Radio Frequency Fingerprinting with Multi-Packet Adaptive Fusion

Radio frequency fingerprinting (RFF), a critical technology for wireless device identification, plays a key role in network security and the Internet of Things (IoT). Due to the complex and resource-constrained working environments of IoT devices, noise is significant in RFF for IoT devices. Suppressing noise while maintaining radio fingerprint information presents a challenge. Multi-packet inference is a method aimed at reducing the impact of noise. In this paper, we propose a multi-packet adaptive fusion method, named MPAF, to enhance the RFF accuracy of IoT devices. This method dynamically adjusts the weights assigned to each data packet, thereby reducing the influence of highly interfered packets and improving the accuracy of the inference. To update the weights, we employ an adaptive weighted sum algorithm that updates weights iteratively, achieving a dynamic balance for each packet. This method, based on the error balancing algorithm, enables the system to adapt to new data features during continuous learning processes. To verify the effectiveness of our proposed approach, we conduct comprehensive experiments using real-world LoRa dataset, and the results indicate that our proposed MPAF method exhibits higher accuracy than traditional methods. Particularly, our proposed approach significantly improves classification accuracy under low signal-to-noise ratio conditions.