Tensor Decomposition Based Channel Estimation In Mmwave MIMO System

With the development of Internet technology,the demand for wireless data services is dramatically increasing.It brings more challenges to the design of physical layer wireless communication systems.In order to provide users with faster,more reliable,more energy-efficient data transmission services,the transmission frequency band for wireless communications tends to be extended from conventional sub-5GHz to the high-frequency as millimeter-wave.Therefore,millimeter-wave communication technology,as one of the core technology for the next generation cellular networks,is the consensus between industry and academia.In order to improve the spectrum efficiency and overcome the path loss in mmWave transmission,both the base station(BS)and the mobile station(MS)need to employ multi-antenna which can provide degree of freedoms gain.At the same time,to reduce the energy consumption of the hardware in radio frequency(RF),the number of RF channels is usually designed to be less than the number of antennas.Furthermore,to effectively perform precoding / beamforming,the BS and MS need to obtain the corresponding channel state information respectively.For the frequency division Duplex(FDD)system,the transmitter(BS or MS)needs to transmit the pilot signal to the receiver.The receiver estimates the channel through the pilot symbols,and then the estimated channel is fed back to the transmitter,ensuring that the transmitter can use channel state information(CSI)for transmission precoding design.For a Time Division Duplex(TDD),the transmitter can utilize the channel reciprocity in precoding design.For the FDD and TDD system mentioned above,this dissertation mainly discusses the scenario,in which the BS as the data transmission side.Thus,in the FDD system,the MS estimates the downlink CSI used for feedback.In the TDD system,the BS received the uplink pilot to obtain the CSI used for downlink multi-user transmission.Furthermore,since both BS and MS are employed with hybrid structure,the conventional orthogonal pilot pattern will introduce serious training overhead.To avoid this problem,we propose a series of algorithm which can take advantage of the limited scattering of multi-antenna millimeter-wave band channel,which means the transmission channel can be described by finite parameters.This feature is also namely,the sparseness of the multi-antenna channel in angular and time domain.This dissertation mainly discusses the training overhead reduction as well as utilizing channel sparsity.In this dissertation,the tensor based pilot transmission architecture has been proposed.The tensor decomposition is employed as an effective mathematical tool to deal with the basic channel estimation problem under FDD and TDD transmission scheme.For the channel estimation problem in FDD,this dissertation proposes a three-dimensional low-rank tensor pilot structure which is based on the inherent structure of multi-carrier multi-antenna transceiver channel.Meanwhile,it proposed a channel estimation algorithm based on the theorem of tensor decomposition and corresponding uniqueness condition.For the TDD channel estimation problem,this dissertation first discusses the uplink channel estimation problem of single carrier multi-antenna system.Based on the inherent three-dimensional low rank tensor structure in multi-user multiantenna transceiver channel,the channel estimation problem also has been formulated as a tensor decomposition algorithm.In addition,the uplink channel estimation method based on tensor in multi-carrier multi-antenna system is briefly discussed.For the tensor decomposition based channel estimation algorithm,we proposed an effective criteria for pilot design by the condition of uniqueness condition of tensor decomposition.At the same time,it is possible to effectively predict the ratio of pilot compression.Furthermore,to compare the proposed method with conventional ways,this dissertation also discussed the channel estimation methods thorough direct compressed sensing.Due to the efficient use of the multi-dimensional low-rank structure of the millimeter-wave channel,it has been demonstrated that the tensor processing method has lower computational complexity and better recovery performance of the channel.