Transformer-Based Multi-Modal Indoor Localization in RIS-Assisted Wireless Networks

This paper introduces a Transformer-based framework for high-precision indoor localization in 6G-enabled Internet of Things (IoT) environments, enhanced by reconfigurable intelligent surfaces (RISs). The proposed system integrates multiple signal modalities, including channel state information (CSI), received signal strength (RSS), geometric data, and adjustable RIS phase shifts, into a unified deep learning model. A multi-head self-attention Transformer is employed to capture the spatial-temporal dependencies inherent in indoor signal propagation, enabling reliable estimation of user coordinates even under challenging multipath and non-line-of-sight (NLoS) conditions. The RIS is further optimized through controlled phase shift strategies to enhance both localization accuracy and signal quality. We conduct comprehensive simulations that model real-world environmental conditions, including Rayleigh fading and varying signal-to-noise ratio (SNR) levels. Results indicate that the proposed framework significantly outperforms traditional localization benchmarks, achieving sub-meter accuracy while reducing training complexity and enhancing scalability. Moreover, the model is robust against channel estimation errors and supports real-time inference, making it ideal for smart building applications. This framework lays the groundwork for future wireless systems that demand intelligent and geometry-aware localization.