IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 3, September 2024, pp. 2854~2861

ISSN: 2252-8938, DOI: 10.11591/ ijai.v 13.i3.pp2854-2861  2854

Autonomous radar interference detection and mitigation using

neural network and signal decomposition

Dayat Kurniawan1 , Budiman Putra Asmaur Rohman1 , Ratna Indrawijaya1 , Chaeriah Ali Wael 1,2 ,
Suyoto1 , Purwoko Adhi 1 , Iman Firmansyah1
1
Research Center for T elecommunication, National Research and Innovation Agency (BRIN), Bandung, Indonesia
2
Institut d’Électronique de Microélectronique et de Nanotechnologie (IEMN), Université Polytec hnique Hauts-de-France (UPHF),
Valenciennes, France

Article Info ABSTRACT

Article history: Autonomous radar interference is a challenging problem in autonomous
vehicle systems. Interference signals can decrease the signal-to-interference-
Received Aug 19, 2023 noise ratio (SINR), and this condition decreases the performance detection
Revised Nov 6, 2023 of autonomous radar. This paper exploits a neural network (NN) and signal
Accepted Dec 2, 2023 decomposition to detect and mitigate radar interference in autonomous
vehicle applications. A NN with four inputs, one hidden layer, and one
output is trained with various signal-to-noise ratio (SNR), interference radar
Keywords: bandwidth, and sweep time of autonomous radar. Four inputs of NN
represent SNR, mean, total harmonic distortion (THD), and root means
Autonomous radar square (RM S) of the received radar signal. Variational mode decomposition
Detection (VM D) and zeroing based on a constant false alarm rate (CFAR-Z) are used
Interference to mitigate radar interference. VM D algorithm is applied to decompose
Neural network interference signals into multi-frequency sub-band. As a result, the proposed
Signal decomposition NN can detect radar interference, and NN-VM D-CFAR-Z can increase
SINR up to 2 dB higher than the NN-CFAR-Z algorithm.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Dayat Kurniawan
Research Center for Telecommunication, National Research and Innovation Agency (BRIN)
KST Samaun Samadikun, 1st Tower, 4th floor, Sangkuriang St., Bandung, West Java, Indonesia
Email: [email protected]

1. INTRODUCTION
The research and development in intelligence transport systems (ITS) are still growing up until now,
such as research in autonomous vehicles, vehicles to vehicles (V2V), and vehicle to infrastructure (V2X)
[1]–[5]. An autonomous vehicle needs more sensors, such as ultrasound, light detection and ranging
(LiDAR), camera, and radar (radio detection and ranging). mmWave radar sensor with frequency-modulated
continuous wave (FMCW) is commonly used in autonomous vehicles [6]. FMCW radar promises high
resolution in short-range detection, needs less power, and better performance in various conditions such as
foggy, rainy, and dark environments than others.
However, implementing FMCW in dense autonomous vehicle cause serious signal interference.
Radar signal interference can decrease radar target detection since radar signal interference decreases
signal-to-interference-noise ratio (SINR), resulting in false detection or miss detection [7]. An adaptive noise
canceller (ANC) with a conventional threshold has been introduced to mitigate radar interference [8]. The
performance mitigation depends on the threshold value. If the signal power of the interference signal is lower
than the desired threshold, the interference signal is not filtered. Wavelet denoising and constant false alarm
rate (CFAR) are also explored to mitigate radar interference [9]–[11]. Both methods can suppress
interference signals with high complexity processing and need adaptive threshold. Signal decomposition was

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938  2855

also introduced to mitigate radar interference [12]–[16]. Signal decomposition extracts interference signals
into several sub-band signals in the frequency spectrum. Frequency hopping mitigates radar interference [17]
by changing frequency with a specific pattern based on time. This method needs a complex receiver to
synchronise the transmitted and received signal.
In recent years, artificial intelligence (AI) has been widely used in many applications, with no
exception in autonomous vehicles. Neural or deep neural networks (DNN) have also been explored to
mitigate radar signal interference. DNN applied in radar doppler matrix (RDM) to reduce signal interference
[18]–[21]. RDM needs more resources and processing time. It is not suitable to be implemented in
low-resource modules. Other methods to mitigate interference in synchronous and asynchronous interference
are also explored in [22], [23]. Meanwhile, an adaptive threshold was also introduced, but still a high -
complexity process [24], [25].
This paper proposed a simple neural network (NN) to detect radar signal interference based on
feature signal interference such as mean, root mean square (RMS), signal-to-noise ratio (SNR), and total
harmonic distortion (THD). Variational mode decomposition (VMD) is introduced to extract signal
interference to multiple frequency sub-bands. The next step is the zeroing process based on CFAR
implemented on each sub-band frequency of VMD output to suppress interference signal.

2. METHOD
We proposed a method to increase SINR consisting of detection and mitigation. A NN is used to
detect signal interference and joined VMD with constant false alarm rate-zero (CFAR-Z) algorithm to
combat radar interference signal. The proposed method is shown in Figure 1. Received signals are digitised
using an analog-to-digital converter (ADC) as raw signals. Raw signals will be extracted to get mean, RMS,
SNR, and THD parameters as NN inputs. VMD-CFAR-Z algorithms process raw signals if contaminated
with interference, while fast fourier transform (FFT) will process non-interference signals to get the radar
range profile.

Figure 1. Radar interference detection and mitigation method

2.1. mmWave frequency modulated continuous wave

mmWave frequency modulated continuous wave (FCMW) radar is used in autonomous vehicles to
detect targets along the road in short and long-range modes. The advantages of using mmWave FMCW radar
are high-resolution detection and less power to transmit a chirp sign al. The transmitted signal 𝑦(𝑡) of FMCW
is expressed in (1).
2 /2)
𝑦 (𝑡) = 𝑒 𝑗2𝜋 (𝑓𝑜 𝑡+𝐾𝑡 ,0 < 𝑡 < 𝑇 (1)

Where 𝑓𝑜 , 𝐾, and 𝑇 represent starting frequency, sweep slope, and time duration, respectively. Meanwhile, a
received signal consists of a reflected, transmitted signal by target and noise. If any radars in confront
position each other, the received signal is added by the in terference signal 𝑠𝑖 (𝑡) . A beat signal after the mixer
and low pass filter in receive part of FMCW radar as formulated in (2).

𝑟𝑏 (𝑡) = 𝑠𝑏 (𝑡) + 𝑠𝑖 (𝑡) + 𝑛 (𝑡) (2)

where 𝑟𝑏 (𝑡) , 𝑠𝑏 (𝑡) , and 𝑛 (𝑡) represents beat signal, echo signal from targets, and noise from the
environment, respectively.

2.2. Neural network

A NN consists of an input, hidden, and output layer. This research's input layer consists of four
inputs from the mean, RMS, SNR, and THD of the received signal, while a hidden layer consists of ten

Autonomous radar interference detection and mitigation using neural network and ... (Dayat Kurniawan)
2856  ISSN: 2252-8938

neutrons and one output layer, as shown in Figure 2. Levenberg-Marquardt backpropagation and mean
squared error (MSE) are used to train the proposed NN.

Figure 2. Proposed NN

2.3. Signal decomposition

VMD is a signal decomposition from 𝑥 (𝑡) signal to 𝑁 narrowband intrinsic signal (NIMFs) as
expressed in (3):

𝑥 (𝑡) = ∑𝐾𝑘=1 𝑢 𝑘 (𝑡) (3)

where 𝑢 𝑘 (𝑡) is the frequency and amplitude-modulated signals. Optimation of the VMD algorithm is
discussed in [26].

2.4. Constant false alarm rate–zero

CFAR-Z has been introduced in [11], where it is proposed to mitigate radar interference with low
complexity and reliability to implement in the actual board. CFAR-Z outperforms ANC and wavelet
denoising. CFAR-Z algorithm is based on cell-averaging (CA)-CFAR to detect the peak signal of the
spectrum signal in the frequency domain. The detected signals are replaced with zero to remove interference
in the received signal.

3. RESULTS AND DISCUSSION

3.1. System model
Autonomous vehicle radar interference occurs when two vehicles or more co -front each other on the
road, as modelled in Figure 3. Radar interference is categorised into two models: interference with different
radar parameters, such as difference in sweep time, and interference caused by the same radar parameter
between the aggressor and victim car. We simulate one victim and three aggressor radars with varying
parameters. Three aggressors were placed in different locatio ns relative to the victim's radar. Generally, all
victim and aggressor radar use mmWave radar with a frequency of 77 GHz, as tabulated in Table 1.

Figure 3. Radar interference scenario

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2854-2861

Int J Artif Intell ISSN: 2252-8938  2857

Table 1. System model parameters

Parameters Victim radar Aggressor radar
Center frequency (GHz) 77 77
Max range detection (m) 250 250
Sample rate (MHz) 40 40
Bandwidth (MHz) 600 300,900,300
Sweep duration (µs) 100 100, 100, 100
Sweep slope (MHz/ µs) 6 3, 9, 3
Interference delay time (µs) - 5, 10, 20

This research simulates radar interference caused by three aggregator radars with different
bandwidths, as shown in Table 1. A victim radar also detects targets from four targets with different location
as follow: 5 m, 10 m, 15 m, and 20 m in front of a victim radar. The generated signal in the time domain for
non-interference and interference conditions under additive white Gaussian noise (AWGN) -10 dB is shown in
Figure 4.

Figure 4. Interference-free and interference radar beat signals in the time domain

The FFT is applied to get the range profile of the received signal, as shown in Figure 5. Some target
signals in a range of 10 m, 15 m, and 20 m are uncleared enough to detect as a target radar caused by
interference signal. The noise floor signal increases from -25 dB to -15 dB, which means the SINR received
signal decreased.

Figure 5. Range profile for non-interference and interference signal after FFT processing
Autonomous radar interference detection and mitigation using neural network and ... (Dayat Kurniawan)
2858  ISSN: 2252-8938

3.2. Neural network

We designed a simple NN with a compact architecture and easily implemented in the actual board,
as shown in Figure 2. The NN input is the received signal features such as SNR, THD, mean, and RMS. The
total data train is 100,000, consisting of interference and non-interference signals, where each category has
50,000 data. The NN is trained with three aggressor radar with varying time slope and different interference
delay time under varying SNR from 1 to 5 dB. Levenberg–Marquardt algorithm and MSE are used along the
training NN. The performance of the NN is shown in Figure 6.

Figure 6. NN performance

3.3. Interference mitigation

This paper proposes the mitigation of interference radar by combining signal decomposition with a
zeroing process based on a CFAR-Z. VMD with 𝑁𝑢𝑚𝐼𝑀𝐹 = 9 is used to extract the detected interference
signal into nine sub-band signals. CFAR-Z processing is implemented for 𝑁𝑢𝑚𝐼𝑀𝐹 = 4 to 𝑁𝑢𝑚𝐼𝑀𝐹 = 9,
while others 𝑁𝑢𝑚𝐼𝑀𝐹 is considered as interference and noise only. After training NN, we test our proposed
method with parameters as listed in Table 1 under AWGN noise -10 dB. The result of interference mitigation
processing is shown in Figure 7.

Figure 7. Mitigation performance for various methods

Figure 7 shows that the CFAR-Z algorithm underperforms to mitigate targets under the
non-interference received signal. A simple NN is proposed to detect whether or not the received radar signal

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2854-2861

Int J Artif Intell ISSN: 2252-8938  2859

is contaminated with other signal radar. The proposed algorithm, NN-VMD-CFAR-Z, outperforms another
method with increased SINR up to 2 dB than the NN-CFAR-Z algorithm. The average increased SINR by the
NN-VMD-CFAR-Z algorithm under various AWGN noises is tabulated in Table 2. All simulation was
processed on a computer with 11th Gen Intel(R) Core (TM) i7-1165G7 @ 2.80 GHz and 8 GB installed
RAM. The processing time is analysed to show the saving time process between interfered and non -interfered
waveforms, and the result is shown in Table 3. From Table 3, the proposed NN can save significant
processing time of up to 1.65 seconds and avoid underperforming CFAR-Z in the non-interfered waveform.

Table 2. Increased SINR by NN-VMD-CFA R-Z algorithm

Parameters AWGN=-15dB AWGN=-10dB AWGN=-5dB AWGN=0dB
SINR-based (dB) -18.85 -17.42 -16.78 -16.65
SINR-NN-CFAR-Z (dB) 8.00 11.30 15.34 18.6
SINR-NN-VMD-CFAR-Z (dB) 10.60 14.27 18.45 21.19

Table 3. Processing time

Parameters T ime (s)
Pre-processing data 0.004810
NN-CFAR-Z 0.105028
NN-VMD-CFAR-Z 1.652097
FFT (non-interference) 0.000061

4. CONCLUSION
Radar interference detection and mitigation has been simulated and evaluated under various noise
and interference signal condition. A simple and compact designed NN performs well in detecting interference
signals. The proposed method, NN-VMD-CFAR-Z, outperforms with increases SINR up to 2 dB on average
higher than NN-CFAR-Z algorithm.

ACKNOWLEDGEMENTS
This research is supported by Research Center for Telecommunication –National Research and
Innovation Agency (BRIN).

REFERENCES
[1] M. Huang, R. Zhang, Y. Ma, and Q. X. Yan, “Research on autonomous driving control method of intelligent vehicle based on
vision navigation,” in 2010 International Conference on Computational Intelligence and Software Engineering, CiSE 2010, IEEE,
2010, doi: 10.1109/CISE.2010.5676770.
[2] P. Alam and P. Rajalakshmi, “Deep learning based steering angle prediction with LiDAR for autonomous vehicle,” in IEEE
Vehicular Technology Conference, IEEE, 2023, doi: 10.1109/VT C2023-Spring57618.2023.10201141.
[3] Y. C. Chung, H. Y. Chang, R. Y. Chang, and W. H. Chung, “Deep reinforcement learning-based resource allocation for cellular
V2X communications,” in IEEE Vehicular Technology Conference, IEEE, 2023, doi: 10.1109/VT C2023-
Spring57618.2023.10200293.
[4] X. Guo and X. Hong, “DQN for smart transportation supporting V2V mobile edge computing,” in Proceedings - 2023 IEEE
International Conference on Smart Computing, SMARTCOMP 2023 , IEEE, 2023, pp. 204–206, doi:
10.1109/SMART COMP58114.2023.00048.
[5] Y. Yao, F. Shu, X. Cheng, H. Liu, P. Miao, and L. Wu, “Automotive radar optimization design in a spectrally crowded V2I
communication environment,” IEEE Transactions on Intelligent Transportation Systems, vol. 24, no. 8, pp. 8253–8263, 2023,
doi: 10.1109/T IT S.2023.3264507.
[6] C. Kärnfelt et al., “77 GHz ACC radar simulation platform,” in 2009 9th International Conference on Intelligent Transport
Systems Telecommunications, ITST 2009, IEEE, 2009, pp. 209–214, doi: 10.1109/IT ST .2009.5399354.
[7] S. Rao and A. V. Mani, “Interference characterization in FMCW radars,” in IEEE National Radar Conference - Proceedings,
IEEE, Sep. 2020, doi: 10.1109/RadarConf2043947.2020.9266283.
[8] F. Jin and S. Cao, “Automotive radar interference mitigation using adaptive noise canceller,” IEEE Transactions on Vehicular
Technology, vol. 68, no. 4, pp. 3747–3754, Apr. 2019, doi: 10.1109/T VT .2019.2901493.
[9] Z. Xu and M. Yuan, “An interference mitigation technique for automotive millimeter wave radars in the tunable Q-Factor wavelet
transform domain,” IEEE Transactions on Microwave Theory and Techniques, vol. 69, no. 12, pp. 5270–5283, 2021, doi:
10.1109/T MT T .2021.3121322.
[10] S. Lee, J. Y. Lee, and S. C. Kim, “Mutual interference suppression using wavelet denoising in automotive FMCW radar systems,”
IEEE Transactions on Intelligent Transportation Systems, vol. 22, no. 2, pp. 887–897, 2021, doi: 10.1109/T IT S.2019.2961235.
[11] J. Wang, “CFAR-based interference mitigation for FMCW automotive radar systems,” IEEE Transactions on Intelligent
Transportation Systems, vol. 23, no. 8, pp. 12229–12238, 2022, doi: 10.1109/T IT S.2021.3111514.
[12] T . Balasooriya, P. Nallabolu, and C. Li, “Application of variational mode decomposition to FMCW radar interference mitigation,”
in Lecture Notes in Electrical Engineering, Springer Publishing, 2022, pp. 425–432, doi: 10.1007/978-3-030-98886-9_33.
[13] A. B. Baral, B. R. Upadhyay, and M. Torlak, “Automotive radar interference mitigation using two-stage signal decomposition
approach,” in Proceedings of the IEEE Radar Conference, IEEE, 2023, doi: 10.1109/RadarConf2351548.2023.10149713.

Autonomous radar interference detection and mitigation using neural network and ... (Dayat Kurniawan)
2860  ISSN: 2252-8938

[14] Z. L. Xia, X. Huai Wang, H. B. Wei, and Y. Xu, “Detection of vital signs based on var iational mode decomposition using FMCW
radar,” in 2021 International Conference on Microwave and Millimeter Wave Technology, ICMMT 2021 - Proceedings, IEEE,
2021, doi: 10.1109/ICMMT 52847.2021.9617967.
[15] Y. Li, B. Feng, and W. Zhang, “Mutual interference mitigation of millimeter-wave radar based on variational mode
decomposition and signal reconstruction,” Remote Sensing, vol. 15, no. 3, 2023, doi: 10.3390/rs15030557.
[16] A. Gaur, P.-H. Tseng, K.-T. Feng, and S. Srirangarajan, “VAFER: signal decomposition based mutual interference suppression in
FMCW radars,” arXiv-Electrical Engineering and Systems Science, pp. 1-11, Dec. 2022.
[17] T . Moon, J. Park, and S. Kim, “BlueFMCW: random frequency hopping radar for mitigation of interference and spoofing,”
Eurasip Journal on Advances in Signal Processing , vol. 2022, no. 1, 2022, doi: 10.1186/s13634 -022-00838-7.
[18] J. Hille, D. Auge, C. Grassmann, and A. Knoll, “FMCW radar2radar interference detection with a recurrent neural network,” in
Proceedings of the IEEE Radar Conference, IEEE, 2022, doi: 10.1109/RadarConf2248738.2022.9764236.
[19] J. Mun, H. Kim, and J. Lee, “A deep learning approach for automotive radar interference mitigation,” in IEEE Vehicular
Technology Conference, IEEE, 2018, doi: 10.1109/VT CFall.2018.8690848.
[20] J. Rock, W. Roth, P. Meissner, and F. Pernkopf, “Quantized neural networks for radar interference mitigation,” arXiv-Electrical
Engineering and Systems Science, pp. 1-12, Nov. 2020.
[21] J. Rock, W. Roth, M. T oth, P. Meissner, and F. Pernkopf, “Resource-efficient deep neural networks for automotive radar
interference mitigation,” IEEE Journal on Selected Topics in Signal Processing , vol. 15, no. 4, pp. 927–940, 2021, doi:
10.1109/JST SP.2021.3062452.
[22] Z. Xu, S. Xue, and Y. Wang, “Incoherent interference detection and mitigation for millimeter -wave FMCW radars,” Remote
Sensing, vol. 14, no. 19, Sep. 2022, doi: 10.3390/rs14194817.
[23] F. Uysal, “Synchronous and asynchronous radar interference mitigation,” IEEE Access, vol. 7, pp. 5846–5852, 2019, doi:
10.1109/ACCESS.2018.2884637.
[24] M. Wagner, F. Sulejmani, A. Melzer, P. Meissner, and M. Huemer, “Threshold-free interference cancellation method for automotive
FMCW radar systems,” in IEEE International Symposium on Circuits and Systems, IEEE, 2018, doi: 10.1109/ISCAS.2018.8351077.
[25] M. Umehira, T. Okuda, X. Wang, S. T akeda, and H. Kuroda, “An adaptive interference detection and suppression scheme using
iterative processing for automotive FMCW radars,” in IEEE National Radar Conference - Proceedings, IEEE, Sep. 2020, doi:
10.1109/RadarConf2043947.2020.9266712.
[26] K. Dragomiretskiy and D. Zosso, “Variational mode decomposition,” IEEE Transactions on Signal Processing, vol. 62, no. 3, pp.
531–544, 2014, doi: 10.1109/T SP.2013.2288675.

BIOGRAPHIES OF AUTHORS

Dayat Kurniawan is currently working as a researcher at the Research Center for

Telecommunication, National Research and Innovation Agency (BRIN). His research interests
lie in the areas of signal processing, wireless communication, and machine learning. He can be
contacted at email: [email protected].

Budiman Putra Asmaur Rohman is currently working as a researcher at the

Research Center for Telecommunication, National Research and Innovation Agency (BRIN).
His research interests lie in the areas of signal processing, machine learning, and embedded
systems. He can be contacted at email: [email protected].

Ratna Indrawijaya is currently working as a researcher at the Research Center

for Telecommunication, National Research and Innovation Agency (BRIN). His research
interests lie in the area of signal processing, radar systems, and Terahertz communication. He
can be contacted at email: [email protected].

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2854-2861

Int J Artif Intell ISSN: 2252-8938  2861

Chaeriah bin Ali Wael received her M aster's in electrical engineering from
Institut Teknologi Sepuluh Nopember, Surabaya - Indonesia. She joined the Research Center
for Electronics and Telecommunication - LIPI in 2015 as a junior researcher. Currently, she is
pursuing her Ph.D. degree in 5G enabling technology for intelligent transportation systems at
Institut d’Électronique de M icroélectronique et de Nanotechnologie (IEM N), Université
Polytechnique Hauts-de-France (UPHF), France. Her research interests include wireless
communications, signal processing, and machine learning. She can be contacted at email:
[email protected].

S uyoto is currently a researcher at Research Center Electronics and

Telecommunication National Research and Innovation Agency (BRIN). He obtained a
Bachelor of Telecommunication Engineering, a M aster in Computer and M ultimedia Systems,
and a Doctor of Telecommunication Engineering, all from the Bandung Institute of
Technology in 2002, 2009, and 2019, respectively. His research interests include signal
processing, synchronisation, multicarrier systems, and wireless communication. He can be
contacted at email: [email protected].

Purwoko Adhi is currently working as a researcher at the Research Center for

Telecommunication, National Research and Innovation Agency (BRIN). His research interests
lie in the area of signal processing. He can be contacted at email: [email protected].

Iman Firmansyah is currently working as a researcher at the Research Center for

Telecommunication, National Research and Innovation Agency (BRIN). His research interests
lie in the areas of embedded systems, high-performance computing, and digital signal
processing. He can be contacted at email: [email protected].

Autonomous radar interference detection and mitigation using neural network and ... (Dayat Kurniawan)