Precoding Matrix Selection Based On Artificial Intelligence

With the development of mobile communication technology,the 5th generation mobile communication(5g)has gradually entered the commercial stage.As one of the key technologies of 5g,massive threedimensional multiple input multiple output(3D MIMO)has become a research hotspot in recent years.This technology extends the traditional antenna array to two dimensions,so as to introducing the concept of pitch angle on top of traditional massive MIMO,which further improves the number of antennas and spatial degrees of freedom,exploits the potential of spatial multiplexing,and improves the spectral efficiency.With the cross application of artificial intelligence technology in various fields,it provides a new solution to the problems of large complexity,long running time and slow convergence speed encountered in traditional massive 3D MIMO beamforming technology.The main contents of this thesis are as follows:Firstly,this thesis studies the technical background related to 3D MIMO,including the antenna array structure of 3D MIMO,and typical 3D MIMO channel modeling and analysis process,and traditional massive MIMO precoding algorithm.Secondly,this thesis proposes a new 3D MIMO precoding codebook construction scheme based on the quantization of base station antenna array parameters.This technology obtains the quantified channel matrix set by quantifying the pitch and azimuth angle in the horizontal and vertical dimensions of the antenna array with strictly modeling the channel,then uses this channel matrix set to construct the codebook through the traditional 3D MIMO precoding algorithm,and then simulates and compares its performance comparing with the traditional 3D MIMO precoding algorithm.Finally,this thesis uses the codebook set and channel matrix set as the training set and test set of artificial intelligence model,and uses machine learning and deep learning to select and predict codewords,and then gets the codeword selection and prediction model based on artificial intelligence through training and testing.Simulation results show that compared with the traditional codebook-based global search and matrixbased pre-coding algorithms,this method significantly reduces the system complexity,shortens the running time,and it can approach the performance of traditional precoding algorithm.