CNN-LSTM-AM approach for outdoor wireless optical communication systems

Area of Science:

• Electrical Engineering and Computer Science
• Artificial Intelligence in Communications

Background:

• Visible Light Communications (VLC) for vehicle-to-vehicle (V2V) systems face challenges like bit errors, high power consumption, and interference.
• Existing V2V communication systems require enhancement for improved performance and dependability in dynamic environments.

Purpose of the Study:

• To introduce an AI-based framework for enhancing VLC in V2V communication.
• To address critical issues including power consumption, bit errors, and signal interference in V2V systems.

Main Methods:

• A novel deep learning framework combining Convolutional Neural Networks (CNNs), Generative Adversarial Network (GAN), Gated Recurrent Unit (GRU), and Deep Denoising Autoencoder (DDAE) was developed.
• The framework includes modules for power reduction (GAN), performance enhancement (GRU), Bit Error Rate (BER) reduction (DDAE), and interference cancellation (CNN-U-Net).
• Simulations were conducted for V2V scenarios with varying longitudinal and lateral separations between vehicles.

Main Results:

• The proposed CNN-U-Net-GAN-GRU-DDAE model demonstrated significant improvements in throughput, power efficiency, BER reduction, and interference cancellation.
• The model achieved superior performance compared to existing models, outperforming CNN-U-Net, CNN-U-Net-GAN, and CNN-U-Net-GAN-GRU by 13.6%, 14.4%, and 4.2% respectively.
• An overall average performance improvement of 31.7% was observed compared to previous works.

Conclusions:

• The developed deep learning framework offers a reliable and scalable solution for enhancing V2V communication performance.
• The integrated AI model effectively mitigates common issues in V2V systems, paving the way for more robust communication.