Signal Detection Based On Deep Learning For Massive MIMO System

With the development of fifth-generation(5G)mobile communication,massive MIMO has received a lot of attention from academia and industry.By increasing the number of antennas,massive MIMO can greatly improve spectral efficiency and energy efficiency,and then become one of the key technologies of 5G.However,as the number of users and antennas at the base station increase,the complexity and power consumption of the signal process will also increase significantly.Moreover,due to the complex channel models and high-dimensional channel matrices,traditional signal detections will face significant challenges.Therefore,it is vital to study low-complexity and high-performance signal detection algorithms.In recent years,deep learning has achieved great success in the field of image and speech processing.It has such a strong ability to learn deep network structure that can achieve the approximation of complex functions.Combining the advantages of deep learning with massive MIMO provides a new solution for signal detection.The main work of this paper is as follows:At the beginning,this paper studies the massive MIMO uplink signal detection based on fully connected layer and residual network under various channels and different modulation modes.Specifically,a simple deep neural network is built using the fully connected layer,and then the number of network layer is determined through experimental simulation.Then,the trained network is compared with the traditional method for BER performance.According to the simulation results,it is difficult to obtain the optimal BER performance for the simple structure of the fully connected layer.Therefore,this paper re-selects the convolutional neural network to build a residual network structure which is more suitable for massive MIMO signal detection,and at the same time,add the channel state information into the training process so that we can improve training efficiency.Finally,we compared the BER performance of the fully connected layer-based method,the residual network based signal detection algorithm and the traditional method experimentally.The simulation results show that under the condition of millimeter wave flat fading channel and ill-conditioned channel environment,the algorithm based on residual network can achieve optimal performance using various modulation methods.In the second part,in order to reduce the cost of massive MIMO,this paper further studies the signal detection combined with low-resolution ADC based on the previous algorithm.First,the input signal of the residual network is quantized,and then experimental simulation is performed in the Rayleigh flat fading channel and the ill-conditioned channel environment.The results show that the residual network is equally suitable for signal detection in low resolution ADC systems.Moreover,the deep learning based detection method has better error performance under lower ADC quantization bits than the traditional linear detection method.