Detecting radio frequency interference for CSI measurements on COTS WiFi devices

In recent years, WiFi-based sensing applications have been proliferated due to growing capacities of the physical layer. Channel State Information (CSI), which depicts the characteristics of propagation environment and reflects different human behaviors, can be easily obtained on commodity WiFi devices with slight driver modification. For the sake of higher accuracy and robustness of CSI-based sensing, a variety of research efforts have been devoted to model refinement, algorithm optimization and data sanitization. Radio frequency interference (RFI) is a crucial problem, which, however, is surprisingly overlooked and largely unexplored. The sensing performance can be significantly boosted by identifying and properly handling the interfered CSI measurements. In this paper, we demonstrate that it is feasible to identify the interfered CSI measurements due to the unique properties induced by RFI. We propose two RFI detection algorithms by utilizing cyclostationary analysis from different angles. Experimental results on off-the-shelf WiFi devices show that both algorithms are robustly stable for different scenarios and can achieve a remarkable overall accuracy of > 90%.