A Signal Processing Algorithm for Interference Suppression of Automotive Radar Using an Auxiliary Receiver
Article information
Abstract
In this work, we propose an automotive radar system equipped with an add-on receiver (RX) capable of sensing and suppressing interference signals. The add-on RX is designed to filter out target signals, allowing only signals from interferers to exist in the add-on RX. By setting the add-on RX’s local oscillator (LO) frequency higher than the original RX’s LO, the intermediate frequency of the interference signals in the RXs occurs in two separate time zones, enabling their identification and cancelation in the time domain. In both RXs, burst-like interference signals are searched for, and their phase information is acquired through Hilbert transform. Subsequently, the interference signal identified in the add-on RX is phase- and time-shifted using a phase recovery and time-shifting algorithm. The compensated interference signal is then subtracted from the signal in the original RX, leaving only the target signals in the original RX. Simulation results show that the interference signal is successfully suppressed, and a hidden target signal is recovered.
I. Introduction
Interference issues in automotive radars, which are widely used in vehicles worldwide, have gained increasing attention over time [1–11]. When a vehicle with a front-facing radar transmitting high output power approaches a victim radar, the latter may face several problems including missed targets or false alarms.
In frequency-modulated continuous-wave (FMCW) radars, an interfering signal with the same chirp waveform and chirp time as that of the victim radar is unlikely. However, interference takes place when the interfering signal enters the receiving frequency band of the victim radar.
Among the numerous interference mitigating techniques proposed and demonstrated in previous studies, most have relied on various forms of time domain signal processing to identify an interfering signal that surpasses a certain threshold level [3, 4, 7]. Notably, finding the appropriate threshold level often involves an iterative procedure that requires complex signal processing and algorithms. Some of the most well-known interference suppression techniques are as follows:
1) Zeroing is the simplest and most well-known method [4]. As the name implies, this method removes intermediate frequency (IF) signal samples in the time domain that are contaminated by interference. In other words, any signal beyond a certain threshold level is zeroed. However, simply zeroing samples cause phase discontinuity in the signal, resulting in a broadening IF spectrum and, in turn, high side lobes.
2) A raised cosine window can be applied to locations in which samples are disturbed by interference [5, 7]. It helps smoothen signal discontinuity and reduces the ringing effects caused by fast Fourier transform [5]. An interfered signal region can be identified by converting the received signal into an image matrix and then applying the maximally stable extremal regions (MSER) algorithm [5].
3) Extrapolation in the short-time Fourier transform (STFT) domain involves multiple steps [9]—zeroing the interference regions in the frequency domain, estimating forward/backward extrapolation coefficients, extrapolating the signal to reconstruct in the disturbed region, and interpolating the estimated phase. Unlike signal processing in the time domain, removing interference regions in the frequency domain allows for the preservation of noise and clutter information.
These techniques are based on innovative but heavily loaded algorithms. In this paper, instead of relying only on signal processing approaches, we propose an automotive radar system equipped with an additional receiver in the radar transceiver to detect and suppress interference. The interfering signal is identified and extracted from the additional receiver and then subtracted from the main receiver signal in the time domain, thereby leaving only the target signal in the main receiver.
This article is organized as follows: Section II presents the proposed interference mitigation approach based on an add-on receiver, Section III presents the simulation results and compares them to those obtained using conventional techniques, and Section IV presents the conclusions of this study.
II. Radar System with an Add-On Receiver
1. FMCW Radar Architecture
The proposed FMCW radar system comprises a single-channel add-on receiver (RX) placed next to the main receiver, as shown in Fig. 1. The additional receiver is driven by an FMCW waveform generator, which supplies a higher operating frequency than to the main receiver. In particular, the offset frequency was set to be more than two times higher than the bandwidth of the low-pass filters (LPF). When a target signal and an interfering signal are received by both RXs, two IF signals emerge from each RX—IF1 in RX1 (the original RX) and IF2 in RX2 (the add-on RX). After these two IF signals are filtered by each LPF, IF2 carries the interference signals, but not the target signals, while IF1 carries both the interference and the target signals. The down-converted interference signal at RX1 is expected to be similar to IF2, except that it would be time-shifted and phase-rotated.
Radar system with an add-on receiver.
In this study, RX1 and RX2 are built on separate boards so that a phase difference between IF1 and IF2 cannot be avoided. The phase recovery block in Fig. 1 is responsible for correcting this difference, while the time shift block is meant to match the time difference between IF1 and IF2. The IF2 signal would then be subtracted from IF1, thus ultimately providing the cleaned-up target signal in IF1 with the interference rejected.
2. Detection and Suppression Principles
Fig. 2 illustrates the relationship between the received signals in RX1 and RX2 in terms of the offset local oscillator (LO) frequency of the add-on RX (RX2) and the time shift between the two signals. It is evident that the FMCW signal of the add-on mixer (blue) has the same chirp slope as the original TX LO (green) but with the frequency offset added to it.
Relation between the LOs of the original and the add-on RX in the presence of an interference.
When an FMCW interfering signal (red) attacks a victim radar, it crosses the chirp signal of the victim radar receiver and that of the add-on RX at two intersections. Notably, to ensure that no target signal existed within the IF2 bandwidth, its offset amount was set to more than two times the bandwidth of the LPF, as depicted in Fig. 2.
Furthermore, the time shift was determined by the frequency slope and frequency offset. As shown in Fig. 2, two intersections take place at two different time zones—one after the other—depending on the slope of the interference signal. Later, one of the IF signals would have to be time-shifted by τ.
Fig. 3 presents a block diagram of the interference suppression mechanism, showing how phase recovery and time shifting are implemented in the proposed algorithm. Notably, phase information of the interference signal in the original and the add-on RX can be acquired using IQ channels or the Hilbert transform of the real signal. The latter is depicted in Fig. 3, where φ1 and φ2 signify the phases of the original and the add-on RX, respectively.
Interference signal suppression algorithm.
The interference suppression process using the Hilbert transform follows four steps: 1) acquiring phase information using the Hilbert transform, 2) correcting phase discrepancy, 3) shifting the time samples, and 4) subtracting the add-on signal from the original signal. In this study, the original and add-on real-time signals are Hilbert transformed, and then their peaks are searched. Following this, the phase information of the two peaks and the time delay between them are identified. For the original RX, the peak signal is assumed to be arising from only the interfering signal since it exhibits in the time domain a burst-like signal which is much stronger than a weak target signal. After acquiring the phase information, φ1 and φ2, the phase of the add-on signal is corrected by shifting the signal by φ2 − φ1. Next, the time shift, τ, between the two peak signals is calculated, as noted below.
Considering that si and so are the chirp slopes of the interference and the original TX, respectively, and the duration of the interference, ti, is known, si can be obtained from the known parameters fLPF, ti, and so depicted in Fig. 2, using the following equation:
Furthermore, the time shift required for the add-on signal can be calculated as follows:
The analog-to-digital converter (ADC) samples of the add-on signals are shifted accordingly.
The phase-recovered and time-shifted interference signal in the add-on RX should then be subtracted from the signal in the original RX to obtain the IF signal with the interference suppressed.
III. Simulation
This section presents the simulation results obtained using MATLAB’s Simulink to validate the proposed interference mitigation technique. The verification was conducted through simulations in environments where 1) there was no target and only interference existed, and 2) there were two targets with an interference present. In both scenarios, the interference signal received by the original RX was removed through time- and phase-shifting the interference signal received by the add-on RX.
1. Simulation without a Target
In the case of the environment where there was no target, only the interference signal existed in the time domain in IF1 as well as in IF2. Fig. 4(a) shows the time delay and phase change of the interfering signal in the time domain when the original TX chirp slope is 2 MHz/μs, the interferer chirp slope is 4 MHz/μs, the LPF bandwidth is 2 MHz, and the offset frequency of the add-on LO is 10 MHz. Since the original RX signal, IF1, is a real signal (black), its imaginary part (gray) is zero. The add-on RX signal, IF2, was phase rotated by 100° and time shifted by 5 μs. After Hilbert transform was performed on IF2, as shown in Fig. 4(a), the real part of IF2 (red) was different from IF1 (black), and its imaginary part (blue) was non-zero. Furthermore, as the phase discrepancy between the two increased, the mismatch between the waveforms increased along with it.
Simulation results: (a) interference signals in the original RX and add-on RX in the time domain, and (b) phase recovering and time shifting of the interference.
Fig. 4(b) illustrates the process of recovering the interference signal received by the original RX using the add-on RX signal. The interference signal in the add-on RX was first phase compensated (blue) and then time shifted (red dotted). Subsequently, the interference signals in both RXs were matched, after which the recovered interference signal was subtracted from the original RX signal.
2. Simulation with a Target
With regard to both target signals and an interference signal received by the original RX, the proposed algorithm was tested and verified using the following example. It was assumed that a target is located at a distance of 22.5 m and a second target is stationed at a distance of 45 m. Furthermore, the FMCW radar was assumed to have a 200 MHz RF bandwidth, generating a signal ranging from 77 to 77.2 GHz. The chirp duration was set to be 100 μs. The add-on RX LO was set to be 2 MHz more than the LO of the original RX. The interfering signal was assumed to have a bandwidth of 200 MHz, but with a 50 μs chirp time. The round-trip time for the first target signal was 0.15 μs (accordingly, time samples of the number of 60 were considered for our setup), while the interfering signal was set to be received after 10,000 samples.
Fig. 5(a) shows the time signal before and after the interference is rejected. Notably, the IF signal in the original RX (green) is composed of both the target and interfering signals. The interfering signal present in the add-on RX (blue dot) was phase recovered, time shifted, and subtracted from the original RX signal using the algorithm implemented via Simulink. As a result, the IF signal in the original RX (red) attained the time signal with the interference rejected. Zooming into the interfering time zone, it was noted that the target signal obtained after the rejection algorithm is applied is slightly deformed, not purely sinusoidal. Fig. 5(b) depicts the power spectrum before and after the interference is rejected for the two target scenarios, along with some noise present. When no interference signal is present, the original RX signal (black circle) clearly displays both targets, with a minimum signal-to-noise ratio (SNR) of 25 dB. However, in the presence of interference, while the original RX signal (green) clearly identifies the first target, the second target is almost hidden by interference. This is because the interference increased the noise floor by about 30 dB, making the detection of the target located at 45 m difficult due to low SNR. Notably, two specific factors pertaining to the interfering signal captured by the add-on RX (blue dot) were observed: 1) the interference spectrum appeared as increased noise in the frequency band lower than 2 MHz, and 2) the 2 MHz frequency offset employed in the add-on RX pushed the two target signals beyond the 2 MHz range and filtered them out. After the proposed algorithm was applied to the original RX signal, the two targets were clearly observed (red), along with an SNR similar to the level observed in the absence of any interference.
Before and after interference suppression: (a) time domain results and (b) power spectrum domain results of the two targets.
In this context, it should be mentioned that this methodology is not without weaknesses, such as the increased complexity involved in generating the LO using the offset frequency, implementing the add-on RX, and recovering the phase discrepancy, all of which are necessary to emulate the interfering signal captured in the original RX. Nonetheless, we believe that the proposed technique also offers several advantages. First, it highlights that the detection of any interference can be achieved quickly when assisted by hardware, since it prevents confusing the interference signal with the target signals, which can therefore be separately retrieved. In contrast, algorithm-only techniques require continuous monitoring of the time signal to track whether it meets a threshold. Second, for the same reason, identifying the presence or absence of an interfering signal can be achieved both easily and quickly.
3. Comparison with Other Techniques
A series of simulations were conducted to compare the results of the proposed method with those of the other techniques mentioned in Section I. Fig. 6 shows the results of the IF power spectrum after the suppression of interference using zeroing, MSER with inverse cosine windowing, and STFT with extrapolation of signals. In the presence of an interference signal (black dot), applying the zeroing technique (red) resulted in a reduction in the noise level by approximately 5 dBm. Meanwhile, when applying MSER with a cosine window (orange), a 15 dBm reduction in noise level was observed. When applying STFT with extrapolation (green) and the algorithm proposed in this paper (blue), a reduction in noise level of up to 25 dBm was attained. This reduction ensured sufficient SNR for target range estimation, suggesting that the proposed algorithm achieves nearly the same performance as the STFT and extrapolation technique but involves a lower computational load.
Comparison of the interference suppression results and performance of the four techniques.
IV. Conclusion
In this paper, we propose a hardware-assisted signal-processing algorithm for mitigating interference using an add-on RX. Since only the interference signal exists in the add-on RX, it can be easily identified and rejected in the original RX through phase recovery and time-shift processing. The proposed interference mitigation algorithm was verified through simulations. However, this technique presents the limitation of increased complexity of the radar hardware as a result of the use of an add-on RX module. One way to significantly reduce this complexity would be to integrate the add-on mixer into the original RX. Overall, the proposed radar is expected to successfully detect a target that is hidden due to interference by enhancing the target’s SNR.
References
Biography
Sanghyun Lee, https://orcid.org/0009-0003-3149-252X received his B.S. degree in mechanical engineering from Dankook University, Yongin, Korea, in 2015. He is currently pursuing his M.S. degree in semiconductor and display engineering at Sungkyunkwan University. He is also working as an engineer in the Department of Device Solution, Samsung Electronics, Pyeongtaek, Korea. His research interests include radar signal processing and FMCW radar systems.
Jaehyun Park, https://orcid.org/0000-0002-8303-6273 received his B.S. degree in semiconductor systems engineering from Sungkyunkwan University, Suwon, Korea, in 2016, and is currently working towards his Ph.D. degree in electronic, electrical, and computer engineering at Sungkyunkwan University. His current research interest is millimeter-wave CMOS integrated circuit design for automotive radar systems.
Mingeon Shin, https://orcid.org/0000-0002-6715-5416 was born in Jeonju, South Korea, in 1990. He received his B.S. degree in electronic engineering from Jeonbuk National University, Jeonju, South Korea, in 2015. He is currently pursuing his M.S. degree in electronic, electrical, and computer engineering at Sungkyunkwan University. He is also working as a researcher at the Department of SoC Platform Research Center, Korea Electronics Technology Institute, Seongnam, South Korea. His research interests include radar signal processing, FMCW radar systems, sensor technology, motion recognition, and vital monitoring.
Byungsung Kim, https://orcid.org/0000-0003-3084-6499 received his B.S., M.S., and Ph.D. degrees in electronic engineering from Seoul National University, Seoul, Korea, in 1989, 1991, and 1997, respectively. In 1997, he joined the College of Information and Communication Engineering, Sungkyunkwan University, Suwon, Korea, where he is currently a professor. He was a visiting researcher at the University of California at Santa Barbara in 2013. His research interests include high-frequency device modeling and RF/millimeter-wave CMOS integrated circuit design.
Reem Song, https://orcid.org/0000-0002-7088-1777 received her Ph.D. degree in electrical engineering from the University of Southern California, Los Angeles, California, in 2006. From 2007 to 2010, she worked as a senior design engineer at Skyworks Solutions Inc. in Thousand Oaks, California. Since 2014, she has been working as a research faculty member at Sungkyunkwan University, Suwon, Korea, performing research on millimeter-wave circuits, antennas, and systems.