A Research Of Efficient Precoder Design In MIMO Systems With Finite-Alphabet Inputs

With the increasing demands on wireless communication,how to improve the spectral efficiency of Multiple Input Multiple Output(MIMO)systems has gradually become the focus of many scholars.It has been shown that precoding the pre-transmission signal in MIMO systems can effectively improve the transmission efficiency of the system.Although many existing studies on MIMO precoder can provide good theoretical support for practical applications,they all have a prerequisite that the input is a Gaussian signal.However,in practical MIMO systems,the transmit signals token come from finitealphabet,and the direct application of algorithms designed based on Gaussian signals to practical MIMO systems can seriously affect the system performance.In recent years,some precoding algorithms for finite-alphabet input have been proposed,which effectively improve the spectral efficiency of MIMO systems,but there are still two shortcomings: first,It needs to solve the optimization problem iteratively,which is difficult to meet the real-time requirements of pratical system;second,only linear precoding schemes are considered,and there is still much room for improvement in nonlinear precoding schemes.In this paper,we focus on the precoding in MIMO systems with finite-alphabet inputs,and propose two efficient precoding schemes based on neural networks,and compare them with existing precoding algorithms through simulation experiments.The main research work of the paper is as follows.(1)To address the problem of high computational complexity of the current optimal linear precoding algorithm,a precoding algorithm based on the universal approximation property of neural networks is proposed.The main idea of this algorithm is to use the approximation property of neural networks to achieve the learning of the optimal linear precoding algorithm in order to improve the precoding efficiency.At the same time,different neural network models are designed and simulated for two scenarios: known complete channel state information(CSI)and known statistical CSI.The experimental results show that the algorithm can significantly reduce the precoding time and achieve almost the same performance as the optimal linear precoding algorithm,which has good practical applications.(2)Based on the optimal linear precoding algorithm,a nonlinear precoding algorithm based on neural network is proposed.The main idea of this algorithm is to use the transformation characteristics of neural network to precode the signal nonlinearly,and to achieve efficient precoding by constructing a loss function that can maximize mutual information.The experimental results show that the performance of the proposed precoding algorithm is significantly better than that of the linear precoding algorithm under both the real channel and the complex channel.