EITN90 Radar and Remote Sensing

Lecture 6: Target fluctuation models

Daniel Sjöberg

Department of Electrical and Information Technology

Spring 2020
 Outline

1 Radar cross section of simple targets

2 Radar cross section of complex targets

3 Statistical characteristics of the RCS of complex targets

4 Target fluctuation models

5 Doppler spectrum of fluctuating targets

6 Conclusions

2 / 37
 Learning outcomes of this lecture
In this lecture we will
I Learn the origin of fluctuations in target RCS
I Describe the statistics of the fluctuations
I Characterize RCS decorrelation lengths
I Understand the basic assumptions of the Swerling fluctuation
models
Target
Transmit signal Antenna

Transmitter T/R

Receive
signal

Receiver
protector
switch Mixer/Preamplifier
Detection and
Signal
Detector A/D measurement
processor
results

Low noise Local

IF amplifier
amplifier oscillator
Receiver (Adapted from Fig. 1-1)

3 / 37
 Outline

1 Radar cross section of simple targets

2 Radar cross section of complex targets

3 Statistical characteristics of the RCS of complex targets

4 Target fluctuation models

5 Doppler spectrum of fluctuating targets

6 Conclusions

4 / 37
 Sphere scattering
The scattering from a sphere can be computed analytically. This
gives very precise results, and spheres are often used as calibrating
targets at measurement sites.

5 / 37
 Sphere scattering

A sphere has the same backscattering in all directions due to

symmetry.

6 / 37
 Corner reflectors
A corner reflector has a strong backscattering in a large angular
interval. Dihedral = two walls, trihedral = three walls.

Discussion
7 / 37
 Multiple targets

E 0 e−jk·R k̂ · R1
R1
k̂ · R2
R2

R3
k̂ · R3

When several scatterers are subjected to an incident wave

E 0 e−jk·R , the backscattering is (complex addition)
 2
N
X √ −j2k·Ri 
σtot =  σi e


 
i=1

8 / 37
 Two scatterers

D
D
2 sin θ
D
2 sin θ

R
θ

 2
N
X √  √ D √ D
2
σtot = σi e−j2k·Ri  =  σ0 e−j2k 2 sin θ + σ0 e−j2k(− 2 sin θ) 
   
 
i=1
 
2 2πD sin θ
= 4σ0 cos
λ
9 / 37
Two scatterers, RCS as function of angle

R
θ

10 / 37
 Outline

1 Radar cross section of simple targets

2 Radar cross section of complex targets

3 Statistical characteristics of the RCS of complex targets

4 Target fluctuation models

5 Doppler spectrum of fluctuating targets

6 Conclusions

11 / 37
 Randomly distributed targets
Put out a number of randomly distributed targets.

With Ri a random variable, the RCS

 2
N
X √ −j2k·Ri 
σ= σi e


 
i=1
also becomes a random variable, with some probability density
function as angle or frequency varies. However, once the scatterers
are fixed the model (including σ) is completely deterministic.
12 / 37
 Randomly distributed targets
The RCS as function of angle appears as a random variable, with
some probability density.

13 / 37
 Outline

1 Radar cross section of simple targets

2 Radar cross section of complex targets

3 Statistical characteristics of the RCS of complex targets

4 Target fluctuation models

5 Doppler spectrum of fluctuating targets

6 Conclusions

14 / 37
 Equal scatterers
A large number of approximately equal scatterers with uniformly
distributed phase have exponentially distributed RCS:
(
1
exp − σσ̄
 
σ̄ σ≥0
p(σ) =
0 σ<0
√
where σ̄ is the mean. The amplitude voltage (ς = σ) is Rayleigh
distributed:
( h 2i
2ς
exp − ςσ̄ ς≥0
p(ς) = σ̄
0 ς<0
For any RCS distribution pσ (σ), the corresponding amplitude
distribution pς (ς) is given by pς (ς) = 2ςpσ (ς 2 ). This is seen from
Z Z
σ→ς 2
1 = pσ (σ) dσ = pσ (ς 2 )2ς dς
| {z }
=pς (ς)

where dσ = d(ς 2 ) = 2ς dς.

15 / 37
 One dominating scatterer
When the situation has one dominating and many small scatterers,
the RCS can be modeled by a fourth-degree chi-square distribution
(easier to handle analytically than the true Rician distribution):
(
4σ
 2σ 
σ̄ 2 exp − σ̄ σ≥0
p(σ) =
0 σ<0

16 / 37
17 / 37
Graphical comparison of the distributions

Discussion

18 / 37
Matlab demo (see http://radarsp.com)

19 / 37
 RCS decorrelation
Decorrelation occurs when the observation of RCS is significantly
changed by an alteration of time, frequency, or angle. To estimate
when this occurs, consider the situation below.

 2  2
M M
 X √ −j2k·Ri   X 
−j2ki∆x sin θ 
σ= σi e  = σ0  e
 

   
i=−M i=−M
sin2 [(2M + 1)k∆x sin θ]
= σ0
sin2 (k∆x sin θ) 20 / 37
 RCS decorrelation
Introduce the total length L = (2M + 1)∆x of the collection of
scatterers:
sin2 (kL sin θ)
σ = σ0 2
sin (k∆x sin θ)
The first zero corresponds to kL sin θ = π, implying the correlation
lengths (using k = 2π/λ = 2πf /c and sin θ ≈ θ for small θ)
c
∆f = (frequency decorrelation)
2L sin θ
λ
∆θ = (angle decorrelation)
2L
L sin θ is the projected length of the target in boresight.

The book derives these values differently (from an autocorrelation

function), but the results are the same. However, different
definitions of decorrelation may apply in different applications.
21 / 37
 RCS correlation properties, width of target

Autocorrelation in angle for a rectangular target 5 m × 10 m. Note

that a wider target has shorter correlation length.

22 / 37
 RCS correlation properties, frequency agility

Variation in RCS due to frequency agility for 5 m × 10 m target.

Frequency agility can be used to decorrelate contributions from

clutter and atmospheric effects.
Discussion

23 / 37
 Outline

1 Radar cross section of simple targets

2 Radar cross section of complex targets

3 Statistical characteristics of the RCS of complex targets

4 Target fluctuation models

5 Doppler spectrum of fluctuating targets

6 Conclusions

24 / 37
 Detection framework
A radar with azimuth beamwidth θ3 and pulse repetition frequency
PRF is rotating at angular speed Ω rad/s. This provides
θ3
N= PRF main beam samples per 360◦ sweep
Ω

25 / 37
 Swerling models
Extreme cases for decorrelation. The non-fluctuating case is
sometimes listed as the Swerling 0 or Swerling 5 case.

I The exponential distribution models many equal scatterers.

I Chi-square degree 4 models one dominant scatterer
surrounded by many small.
I Scan-to-scan decorrelation may occur for long scan times.
I Pulse-to-pulse decorrelation may occur for frequency agile
systems.
www.radartutorial.eu Discussion

26 / 37
Different decorrelations, scan or pulse

27 / 37
 Example: choice of model
Consider the following scenario:
I Target: 10 m long, complex aircraft.
I Frequency: 10 GHz.
I Range: 30 km.
I Cross-range velocity: 200 m/s.
Choice of model:
I The complex nature of the aircraft with no clear dominating
scatterer suggests an exponential PDF.
I Decorrelation angle: ∆θ = 2L λ
= 1.5 mrad = 0.86◦ . This
corresponds to a flight time ∆t = R∆θ/v = 225 ms.
I If dwell time Td < 225 ms we expect coherence during
pulse-to-pulse and choose Swerling 1.
I If dwell time Td > 225 ms the signal is decorrelated within the
PRI, and we choose Swerling 2.
28 / 37
 Outline

1 Radar cross section of simple targets

2 Radar cross section of complex targets

3 Statistical characteristics of the RCS of complex targets

4 Target fluctuation models

5 Doppler spectrum of fluctuating targets

6 Conclusions

29 / 37
 Doppler shift
A target traveling at radial velocity v cos ψ towards the radar
causes a frequency shift f → f + fd . After demodulation (removal
of carrier frequency f ), the baseband signal will ideally be
y[m] = Aej2πfd mT
where fd is easily detected by a Fourier transform of the data y[m].
However, amplitude fluctuations may change this (A → A[m]).

30 / 37
 Doppler spectrum of fluctuating targets

More on Doppler processing in next lecture! 31 / 37

 Outline

1 Radar cross section of simple targets

2 Radar cross section of complex targets

3 Statistical characteristics of the RCS of complex targets

4 Target fluctuation models

5 Doppler spectrum of fluctuating targets

6 Conclusions

32 / 37
 Conclusions

I Interference causes fluctuations in RCS with viewing angle

and frequency, which may be modeled as random.
I Typical probability densities have been reviewed.
I An interactive matlab program is available for investigating
effects of number of scatterers, frequency, range etc.
I Decorrelations in angle and frequency have been quantified.
I The bounding cases of Swerling models have been presented.

33 / 37
 Discussion

Why do you think there are different sizes of the trihedrals in the
picture?

Go back

34 / 37
 Discussion

Why do you think there are different sizes of the trihedrals in the
picture?

Answer: What matters is the size of the scatterer in terms of

wavelengths. The different sizes target different wavelengths.
Go back

34 / 37
 Discussion
What is the implication of a long tail in the probability distribution?

Go back

35 / 37
 Discussion
What is the implication of a long tail in the probability distribution?

Answer: with a long tail (slow decrease), there is a higher

probability for extreme events.
Go back

35 / 37
 Discussion

With the frequency response below, what approximate mean (dB)

RCS would a frequency agile system detect, switching between the
different frequency bins? 0 dB, 10 dB, 20 dB, 30 dB, or 40 dB?

Go back

36 / 37
 Discussion

With the frequency response below, what approximate mean (dB)

RCS would a frequency agile system detect, switching between the
different frequency bins? 0 dB, 10 dB, 20 dB, 30 dB, or 40 dB?

Answer: Around 20 dB (more exactly 18 dB).

Go back

36 / 37
 Discussion

With a decorrelation time of ∆T , pulse repetition interval PRI,

and dwell time Td = np PRI, what is the condition for a Swerling 1
situation (scan-to-scan decorrelation), and a Swerling 2 situation
(pulse-to-pulse decorrelation)?

Go back

37 / 37
 Discussion

With a decorrelation time of ∆T , pulse repetition interval PRI,

and dwell time Td = np PRI, what is the condition for a Swerling 1
situation (scan-to-scan decorrelation), and a Swerling 2 situation
(pulse-to-pulse decorrelation)?

Answer:
I Swerling 1: Td > ∆T (or N Td > ∆T , with N being the
number of scan positions for a search radar)
I Swerling 2: PRI > ∆T
Go back

37 / 37