An Artificial Intelligence Enabled Signature Estimation of Dual Wideband Systems in Ultra-Low Signal-to-Noise Ratio

Millimeter-wave (mmWave) massive Multiple Input Multiple Output (MIMO) systems encounter both spatial wideband spreading and temporal wideband effects in the communication channels of individual users. Accurate estimation of a user's channel signature -- specifically, the direction of arrival and time of arrival -- is crucial for designing efficient beamforming transceivers, especially under noisy observations. In this work, we propose an Artificial Intelligence (AI)-enabled framework for estimating the channel signature of a user's location in mmWave massive MIMO systems. Our approach explicitly accounts for spatial wideband spreading, finite basis leakage effects, and significant unknown receiver noise. We demonstrate the effectiveness of a denoising convolutional neural network with residual learning for recovering channel responses, even when channel gains are of extremely low amplitude and embedded in ultra-high receiver noise environments. Notably, our method successfully recovers spatio-temporal diversity branches at signal-to-noise ratios as low as -20 dB. Furthermore, we introduce a local gravitation-based clustering algorithm to infer the number of physical propagation paths (unknown a priori) and to identify their respective support in the delay-angle domain of the denoised response. To complement our approach, we design tailored metrics for evaluating denoising and clustering performance within the context of wireless communications. We validate our framework through system-level simulations using Orthogonal Frequency Division Multiplexing (OFDM) with a Quadrature Phase Shift Keying (QPSK) modulation scheme over mmWave fading channels, highlighting the necessity and robustness of the proposed methods in ultra-low SNR scenarios.