Machine Learning Implementation Method Of Rate Adaptive Precoding For Multi-user MIMO Transmission

In MIMO transmission,the number of streams determines the transmission rate,and the multi-user precoding determines the transmission quality.Since the MIMO channel is time-varying,the number of streams and the multi-user precoding must be adjusted adaptively as the channel changes.This paper proposes an optimization algorithm based on machine learning around the joint optimization of the number of streams for MIMO transmission and multi-user precoding.The main works of the paper are as follows:Firstly,in the single-user downlink MIMO transmission scenario,with the weighted block transmission accuracy as the target,the MCS,the number of streams and the multi-antenna transmission power are jointly optimized.The main works are: 1)Through the interpolation simulation method,under different stream numbers and MCS,the parameter values of the function relationship between the block error rate and the equivalent signal-to-interference-noise-ratio are determined,and the weighted block transmission accuracy is the objective function,and the MCS,the number of streams and the multi-antenna transmission power are the optimization variables,the optimization problem is established;2)Decompose the optimization problem into three sub-problems:power optimization,MCS optimization,and stream number selection,then using interior point method,dichotomy and exhaustive method to solve iteratively;3)PDRL reinforcement learning algorithm based on machine learning is proposed to solve the original problem,and the reinforcement learning update method of parameters is given.Secondly,in the multi-user MIMO downlink transmission scenario,with the goal of maximizing the total achievable rate of the system,the precoding matrix and the number of streams are jointly optimized.The main works are: 1)Using the precoding matrix and the number of streams as optimization variables,taking the total achievable rate of the system as the objective function,a mathematical optimization problem model is established under space constraint and power constraint;2)In order to solve this non-convex hybrid optimization problem,it is decomposed into two sub-problems of precoding matrix optimization and stream number selection optimization;For the sub-problem of precoding matrix optimization,the closed-form solution of the optimal precoding matrix under a fixed number of streams is obtained through approximate transformation;For the stream number optimization sub-problem,a stream number selection algorithm based on the DQN reinforcement learning method is proposed.Thirdly,in the multi-user MIMO downlink relay cooperative transmission scenario,with the goal of maximizing the total achievable rate of the system,the base station precoding matrix,the number of streams,and the relay beamforming matrix are jointly optimized.The main works are: 1)Taking the total achievable rate of the system as the objective function,taking the base station precoding matrix,the number of streams and the relay beamforming matrix as the optimization variables,a mathematical optimization problem model is established under the constraints of space and power;2)This problem is transformed into an equivalent optimization problem,and the equivalent optimization problem is divided into two sub-problems,namely the joint optimization of base station precoding and relay beamforming matrix and the optimization of the number of streams;In the joint optimization of base station precoding and relay beamforming matrix,it is proved that when only one of the matrix variables is optimized,it is a convex problem and solved by the KKT condition;For the stream number optimization problem,the stream number selection algorithm based on the DQN reinforcement learning method is proposed,and the reinforcement learning update method of the parameters is given.Fourthly,in the multi-cell MIMO downlink RIS collaboration scenario,with the goal of maximizing the total achievable rate of the system,the precoding matrix of all users,the number of streams and the RIS phase shift angle are jointly optimized.The main works are:1)Taking the total achievable rate of the system as the objective function,taking precoding matrices,the number of streams and the RIS phase shift angle as optimization variables,the optimization problem is established under the constraints of space,power and phase shift angle;2)The optimization problem is transformed into an equivalent optimization problem,and the equivalent optimization problem is divided into two sub-problems,namely the joint optimization of precoding and RIS phase shift angle and the optimization of stream number;In the joint optimization of precoding and RIS phase shift angle,the KKT condition and the MM algorithm are used to update the precoding matrix and the phase shift angle in a single step,the precoding matrix and the phase shift angle are solved iteratively alternately;for the stream number optimization problem,a stream number selection algorithm based on the multi-agent DQN reinforcement learning method is proposed and the reinforcement learning update method of the parameters is given.