Research Of GPU-Based Radar Data Processing Technology

Radar data processing system is an important part of modern radar. As the complexity of modern battlefield increased, modern radar is faced with more fake targets. Consequently, radar data processing system is facing increasingly heavier computational load. Exploring a flexible, High-performance, software radar has become a trend. To meet the requirements of modern data processing system, this thesis proposed a GPU-based method to implement the radar data processing system. Utilizing GPU-based parallel optimization of data association algorithm and track initiation algorithm, this method significantly shortens the running time.This thesis studies the typical radar processing methods, including the conversion of Coordinate System, Plots Centroid, Track Initiation, Single Target Tracking and Multi-target Association algorithms. For track initiation problem, this thesis studies logic-based sliding-window detection method, Hough Transform-based method and Modified Hough Transform-based method, and validates the correctness of these methods by using Matlab. The thesis then studies the parallelism of each method, providing the theoretical basis for selecting Modified Hough Transform-based method as the track initiation method. For Single Target Tracking problem, this thesis studies Extended Kalman Filter(EKF) method. The effectiveness of the EKF method on tracking problem of targets with constant acceleration is validated by Monte Carlo simulation. For Multi-target Association problem, this thesis studies the theoretical basis of the Joint Probability Data Association(JPDA) algorithm.This thesis studies the GPU-based implementation of the JPDA algorithm. According to the structure of JPDA algorithm, this thesis studies the parallelism of JPDA, particularly the data storage and the thread scheduling methods of GPU-based implementation. The simulation results show that GPU-based implementation of JPDA algorithm can greatly shorten the running time. The speedup ratio varies from 30 to 300 with different target numbers. Besides, the correctness of GPU-based system is also verified in the simulation experiments.This thesis studies the GPU-based implementation of the Modified Hough Transform method, and simulation studies are carried out. By comparing the running time of CPU-based method and GPU-based method, the simulation experiment conforms that GPU-based implementation of the Modified Hough Transform algorithm can greatly shorten the running time.Finally, this thesis studies the design and implementation of radar data processing system. This system consists of three important modules, including data receiver module, data processing module and display module. Proven by the practical work, the system can correctly initiate the track of targets, perform target tracking and show the tracks.