DFFNet: Deep Federated Radio Fingerprinting Based on Fractional Wavelet Scattering Network

The rapid development of the Internet of Things (IoT) has highlighted the critical importance of security and privacy in cognitive cities. In this context, radio frequency fingerprinting (RFF) identification has emerged as an excellent authentication scheme that provides intelligent and efficient identification in IoT systems. By leveraging RFF, we can improve the security and privacy of cognitive cities while also enhancing their operational efficiency. The RF nonlinear features are unique and unchanging, operating at the hardware level. This attribute renders them amenable to sufficient learning through convolution neural networks (CNNs), which have demonstrated remarkable identification accuracy. Nonetheless, CNNs suffer from a lack of strong interpretability and necessitate vast quantities of training data. Additionally, the enormous amount of data required for training imposes greater demands on computing resources, which are often inadequate in IoT. Moreover, traditional training schemes employ centralized datasets, which cannot ensure corresponding privacy. More recently, federated learning and fractional wavelet scattering network have been proposed to solve the problems above. To address this issue, we in this paper proposed a deep federated radio fingerprinting based on fractional wavelet scattering network (DFFNet) which can acquire the subtle features from non-stationary signals. The advantage of DFFNet is that the federated learning is applied to achieve privacy preserving during the learning process. Meanwhile, fractional wavelet is suitable for non-stationary signal's features extraction with high interpretability. The representative experiment results demonstrate that hybrid federated framework DFFNet achieve about 99.1% identification accuracy under practical application.