Real-Time Signal Processing of Massive Sensor Arrays via a Parallel Fast Converging SVD Algorithm: Latency, Throughput, and Resource Analysis
Real-Time Signal Processing of Massive Sensor Arrays via a Parallel Fast Converging SVD Algorithm: Latency, Throughput, and Resource Analysis
复制标题
通过并行快速收敛 SVD 算法对大规模传感器阵列进行实时信号处理：延迟、吞吐量和资源分析
作者：
Mrudula V. Athi;S. Zekavat;A. Struthers
介绍了一种适用于大规模传感器阵列实时信号处理的并行快速收敛类雅可比奇异值分解(SVD)算法。与传统的基于Jacobi的方法相比，该算法大大提高了大型矩阵的奇异值分解收敛速度。提出了一种高度模块化的系统设计，它保留了雅可比方法的并行本质，这是现场可编程门阵列(FGA)实时实现的关键。通过在Virtex-6现场可编程门阵列上的实现验证了设计的正确性，并通过仿真验证了性能的提高。从资源消耗、最大可达频率、时延吞吐量权衡等方面对该设计与传统设计进行了比较。
This paper introduces a parallel fast converging Jacobi-like singular value decomposition (SVD) algorithm applicable to real-time signal processing of massive sensor arrays. The proposed algorithm highly increases the SVD convergence rate for larger matrices when compared with traditional Jacobi-based methods. A highly modular system design is proposed, which retains the parallel nature of the Jacobi methods key to real-time implementation intended for field programmable gated arrays (FPGAs). The proof of design was provided via an implementation on Virtex-6 FPGA, and the improvement in performance was verified via simulations. The proposed design was compared with the traditional design in terms of FPGA resource consumption, maximum achievable frequency, and latency throughput tradeoff.