The Research And Implementation Of Deep Learning-Based CSI Non-Ideal Feedback Scheme

Massive multi-input multi-output(MIMO)has become the key technology of the 5G and Beyond 5G mobile communication.In the frequency division duplexing(FDD)mode,since the uplink and downlink channels are not reciprocal,the base station can only obtain channel state information(CSI)through channel feedback and then perform channel adaptive technologies such as downlink precoding and resource allocation,which will cause colossal system overhead.In recent years,the application of deep learning in the field of communication has also become a research hotspot.This paper will focus on using deep learning methods for channel feedback technology in massive MIMO systems in FDD mode.This paper first focuses on the feedback accuracy of the deep learning scheme in the non-ideal feedback channel,models non-ideal factors in the feedback channel,analyze how they affect deep learning-based feedback schemes,and designs the feedback network structure ATNet and error correction module ECBlock with more advantageous performance.Then the two-step training strategy is proposed to be used in conjunction with ECBlock to maximize the error correction effect.Simulation results show that the combination of ECBlock and training strategy can significantly reduce the impact of bitstream errors on feedback performance,increasing the robustness of DL methods in the non-ideal feedback channel.The results were published in IEEE Wireless Communications Letters.This paper also studies the deep learning-based channel feedback method in the scenario of multiple receiving antennas on the user side.For multiple receiving antenna situations,the correlation between the CSI of each antenna is analyzed,and a synergistic network SynNet is designed to improve the feedback accuracy by using the correlation between the antennas.The simulation compares the effect of using SynNet under a different number of receiving antennas.The experimental results show that SynNet can effectively improve the CSI feedback accuracy of multiple receiving antennas in different antenna number scenarios.Finally,this paper considers the practicality of the deep learningbased feedback scheme and combines the non-ideal feedback scenario scheme ECBlock with the multiple receiving antenna scheme SynNet to further expand the applicability of the deep learning scheme.The above scheme is integrated into the existing link-level simulation platform,and the link-level performance is compared between the deep learning-based feedback scheme and the traditional codebook scheme.The simulation results show that the deep learning-based scheme is better than the codebook scheme in the case of non-ideal feedback scenarios with multiple receiving antennas,which further illustrates the feasibility and practicality of the above deep learning-based scheme.