Blind Recognition of Frame Synchronization Based on Deep Learning.

Area of Science:

• Digital Communications
• Signal Processing
• Machine Learning

Background:

• Non-cooperative communication systems face challenges in frame synchronization, especially at high bit error rates (BER).
• Existing frame detection methods struggle with accuracy under noisy conditions.

Purpose of the Study:

• To propose a deep-learning-based blind recognition algorithm for frame synchronization in non-cooperative systems.
• To improve frame detection performance under high BER conditions.

Main Methods:

• Binary data is interpolated and converted into grayscale images, then scaled to RGB images.
• Images with specific stripe features, based on matching radius, frame length, and synchronization code, are classified.
• A neural network is trained on these classified images for effective test data classification.

Main Results:

• The algorithm achieves 100% frame recognition probability for BER below 0.2.
• Recognition probability remains above 90% even at a BER of 0.25.
• Demonstrates over 60% performance improvement compared to traditional algorithms.

Conclusions:

• The proposed deep learning algorithm effectively addresses frame synchronization challenges in high-error environments.
• Converting data sequences into RGB images offers a robust solution for frame synchronization in difficult communication scenarios.