StableFP: NN-Based Hardware Fingerprint Extractor for LoRa Device Identification

Hardware fingerprint is a new dimension of security mechanisms in low power wide area networks (LPWANs). It is hard to emulate for attackers and does not increase the computing and energy burden of transmitters. long range (LoRa) is a long-range communication technology designed for battery-powered devices. In practice, LoRa is vulnerable to malicious attacks such as replace attack. Therefore, the hardware fingerprint is an excellent supplementary mechanism of LoRa security. However, the variable wireless environment contaminates the extracted fingerprints. The long wireless channel adds a large amount of the environment dependent information to the hardware features extracted from LoRa devices. In this paper, we propose StableFP which is a neural network (NN) based device identifier for long range wide area network (LoRaWAN). StableFP extracts stable and representative hardware features from channel frequency response (CFR) as the fingerprint, and it eliminates the environment dependent information caused by wireless environments. We implement StableFP on a software defined radio (SDR) testbed which consists of 4 commercial LoRa nodes. The result demonstrates that StableFP achieves over 90% identification accuracy in unseen environments under an over 5 dB signal to noise ratio (SNR).