A Unified Bayesian Method for Multipath Parameter Extraction With Cochannel Interference

Robust tensor variational methods based on CANDECOMP/PARAFAC (CP) decomposition have attracted significant interest for efficiently exploiting high-dimensional multipath channel data, capturing multipath parameters, and enhancing robustness and convergence under sparse interference and noise. In this letter, to address the overfitting and slow convergence issues of the traditional robust CP (RCP) method and the modeling error in the RCP based on alternating prior hypotheses method, we propose a unified RCP (URCP) method, which offers a unified model for various wavefront types of cochannel interference (CCI) and performs alternating maximization over different inference domains. Meanwhile, to improve the accuracy of channel parameter estimation, CCI frequency bands are preidentified before alternating iterations, and interference hyperparameters are adaptively updated during iterations. To validate the proposed method, we compare it with the aforementioned methods in both SI and UI scenarios. Simulation results show that URCP effectively avoids overfitting and reduces convergence time. In addition, a channel measurement-based experimental validation confirms the reliability of the proposed multipath parameter extraction method.