Biography
Enrollment Date: 2009
Graduation Date：2012
Degree：M.S.
Defense Date：2012.05.28
Advisors：Yangdong Deng
Department：Institute of Microelectronics，Tsinghua University
Title of Dissertation/Thesis：The Parallel Implementation of Radar Space-Time Adaptive Processing Based on Several Platforms
Abstract:
Space-time adaptive processing (STAP) is an essential technique for such key applications as the next generation airborne early warning and control system (AWACS) radars which has quite high requirements of computing power. In this paper, we develop an efficient GPU-accelerated solution as well as a multicore CPU-accelerated solution for a complete STAP backend. First, based on a detailed analysis of the performance, we found that the bottleneck of STAP is the processing of Hermitian matrices inversion. So in this paper we design a parallel algorithm of the Hermitian matrices inversion based on Cholesky factorization and implement it on both multicore CPU and GPU. Our quad-core CPU solution has a speedup of 3X against its single-core CPU counterpart while Our GPU solution achieves a speedup of up to 33X and a throughput of more than 30 GFLOPs. The speedup can be taken advantage in two ways: 1. Dealing with higher resolution radar data in the same processing time to achieve a higher radar accuracy; 2. Dealing with the original resolution radar data in a shorter time to realize a better real-time performance.Meanwhile, the Hermitian matrices inversion in STAP is an important mathematical algorithm. The algorithms of Cholesky factorization, inversion of triangular matrices and triangular matrix multiplication in it are widely used in mathematical computation and various engineering field. So the parallelization of these algorithms as in stated in this thesis has significant general meaning.Besides the parallel design of STAP based on the GPU platform and multicore CPU platform, this work gives a trial of standardizing the parallel programming practice, which can provide the basis to the parallel implementation of complex algorithms later.