A method and device for SVD decomposition of channel matrix

Abstract

The invention discloses an SVD decomposition method and device for a channel matrix. The method comprises the steps that acquring the channel matrix H; performing CUR decomposition on the channel matrix H to obtain three low-dimensional matrixes C, U and R; performing QR decomposition on the low-dimensional matrix C and RT respectively to obtain two lower triangular matrixes Rc and RR and two orthogonal matrixes Qc and QR; combining the two lower triangular matrixes Rc; combining the RR with the low-dimensional matrix U and carrying out SVD decomposition; obtaining a diagonal matrix s and twoorthogonal matrixes Us and VsT, and taking the diagonal matrix s and the four orthogonal matrixes Qc, Us, VsT and QRT as decomposition results of the channel matrix H, so that the real-time processingperformance of channel matrix decomposition is improved, and the calculation load, the processing delay and the power consumption are reduced.

Description

technical field [0001] The invention relates to the field of communication technology, in particular to a method and device for SVD decomposition of a channel matrix. Background technique [0002] SVD decomposition is an important decomposition method in matrix theory. With the rapid development of disciplines such as machine learning, big data, image processing, and signal processing, the application fields of SVD are becoming more and more extensive. However, the computational bottleneck of SVD decomposition greatly limits the practical application of these disciplines. Although SVD decomposition is a very powerful mathematical tool, its super high computational complexity makes its practical application a certain obstacle. [0003] The traditional mainstream SVD decomposition method will directly process the matrix, but the calculation of SVD decomposition is too complicated, and its computational complexity is O(KM 2 ) or O(M 3 ), where M represents the dimension of t...

AI Technical Summary

Problems solved by technology

When the size of the matrix grows, its requirements for hardware and time are in a cubic growth state. Although MATLAB can be used to perform SVD decomposition on the matrix, when the matrix is ​​larger than a certain scale, it seems unable to do what it wants.

As a leader in the field of machine learning, Google uses a parallel computing method for SVD decomposition, but this method has high requirements for hardware and cannot fundamentally overcome the problem of computational complexity expansion.

Therefore, the computing resources and processing delay required by traditional SVD decomposition will become unbearable for real-time signal processing and analysis scenarios.

Images