Joint Channel and Data Estimation via Parametric Bilinear Inference for OTFS Demodulation

Abstract

This paper proposes a novel joint channel and data estimation (JCDE) algorithm via parametric bilinear Gaussian belief propagation (PBiGaBP) for orthogonal time frequency space (OTFS) demodulation. In doubly-selective fading channels, since the Doppler shift breaks the orthogonality between subcarriers, the signal demodulation process requires estimating the path gains and data symbols from the high-dimensional signals that are multiplexed in the frequency-time (FT) domain. To obtain highly accurate channel state information (CSI) using a typical channel estimation scheme based only on reference signals, an increase in pilot overhead to suppress inter-symbol interference (ISI) is unavoidable. In addition, large-scale matrix operations based on the size of OTFS-equivalent channels also pose problems in terms of computational cost. To address these issues, we focus on the fact that OTFS demodulation can be formulated as a large-scale parametric bilinear inference (PBI) problem, and solve it using the Gaussian belief propagation (GaBP) approach, which enables an approximate implementation of the sum-product algorithm (SPA), based on the central limit theorem (CLT), to design a low-complexity and high-accuracy JCDE algorithm using short pilots. Simulation results show that the proposed PBiGaBP-based JCDE algorithm outperforms the state-of-the-art (SotA) schemes and approaches the performance of idealized (genie-aided) scheme in terms of mean square error (MSE) and bit error rate (BER) performances.

Keywords

• OTFS demodulation
• high-mobility communications
• joint channel and data estimation
• parametric bilinear inference
• Gaussian belief propagation