| In the development of mobile communication technology,the performance index of wireless communication become more and more stringent for people.However,spectrum resources are limited and highfrequency communication technology is limited in some communication scenarios.How to use efficiently high-quality spectrum resources is of great significance.As one of the key technologies of 4th-Generation(4G)and 5th-Generation(5G)communication systems,Multiple Input Multiple Output(MIMO)wireless transmission technology can significantly improve the spectral efficiency.The receiver needs to estimate the best channel state information(CSI)for feedback to obtain higher spectral efficiency for the transmitter.Precoding Matrix Indicator(PMI)and Rank Indicator(RI)are closely related to precoding technology.Selecting accurate PMI and RI effectively will dramatically improve the system throughput.However,the existing channel measurement and feedback technology still has some limitations.One is that the channel fading and communication delay affect the accuracy of feedback.The other is that the large scale precoding codebooks cause the high computational complexity with the antennas increasing.In recent years,deep learning technology has been widely used in the field of communication signal processing,and has achieved better performance than traditional methods in some aspects.Firstly,this thesis investigates the research status of wireless channel prediction,channel measurement feedback and the application of artificial intelligence technology in the field of signal processing to understand the existing technology and the latest research direction,and then introduces the channel fading characteristics,the principles and algorithms of channel measurement and deep learning,which provides theoretical support for the channel measurement feedback scheme based on deep learning proposed in this thesis.Secondly,aiming at the problems of channel fading and communication delay,this thesis designs a long-range channel response prediction scheme based on Long Short Term Memory(LSTM)neural network model.Through off-line training of the LSTM neural network model,the model can learn the change rule of channel response over time,and obtain the ability to predict the future channel response.In practical use,the channel response with a certain time delay is obtained through the model,and the feedback calculation is carried out according to this result,so as to effectively reduce the error of PMI/RI feedback and improve the capacity performance of the system.Finally,for the problem of PMI/RI selection,this thesis designs the PMI/RI selection schemes for codebooks for 2,4,8 antenna ports respectively based on the fully connected neural network model to reduce the complexity of searching codebook.The constructed neural network model makes full use of channel information and noise information to learn PMI/RI feedback rules based on maximum mutual information.We select RI with neural network at first,cluster the codebook,and then classify the codebook with neural network,so as to reduce the search space of codebooks,and the complexity of calculation is reduced eventually.The results show that the proposed channel response prediction scheme has higher accuracy than the scheme based on autoregressive model and the proposed PMI/RI selection scheme can effectively reduce the computational complexity without obvious performance loss when the number of antennas is large.The combination of the two methods can effectively improve the overall performance of the actual system,especially under the poor noise condition. |
