| Beamforming technology is one of the most important technologies in wireless communication technology.It has received extensive attention in the past few decades.The wide application of neural network has provided new research methods for beamforming technology.Adaptive beamforming technology combined with neural network technology can improve the anti-interference performance of the signal,which meet the high requirements of the communication system for the real-time signal in this thesis.Transmitting terminal receives mixed signal containing different modulation types in the wireless communication system.Therefore,a method of separating target signal and interference signal based on neural network is proposed.First,a neural network model is designed and trained,which can separate the target signal and the interference signal,then make deep neural network-based minimum variance distortionless response(DNN-MVDR)adaptive beamforming according to the interference signals of different energy intensities after separation.Then,singular value decomposition is introduced for signals with poor separation effect,and singular value decomposition-based minimum variance distortionless response(SVD-MVDR)is made.The results show that: the adaptive beamforming completed after neural network separation is 20～40d B higher in the target direction and the interference direction than the direct adaptive beamforming with the mixed signal,and the stronger interference signal energy,the greater the gain difference,the better suppression effect.The improved SVD-MVDR adaptive beamforming increases the gain difference between the target direction and the interference direction by 1～5d B compared to the DNN-MVDR adaptive beamforming after neural network separation.The nulling of the interference direction is deeper,which has a better suppression effect on interference.In addition,the current beamforming technology needs to be solved according to specific iterative algorithms and convex optimization algorithms.However,these iterative algorithms have shortcomings such as high computational complexity and extended calculation time,which results in beamforming technology that cannot satisfy the high real-time requirements of the communication process in certain scenarios.It is known that the traditional minimum variance distortionless response(MVDR)adaptive beamforming algorithm needs to get the weights according to the covariance matrix of the signal.Aiming at the large amount of calculation for matrix inversion,this thesis uses a fully connected neural network with fewer neurons to replace the process of signal covariance matrix inversion,and proposes an improved neural network-based beamforming algorithm.The network model is trained offline,and the network model parameters with better performance are saved for online generation of weights.Through calculation and comparison of the calculation amount of the two algorithms,it is found that the calculation amount of the beamforming based on the neural network is less than that of the traditional MVDR adaptive beamforming algorithm.In the same equipment experimental platform,the weights obtained by the neural network-based beamforming algorithm have higher accuracy,forming a deeper null at the interference.The neural network-based beamforming algorithm has a higher computing speed,and its algorithm speed is increased by 43.9%,which can improve the real-time performance of the signal in the communication system significantly. |
