SenseMamba: A General Lightweight State-Space Model for Wireless Human Sensing

Wireless human sensing has demonstrated significant potential across various domains, including healthcare, smart homes, and human-computer interaction. However, most existing research is task-specific, often requiring the design of multiple networks for different human sensing tasks. Moreover, the complex processing of wireless signals and the substantial number of network parameters increase recognition time and deployment costs. In this work, we develop SenseMamba, a general lightweight state-space model tailored for a variety of wireless human sensing tasks, aiming to overcome computational resource constraints in real-time wireless human sensing scenarios. Our approach integrates selective state-space modeling with Kolmogorov-Arnold networks (KANs), enabling SenseMamba to effectively extract spatial-temporal features from raw wireless signals. With state-space modeling at its core, SenseMamba captures dynamic temporal dependencies in the data, facilitating a more sophisticated understanding of signal behavior. By incorporating KANs, our model simplifies the complexity of signal-to-output mappings, maintaining a lightweight design critical for real-time wireless sensing. The design of KANs allows the model to better capture and represent complex relationships in high-dimensional data by learning activation functions while maintaining a fully connected structure. Leveraging the local plasticity and global stability of splines, SenseMamba enhances the model’s local plasticity and global stability, achieving state-of-the-art (SOTA) accuracy in real-time across various complex scenarios for different wireless human sensing tasks, making wireless sensing deployment more convenient. These performance records demonstrate SenseMamba’s advantages over existing methods for different wireless human sensing tasks and various complex scenarios.