Parallel PHD Function on Target Association with Multistatic Radar System (MRS) using CUDA

Target association is one of the main process of the signal processing of the passive Multi-static Radar System (MRS) which requires a complex geometry calculation. Among advanced techniques for passive radar system's target association, several experiments have been done based on Probability Hypothetic Density (PHD) function. The complex calculation makes the computation process a very demanding task to be done, thus, this paper is focused on PHD function performance comparison between preceding attempts to the implementation using pure C programming language with CUDA library. Naive parallelization is used on mapping each matrices data to CUDA memories, for each major operation is done in parallel behavior via self-made CUDA kernels to suits the data dimensions. Results for kernels are captured with NVIDIA profiling tools for increasing number of random targets on 4 transmitter-receiver (PV) combination (without any knowledge about approximation of targets direction). All results are taken according to the average running time of kernel calls and speed up for each size of input, compared with serial and CPU parallel version data of the previous work.