FDD Downlink Channel Estimation Solution With Common Sparsity Learning Algorithm and Zero-Partition Enhanced GAMP Algorithm

To acquire precise channel state information in the forthcoming 5G frequency division duplexing (FDD) communication system, we study the virtual angular domain channel common sparsity in the massive multiple-input multiple-output orthogonal frequency division multiplexing downlink transmission system. Here, we propose an off-line learning algorithm, which is based on expectation maximization and generalized approximate message passing (GAMP) algorithms. Besides, the learning algorithm utilizes the common sparsity to obtain the support information of the sparse channel vectors. This process is defined as zero partition. With this learned information, the sparse channel vectors are estimated using GAMP again. We prove that when the learned support information is correct, the zero partition-enhanced GAMP algorithm will achieve better sparsity and indeterminacy trade-off performance when compared with the original GAMP algorithm. Based on these results, our proposed channel estimation (CE) solution has two stages. In the first stage, because of the practical propagation characteristics, the zero partition learning algorithm is amended to ensure the correctness of the learned result. In the second stage, channel vectors on the virtual angular domain are estimated with the proposed zero partition-enhanced GAMP algorithm. Analysis and numerical results prove that our CE solution will consume less pilots to ensure the estimation accuracy, when compared with other existed FDD downlink CE solution.