Floating point operation circuit implementation method for real number matrix inversion

Abstract

The invention discloses a floating point operation circuit implementation method for real number matrix inversion, and aims to provide a square matrix inversion method which is beneficial to HDL implementation and short delay. The invention is realized through the following technical scheme: inputting a one-out-of-three reselector according to a row or column beat sequence; inputting an N * N square matrix according to a line-by-line beat sequence; sequentially storing the matrix into an internal RAM buffer area; performing N times of cyclic transformation operation on the matrix line by lineby using an in-situ operation algorithm of a Gaussian-approximate elimination algorithm, reading the matrix row data of the kth row from an RAM cache unit by the one-to-three reselector in each cyclictransformation operation process, traversing all the row data of the matrix for one time to serve as one time of elimination circulation, and totally executing n times of circulation; and finally, outputting a result inverse matrix according to a line-by-line beat sequence. The time division multiplexing technology is adopted to reasonably distribute the operation time of the multiplier at different stages of operation, so that hardware resources are saved, the resource utilization efficiency is improved, and the data throughput rate and the operation speed are improved by a local assembly line technology.

Description

technical field [0001] The present invention relates to a hardware description language (Hardware Description Language, referred to as HDL) and is suitable for realizing the operation circuit implementation method of real number matrix inversion, especially relates to the floating-point operation circuit structure realization technology of real number matrix inversion. Background technique [0002] Matrix operation is a basic problem in scientific and engineering calculations, which can be widely used in various circuit calculations, and can be widely and deeply applied in such as DCT, LDPC coding and decoding, and adaptive array antenna technology that has emerged in recent years to solve one of the key issues. For example, in algorithms such as array signal processing and radar imaging, many calculations of weight coefficients involve matrix inversion, but due to the complexity of the algorithm, in many cases it can only be achieved by using the least mean square algorithm...

AI Technical Summary

Problems solved by technology

[0020] The purpose of the present invention is to solve the technical problems of complex circuit design, low utilization efficiency of hardware resources and excessive calculation delay in the existing method, and to provide a method with high calculation accuracy, strong reusability, low difficulty in implementation, and more efficient circuit design. Realization method of real number matrix inversion floating-point operation circuit with simplicity and higher utilization rate of hardware resources

Images