Variational Bayesian Learning Based Delay-Doppler Channel Estimator for Multi-User OTFS Systems

We consider a practical-pulse-shaped multi-user orthogonal time frequency space system wherein the delay-Doppler domain channels of different users observe a shared sparsity structure. We propose a variational Bayesian learning (VBL) based channel estimation algorithm, which exploits this shared sparsity by constructing a novel Gaussian mixture prior. This prior makes the calculation of posterior channel distribution intractable, which the proposed VBL algorithm handles by introducing a structured variational distribution. Our simulation results demonstrate that the proposed estimator radically outperforms the existing state-of-the-art ones, in terms of normalized mean squared error and bit error rate.