Parallel Implementations of Detection Algorithms for MIMO Systems on The Graphics Processing Unit

Abstract

In this thesis, the acceleration of detection algorithms for multiple-input multipleoutput (MIMO) wireless systems is investigated. First we use the graphics processing unit (GPU), which provides thousands of parallel threads, to accelerate our detectors. The simulation environment for the parallel detectors is MATLAB with the Jacket library extension, which can modify conventional serial simulation codes to access the GPU and be executed in parallel. Comparisons between serial and parallel versions of different MIMO detectors are described in this thesis to determine how much speed-up that can be achieved from the parallelism. Furthermore, a parallel hybrid VBLAST-KBest detection algorithm is proposed that increases the accuracy beyond the conventional K-Best algorithm. The use of different forms of parallelism to speed-up matrix multiplication is investigated to provide insight into making the best use of the GPU. As a comparison, a multicore CPU acceleration using the parallel computing toolbox (PCT) is also briefly investigated.