Distributed Multi-Agent Deep Reinforcement Learning-Based Transmit Power Control in Cellular Networks.

Area of Science:

• Wireless Communication
• Artificial Intelligence
• Network Engineering

Background:

• Interference management is critical for multi-cell network performance.
• Controlling base station downlink transmit power is essential for high data rates but becomes complex with increasing cell density.

Purpose of the Study:

• To propose a transmit power control scheme using multi-agent deep reinforcement learning (MADRL) to maximize the sum rate in multi-cell networks.
• To enhance MADRL by integrating a long short-term memory (LSTM) architecture for improved state retention and decision-making.

Main Methods:

• A multi-agent deep reinforcement learning (MADRL) framework is developed for downlink transmit power control.
• A long short-term memory (LSTM) architecture is incorporated into the MADRL agents to leverage historical state information.
• Each base station agent utilizes only local channel state information, minimizing inter-agent communication.

Main Results:

• The proposed MADRL-LSTM scheme demonstrates improved sum rate performance compared to existing MADRL methods.
• The scheme effectively reduces the amount of signal messages exchanged between base station links.
• Local channel state information utilization proves efficient for decentralized control.

Conclusions:

• The integration of LSTM with MADRL offers a robust solution for transmit power control in dense cellular networks.
• Decentralized control using local information is feasible and efficient, reducing communication overhead.
• The proposed method enhances overall network capacity and user data rates through effective interference management.