EITN90 Radar and Remote Sensing

Lecture 12: An overview of radar imaging

Daniel Sjöberg

Department of Electrical and Information Technology

Spring 2020
 Learning outcomes of this lecture
In this lecture we will
I See how images can be formed from radar signals collected at
different locations.
I Learn what determines the achievable resolution of the image.
I See the basics of a few imaging algorithms.
I Observe typical phenomena in radar images.
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)

2 / 58
 Outline

1 General imaging considerations

2 Resolution relationships and sampling requirements

3 Data collection

4 Image formation

5 Image phenomenology

6 Conclusions

3 / 58
 Outline

1 General imaging considerations

2 Resolution relationships and sampling requirements

3 Data collection

4 Image formation

5 Image phenomenology

6 Conclusions

4 / 58
 Radar vs optical photo

Some benefits of radar (cost of providing its own illumination, long

measurement time, computationally demanding):
I Weather tolerant.
I Emphasizes man-made objects.
I Provides fine resolution at large range.
I Can penetrate foliage and other obstacles.
I Requires coherent operation.
I Polarization may provide additional information.
5 / 58
 Synthetic Aperture Radar (SAR)

SAR can be thought of in at least three different aspects:

I SAR as a large synthetic antenna aperture.
I SAR as range-Doppler imaging.
I SAR as a signal processing exercise.

6 / 58
 SAR as a large synthetic antenna aperture

The cross-range resolution can be seen as the effective beam width

of an antenna with
λ
∆CR ≈ Rθ3 ≈ R
2DSAR 7 / 58
 SAR as range-Doppler imaging

The platform is moving in a certain direction relative to the scene,

causing a Doppler shift fd = 2v cos θcone /λ in the received signal.
λ∆fd λ λ
∆CR = R∆θ = R =R =R
2v sin θ 2vTd sin θ 2DSAR sin θ
In the book, the broadside angle θb = θ + 90◦ is used, resulting in
sin θ → cos θb . Discussion 8 / 58
 SAR as a signal processing exercise

The image is the result of the relation between three data spaces:
I Object space, defined by the scatterers.
I Data space, the raw radar measurements.
I Image space, the final synthetic image.
9 / 58
 Down-range and cross-range coordinates

I The down-range coordinate corresponds to fast time (delay for

individual pulses).
I The cross-range coordinate corresponds to slow time (phase
difference between pulses).
10 / 58
 Outline

1 General imaging considerations

2 Resolution relationships and sampling requirements

3 Data collection

4 Image formation

5 Image phenomenology

6 Conclusions

11 / 58
 Down-range and cross-range resolutions

The resolutions in down-range and cross-range can be written

c c c Rλc
∆R = ∆CR = = =
2Br 2Bx 2fc θint 2DSAR

fc = carrier frequency.
12 / 58
Different waveforms

13 / 58
 Illumination options for data collection

Both the flightpath and the direction of the antenna can affect the
total illumination of the scene.

14 / 58
 Integration angle

In the stripmap approach, θint is limited by the real antenna beam.

Spotlight scenarios can greatly improve the integration angle.

The linear path and the circular path give very different signal
histories, giving rise to different algorithms.

15 / 58
 Cross-range sampling requirement
When using sampling distance d, the phase change between two
samples is  
2δR d sin(θ3 /2)
∆ψ = 2π = 4π
λ λ
To keep the successive phase changes |∆ψ| < π requires
λ λ ∆CR
d< ≈ ≈
4 sin(θ3 /2) 2θ3 2
For wideband pulses, use the minimum wavelength λmin .

16 / 58
 Outline

1 General imaging considerations

2 Resolution relationships and sampling requirements

3 Data collection

4 Image formation

5 Image phenomenology

6 Conclusions

17 / 58
 Radar vs camera

A radar sensor is typically side-looking instead of overhead-looking.

Discussion

18 / 58
 Radar vs camera

A radar sensor collects data pulse-by-pulse instead of

2D-image-by-2D-image. Note the radar provides its own
illumination.

19 / 58
 Stop-and-hop model

A stream of I/Q data is recorded (phase is important). The radar

is moving continuously, but each signal is considered to be
recorded in a stationary frame.

20 / 58
 Cross-range phase changes

A target off the centerline causes phase changes in cross-range,

corresponding to the Doppler shift. The target may still be in one
single range bin.
21 / 58
 Outline

1 General imaging considerations

2 Resolution relationships and sampling requirements

3 Data collection

4 Image formation

5 Image phenomenology

6 Conclusions

22 / 58
 Stripmap coordinates

I Platform moves linearly, parameterized by position u.

I Complex reflectivity g(x, r).
I Complex data d(u, r).
I Complex image f (x, r).
23 / 58
 Slant range

The slant range R(u; x, r) is given by

p 2p
R(u; x, r) = (u − x)2 + r2 ⇒ t(u; x, r) = (u − x)2 + r2
c
The translational invariance from the dependence on u − x enables
us to consider the generic response at x = 0.
24 / 58
 Point spread response (PSR)

The point spread response t(u; x, r) = 2c (u − x)2 + r2 describes

p

a hyperbola in the data domain d(u, t), with a vertex depending on

(x, r).

25 / 58
 Dependence of PSR on down-range

The PSR is not a simple translation when considered as a function

of down-range r:
2p 2 2p 2
t(u; r+∆r) = u + (r + ∆r)2 = u + r2 + 2r∆r + (∆r)2
c c
2
6= t(u; r) + ∆r
c
More bowing of the PSR for scatterers with close range.
26 / 58
 Variations of PSR in cross-range and down-range

The shape invariance in cross-range (along-track) can be used for

image formation.

27 / 58
 Range-Doppler data

Range bins are formed as functions of fast time, and Doppler bins
are formed as functions of slow time (phase shifts between pulses)
due to Doppler shift
2v
fd = sin θ
λ 28 / 58
 Short dwell time compared to range

At long range (r  u, x), the PSR splits into two terms:

r
2p 2 2
2r (u − x)2
t(u; x, r) = (u − x) + r = 1+
c c r2
2 2 ux x2
   
2r 1 (u − x) 2 u
≈ 1+ = r+ − +
c 2 r2 c 2r r 2r
2
   2 
2 x 2 u ux
= r+ + − = t0 (x, r) + ∆t(u; x, r)
c 2r c 2r r 29 / 58
 Cross-range position as spatial frequency
√
Assuming the u2 term can be ignored (requires DSAR < rλc )
4π u2 ux
 
4π x
∆ψ = 2πfc ∆t(u; x, r) = − ≈ −u
λc 2r r λc r
shows that the phase change is linear as a function of platform
position u. The spatial frequency is proportional to x:
∂ 4π x
ku = ∆ψ = −
∂u λc r

30 / 58
 Doppler Beam Sharpening (DBS) image formation

Fourier transform in platform position (slow time)

Z ∞
D(ku , t) = d(u, t)e−jku u du
−∞

Coordinate mappings:
√
2R 2 x2 + r 2 2
t= = ≈ r
c c c
4π x 4π x
ku = − ≈−
λc r λc r0
Rescale coordinates to find final image:
   
λc r0 c
f (x, r) = D ku − → x, t →r
4π 2

31 / 58
 DBS example 1

I Carrier frequency = 9.6 GHz.

I RF bandwidth = 3 MHz.
I Integration angle = 0.02◦ .
I Resolution = 50 m (both dimensions).
I Number of along-track sample points (pulses) = 80.
I Range to scene center = 50 km.
32 / 58
DBS example 1: raw data, absolute value

33 / 58
 DBS example 1: raw data, real part

Phase shifts due to different cross-range positions of the targets.

34 / 58
 DBS example 1: processed data

The crosses correspond to two-dimensional sinc functions, typical

from Fourier transformations of rectangular pulses. Can be
suppressed by windowing, at the cost of reduced resolution.
35 / 58
 DBS example 2: challenging the algorithm

I Carrier frequency = 3 GHz.

I Integration angle = 51◦ .
I Resolution = 0.2 m (both dimensions).
I Number of along-track sample points (pulses) = 275.
I Range to scene center = 30 m.
Discussion

36 / 58
 DBS example 2: raw data, real part

Significant range migration and higher order phase modulation.

37 / 58
 DBS example 2: processed data

Standard DBS processing does not produce a useful image.

38 / 58
 Matched filter imaging
More sophisticated imaging methods are necessary when the
assumptions of the DBS are not valid. A reference signal dREF is
defined by the PSR
2p
t(u; x, r) = (u − x)2 + r2 ⇒ dREF (u, t; x, r) = δD (t−t(u; x, r))
c
The image can then be computed from the matched filter approach
Z ∞Z ∞
f (x, r) = d(u, t)d∗REF (u, t; x, r) du dt
−∞ −∞

which reduces to the (time-domain) back-projection algorithm

Z ∞
f (x, r) = d(u, t(u; x, r)) du
−∞

This usually entails resampling and interpolating the data to fit the
discretized image coordinates (x, r). Computationally intensive,
but less geometry dependent than DBS.
39 / 58
 Example 2: matched filter result

The constraints of the DBS assumptions are relaxed.

40 / 58
Image formation survey

41 / 58
 Outline

1 General imaging considerations

2 Resolution relationships and sampling requirements

3 Data collection

4 Image formation

5 Image phenomenology

6 Conclusions

42 / 58
No return areas (NRA): shadows

43 / 58
Paved areas, clutter, man-made objects

44 / 58
 Lakes

Still water has specular scattering, producing very little return.

45 / 58
 Shadow, fore-shortening

Shadow shapes provide information on target shapes. The

increased slope of the terrain makes a stronger return than flat
ground, causing a combination of bright and dark spots.
Fore-shortening is the range compression of the illuminated hill.
46 / 58
 Layover

Steep angles cause tall objects to appear earlier in the image than
the expected ground return. Some effects can be studied at
http://www.nrcan.gc.ca/node/9325. Discussion

47 / 58
 Speckle

Constructive and destructive interference from many scatterers

produce a speckled pattern, varying from pixel to pixel.

48 / 58
Speckle reduction by multilook and averaging

49 / 58
 Man-made returns

Man-made objects tend to have significant structure and finite

extent, as compared to lawns, fields, forests etc. Typically metal
structures, smooth surfaces, and right-angle corners, can cause
strong reflections.
50 / 58
 Man-made returns

Strong returns from the dihedral formed by

ground and flat side. Several minor features
in the vehicle may cause retroreflection.

51 / 58
 Signal-to-noise and clutter-to-noise ratios
The radar range equation can be rewritten in terms of parameters
relevant to SAR imaging.

Signal-to-noise ratio
Pavg Gt Gr λ3 σ
SNR =
2v sin θcone (4π)3 R3 N0 Ls ∆CR
Clutter-to-noise ratio (σc = σ 0 Ac = σ 0 ∆CR∆R/ cos δ)
Pavg Gt Gr λ3 σ 0 ∆R
CNR =
2v sin θcone cos δ(4π)3 R3 N0 Ls
Noise-equivalent backscatter coefficient (σ 0 for which CNR = 1)
2v sin θcone cos δ(4π)3 R3 N0 Ls
σn =
Pavg Gt Gr λ3 ∆R
This limits the possible contrast between different terrain types.
52 / 58
 Outline

1 General imaging considerations

2 Resolution relationships and sampling requirements

3 Data collection

4 Image formation

5 Image phenomenology

6 Conclusions

53 / 58
 Conclusions

I Radar images can be formed using synthetic aperture.

I Down-range resolution from pulse bandwidth, typically using
pulse compression of chirped pulses.
I Cross-range resolution from along-track sampling.
I DBS algorithm suitable for long ranges compared to DSAR ,
uses Fourier transform in cross-range.
I Matched filter or backprojection is more computationally
demanding, but relaxes geometry constraints.
I Radar images tend to emphasize returns from man-made
objects.

54 / 58
 Discussion

With Doppler resolution ∆fd = 1/Td and fd = 2v cos θ/λ, how

would you compute the angle resolution ∆θ?

Go back

55 / 58
 Discussion

With Doppler resolution ∆fd = 1/Td and fd = 2v cos θ/λ, how

would you compute the angle resolution ∆θ?
   
 ∂θ   1  λ 1 λ
Answer: ∆θ ≈  ∂f d
 ∆f d =  ∂fd /∂θ  ∆fd = 2v sin θ Td = 2DSAR sin θ ,
where DSAR = vTd .
Go back

55 / 58
 Discussion

A radar sensor is typically side-looking instead of overhead-looking.

Why?

Go back

56 / 58
 Discussion

A radar sensor is typically side-looking instead of overhead-looking.

Why?

Answer: In the overhead-looking scenario, most of the down-range

data end up in the same range bins. Looking sideways, the
down-range is more efficiently used.
Go back

56 / 58
 Discussion

I Carrier frequency = 3 GHz.

I Integration angle = 51◦ .
I Resolution = 0.2 m (both dimensions).
I Number of along-track sample points (pulses) = 275.
I Range to scene center = 30 m.
What can be seen as challenging in this scenario?

Go back 57 / 58
 Discussion

I Carrier frequency = 3 GHz.

I Integration angle = 51◦ .
I Resolution = 0.2 m (both dimensions).
I Number of along-track sample points (pulses) = 275.
I Range to scene center = 30 m.
What can be seen as challenging in this scenario?

Answer: Large scene compared to range, large RF bandwidth, wide

integration angles, fine resolution. Breaking the assumptions of
DBS. Go back 57 / 58
 Discussion

Fore-shortening Layover

Which effect tends to make the object larger than it is?

Go back

58 / 58
 Discussion

Fore-shortening Layover

Which effect tends to make the object larger than it is?

Answer: The layover effect appears as reflections from the ground

in front of the object, whereas the fore-shortening makes the peak
of the object shift towards the radar but within the object
footprint. If there is a shadow, it may cover both part of the object
footprint and some distance behind it.
Go back

58 / 58