cuRadar

Problem

Modern radar systems generate large volumes of raw ADC data that must be transformed through range-doppler processing, detection, and beamforming stages under tight latency constraints. cuRadar explores how a full radar signal processing chain — from raw samples through to detections — can be implemented and accelerated on the GPU using CUDA, as a way to build hands-on intuition for heterogeneous, high-performance signal processing pipelines.

Highlights

• Simulated raw radar signal generation for a configurable pipeline input
• GPU-accelerated Range-Doppler processing using CUDA-based FFTs
• CA-CFAR (Cell-Averaging Constant False Alarm Rate) detection implemented as a CUDA kernel
• Bartlett beamforming for angle-of-arrival estimation on the GPU