基于GPU并行计算的OMP算法  

Parallel Optimization OMP Algorithm Based on GPU

作　　者：谈继魁 方勇[1] 霍迎秋[1] 

机构地区：[1]西北农林科技大学信息工程学院,陕西杨陵712100

出　　处：《电视技术》2015年第15期42-45,54,共5页Video Engineering

基　　金：国家自然科学基金项目(61271280);中央高校基本科研业务费支持项目(QN2013086)

摘　　要：重建算法在压缩感知理论中有着重要的作用，经典的正交匹配追踪（OMP）重建算法在每次迭代中对已选择的原子进行正交化处理以加速算法的收敛速度，但同时增加了算法的计算复杂度。针对这一问题，提出了一种基于图形处理单元（GPU）并行计算的OMP算法，重点对算法中复杂度高的投影和矩阵求逆部分在GPU平台上进行并行设计。实验结果i表明，基于GPU的并行OMP算法相对于其串行算法加速比可以达到30～44倍，有效地提高了算法的计算效率，拓宽了该算法的应用范围。Recovery algorithms play a significant role in compressed sensing. The orthogonal matching pursuit （OMP） algorithm as ： the most popular algorithm makes orthogonal treatment of atoms selected at every step of decomposition to accelerate its convergence ：i speed. However, the computation complexity is considerably increased at the same time. To address this problem,a parallel OMP algorithm based on graphics processing unit（GPU） which focus on the parallel design to solve the high complexity problem of the projection and matrix inversion sections is introduced. The experiment results show that 30 - 44 times speedup is achieved by the parallel OMP algorithm based on GPU over its serial algorithm, which not only improves the computational efficiency of the algo-rithm effectively, but also broadens its scope of application.

关 键 词：压缩感知 正交匹配追踪算法 图形处理单元 并行计算 加速比 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]