Specific Emitter Identification Based on Multi-Scale Attention Feature Fusion Network.

With the rapid advancement of the Industrial Internet of Things (IIoT), the proliferation of connected devices across various networks has significantly increased the security risk of unauthorized access by malicious entities. Therefore, ensuring secure and reliable wireless access becomes crucial by accurately identifying illegal devices and preventing potential attacks. One well-established device identification technique is the use of radio frequency fingerprinting (RFF) for specific emitter identification (SEI), which provides a reliable and secure method. In order to enhance the performance of SEI, a two-scale self-attentional feature fusion network (TSFFN), based on different scales, was introduced. The TSFFN leverages distinct convolution kernel sizes to extract RFF from two receptive fields. To assess the identification accuracy, an extensive dataset comprising authentic radio signals generated from a specialized aviation monitoring system, the automatic dependent surveillance - broadcast (ADS-B), is employed. Experimental results show that the proposed TSFFN achieved better identification accuracy than other models and better performance in imperfect channel environment.