Automatic Modulation Classification Based on Deep Feature Fusion for High Noise Level and Large Dynamic Input.

Area of Science:

• Electrical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Automatic modulation classification (AMC) is crucial for spectrum monitoring and cognitive radio.
• Increasingly complex electromagnetic environments pose challenges due to high background noise and large dynamic input ranges.
• Existing AMC methods struggle with noise interference and signal variability.

Purpose of the Study:

• To propose a novel feature fusion scheme for robust automatic modulation classification.
• To enhance signal representation stability and efficiency by integrating complementary features.
• To overcome limitations of high background noise and large dynamic input in AMC.

Main Methods:

• Signals transformed to frequency domain using Fast Fourier Transform (FFT) and Welch power spectrum analysis.
• Convolutional Neural Network (CNN) and Stacked Auto-Encoder (SAE) used for frequency-domain feature extraction.
• Instantaneous amplitude/phase statistics and higher-order cumulants (HOC) extracted for time-domain statistical features.
• Feature fusion followed by a Probabilistic Neural Network (PNN) for classification.

Main Results:

• The proposed feature fusion method demonstrates superior performance in automatic modulation classification.
• Achieved a classification accuracy of 99.8% at a signal-to-noise ratio (SNR) of 0 dB.
• Effectively suppresses interference from background noise and handles large dynamic input ranges.

Conclusions:

• The deep learning-based feature fusion scheme provides a stable and efficient representation for modulation types.
• The method significantly improves AMC performance in challenging electromagnetic environments.
• This approach offers a promising solution for advanced spectrum monitoring and cognitive radio applications.