Cluster Analysis And Modeling Of The Massive MIMO Channels

With the growing requirements of mobile services,the 5th Generation(5G)wireless communication systems propose the requirement goals such as the peak rate reaching 10 Gbps,user experienced data rate up to 100 Mbps,etc.To achieve these goals,the Massive Multiple Input Multiple Output(Massive MIMO)technology,as one of the 5G key technologies,has been widely researched all over the world.By configuring dozens or even hundreds of antennas at the base station,it has improved the transmission rate,transmission quality,and anti-interference ability among cells and users significantly.It is also promised to be energy-efficient and spectrum-efficient.However,with the increasing number of antennas,the size of the antenna array becomes larger,relatively,which makes the plane wave assumption unsuitable to the propagation of the radio waves anymore.Besides,each element of the larger antenna array cannot observe all the multipath clusters in the channel.Thus,the characteristics and the theoretical models of the conventional MIMO channels cannot describe the massive MIMO channels accurately anymore.The characteristics of clusters in the massive MIMO channels can reflect non-stationary characteristics of the channels in the spatial domain,which is an important part of the channel measurement analysis and modeling researches.To support the development of massive MIMO,the research on its channel has to be performed.As an important part of channel measurement analysis and theoretical modeling,cluster characteristics can reflect the spatial non-stationary of the channels effectively.Considering this,the issues of this paper are given as follows:1.The cluster characteristics of the massive MIMO channels.The current changes of clusters when the number of antennas increases are still unknown.Besides,the research work lacks comparisons of massive MIMO channels in different scenarios.Considering this,firstly,the author designed and performed the measurement campaigns among three typical scenarios:urban macro(UMa),urban micro(UMi)and indoor hotspot(InH).Then the variations of clusters with the increasing number of antennas from 32 to 256 are shown.With the increasing number of antennas,the positions of cluster power become more concentrated,ASD and DS become larger while ASA becomes smaller.In the different positions of the antenna array,the observed clusters present different distributions.The above observations give a direct sight of the spatial non-stationary characteristics of massive MIMO channels.2.The cluster evolution-based massive MIMO channel modeling.Aiming at the influence of the geometry structure of the antenna array and the evolution of clusters,the variations of the number of clusters are simulated based on the birth-death process.So does the changing of clusters on the different elements of the antenna array.Then a novel massive MIMO channel model is proposed,which considers the effect of changing clusters in the spatial coordinate system.From the simulations and theoretical analysis,we can see that the spatial non-stationary characteristics fit the real conditions well.3.New feature updates to the ITU-R IMT-2020 channel model.The ITU-R standard channel must follow up on the characteristics of the massive MIMO channels and update its features.To make the channel model closer to the real conditions,two advanced features called spatial consistency and random cluster number are introduced for spatial non-stationary characteristics.Based on the primary module of the IMT-2020 channel model,the spatial consistency module is added for simulating the continuous and smooth variations of clusters during the movement,while the random cluster number is added for simulating non-stationary variations of the number of clusters in the channels.4.The optimization of the clustering algorithm.In terms of the design of clustering algorithms,the original K power means algorithm can only be calculated independently in each snapshot,which implies that the variations of clusters between continuous snapshots cannot be directly calculated.To continuously parameterize the results of clusters in the mobile scenarios,the K power means algorithm is optimized.Kalman filter algorithm is combined with the original algorithm to form an improved clustering framework.It is characterized by the ability to perform the Kalman gain weighting on the predicted value at the next snapshot and the measured value at the next snapshot to obtain the optimal solution.After the optimization,the variations of clusters become more frequent since the resolution of clustering results is improved.Another two optimized clustering algorithms are proposed to decrease the computational complexity in the massive MIMO channels.One is called Gradient Boosted Decision Tree(GBDT)-based algorithm,which can form a clustering model by training and validating.The other one is called Density Peaks(DP)-based algorithm.It only calculates the neighboring density of each multipath component(MPC)and the distances to other candidate cluster centroids,then uses a decision graph to judge the cluster centroids.Both of them can decrease the computational complexity significantly.In conclusion,this paper focuses on the propagation characteristics of clusters in massive MIMO channels.Firstly,the author analyzes the spatial non-stationary characteristics in massive MIMO channels.Secondly,a channel model is introduced which is based on the measurement analysis.What is more,the author validates the spatial consistency and random cluster number to make the IMT-2020 channel model closer to the real channel conditions.At last,three clustering algorithms are given for optimization,respectively.By the above research work and corresponding conclusions,a reference for improving the design of channel measurement campaigns,channel model simulations and performance evaluations in massive MIMO channels is given.