Vous êtes sur la page 1sur 4

A New Method of multi-target detection for FMCW Automotive

Radar
Yuanzhang Fan*, Kai Xiang, Jianping An, Xiangyuan Bu

* School of Information and Electronics


Beijing Institute of Technology
Beijing, China
fanyz@bit.edu.cn

stepped-FMCW are considered [3-5]. FSK radar provides a


Abstract very high velocity resolution and avoids any so-called ghost
target but cannot distinguish multiple fixed or same velocity
In multi-target situations, typically triangular FMCW radar targets with different distances. FMCW radar is an effective
brings out false targets and degrades the radar performance. A automotive sensor method because of the advantages of very
novel transmitted waveform and multi-target detection high range resolution, short measurement time and low
algorithm for automotive applications is presented in this computation complexity, yet ghost targets will occur in
paper. The proposed waveform is composed of three multiple target situations. This paper proposes a new multi-
segments: up ramp, flat frequency and down ramp in every target detection algorithm for FMCW radar to reduce the
small period. Especially, the flat frequencies are modulated ghost targets, which combines the advantages of both FSK
by a unique PN code sequence within a determinate and FMCW radar.
frequency range. The unambiguous relative velocities
detected in the flat frequency segment can cancel some false 2 FMCW radar in multi-target situations
targets initially. Corresponding to the waveform, which
combines the advantages of both FSK and FMCW formats, a Fig. 1 shows the time-instantaneous frequency relationships
progressive judgment algorithm is used in the continuous of transmitted signalķ echo signalĸ (reflected by a target
small period to further eliminate the false targets. with the distance R and the relative velocity v ) and IF
(Intermediate Frequency, which is the difference frequency of
Keywords: Multi-target detection, FMCW, Automotive radar, transmitted signal and echo signal in the system with
waveform design superheterodyne receiver) signal f IF (t ) (Fig. 1 (b)) of
FMCW radar using typically triangular waveform. When the
1 Introduction v 0 , this means the Doppler frequency f d 2vf0 / c 0
Because of possible casualties, property damage and road and then it is a stationary target, its echo signal is presented as
congestion by traffic accidents, automotive safety has become Ĺin figure 1(a), Where c is the speed of electromagnetic
a focal point of consumers, governments, and society. waves.
Automotive safety systems are classified in passive safety
systems and active safety systems[1]. The objective of
passive safety systems which have become standard
equipment in passenger vehicles, such as seat belts, head rests,
and air bags, is only to reduce injuries and fatalities during
accidents. Otherwise, many automotive active safety systems
recently have been designed to use for providing warning,
decelerating or braking automatically, preventing accidents
and better protecting occupant and pedestrian in advance of W
potential collision.
In these systems, environmental sensors are employed to
detect targets distance and relative velocity simultaneously
and with high accuracy. Comparing with other alternative
sensors like video, lidar and ultrasound, radar, especially
millimeter wave radars, has the advantages of high tolerance Fig. 1 Principle of FMCW radar for a single moving /
to all weather conditions and easy mount without requiring stationary target using triangular waveform
specific cut-outs or similar accommodations[2]. (a) the transmitted signalķ and echo signal(moving
For the detection and parameter estimation of automotive targetĸ; stationary targetĹ); (b) the corresponding IF
targets, many radar modulation formats, such as FSK, FMCW, signal of the moving target
Other parameters are: and down ramps) in every small period, as shown in Fig. 2.
— The delay time of target echo signal ( W ), Especially, the flat frequencies f mn are modulated within a
— Chirp bandwidth ( W ), determinate frequency range by a M-bit PN code sequence
— The ramps duration ( T ), which are configured for different radars. The modulated
— The carrier starting frequency ( f0 ) , duration Tb of M-bit PN code sequence is regard as a big
— The IF frequency of a stationary target ( f s ) and period. The subscript m(m 1, 2,!, M ) represents the
— The IF frequency of up/down ramps ( f  / f  ) of a sequence number of the three-segment small period and the
superscript n(n 1, 2,!) represents the sequence number of
moving target.
According to the character of similar triangles in Figure 1, the the big period. f mn denotes the fixed frequency of the m -th
ramp slope of the chirp is three-segment in the n -th big modulated period.
W fs
P (1)
T W f11 f M1 f12 f M2
And the relationship between f  , f  , f d and fs is: f 21 f 22
fs f  fd f  fd (2)
So
­ ( f  f ) Fig. 2 The proposed radar waveform
°° f d 2
® (3)
This design has at least two advantages. First, the transmitted
° f f  f
signals from different radar can be set up uniquely with low
°̄ s 2 cost. The mutual interferences from other radars on the
Once the IF frequencies have been detected, according to the opposite or adjacent lanes can be differentiated. Second,
equation (1) and (3), the distance R to the target and the because the carrier starting frequencies are all f0 and the flat
radial relative velocity v can be calculated using the
frequencies are not equivalent in a certain big period, the
following equation:
Chirp bandwidths and the ramp slopes are also. And then in
­ Tc( f   f  )
°° R
different three-segments the IF frequency caused by the same
4W (4)
® target is different except the flat frequency segments.
°v O ( f   f ) The relationships between transmitted signal, echo signal and
°̄ 4 IF signal for a single moving target in two small periods of
Here O is the wavelength of modulation carrier frequency. proposed waveform are illustrated in Fig. 3. In the first three-
And the relative velocity v for approaching the radar is segment, the IF frequencies corresponding to the up ramp, the
defined as positive. flat frequency segment and the down ramp are f1 , f12
While in a multi-target situation (N targets is assumed), N IF
frequencies can be detected at the up ramp segment and f1 respectively. In the second three-segment, they are
(expressed as f 1 , f 2 ,!, f  N ) and then the same at the down f 2 , f 22 and f 2 . Because of the Chirp bandwidths W1 z W2 ,
ramp segment (expressed as f 1 , f 2 ,!, f  N ) respectively. the IF frequencies are unequal that f1 z f 2 and f1 z f 2 , at
Their matches are ambiguous and there are N 2 kinds of the same time f12 f 22 by the reason of they are equivalent
possible combinations. According to equation (4), each pair to the Doppler frequency of the moving target.
of ( f  i , f  j ) ( i 1, 2,!, N ; j 1, 2,! , N . ) can determine
one target ( Rij , vij ). It is a problem to eliminate the N 2  N
false targets among the N 2 results detected.
To resolving the limitations in multi-targets situations, two
existing approaches are typically proposed. The first means is
based on improved waveform designs[4-7]. The other method
implemented by some new arithmetic algorithms without
changing the system architecture[8] . This paper describes a W
new transmitted waveform and corresponding multi-target
detection algorithm to eliminate the false targets. f1 f1 f 2
f12 f 22 f 2
3 The proposed method
The proposed waveform for FMCW automotive radar is Fig. 3 The relationships between transmitted signal, echo
composed of three segments (up ramps, flat frequencies f mn signal and IF signal for a single moving target in two small
periods of proposed waveform
From equation (4), it can be derived to the relations: performed using equation (4). The others are eliminated as
­ WO 1 false targets.
v R  fO
°° Tc 2 (5) In the ( j  1) -th three-segment period, the processing method
®
°v  W O 1 is similar with the previous period.
R  fO
°̄ Tc 2 After the implementations of up- and down- ramp IF
Equation (5) can be drawn as two lines in a single R-v- frequencies matching in the two continuous three-segment
diagram and the coordinate value of their intersection is the periods, most ghost targets are discarded. A Further step is
distance R and the relative velocity v of the target. Fig. 4 that the distances and relative velocities are compared and

¦ R  R( j 1)G ( n )
2
illustrates the principle of certain true target detection for N
estimated by minimizing and
proposed radar waveform. Combined with the continuous two n 1 jn

¦ v  v( j 1)G ( n ) over all the permutations G (˜) of the


three-segment, estimating the real target by the condition that N 2

all six lines must intersect at a reasonable position. The n 1 jn

detailed process description is as follows. set 1, 2,! , N .


A summary of this algorithm is shown in Fig. 5.
j -th three-segment
( j  1)-th three-segment j -th three-segment ( j  1)-th three-segment
Up ramp segment
Calculate Calculate
cT cT
R j ( f j  f j 2 ) R j 1 ( f ( j 1)  f ( j 1)2 )
Flat frequency segment 2W j 2W j 1
cT cT
R j ( f j  f j 2 ) R j 1 ( f ( j 1)  f ( j 1)2 )
Down ramp segment 2W j 2W j 1

Distance Search matches


Search matches
Fig. 4 The principle of certain target detection for proposed between positive between positive
radar waveform R +j and R j R +j 1 and R j 1

In the j -th three-segment period, the multi-target detection is


based on the fact that Doppler frequencies f j 2 f ( j 1)2 fd Calculate for all targets Calculate for all targets
R j and v j R j 1 and v j 1
are obtained at the flat frequency segment of the modulation,
thus providing the relative velocities of targets. Then, from
the Doppler frequency information it is easy to find all
possible corresponding Doppler-distance pairs on the up and
down ramps. The distance estimation is given by equations (5) Search matches between positive Rj and Rj 1 , vj and vj 1
and (6).
Up ramp and flat frequency segment: Fig. 5 The algorithm of calculating distance and relative
 cT velocity
R j ( f j  f j 2 ) (6)
2W j
Flat frequency and down ramp segment:
cT 3 Simulation results
R j ( f j  f j2 ) (7)
2W j The performance of the proposed algorithm was simulated by
Where W j is the chirp bandwidth of the j -th three-segment. using Matlab. The transmitted bandwidth was 300 MHz, the
carrier starting frequency was 76 GHz, and the modulation
The relative velocities estimation is given by the following period was 1ms. The distances and relative velocities of 10
equations. targets on a road are generated randomly in the range of
O f j2 R  [0, 200] m and v  [200, 200] km/h. And this time
v (8)
2 (R, v) {(163,  137) (181,188) (25,183) (183, 6)
Then, we seek the best possible match between the distances (127,120) (20, 143) (56, 31) (109,166)
estimated on the up ramp and those estimated on the down
(192,117) (193,184)}
¦ R  R jG ( n )
N 2

ramp by minimizing jn
over all the Fig. 6 shows the results of multi-targets detection using the
n 1

permutations G (˜) of the set 1, 2,! , N , where N is the typical triangular waveform method (small green markers)
and the proposed algorithm (big red markers). The horizontal
number of targets. So the association made the relative axis is the range and the vertical pole is the velocity. In
velocity and distance estimates of the targets can be typical triangular waveform method, the number of targets
detected is 58 including 48 ghost targets. In fact, there exist [2] R. H. Rasshofer and K. Gresser, "Automotive Radar and
other 42 false targets out of the above range and they are not Lidar Systems for Next Generation Driver Assistance
shown in the figure. According to the results in the figure, the Functions," Adv. Radio Sci., vol. 3, pp. 205-209, 2005.
miss-detection rate is 83%. All ghost targets are cancelled and [3] A. Bazzi, C. Karnfelt A. Peden, T. Chonavel, P. Galaup
the miss-detection rate is 0 in the proposed algorithm. and F. Bodereau, "Estimation techniques and simulation
platforms for 77 GHz FMCW ACC radars," European
200 Physical Journal, Applied Physics, vol. 57, p. 11001 (16
pp.), 2012.
150
[4] H. Rohling and C. Moller, "Radar waveform for
100
automotive radar systems and applications," in Radar
Conference, 2008. RADAR '08. IEEE, 2008, pp. 1-4.
50 [5] J. Zhang, G. Liu, H. Gu and W. Sun, "Stepped-FMCW
waveform applied for MM-wave automotive collision
v(km/h)

0 warning radar," Journal of Infrared and Millimeter


Waves, vol. 19, pp. 413-18, 2000.
-50 [6] H. Eugin, O. Woojin and L. Jong-Hun, "Multi-target
detection algorithm for FMCW radar," in 2012 IEEE
-100
Radar Conference: Ubiquitous Radar, RADARCON
2012, May 7, 2012 - May 11, 2012, Atlanta, GA, United
-150
states, 2012, pp. 0338-0341.
-200
[7] T. Xu, C. Jin, X. Sun and G. Xia, "A Novel Method to
0 20 40 60 80 100 120 140 160 180 200 Identify Multitarget by FMCW Radar". Acta Electronica
R(m)
Sinica, 2002(6)
Fig. 6 The simulation results of multi-targets detection
[8] L. Shi and L. Zhang, "The signal processing based on
the partner for frequency modulation continuous wave
radars", Journal of XiDian University, 2003, 30(8):534-
4 Conclusions 538ˈ2003
Since transmitted waveform is critical to the viability of
automotive radars, and high false alarm rate is a problem that
must be solved before the automotive radar can be marketed
as a commercial product. In this paper, a novel waveform
and multi-target detection algorithm applied for FMCW
automotive radar is proposed. The waveform is composed of
three segments: up ramp, flat frequency and down ramp in
every small period. Especially, the flat frequencies are
modulated by a unique PN code sequence within a
determinate frequency range. The unambiguous relative
velocities detected in the flat frequency segment can cancel
some false targets initially. Corresponding to the waveform,
which combines the advantages of both FSK and FMCW
formats, a progressive judgment algorithm is used in the
continuous small period to further eliminate the false targets.
And improvements have been made in multi-target distinction
performance, together with the enhancement of target
recognition performance.
The performance simulation including the target SNR will be
carried out in the next step.

Acknowledgements
This work was supported by Guangdong Province and
Chinese Ministry of Education project funding with the
combination of industries, universities and institutes.

References
[1] F. Gustafsson, "Automotive safety systems," Signal
Processing Magazine, IEEE, vol. 26, pp. 32-47, 2009.