| Massive Multiple Input Multiple Output(MIMO),as one of the key technologies of the next generation mobile communication system,can significantly improve system throughput and spectrum efficiency.Massive MIMO systems serve different users by configuring hundreds of antennas at the base station.However,it is very challenging for the base station to recover transmitted signals due to the inter-user interference.Traditional signal detection schemes either have low detection accuracy or high complexity,so it is difficult to strike a good balance between detection accuracy and complexity.Deep Learning(DL)can process massive data due to its powerful learning ability.Model-driven DL obtains a dedicated neural network structure by expanding the existing algorithm model and optimize the parameters,which can effectively improve the performance.At the same time,model-driven DL contains few training parameters.Therefore,this paper will take the model-driven DL technology as the core to study the problem of signal detection at the receiving end of massive MIMO systems.Non-orthogonal Multiple Access(NOMA)technology enables multiple devices to communicate on the same time-frequency resource at the same time,which meets the requirements of "large-scale wireless network" and is a potential multiple access technology of B5 G and 6G.The introduction of NOMA into massive MIMO systems can significantly improve the performance of massive MIMO systems.However,each sub-channel is shared by multiple users,the sub-channels of different users are no longer orthogonal,and the number of available orthogonal resources is usually less than the number of service users,which greatly increases the difficulty of signal detection at the receiving end.In this paper,the signal detection problem of massive MIMO-NOMA system based on model-driven DL technology is studied,and the advantages of model-driven DL technology are fully utilized to improve the system performance.The main research work of this paper is as follows:Firstly,the signal detection scheme for uplink massive MIMO systems is proposed based on the model-driven DL method.First of all,the JC-Net detection network is obtained by improving the traditional damped Jacobi(JC)iterative method and learning the parameters in the detector through the neural network method,which avoids tedious calculations and makes the network interpretable.The JC-Net detector has the advantages of few learnable parameters,low computational complexity and fast convergence speed.Then,this paper further improves the performance of the JC-Net detector by increasing the dimension of the residual vector in JC-Net network.Finally,the model-driven idea is used in the Successive Over Relaxation(SOR)iterative detector to obtain the SOR-Net detection network.Simulation results show that the model-driven signal detection scheme proposed in this paper can achieve high detection accuracy with low complexity,thus solving the difficult problem of signal detection in massive MIMO systems.Secondly,in the uplink grant-free massive MIMO-NOMA system,this paper improves the Iterative Shrinkage Thresholding Algorithm(ISTA)based on the sparse characteristics of active users and the principle of compressed sensing,and applies it to signal recovery of active users.The ISTA used in this paper can detect signals without knowing the number of active users.Then,combined with model-driven DL technology,the improved ISTA is expanded and parameters in the model are optimized by neural network to improve the detection performance and speed up the convergence.At the same time,for the multi-antenna situation at the base station of the massive MIMO-NOMA system,the "first significant jump" scheme is innovatively applied to make full use of the advantages of a large number of antennas at the base station of the massive MIMO-NOMA system.Finally,the superiority of the proposed model-driven detection scheme in the uplink grant-free massive MIMO-NOMA system is verified by simulation experiments. |
PDF Full Text Request