LAUSR

Recent advancements in wireless communication, notably the shift towards 5G and the anticipated evolution to 6G, have significantly increased the importance of channel state information (CSI) in achieving high-accuracy indoor localization. These advancements promise better localization precision due to more complex antenna systems. However, they also bring challenges, such as the need to process large and complex CSI data and accurately interpret it in environments that constantly change. To address these, we present D2-MLoc, a Dual-Domain multi-layer perceptron (MLP)-Mixer Localization framework for CSI-fingerprinting localization. D2-MLoc first transforms complex frequency-space CSI into a sparse Angular Delay Fingerprint Matrix (ADFM) using discrete Fourier transform (DFT), thereby reducing the storage and computational complexity. Subsequently, we design an MLP-Mixer-based Localization model, named MLoc, to capture the interlaced space and frequency characteristics inherited in ADFM. Furthermore, a dual-domain transfer architecture for MLoc is proposed to improve its adaptability and accuracy in dynamic indoor environments, where the data distribution varies due to environmental changes (e.g., noise and multipath effects). Experimental results from both simulation and real-world datasets demonstrate that D2-MLoc outperforms state-of-the-art methods in terms of localization accuracy and robustness under various environmental dynamics.