Research On Mechanism And Method Of Compressed Sensing CSI Acquisition In FDD Massive MIMO System

Massive MIMO is one of the most important technology for the 5th generation mobile communication system.Both the base stations and the user terminals in the massive MIMO system are equipped with antenna arrays of large dimensions.This deployment introduces extra spatial freedom into the system,providing that the massive MIMO system has the potential to achieve a much higher transmission throughput Consequently,it is of great importance and difficulty to design new channel acquisition solutions to provide base stations and user terminals inside the massive MIMO system with the ability to acquire accurate enough channel state information under a relatively low pilot consumptionIn the downlink training process of the FDD massive MIMO system,the limited pilot consumption is conflicted with both the great antenna numbers of the base station and the necessity of the accurate acquired channel state information.To solve this problem,we do researches on the following three aspects1)The channel model of the massive MIMO OFDM system is established.Then the virtual angular representation and the virtual angular transform is introduced into this channel model to build up the relationship between the virtual angular domain channel impose response and the spatial domain channel impose response.Besides,the channel interpolation property and the channel training structure of the OFDM system are established.At last,the sparsity of the virtual angular domain channel is analyzed2)We find out the relationship between the reconstruction efficiency and the indeterminancy-sparsity trade-off performance of compressive sensing algorithms Based on this result,GAMP(generalized approximate message passing)algorithm is picked up as the channel estimation architecture and extended to the complex targets,forming the C-BG-GAMP(complex Bernoulli-Gaussian generalized approximate message passing)algorithm3)We define the zero-partiton information and zero-partition ration for the sparse targets.Besides,the influence of the zero-partiton information to the reconstruction efficiency of C-BG-GAMP algorithm is analyzed.This result is proven by the simulations.In addition,we design a zero-partition information learning algorithm and analyze its effectiveness.At last,we revise this zero-partition information learning algorithm in the application of channel estimation to build up the zero-partition enhanced C-BG-GAMP channel estimation solutionOur researches on the first aspect concentrate on the channel modeling and channel sparsity.The sparsity estabilish the channel estimation utilization of compressive sensing techinique.Works on the C-BG-GAMP(the second aspect)extend the GAMP algorithm to the complex targets.These researches lay the foundation of the improvements of the C-BG-GAMP channel estimation,which belong to the third aspectWith the same pilot consumption,our new solution is provided to have better accuracy performance than the existed M-SBL(multi-task sparse Bayesian learning)and J-OMP(joint-orthogonal matching pursuit)channel estimation solution,as well as the original solution which only the C-BG-GAMP directly.