DC-BL: Flexible delay-doppler domain channel estimation via decoupled bayesian learning

Orthogonal Time Frequency Space (OTFS) modulation has shown remarkable robustness in high-mobility scenarios by effectively mitigating the detrimental effects of Doppler shifts. However, accurate channel estimation for OTFS systems remains a challenging problem due to issues such as energy dispersion caused by fractional Doppler shifts, limited exploitation of inherent delay-Doppler domain sparsity, and prohibitive computational complexity. In this work, we propose a novel decoupled channel estimation framework that decomposes the delay-Doppler (DD) domain channel into separate delay and Doppler components. This decoupling facilitates a more efficient and targeted utilization of channel structures in each domain. Building upon this framework, we develop a novel estimation algorithm based on decoupled Bayesian learning (DC-BL), which leverages virtual grids and integration-based sensing matrices to achieve super-resolution channel recovery. Notably, the proposed method retains adaptability even under low-resolution grid settings. To address such scenarios, we further design a specialized Bayesian estimation algorithm optimized for coarse-granularity grids. Both algorithms incorporate a sparsity-aware mechanism to eliminate redundant channel taps, thereby significantly reducing model dimensionality and computational burden. Extensive simulation results demonstrate that the proposed methods consistently outperform existing state-of-the-art approaches in terms of estimation accuracy and efficiency.