Research On Multi-user High-dimensional Channel Clustering And Dimension Reduction Method For Large-scale MIMO Two-layer System

With the rapid growth of mobile data traffic,in order to improve the quality of service of 5G,dense small cells are introduced into massive MIMO macro-cellular networks,so that the same spectrum can be shared in a two-tier network to improve the spectrum efficiency,obtain higher system capacity and optimize network coverage.Furthermore,utilizing the short-distance coverage,low-power,and high-rate characteristics of the small cell base station to provide high-speed data transmission for hotspots.However,there are serious interferences between large-scale MIMO macrocell and small cells,and multi-user high-dimensional channels.In this paper the interference existing in large-scale MIMO dual-layer network systems is studied as follows.Firstly,in order to study the multi-user distributed cluster geometry model in large-scale MIMO two-tier system,an improved K-means user clustering algorithm based on maximum-minimum distance method is proposed.The algorithm propose a weighted likelihood function as the similarity measure between a user and a group,and partition the users into same group that have similar channel correlation matrices,and the user channel characteristics of different clusters have large differences so as to reduce multi-user interference.The simulation results show that compared with the traditional clustering algorithm with chordal distance similarity measure,the improved algorithm has lower complexity,higher efficiency,and can greatly improve the system capacity.Then,for large-scale MIMO two-tier systems,when the number of antennas at the base station is approaching infinity,the channel transmission matrix and precoding matrix require high-dimensional operations,which leads to the high computational complexity of the system.A dimensionality reduction method based on manifold learning is proposed to reduce the dimensionality of multi-user high-dimensional channel.LLE algorithm is used to embed low-dimensional user channel information,reducing the system computational complexity and maintain the users’ original signal space domain characteristics.The simulation results show that LLE algorithm has certain advantages on low dimensional features extraction for users raw data,and it can not only reduce the computational complexity of the system,but also makes the macrocell layer system capacity has been improved.Finally,in order to reduce the bit error rate of large-scale MIMO two-tier system and improve the system throughput,a low-complexity two-stage precoding algorithm is studied based on the user channel transmission matrix after dimensionality reduction.By designing the pre-beamforming matrix,separated effectively different user groups in space to suppress inter-group interference,and precoding scheme operates on the lower dimensional transformed channel to eliminate the intra-group interference.As a consequence of pre-beamforming,concentrating transmission energy only in certain directions while implicitly mitigating interference caused to the small cells located in the other directions.Simulation results show that compared with the ZF precoding scheme,the proposed algorithm increases the total capacity of the two-tier system.At the same time,pre-beamforming matrix of the proposed algorithm is related to the number of main eigenvalues of the covariance matrix,and when the SNR is not less than 10 dB,the larger the number of main eigenvalues,the higher the system sum rates.