A Self-Supervised Specific Emitter Identification Method Based on Contrastive Asymmetric Masked Learning.

Area of Science:

• Cybersecurity
• Signal Processing
• Machine Learning

Background:

• Specific emitter identification (SEI) is vital for wireless security but current deep learning methods require extensive labeled data, limiting their use in non-cooperative scenarios.
• Existing SEI techniques struggle with data scarcity and feature discriminability in real-world applications.

Purpose of the Study:

• To propose a novel contrastive asymmetric masked learning-based SEI (CAML-SEI) method for effective wireless device identification with scarce labeled samples.
• To enhance the learning of fine-grained local radio frequency fingerprint (RFF) features and improve feature discriminability.

Main Methods:

• An asymmetric auto-encoder architecture with a channel squeeze-and-excitation residual block encoder for RFF feature extraction.
• A lightweight convolutional decoder for masked signal reconstruction and a learnable non-linear mapping for feature compression.
• A contrastive loss function for positive sample aggregation and negative sample separation, jointly optimized with signal reconstruction.

Main Results:

• The CAML-SEI method effectively learns generalized RFF features from signals.
• Experimental results on ADS-B and Wi-Fi datasets show superior performance compared to existing SEI methods.
• The approach demonstrates robustness and effectiveness in scenarios with limited labeled data.

Conclusions:

• The proposed CAML-SEI method offers a powerful solution for specific emitter identification, particularly in data-scarce environments.
• This work advances SEI technology by enabling robust device identification through enhanced feature learning and contrastive optimization.
• The findings have significant implications for improving wireless communication security and device authentication.