EITN90 Radar and Remote Sensing

Lecture 7: Doppler phenomenology and data

acquisition
Daniel Sjöberg

Department of Electrical and Information Technology

Spring 2020
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

2 / 56
 Learning outcomes of this lecture
In this lecture we will
I See how relative motion induces the Doppler frequency shift
I Introduce the Fourier transform to describe signals
I Study the spectrum of pulsed radar signals
I Understand I/Q channels for data acquisition
I See examples of range-Doppler spectra
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 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

4 / 56
 Doppler shift

2vr 2v
fd = f= cos ψ
c λ
Only the radial motion (towards/from the radar) matters.

5 / 56
 Doppler shift, examples

2vr 2v
fd = f= cos ψ
c λ
Since most speeds v are very small compared to speed of light c,
the Doppler shift is small compared to the carrier frequency.

6 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

7 / 56
 The Fourier transform
The Fourier transform is the archetypical method to consider a
time domain function in frequency domain:
Z ∞
X(f ) = x(t)e−j2πf t dt
Z−∞
∞
x(t) = X(f )ej2πf t df
−∞

Often, the transform is instead expressed in terms of angular

frequency ω = 2πf :
Z ∞
X̂(ω) = x(t)e−jωt dt
−∞
Z ∞
1
x(t) = X̂(ω)ejωt dω
2π −∞
Discussion

8 / 56
Some explicit Fourier transforms

x(t) X(f )

1 δ(f )

ej2πf0 t δ(f − f0 )

1
cos(2πf0 t) 2 δ(f − f0 ) + 12 δ(f + f0 )
sin(πf τ )
rect(t/τ ) τ = τ sinc(πf τ )
πf τ
2 /2 √ 2
e−(t/τ ) τ 2πe−(2πf τ ) /2

9 / 56
 Some properties of the Fourier transform
I Linearity:
x(t) = ax1 (t) + bx2 (t) ⇔ X(f ) = aX1 (f ) + bX2 (f )
I Time shifting:

x(t) = y(t − t0 ) ⇔ X(f ) = e−j2πf t0 Y (f )

I Frequency shifting:

x(t) = ej2πf0 t y(t) ⇔ X(f ) = Y (f − f0 )

I Scaling (a > 0 is a real number):
1
x(t) = y(at) ⇔ X(f ) = Y (f /a)
a
R
I Convolution vs product ([x1 ∗ x2 ](t) = x1 (t − τ )x2 (τ ) dτ ):
x(t) = [x1 ∗ x2 ](t) ⇔ X(f ) = X1 (f )X2 (f )
x(t) = x1 (t)x2 (t) ⇔ X(f ) = [X1 ∗ X2 ](f )
10 / 56
 Discrete Fourier transform and FFT

The Discrete Fourier Transform (DFT) is a discretization of the

continuous transform in time and frequency:

N
X −1
Xk = xn e−j2πkn/N
n=0
N −1
1 X
xn = Xk ej2πkn/N
N
k=0

The Fast Fourier Transform (FFT) is any implementation of the

DFT that can be considered “fast”.

The most known is the Cooley-Tukey radix-2 algorithm, requiring

the number of samples to be N = 2r for some integer r. If this is
not the case, zero-padding can be applied with little penalty.

11 / 56
 The effect of zero-padding a DFT

−1
Assume a sampled signal x = {xn }N
n=0 is augmented by a number
of zeros,
(
−1 xn n = 0, 1, . . . , N − 1
y = {yn }M
n=0 , yn =
0 n = N, N + 1, . . . , M − 1 > N

The corresponding DFT is then

M
X −1 N
X −1
Yk = yn e−j2πkn/M = xn e−j2πkn/M
n=0 n=0

which can be seen as an interpolation in frequency since the step

length 1/M is smaller than the original 1/N .

12 / 56
 The effect of zero-padding a DFT

1.0
0.8
0.6
0.4
0.2
0.0
3 2 1 0 1 2 3
2.0 t

1.5
1.0 N=8
N=16
0.5 N=32
N=64
0.00.6 0.4 0.2 0.0 0.2 0.4 0.6
f

Original time-domain function (top graph) sampled at 8 points.

Augmented with zeros to 16, 32, and 64 points, makes the DFT
interpolate between the original points.
13 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

14 / 56
 Pulsed radar signals

We intend to find out the spectrum of a pulsed radar signal:

x(t)

The strategy is to express the signal as a modulated carrier:

x(t) = ej2πf t p(t)

where ej2πf t is the carrier wave, and p(t) is the modulation

(change in amplitude).

15 / 56
 Infinite length continuous wave

A
2 δ(f + f0 )

A j2πf0 t
x(t) = A cos(2πf0 t) = (e + e−j2πf0 t )
2
A A
X(f ) = δ(f − f0 ) + δ(f + f0 )
2 2
16 / 56
 Single rectangular pulse

(
A, −τ /2 < t < τ /2
pτ (t) =
0, otherwise
Z τ /2
sin(πf τ )
Pτ (f ) = Ae−j2πf t dt = Aτ = Aτ sinc(πf τ )
−τ /2 πf τ
Note the contradictory definition of sinc(z) in the book’s (8.15). 17 / 56
 Infinite pulse train

∞ ∞
" #
X X
pI (t) = pτ (t − nT ) = pτ (·) ∗ δ(· − nT ) (t)
n=−∞ n=−∞
 
∞
1 X 
PI (f ) = {Aτ sinc(πf τ )} δ(f − k · PRF )
| {z } T | {z } 
k=−∞ =1/T
=Pτ (f )
18 / 56
 Finite pulse train

(
1, −Td /2 < t < Td /2
pF (t) = pI (t) · pTd (t), pTd (t) =
0, otherwise
∞
ATd τ X
PF (f ) = [PI (·) ∗ PTd (·)](f ) = sinc(πτ k · PRF) sinc[π(f − k · PRF)Td ]
T
k=−∞

19 / 56
 End result: modulated finite pulse train

The carrier wave cos(2πf0 t) shifts the spectrum to ±f0 .

20 / 56
 Frequency scales

Four frequency scales:

I Bandwidth of spectral lines (1/Td )
I Spacing of spectral lines (1/T )
I Rayleigh bandwidth of single pulse envelopes (1/τ )
I Center frequencies (±f0 )
Discussion

21 / 56
 Pulsed waveform spectrum with moving targets

The stationary clutter stays centered at ±f0 , whereas a moving

target signal is shifted by fd . If the separation is large enough, the
stationary clutter can be filtered out, allowing detection of the
weaker target.

22 / 56
 Doppler resolution

Since linewidth is proportional to 1/Td , Doppler resolution

improves with increasing dwell time Td . 23 / 56
 Receiver bandwidth effects

I In principle, the signal carries power at all frequencies.

I About 91% of the total energy is inside the main lobe.
I Extending the bandwidth outside the main lobe increases
signal with at most additional 9%, but noise increases
proportionally to bandwidth.
I To maximize SNR, the receiver bandwidth should match
expected signal.
24 / 56
 Why multiple pulses?

I Consider an X-band radar (10 GHz) with 10 µs pulses. A

target moving at Mach 1 (340 m/s), implies a Doppler shift of
22.7 kHz.
I Without Doppler shift, one pulse has
(10 · 10−6 ) · (10 · 109 ) = 100 000 cycles.
I With Doppler shift, one pulse has
(10 · 10−6 ) · (10 · 109 + 22.7 · 103 ) = 100 000.227 cycles, only
about a quarter of a cycle more.
I The Doppler resolution from one pulse is
1/τ = 1/(10 · 10−6 ) Hz = 100 kHz, not sufficient to resolve
the Doppler peak at 22.7 kHz shift.
I Using multiple pulses, the resolution becomes
1/Td = 1/(N · PRI) = PRF/N , which can be made small
enough.

25 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

26 / 56
 Received signal

R0

Transmitted signal:
x(t) = A cos(2πf0 t + θ) = Re{Aej(2πf0 t+θ) } = Re{(Aejθ )ej2πf0 t }
Received signal:
 
2R0 n 4πR0
 o
y(t) ∼ x t − = Re Aej(θ− λ ) ej2πf0 t
c
  
4π
Complex amplitude of received signal: A exp j θ − R0
λ
Discussion
27 / 56
 Video detector

By mixing the received signal with a reference signal 2 cos(2πf0 t),

we obtain (writing θ − 4πR0 /λ = θ0 )

A cos(2πf0 t + θ0 ) · 2 cos(2πf0 t)
A h j(2πf0 t+θ0 ) 0
ih i
= e + e−j(2πf0 t+θ ) ej2πf0 t + e−j2πf0 t
2
A h j(4πf0 t+θ0 ) 0 0 0
i
= e + e−j(4πf0 t+θ ) + ejθ + e−jθ
2

After low pass filtering: A cos(θ0 ) = A cos(θ − 4πR0 /λ)

Not enough to determine both amplitude A and phase θ0 !

28 / 56
 Coherent detector

I In-phase (I) channel, reference signal cos(2πf0 t).

I Quadrature (Q) channel, reference signal
cos(2πf0 t + π/2) = − sin(2πf0 t).
The I/Q channels can be combined to form the analytic signal
0
a = I + jQ = A cos θ0 + jA sin θ0 = Aejθ
from which both amplitude A and phase θ0 can be determined.
Requires an accurate phase difference between I and Q references.
29 / 56
 Range bins, fast time

I For one transmitted pulse, sample the receiver output

(down-converted frequency, almost constant during sample
time).
I Each reflected pulse contributes to only one sample.
I Store the samples as a vector, with elements being called
range bins, range gates, range cells, or fast-time samples.
30 / 56
Slow time, datacube

31 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

32 / 56
 Measuring Doppler with multiple pulses

y[m] = A exp {j[θ − (4π/λ)(R0 − vmT )]}

     
2v 4πR0
= A exp j 2π (mT ) + θ −
λ λ
0
= A exp[j(2πfd tm + θ )], 0 ≤ m ≤ M − 1
The Doppler shift fd can be found from a frequency analysis of the
received analytic signal!
33 / 56
 Coherent pulses

If the pulses are non-coherent, the phase in consecutive pulses is

uncorrelated, and frequency analysis cannot be used.
34 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

35 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

36 / 56
 Doppler spectrum in one range bin

The signal is sampled every PRI seconds, hence the Doppler

spectrum (DFT in slow time) is contained in [−PRF/2, PRF/2],
where PRF = 1/PRI.

Clutter and stationary targets are centered at fd = 0, moving

targets and noise appear throughout the spectrum.

37 / 56
 Range-Doppler spectrum

The range-Doppler spectrum is obtained by plotting a radar signal

as function of both range (fast-time) and Doppler frequency
(Fourier transformed slow-time).
38 / 56
 Range-Doppler spectrum, realistic data

Note the clutter ridge around fd ≈ 0, and targets in different range

bins.

39 / 56
 Simulated data, http://radarsp.com
FRSP Demos/FRSP Non-GUI demos/Pulse Doppler/

RANGE-DOPPLER PLOT OF UNPROCESSED DATA RANGE-DOPPLER CONTOUR PLOT OF UNPROCESSED DATA

5.5

0 5

-20 4.5

-40

range (km)
4
-60
3.5
-80
3
-100
6
2.5
200
4
100
2 0 2
-100
range (km) 0 -200 velocity (m/s) -150 -100 -50 0 50 100 150
velocity (m/s)

Target extent in range corresponds to uncompressed pulse length

10 µs · c/2 = 1.5 km.
40 / 56
 Simulated data, http://radarsp.com

Clutter cancellation through high-pass filtering.

RANGE-DOPPLER PLOT OF CLUTTER-CANCELLED DATA RANGE-DOPPLER CONTOUR PLOT OF CLUTTER-CANCELLED DATA

5.5

0 5

-20 4.5

range (km)
4
-40

3.5
-60

3
-80
6
2.5
200
4
100
2 0 2
-100
range (km) 0 -200 velocity (m/s) -150 -100 -50 0 50 100 150
velocity (m/s)

Note target on clutter ridge edge is almost deleted.

41 / 56
 Simulated data, http://radarsp.com

Pulse compression (a chirped pulse is used, better range resolution,

covered in Chapter 20.)

RANGE-DOPPLER PLOT OF FULLY-PROCESSED DATA RANGE-DOPPLER CONTOUR PLOT OF FULLY-PROCESSED DATA

0 3.5

3
-50

range (km)
2.5

-100 2

1.5
-150
6
1
200
4
100
2 0 0.5
-100
range (km) 0 -200 velocity (m/s) -150 -100 -50 0 50 100 150
velocity (m/s)

42 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

43 / 56
 Main lobe clutter spreading

The Doppler shift depends on squint angle ψ, which implies a

frequency broadening due to the beam width θ3 :
 
2v 4v θ3 2vθ3
fd = cos ψ ⇒ BMLC = sin sin ψ ≈ sin ψ
λ λ 2 λ

44 / 56
 Clutter spectrum elements

The clutter spectrum of a moving platform has many elements,

induced by antenna sidelobes and different velocity components.
45 / 56
Example of range-Doppler clutter distribution

46 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

47 / 56
 Frequency Modulated Continuous Wave (FMCW)
In a continuous wave system, the transmitter is always on. The
frequency can be tuned by a control signal.

transmitting
Vtune VCO antenna

90◦

I
receiving
antenna

This makes for a cheap system, where the range information is

obtained by frequency synthesis, rather than time-of-flight
measurements.
Discussion 48 / 56
 Signal processing I
With a target at range R, the received signal is
r(t) = Aej2πf (t−2R/c)
After downconversion (removing ej2πf t ) the analytic signal is
a = I + jQ = Ae−j4πf R/c
Measure at discrete times tn with changing frequency fn :
tn = n∆t n = 0, 1, . . . , N − 1
fn = f0 + nB/N B = bandwidth
Rn = R0 − nv∆t v = radial velocity
The sampled signal is then (after

expansion of the product fn Rn )
2BR0 2vf ∆t n2 2Bv∆t
−j4πf0 R0 /c jn2π − + c0
an = Ae
| {z }e
Nc
ej2π Nc

=A0
The last term is a chirping effect, which can be neglected if the
target’s movement in one pulse (N v∆t) is much less than the
2
spatial resolution c/(2B): n 2Bv∆t
Nc
N v∆t
< c/(2B)  1. 49 / 56
 Signal processing II

Using both up-chirped and down-chirped signals,

 
2BR0 2vf0 ∆t
0 jn2π − N c + c
an = A e (up-chirp, fn = f0 + nB/N )
 
2BR0 2vf ∆t
jn2π + c0
an = A0 e Nc
(down-chirp, fn = f0 + (N − n)B/N )

2BR0 2vf0 ∆t
we can use FFT to compute both Nc and c .

Conclusion:
I Both range R and velocity v can be detected by FMCW.
I Necessary to use I and Q signals, sweeping the frequency over
bandwidth B in N steps up and down. Dwell time is 2N ∆t.
I The technique will be used in the lab on Friday.

50 / 56
 Outline

1 Doppler shift

2 The Fourier transform

3 Spectrum of a pulsed radar signal

4 Pulsed radar data acquisition

5 Doppler signal model

6 Range-Doppler spectrums
Stationary radar
Moving radar

7 FMCW radar

8 Conclusions

51 / 56
 Conclusions

I The Doppler shift is due to relative motion between radar and

target.
I The Fourier transform can be used to compute frequency
domain data.
I The spectrum of a pulsed radar has been characterized,
having several frequency scales.
I The I/Q channels and analytic signal have been introduced.
I The range-Doppler spectrum provide information on both
range and velocity.
I Range and velocity can be extracted from FMCW data using
Fourier analysis.

52 / 56
 Discussion

The Fourier transform is defined by

Z ∞
X(f ) = x(t)e−j2πf t dt
Z−∞
∞
x(t) = X(f )ej2πf t df
−∞

If x(t) has units of volts (V), what unit does X(f ) have?

Go back

53 / 56
 Discussion

The Fourier transform is defined by

Z ∞
X(f ) = x(t)e−j2πf t dt
Z−∞
∞
x(t) = X(f )ej2πf t df
−∞

If x(t) has units of volts (V), what unit does X(f ) have?

Answer: [X(f )] = Vs (x(t) is multiplied by dt, having units of s).

Go back

53 / 56
 Discussion

How does the spectrum look if 1) Td = T = τ , 2) Td > T = τ ?

Go back 54 / 56
 Discussion

How does the spectrum look if 1) Td = T = τ , 2) Td > T = τ ?

Answer: 1) Equal to the dashed sinc envelope. This corresponds to

a single rectangular pulse of length Td = T = τ with carrier
frequency f0 . 2) With Td > T = τ , there is a single peak inside
the dashed sinc envelope (one long rectangular pulse). Go back 54 / 56
 Discussion

The complex amplitude of the received signal is

  
4π
Ac = A exp j θ − R0
λ

If we could measure Ac , how can we extract information on the

range R0 ?

Go back

55 / 56
 Discussion

The complex amplitude of the received signal is

  
4π
Ac = A exp j θ − R0
λ

If we could measure Ac , how can we extract information on the

range R0 ?

Answer: By varying the wavelength λ, we control the phase due to

R0 . For instance, using two wavelengths λ1 and λ2 we get
  
Ac1 1 1 arg(Ac1 /Ac2 )
= exp −j4πR0 − ⇒ R0 =  
Ac2 λ1 λ2 −j4π λ11 − λ12

(under the requirement that |4πR0 ( λ11 − 1

λ2 )| < π).
Go back

55 / 56
 Discussion
Would a continuous wave radar be best for long or short ranges?

transmitting
Vtune VCO antenna

90◦

I
receiving
antenna

Go back

56 / 56
 Discussion
Would a continuous wave radar be best for long or short ranges?

transmitting
Vtune VCO antenna

90◦

I
receiving
antenna

Answer: High duty cycle implies low peak power and short range.
Go back

56 / 56