Beruflich Dokumente
Kultur Dokumente
Lecture 6
Detection of Signals in Noise
Transmitter
Propagation
Medium Power Waveform
Amplifier Generation
Target
T/R
Radar
Switch
Cross
Section
Antenna Signal Processor Computer
Parameter
Tracking Thresholding Detection
Estimation
Data
Recording
Photo Image This Lecture
Courtesy of US Air Force
Used with permission.
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
2
IEEE AES Society
Outline
• Basic concepts
– Probabilities of detection and false alarm
– Signal-to-noise ratio
• Integration of pulses
• Fluctuating targets
• Summary
Transmits
Pulses at
~ 1250 Hz
Rotation
Rate
12. rpm
0
Resolution
-25
1/16 nmi
-50 Rotation
Rate
0 50 100 12.7 rpm
Radial Velocity (kts)
Radar
As Antenna Rotates Transmits
~22 pulses / Beamwidth Pulses at
Az ~ 1250 Hz
Range - Azimuth - Doppler Cells Beamwidth
~1.2 °
~1000 Range cells
~500 Azimuth cells
~8-10 Doppler cells
5,000,000 Range-Az-Doppler Cells Is There a Target Present
to be threshold every 4.7 sec. in Each Cell?
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
5
IEEE AES Society
Target Detection in Noise
Received Backscatter Power (dB)
40
Detected
30 False Strong
Alarm Target
Threshold
20
Weak target (not detected)
Noise
10
0
0 25 50 100 125 150 175
Range (nmi.)
• Received background noise fluctuates randomly up and down
• The target echo also fluctuates…. Both are random variables!
• To decide if a target is present, at a given range, we need to set a
threshold (constant or variable)
• Detection performance (Probability of Detection) depends of the
strength of the target relative to that of the noise and the threshold
setting
– Signal-To Noise Ratio and Probability of False Alarm
Measurement Decision
Radar Detection
Receiver x Processing
H0 or H1
For each measurement
Probability
There are two possibilities: Measurement
Density
Target absent hypothesis, H 0
Noise only x=n p( x H 0 )
Target present hypothesis, H 1
Signal plus noise
x = a+n p( x H 1 )
Decision
For each measurement H0 H1
There are four decisions:
Don’t False
H0
Report Alarm
Truth
Missed
H1 Detection
Detection
Decision
Probability of Detection: The probability we choose
H0 H1 H1 when H1 is true
Don’t False PD
H0 Report Alarm
Probability of False Alarm: The probability we choose
Truth H1 when H 0 is true
Missed
H1 Detection Detection PFA
Objective:
Maximize PD subject to PFA no greater than specified
Neyman-Pearson
criterion (PFA ≤ α )
L( x ) =
p( x H 1 )
> η
p( x H 0 ) <H
0
0.7
Noise Only
Noise Probability Density Target Absent
0.6
Detection
Probability Density
0.5
Threshold
0.4 Probability
Probability ofof False
False Alarm
Alarm (( P )
FA )
PFA
P
PFA = Prob{ threshold exceeded given target absent }
FA = Prob{ threshold exceeded given target absent }
i.e.
i.e. the
the chance
chance that
that noise
noise is
is called
called aa (false)
(false) target
target
0.3
We
We want
want P to be very, very low!
FA to be very, very low!
PFA
0.2
0.1
0
0 2 4 6 8
Voltage
Courtesy of MIT Lincoln Laboratory
Used with permission
0.7
Noise
Probability
Probability of
of Detection
Detection (( P
PDD ))
Probability Density
0.6 PDD == Prob{
P Prob{ threshold
threshold exceeded
exceeded given
given target
target present
present }}
Detection I.e.
I.e. the
the chance
chance that
that target
target is
is correctly
correctly detected
detected
Probability Density
0.5
Threshold We
We want PDD to
want P to be
be near
near 11 (perfect)!
(perfect)!
0.4
0.3 Signal-Plus-Noise
Probability Density
0.2
p( x H 1 )
0.1
Signal + Noise
Target Present
0
0 2 4 6 8
Probability Voltage
Probability of
of Courtesy of MIT Lincoln Laboratory
Used with permission
False
False Alarm (( P
Alarm )
FA )
PFA
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
10
IEEE AES Society
Detection Examples with Different SNR
0.7
Noise Only Courtesy of MIT Lincoln Laboratory
0.6 Detection Threshold Used with permission
PFA = 0.01
0.5
Probability Density
0.2
0.1
0
0 5 10 15
Voltage
0.7
Noise pdf
Probability
0.6
0.5 Threshold
0.4 Signal-Plus-Noise pdf
0.3
0.2
0.1
00 2 4 6 8
Rayleigh Rician Voltage
⎛ r2 ⎞
p ( r H o ) = r exp ⎜⎜ − ⎟⎟
⎝ 2⎠
⎛ r2 + R ⎞
p ( r H 1 ) = r exp ⎜⎜ −
2 ⎠
(
⎟⎟ I o r R ) SNR =
R
2
⎝
Set threshold rT based
rT = − 2 log e PFA
Courtesy of MIT Lincoln Laboratory
on desired false-alarm probability Used with permission
( )
∞
P = p ( r H 1 ) dr = Q
Compute detection probability for
2 ( SNR ), − 2 log e PFA
given SNR and false-alarm probability D ∫
rT
∞
⎛
where Q ( a, b ) = r exp ⎜ −
2
+ 2
⎞
⎟⎟ I o ( a r ) dr
r a
∫b ⎜
⎝ 2 ⎠
Is Marcum’s Q-Function
(and I0(x) is a modified Bessel function)
Remember
PFA=10-4 This!
Probability of Detection (PD)
0.8
SNR = 13.2 dB
PFA=10-6 needed for
0.6 PD = 0.9 and
PFA = 10-6
Steady Target
0.4
PFA=10-8
0.2 PFA=10-10
PFA=10-12
0.0
4 6 8 10 12 14 16 18
0.999 PFA=10-4
Probability of Detection (PD)
0.99 PFA=10-6
0.98
0.95
0.9
Remember
0.8 This!
0.7
PFA=10-8 SNR = 13.2 dB
0.5 needed for
PFA=10-10 PD = 0.9 and
0.3 PFA = 10-6
0.2
PFA=10-12 Steady Target
0.1
0.05
4 6 8 10 12 14 16 18 20
Signal-to-Noise Ratio (dB)
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
14
IEEE AES Society
Tree of Detection Issues
Detection
Non-Coherent Coherent
Fluctuating Non-Fluctuating Integration Integration
Fluctuating Non-Fluctuating
Determine PDF at
Detector Output
• Single pulse
• Fixed S/N
Scan to Scan Fluctuations Pulse to Pulse Fluctuations
(Swerling Case I and III) (Swerling Case II and IV)
• Basic concepts
• Integration of pulses
• Fluctuating targets
• Summary
2
2 n
n
n =1
…
n =1 Calculate
xN x 2N
N 2 xN
1
Target Detection
Declared if
N
∑x
n =1
n >T Target Detection
Declared if
1 N 2
∑ xn > T
N n =1
• Adds ‘voltages’ , then square
• Phase is preserved • Adds ‘powers’ not voltages
• pulse-to-pulse phase coherence required • Phase neither preserved nor required
• SNR Improvement = 10 log10 N • Easier to implement, not as efficient
Gain
PFA=10-6
0.6
Non-Coherent
Integration
Gain
0.4 One Pulse
Ten Pulses
Non-Coherent
Integration
0.2
0 5 10 15
Signal to Noise Ratio per Pulse (dB)
Pulse 1
Calculate Threshold i1
x1 x12 T
Calculate
Pulse 2
Calculate Threshold
i2 Sum
x2 2 N 2nd Threshold
m = ∑ in
x 2
T M
…
n =1
Pulse N
Calculate Threshold iN
xN T
x 2N
Individual pulse detectors: 2nd thresholding:
2
x n ≥ T, i n = 1 m ≥ M , target present
2
x n < T, i n = 0 m < M , target absent
At Least N
pk ( 1 − p) PC = 1 − ( 1 − p )
N! At Least
∑
N −k
PM / N =
N
M of N
k ! (N − k ) !
1 of N
Detections k =M
3/4
Out of 4
0.90 4/4 Pulses
0.50
0.10 Steady
(Non-Fluctuating )
Target
PFA=10-6
0.01
2 4 6 8 10 12 14
Signal to Noise Ratio per Pulse (dB)
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
22
IEEE AES Society
Optimum M for Binary Integration
100
Steady
(Non-Fluctuating )
Target
Optimum M
PD=0.95
10
PFA=10-6
1
1 10 100
Number of Pulses
0.95
0.90 Non-Coherent
Integration
4 Pulses
0.50
0.95
0.90 Non-Coherent
Integration
4 Pulses
0.50
0.95
0.90 Non-Coherent
Integration
4 Pulses
0.50
5
Relative
Binary N1/2 2
To Coherent
Integration
0 0
1 10 100 1 10 100
Number of Pulses Number of Pulses
Dependent
50
Sampling σv
Region Velocity
Spread
00
20
=3
Of
N
Clutter
0
10
10
50
Independent
30
Sampling 1
5
fr =
20
Region
10
T
2 3 Pulse
Repetition
1 Rate
0.001 0.01 0.1 1
σ v / λ fr
• Non-coherent Integration Can Be Very Inefficient
in Correlated Clutter
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
29
IEEE AES Society
Effect of Pulse to Pulse Correlation on
Non-Coherent Integration Gain
ρ (T) .99 0.9 0.1 0.02
Equivalent Number of Independent Pulses 100
Dependent
50
Sampling σv
Region Velocity
Spread
00
20
=3
Of
N
Clutter
0
10
10
50
Independent
30
Sampling 1
5
fr =
20
Region
10
T
2 3 Pulse
Repetition
1 Rate
0.001 0.01 0.1 1
σ v / λ fr
•
Non-coherent Integration Can Be Very Inefficient
Adapted from nathanson, Reference 8 in Correlated Clutter
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
30
IEEE AES Society
Albersheim Empirical Formula for SNR
(Steady Target - Good Method for Approximate Calculations)
• Basic concepts
• Integration of pulses
• Fluctuating targets
• Summary
Fluctuation Rate
Nature of RCS
Scattering Model Slow Fluctuation Fast Fluctuation
“Scan-to-Scan” “Pulse-to-Pulse”
Similar amplitudes
Exponential
(Chi-Squared DOF=2)
Swerling I Swerling II
⎛ σ⎞
p(σ ) = exp⎜ − ⎟
1
σ ⎝ σ⎠
Fluctuation Rate
Nature of Amplitude
Scattering Model Slow Fluctuation Fast Fluctuation
“Scan-to-Scan” “Pulse-to-Pulse”
Similar amplitudes Rayleigh
⎛ a2 ⎞ Swerling I Swerling II
p(a ) =
2a
exp⎜⎜ − ⎟⎟
σ ⎝ σ⎠
10 Constant
0
20
RCS (dBsm) 15
Swerling I/II
High
10 Fluctuation
5
0
20
Swerling III/IV 15
Medium
10
Fluctuation
5
0
0 20 40 60 80 100
Sample # Courtesy of MIT Lincoln Laboratory
Used with permission
H 1 : x = ae jφ + n Rayleigh
p(a ) =
2a ⎛ a2 ⎞
exp⎜⎜ − ⎟⎟
amplitude
H0 : x = n model
σ ⎝ σ⎠
z= x >T H1
Detection Test
x <T
Average σ
H0 SNR per ξ= 2
pulse σN
Probability of
False Alarm ⎛ T2 ⎞
(same as for
PFA = exp⎜⎜ − 2 ⎟⎟
⎝ σN ⎠
non-fluctuating)
⎛ 1 ⎞
⎜⎜ ⎟⎟
Probability of
PD = ∫ p ( z > T H 1 , a) p(a) da dz PD = P FA ⎝
1 + ξ ⎠
Detection Test
⎛ T2 ⎛ 1 ⎞⎞
PD = exp⎜⎜ − 2 ⎜⎜ ⎟⎟ ⎟⎟
⎝ σN ⎝ 1+ ξ ⎠⎠
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
38
IEEE AES Society
Fluctuating Target Single Pulse Detection
1.0
Probability of Detection (PD)
Fluctuation Loss
0.8
0.6
Non-Fluctuating
0.4
Swerling Case 3,4
PFA=10-6
0.0
0 5 10 15 20 25
Signal-to-Noise Ratio (dB)
For high detection probabilities, more signal-to-noise is required for
fluctuating targets.
The fluctuation loss depends on the target fluctuations, probability of
detection, and probability of false alarm.
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
39
IEEE AES Society
Fluctuating Target Multiple Pulse Detection
Probability of Detection (PD) 1.0
0.8
Steady Target
Coherent Integration
0.6 Swerling 2 Fluctuations
Non-Coherent Integration, Frequency
Diversity
Swerling 1 Fluctuations
0.4 Coherent Integration, Single Frequency
0.2
PFA=10-6
N=4 Pulses
0.0
0 4 8 12 16 20
Signal-to-Noise Ratio per Pulse (dB)
• In some fluctuating target cases, non-coherent integration with
frequency diversity (pulse to pulse) can outperform coherent
integration
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
40
IEEE AES Society
Detection Statistics for Different Target
Fluctuation Models
1.0
10 pulses
Non-coherently
Probability of Detection (PD)
Integrated
0.8 PFA=10-8
Steady Target
Swerling Case I
0.6
Swerling Case II
Swerling Case III
Swerling Case IV
0.4
0.2
0.0
-2 0 2 4 6 8 10 12 14
Signal-to-Noise Ratio (SNR) (dB) Adapted from Richards, Reference 7
K= N, Swerling Case 2
2, Swerling Case 3
2N Swerling Case 4
0 N ≤ 40
α= 1
N > 40
4
⎛ N ⎛ 1 ⎞ ⎞⎟
⎜
X ∞ = η⎜ η + 2 + ⎜ α − ⎟⎟
⎝ 2 ⎝ 4 ⎠⎠
C1 = (((17.7006 PD − 18.4496 ) PD + 14.5339 ) PD − 3.525 ) / K
1 ⎛⎜ 27.31 PD − 25.14 ) ⎛ ⎛ 10 − 5 ⎞ (2 N − 20 ) ⎞ ⎞⎟
C2 = e + (PD − 0.8) ⎜⎜ 0.7 ln⎜⎜ ⎟⎟ + ⎟
⎟⎟
K⎝ ⎜ ⎝ PFA ⎠ 80
⎝ ⎠⎠
C1 + C 2 0.872 ≤ PD ≤ 0.99
0.0
-0.5
-1.0
0.0 0.2 0.4 0.6 0.8 1.0
Adapted from Shnidman in Richards, Reference 7 Probability of Detection
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
44
IEEE AES Society
Outline
• Basic concepts
• Integration of pulses
• Fluctuating targets
• Summary
Power
Rain Backscatter Data
40
S-Band Data
Power
20 Rain
Receiver
Cloud
Noise
0
0 1 2
Courtesy of NOAA Range (nmi)
“Guard”
Cells Cell to be
Thresholded
Range cells Î
Data Cells Used
• Estimate background (noise, etc.) from data to Determine
Mean Level of
– Use range, or range and Doppler filter data Background
– Set threshold as constant times the mean value of background
1 N
• Mean Background Estimate = ∑ xn
N n =1
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
47
IEEE AES Society
Effect of Rain on CFAR Thresholding
Radar Backscatter (Linear Units) Mean Level Threshold CFAR
Receiver Noise
Receiver Noise
Range Cells
C Band
5500 MHz
9 dB
Rain Cloud
2.2 dB
Receiver Noise
Receiver Noise
• Find mean value of N/2 cells before and after test cell
separately
• Use larger noise estimate to determine threshold
Window Slides Through Data
PFA=10-6
0.9999
Matched Filter
1 Pulse
0.999 Mean Level CFAR – 10 Samples
Mean Level CFAR – 50 Samples
Probability of Detection (PD)
0.90
0.50
CFAR
Loss
0.10
0.01
4 6 8 10 12 14 16 18
Signal-to-Noise Ratio per Pulse (dB)
4
CFAR Loss (dB)
10-6
10-4
2
0
0 10 20 30 40 50
Number of Reference Cells Adapted from Richards, Reference 7
The greater the number of reference cells in the CFAR, the better is
the estimate of clutter or noise and the less will be the loss in
detectability. (Signal to Noise Ratio)
IEEE New Hampshire Section
Radar Systems Course
Detection 11/1/2010
53
IEEE AES Society
CFAR Loss vs Number of Reference Cells
5
4
PFA For Single Pulse Detection
10-4 Approximation
3 10-6
10-8 CFAR Loss (dB) =
2 - (5/N) log PFA
CFAR Loss, dB
N=1
Dotted Curve PFA = 10-4
Dashed Curve PFA = 10-6
1 N=3 Solid Curve PFA = 10-8
0.7 N = 15 to 20 (typically)
0.5 N=10 Since a finite number of
0.4 cells are used, the estimate
N=30
0.3 of the clutter or noise is
N=100 not precise.
0.2
2 5 7 10 20 50 70 100
Number of Reference Cells, N
Adapted from Skolnik, Reference 1