Modulation Classification Based on Signal Constellation Diagrams and Deep Learning.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Machine learning (ML) has applications in various fields, but its use in communication systems, particularly for modulation classification, remains underexplored.
• Modulation classification is a critical task in communication systems, often requiring complex manual feature selection in traditional ML approaches.

Purpose of the Study:

• To investigate the application of deep learning (DL) for modulation classification in communication systems.
• To evaluate the performance of DL models, specifically convolutional neural networks (CNNs), in this domain.
• To compare DL-based methods with traditional cumulant and ML-based algorithms.

Main Methods:

• Utilized two CNN-based DL models: AlexNet and GoogLeNet.
• Developed novel methods for representing modulated signals in grid-like data formats suitable for CNN input.
• Analyzed the impact of different signal representations on classification accuracy.

Main Results:

• DL-based methods, particularly CNNs, demonstrated significant performance advantages over traditional cumulant and ML-based algorithms.
• The chosen data representations for modulated signals positively impacted classification performance.
• The study confirmed the feasibility of applying DL for modulation classification in practical communication systems.

Conclusions:

• Deep learning offers a powerful and efficient approach to modulation classification in communication systems.
• CNNs, with appropriate data representation, can overcome the limitations of manual feature engineering in traditional methods.
• The DL-based approach shows considerable promise for enhancing the performance and simplifying the complexity of communication system tasks.