Research On MIMO Detection And Channel Estimation Based On Neural Networks

Multiple-Input Multiple-Output(MIMO) is one of the most important technologies of wireless communication.Without increasing the system bandwidth and the transmitting power,MIMO can multiply the channel capacity by using diversity technique fully.Data detection algorithm and channel estimation algorithm are the two key technologies of the MIMO systems.The study which not only maintains optimal performance of the algorithms but also has the moderate computational complexity is of great significance to implement MIMO systems.In recent years,with the rapid development of the hardware coprocessors and the neural network training speed,artificial intelligence(AI),especially Deep Neural Networks(DNN),has become a research hotspot.Combining AI with wireless communication has become one of the trends of future development.In order to reduce the complexity of algorithm and improve precision,this thesis applies DNN to MIMO systems and proposes a data detection algorithm and a channel estimation algorithm.Firstly,this thesis reviews the development and characteristics of MIMO systems and DNNs,and focuses on the roles and characteristics of neurons,hidden layers,activation functions,loss functions and so on.Secondly,DNN is applied to the MIMO detection.During building the neural network model,residual networks are introduced to resolve the network degradation problem.Simulation results show that the performance of the DNN-based algorithm is better than that of the traditional linear data detection algorithms and is similar to that of the Minimum Mean Square Error Serial Interference Cancellation(MMSE-SIC)algorithm which has higher algorithm complexity.Finally,this thesis studies the MIMO channel estimation algorithms,and proposes a DNN-based MIMO channel estimation algorithm.Compared with the Least Square-Linear Interpolation algorithm,the DNN-based algorithm has a high precision.