A Complex-Valued Self-Supervised Learning-Based Method for Specific Emitter Identification.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Specific emitter identification (SEI) methods excel with large datasets and high signal-to-noise ratios (SNR).
• Existing SEI techniques face performance degradation with small sample sizes and noisy environments.

Purpose of the Study:

• To develop a robust SEI scheme overcoming limitations of small datasets and high noise.
• To leverage unlabeled data effectively through self-supervised learning.

Main Methods:

• A complex self-supervised learning framework combining a pretext task and a downstream task.
• An optimized data augmentation strategy tailored for communication signals within a contrastive learning paradigm.
• Integration of a complex-valued network to enhance noise robustness.

Main Results:

• The proposed scheme demonstrates generality across varying labeled sample sizes (10-400).
• Significant improvements in accuracy and robustness were observed, with recognition rates increasing by 10-16% at SNRs of 10-15 dB.

Conclusions:

• The complex self-supervised learning approach effectively addresses SEI challenges in low-sample and high-noise scenarios.
• The method offers a robust and generalizable solution for identifying wireless emitters under adverse conditions.