Abstract:A deep denoising based channel estimation framework is proposed for orthogonal time frequency space (OTFS) modulated systems, wherein channel state information (CSI) recovery is formulated as an image restoration problem. A salient attribute of the approach is the exploitation of structural invariance in the delay Doppler (DD) domain channel over a geometric coherence time, allowing multiple OTFS frames captured during this period to serve as noisy snapshots of the approximately identical channel. These snapshots jointly enhance the effectiveness of the proposed lightweight denoiser based on nonlinear activation free network (NAFNet). The method exhibits low computational complexity, operates reliably even at low pilot signal-to-noise ratio (PSNR), and can accommodate both fractional delay and fractional Doppler effects. Simulation results demonstrate significant performance gains over the existing methods.
Abstract:A novel Gaussian mixture model (GMM) aided sparse Bayesian learning (SBL) framework is proposed for channel state information (CSI) estimation in orthogonal time-frequency space (OTFS) modulated systems. The key attribute of the proposed algorithm lies in casting CSI recovery as an SBL inference problem, where posterior distributions are iteratively refined under a hierarchical GMM prior. Using this approach, the sparsity-inducing variances beneficially promote sparsity in the delay Doppler (DD) domain, while additionally augmenting the capability of SBL to exploit channel statistics more effectively. Moreover, to fully exploit the GMMs ability to approximate arbitrary probability density functions and model complex multipath fading scenarios, the channel statistics are represented using a complex Gaussian mixture. Simultaneously, the method leverages time-domain (TD) pilots without requiring wasteful DD domain guard intervals, thereby ensuring low pilot overhead and high spectral efficiency. The CSI recovered is subsequently applied in a linear minimum mean square error (MMSE) detector for reliable data detection. To benchmark performance, the Oracle-MMSE and the Bayesian Cramèr Rao lower bound (BCRLB) are also derived. Our simulation results demonstrate significant performance improvement over the state of the art sparse estimation methods.