Research On Sparse Channel Estimation For Massive MIMO System

Channel state information?CSI?estimation is important research content in the field of large-scale multiple input multiple output?MIMO?in array signal processing.It is widely used in radar,navigation,mobile communication,driverless,industrial control,Internet of Things,smart city,etc.However,compared to the conventional array antenna structure,the massive MIMO system brings a large increase in the number of antennas which results in a significant increase in the pilot consumption at the transmitting devices.In order to effectively avoid the problem of excessive pilot overhead,researchers have proposed many frameworks of channel estimation algorithms based on sparse structure,which use the sparse characteristics of the signal to reduce the pilot overhead.However,due to the signal sampling in continuous space,the quantization angle of arrival and Angles of arrival/depatures?AoAs/AoDs?will not coincide with pre-defined grids,resulting in grid mismatch problem which will severely limit the performance of algorithms.In addition,the noise interference with pulse characteristics in array signal processing will destroy the performance of algorithms.Therefore,the study of the gridless robust channel estimation method has crucial theoretical and practical significance for the field of array signal processing.This paper focuses on the large-scale MIMO system based on hybrid precoding structure,and studies from the following three aspects:1.Research on the problem of high-dimensional computational burden in the traditional joint 2D sparse channel estimation framework.We have proposed a dimensionality reduction mathematical model representation for a massive MIMO system,which using the received intrinsic tensor structure of high-dimensional data and transform the traditional 2-dimensional joint estimation problem into two parallel?One Dimention,1D?estimation sub-problems.The proposed reduce-dimension schemes can effectively address the high computational complexity problem.The numerical results show that the proposed parallel one-dimensional model framework can effectively reduce the time complexity compared with the traditional joint estimation 2D framework,which especially is suitable for large-scale array structure and easy to extend to other algorithm frameworks.The proposed schemes have strong applicabilities.2.Research on the grid mismatch problem in the traditional sparse channel estimation scheme.An improved l1-singular value decomposition?SVD?sparse channel estimation algorithm has been proposed.The first-order linear approximation of the steering vector is used to construct the mesh mismatch error model.The cost function corresponding to the grid error is constructed by the orthogonal relationship between the signal space and the noise space,then the grid mismatch problem is transformed into the analytical expression of the known structure.Due to the convex structure characteristics of the constructed objective function,the convergence of the algorithm is guaranteed,which can solve the problem of optimal values.At the same time,due to the strict closed solution of the grid error,the proposed algorithm can converge in a few iteration steps.Finally,the correctness and effectiveness of the proposed algorithm are verified by numerical simulation of calculated values.In addition,an improved meshless sparse channel estimation algorithm based on Fast Fourier Transform?FFT?has been proposed and the estimation performance of the proposed algorithm is theoretically analyzed.Compared with the existing sparse channel estimation method of FFT domain,the proposed framework has the same theoretical estimation performance,however,it is more powerful because the proposed method does not have the contradiction between time calculation complexity and estimation performance accuracy,which resulting in practicality and effectiveness.3.Research on the impulse noise problem existing in the girdless sparse channel estimation method.An iterative reweighted?IR?-based robust algorithm has been proposed.The theoretical lower bound of the proposed algorithm framework has been analysised.By using the log-sum function to smooth the non-smooth points of the l₁-norm constraint in the traditional framework and adding the weighting factor to construct the penalty term.Next,we use the gradient descent method to obtain update solution of AoAs/AoDs.Compared with the existing iterative reweighted algorithm,the proposed algorithm can effectively suppress the interference of impulse noise and can preserve the original good estimation performance under the background of Gaussian noise.The proposed scheme can effectively extend the applicable range of the algorithm and has better prospects.