Nonparametric Variational Bayesian Learning for Channel Estimation with OTFS Modulation

TL;DR

This work tackles OTFS channel estimation in high-mobility environments with realistic CDL-type clustered channels and off-grid delay/Doppler offsets. It introduces a nonparametric Bayesian learning (NPBL) framework that combines a truncated stick-breaking process to automatically infer the number of paths and their cluster assignments with a Gaussian-mixture model for intra-cluster channel coefficients, all estimated via variational inference and augmented by a successive linear approximation to handle off-grid effects. The approach yields automatic model order selection, improved NMSE over competitive baselines, and pruning-based complexity reduction, demonstrating robust performance across varying cluster counts and sparsity. The proposed NPBL method holds promise for practical OTFS deployment in next-generation networks (e.g., SAGIN in 6G), where high mobility, diverse propagation conditions, and dynamic cluster structures prevail.

Abstract

Orthogonal time frequency space (OTFS) modulation has demonstrated significant advantages in high-mobility scenarios in future 6G networks. However, existing channel estimation methods often overlook the structured sparsity and clustering characteristics inherent in realistic clustered delay line (CDL) channels, leading to degraded performance in practical systems. To address this issue, we propose a novel nonparametric Bayesian learning (NPBL) framework for OTFS channel estimation. Specifically, a stick-breaking process is introduced to automatically infer the number of multipath components and assign each path to its corresponding cluster. The channel coefficients within each cluster are modeled by a Gaussian mixture distribution to capture complex fading statistics. Furthermore, an effective pruning criterion is designed to eliminate spurious multipath components, thereby enhancing estimation accuracy and reducing computational complexity. Simulation results demonstrate that the proposed method achieves superior performance in terms of normalized mean squared error compared to existing methods.

Figures (5)

• Figure 1: Channel in the DD domain.
• Figure 2: Tx and Rx symbols in the DD domain.
• Figure 3: Graphical model for the proposed algorithm.
• Figure 4: NMSE of SNR for different algorithms.
• Figure 5: NMSE of SNR for different cluster numbers.