一种NOMA系统中的反馈深度神经网络检测算法

A Feedback Deep Neural Network Detection Algorithm in NOMA Systems

在上行非正交多址(Non-orthogonal Multiple Access,NOMA)系统中,针对传统基于串行干扰消除(Successive Interference Cancellation,SIC)检测存在同个时频块内用户间干扰的问题,提出了一种新型的NOMA检测算法。通过将SIC检测的反馈消除结构和深度神经网络结合起来,设计出了一种新型的反馈深度神经网络(Feedback Deep Neural Network,FDNN)结构。FDNN模型分为两个模块,检测模块通过深度神经网络实现非线性检测,反馈模块通过权重矩阵重构信号并消除用户干扰。通过重复检测和反馈过程,FDNN依次检测出各个用户的符号,并达到了良好的性能。仿真结果表明FDNN检测算法相较于SIC检测具有更低的误符号率和误比特率,并验证了其具有更良好的抗用户间干扰的性能。

In order to address the problems of inter-user interference at the same time-frequency block in uplink non-orthogonal multiple access(NOMA)systems with successive interference cancellation(SIC)receiver,a novel detection algorithm is proposed.Combining the feedback elimination structure of SIC detection with deep neural network(DNN),a feedback DNN(FDNN)is designed.The FDNN has two modules.The detection module performs non-linear detection through DNN,and the feedback module reconstructs the signal through weight matrix and eliminates user interference.By repeating the detection and feedback process,FDNN detects the symbols of each user in turn,and achieves good performance.Simulation results show that FDNN detection algorithm has lower symbol error rate and bit error rate than SIC detection,which verifies its better performance against inter-user interference.