A Lightweight Deep Learning RF Fingerprint Recognition Method

Current deep learning-based RF fingerprint recognition methods have the problems of large models and high operational complexity, for which this paper proposes an RF fingerprint recognition method (CNN-GRU network) using a combination of convolutional neural networks and gated recurrent unit networks. The algorithm uses a convolutional layer to extract IQ orthogonal correlation features and learning temporal data, then extracts temporal features through the GRU network, and finally has an output layer to output recognition results. The number of parameters of the algorithm is reduced by using small convolutional kernels and replacing multiple fully connected layers with gated cyclic units. Experimental results show that the algorithm achieves a recognition rate of 98.7
^%
on a software radio dataset with multiple USRP X310s, and the algorithm has fewer parameters and is more lightweight than network models with the same recognition rate, which is of better engineering application.