EITN90 Radar and Remote Sensing
Lecture 11: Fundamentals of pulse
compression waveforms
Daniel Sjöberg
Department of Electrical and Information Technology
Spring 2020
Learning outcomes of this lecture
In this lecture we will
I Introduce matched filters
I See how pulse compression can improve range resolution
I Study the linear frequency modulated waveform
I See how the ambiguity function can be used to analyze
waveforms
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 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
3 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
4 / 56
Radar waveforms
Many different waveforms are used in radars, taking many system
requirements and constraints into account: bandwidth, power,
Doppler tolerance, sidelobes, range resolution etc.
5 / 56
General time-invariant filtering
After filtering the received signal xr (t) through any linear,
time-invariant filter h(·) the signal is
Z ∞
y(t) = h(t − α)xr (α) dα
−∞
With a time delayed received signal xr (t) = x(t − td ) we have
Z ∞
y(t) = h(t − α)x(α − td ) dα
−∞
The amplitude |y(t)| can be estimated using the Schwartz
inequality
Z ∞ 1/2 Z ∞ 1/2
2 2
|y(t)| ≤ |h(t − α)| dα · |x(α − td )| dα
−∞ −∞
1/2
= (energy of filter) · (energy of signal)1/2
where the values of t or td do not matter in the last expression.
6 / 56
Matched filter
With knowledge of the transmitted signal x(t), we can choose the
matched filter
h(t) = x∗ (−t)
With this particular choice, we have the output
Z ∞ Z ∞
y(t) = h(t − α)x(α − td ) dα = x∗ (α − t)x(α − td ) dα
−∞ −∞
This is maximized at t = td (demonstrating the optimality of the
matched filter since the maximum is attained)
Z ∞ Z ∞
max |y(t)| = y(td ) = |x(α − td )|2 dα = |x(α)|2 dα
t −∞ −∞
which is proportional to the energy of the pulse waveform x(t).
7 / 56
Matched filter as maximizing SNR
Convolution in time domain corresponds to multiplication in
frequency domain, or
Y (ω) = H(ω)Xr (ω) = H(ω)X(ω)e−jωtd
Z ∞
1
⇒ y(t) = H(ω)X(ω)ejω(t−td ) dω
2π −∞
With white noise N (ω) = N0 , the total received noise power is
N0 ∞
Z
2
n (t) = |H(ω)|2 dω
2π −∞
Hence the SNR at t = td is
R∞ 2
|y(td )|2 | −∞ H(ω)X(ω) dω|
SNR = = R∞
n2 (t) N0 −∞ |H(ω)|2 dω
which is maximized for H(ω) = X ∗ (ω) or h(t) = x∗ (−t).
Discussion
8 / 56
Example: rectangular pulse
For the simple rectangular pulse (setting td = 0)
τ τ
x(t) = A, − ≤ t ≤
2 2
the matched filter is
τ τ
h(t) = A, − ≤ t ≤
2 2
and the filtered response is
(
A2 (τ + t) −τ ≤ t ≤ 0
y(t) =
A2 (τ − t) 0 ≤ t ≤ τ
9 / 56
Generic response
For general waveforms, the filtered response is typically described
in terms of mainlobe and sidelobes.
10 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
11 / 56
Resolution
The Rayleigh resolution criterion is that the peak of one target is
at the null of the second target.
The above figure corresponds to the matched filter response of
rectangular pulses.
12 / 56
Fourier uncertainty principle
The widths of a signal of zero mean in time and frequency domain
can be defined by sR ∞
t2 |y(t)|2 dt
Dt = R−∞
∞ 2
−∞ |y(t)| dt
sR ∞
2 2
−∞ ω |Y (ω)| dω
Dω = R∞
2
−∞ |Y (ω)| dω
The product of these widths is bounded below as
r
π
Dt Dω ≥
2
with equality for Gaussian signals. This motivates that resolution
in time (range) is inversely proportional to frequency bandwidth.
c
δR = κ
2B
κ ≈ 1, definitions of resolution and bandwidth often chosen to
conform with this formula. 13 / 56
Phase difference between two targets
Two targets separated by the Rayleigh resolution can present
radically different responses depending on phase difference.
14 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
15 / 56
LFM waveform
A baseband linear frequency modulated waveform (LFM) is
" 2 #
t τ τ
x(t) = A cos πτ B , − ≤t≤
τ 2 2
The waveform is characterized by the time-bandwidth product τ B
and normalized time t/τ .
16 / 56
Instantaneous frequency
The instantaneous phase is φ(t) = πτ B(t/τ )2 , and
instantantaneous frequency is
1 dφ B τ τ
fi (t) = = t, − ≤t≤
2π dt τ 2 2
The linear change motivates the term linear frequency modulation.
17 / 56
LFM spectrum
The LFM spectrum has a relatively flat spectrum across bandwidth
B. Flatness and roll-off improves as time-bandwidth product τ B
increases.
18 / 56
Matched filter response
The matched filter response for the LFM waveform is
h i
∞ sin 1 − |t| πτ B t
|t|
Z
τ τ
y(t) = x∗ (α − t)x(α) dα = 1 − , |t| ≤ τ
−∞ τ 1 − |t| πτ B t
τ τ
The peak is much more narrow than total pulse width τ !
Discussion
19 / 56
Range resolution
For large values of τ B, the first null occurs at t ≈ 1/B. With
range R = ct/2, the Rayleigh resolution in range is
c
δR =
2B
The −4 dB pulsewidth is 1/B. The width of the main lobe is
τ
compressed by a factor of about 1/B = τ B.
20 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
21 / 56
Dispersive filters
Filters having frequency dependent group delay
dφ(ω) τ
tgd = − = ω
dω 2πB
can both stretch and compress waveforms. One implementation is
surface acoustic wave (SAW) technology:
The device couples electromagnetic energy to acoustic waves,
where the coupling is strongest when the distance between the
metal fingers correspond to λ/2 for the acoustic wave. Chirping is
obtained by different acoustic propagation lengths. Works up to
about 3 GHz, high insertion loss. 22 / 56
Digital filters
With a digitized signal, the matched signal can be implemented
using the Fast Fourier Transform (FFT) of the analytic signal
x[n] = xI [n] + jxQ [n]:
y[n] = FFT−1 {H[·] X[·]}[n], X[k] = FFT{x[·]}
Error correction is obtained by transmitting a pilot pulse and
recording the received (distorted) signal, taking into consideration
imperfections in the transmit/receive chain.
23 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
24 / 56
Sidelobe reduction
Sidelobes of the compressed pulse can be reduced by weighting the
filter in amplitude. The cost is an increased mainlobe width.
Two parameters:
I PSR = Peak Sidelobe Ratio.
I n̄ = number of sidelobes in frequency domain.
25 / 56
Reduced resolution
When increasing the sidelobe suppression, the resolution is
decreased.
26 / 56
Sidelobe suppression, time-bandwidth product
The theoretical sidelobe reduction is achieved when the weighting
is applied to a rectangular spectrum. A real LFM has some
additional spread, which is reduced as τ B → ∞.
27 / 56
Matlab demo, http://radarsp.com
FRSP Demos/FRSP GUI Demos/FRSP LFM-GUI
28 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
29 / 56
Ambiguity function
Taking into account the possibility of both time delay and Doppler
shift, the received signal is
xr (t) = ej2πfd t x(t − td )
Centering the waveform over td = 0 and applying the matched
filter and normalizing x with its energy, we find the ambiguity
function
R∞
x(α)ej2πfd α x∗ (α − t) dα
A(t, fd ) = −∞ R ∞ 2
−∞ |x(α)| dα
This function satisfies
Z ∞Z ∞
|A(t, fd )|2 dt dfd = 1 and |A(t, fd )| ≤ |A(0, 0)| = 1
−∞ −∞
Discussion
30 / 56
Ambiguity for a simple rectangular pulse
For an unmodulated pulse,
1 τ τ
x(t) = √ , − ≤t≤
τ 2 2
the ambiguity function can be calculated as
h i
sin πf τ 1 − |t|
|t| d τ
A(t, fd ) = 1 − , |t| ≤ τ
τ πf τ 1 − |t|
d τ
Depends on normalized time t/τ and normalized Doppler shift fd τ .
31 / 56
Ambiguity for a simple rectangular pulse
0.8
0.6
0.4
0.2
0
5
1
0 0.5
0
-0.5
-5 -1
-1
-2
-3
-4
-5
-1 -0.5 0 0.5 1
32 / 56
Ambiguity function for LFM waveform
For a linear frequency modulated pulse
1 B 2
x(t) = √ exp jπ t , |t| ≤ τ
τ τ
the ambiguity function can be calculated as
h i
sin πτ B 1 − |t| fd
+ t
|t| τ B τ
A(t, fd ) = 1 − , |t| ≤ τ
τ πτ B 1 − |t| fd t
τ B + τ
Depends on normalized time delay t/τ and normalized Doppler
shift fd /B, with time-bandwidth parameter τ B.
33 / 56
Ambiguity function for LFM waveform, τ B = 20
0.8
0.6
0.4
0.2
0
1
0.5 1
0 0.5
0
-0.5 -0.5
-1 -1
0.8
0.6
0.4
0.2
-0.2
-0.4
-0.6
-0.8
-1
-1 -0.5 0 0.5 1
34 / 56
Ambiguity function for LFM waveform, different τ B
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
35 / 56
Range-Doppler coupling
From the formula
h i
sin πτ B 1 − |t| fd
+ t
|t| τ B τ
A(t, fd ) = 1 −
, |t| ≤ τ
τ πτ B 1 − |t| fd t
τ B + τ
we see that a non-zero Doppler shift fd can be interpreted as
I Time shift ∆t = −fd τ /B
I Amplitude reduction by (1 − |∆t|/τ ) = (1 − |fd /B|)
This leads to shifts in peak location, peak amplitude, and
decreased resolution due to peak widening.
36 / 56
Degradations in presence of Doppler shift
37 / 56
Matched filter response in presence of Doppler shift
38 / 56
Degradation of side lobes
The effect of applying an amplitude taper to control sidelobes is
reduced in presence of Doppler shift.
39 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
40 / 56
Phase modulation
Instead of modulating the frequency, the phase can be controlled,
typically in a digital way:
41 / 56
Matched filter
Discussion
42 / 56
Different phase codes
I Biphase codes
I Two phase states (+/−)
I Minimum peak sidelobes (MPS)
I Bi-phase Barker codes: achieve a 1 : N peak sidelobe to
mainlobe ratio for codes of length N
I Maximum length n
√ sequence (MLS): length ` = 2 − 1, peak
sidelobes ∼ 1/ `
I Polyphase codes (not treated in this lecture)
I More than two phase states: more degrees of freedom
I Frank, P1, P2, P3, P4
43 / 56
Thumb-tack ambiguity function
Careful design of the phase codes can result in a thumb-tack like
ambiguity function.
A number of filter banks can be used to search the Doppler space,
applying the matched filter at the output of each filter. Enables
simultaneous estimation of range and Doppler.
44 / 56
Barker codes
45 / 56
Minimal peak sidelobe (MPS)
46 / 56
Examples of minimal peak sidelobe (MPS). . .
47 / 56
Maximal length sequences
48 / 56
Comparison LFM and biphase MLS, waveform
Higher average sidelobe levels in MLS, higher peak sidelobe in
LFM.
49 / 56
Comparison LFM and biphase MLS, spectrum
The wide spectrum of MLS can be attributed to the abrupt
changes in phase. Can create an electromagnetic interference
problem.
50 / 56
Outline
1 Matched filters
2 Range resolution
3 Linear frequency modulated waveforms
4 Matched filter implementations
5 Sidelobe reduction in an LFM waveform
6 Ambiguity functions
7 Phase-coded waveforms
8 Conclusions
51 / 56
Conclusions
I Matched filters maximize SNR for a given waveform
I The resulting pulse compression improves range resolution
I The LFM is a generic waveform, sidelobes can be improved by
tapering
I Phase coding can produce very narrow ambiguity peaks, but
with a wide spectrum
52 / 56
Discussion
Why is R∞
| 2
|y(td )|2 −∞ H(ω)X(ω) dω|
SNR = = R∞
n2 (t) N0 −∞ |H(ω)|2 dω
maximized by H(ω) = X ∗ (ω)?
Go back
53 / 56
Discussion
Why is R∞
| 2
|y(td )|2 −∞ H(ω)X(ω) dω|
SNR = = R∞
n2 (t) N0 −∞ |H(ω)|2 dω
maximized by H(ω) = X ∗ (ω)?
Answer: The Schwartz inequality
Z ∞ 2 Z ∞ Z ∞
2 2
H(ω)X(ω) dω ≤
|H(ω)| dω · |X(ω)| dω
−∞ −∞ −∞
implies Z ∞
1
SNR ≤ |X(ω)|2 dω
N0 −∞
with equality for the choice H(ω) = X ∗ (ω).
Go back
53 / 56
Discussion
Given the matched filter response for the LFM waveform is
h i
∞ sin 1 − |t| πτ B t
|t|
Z
τ τ
y(t) = x∗ (α − t)x(α) dα = 1 − , |t| ≤ τ
−∞ τ 1 − |t| πτ B t
τ τ
where in the graph below can you find the value of πτ B in dB?
Go back
54 / 56
Discussion
Given the matched filter response for the LFM waveform is
h i
∞ sin 1 − |t| πτ B t
|t|
Z
τ τ
y(t) = x∗ (α − t)x(α) dα = 1 − , |t| ≤ τ
−∞ τ 1 − |t| πτ B t
τ τ
where in the graph below can you find the value of πτ B in dB?
Answer: For |t/τ | ≈ 1, we have |y(t)| ≈ 1/(πτ B) ≈ −44 dB in
this graph. Go back
54 / 56
Discussion
R∞
x(α)ej2πfd α x∗ (α−t) dα
The ambiguity function A(t, fd ) = −∞ R ∞ |x(α)|2 dα satisfies
−∞
Z ∞Z ∞
|A(t, fd )|2 dt dfd = 1 and |A(t, fd )| ≤ |A(0, 0)| = 1
−∞ −∞
Given the above and a pulse shape with finite support,
|A(t, fd )|2 = 0 when |t| > τ and |fd | > B
what average value of |A(t, fd )|2 do you expect for |t| < τ and
|fd | < B? How should τ be chosen to minimize the average
ambiguity?
Go back
55 / 56
Discussion
R∞
x(α)ej2πfd α x∗ (α−t) dα
The ambiguity function A(t, fd ) = −∞ R ∞ |x(α)|2 dα satisfies
−∞
Z ∞Z ∞
|A(t, fd )|2 dt dfd = 1 and |A(t, fd )| ≤ |A(0, 0)| = 1
−∞ −∞
Given the above and a pulse shape with finite support,
|A(t, fd )|2 = 0 when |t| > τ and |fd | > B
what average value of |A(t, fd )|2 do you expect for |t| < τ and
|fd | < B? How should τ be chosen to minimize the average
ambiguity?
Answer: Since
the integral of |A|2 equals 1, we should have the
average value |A|2 = 1/(4τ B). To minimize this, increase the
pulse length τ .
Interpretation: If you try to decrease |A|2 somewhere, it needs to
increase somewhere else.
Go back
55 / 56
Discussion
With each chip consisting of a fixed frequency carrier wave
cos(2πf0 t + φ), with pulse length τchip , what is the chip bandwidth
Bchip and time-bandwidth product τ B with τ = N τchip and
B = Bchip ?
Go back
56 / 56
Discussion
With each chip consisting of a fixed frequency carrier wave
cos(2πf0 t + φ), with pulse length τchip , what is the chip bandwidth
Bchip and time-bandwidth product τ B with τ = N τchip and
B = Bchip ?
Answer: Bchip = 1/τchip , τ B = N τchip Bchip = N . Hence, the
time-bandwidth product only depends on the number of chips, N .
Go back
56 / 56
