CSI-Based NTC Using Ambient WiFi: Opportunities and Challenges

With the ubiquity of WiFi-enabled devices, WiFi Channel State Information (CSI) based sensing of the physical environment has been researched broadly. Following our previous study [1] demonstrating the possibility of WiFi sensing for Network Traffic Classification (NTC) application, we introduce a Kth Nearest Neighbour (KNN) classification model, and examine the CSI-based NTC framework with channel interference and show a deterioration of NTC performance of a trained KNN model from 73.1% accuracy to as low as 16.2% for a four-class classification. Additionally, we also investigate the capability of this framework by testing the performance with finer, overlapping network traffic classes, achieving overall 84.1% for several Spotify and YouTube network traffic classes.