TR2019-042

Sparse Bayesian Estimation of Millimeter-Wave Channel Correlation Matrix



We propose an algorithm for estimating correlation matrix of mmWave channels. In addition to being sparse in angular/spatial domain, mmWave channel is commonly assumed to be time-invariant over a certain time period. However, more recent experiments indicate that while mmWave channel spatial directions can be assumed constant over some time period, their coefficients vary in time. Building upon this result, we probabilistically treat the correlation matrix estimation problem by associating sparse Bayesian learning prior to channel realizations, and performing statistical inference to recover channel directions and estimate their correlation matrix. The proposed algorithm is validated with simulations and shown to outperform benchmark methods based on greedy optimization-based sparse recovery.