A Neural Network-Based Random Access Protocol for Crowded Massive MIMO Systems.

Area of Science:

• Telecommunications Engineering
• Wireless Communication Systems
• Machine Learning in Networks

Background:

• Fifth-generation (5G) and beyond networks require efficient handling of numerous user equipments (UEs).
• Grant-based random access (RA) protocols are suitable for human users with large data needs.
• The strongest user collision resolution (SUCRe) protocol leverages base station (BS) antennas for improved connectivity.

Purpose of the Study:

• To enhance the SUCRe protocol by replacing its retransmission rule with a neural network (NN).
• To enable decentralized collision resolution on the UE side for improved identification of the strongest user.
• To develop a single NN setup adaptable to varying system congestion and UE numbers.

Main Methods:

• Substitution of the SUCRe protocol's retransmission rule with a neural network (NN).
• Offline training of the NN across diverse system congestion levels.
• Decentralized collision resolution implemented on the user equipments (UEs).

Main Results:

• Substantial improvements in connectivity performance compared to existing protocols.
• No additional complexity or overhead required by the proposed NN-based approach.
• Enhanced energy efficiency due to fewer random access (RA) attempts.

Conclusions:

• The proposed NN-based method significantly enhances random access performance in 5G and beyond networks.
• The approach demonstrates robustness against variations in base station (BS) antennas and transmission power.
• The method offers improved energy efficiency and connectivity without increased system complexity.