Joint Near-Field Channel Estimation and Activity Detection for mMTC With Distributed Arrays

Abstract

This paper addresses joint channel estimation and activity detection for massive machine-type communications in near-field regimes, where spherical wavefronts introduce range-angle coupling and dramatically expand the parameter space. The far-field sparse recovery approaches become inaccurate or inapplicable, while existing near-field methods usually face approximation errors and can only be applied to single uniform array configurations. To overcome these challenges, an exact near-field channel model based on practical spatial geometry is introduced to avoid any near-field approximation and support arbitrary distributed array configurations. We further develop an efficient Bayesian learning framework that exploits both channel sparsity and sporadic device-activity through a hybrid common-sparse prior. After that, we propose a multi-layer message-passing algorithm that decouples channel estimation and activity detection using an auxiliary variable and enforces consistency through a Dirac-delta constraint. Finally, a collaborative expectation-maximization (EM) procedure is embedded to enable scalable parameter updates across distributed arrays. The proposed method can achieve accurate and computationally efficient joint near-field channel estimation and activity detection for large-scale mMTC systems.