Channel Estimation for Sparse Massive MIMO Channels in Low SNR Regime

With perfect channel state information, a huge signal to noise ratio (SNR) gain can be obtained in massive multiple input multiple output (MIMO) systems. Therefore, massive MIMO systems are generally assumed to work in low SNR regime. However, channel estimates are contaminated by white noise in practical scenarios, which will induce great performance degradation, especially in low SNR regime. To improve channel estimation quality, we propose a channel estimator to filter out noise in the conventional matched filter-based channel estimates by exploring the spatial sparsity of massive MIMO signals. The viability of this new method is based on the fact that wireless channels are sparse in space domain. To be specific, most energy of the desired signals concentrates on a small number of paths (or directions, equivalently), while the energy of noise is equally spread on all directions. Therefore, we propose an algorithm to identify the desired signals and eliminate most noise. One of the largest advantages of the proposed algorithm is that statistical information concerning the channel vectors is unnecessary. Both theoretical analysis and simulation results justify the efficacy of the proposed channel estimator.