Research On Deep Learning Based Antenna Selection And Signal Detection In DM-GSM System

Spatial modulation(SM)technology is a nascent wireless communication technology that can significantly improve system spectral performance and transmission at high quality.In a multiple-input multiple-output system,SM employs the spatial location of antennas to facilitate extra bit transmission,resulting in better spectral efficiency and avoidance of issues such as inter-channel interference and inter-antenna synchronization.Dual mode generalized spatial modulation(DM-GSM),which leverages all the benefits of SM while also optimizing antenna selection and symbol modulation methods,is aimed at further improving the data transfer rates and reducing the bit error rate.With the increase in the number of antennas,the system computational complexity increases dramatically.Thus,efficient antenna selection and signal detection techniques are crucial for boosting DM-GSM system performance and reducing complexity.With the ongoing advancements and increasing utilization of artificial intelligence technology in communication,it holds significant importance to incorporate deep learning(DL)techniques in areas such as antenna selection and signal detection for multiple-input multipleoutput systems.By doing so,it is possible to significantly enhance the performance of communication systems while minimizing computational complexity.As a result,this research thesis adopts two typical neural network structures,deep neural network(DNN)and convolutional neural network(CNN),to investigate the antenna selection and signal detection algorithms of DM-GSM system.The primary focus of this study is to:(1)The traditional signal detection algorithms in DM-GSM system have limitations in maintaining a balance between computational complexity and system performance.In this thesis,we proposed a novel signal detection algorithm based on DL to overcome this problem.The signal detection algorithm based on DNN is firstly proposed to capture the mapping relationship between the input feature vector and the information bits.This trained model is then used for real-time online detection.To adapt to various channel environments and noise conditions,we also proposed a signal detection algorithm based on CNN in DM-GSM system.This algorithm reshapes the received signal and channel matrix into a two-dimensional input matrix at the receiving end.The two-dimensional convolutional layer is then used to fully extract local features.The simulation results illustrate that both of the proposed detection algorithms can achieve an approximate optimal detection performance at a low computational complexity.(2)Aiming at the problem of improving the overall performance of DM-GSM system,a DNN-based antenna selection algorithm is presented.The proposed method transforms the antenna selection issue into a multi-classification problem,utilizing a DNN network to select the antenna with higher channel gain while preserving the spectral efficiency.The channel state information(CSI)of DM-GSM system is firstly obtained,and the dataset is constructed the normalization and multiclassification label calibration operations,while the constructed network model is trained to obtain the classification model.The simulation results demonstrate that the proposed DNN-based antenna selection algorithm exhibits higher classification accuracy,while the complexity is significantly lower compared to the optimal exhaustive antenna selection algorithm.(3)To overcomes the issue of system performance degradation in correlated channel conditions,we proposed a DM-GSM antenna combination selection algorithm based on deep learning.The algorithm integrates two techniques,deep neural network and correlation analysis,to determine the optimal transmitting antenna by utilizing the output of the DNN.Furthermore,it selects the antenna combination with a low correlation degree through correlation calculation to maximize the channel gain while improving the discriminability among antenna combinations.The results demonstrate that the proposed algorithm takes into account the high channel gain and low correlation between antenna combinations as the number of antennas increases,and effectively balances the system performance and complexity of computation in the correlated channel environment.