Joint estimation of multipath signal parameters using variational SBL-inspired SAGE algorithm

Abstract

In this paper, we jointly estimate the model order, amplitude gain and dispersion parameters of the received signal vector using a variational sparse Bayesian framework. Contrasting with the Gamma-Gaussian model typically employed in classical sparse Bayesian learning, we select the Bernoulli–Gaussian model as the hierarchical prior and infer a pruning condition for a single specular component within the SAGE framework. The adaptive thresholds derived from this approach are better suited to varying signal-to-noise ratios and provide improved model order estimation. Moreover, two novel joint estimation algorithms are proposed within this framework: (1) optimizing the alternating iterative process inherent in the variational solving approach, while jointly optimizing a portion of the dispersion parameters and amplitude gain to enhance model order estimation without adding to the computational complexity; (2) additionally, introducing a time-delay estimation computation method based on the autocorrelation characteristics of the sounding sequence, aimed at reducing algorithm complexity and speeding up convergence. Finally, the performance advantages of the proposed algorithm are validated through simulations and measured data. Comparisons with related algorithms demonstrate that the proposed algorithm effectively accomplishes joint estimation of model order and channel parameters. Particularly, it achieves more accurate estimation of model order and dispersion parameters in scenarios with high signal-to-noise ratios.