Local and Remote Digital Pre-Distortion for 5G Power Amplifiers with Safe Deep Reinforcement Learning.

Area of Science:

• Wireless Communication
• Signal Processing
• Artificial Intelligence

Background:

• High data rates and energy efficiency demands push power amplifiers (PAs) into nonlinear operation, causing signal distortions.
• Existing Digital Pre-Distortion (DPD) techniques struggle with complexity, adaptability, and resource limitations.
• Nonlinear distortions in PAs degrade wireless communication performance.

Purpose of the Study:

• Introduce Deep Reinforcement Learning-based Digital Pre-Distortion (DRL-DPD) to overcome limitations of current DPD systems.
• Reduce computational burden, enhance adaptability in dynamic environments, and minimize resource consumption.
• Ensure safety and regulatory compliance using a novel Safe Reinforcement Learning algorithm.

Main Methods:

• Developed DRL-DPD, integrating Deep Reinforcement Learning for adaptive DPD.
• Incorporated the Cautious-Recoverable-Exploration Deep Deterministic Policy Gradient (CRE-DDPG) algorithm for safe operation.
• Validated through extensive simulations and hardware experiments in local and remote configurations.

Main Results:

• DRL-DPD with CRE-DDPG demonstrated superior performance compared to existing DPD methods.
• The system effectively reduced computational complexity and resource usage.
• ACLR measurements were maintained above safety thresholds, ensuring regulatory compliance.

Conclusions:

• DRL-DPD, enhanced with CRE-DDPG, offers a promising solution for efficient wireless communication.
• The approach significantly surpasses current DPD limitations, particularly for 5G and future systems.
• This technology paves the way for more robust and resource-efficient wireless networks.