Enhancing Channel Estimation in Quantized Systems with a Generative Prior

TL;DR

This work tackles channel estimation in one-bit quantized systems by integrating a Gaussian Mixture Model as a generative prior into an EM-based estimation framework. By inferring the most responsible GMM component from the observed pilots, the method yields a conditional Gaussian prior that preserves analytic M-step solutions, enabling a closed-form, low-complexity update for the channel. The offline-trained GMM prior can incorporate structural covariances (Toeplitz or circulant) to reduce memory and computation, and the proposed GMM-EM framework shows significant gains over both simplistic Gaussian priors and state-of-the-art estimators, with robustness to higher-resolution quantization and alternative priors. Practically, this approach improves estimation accuracy and can reduce pilot overhead, energy consumption, and latency in massive MIMO systems with quantized receivers, while remaining adaptable to different generative priors and channel models.

Abstract

Channel estimation in quantized systems is challenging, particularly in low-resolution systems. In this work, we propose to leverage a Gaussian mixture model (GMM) as generative prior, capturing the channel distribution of the propagation environment, to enhance a classical estimation technique based on the expectation-maximization (EM) algorithm for one-bit quantization. Thereby, a maximum a posteriori (MAP) estimate of the most responsible mixture component is inferred for a quantized received signal, which is subsequently utilized in the EM algorithm as side information. Numerical results demonstrate the significant performance improvement of our proposed approach over both a simplistic Gaussian prior and current state-of-the-art channel estimators. Furthermore, the proposed estimation framework exhibits adaptability to higher resolution systems and alternative generative priors.

Figures (4)

• Figure 1: mse performance with one (top) and three (bottom) propagation cluster, $N=64$ antennas, and $P=16$ pilot observations.
• Figure 2: Average number of em iterations until convergence for one propagation cluster, $N=64$ antennas, and $P=16$ pilot observations.
• Figure 3: mse performance for varying numbers of pilot observations for one propagation cluster, $N=64$ antennas, and $\text{SNR} = \qty{5}{dB}$.
• Figure 4: mse performance for different numbers of gmm components for one propagation cluster, $N=64$ antennas, and $P=16$ pilot observations.