<![CDATA[Accurate wireless channel modeling is fundamental to the design and optimization of fifth-generation (5G) communication systems. Traditional geometry-based stochastic models (GBSMs) and empirical formulations, while effective in static environments, often fail to capture the nonlinear, non-stationary, and environment-dependent propagation behaviours inherent in modern multi-antenna and millimeter-wave systems. This study introduces a physics-informed AI hybrid framework that fuses physical propagation principles with deep learning architectures, enabling channel modeling that is interpretable, adaptive, and data-efficient. Using large-scale datasets including DeepMIMO, COST (Cooperation in Science and Technology) 2100, and New York University (NYU) Wireless, the model integrates Physics-Informed Neural Networks (PINNs) and Convolutional Neural Networks (CNNs) to simultaneously capture spatial, temporal, and frequency-domain relationships under realistic propagation environments. Reinforcement and federated learning layers enable real-time adaptation and decentralized training across multiple base stations while preserving data privacy. Experimental results demonstrate substantial improvements over benchmark models such as 3GPP (3rd Generation Partnership Project) TR 38.901, COST 2100, and QuaDRiGa (QUAsi Deterministic RadIo Channel GenerAtor), achieving an RMSE of 1.72 dB and NMSE of –20.6 dB, corresponding to a 25–30% accuracy gain. Visual analyses of power delay profiles, residual error distributions, and spatial correlation maps confirm the model’s robustness and physical consistency. The proposed framework offers a scalable, interpretable, and adaptive paradigm for next-generation wireless channel modeling, paving the way toward intelligent, self-optimizing, and 6G-ready communication networks that bridge the gap between physics-based theory and AI-driven modeling.]]>
