TEA-RFFI: Temperature adjusted radio frequency fingerprint-based smartphone identification

Recently, radio frequency fingerprints (RFFs) have been widely applied for smartphone identification, since RFFs are distinguishable and hard to imitate. Compared with other types of RFFs, carrier frequency offset (CFO) in Wi-Fi signals is more robust and practical. However, when many smartphones need to be identified, the probability of CFO collision is high due to the weak distinguishability of CFO, and thus the primitive CFO-based identification solution will perform poorly. Fortunately, we find that the CFO varies with crystal oscillator temperature. Inspired by the phenomenon, we can actively adjust the crystal oscillator temperature to increase the difference between Wi-Fi device fingerprints for CFO collision mitigation. Firstly, we propose a non-intrusive temperature sensing and adjustment solution, which can obtain crystal oscillator temperature accurately and actively adjust its temperature to specified values without any additional hardware. Then, we investigate the temperature selection problem which aims to maximize overall differences among all smartphones, and propose the corresponding algorithm that combines greedy strategy and simulated annealing to assign a proper temperature value to each smartphone for identification. Finally, we implement the TEA-RFFI system and conduct several sets of experiments under the cases of different positions, time periods and scenarios. Experimental results demonstrate that TEA-RFFI can effectively identify 20 smartphones with over 90% precision, recall and F1-score. Even when the smartphones are moving, our proposed system still can identify them with over 89% precision, recall and F1-score.