FICDF: A Federated Incremental Learning Framework for IoT Device Fingerprinting

Although Internet of Things (IoT) devices have been widely used, their simple structure restricts the deployment of advanced security protocols, making them vulnerable to cyber-attacks. Consequently, network administrators must adopt a zero-trust approach to identify each IoT communication entity. Recently, radio frequency (RF) or traffic data based fingerprinting has proven to be an effective IoT identification technique. Nevertheless, existing fingerprinting methods face limitations due to privacy concerns, the extreme non-independent distribution of fingerprint data, and the dynamic updating of IoT devices, hindering real-world deployment. We propose a Federated IoT Continuous Device Fingerprinting (FICDF) mechanism to address these challenges. In the traffic data preprocessing stage, we design a binary encoding and temporal tensor channel stacking mechanism to enhance the device-specific features in each training sample. Within the framework of federated incremental learning, we introduce a k-means multi-centroid exemplar-based Gaussian noise feature-sharing mechanism to simultaneously address the extreme non-IID nature of the data and the issue of catastrophic forgetting. To the best of our knowledge, this is the first study to tackle federated incremental device fingerprinting under extreme non-IID conditions. The code for this paper can be found at https://github.com/Squiding/FICDF_WiOpt2024