Scheduled Spatial Sensing against Adversarial WiFi Sensing

• Published in: Proceedings of the IEEE International Conference on Pervasive Computing and Communications pp. 91 - 100
• Main Authors: Hernandez, Steven M., Bulut, Eyuphan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 13.03.2023
• Subjects: Data models; human activity detection; Performance evaluation; Predictive models; Receiving antennas; Reflection; Schedules; security and privacy; Transmitting antennas; WiFi sensing
• ISSN: 2474-249X
• DOI: 10.1109/PERCOM56429.2023.10099079

WiFi sensing aims to utilize the changes in the Channel State Information (CSI) of WiFi signals due to the reflections from objects in the environment for sensing purposes. It uses machine learning classification models to predict physical actions being performed in a given environment (e.g., human activities such as walking, running). Thanks to the existing WiFi infrastructure in most indoor areas, this device-free technology can be used to provide low-cost motion detection and activity recognition opportunities for smart-homes. However, as the WiFi signals can be sniffed by adversaries, it can also be utilized by malicious actors to learn private information about the residents. To address this issue, motivated by the fact that the accuracy of WiFi sensing systems is highly reliant on the location of transmitter and receiver devices, we propose a simple yet effective solution based on the utilization of spatially distributed transmitter antennas (connected to a single source device) which communicate to a receiver device. The legitimate or allowed receiver is provided the schedule of transmitter antennas; thus, it can leverage this information to more accurately recognize activities performed within the environment. On the other hand, an eavesdropper who is unaware of the transmission schedule will encode the CSI frames from all transmitter antennas as if they were transmitted by a single source and thus will fail to recognize the activities properly. Through experiments, we show the effectiveness of this approach considering different number of transmitter antennas as well as against different levels of eavesdroppers.