Measuring Wireless
Fingerprints Inside a Wireless Sensor Network
Wireless fingerprints authenticate transmitters
in wireless networks, using attributes of the physical wireless signal rather
than information being conveyed with that signal. They can be used effectively
in an intrusion detection system for networks, where nodes are physically
vulnerable. While previous research shows that wireless fingerprints work
reliably in a laboratory setting, little work has been published showing the
feasibility of their implementation within a real network. We present methods
for wireless fingerprints using data sampled at the demodulation rate. This
eliminates the need for high bandwidth data processing, making them feasible
from inside a wireless sensor network. We analyze their classification
performance empirically for different network conditions and theoretically
examine aspects of their secure usage in a network.
To the best of our knowledge, our
research is the only published work that uses representative wireless
networking hardware for wireless fingerprints. We discriminate between
different IEEE 802.15.4 2.4 GHz Radio Frequency (RF) sources, using the SiLabs
IEEE 802.15.4 WSN node development platform and the Ettus
Labs USRP1 Software-Defined Radio. The wireless fingerprinting method
implemented on the SiLabs WSN node discriminates between RF sources using
differences in the Automatic Gain Control circuitry time response during the
initial appearance of RF signals. The results show that different RF sources
can be distinguished over short transmission distances. The more sophisticated
USRP1 device exploits differences in the phase attributes of RF signals using a
larger set of demodulated data samples than was available with the WSN node.
The USRP1 classifies more accurately over a wider range of network conditions:
time, transmission distance and also different receiving devices. Our average classification accuracies are: 99.6% at short range, 95.3% at medium
range and 81.9% at long range using five SiLabs devices.
We demonstrate the independence of
classification errors made over different RF channels and the benefit of using
multiple nodes for classification. This suggests that nodes can collaborate to
increase the reliability of wireless fingerprints, either by comparing their
classification decisions or by aggregating their fingerprints. We present a
secure group key establishment protocol, using wireless fingerprints for
authentication, for use in a network containing malicious nodes.