Enhancing Generalization in Human Activity Recognition Through Improved WI-FI Channel State Information Phase Processing and Antenna Pair Selection

Utilizing Wi-Fi signals to recognize human activities indoors has become increasingly popular for interfacing with smart devices. However, the lack of acceptable generalization in new conditions and for new users has made Wi-Fi-based activity recognition impractical in daily usage. Although previous studies have attempted to introduce various methods for HAR using Wi-Fi channel state information (CSI), to the best of our knowledge, these methods are reliable only under highly controlled, constant experimental conditions. This work focuses on improving the generalization of HAR for unseen users and proposes a novel method for CSI-based HAR, specifically designed for the CSI collected using Wi-Fi routers with multiple antennas. This process involves selecting the best pair of antennas that can effectively eliminate most of the common noise present in the antennas and training a separate classifier for each antenna pair. The results demonstrate that this method significantly enhances the accuracy of Wi-Fi-based HAR on previously unseen users.