Fast and Reliable Bayesian Orthogonal Approximate Message Passing Detector for Zero-Padded OTFS Systems
Abstract
The Bayesian orthogonal approximate message passing (BOAMP) detector provides reliable data detection for orthogonal time-frequency-space (OTFS) systems, but its linear estimator (LE) requires a matrix inversion with complexity <inline-formula> <tex-math notation="LaTeX">$\mathcal {O}(N^{3}M^{3})$ </tex-math></inline-formula>, where <inline-formula> <tex-math notation="LaTeX">$N$ </tex-math></inline-formula> and <inline-formula> <tex-math notation="LaTeX">$M$ </tex-math></inline-formula> denote the number of Doppler and delay bins, respectively. In this letter, inspired by cross-domain iterative detection, we exploit the unitary property of the fast Fourier transform (FFT) to derive that the covariance of the Gaussian estimation error remains invariant across the time and delay-Doppler (DD) domains. Then, we develop the cross-domain BOAMP framework based on the covariance invariance. Finally, we further develop a fast and reliable time-domain block BOAMP (TB-BOAMP) detector by exploiting the block-diagonal structure of the equivalent time-domain channel matrix in the zero-padding OTFS (ZP-OTFS) system. The proposed TB-BOAMP detector reduces the dominant asymptotic complexity of the LE stage to <inline-formula> <tex-math notation="LaTeX">$1/N^{2}$ </tex-math></inline-formula> of that of conventional BOAMP. Simulation results verify that the proposed scheme maintains nearly identical bit error rate (BER) performance to the conventional BOAMP, while significantly outperforming benchmark schemes and reducing the computational burden.