Research On Low-complexity Massive MIMO Channel Estimation Methods

Massive MIMO technology has become the key technology of the 5th generation mobile communication system by increasing the number of antennas to improve the spectral efficiency,energy efficiency and reliability of the system.However,with large-scale antennas array,the channel will show spherical wave characteristics and spatial non-stationarity.With the increase of the number of system antennas,the power consumption and complexity of the system also increase,which brings great challenges to channel estimation.Therefore,low-complexity channel estimation methods are studied in this thesis,which focuses on the spherical wave channel characteristics and spatial non-stationary channel characteristics of massive MIMO system.Firstly,this thesis studies the spherical wave channel characteristics and non-stationary channel characteristics of massive MIMO system,and introduces theoretical derivation and the optimal pilot design conditions of the classical channel estimation methods,including the least squares and minimum mean square error,and the influence of antenna number on the performance of the two estimation methods is analyzed by simulation.Secondly,considering the communication scenario of uniform circular array antennas at both ends of the transmitter and receiver,the low-complexity linear minimum mean square error(LMMSE)channel estimation is studied by using the spherical wave channel characteristics of massive MIMO system under the conditions of known channel covariance information and unknown channel covariance information,respectively.It is shown by the research results that:1.When the channel covariance information is known,the computational complexity of LMMSE channel estimation can be reduced from5N~3-N~2+2N floating-point operations(FLOPs)to2N~3-N~2+3Nlog N+2N FLOPs by using the characteristics of spherical wave channel.2.When the channel covariance information is unknown,the computational complexity of LMMSE channel estimation can be reduced to2N~3-N~2+6N-1+(3/2)N(N+1)log N FLOPs by using spherical wave channel characteristics,which is lower than that of channel estimation when the channel covariance information is known and spherical wave channel characteristics are not used.Finally,for the nonstationary channel characteristics of massive MIMO system,it is assumed that each antenna of the base station is equipped with a pair of 1-bit analog-to-digital converter.The non-stationary characteristics are characterized by the visible regions between the sub-array and the user.In addition,a supervised deep neural network(DNN)model is researched based on the strong generalization ability of DNN.The simulation results show that the proposed network can achieve better estimation performance with less pilot,and achieve a good balance between performance and computational complexity.