Research On 5G Massive MIMO Downlink Channel Prediction Based On Koopman Operator And Empirical Mode Decomposition

The innovation and development of mobile communication has been advancing constantly,and the fifth Generation Mobile Communications(5G)technology is becoming more and more mature.Compared with previous communication technologies,5G uses Massive Multi-Input Multi-Output(Massive MIMO)technology.If the features of Massive MIMO,such as high spectral efficiency and low bit error rate,are to be fully realized,then accurate channel state information(CSI)should be obtained in time or in advance.Especially when the user is in the mobile state,the CSI will fluctuate sharply in a short time.So it is an effective way to support 5G high-performance service to predict the CSI of the next moment or the next time period from the channel data of the known moment.In this thesis,the channel prediction of Massive MIMO system with different mobile velocities is studied.The main work contents are as follows:The ray data from the ray file are combined with the statistical 3D MIMO channel model and the ray file contains the three dimensional position,path loss,delay and other parameters of each ray.Then the model combines the mobile user's distribution pattern and mobile mode.Finally,the channel data of users with different mobile speeds are output in 5G simulation platform.Aiming at the single point prediction of mobile user channel amplitude,this thesis proposes an adaptive Koopman operator prediction method based on empirical mode decomposition(EMD)grouping.This method takes the number of groups of subchannel responses decomposed as the threshold value to distinguish the prediction scenarios of Koopman operator finite dimensional approximation algorithm and Koopman operator algorithm based on EMD grouping.The simulation result shows that this method can adaptively predict low speed low complex channel amplitude and high speed high complex channel amplitude.The average accuracy of predicting the amplitude of mobile user channel from 5 km/h to 120 km/h is 98.10%,which is 0.56% and 1.13% higher than that of the traditional autoregressive(AR)prediction method and long short term memory(LSTM)network prediction method,respectively.Aiming at the continuous prediction of mobile users' channel capacity,a selective normalized Koopman operator prediction algorithm based on EMD is proposed in this thesis.In this algorithm,the average number of intrinsic mode functions(IMF)decomposed by channel capacity of users with different speed is used as a threshold to distinguish the prediction scenarios of the finite dimensional approximation algorithm of Koopman operator and the selective normalization Koopman operator algorithm.Through the simulation result,it can be seen that this algorithm can predict the channel capacity of mobile users from 5 km/h to 120 km/h adaptively,and the average accuracy is 82.54%,which is 9.82% and 4.89%higher than the traditional AR prediction method and LSTM network prediction method,respectively.