Research On Channel Estimation For Mmwave MIMO Systems Under Low-resolution ADCs

Massive multiple-input multiple-output(MIMO)and millimeter-wave(mm Wave)communication technologies play a vital role in fifth-generation(5G)mobile communication systems because they can significantly increase communication capacity and provide greater bandwidth.However,the excessive power consumption caused by a large number of RF chains and highresolution analog-to-digital converters(ADCs)in a mm Wave massive MIMO receiver is unaffordable,as the power consumption of a typical ADC scales linearly with the bandwidth and grows exponentially with the quantization bits.Therefore,using a low-resolution quantization ADC in the receiver is a promising solution to effectively reduce power consumption.However,severe non-linear distortion caused by low-resolution quantization will challenge the receiver to obtain accurate channel state information(CSI).This paper focuses on channel estimation for mm Wave MIMO systems with low-resolution ADCs,the aim of which is to improve the estimation performance as much as possible and reduce the required pilot overhead.Two estimation methods are mainly included,which are based on statistical learning and deep learning,respectively.First,we study the basis of common channel estimation methods for mm Wave MIMO systems.Due to the sparse distribution characteristics exhibited by millimeter-wave channels,many millimeter-wave channel estimation methods are proposed based on the theory of compressed sensing(CS).We first study the basic principles of CS,and then investigates the current application status of CS in massive MIMO channel acquisition.In addition,based on Bayes' theorem,many iterative sparse signal recovery algorithms complete signal reconstruction by inferring the posterior probability distribution of the signal to be estimated.We further explore several signal recovery algorithms based on Bayesian inference framework and the basic principles and algorithm steps of the Belief Propagation(BP)algorithm,Approximate Message Passing(AMP)algorithm,Expectation Propagation(EP)algorithm are digested.Furthermore,the advantages and disadvantages as well as applicable conditions of different algorithms are analyzed.Then,We propose a channel estimation method for mm Wave MIMO systems with lowresolution ADCs based on statistical learning.The modeling and sparse characteristics of mm Wave channels are introduced and analyzed,and the mathematical model of mm Wave MIMO system receiver with low-resolution ADCs is described.Then two schemes are proposed,namely channel estimation based on generalized approximate message passing(GAMP)algorithm and channel estimation based on generalized expectation consistent signal recovery(GECSR)algorithm.And the Laplacian prior distribution is introduced to model the mm Wave channel coefficients in the angular domain.Expectation Maximization(EM)algorithm is embedded into the GAMP and GEC algorithms to learn the unknown parameters in the prior distribution.The simulation results show that the normalized mean-square error(NMSE)performance of the two algorithms under the Laplace prior is significantly improved compared to the Gaussian-mixture(GM)prior.In addition,the GEC algorithm can significantly reduce the required pilot length under the orthogonal DFT pilot to reduce pilot overhead.Finally,we study the channel estimation method for mm Wave MIMO systems with lowresolution ADCs based on deep learning.In view of the significant advantages of deep neural networks in mining unknown distribution feature information of channels,a model-driven deep learning-based LDGAMP(learned denoising GAMP)channel estimation method is proposed.A deep denoising convolutional neural network(Dn CNN)is embedded into the GAMP algorithm framework,replacing the minimum mean-square error(MMSE)original estimator.Simulation results show that the LDGAMP channel estimation method performs significantly better than advanced iterative estimation algorithms based on statistical learning,and its NMSE performance can maintain superior even with very few pilot.Finally,the performance of Dn CNN under different mm Wave channel beam clusters and the number of sub-paths is explored,showing that it has good robustness.