SARComp: High-Performance Algorithms for SAR Processing from FFT Kernels to Matched Filtering

Abstract

This paper presents optimization strategies for accelerating Synthetic Aperture Radar (SAR) image formation on modern processors. Standard techniques like OpenMP and auto-vectorization often fail to optimize core SAR functions, especially FFT-based kernels. We address this by designing SIMD-optimized routines that restructure data flows to overcome vectorization bottlenecks. Our implementation replaces compiler heuristics with explicit vector control, achieving substantial speedups. These improvements are critical to meeting the demands of advanced SAR systems in real-time and with limited resources.