Efficient wideband off-grid DOA estimation via sparse Bayesian learning with fixed-point iterations for signals with different frequency bands

This paper proposes an efficient wideband off-grid direction-of-arrival (DOA) estimation method for signals with different frequency bands. The method, termed WOG-FPSBL, is developed within the sparse Bayesian learning (SBL) framework and incorporates fixed-point iterations to accelerate convergence. Unlike conventional SBL approaches that assume a common signal variance vector across all frequencies, WOG-FPSBL adopts subband-specific variance parameterization to effectively capture the joint spatial-spectral structure and enhance estimation accuracy. Bayesian inference is performed using the expectation-maximization algorithm with fixed-point hyperparameter updates to improve computational efficiency. Coarse DOA estimates are obtained from the spatial power spectrum and further refined through dichotomous search-based marginal likelihood maximization, thereby reducing off-grid errors. Simulation results confirm that WOG-FPSBL consistently delivers superior accuracy and lower runtime compared with state-of-the-art SBL-based methods, demonstrating its suitability for practical wideband DOA estimation.