Wavelet Analysis of Low Observable Targets Within Sea Clutter

1
Wavelet Analysis of Low Observable Targets Within
Sea Clutter

• G Davidson,H D Griffiths
• University College London

2
Overview

• Aim is to detect low observable, slow moving
  targets (debris, ships, others) in heavy sea
  clutter, within clutter Doppler spectrum.
• 3GHz and 9GHz real aperture, experimental
  multifunction radars at metre resolution
  (QinetiQ).
• Either the sea surface or the observing platform
  is moving and, this non-stationary velocity
  spectrum can mask targets of unknown varying
  velocity.
• Strong clutter backscatter complicates CFAR
  detection. Doppler spectrum is analysed by a
  Wavelet transform to minimise the uncertainty in
  both time and velocity.
• Clutter returns appear to consist of discrete
  scatterers with a characteristic lifetime.
  Thresholding this scatterer lifetime reveals a
  real target in real clutter that was difficult to
  detect in Intensity and Doppler.
• This is not the correlation time of the surface.

3
Medium Sea Surface
4.3 metres Significant Wave Height

• Whitecaps suggest different scattering event

4
Heavy Sea Surface
6.1 metres Significant Wave Height

• Heavy seas cause shadowing

5
Recorded dB Backscatter
6
Non-Stationary Doppler
- Velocity
Time
Dsitribution Measure
- Velocity
Proportion Above Noise
- Velocity
7
Event Based Processing

• Some justification for considering the
  backscatter as discrete events these are more
  obvious in Doppler.
• Over sufficient time (30s), these events average
  out to a stable Doppler spectrum.
• At shorter time scales (1s) identifying
  individual events may be useful for target
  detection.
• But neither the velocity or the lifetime of the
  scatterers can be known a priori. Windowed
  Fourier is not well suited to this.
• Wavelet Transform is useful for this case as it
  maintains constant uncertainty in time-frequency.
• WT gives optimally smoothed Doppler Velocity-Time
  image of the sea surface.

8
Fourier vs Wavelet Transform
Frequency
Time
Time

• Fourier
• Arbitrary time window
• Window determines frequency
• Problems at boundaries

• Wavelet
• Forced constant ???t
• Can choose frequency subset
• Convolution over all time

9
Wavelet Filter Bank

• Filter width proportional to frequency, log
  spaced filter bank

10
Simulated Target (No Noise)
FT1
FT3
Time
Time
Velocity Measure
FT2
WT

• For stable, well defined signals FFT is best,
  but
• At low PRF (1 second of 256 samples) the optimum
  window size for the Fourier transform (F1, F2,
  F3) is unknown.
• The Wavelet transform (WT) gives an acceptable
  view.

11
Target Parameters
Velocity
Time
Intensity

• Target simulated from observed parameters
  (Swerling 1-2)

12
Wavelet Maxima
Full Doppler via FFT
Power
Velocity
Doppler via WT
Velocity
Time

• Single FFT Doppler is misleading, but WT maxima
  useful

13
Real Clutter/Simulated Target
Real Clutter 0dB Sim. Target (within clutter
spectrum)
Time
Real Clutter
Time

• Doppler is misleading, but WT maxima useful

14
Detection Scheme

• Instantaneous WT-Doppler spectrum is smooth
  (red).
• Dominant event can be isolated without any
  thresholding
• Length of dominant event (arrow) related to
  individual scatterer lifetime
• Threshold this to reveal target

15
Lifetime Distribution
20 minutes of data 3GHz, VV 156Hz prf Grazing
angle
Real Data 0dB Sim.Target
Log10 Complementary Cumulative Distribution
Real Data
Scatterer Lifetime

• Exponential distributed scatterer lifetime,
  agrees with Doppler spectral lineshape models
  Lee et al.

16
Doppler Real ClutterTarget
9GHz, VV 6m Range Resolution 500Hz
prf Significant wave ht. 2.4m 8ms-1 wind 1.5?
angle (grazing)
dB Intensity
17
Intensity Real ClutterTarget
Arbitrary Intensity
18
WT Real ClutterTarget
Lifetime
19
Conclusions

• The sea surface backscatter can be considered as
  a collection of individual scattering events.
• Event velocity and lifetime unknown so wavelet
  analysis is easier than FFT Doppler especially
  for fading targets with velocity varying over 1
  second.
• WT minimises uncertainty in time and velocity.
• Dominant scatterer lifetime can be easily
  measured, distribution is exponential in
  agreement with models.
• Thresholding the lifetime of scattering events
  suggests targets can be detected within the
  Doppler spectrum. Real data and real target gave
  encouraging results.
• Obviously, FFT/MTI is better for fast moving
  targets of relatively constant velocity outside
  clutter spectrum. The lifetime of these is
  determined by the range cell size.
• Not measuring correlation length this averages
  all scatterers together and requires sampling
  window to be chosen.