Conditional denoising diffusion-based channel estimation for fast time-varying MIMO-OFDM systems

Abstract

We propose an innovative conditional denoising diffusion-based channel estimation (CDDCE) scheme for fast time-varying multiple-input multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) systems. This intelligent CDDCE model delicately adapts the denoising diffusion probabilistic model (DDPM) to conditional channel state information (CSI) generation and performs efficient channel estimation with a stochastic iterative denoising process. Specifically, the CDDCE model utilizes a Markov chain that gradually adds Gaussian noise to the customized preprocessed genuine CSI according to the cosine variance schedule for the forward Gaussian diffusion process. Then, the channel estimation begins with pure Gaussian noise and repeatedly refines the conditional roughly estimated CSI by a specialized U-Net trained on denoising at different noise levels for the reverse iterative refinement process. Numerical results show that our CDDCE scheme significantly outperforms classical approaches and three cutting-edge deep learning (DL)-based ones, indicating its eminent capability to learn the statistical characteristics of wireless channels. Besides, we demonstrate that the CDDCE scheme exhibits excellent robustness against various channel distortions and interference: when (i) there are a restricted number of pilot symbols, (ii) the cyclic prefix (CP) is omitted, (iii) the clipping noise is introduced, and (iv) the offline and online channel conditions are mismatched.