The Importance of Expert Knowledge for Automatic Modulation Open Set Recognition.

Area of Science:

• Electrical Engineering
• Signal Processing
• Machine Learning

Background:

• Automatic modulation classification (AMC) is crucial for communication system management.
• Machine learning enhances AMC effectiveness for radio signals.
• Automatic modulation open-set recognition (AMOSR) for known and unknown signals is underexplored.

Purpose of the Study:

• To propose a novel framework, MMPRF, for improved AMOSR performance.
• To address the challenge of simultaneously recognizing known (closed-set) and unknown (open-set) modulation signals.
• To enhance AMOSR capabilities by leveraging domain knowledge and advanced techniques.

Main Methods:

• Developed a multi-modal marginal prototype framework (MMPRF) using one-vs-other partitioning and marginal restrictions.
• Extracted signal-related features based on wireless signal domain knowledge.
• Implemented a Generative Adversarial Network (GAN)-based strategy for unknown sample generation.

Main Results:

• The proposed MMPRF framework demonstrates superior performance in AMOSR tasks.
• Experimental results validate the effectiveness of MMPRF on public radio modulation datasets.
• MMPRF outperforms existing state-of-the-art AMOSR methods.

Conclusions:

• The novel MMPRF framework significantly advances the field of AMOSR.
• The approach effectively handles both known and unknown radio frequency signal modulations.
• This work provides a robust solution for enhanced communication system monitoring and management.