Research On Anti-Interference Method Of Vehicle FMCW Radar Based On Reinforcement Learning

As a key sensing device for autonomous driving,the accurate sensing of position,velocity,and azimuth of on-board millimeter wave radars can not only effectively improve the safety performance,but also provide effective support for the subsequent fusion sensing and planning decision process.When there are multiple radars in the effective working area of the vehicle mounted millimeter wave radar,it will lead to serious mutual interference problems of the vehicle mounted radar.How to effectively solve the problems of radar detection false alarms,missed alarms,and shortened radar detection distance caused by mutual interference of the vehicle mounted radar has become a priority for safe driving.However,most of the existing anti-interference technology methods rely on specific interference environment parameters and have a high dependence on the environment,hardware,and communication infrastructure.They can only achieve effective anti-interference in limited situations,leading to problems such as instability and poor robustness of existing technologies Based on the above research background,this paper relies on reinforcement learning and vehiclemounted Frequency Modulated Continuous Wave(FMCW)radar signal processing technology to study the vehicle-mounted FMCW radar anti-interference algorithm based on reinforcement learning,overcomes the application limitations of traditional algorithms,and effectively improves the autonomous anti-interference performance of vehicle-mounted FMCW radar.The main contributions of this paper are as follows:(1)Aiming at the problem that traditional anti-interference algorithms have high requirements for vehicle FMCW radar interference signal statistics and strong infrastructure support,this paper proposes an anti-interference algorithm for vehicle FMCW radar bandwidth allocation based on DQN.The algorithm first divides the available bandwidth into several candidate subbands according to the resolution requirements,and then takes the detection domain signal-to-interference ratio as the measurement standard and reward function,and fuses the information such as the subband information,reward value and target position at the historical moment into an environmental state variable as the input of the neural network,and finally through continuous iterative training,the vehicle-mounted FMCW radar intelligently selects the sub-band with low or no interference degree for transmission,so as to reduce or avoid the interference impact as much as possible.The simulation results show that the algorithm does not need to obtain the interference center frequency,bandwidth and other information,nor does it require complex communication overhead,and can achieve effective anti-interference by relying only on self-intelligence.(2)Aiming at the complex interference situation such as similar chirp slope and overlapping occupied bandwidth between own FMCW radar and FMCW interference radar,this paper expands the anti-interference algorithm based on bandwidth allocation,and proposes an anti-interference algorithm based on DDQN for vehicle FMCW radar bandwidth and chirp slope allocation.The algorithm discretizes the available bandwidth and CHIRP slope to anti-interference from the time-frequency domain parameters of the vehicle FMCW radar signal,effectively ensuring the effectiveness and robustness of the anti-interference algorithm in the face of complex electromagnetic environments.