Robust Angular Power Spectrum Estimation in Massive MIMO Under Covariance Errors: Adapting Gaussian Centers and Scales to Time-Varying Environments

This paper addresses the problem of estimating the angular power spectrum (APS) from an erroneous estimate of the covariance matrix in massive multiple-input multiple-output (MIMO) systems—a key challenge, for instance, for improving the efficiency of frequency-division duplex (FDD) operation. The APS is modeled as a superposition of Gaussian functions, which implicitly imposes a smoothness prior that alleviates the ill-posedness of the estimation problem and that robustifies our APS estimate against the covariance estimation errors. The weights, scales, and centers of the Gaussian components are iteratively updated within a multikernel adaptive filtering framework, enabling real-time tracking of the temporal variation of the APS. Simulations demonstrate that the proposed method achieves high estimation accuracy and remarkable robustness against the covariance errors simultaneously.