High Throughput-rate Parallel Algorithms for Space Time Adaptive Processing

Space Time Adaptive Processing (STAP) techniques that are being developed for the next generation radar systems require very high computing power. In order to meet the real-time requirements in STAP applications, High Performance Computing (HPC) systems are used. In this paper, we present a methodology to design high throughput-rate parallel algorithms for STAP on HPC platforms. Data remapping is used between successive computation steps of STAP. Task level parallelism is used for these steps. In order to achieve high throughput performance, we regard the HPC platform as executing in a pipelined fashion. For each pipeline stage, we schedule the computation and the communication, and assign a number of processors based on the computation requirements and associated communication time. A simple and accurate communication model is used to characterize the communication time. A technique called replication of pipeline stages is employed to eeectively reduce the communication cost and the period of each pipeline stage. As an example, our methodology is applied to a general STAP technique called Beam space post-Doppler STAP 12]. Experimental results conducted on the IBM SP2 show that our methodology leads to high throughput performance. Our code is written using C and MPI, and is portable across HPC platforms.