Frequency hopping signal detection based on optimized generalized S transform and ResNet.

Area of Science:

• Signal Processing
• Machine Learning
• Telecommunications

Background:

• Traditional frequency hopping signal detection faces limitations in time-frequency resolution and spectrum leakage.
• Machine learning approaches for frequency hopping signal detection often exhibit high complexity.

Purpose of the Study:

• To propose a novel method for frequency hopping signal detection using an optimized generalized S transform and a residual network.
• To enhance detection performance while reducing computational complexity.

Main Methods:

• Optimized Generalized S Transform (OGST) using a multi-population genetic algorithm to tune parameters $ \lambda $ and $ p $.
• Noise-robust normalization of the time-frequency spectrum.
• Residual network architecture for automatic feature learning from time-frequency spectra.

Main Results:

• The multi-population genetic algorithm demonstrated superior optimization efficiency, faster convergence, and more stable results than a standard genetic algorithm.
• The proposed residual network and OGST method achieved better detection performance.
• The new technique exhibited lower computational and storage complexity compared to hybrid convolutional/recurrent neural network algorithms.

Conclusions:

• The proposed method effectively detects frequency hopping signals by leveraging an optimized time-frequency representation and a deep learning approach.
• This technique offers a more efficient and accurate alternative for frequency hopping signal detection, addressing limitations of traditional and existing machine learning methods.