Implementation of the CA-CFAR algorithm for pulsed-doppler radar on a GPU architecture

Abstract

The Cell-Averaging Constant False-Alarm Rate (CA-CFAR) algorithm was implemented and optimized in software on the NVIDIA Tesla C1060 GPU architecture for application in pulsed-Doppler radar signal processors. A systematic approach was followed to gradually explore opportunities for parallel execution and optimization by implementing the algorithm first in MATLAB (CPU), followed by native C (CPU) and finally NVIDIA CUDA (GPU) environments. Three techniques for implementing the CA-CFAR in software were identified and implemented, namely a naive technique, sliding window technique and a new variant which employs the Summed-Area Table (SAT) algorithm. The naive technique performed best on the GPU architecture. The SAT technique shows potential, especially for cases where very large CFAR windows are required. However, the results do not justify using the GPU architecture instead of the CPU architecture for this application when data transfer to and from the GPU is taken into consideration.