You are on page 1of 5

An Anti-collision Automotive FMCW Radar

Using Time-domain Interference Detection and Suppression


Takuya Nozawa† Yuya Makino† Nobuyuki Takaya† Masahiro Umehira† Shigeki Takeda†
Xiaoyan Wang† Hiroshi Kuroda‡
†Graduate School of Science and Engineering ‡Hitachi Automotive Systems, Ltd.
Ibaraki University 4-12-1 Nakanarusawa, Hitachi, 2520 Takaba, Hitachinaka,
Ibaraki 316-8511 Japan Ibaraki 312-8503 Japan
Email: †{16nm668n, 13t8045h, 13t8053n, masahiro.umehira.dr, shigeki.takeda.tmkyo, xiaoyan.wang.shawn}
@vc.ibaraki.ac.jp
‡hiroshi.kuroda.gy@hitachi-automotive.co.jp

Keywords: Automotive Radar, Frequency Modulated frequency range of FMCW radar was proposed in [5], which
Continuous Wave (FMCW), Interference Suppression, Time- can avoid ghost target detection however still suffers from
Domain significant SNR degradation due to inter-radar interference.
This paper proposes a time-domain interference detection
Abstract and suppression scheme for FMCW radar systems.
Specifically, phase noise effect is firstly removed thus
As the number of automotive radars in vehicles will rapidly interference detection can be made by using a simple
increase for self-driving car applications in near future, threshold based approach. Furthermore, time domain
mutual interference between radars would result in frequent interference suppression is applied to improve SNR in
false and miss detection of the target due to dense deployment interference environment. The proposed scheme post-
of radars. This paper focuses on interference between processes the mixer output signal, and thus could be used in
frequency modulated continuous wave (FMCW) radars, and any existing FMCW radar system by adding signal processing
proposes a time domain interference detection and software. To validate the performance improvement of the
suppression scheme to achieve anti-collision capability. We proposed scheme, we conducted extensive experiments by
conducted real-world experiments by using 77.5GHz FMCW using 77.5 GHz FMCW radars. The results show that the
radar to validate the proposed scheme. The experimental proposed scheme could improve the target detection rate in
results show that the proposed scheme can achieve almost the interference environment by approximately 30% compared to
same target detection rate as in interference-free environment, the baseline scheme.
even in interference environment.
2 Effect of inter-FMCW radar interference
1 Introduction
In severe inter-FMCW radar interference environment,
The autonomous vehicle has been getting a great deal of reflected signal from the target and the radar signal from other
interests and considerable research investment recently. A vehicles are received simultaneously, which results in false
self-driving car must be able to see what is in front, behind and miss detection of the target. Based on the relationship of
and both sides, in other words, it needs a 360 degree view, the chirp slopes between the transmitted signal and
just like a human driver. Compared to laser radar, infrared interference signal, the FMCW radar interference could be
sensor, camera, etc., FMCW radar is one of the most classified into two categories, i.e., narrow band interference
attractive solutions, since it can measure the target’s relative and wide band interference [2].
distance and speed simultaneously, and has the advantage of In the narrow band interference case, the interference
lower cost and higher performance in dark or harsh comes with a small-time delay such that the chirp of the
environments. As the number of vehicles equipped with interference is in-phase with the chirp of the desired signal.
automotive radars is expected to increase in near future, As shown in Fig. 1, all the interference beat frequency leaks
mutual interference has become a crucial issue [1], [2], [3]. into the pass-band region of low pass filter (LPF) in FMCW
Specifically, it leads to the generation of ghost target, as well radar. This narrow band interference causes ghost target, i.e.
as the signal to noise ratio (SNR) degradation and thus the false detection of the target, and is hard to remove. However,
target detection failure. To mitigate the inter-radar the probability of this narrow interference is extremely low
interference, various techniques have been proposed for [1], and thus we do not tackle it in this paper and this problem
automotive radar systems [4], [5]. An interference is remained for future study.
suppression scheme using clipping and weighted-envelope In the wide band interference case, the frequency of desired
normalization was proposed in [4]. However, the effect of signal increases while the one of interference signal decreases,
phase noise on FMCW radar is not considered, which needs or vice versa. As illustrated in Fig. 2, the beat frequency lies
to be pre-processed before interference detection. Another in both in-band and out-of-band regions. In this case, the out-
anti-interference technique using randomized chirp rate and of-band beat frequency can be cut by LPF, and the remaining

1
interference leaks into pass-band of LPF and results in
impulse-like interference signal. Usually, the received level of
the interfering radar signal is much higher than the reflected
signal from the target when the interferer is located near the
desired FMCW radar. This interference signal can be
regarded as an impulse signal in time domain. The received
signal of FMCW radar is converted into frequency domain
signal by FFT (Fast Fourier Transform) to detect the peak
frequency, i.e. distance from the target. As impulse is
equivalent to white noise in frequency domain, the wide band
interference leads to miss detection of the target due to the
increase of noise level in frequency domain.

3 Proposed time-domain interference detection Fig. 1. Narrow band interference in FMCW radar.
and suppression method
In this section, we present the proposed FMCW radar
interference suppression method. Figure 3 shows the block
diagram of the proposed FMCW radar system. Principle of
the proposed interference suppression method is to detect the
timing of the impulse noise caused by the interfering radar
signal and to suppress it in time domain before FFT
processing in the FMCW radar. However, we found that DC
component drift appears at the mixer output of FMCW radar
because of large phase noise of the high frequency oscillator
used in FMCW radar. This DC component drift should be
removed before setting the threshold for impulse noise
Fig. 2. Wide band interference in FMCW radar.
detection. Therefore, firstly, we propose a phase noise
suppression method to remove the DC component drift at the
mixer output. Then we propose a time domain interference
suppression method which detects the position of the impulse
noise caused by the interference and suppresses it in the mixer
output signals. Finally, we conduct FFT and detect the target
from the beat frequency peak power in frequency domain.
As shown in Fig. 3, we assume that triangle chirp signal is
transmitted from VCO and the received signal reflected from
the target is mixed with the output of VCO to obtain the beat
signals. Low pass filtering and A/D conversion is performed, Fig. 3. Block diagram of the proposed FMCW radar system
then all the signal processing i.e., removing DC component 4000
w/o phase noise suppression
drift, interference pulse noise detection and suppression, FFT with pahse noise suppression
3000
and target detection, is performed in the digital signal
processor of the FMCW radar. 2000
ADC data

Impulse noise
3.1 Phase noise suppression method 1000

Based on our experiment (details are presented in Section 0

4), we found that the DC component of the mixer output


-1000
varies in time due to large phase noise in a millimetre wave
oscillator using PLL (Phase Locked Loop) technique. An -2000
example of actual measured mixer output is illustrated as the 0.028 0.0285 0.029 0.0295
time[s]
0.03 0.0305 0.031

red line in Fig. 4 (a), when the target signal and the interfering (a) Mixer output
signal are simultaneously received. Impulse noise caused by 200
the interfering radar appears around the time of 0.0294(s) and 100
ADC data

the received level of the beat signals is much lower than the 0
impulse noise as well as DC drift component. Therefore, DC
component drift needs to be removed first to set the threshold -100

to detect the impulse signal due to interference. -200


0.028 0.0285 0.029 0.0295 0.03 0.0305 0.031
To solve this problem, we propose a phase noise time[s]
suppression method which removes the DC component from (b) Enlarged view with phase noise suppression
Fig. 4. Mixer outputs with phase noise suppression

2
the mixer output by using LPF, and subtracts the DC 60
component from the mixer output signal. Here, the LPF is
implemented by using simple moving average. An example of 55
the mixer outputs after removing phase noise by the proposed
method is shown as the blue line in Fig. 4(a) and (b). Peak-to 50

–peak amplitude of the desired beat signals is as small as 40

variance
while that of impulse noise is as large as 300, i.e. about 20dB 45

higher. Amplitude of DC component drift is as large as 5000


40
and is much higher than the desired beat signals and the
impulse noise. By using the proposed method, it is obvious
35
that the phase noise is perfectly removed and the detection
and suppression of the interference pulse becomes possible by 30
setting an appropriate threshold. 4 8 16 32 64 128 256 512
number of moving average
It is necessary to optimize the number of moving average
Fig. 5. Spectrum variation according to number of moving average.
in the proposed method. As larger moving average number
results in lower power level in low frequency region and where k is threshold parameter. The threshold, mthreshold is used
smaller moving average number results in higher power level to detect impulse noise where reference amplitude is mave
in low frequency region in its power spectrum, the variation given by equation (1). There is a trade-off to set the parameter,
caused by phase noise should be minimized. Therefore, we k. Specifically, small k leads to false-detection of interference
performed parameter optimization by using actual measured impulse noise signal, thus the desired beat signal is also
mixer output to find the number of moving average suppressed and miss-detection rate increases since the desired
processing which minimizes the variation of power spectrum signal can be recognized as interference impulse signal.
after FFT processing. Figure 5 shows FFT signal variation Meanwhile, large k leads to miss-detection of interference
according to the number of moving average. As shown in Fig. impulse noise signal, thus noise level cannot be suppressed.
5, number of moving average of 32 gives minimized spectrum
variation. Therefore, number of moving average of 32 is used STEP 3: Impulse noise detection
in our FMCW radar experiments. Note that the optimized Find the position of impulse noise, Ni where the level of
number of moving average might be different in another impulse noise signal is higher than mthreshold, and calculate a
FMCW radar. new signal, msup(n) that is weighted by window function, w(n)
as follows:
3.2 Interference detection and suppression method
msup (n)  m(n)  w(n) . (3)
After DC components drift caused by the phase noise is
removed, impulse noise caused by the interfering FMCW Though there are many alternatives for w(n), we employ
radar is detected and suppressed in the proposed scheme. This raised cosine weighting function, wR(n) and zero weighting
paper proposes a time-domain interference detection and function, w0(n), which are given below:
suppression method by detecting the interference impulse (1  cos(2 (n  N i / M )) / 2 (| n  N i | M / 2) (4)
signal position and suppressing it using window function in wR ( N i )  
 1 (else)
time domain. The detailed algorithm is described as follows.
w0 ( N i )  0 . (5)
STEP1: Received signal level detection
Calculate the averaged absolute value, mave of the mixer Raised cosine function has a design parameter, M, which is
output, m(n) (n=1~N) after suppressing DC component drift the width of raised cosine function. M is related to the
and mave is given as: interference pulse signal duration. Large M leads to beat
signal suppression resulting in decrease of desired signal
N
mave  1 / N  | m(n ) | , (1) power, and small M leads to less suppression of impulse noise
n 1 signal resulting in increase of noise power.
where N is the number of mixer output signals sampled by
3.3 Design of threshold parameter, k
A/D converter. As the received signal level can vary
according to the distance from the target, we need a reference Threshold parameter, k is used for impulse noise detection
level to set the threshold for impulse noise detection. This and its design is a trade-off issue as described before. Figure 6
method is simple but effective to calculate the received signal shows simulation results of signal power at the beat frequency
level for threshold setting to detect impulse noise caused by of the target and averaged noise power according to the
interfering radar. parameter, k where the experimental data of the FMCW radar
is used. As shown here, the signal power drastically decreases
STEP2: Threshold setting for impulse noise detection when k is smaller than 5 since the desired beat frequency
Set the threshold mthreshold by the following equation (2): signal can be effectively suppressed. On the other hand, when
k is larger than 20, the noise power level increases since
mthreshold  mave  k , (2)
impulse noise is not suppressed. Therefore, the threshold

3
10 Table. 1. Major experiment parameters.
Parameter value unit
5
Sweep frequency 560 MHz
0 Center frequency 76.5 GHz
Sweep time 3 ms
Relative Power[dB]

-5
Sampling rate 781.25 kHz
-10 FFT size 2048
FFT window blackman
-15 SNR threshold 15 dB
-20

-25

-30
0 10 20 30 40 50
Threshold parameter, k
Fig. 6. Signal and noise power level according to threshold
parameter, k. Fig. 8. Exprimental set-up.
3

2 200

ADC data
1
0
Relative Power[dB]

-1 -200
0.0295 0.03 0.0305 0.031 0.0315 0.032
-2 time[s]
(a) w/o interference suppression
-3
50
ADC data

-4
0
-5
2 4 16 32 64 128 256 512 1024
raised cosine window width, M
-50
Fig. 7. Signal power level according to raised cosine window width, 0.0295 0.03 0.0305 0.031 0.0315 0.032
M. time[s]
(b) with interference suppression
parameter, k should be set in the range from 5 to 20. Fig. 9. Examples of mixer output without and with interference
suppression.
Considering the margin of threshold level, the parameter, k
is set at 10 in this paper.

3.4 Design of window width, M


As shown in equation (4), the width of raised cosine
function, M, is a design parameter. M is related to the
interference pulse signal width. Large M can remove relevant
impulse noise caused by the interfering radar, but also reduce
desired signal power. Figure 7 shows signal power according
to raised cosine window width, M. As a design trade-off,
M=32 is chosen since no significant decrease of signal power
Fig. 10. Mixer output in frequency domain.
is observed if M is less than or equal to 32.
experimental FMCW radar are summarized in Table 1.
4 Performance evaluation of the proposed anti- Figure 9 shows an example of mixer output without and
collision FMCW radar with the proposed interference suppression method. We can
observe that large interference impulse noise is successfully
The proposed anti-collision FMCW radar is evaluated by suppressed by the proposed method. Figure 10 shows the
real-world experiments to demonstrate the anti-collision power spectrum of these two mixer signals in frequency
capability by suppressing the interference impulse noise in domain. It is clear that the noise level is reduced by using the
time domain. We used two FMCW radars as the observing proposed method.
radar and the interfering radar and both of them have the Figure 11 compares the target detection rate of FMCW
same parameters such as chirp rate and center frequency. We radar without interference suppression in non-interference
used a reflector target which has the same RCS (Radar Cross condition and interference condition as described above
Section) of a mid-size car. Figure 8 shows the experimental where 50 trials were conducted. It also compares the target
setup where distance between the observing radar and target detection rate of FMCW radar using the proposed interference
is set to 15 and 25 [m], and that between the observing radar suppression with raised cosine weighting function and zero
and the interference radar is 5 [m]. Major parameters of the

4
weighting function in interference condition. As shown here,
the proposed method improves the target detection rate from
0.5~0.6 to 0.75~0.8, which are almost the same as that in non-
interference condition. No significant difference between
raised cosine weighting function and zero weighting function
was observed.

5 Conclusion
This paper proposed an anti-collision automotive FMCW
radar using time-domain interference detection and
suppression, which consists of a phase noise suppression
method to remove the DC component, and an interference
suppression method to detect and suppress the interference Fig. 11. Evaluated results of target detection rate
impulse signal in time domain. We conducted design
optimization regarding the design parameters such as the
number of moving average, threshold parameter, k, and
window parameter, M. We also conducted extensive
experiments and confirmed that the proposed method can
improve the target detection rate by approximately 30%
compared to the baseline scheme.
Future work includes further performance evaluation in
multiple interference scenario since more interference radars
around a car are expected in actual automotive scenarios.

Acknowledgements
This research and development work was supported by the
MIC/SCOPE # 175003004. The authors also appreciate
continuous support for this work by Hitachi Automotive
Systems, Ltd, Japan.

References
[1] G.M. Brooker, “Mutual interference of millimeter-wave radar
systems,” IEEE Trans. Electromagn. Compat., vol.49, no.1,
pp.170-181, 20007
[2] M. Goppelt, H.–L. Blöcher, W. Menzel, ”Automotive radar –
investigation of mutual interference mechanisms”, Advances in
Radio Science., vol.8, 2010-1, pp.55-61
[3] Li Mu, Tong Xiangqian, Shen Ming, Yin Jun, “Research on Key
Technologies for Collision Avoidance Automotive Radar”, IEEE
Intelligent Vehicles Symposium, June 2009, pp.233-236
[4] Jung-Hwan Choi, Han-Byul Lee, Jiwon Choi, Seong-Cheol Kim,
“Mutual Interference Suppression Using Clipping and Weighted-
Envelope Normalization for Automotive FMCW Radar
Systems”, IEICE TRANS. COMMUN., vol.E99-B, No.1 2016-1,
pp.280-287
[5] Tang-Nian Luo, Chi-Hung Evelyn Wu, and Yi-Jan Emery Chen,
“A 77-GHz CMOS Automotive Radar Transceiver With Anti-
Interference Function”, IEEE TRANSACTIONS ON CIRCUITS
AND SYSTEMS., vol.80, No.12, 2013-12 pp.3247-3255.

You might also like