Radio Frequency Fingerprinting in WBANs Using Complex-Valued Convolutional Neural Networks

Healthcare sensor authentication can be challenging due to the limited energy capacity and small size of the devices. The need for additional hardware, such as NFC or displays, for secure pairing, as imposed by the Bluetooth Low Energy (BLE) standard, further worsens this challenge. One promising alternative for node authentication is the use of Radio Frequency Fingerprint Identification (RFFI). We investigate RFFI’s effectiveness within Wireless Body Area Networks (WBANs) using the real-world BLE-WBAN dataset, which includes signals emitted by body-mounted BLE sensors inside an anechoic chamber. Diverging from most prior research, our approach leverages higher fidelity input (real and imaginary parts together) in Complex-Valued Neural Networks (CV-NNs), despite their increased computational cost, to realize a substantial improvement in fingerprinting accuracy over traditional Real-Valued Neural Networks (RV-NN). Our experiments demonstrate that CV-NNs excel, particularly at higher sampling rates, and reveal that the human body can positively influence RFFI, achieving up to 88% accuracy in on-body node fingerprinting among 12 nodes. While our results are promising, they highlight the need for further validation to confirm CV-NNs’ potential to bolster RFFI in WBANs.