Presentation Information
[B-5A-38]Variational Bayesian Channel Estimation with Horseshoe Prior for AFDM Mobile Communications
◎△Hao Jia1, Kazuhiko Fukawa1 (1. Institute of Science Tokyo)
Keywords:
AFDM,channel estimation,sparse Bayesian learning,ISAC,Horseshoe prior
Affine frequency division multiplexing (AFDM) can be considered promising for next-generation high-mobility communications, because this multi-carrier scheme is robust against doubly-selective channels by exploiting channel sparsity in the discrete affine Fourier transform (DAFT) domain. Thus, sparse Bayesian learning (SBL) has been applied to AFDM channel estimation. However, when fractional Doppler shifts need to be estimated with high resolution, especially for integrated sensing and communication (ISAC), dictionary step size k_d should be set smaller, which causes higher dictionary coherence and makes it hard to satisfy the mutual incoherence condition. Since conventional sparse channel reconstruction schemes such as expectation-maximization (EM)-based Laplace-prior SBL are sensitive to high dictionary coherence, their channel estimation accuracy can be severely deteriorated. To compensate for such degradation, this article proposes a variational Bayesian (VB)-based Horseshoe prior SBL method. Computer simulations show that the proposed method outperforms EM-based Laplace prior SBL and orthogonal matching pursuit (OMP).