Design Of Hybrid Beamforming Algorithm For Non-orthogonal Multiple Access Millimeter Wave Communication Based On Deep Learnin

The commercialization of 5G is accelerating globally,and various countries and organizations have started research on 6G networks.Non-Orthogonal Multiple Access(NOMA)is a wireless communication technology that allows multiple users to share the same frequency band,thereby improving the coverage and capacity of wireless networks and supporting a larger number of user connections.Therefore,NOMA is considered as an alternative technology to enhance the spectrum efficiency of future 5G and even 6G networks.Additionally,millimeter-wave communication can effectively address the issue of insufficient spectrum resources.However,challenges such as attenuation and multipath effects during the propagation of millimeter waves pose significant obstacles to the application of millimeter-wave communication.In this context,Multiple-Input-Multiple-Output(MIMO)technology can be introduced for compensation.In millimeter-wave MIMO systems,hybrid precoding schemes can reduce the number of radio frequency(RF)chains without sacrificing system performance excessively.However,reducing the number of RF chains also means a decrease in the number of served users.Therefore,introducing NOMA technology into millimeter-wave massive MIMO systems can further increase the number of served users.Nevertheless,the rapidly changing channel conditions and complex spatial structures of millimeter-wave MIMO-NOMA introduce significant complexity and hinder its application.Consequently,deep learning-based millimeter-wave MIMONOMA systems have become a hot spot.The design of millimeter-wave MIMO-NOMA hybrid beamforming algorithms based on deep learning can utilize neural network models to achieve end-to-end beamforming,avoiding the cumbersome parameter tuning required by traditional methods.Moreover,deep learning-based methods can adaptively model and process complex communication channels,thereby improving system robustness and performance stability.Existing studies have shown that deep learningbased millimeter-wave MIMO-NOMA hybrid beamforming algorithms can significantly reduce communication latency and computational complexity while ensuring communication quality,demonstrating promising practical application prospects.This thesis proposes a deep learning-based MIMO-NOMA hybrid beamforming algorithm.Specifically,deep learning techniques are applied to the Multilinear Generalized Singular Value Decomposition(ML-GSVD)method to adaptively select beamforming parameters for MIMO-NOMA systems.The specific work is as follows:First,a neural network is used to fit the ML-GSVD algorithm to demonstrate the feasibility of the proposed approach.The channel matrix is taken as the input of the neural network,and the decomposition process of the ML-GSVD is approximated.The neural network is trained using a loss function to obtain accurate results.This process can be regarded as a fully digital millimeter-wave MIMO-NOMA beamforming design process.Next,power constraints and decoding order constraints in NOMA are introduced,modeling the MIMO-NOMA beamforming design as an optimization problem involving the sum data rate.The main optimization objective and constraints are treated as the primary and penalty terms of the neural network,respectively,guiding the training of the neural network to obtain the optimal beamforming design.Different schemes and parameter settings are compared,and their complexity in millimeter-wave MIMO-NOMA systems is analyzed.Simulation results show that the proposed deep learning-based millimeter-wave MIMO-NOMA beamforming algorithm is comparable to fully digital beamforming solutions in terms of performance.Furthermore,it shows improvements compared to traditional ML-GSVD algorithms and other deep learningbased approaches in terms of performance...