User Pairing for Delay-Limited NOMA-Based Satellite Networks with Deep Reinforcement Learning

Area of Science:

• Satellite Communications
• Wireless Networks
• Network Optimization

Background:

• Non-Orthogonal Multiple Access (NOMA) enables efficient resource sharing in satellite networks.
• Varying Quality of Service (QoS) requirements, particularly delay constraints, pose challenges for NOMA performance.
• Effective capacity is a key metric for evaluating performance under delay-limited QoS.

Purpose of the Study:

• To address the user pairing problem in power-domain NOMA satellite networks with diverse delay QoS requirements.
• To optimize resource utilization and network performance by efficiently forming NOMA user pairs.
• To develop a dynamic user selection strategy that overcomes the intractability of traditional methods.

Main Methods:

• Utilized effective capacity to model the impact of delay QoS on performance.
• Determined a power allocation coefficient to maintain performance for delay-sensitive users compared to Orthogonal Multiple Access (OMA).
• Employed a Deep Reinforcement Learning (DRL) algorithm for dynamic and optimal user selection in delay-limited NOMA scenarios.

Main Results:

• The DRL algorithm successfully identified optimal user pairings for NOMA in each state.
• The proposed DRL-based scheme demonstrated superior performance compared to random selection and OMA.
• The approach effectively managed channel conditions and delay QoS requirements for user pairing.

Conclusions:

• The DRL-based user selection scheme provides an effective solution for optimizing NOMA satellite networks.
• This method significantly improves performance by dynamically pairing users based on QoS and channel conditions.
• The findings highlight the potential of DRL in advancing complex wireless network management.