Research On Signal Detection Algorithm Of MIMO-NOMA System Based On Deep Learning

Non-Orthgonal Multiple Access(NOMA)technology can improve spectral efficiency by communicating with multiple users under the same time-frequency resource;MIMO technology can effectively overcome channel fading and improve the capacity and reliability of the system,so MIMO-NOMA has become one of the key technologies of mobile communication.In MIMO-NOMA systems,serial interference cancellation(Successive Interference Cancellation SIC)is the main signal detection method.However,SIC has some limitations in practical applications,such as error propagation and large calculation amount.To this end,this paper proposes a signal detector method based on deep learning,the main work is as follows:Firstly,the MIMO-NOMA downlink signal detection algorithm based on Deep Neural Networks(DNN)is adopted,and DNN is used instead of the fragmented channel estimation and signal detection module,specifically,the pilot data is used as a label,the MIMO-NOMA received signal is used as the input signal,the signal detection is realized in a classified manner,and the cross-entropy loss function is used to train the DNN network.The bit error rate performance in B PSK and QPSK modes is simulated,and compared with the traditional receiving algorithm.The simulation results show that the DNN-based receiver has better performance,lower bit error rate and strong robustness.In order to better improve the system performance,this paper tries to increase the number of DNN network layers.In the simulation,it is found that when the number of network layers reaches eight layers,the system performance cannot be effectively improved,and even will slightly decreased,that is,the network is degraded.In order to solve this problem,considering the characteristics of the residual network,this paper puts forward a signal detection algorithm based on the residual network,in the introduced on the basis of the residual network principle of the signal detection model based on the residual network,the model for the first input layer,the back layer for a full connection layer,add two residual block,the last three layers for full connection layer.Finally,the performance of the proposed receiver based on DNN network,residual network and traditional algorithm is simulated and analyzed.The simulation results show that the signal detection algorithm based on residual network performs better than DNN network and traditional algorithm detection algorithm under BPSK modulation mode,and the bit error rate performance of all receiving algorithms decreases under QPSK modulation mode,but the signal detection algorithm based on residual network is still the algorithm with the highest "cost performance".