Row-Vector Transformer (RVT): AI-Driven Energy-Efficient Received Power Prediction for Sub-6 GHz RIS in 6G Wireless Networks

Accurate and energy-efficient prediction of received signal power is critical for optimizing reconfigurable intelligent surfaces (RISs) in beyond-5G and emerging 6G wireless communication systems. These surfaces, capable of dynamically reconfiguring the propagation environment, demand real-time, high-precision modeling to maximize spectral and energy efficiency. Traditional model-based and deep learning approaches often rely on synthetic data or treat features independently, limiting their ability to capture complex, nonlinear electromagnetic interactions inherent in real-world RIS deployments. In this work, we propose the row-vector transformer (RVT)—a novel AI framework designed for RIS-assisted 6G propagation modeling—that constructs a single, dense token from the entire feature set, including receiver angles, RIS phase configurations, and normalized received power measurements. Unlike conventional transformer architectures [e.g., feature tokenizer transformer (FT-Transformer)] that process features columnwise, RVT processes each sample holistically, enabling more effective extraction of electromagnetic channel dependencies and propagation patterns. Experimental validation on three real RIS measurement datasets demonstrates that RVT consistently outperforms advanced deep learning baselines, including one-dimensional convolutional neural network (1D-CNN), long short-term memory (LSTM), bidirectional LSTM (BI-LSTM), gated recurrent unit (GRU), fully connected feedforward neural network (FCN), FT-Transformer, and TabNet, achieving the lowest mean squared error (MSE), mean absolute error (MAE), and root mean square error (RMSE), along with faster convergence and greater stability. The results confirm RVT’s potential as a cutting-edge, AI-enabled solution for real-time RIS power prediction, contributing toward intelligent, adaptive, and energy-efficient channel modeling in future 6G wireless networks.