Neural network-based optimal and adaptable power allocation for real-time FSO-RF communications using Jetson nano

Area of Science:

• Wireless communication networks
• Optimization algorithms
• Deep learning applications

Background:

• Power allocation (PA) is crucial for managing optical-radio wireless networks but analytical methods are slow.
• Deep Neural Networks (DNNs) offer a potential solution for real-time PA with high accuracy.

Purpose of the Study:

• To develop and implement a novel, real-time optimal power allocation (PA) system for Free Space Optics-Radio Frequency (FSO-RF) wireless networks.
• To leverage DNNs for fast and accurate PA, overcoming the limitations of traditional analytical approaches.

Main Methods:

• Developed a real-time optimal PA system using two parallel three-layer DNNs for FSO-RF PA.
• Utilized the Weighted Minimum Mean Square Error (WMMSE) algorithm with fading and user priorities for RF channel training data.
• Employed an analytical Bit Error Rate (BER) algorithm to adjust transmitter power intervals for FSO channels, considering various modulation schemes and device sizes.
• Implemented the DNNs on a Jetson Nano platform for performance evaluation.

Main Results:

• The implemented DNN-based system achieved a sum rate of 1.6 Gbps and 97.82% average accuracy for the RF channel across different user loads.
• The FSO channel component reached a data rate of 1.6 Gbps with 98.87% average accuracy.
• The system demonstrated significant speed and accuracy advantages over analytical methods.

Conclusions:

• The proposed DNN-based PA system provides a viable and efficient solution for real-time optimal PA in FSO-RF wireless networks.
• The system's high accuracy and speed make it suitable for practical deployment in demanding wireless communication scenarios.