The Effect of Ambient Temperature On Device Classification Based On Radio Frequency Fingerprint Recognition

Physical layer authentication is an important technique for cybersecurity, especially in military scenarios. Device classification using radio frequency fingerprinting, which is based on recognizing device-unique characteristics of the transient waveform observed at the beginning of a transmission from a radio device, is a promising method in this context. In this study, the effect of the ambient temperature on the performance of radio device classification based on RF fingerprinting is investigated. The waveforms of the transient regions of the transmissions are recorded as images, and ResNet50 and InceptionV3 networks for image classification are used to determine the radio devices. The radio devices used in the study belong to the same brand, model, and production date, making the problem more difficult than classifying radio devices of different brands or models. Our results show that high levels of accuracy can be attained using convolutional neural network models such as ResNet50 and InceptionV3 when the test data and the training data are collected at the same temperature, whereas performance suffers when the test data and the training data belong to different temperature values. We provide the performance figures of a blended training model that uses training data taken at various temperature values. A comparison of the two networks is also provided.