Construction and Sharing of Radio Frequency Fingerprinting Dataset Based on Federated Learning

The Radio frequency fingerprinting (RFF) identification aims at performing feature measurement on received signal and then determines the emitter individual. Most RFF techniques focus on specific system with niche targeting at present, the generalization ability of which should be strengthened. Meantime, some technique may not achieve best effect because of the scale and quality of radio dataset. To address the problem, the framework to construct a large-scale RFF dataset is proposed based on Federated Learning (FL) to meet the needs of improving RFF identification ratio and accuracy. It can benefit researches in the field and also promote practice and industrialization of communication emitter individual identification techniques.