基于多核DSP的基带信号处理算法并行性研究

发布时间：2018-01-14 07:27

本文关键词：基于多核DSP的基带信号处理算法并行性研究　出处：《电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着计算机性能需求的不断增加,多机并行势必成为计算机的发展方向。同时,随着软件无线电技术SDR的兴起以及该技术在LTE移动通信系统中的使用越来越深入,高速并行计算的应用需求也越发急迫。为了推动LTE移动通信系统的全软件化进程,本文以多核DSP为基础,搭建了并行处理平台,并对相关的并行处理技术和信号处理算法的并行性进行了讨论和研究。矩阵求逆算法和FIR滤波器在LTE中都有着广泛的应用。通常情况下,这两种算法运算量较大,运算耗时较长。这并不能满足LTE系统对高速率的要求,所以需要采用并行方式对这两种算法进行并行实现。通信开销在一定程度上影响着算法的并行性能。本文通过性能测试对比了几种多核DSP的核间通信机制和片间通信机制,最终分别选用具备灵活性和高速率的EDMA3和SRIO来实现并行算法中的数据交互工作。针对矩阵求逆算法,本文对比了几种经典的求逆算法,综合计算复杂度和稳定性选择了全选主元高斯约旦算法作为并行计算的研究对象,并提出了一种基于全选主元高斯约旦算法的矩阵求逆并行算法。最后在TMS320C6474 EVM评估板上对该并行算法进行了多核实现,并对该并行算法加速比进行了对比分析。结果表明算法耗时与并行核数成反比,并且处理器数目可观时,该算法具备一定的加速比。针对FIR滤波器,本文对比了现有的几种并行滤波结构,并选择了其中具备运算量小等优势的基于多相滤波器的快速FIR并行滤波结构进行仿真,验证了该并行滤波结构的正确性和可靠性。最后在TMS320C6474 EVM评估板上对该并行滤波结构进行了多核实现,并对该并行算法的加速比进行了对比分析。结果表明算法耗时与并行核数成反比,并且该算法具有可观的加速比。
[Abstract]:With the increasing demand for computer performance, multi-computer parallelism is bound to become the direction of computer development. With the rise of software radio technology SDR and the application of this technology in LTE mobile communication system more and more in-depth. In order to promote the full software process of LTE mobile communication system, this paper builds a parallel processing platform based on multi-core DSP. The parallelism of related parallel processing techniques and signal processing algorithms is discussed and studied. Matrix inverse algorithm and FIR filter are widely used in LTE. These two algorithms can not meet the requirement of high speed in LTE system because of the large amount of computation and the long time consuming. Therefore, parallel implementation of these two algorithms is needed. The communication overhead affects the parallel performance of the algorithm to a certain extent. This paper compares several inter-core communication mechanisms of multi-core DSP and compares them with each other through performance tests. Inter-chip communication mechanism. Finally, EDMA3 and SRIO, which have flexibility and high speed, are selected to realize the data interaction in parallel algorithms. For matrix inversion algorithm, this paper compares several classical inverse algorithms. In order to improve the computational complexity and stability, the fully selected principal component Gaussian Jordan algorithm is chosen as the research object of parallel computing. A matrix inverse parallel algorithm based on fully selected principal component Gaussian Jordan algorithm is proposed. Finally, the parallel algorithm is implemented on the TMS320C6474 EVM evaluation board. The speedup ratio of the parallel algorithm is compared and analyzed. The results show that the time consuming of the algorithm is inversely proportional to the number of parallel kernels, and when the number of processors is considerable, the algorithm has a speedup ratio. The algorithm has a certain speedup for FIR filters. In this paper, several parallel filtering structures are compared, and the fast FIR parallel filter structure based on polyphase filter is selected for simulation. The correctness and reliability of the parallel filter structure are verified. Finally, the multi-core implementation of the parallel filter structure is carried out on the TMS320C6474 EVM evaluation board. The speedup of the parallel algorithm is compared and analyzed. The results show that the time consuming of the algorithm is inversely proportional to the number of parallel kernels and the algorithm has considerable speedup.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7;TN713

【相似文献】