Professional Documents
Culture Documents
1
Block Diagram of Radar System
2
Range-Azimuth-Doppler Cells to Be
Thresholded
3
Noise
If you were to look at noise on
an oscilloscope, you’d see
something like
𝑛=𝐼 +𝑖𝑄
´2
𝑛 =2 𝜎
2
4
Linear Detector
The amplitude of the noise z (i.e. )
the square root of the sum of the
squares of the I and Q data) z
2
𝑧 𝑧
{
𝑝 ( 𝑧 )= 𝜎 2
exp −
2𝜎(
0,𝑧<0
2
, 𝑧 ≥0
)
p(z)
𝜋
´𝑧 =
2 √
𝜎
𝑧´2=2 𝜎 2
𝑧´𝑛=𝜎 𝑛 2𝑛 /2 Γ (1+𝑛/ 2) z
5
Square-Law-Detector
If z is instead the squared magnitude of the complex Gaussian
noise (i.e., the signal at the output of a square law detector),
the detected noise has an negative exponential PDF,
)
p(z)
1 𝑧
{
𝑝 ( 𝑧 )= 2 𝜎 2
exp −
(
2𝜎
0 𝑧 <0
2 )
, 𝑧≥0
𝑧 =2 𝜎 2
´
𝑧´2=8 𝜎 4
z
𝑧´𝑛=𝑛! (2 𝜎 2 )𝑛
6
Target Detection in Noise
8
Threshold Test is Optimum
9
Basic Target Detection Test
10
Basic Target Detection Test
p(x|H0)
Ricean distribution
11
with Different
Detection Examples
12
Probability of False Alarm,
We can evaluate the probability of false alarm from noise
alone. Writing the Rayleigh pdf in the form, we have:
then :
13
Time,
False Alarm
The average time interval between crossings of the threshold by
noise alone is the false alarm time Tfa
14
15
Probability of False Alarm Again
Now the false alarm probability Pfa is also given by the ratio of the
time that the envelope is above the threshold to the total time
since the average duration of a
noise pulse is approximately the
reciprocal of the bandwidth.
16
False alarm time
(Examples)
By substituting
1
𝑝𝑓𝑎 =
𝑇 𝑓𝑎 𝐵 𝐼𝐹
2
𝑉𝑇
if (
𝑝𝑓𝑎 =exp −
2 𝜓0 )
2
1 𝑉𝑇
𝑇 𝑓𝑎 =
𝐵 𝐼𝐹
exp ( )
2 𝜓0
17
Numerical Example
for BIF = 1 MHz and required false alarm rate of 15 minutes,
18
Detection,
Probability of
A is the amplitude of
the RF pulse signal
19
Detection,
Probability of
the probability
that the envelope
will exceed VT
20
Detection,
Probability of
The signal-to-rms noise voltage ratio is related to signal-to-noise
power by :
at IF stage
and by substituting:
22
Design example
• the customer needs that the average time between the false
Let
alarm is 15 min. If IF band is 1 MHz, then
1
Pfa 1.11 10 9
T fa B
If we need Pd = 0.5 then the required S/N=13.05 dB
If we need Pd = 0.9 then the required S/N=14.7 dB
If we need Pd = 0.99 then the required S/N=15.75 dB
23
Multiple-Pulse Effects
• The previous radar equation is for a single radar pulse
• Seldom is a radar system required to detect a target on the basis of a single
transmitted pulse.
• many pulses are usually returned from any particular target and can be used
to improve detection.
• the number of pulses nB as the antenna scans is
24
Example
• For a ground based search radar having
θB = 1.5 ˚
fP = 300 Hz
θ˙S = 30˚/s (ωm = 5 rpm)
then the number of hits from a point target in each scan nB = 15
25
Pulse integration
26
Coherent Integration
27
Non-Coherent Integration
28
Integration Improvement Factor
Signal to Noise Gain
𝑆𝑁𝑅 𝑖 (𝑛)
• defined as: 𝐼𝑖 ( 𝑛 ) = ≤𝑛
𝑆𝑁𝑅 1
- is the SNR of a single pulse
29
Integration of Pulses
30
Different Types of Non-Coherent Integration
31
Binary (M-of-N) Integration
32
Detection Statistics for Binary Integration
33
Optimum M for Binary Integration
34
Optimum M for Binary Integration
35
Detection Statistics for Different Types of
Integration
36
Signal to Noise Gain / Loss vs. # of Pulses
37
Albersheim’s Equation
• a closed-form expression relating , , and under the following
Gives
conditions:
• Nonfluctuating target in Gaussian (IID in I and Q) noise.
• Linear (not square law) detector.
• Noncoherent integration of samples.
𝑆𝑁𝑅
(𝑑𝐵)
38
Albersheim’s Equation (example)
•
Suppose and are required for a nonfluctuating target in a system
using a linear detector. If detection is to be based on a single
sample and 100 samples, what is the required of each case?
39
A Graphical Example
40
Matlab script
c=3*10^8; %light speed
PtdB=10*log10(150000); % 150 kw
f=9.4*10^9; % RF frequency
lamda=c/f; % lamda
LamdadB=20*log10(lamda); % lamda in db
thao=1.2*10^-6; %pulse width
B=1/thao; % frequency bandwidth
D=2.4; %diameter of circular antenna
GdB=10*log10(4*pi*0.6*(pi*(D/2)^2)/(lamda^2)); % antenna gain
dwell_time=18.3*10^-3; % dwell time
prf=2000; % pulse repeat frequency
n=floor(dwell_time*prf); % nbr of integrated pulses
ndB=10*log10(n); % nbr of integrated pulses in dB
LdB=3.1+2.4+3.2; % losses in dB
FdB=2.5; % noise figure
T0=10*log10(290); % Kelvin temperature in dB
K=10*log10(1.38*10^-23); % Boltzmann’s constant in dB
41
Matlab script (cont.)
% calcul the SNRdB for range [5:150]km
R=(5:105)*1000; % range from 5 to 150 km
RdB=40*log10(R); % range in dB
sigmadB=10*log10(1); % target 1
% calcule the SNRdB for target 1 for range from 5 to 150 km
SNR=PtdB+2*GdB+LamdadB+sigmadB+ndB - (30*log10(4*pi)+K+LdB+T0+10*log10(B)
+FdB+0.16*2*R/1000+40*log10(R));
plot(R/1000,SNR);
grid;
axis([5 105 -20 100])
xlabel('range (km)');
ylabel('SNR (dB)');
sigmadB=10*log10(0.1); %target 2
% calcule the SNRdB for target 2 for range from 5 to 150 km
SNR=PtdB+2*GdB+LamdadB+sigmadB+ndB - (30*log10(4*pi)+K+LdB+T0+10*log10(B)
+FdB+2*0.16*R/1000+RdB);
hold on;
plot(R/1000,SNR,'r');
plot([5 150],[15 15],'g--');
42
Graphical solution
to radar range equation
100
80
60
SNR (dB)
40
Target 1: 65 km
20
15 dB
Target 2: 50 km
0
-20
10 20 30 40 50 60 70 80 90 100
range (km)
43
Cross-Section Fluctuations
The analysis of detection has so far assumed that the echo signal from a given
target is constant. In practice, real targets are made up of several scatterers,
and the net echo depends on the way in which the contributions from these
scatterers add vectorially, and the way in which the motion of the target
(and/or radar). The echo variations are also caused by meteorological
conditions, lobe structure of the antenna, equipment instability, the variation
in target cross section and view aspect. That cause the echo varies from pulse
to pulse and scan to scan
44
Fluctuating Target Models
45
Swerling models: Case 0
46
Swerling models: Case 1
Echo pulses are assumed to be correlated (of constant
amplitude) from pulse to pulse, but uncorrelated (independent)
from scan to scan. Scan to scan fluctuations are described by an
exponential pdf:
Notional sequences of Swerling-1
1 target samples. Results from three
p ( ) exp , 0
ave ave scans with 10 pulses per scan.
47
Swerling models: Case 2
Echo pulses are uncorrelated from pulse to pulse (and therefore
also from scan to scan). The pulse to pulse fluctuations are
described by the same exponential pdf as for Swerling-1 targets.
48
Swerling models: Case 3
Swerling 1 and 2 evidently correspond to cases where the target
is composed of a large number of similar scatterers, and hence
where Gaussian statistics apply. We can also have targets where
there is one dominant scatterer plus a number of other smaller
scatterers. In such a case the pdf follows Chi-square, degree 4.
4 2
p ( ) 2 exp
ave ave
49
Swerling models: Case 4
echo pulses are uncorrelated from pulse to pulse (and therefore
also from scan to scan), and the pulse to pulse variations also
follows Chi-square, degree 4:
4 2
p ( ) 2 exp
ave ave
50
Swerling Target Models
53
Fluctuating Target Single Pulse Detection : Rayleigh
Amplitude
54
Fluctuating Target Single Pulse Detection
55
Additional Single Pulse SNR Required
As one would expect, the single
pulse SNR required to achieve a
particular Pd (for Pd > 0.4) will
be higher for a fluctuating target
than for a constant amplitude
signal.
However, for Pd < 0.4, the
system takes advantage of the
fact that a fluctuating target will
occasionally present echo
signals larger than the average,
and so the required SNR is
lower.
56
Integration improvement factor as a function of the number
of pulses integrated
57
Detection Statistics for Different Target Fluctuation Models
6 dB 14.5 dB
loss of
8.5 dB
Shnidman’s Equation
Applied to noncoherent
integration and square law
detector case
59
Shnidman Empirical Formulae for SNR
(for Steady and Swerling Targets)
60
Shnidman’s Equation
61
Neyman-Pearson Detector (review)
Note:
False alarm
consume finite
resources of
the radar: bus,
computation, …
is the estimated interference power from measured samples, it is a random variable
and so , the detector will be considered CFAR if the expected value of does not
depend on the actual value of . 64
Homogeneous Performance
It is for Swerling 1 or
Swerling 2 target
66
CA-CFAR Problems : Self Target Masking
Target returns in the reference window bias the threshold
estimate and may prevent the target in the cell under test from
being detected.
67
CA-CFAR Problems : Mutual Target Masking
68
CA-CFAR Problems : Clutter boundaries
Sharp clutter or interference power lead to an increased
number of false alarms and to masking of targets located
near the boundary.
69
Greatest-of Mean Level CFAR
• Find mean value of M/2 cells before and after test cell
separately
• Use larger noise estimate to determine threshold
70
Greatest-of Mean Level CFAR (example)
71
Censored-CFAR (CS-CFAR)
• Compute and use noise estimates as in Greatest-of, but
remove the largest N samples before computing each average
73
CS- and OS-CFAR (example)
The Order
Statistic (OS),
and Censored
CFARs mitigate
mutual target
masking while
the CA-CFAR
misses one
target
74
Summary
75