Research On Detection Method Of NOMA-OFDM Signal

With the development of mobile communication,the scarcity of spectrum resources poses a challenge to the spectrum efficiency of orthogonal multiple access(Orthogonal Multiple Access,OMA)technology,while non-orthogonal multiple access(Non-orthogonal Multiple Access,NOMA)technology further improves the spectrum utilization by superimposing user signals at the transmitter.At the receiving end,due to the active introduction of multiple access interference(Multiple Access Interference,MAI)in the signal,it is difficult to detect the signal at the receiving end.The NOMA receiver separates the user’s signal through continuous interference cancellation(Successive Interference Cancellation,SIC)technology.At present,the traditional detection methods of uplink NOMA have the problem of error propagation(Error Propagation,EP).When NOMA is combined with orthogonal frequency division multiplexing(Orthogonal Frequency Division Multiplexing,OFDM)modulation,it is not robust to the length of cyclic prefix(Cyclic Prefix,CP)and nonlinear channel.Aiming at the problems existing in the traditional detection methods of uplink power domain NOMA-OFDM signals,the main research contents of this dissertation are as follows:1.The receiver uses the message passing(Message Passing,MP)algorithm based on factor graph to detect the uplink NOMA-OFDM signal.Firstly,based on the theory of message passing algorithm,the factor graph is used as an analysis tool to analyze the factor graph of NOMA-OFDM signal and the message passing in NOMA-OFDM system.Then,the message passing in the factor graph is iteratively updated to recover the NOMA-OFDM signal at the sender.The idea of SIC is combined in calculating the a posteriori probability,and the channel is estimated.The simulation results show that the receiver detection algorithm based on MP is more robust to signal distortion caused by ISI and nonlinear channel than ML algorithm.2.Aiming at the problem of error propagation in SIC algorithm,a detection method based on deep learning(Deep Learning,DL)is studied.A NOMA-OFDM signal detection model based on deep neural network(Deep Neural Network,DNN)is established.The hidden layer uses three full connection layers,Sigmoid,Re LU and Softmax activation functions.In the process of DNN model training,firstly,the feature vector of the received data is extracted and input into the model together with the label of the sender.The forward propagation and back propagation in the network are analyzed.At the output,soft detection is used to estimate the label that the sender may send according to the probability distribution,and the signal of the sender can be recovered at the same time.The simulation results show that compared with the traditional detection algorithm,the improved receiver has better detection performance in both linear and nonlinear channels,and solves the problem of error propagation.In addition,in the case of reducing CP,the traditional detection method can’t recover the signal well,and the model has strong robustness to the signal distortion caused by inter symbol interference(Inter-Symbol Interference,ISI),and still has good detection performance.