US 12,438,584 B2
CSI compression and feedback method based on deep learning
Zhaoyang Zhang, Hangzhou (CN); and Zirui Chen, Hangzhou (CN)
Assigned to ZHEJIANG UNIVERSITY, Hangzhou (CN)
Filed by ZHEJIANG UNIVERSITY, Zhejiang (CN)
Filed on Dec. 5, 2024, as Appl. No. 18/969,330.
Application 18/969,330 is a continuation of application No. PCT/CN2023/093244, filed on May 10, 2023.
Claims priority of application No. 202211065211.4 (CN), filed on Sep. 1, 2022.
Prior Publication US 2025/0096872 A1, Mar. 20, 2025
Int. Cl. H04B 7/06 (2006.01); H04L 5/00 (2006.01); H04L 41/16 (2022.01)
CPC H04B 7/0626 (2013.01) [H04L 5/0007 (2013.01); H04L 41/16 (2013.01)] 7 Claims
OG exemplary drawing
 
1. A Channel State Information (CSI) compression and feedback method based on deep learning, the CSI compression and feedback method is applied to a multiple-input, multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) system, comprising:
step S1, establishing and training a two-dimensional sequence-to-sequence structure neural network;
sub-step S1.1, collecting, by a base station, 40,000 historical CSI data from a user terminal with a single receiving antenna by configuring the MIMO-OFDM system with 32 transmitting antennas and 256 subcarriers as training data of the two-dimensional sequence-to-sequence structure neural network;
sub-step S1.2, building, by the base station, the two-dimensional sequence-to-sequence structure neural network, and randomly initializing parameters of the two-dimensional sequence-to-sequence structure neural network, wherein the two-dimensional sequence-to-sequence structure neural network comprises:
an encoder comprising two cascaded 2D long short-term memory (LSTM) layers and one fully connected layer, wherein each layer of the two cascaded 2D LSTM layers has 128 hidden units, each cell of the 2D LSTM layers shares parameters in two dimensions, a CSI matrix is sub-blocked to obtain 1×32 sub-blocks, and each sub-block corresponds to a channel response of an antenna port on 32 consecutive subcarriers; and
a decoder comprising one fully connected layer and two cascaded 2D LSTM layers, wherein an output dimension of the fully connected layer of the decoder matches an input dimension of the fully connected layer at an end of the encoder; and the 2D LSTM layer of the decoder reconstructs 1×32 sub-blocks through parameter-sharing units; and
sub-step S1.3, based on an adaptive momentum estimation (Adam) optimizer, training the two-dimensional sequence-to-sequence structure neural network in the sub-step S1.2, and continuously training the parameters of the two-dimensional sequence-to-sequence structure neural network by a gradient descent method until convergence, wherein a learning rate is set to 0.001, and the training is repeated for 800 cycles, and the learning rate is decayed to ⅕ of a previous one every 100 cycles after a 500th training cycle;
step S2, distributing, by the base station, the encoder of the two-dimensional sequence-to-sequence structure neural network trained in the sub-step S1.3 to a client;
step S3, compressing, by the client, downlink CSI by using the encoder distributed in the step S2, and feeding the compressed downlink CSI back to the base station; and
step S4, reconstructing, by the base station, the compressed downlink CSI fed back by the client in the step S3 by using the decoder of the two-dimensional sequence-to-sequence structure neural network trained in the sub-step S1.3 to obtain complete downlink CSI, inputting the reconstructed complete downlink CSI into a downlink precoding module to generate beamforming parameters, and performing beamforming, subcarrier allocation and power control on a downlink signal based on the beamforming parameters, to optimize a communication performance of the MIMO-OFDM system.