Optimized Parallel Architecture of Kalman Filter for Radar Tracking Applications
Amin A. Jarrah |Pages: 215-230 |

Abstract— Kalman filter has been proven to be a very effective method to identify targets in an efficient and accurate manner. It provides efficient estimations when the precise nature of the modeled system is unknown in the presence of measurement and process noise. However, Kalman filter is computationally extensive especially in Multi Target Tracking (MTT) radar system. Therefore, it is desirable to apply it on advanced parallel architecture such as FPGA, GPU, and multi-cores to increase performance and achieve real time requirements. In this paper, we present an efficient parallel architecture of Kalman filter on different platforms such as FPGA, GPU, and multi- core. Kalman filter operations are carried out on a single core CPU before they are decomposed, parallelized, scheduled, and mapped into FPGA and GPU platforms. Different optimization techniques for both the computation and memory utilization are adopted and applied to achieve high performance. The experimental results show the viability of using FPGA and GPU platforms to perform signal processing in real time. Parallel architectures can significantly outperform an equivalent sequential implementation due to their pipelined architecture, custom functionality of VLSI ASIC devices, flexibility, and adaptability. Our simulation results indicate that the achieved speed-up of FPGA and GPU over the sequential one is improved by up to 37.76 and 31.93, respectively.