CNN-based AMR of simultaneously transmitted SC-QAM and OFDM-QAM signals in a 220 GHz terahertz system

Area of Science:

• * Wireless Communication Engineering
• * Signal Processing
• * Machine Learning Applications

Background:

• * The terahertz (THz) frequency band offers abundant spectrum for ultra-high-speed wireless communications.
• * Band-pass delta-sigma modulation (BP-DSM) enables efficient transmission of high-order Quadrature Amplitude Modulation (QAM) signals in THz systems.
• * Adaptive demodulation of diverse QAM orders is crucial for flexible THz communication scenarios.

Purpose of the Study:

• * To introduce a Convolutional Neural Network (CNN) for automatic modulation recognition (AMR) in a 220-GHz THz system.
• * To enable the recognition of simultaneously transmitted Single-Carrier QAM (SC-QAM) and Orthogonal Frequency-Division Multiplexing QAM (OFDM-QAM) signals.
• * To achieve robust AMR across various modulation orders (32QAM to 512QAM) and transmission distances.

Main Methods:

• * Development of a CNN architecture with five convolutional and four max-pooling layers.
• * Utilizing progressively decreasing convolutional kernels to extract deep abstract features.
• * Testing the model's performance on nine different modulation combinations of SC-QAM and OFDM-QAM signals.

Main Results:

• * The CNN achieved stable, high-performance modulation classification for OFDM-QAM signals across tested orders.
• * Accurate recognition of SC-QAM signals from 32QAM to 256QAM was demonstrated.
• * The model showed consistent robustness across different wireless transmission distances, though 512QAM recognition accuracy decreased for SC-QAM.

Conclusions:

• * The developed CNN provides effective AMR for complex THz communication systems.
• * The study validates the feasibility of using deep learning for modulation recognition in THz bands.
• * This research lays a foundation for intelligent and efficient THz wireless transmission systems.