Prediction of Quadcopter State through Multi-Microphone Side-Channel Fusion

Abstract

Improving trust in the state of Cyber-Physical Systems becomes increasingly important as more tasks become autonomous. We present a multi-microphone machine learning fusion approach to accurately predict complex states of a quadcopter drone in flight from the sound it makes using audio content analysis techniques. We show that using data fusion of multiple microphones, we can predict states with near-perfect results. Furthermore, we significantly improve the state predictions of single microphones, outperforming several other integration methods. These results show that side-channel information can be effectively used to improve the state assurance and security in Cyber-Physical Systems.