Underdetermined DOA Estimation of Wideband Quasi-Stationary Signals via Tensor Completion

A direction of arrival (DOA) estimation method for wideband quasi-stationary signals via tensor completion is proposed in this article. Second-order virtual arrays across all frequencies of interest are initially generated and extended into a unified uniform linear array. Subsequently, the autocovariance and crosscovariance matrices of these virtual array models are computed and stacked to form an incomplete 3-D tensor. Following low-rank tensor completion, the completed tensor is combined together along the third dimension to obtain an equivalent covariance matrix for DOA estimation. This proposed method is gridless, avoiding off-grid mismatch errors while enhancing the estimation accuracy. It also boasts low computational complexity, with simulation and experiments validating its accuracy and efficiency.