Channel information compression feedback method adopting bit-level optimized network

Abstract

The invention discloses a channel information compression feedback method adopting a bit-level optimized network. Firstly, two-dimensional Fourier transform is used for converting a channel matrix ofa frequency domain into an angle-time delay domain, then a complex matrix is split into two real number sub-matrices and splicing is performed; a coding network is used for extracting channel characteristics through a composite residual network, channel characteristic dimension is compressed through a full-connection network, and finally each real number is quantized into a 4-bit binary symbol through a quantizer; and a decoding network is used for mapping received binary bit streams into real number code words, then dimension expansion is carried out through one full-connection network, and finally channel information is recovered through two composite residual networks. According to the method, the composite residual network is introduced, and a bit-level optimized network training method is provided, so that the compression and recovery performance of the channel information is effectively improved; and channel compression feedback can be carried out more efficiently by the trainedbit-level optimized neural network; and the method has obvious performance advantages.

Description

technical field [0001] The invention belongs to the technical field of compressed sensing and channel information recovery, relates to a channel information compression feedback network method, in particular to a channel information compression feedback method using a bit-level optimized network. Background technique [0002] In massive MIMO, the base station is equipped with a large number of antennas, so that the uplink channel information feedback needs to consume a lot of resources. Usually, the channel information matrix of massive MIMO is very sparse, and its efficient compression can well reduce the cost of feedback. However, the traditional codebook-based method cannot be used when the channel matrix is ​​very large due to the exponential relationship between its calculation and storage costs and the number of feedback bits; the method based on compressed sensing is difficult to recover when the feedback bits are small. However, the current neural network-based comp...

AI Technical Summary

Problems solved by technology

[0003] In order to solve the above problems, the present invention discloses a channel information compression feedback method using a bit-level optimized network, designs a more efficient convolution structure, and adopts a bit-level optimized network design, thereby greatly improving network performance , which can solve the problem of low feedback accuracy of existing methods on the channel compression feedback problem

Images