Random Access Using Deep Reinforcement Learning in Dense Mobile Networks.

Area of Science:

• Telecommunications Engineering
• Wireless Communication Systems
• Artificial Intelligence in Networks

Background:

• 5G and Beyond 5G networks employ high-frequency millimeter-wave (mmWave) bands to address bandwidth scarcity.
• Limited range of mmWave necessitates heterogeneous networks with small and macrocells (TRxPs) to ensure coverage.
• Efficient random access for users to select optimal TRxPs is crucial but challenging due to network dynamics.

Purpose of the Study:

• To develop a decentralized scheme for optimizing random access in heterogeneous 5G and Beyond 5G networks.
• To address the challenge of selecting less congested Transmission and Reception Points (TRxPs) without centralized control.
• To minimize random access delay and enhance the probability of successful user access.

Main Methods:

• Formulation of an optimization problem to approximate and minimize random access delay.
• Implementation of a reinforcement learning-based scheme for estimating TRxP congestion.
• Development of a deep learning algorithm to select the optimal access point for users.

Main Results:

• The proposed deep learning algorithm significantly improves random access performance compared to existing methods.
• Average access delay was reduced by 58.89% compared to the standard 3GPP algorithm.
• The probability of successful random access also demonstrated notable improvement.

Conclusions:

• Reinforcement learning offers an effective solution for decentralized congestion estimation and optimal TRxP selection in 5G and Beyond 5G networks.
• The deep learning approach enhances network efficiency by minimizing user access delays.
• The proposed method provides a practical and performant solution for improving random access in future mobile networks.