Abstract: High-performance radar signal processing is a key technique in radar systems. Previous research on high-performance radar signal processing algorithms usually rely on specialized signal processing devices, such as FPGAs or DSPs, which are not only expensive but also difficult to program. GPU is a general-purpose device, especially suitable for processing large-scale data, such as radar signals, etc. At present, most of the existing GPU-accelerated radar signal processing research is on SAR imaging related applications, with little research on target detection radar, such as pulsed Doppler radar. In order to fulfill the requirement of radar echo data on throughput and real-time processing, a novel fine-grained parallelization by grid stride, a coarse-grained parallelization by multi-CUDA streams, and a data preprocessing method based on parallel scan are proposed. The real-time performance and accuracy of the algorithm are evaluated from multiple perspectives such as performance testing and error analysis. Compared with the state-of-the-art implementation on CPU, the hardware platform has achieved a speedup of more than 300 times, and is superior than other existing CUDA accelerated pulse Doppler radar signal processing algorithms.

Key words: pulsed Doppler radar, GPU parallel computing, grid striding parallism, multi-stream parallism, parallel scan