cancellers

Cumulative Probability of Detection

SIR (dB) Single-Pulse Probability of Detection for 4 Independent Looks

9 0.213 0.617

13.5 0.554 0.960

18 0.806 0.999

Table 15.3 Horizon Search Fence with Swerling I Targets in a Clutter Environment with MTI
Cancellation

Cumulative Probability
Input of Detection SCR after MTI Cumulative Probability of Detection
SCR (dB) for 4 Independent Looks Cancellation (dB) for 4 Independent Looks

–23 0.181 8.5 0.550

–18 0.517 13.0 0.946

–13 0.786 17.5 0.998

As can be seen from the table, the radar that in the clear produces 15 dB SNR,
produces 13 dB SCR after MTI cancellation and achieves a cumulative probabil-
ity of detection of 0.946. Although not quite equal to the performance in the
clear, this predicted performance indicates very little degradation and would
probably be considered adequate in most situations.

15.2.2 Tracking
Tracking performance predictions usually focus on track smoothing and predic-
tion accuracies. Two target types will be considered: (i) aircraft with constant ve-
locity, and (ii) intact tactical ballistic missile (TBM) after rocket burnout. Assume
the radar achieves 15 dB SNR on both targets and has a 3 dB beamwidth of 2 de-
grees. The aircraft is tracked at 5 Hz and the missile is tracked at 2 Hz. A time-in-
track of 10 seconds is achieved on the aircraft prior to a maneuver, and a time-in-
track of 30 seconds is achieved for the missile target. Both targets are 250 km
from the radar. The track smoothing estimation error is approximately given by:
 Predicted Performance of Phased-Array Radars 305

R θ3
σP = , (15.1)
2 SNR fr Tt

where R, θ 3, SNR, f r, and T t are the target average slant range, antenna 3 dB
beamwidth, signal-to-noise ratio, tracking rate, and time-in-track, respectively.
In addition, the predicted position error is approximately:

R θ3 TP 12 R θ3
σ P − Predicted = + . (15.2)
2 SNR fr Tt Tt 2 SNR fr Tt

The position is to be predicted 10 seconds ahead (TP). Table 15.4 illustrates the
predicted performance for this tracking problem.

Table 15.4 Smoothed and Predicted Position Errors for Aircraft

and Missile

One-Sigma Smoothed One-Sigma Predicted

Target types Position Error (m) Position Error (m)

Aircraft 157 693

TBM 143 305

It should be noted that the longer time-in-track for the TBM (i.e., 30 seconds
versus 10 seconds) outweighs the higher track rate for the aircraft (5 Hz for the
aircraft, while the TBM’s trajectory, if outside the atmosphere, is effectively bal-
listic acting only under gravity, versus 2 Hz for the TBM) for equal SNR. This is
to be expected. In reality performance could favor TBM tracking even more
since aircraft inevitably maneuver, which would further limit the smoothing
time-in-track for the aircraft. However, a larger SNR would probably result
from the aircraft as compared with the TBM.

15.2.3 Interference Suppression

Prediction of the interference mitigation performance of phased-arrays radar
can be performed at a first-pass, or “back-of-the-envelope,” by using approxi-
mate formulas, as well as by using high-fidelity simulations.
306 Chapter 15

Sidelobe canceller performance is addressed in Chapter 8. The residual jam-

mer power after an ideal sidelobe canceller is, from Chapter 8 and [8], given
by:

 G 
for PJ >> Pn and ρ = 1 Pmin ≈ Pn  1 + m  (15.3)
 Ga 

where P J, P n, ρ, G m, and G a are the jammer and thermal noise power, correlation
of the jammer in the main and auxiliary channels, and the main antenna sidelobe
and auxiliary antenna gains, respectively. Table 15.5 evaluates equation (15.9) for
several antenna gain values when the SNR in the clear is 20 dB to calculate out-
put signal-to-interference ratio.

Table 15.5 Output SIR (dB) Using an Ideal

Sidelobe Canceller

Gm/Ga (dB) Output SIR (dB)

3 15.22

0 16.99

–10 19.59

–20 19.96

As can be seen, good performance is achieved when the auxiliary antenna

gain is at least 10 dB higher than the main antenna sidelobes. However, even
with the sidelobe canceller, the radar’s performance is still degraded by 3 dB for
auxiliary to sidelobe gain of unity. For a Swerling I target and a probability of
false alarm of 10–6, this results in a significant loss in target detectability: proba-
bility of detection of 0.87 in the clear versus 0.76 for operating in jamming with a
low-gain auxiliary antenna when using a sidelobe canceller.
Open-loop nulling can be employed to cancel interference by estimating the
optimal weight vector using a number of interference samples. Consider an ex-
ample where the ideal mitigation using the optimal weight is –35 dB cancella-
tion of the interference. The open-loop nulling performance for the case of a 32-
element array is estimated versus number of samples used to compute the
weights with the approximate results appear in Table 15.6. The calculation uses
 Predicted Performance of Phased-Array Radars 307

Table 15.6 Open-Loop Nulling Performance vs. Number of

Samples used to Compute the Weights for an Ideal –35 dB
Cancellation

Number of Samples Open-Loop Cancellation (dB)

32 –22.82

48 –30.65

64 –32.18

80 –32.90

96 –33.32

128 –33.81

160 –34.07

equation (15.4) for the approximate loss relative to the ideal cancellation, where
N and M are the number of samples used and number of antenna elements, re-
spectively, from Chapter 8 and [8] given by:

N + 2 − M 
Loss (dB) = 10 log10  . (15.4)
 N +1 

As can be seen in the table, open-loop cancellation is significantly degraded

when the number of samples equals the number of elements (i.e., 12 dB), versus
80 samples (i.e., 2 dB) or 160 samples (i.e., 1 dB). The trade-off is the time re-
quired to collect the interference samples, during which the interference param-
eters could change resulting in a loss in cancellation performance.
Similar performance can be achieved for open-loop Doppler nulling of clutter
backscatter using a limited number of clutter samples. Closed-loop interference
or clutter mitigation can often achieve superior cancellation performance com-
pared to open-loop nulling, at the cost of higher computational requirements.
Again, estimating the adaptive weights using a limited number of interference
or clutter samples will suffer a similar loss in performance as predicted by equa-
tion (15.4).
308 Chapter 15

Another type of interference suppression is the use of time or frequency-

domain censoring in the signal processor. This can be extremely effective when
interference (or clutter backscatter) is of short duration or is narrowband in na-
ture. Equivalently, for clutter suppression, this corresponds to discrete clutter re-
turns like those arising from man-made structures. The performance is limited
by the number of interference-dominated samples versus the total number of
signal-plus-interference samples. Table 15.7 illustrates the predicted loss in per-
formance versus the ratio of censored samples to total samples (e.g., range cells,
frequency bins, or Doppler filters). As seen in the table, up to approximately 10%
of samples can be censored with an acceptable loss when compared with the
degradation of uncensored interference sources. In some cases, accepting a 1 or
2 dB loss would also be acceptable.

Table 15.7 Loss in SNR due to Time or Frequency

Censoring as a Function of the Ratio of Censored
Samples to Total Samples

Ratio of Censored Number Loss in SNR Due

of Samples to Total Number to Censoring (dB)

0.05 –0.22

0.10 –0.46

0.20 –0.96

0.30 –1.55

0.40 –2.22

0.50 –3.01

It should be noted that the cancellation performance discussed in this section

assumes no amplitude or phase errors, nor channel-to-channel mismatch errors,
that exist due to hardware subsystem designs. These effects are addressed in
Section 2.6.

15.2.4 Clutter Cancellation

Clutter mitigation is usually performed via the signal processing techniques de-
scribed in Chapter 2. The two types of mitigation are MTI waveforms and pro-
 Predicted Performance of Phased-Array Radars 309

cessing and pulse-Doppler waveforms and processing. The performances of

both types of clutter cancellation are addressed in this section.
MTI processing is effective for zero Doppler-shifted clutter (or at most moving
very slowly relative to targets of interest). This technique is a digital high-pass
filter that has zero gain at zero Doppler and all multiples of the pulse repetition
frequency (PRF), as depicted in Figure 2.7 of Chapter 2. For these conditions,
good performance can be attained as approximately given for two and three-
pulse cancellers by the relationships:

−1 2
 
(
 2 πσ T ) 
2
I2 = CA ≈ =  PRF σ  (15.5)
 f
  f 19.75

and

4
 PRF 
I3 ≈   780 , (15.6)
 σf 

where σ f , T, and PRF are the one-sigma clutter spectral width, the pulse repeti-
tion interval (PRI), and pulse repetition frequency, respectively. Table 15.8 illus-
trates the typical theoretical performance of the 2 and 3-pulse MTI cancellers.
Table 15.8 Predicted MTI Clutter Cancellation Performance

Ratio of PRF to 2-Pulse MTI 3-Pulse MTI

clutter spectral width cancellation (dB) cancellation (dB)

0.05 –38.97 –80.96

0.10 –32.96 –68.92

0.25 –25.00 –53.00

0.40 –20.91 –44.84

As seen in Table 15.8 canceller performance degrades with increasing clutter

spectral width. However, for relatively narrow clutter spectra the two-pulse MTI
canceller provides good performance. The three-pulse MTI can mitigate wider-
spectrum clutter. Note that these cancellation values will be degraded for non-
zero velocity clutter and other real-world effects such as timing jitter or non-
stationary clutter statistics.
310 Chapter 15

Pulse-Doppler waveforms and processing provide superior performance to

MTI, especially for nonzero velocity clutter, at the cost of more radar resources
(e.g., 16 or 32 coherent pulses) and signal processing (i.e., both pulse matched fil-
tering and processing for N Doppler filters, where N is the number of coherent
pulses in the train).
For clutter spectra that are narrower than the Doppler filter bandwidth, near-
perfect clutter cancellation can result. However, for broad clutter spectra, such as
for blowing rain, more than one filter might have significant clutter. Elimination
of all Doppler filters with clutter will eventually reduce the target Doppler cov-
erage and will require a larger number of PRFs (and coherent batches of pulses)
to regain adequate target velocity coverage. Therefore to minimize this effect, fil-
ters with some clutter are typically used for detection purposes, and will result
in less clutter cancellation.

15.2.5 Hardware Subsystems

This section addresses some common hardware-limited performance issues.

[Link] Range (Time) Sidelobes. Hardware errors affect achievable range sidelobe
levels. Key contributors to this phenomenon are the signal paths in the antenna
and receiver analog hardware. Approximate range sidelobe level is limited to no
better than the root-sum-squared errors from the antenna, receiver, and sidelobe
taper via:

SLL (dB) = 10 log10 2

Errorant + Errorrec
2
+ Taper 2 . (15.7)

Table 15.9 illustrates the achievable range sidelobes for reasonable antenna and
receiver error magnitudes for range sidelobes of –30 dB and –40 dB relative to
the peak response at the matched filter output. As can be seen, about 5 dB mar-
gin exists for the desired –30 dB sidelobes, and about 3.4 dB margin is available
for the –40 dB desired sidelobe case. If the receiver and antenna errors rose to
–42 dB, and a –45 dB taper was again selected, then the –40 dB sidelobe level
would be met with no margin.
 Predicted Performance of Phased-Array Radars 311

Table 15.9 Predicted Range Sidelobe Performance

Error and Taper Levels (dB) SLL for –30 dB SLL for –40 dB

Receiver: –50

Antenna: –45 –34.97 –43.39

Taper: SSLdesired – 5 dB

[Link] Channel-to-Channel Alignment. Predicted interference suppression perfor-

mance will be degraded when channel-to-channel mismatch arises. Chapter 11
addresses this and provides a formula for the achievable interference cancella-
tion as a function of channel-to-channel alignment errors:

 4
( )
2
CR =   ε p − 1 . (15.8)
 3

Consider the adaptive array that provides –35 dB cancellation of a sidelobe noise
jammer. Table 15.10 shows the effects of various degrees of channel mismatch.

Table 15.10 Degradation of Cancellation as a

Function of Channel Mismatch

Channel Mismatch Cancellation

Error (dB) Performance (dB)

0.02 –35.0

0.05 –35.0

0.10 –35.0

0.20 –31.4

0.30 –27.8

0.40 –25.3

As seen from Table 15.10, when more than 0.1 dB of mismatch error exists for
this example, cancellation performance is degraded.
312 Chapter 15

[Link] Amplitude and Phase Errors. Similar to the effect of the channel matching
evaluated errors in Section 2.6.2, amplitude and phase errors will also degrade
interference mitigation performance. In Chapter 11, the maximum cancellation
versus sinusoidal amplitude error in the passband is given by:

Amplitude Error (dB) = 20 log10  (1+ δ 2 ) / (1− δ 2 )  , (15.9)

where δ is the peak amplitude of the sinusoid. As can be seen from Table 15.11,
as for channel mismatch errors, for peak sinusoidal errors of greater than 0.1 dB,
cancellation performance is degraded.

Table 15.11 Degradation of Cancellation as a

Function of Amplitude Errors

Peak Sinusoidal Achievable

Mismatch Error (dB) Cancellation Ratio (dB)

0.02 –35.0

0.05 –35.0

0.10 –35.0

0.20 –32.7

0.30 –29.1

0.40 –25.3

For phase errors, a similar relationship relating to cancellation performance

exists. In order to achieve the –35 dB cancellation ratio requires that phase errors
be on the order of 5 degrees rms.

[Link] Phase Noise. Phase noise is the primary limitation to achievable clutter
cancellation as discussed in Chapter 11. The calculation method used here fol-
lows the procedure outlined in [19]. This method is graphical in nature and ef-
fectively adjusts the receiver phase noise frequency response by first multiplying
it by the system frequency response function (defined like a Bode plot in control
systems design) and then integrating over frequency to obtain the resulting
phase noise. The system frequency response is defined by the range to clutter,
the type of MTI canceller used, and the waveform’s pulse bandwidth.
 Predicted Performance of Phased-Array Radars 313

First, to account for clutter cancellation occurring due to correlation between

the transmit and receive (assuming a common local oscillator source), the break
point frequency is computed as:

c
fc = , (15.10)
2 π Rc

where c and Rc are the speed of light and mean slant range to the clutter, respec-
tively. Consider the following example. For a mean range of 100 km, the clutter
break frequency is approximately 500 Hz. The frequency response slope for this
“leg” of the plot is 20 dB/decade.
Next, for the three-pulse MTI canceller, the approximate break frequency is
computed as f2 = 0.249 (PRFave), or about 119 Hz. The frequency response slope
for this portion of the plot is 40 dB/decade. Last, the intermediate frequency fil-
ter response is assumed to have a ± 1.25 MHz 3 dB bandwidth (i.e., correspond-
ing to a 2.5 MHz pulse bandwidth) and is assumed to roll off at –40 dB/decade
above 1.25 MHz.
A nominal receiver phase noise frequency response is assumed with its floor
at a level of –140 dBc/Hz above the break frequency of 1.75 kHz. A roll-off of –30
dB/decade is assumed below the break frequency. To account for the phase
noise spectrum being two-sided, and noise being present on the oscillator for
both transmit and receive, the assumed response is increased by 6 dB. The calcu-
lated system frequency response and the adjusted phase noise frequency re-
sponse are plotted in Figure 15.1.
As indicated on Figure 15.1, the integrated phase noise is –62.6 dBc (i.e., dB
relative to the carrier). Therefore, –62.6 dB is the predicted maximum clutter can-
cellation ratio. In general, 5 to 10 dB of margin would be allocated in the design,
which would result in a clutter cancellation capability of approximately –55 dB,
quite adequate for many applications. A similar analysis approach can be used
for pulse-Doppler waveforms and processing.

[Link] Wideband Operation versus Subarray Size. Due to the frequency-steering

phenomenon that exists for phased-arrays operated over very wide bandwidths,
a loss referred to as dispersion loss results from the antenna main lobe scanning
across a target versus instantaneous waveform frequency. For linear frequency
314 Chapter 15

Figure 15.1 System Adjusted Phase Noise Response

modulation waveforms, the antenna mainlobe linearly scans with increasing or

decreasing frequency (i.e., depending on whether an “up-chirp” or “down-
chirp” waveform is employed by the radar) across the target. Hence, a “scallop-
ing” effect incurs an average loss in target return amplitude.
The problem is to evaluate how small a subarray needs to be to scan a 2,000
MHz waveform out to 60 degrees for a unity figure of merit (FoM). The relation-
ship between the figure of merit, subarray width, maximum electronic scan an-
gle, and pulse length is given by:

( subarray width ) sin (θ max scan ) B

FoM = , (15.11)
c
 Predicted Performance of Phased-Array Radars 315

where B and c are the waveform bandwidth and the speed of light, respectively.
Note that when the FoM is approximately unity, about 1 dB of dispersion loss oc-
curs at maximum electronic scan. Solving equation (15.17) for the subarray
width yields:

FoM c
subarray width =
( )
. (15.12)
B sin θ max scan

For the above values, the required subarray width is 0.173 meter.
Therefore if a phased-array uses an antenna aperture of 10 meters2, then to op-
erate with a 2 GHz LFM waveform at a maximum electronic scan of 60 degrees
and to incur only about 1 dB of dispersion loss, then the number of square sub-
arrays required would be approximately:

10
N subarrays = = 334 .
( 0.173) 2
For less than 1 dB dispersion loss, a larger number of subarrays would be re-
quired.

15.3 REFERENCES
[1] J. V. Candy, Signal Processing—The Modern Approach, McGraw-Hill,
[2] S. Haykin & A. Steinhardt, Adaptive Radar Detection and Estimation, Wiley, 1992
[3] S. Haykin, Adaptive Radar Signal Processing, Wiley-Interscience, 2006
[4] S. Kay, Modern Spectral Estimation: Theory and Application, Prentice-Hall, 1999
[5] D. Manolakis, Statistical and Adaptive Signal Processing, Artech House, 2005
[6] S. L. Marple, Digital Spectral Analysis with Applications, Prentice-Hall, 1987
[7] R. A. Monzingo & T. M. Miller, Introduction to Adaptive Arrays, SciTech, 2003
[8] R. Nitzberg, Radar Signal Processing and Adaptive Systems, 2nd Edition, Artech House,
1999
[9] A. Oppenheim & R. Shafer, Digital Signal Processing, Prentice-Hall, 1975
[10] A. Papoulis, Probability, Random Variables, and Stochastic Processes, McGraw-Hill,
1965
[11] A. Papoulis, Signal Analysis, McGraw-Hill, 1977
[12] H. Van Trees, Detection, Estimation and Modulation Theory, Part 1, Wiley-Interscience,
2001
[13] Y. Bar-Shalom, Multitarget-Multisensor Tracking: Principles and Techniques, YBS, 1995
316 Chapter 15

[14] Y. Bar-Shalom, Multitarget/Multisensor Tracking: Applications and Advances, Artech

House, 2000
[15] S. Blackman & R. Popoli, Design and Analysis of Modern Tracking Systems, Artech
House, 1999
[16] R. Duda, et al., Pattern Classification, 2nd Edition, Wiley-Interscience, 2000
[17] K. Fukunaga, Introduction to Statistical Pattern Recognition, 2nd Edition, Academic
Press, 1990
[18] S. Theodoridis & K. Koutroumbas, Pattern Recognition, 2nd Edition, Academic Press,
2003
[19] M. Skolnik, Radar Handbook, 2nd Edition, McGraw-Hill, 1990