Digital twin-enabled self-evolved optical transceiver using deep reinforcement learning.

Area of Science:

• Optical Communications
• Artificial Intelligence
• Network Engineering

Background:

• Programmable optical transceivers (POT) are crucial for flexible optical fiber communications.
• Existing POT modeling lacks adaptivity to time-varied network states due to limited data and models.
• Adaptive control is essential for optimizing performance in dynamic optical networks.

Purpose of the Study:

• To introduce a novel digital twin (DT) approach for adaptive modeling and control of POTs.
• To overcome limitations of traditional POT modeling and control methods.
• To enhance the real-time adaptability of optical transceivers in evolving network conditions.

Main Methods:

• Development of a digital twin (DT) model for POTs.
• Integration of deep reinforcement learning (DRL) to enable dynamic control within the DT framework.
• Experimental and simulation-based validation of the proposed adaptive POT system.

Main Results:

• The proposed DT-enabled POT system achieved the lowest spectrum consumption compared to traditional methods.
• Minimum latency was observed in the DT-based POT system, outperforming neural network and maximum capability provisioning approaches.
• Demonstrated superior performance in adaptive modeling and control for dynamic network states.

Conclusions:

• Digital twin technology, powered by DRL, offers a powerful solution for adaptive POT modeling and control.
• The proposed method significantly improves spectral efficiency and reduces latency in optical networks.
• This work pioneers a new direction for adaptive optical component management in dynamic optical networks.