A Multi-Task Learning Method for Communication Emitter Individual Identification

This paper proposes a multi-task convolutional neural network identification method based on raw I/Q data, which is used to solve modulation identification and specific emitter identification in communication emitters. The modulation identification and communication emitter identification as two interrelated learning tasks. This method extracts features from raw I/Q data through a deep neural network with shared parameters. Two branch networks with different structures are then used for classification and identification, while joint optimization training is conducted on these two tasks to facilitate mutual learning. The paper employs 15 different USRPs for validation, conducting both single-task and multi-task model analyses to verify the effectiveness of the multi-task architecture. The results demonstrate that, while ensuring the accuracy of modulation identification, the multi-task model improves the identification rate of communication emitters by 1.1% compared to the single-task model identification method.