Optimal User Scheduling in Multi Antenna System Using Multi Agent Reinforcement Learning.

Area of Science:

• Wireless communication systems
• Signal processing
• Machine learning applications in telecommunications

Background:

• Multiple Input Multiple Output (MIMO) systems offer enhanced data rates but require efficient resource management.
• Effective user scheduling is crucial for maximizing capacity in Multi-User MIMO (MU-MIMO) environments.
• Current scheduling mechanisms may not fully exploit the spatial degrees of freedom offered by multi-antenna systems.

Purpose of the Study:

• To investigate the user selection problem in MU-MIMO systems.
• To develop an optimal scheduling policy for simultaneous transmissions.
• To improve system capacity and spectral efficiency through intelligent user grouping.

Main Methods:

• Formulation of the MU-MIMO scheduling as a single-state Markov Decision Process (MDP).
• Application of multi-agent Reinforcement Learning (RL) to solve the formulated MDP.
• Optimization of user selection based on channel conditions and available resource blocks.

Main Results:

• The RL-based methodology successfully derives an optimal scheduling policy.
• The proposed method demonstrates a significant improvement in aggregated sum-rate.
• Achieved a 20% higher sum-rate performance compared to conventional scheduling methods.

Conclusions:

• Reinforcement learning provides an effective framework for optimal user selection in MU-MIMO systems.
• The developed scheduling policy enhances system capacity by efficiently utilizing spectrum resources.
• This approach represents a promising advancement for future wireless communication systems.