Two-Dimensional Channel Estimation via Dictionary Learning for FDD MIMO Systems

Abstract

Using compressed sensing and dictionary learning algorithms to reduce the number of pilot symbols is an effective way to improve the efficiency of channel estimation in frequency division duplexing (FDD) system. The typical predefined dictionaries used in compressed sensing may lead to energy leakage and hence cannot achieve satisfactory performance using sparse representation for current complex channel models. Different from the classic method, dictionary learning algorithm generates a dictionary for a specific training data set, which can make the signal be represented more sparsely. So the signal can be recovered more accurately. In this paper, we consider the two-dimensional (2D) multi-carrier channel model which is more suitable for practical application. We develop the dictionary learning algorithm for 2D channel estimation to reduce the number of pilot symbols and propose an improved K-SVD algorithm to solve the resultant optimization problem. Compared with the traditional channel estimation algorithm and compressed sensing algorithm using a typical predefined dictionary, the proposed algorithm can improve the performance. Simulation results verify that the proposed algorithm can improve the accuracy of channel estimation and save the number of pilot symbols.