Research On Signal And Data Processing Technology Of Automotive Millimeter Wave Radar

In recent years,with economic development and technological progress,unmanned driving technology has gradually become a hot topic in society and a research focus of related institutions.Automotive millimeter-wave radar has become one of the core sensors of unmanned vehicles due to its all-weather and all-weather characteristics.However,due to the cost constraints of the system hardware and the complexity of the actual road environment,to achieve effective detection goals and achieve advanced driving assistance functions,it is necessary to design accurate and robust algorithms.Therefore,the research based on the automotive millimeter wave radar signal and data processing technology has important theoretical and practical significance.To this end,this article has conducted the following research.(1)Research the signal processing algorithm of MIMO radar based on linear frequency modulation continuous wave(LFMCW),and simulate and verify the entire processing flow.Aiming at the problem that the estimated angle of MUSIC algorithm cannot match the target distance and speed,a simple 2D-FFT-MUSIC matching method is proposed.Finally,the actual millimeter wave radar signal processing flow is analyzed based on the AWR2243 cascade radar with lane-level resolution capability.(2)In view of the characteristics of broadband millimeter wave radars that can obtain target point clouds,carry out millimeter wave radar point cloud clustering research,mainly including K-means algorithm,AGNES hierarchical clustering algorithm and DBSCAN density clustering algorithm,aiming at weighted Euclidean distance There are many problems with parameters,the DBSCAN algorithm based on kernel distance is introduced,and an automatic method of K-distance graph to find Eps is given when calculating the neighborhood distance parameter Eps.Finally,the clustering effect of the algorithm is verified by the measured data.(3)Secondly,the target tracking theories and methods related to vehicle applications are studied,mainly including target association and filter tracking.The principles of the nearest neighbor data association(NNDA)algorithm and the probabilistic data association(PDA)algorithm are analyzed and compared by simulation.Study the principles of extended Kalman filter(EKF)and lossless Kalman filter(UKF)algorithms.In the case of abnormal detection values,use the method of weighting the measurement noise covariance matrix to achieve a robust extended Kalman filter for outlier detection(ROD-EKF)algorithm and outlier detection robust lossless Kalman filter(ROD-UKF)algorithm,and simulations verify that ROD-EKF and ROD-UKF outperform EKF and UKF under student t distribution noise with outliers,And the performance is equivalent to EKF and UKF under Gaussian noise.(4)Based on the real-time requirements of automotive millimeter wave radar signal processing,a GPU-based parallel signal processing method was designed and implemented.Parallel modules include 2D-FFT,non-coherent accumulation,2D-CFAR,etc.,and finally its calculation accuracy And the acceleration effect is evaluated.(5)Verification and analysis of the above-mentioned main algorithms based on the actual measurement data of TI’s AWR2243 cascade radar.