You are on page 1of 5

A More Exact Linear FMCW Radar Signal Model

for Simultaneous Range-Velocity Estimation


Peter Asuzu∗† and Charles Thompson∗
∗ Department of Electrical and Computer Engineering
University of Massachusetts Lowell, Lowell, Massachusetts
† Autoliv Radar Systems, Lowell, Massachusetts

Abstract—A Frequency-Modulated Continuous Wave radar 9&2


signal model is presented. The model demonstrates the under-
lying principle for the simultaneous measurement of range and V W
7;
velocity typically employed in automotive applications. Model
validation against radar measurements are presented. The model
is used to demonstrate phenomena such as the influence of target
velocity on radar backscatter and the range-velocity uncertainty
principle. VP W /2
$' /3)
I. I NTRODUCTION [ W
With the increasing interest in autonomous driving, radar
is becoming a technology of choice for safety and driver- $V W
assistance applications such as blind-spot detection, collision 5;
avoidance, and emergency braking [1], [2]. A popular mod-
ulation scheme employed in automotive radar systems is the Fig. 1. Schematic of the FMCW radar transceiver.
Frequency Modulated Continuous Wave (FMCW). The case
for the range estimation of a target using linear FMCW signals
has been shown in [3]–[5]. A simultaneous estimate of range
Where A is an amplitude scaling coefficient that accounts for
and velocity has been further shown in [4], [6].
factors such as atmospheric attenuation and target scattering.
In multi-target scenarios, the model demonstrated in [4],
τ is the time-delay due to the target’s range, and has been
which uses a train of pulsed FMCW signals, is seen to have
assumed to be negligibly small within the chirp’s duration
strong immunity to target ambiguities, however it does not
such that τ  T , and hence wr (t − τ ) ∼
= wr (t).
capture the complete range-Doppler spectrum. By the aid of
Taylor expansions, we demonstrate a model that reveals the Within each chirp, the target is assumed stationary, but its
complete information of the range-Doppler spectrum. The position is changing from chirp to chirp. The change in the
exact spectrum is then used to demonstrate the influence target’s position between chirps can be related to its velocity,
of target-velocity on radar backscatter and the uncertainty where this velocity is assumed constant for the duration of
principle for simultaneous range-velocity measurement. N chirps. This phenomenon is illustrated in Fig. 3, and is
analogous to taking snapshots of the target and comparing its
II. M ODEL DERIVATION positions relative to the first frame.
We begin with an FMCW radar transceiver as shown in
Fig. 1. For a simultaneous measurement of range and velocity,
N chirps as shown in Fig. 2 can be used [4], [6], [7]. The f

transmit signal is given as Q  Q  Q 1


I 7
2
s(t) = cos(ω0 t + πβt )wr (t) (1)
  
1 T
wr (t) = rect t− (2)
T 2
 
t
rect = 1 : |t| ≤ T /2; 0 : |t| > T /2 (3) I W W W
T 0 T Tˆ 2T̂ NTˆ tˆ

β is the chirp rate, ω0 is the starting angular frequency of the


chirp, and T is the chirp’s duration. In the presence of a point Fig. 2. FM profile for multiple chirps. t is an intra-chirp time scale, and t̂
is an inter-chirp time scale. The dashed line is the rectangular window gating
target, the corresponding IF signal at the receiver’s output can the chirp.
be shown to be [7]
A  
x(t) = cos 2πβτ t + ω0 τ − πβτ 2 wr (t) (4) To account for the change in position between chirps, the
2

978-1-5386-4167-5/18/$31.00 ©2018 IEEE 0007


WDUJHW
to the target’s velocity and can be obtained from a Fourier
Q  transform in the t̂ domain, since the phase shift is varying in
this timescale. By assembling the Fourier transforms obtained
5
in t-domain, and then transforming the assembly in the t̂-
t
domain, a two-dimensional spectrum with information about
 ǻ5 the target’s range and velocity is obtained. We show this
procedure as follows: for all N chirps, the collection of Fourier
t transforms from the t-domain can be obtained with the aid of

(6) as
ǻ5 
A 2
X(ω, t̂) = Wr (ω + 2πβτ )e−j(ω0 τ −πβτ ) +
t 4

2
Wr (ω − 2πβτ )ej(ω0 τ −πβτ ) wv (t̂) (9)
tˆ where
  
1 N T̂
wv (t̂) = rect t̂ − ×
N T̂ 2
∞   
1 T
rect t̂ − − nT̂
1 ǻ5 n=−∞
T 2
    ∞
1 N T̂ 2π
= rect t̂ − Cn ej T̂ nt̂ (10)
N T̂ 2 n=−∞
Fig. 3. Target’s positions observed for each frame. Within each frame (t-
domain) ΔR is constant, but across all frames (t̂-domain) ΔR is varying. with wv (t̂) being the window encompassing all chirps. The
complex exponential sum in (10) is the Fourier series repre-
sentation of the repeating rectangular window in Fig. 2. We
target’s time delay can be written as
can rewrite (9) by substituting for Wr from (7) and τ from
2 2 (5). Hence
τ= (R0 + ΔR) = (R0 + v0 t̂) (5) 
c c AT −jω T
R0 is the initial range of the target and ΔR is the range X(ω, t̂) = e 2
4
increment due to the target’s velocity v0 . Note that τ is
T −jb0 −jb1 t̂ −jb2 t̂2
dependent on t̂ not t, where t is the time scale for an sinc ω + a0 + a1 t̂ e e e +
2 
individual chirp and t̂ is the longer time scale encompassing

T jb0 jb1 t̂ jb2 t̂2
all chirps (Fig. 2). Within each chirp, the target’s range can be sinc ω − a0 − a1 t̂ e e e wv (t̂) (11)
2
estimated from the magnitude-Fourier transform of (4), where
the transform of (4) can be written as [7] where
 4πβ
Aπ 2 a0 = R0 (12)
X(ω) = δ(ω + 2πβτ )e−j(ω0 τ −πβτ ) + c
2 4πβ
 a1 = v0 (13)
2 1 c
δ(ω − 2πβτ )ej(ω0 τ −πβτ ) ⊗ Wr (ω)
2π 2 4πβ
 b0 = R0 (πβT + ω0 ) − 2 R02 (14)
A 2 c c
= Wr (ω + 2πβτ )e−j(ω0 τ −πβτ ) +
2 8
4 b1 = (πβT + ω0 ) − πβ 2 R0 v0 (15)
 c c
j(ω0 τ −πβτ 2 )
Wr (ω − 2πβτ )e (6) 4v 2
b2 = −πβ 20 (16)
c
with The Fourier transform across the t̂ dimension is desired,
 
ωT −jω T thus we rewrite (11) as
Wr (ω) = T sinc e 2 (7)
2

sin(·) AT −jω T
sinc(·) = (8) X(ω, t̂) = e 2 X + (ω, t̂)e−jb1 t̂ +
(·) 4

and ⊗ denoting the convolution operator. The target’s velocity X − (ω, t̂)ejb1 t̂ wv (t̂) (17)
can be estimated by making multiple measurements of its

T ∓jb2 t̂2
range as given in (4), and observing the change in the phase-
X ± (ω, t̂) = e∓jb0 sinc ω ± a0 ± a1 t̂ e (18)
shift of x(t). The rate of change of the phase-shifts is related 2

978-1-5386-4167-5/18/$31.00 ©2018 IEEE 0008


We recognize from the arguments of the sinc functions in 1/T̂ is classically referred to as the radar’s pulse repetition
(18) that the a0 term is related to the target’s range and a1 t̂ is frequency (PRF). We note that the zeroth-order expansion of
an additive term to the range due to the target’s motion. For (22) is equivalent to applying the following approximations
small t̂ time scales we can approximate (18) by expanding to (11): a0  a1 t̂ in argument of the sinc function and the
into their Taylor series. Thus, quadratic phase term b2 is negligible. These approximations
∞ were employed by Barrick in [4].
 1 ±
X ± (ω, t̂) = e∓jb0 Um (ω)t̂m (19)
m=0
m! III. T EST CASE SIMULATION AND EXPERIMENT
where The model in (22) was simulated with a three-term Taylor
∂ m
T expansion for a target at 10 m, moving with +3 mph velocity.
±
Um (ω) = sinc ω ± a0 ± a1 t̂ × A rectangular window in the Doppler time-domain t̂ was used
∂ t̂m  2
 with a high PRF assumed, such that the Doppler spectrum
e∓jb2 t̂ 
2
(20) Wv (ω̂) was restricted to the first harmonic (n = 0). A starting
t̂=0 transmit frequency of 76 GHz with 1 GHz bandwidth was
The Fourier transform of (19) in the t̂-domain yields, specified.
Fig. 4a shows the magnitude-spectrum of the IF signal,
X± (ω, ω̂) = F̂{X ± (ω, t̂)} with the target’s backscatter on the range-velocity axes. The
∞
∓jb0 jm ± dm spectrum as shown in Fig. 4a is also known as the range-
=e 2π Um (ω) m δ(ω̂) (21) velocity map, where the factors c/2β and c/2f0 have been
m! dω̂
m=0
used to respectively scale frequency-axes f and fˆ to range
F̂{·} signifies a Fourier transform operation in the t̂ domain. and velocity [7]. The target’s range and velocity can be found
The time-derivative property of the Fourier transform has been from the argument of the maximum of the sinc functions. The
employed in (21). Referring to (17), the Fourier transform of image sinc functions in the third quadrant (negative range,
X(ω, t̂) is now negative velocity) are redundant for range-velocity estimation.
Fig. 4b shows the range-velocity maps simulated by generating
X(ω, ω̂) = F̂{X(ω, t̂)} sinusoids according to Eqn. (4) and computing their DFTs; and

AT −jω T Fig. 4c shows the range-velocity maps obtained from field data
= e 2 X+ (ω, ω̂ + b1 )+
4 of a stationary corner reflector target with the radar receding
 from the target. Good agreement can be qualitatively seen
1
X− (ω, ω̂ − b1 ) ⊗ Wv (ω̂) among all three maps.

AT −jω T  (−j)m 

= e 2
IV. BACKSCATTER - VELOCITY COUPLING
4 m=0
m!
Using the FMCW signal model in (22), the peak backscatter
dm  
e−jb0 Um +
(ω) m Wv (ω̂ + b1 ) + was observed to vary with the target’s velocity, where the peak
dω̂ backscatter is defined as the maximum of the sinc function
dm  
ejb0 Um −
(ω) m Wv (ω̂ − b1 ) (22) in the magnitude spectrum. To demonstrate this phenomenon,
dω̂ the FMCW model was used to simulate the radar spectrum
In (22), the convolution property for mth -order derivatives of for a target at 2m, transmit bandwidth of 1 GHz, and various
the delta function has been employed [7], [8], namely target velocities. The result is shown in Fig. 5, from which
a nonlinear relationship can be inferred between the radar
m m peak-backscatter and the target’s velocity. Note the pronounced
d m d
W (ω̂) ⊗ δ(ω̂ − a) = (−1) W (ω̂ − a) (23) trend seen from the higher expansion Taylor terms when com-
dω̂ m dω̂ m
pared to the 1-term. The trend in Fig. 5 suggests an influence
Wv (ω̂) is the Fourier transform of wv (t̂) in (10), and can be from target velocity which can distort radar measurements that
written as rely on peak backscatter.
  ∞ To further examine the backscatter-velocity phenomenon,
ω̂N T̂ −j ω̂N T̂ 2π
Wv (ω̂) = N T̂ sinc e 2 ⊗ Cn δ(ω̂ − n) experimental radar data was obtained for an identical transmit
2 n=−∞ T̂ bandwidth. The target in this case was a dry road course,
∞    
2π N T̂ with the radar mounted to a vehicle and traveling over it. The
= N T̂ Cn sinc ω̂ − n × ground was located at a radial distance of 2 m. We note that for
n=−∞ T̂ 2
  the radar system used, the ground is more aptly described as
e [ T̂ ] 2
−j ω̂− 2π n N T̂
(24) a distributed target consisting of several point scatterers. The
  use of this setup and a distributed target allows for a fixed
T T −jπn T target range with varying target velocity. More details on the
Cn = sinc πn e T̂ (25)
T̂ T̂ experiment can be found in [7], [9].

978-1-5386-4167-5/18/$31.00 ©2018 IEEE 0009


0
dB
30 0
-10 -1
20
-20

Magnitude (dB)
10 -30 -2
Range (m)

Taylor terms: 1
-40 2
0
-50 -3 3
4
-10 -60
-4
-70
-20
-80
-5
-30 -90 0 5 10 15 20 25 30 35 40
-10 -5 0 5 10 Velocity (mph)
Velocity (mph)
Fig. 5. Radar peak backscatter dependence on target velocity. Obtained from
(a) Simulated from 3-term model in Eqn. (22) Eqn.(22.) for R0 = 2 m and βT = 1 GHz

dB 6
30 0
raw data
-10 4 mean
20 median
-20 2
10 -30
Magnitude (dB)
0
Range (m)

-40
0 -2
-50
-4
-10 -60
-70 -6
-20
-80 -8

-30 -90 -10


-10 -5 0 5 10 0 5 10 15 20 25 30 35 40

Velocity (mph) Velocity (mph)

(b) Simulated from DFT of Eqn. (4) Fig. 6. Radar peak backscatter dependence on target velocity from experi-
mental data. R0 = 2 m and βT = 1 GHz.

dB
30 0
-10 The peak ground returns are plotted against their associated
20 velocities in Fig. 6. The spread in the raw measurements
-20
at each velocity bin can be related to the nature of the
10 -30 experiment, where factors such as vehicle vibrations and
Range (m)

-40 pitch and undesired returns contribute to the ground’s returns.


0 The mean and median backscatter highlight the backscatter-
-50
velocity dependence. Comparing Figs. 5 and 6, the backscatter
-10 -60
falloff in the experiment is better predicted from the higher
-70 order Taylor expansions of the FMCW model, though the
-20
-80 semi-parabolic falloff shape from the model is not well-
reproduced in the experimental data.
-30 -90
-10 -5 0 5 10 V. R ANGE -V ELOCITY UNCERTAINTY PRINCIPLE
Velocity (mph) The model in Eqn. (22) can be used to gain insight into the
uncertainty principle for FMCW range-velocity measurements,
(c) Measured with corner-reflector target
which has been traditionally evaluated through the ambiguity
Fig. 4. Magnitude-spectrum (range-velocity maps) of x(t, t̂) for a target at function [10]–[12]. To model this effect, range-velocity maps
10 m with 3 mph velocity

978-1-5386-4167-5/18/$31.00 ©2018 IEEE 0010


were simulated for a 2 m, 30 mph target with various transmit
bandwidths. The maps are shown in Fig. 7, from which it
1-term, 500 MHz 4-term, 500 MHz
can be seen that increasing bandwidth is associated with 2.4 2.4
reductions in the target’s range profile and expansions in its 2.2 2.2
velocity profile. The effect is more readily observed with the
higher order expansion of Eqn. (22). Since the radar’s range 2 2

Range (m)
resolution can be shown to be related to bandwidth [7], the 1.8 1.8
trend in Fig. 7 suggests that a simultaneous improvement of
1.6 1.6
range and velocity resolutions cannot be obtained for the train-
of-chirps linear FMCW radar signal. 1.4 1.4

1.2 1.2
VI. C ONCLUSION
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
An FMCW radar signal model has been presented. The
model demonstrates the underlying principle for a simultane-
ous measurement of range and velocity by employing a train-
of-chirps modulation such that the range information can be 1-term, 1000 MHz 4-term, 1000 MHz
2.4 2.4
extracted from each chirp, and the velocity information from
the phase relationship of the target’s positions across all chirps. 2.2 2.2

The derivation employed for the model is somewhat iden- 2 2

Range (m)
tical to that used in [4], however Taylor expansions have
1.8 1.8
been used to demonstrate a version that captures the complete
range-Doppler spectrum. Qualitatively good agreement has 1.6 1.6

been observed between the model and radar measurements, 1.4 1.4
and phenomena such as backscatter-velocity coupling and
1.2 1.2
range-velocity uncertainty have been demonstrated.
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
ACKNOWLEDGMENT
The authors would like to thank Autoliv Radar Systems for
support received in this work.
1-term, 1500 MHz 4-term, 1500 MHz
2.4 2.4
R EFERENCES
2.2 2.2
[1] J. Dickmann, J. Klappstein, M. Hahn, M. Muntzinger, N. Appenrodt,
C. Brenk, and A. Sailer, “”present research activities and future re- 2 2
Range (m)

quirements on automotive radar from a car manufacturer??s point of


view”,” in 2015 IEEE MTT-S International Conference on Microwaves 1.8 1.8
for Intelligent Mobility (ICMIM), April 2015, pp. 1–4.
[2] H. Rohling, “Milestones in radar and the success story of automotive 1.6 1.6
radar systems,” in 11-th INTERNATIONAL RADAR SYMPOSIUM, June
2010, pp. 1–6. 1.4 1.4
[3] A. Stove, “Linear fmcw radar techniques,” Radar and Signal Processing, 1.2 1.2
IEE Proceedings F, vol. 139, no. 5, pp. 343–350, Oct 1992.
[4] D. Barrick, “FM/CW Radar Signals and Digital Signal Processing,” 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
National Oceanic and Atmospheric Administration, Tech. Rep. 283-
WPL 26, Jul 1973.
[5] H. Rohling and M.-M. Meinecke, “Waveform design principles for
automotive radar systems,” in Radar, 2001 CIE International Conference
1-term, 2000 MHz 4-term, 2000 MHz
on, Proceedings, 2001, pp. 1–4. 2.4 2.4
[6] E. M. Poulter, FMCW Radar Signal Processing. Lower Hutt, N.Z.:
Physics and Engineering Laboratory, DSIR, 1982. 2.2 2.2
[7] P. Asuzu, “Millimeter-wave Radar Investigations of Road Conditions
and Its Impact on Wheel Slip,” Ph.D. dissertation, University of Mas- 2 2
Range (m)

sachusetts Lowell, May 2018, To be published.


1.8 1.8
[8] Michael O’Flynn, Eugene Moriarty, Linear Systems. Time Domain and
Transform Analysis. Harper & Row, 1987. 1.6 1.6
[9] P. Asuzu and C. Thompson, “Road Condition Identification from
Millimeter-wave Radar Backscatter Measurements,” in IEEE Radar 1.4 1.4
Conference, Apr 2018, To be published.
[10] F. Nathanson, Radar Design Principles - Signal Processing and the 1.2 1.2
Environment (2nd Edition). SciTech Publishing, 1999. [Online]. Avail- 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
able: http://app.knovel.com/hotlink/toc/id:kpRDPSPEEC/radar-design- Vel (mph) Vel (mph)
principles/radar-design-principles
[11] E. M. N. Levanon, Radar Signals. Hoboken, US: Wiley-Interscience,
20040921 2004. Fig. 7. Simulated FMCW radar spectra from Eqn. (22) illustrating the
[12] Bassem R. Mahafza, Radar Systems Analysis and Design Using MAT- uncertainty principle in range-velocity measurements. Target is at 2 m with
LAB. Chapman & Hall/CRC, 2000. 30 mph velocity. Backscatter scale: -60 to 0 dB

978-1-5386-4167-5/18/$31.00 ©2018 IEEE 0011

You might also like