Acceleration of signal processing in modern radar based on GPU platform

Abstract

Modern radar systems aiming for high resolution, wide bandwidth, and real-time processing of large data volumes pose significant computational challenges to traditional signal processing approaches. Implementations based on CPUs, FPGAs (Field-Programmable Gate Array), or dedicated DSPs (digital signal processors) often fail to provide sufficient throughput and computational resources for intensive tasks such as matched filtering, fast Fourier transforms (FFTs), Doppler processing, digital beamforming, and synthetic aperture radar (SAR) image formation. To address these limitations, this paper proposes the use of graphics processing units (GPUs) as an acceleration platform for radar signal processing algorithms by exploiting the massive parallelism inherent in GPU architectures. The paper further presents performance measurements and evaluations conducted on representative radar datasets using various signal processing algorithms. The results demonstrate that GPU-based implementations can achieve speedups ranging from tens to hundreds of times compared to MATLAB-based CPU implementations. These findings indicate that GPU-accelerated signal processing is a promising solution for meeting real-time processing requirements in modern radar systems. In addition, computational complexity analysis and numerical accuracy validation between CPU and GPU implementations are provided to ensure the correctness and scientific rigor of the reported performance improvements.