基于FPGA的大矩阵奇异值分解的实现  

Implementation to solve singular value decomposition of large matrix size in FPGAs

作　　者：许乔[1] 巩玉振[1] 蔡惠智[1] 

机构地区：[1]中国科学院声学研究所,北京100190

出　　处：《电子测量技术》2014年第6期68-72,共5页Electronic Measurement Technology

摘　　要：提出了一种在FPGA中实现大矩阵奇异值分解的方法。主要思想是通过有限状态机进行控制,复用流水线CORDIC计算模块。这种方法极大地节省了FPGA的面积资源,并且增加了算法求解的时间并行度。计算256阶实对称矩阵的奇异值仅占用3601个ALUT,运行时间仅为95.28ms。A method to implement singular value decomposition of large matrix size in FPGA is proposed. The main idea is to multiplex a pipelined CORDIC module and automatically control the order of multiplexing it using a finite state machine. This method could minimize the area resources of an FPGA,and improve timing parallelism of the algorithm. To solve a real symmetric matrix with a size of 256 × 256,it takes only 95.28 milliseconds to finish the computation and the ALUTs consumed is just about 3601.

关 键 词：奇异值分解 FPGA 流水线 CORDIC 

分 类 号：TP271.82[自动化与计算机技术—检测技术与自动化装置]