Application of a Deep Transfer Network to enable Cyber-Physical Sensing using Opportunistic Wireless Signals

The pervasive deployment of the wireless sensor networks (WSN) in modern society has significantly changed the nature of the radio frequency spectrum, especially within the ISM band. Information collected from wireless sensors is distributed and exchanged via wireless nodes and anchors for numerous applications in smart homes and cities, transportation, urban planning and healthcare. However, the unplanned manner in which WSN nodes and anchors are deployed have led to a pell-mell of wireless bursts which limit the capacity and
robustness of WSN’s, whilst also increasing inter-system interference. The work described in this paper presents a novel methodology to disentangle the chaotic radio transmissions in congested wireless signal environments in order to provide a sensing capability. The wireless transmissions are treated as pseudo noise waveforms which eliminates the requirement to extract network information through signal demodulation or decoding. Instead, we leverage the phase and frequency components of these wireless bursts in conjunction with
cross ambiguity function (CAF) signal processing and a Deep Transfer Network (DTN). Lastly, we use 2.4GHz 802.11 (WiFi) signals to demonstrate experimentally the capability of this technique for human respiration detection (including hrough a wall), and classifying everyday but complex human motions such as standing, sitting and falling.