Coexistence Scheme for Uncoordinated LTE and WiFi Networks Using Experience Replay Based Q-Learning

Area of Science:

• Wireless communication networks
• Spectrum sharing technologies
• Machine learning applications in networking

Background:

• Increasing demand for broadband applications necessitates efficient spectrum utilization.
• Long-Term Evolution-Unlicensed (LTE-U) technology aims to offload traffic to unlicensed spectrum.
• Coexistence of LTE-U and WiFi in unlicensed bands presents significant challenges due to interference.

Purpose of the Study:

• To propose novel Q-learning based coexistence schemes for uncoordinated LTE-U and WiFi networks.
• To address the complexity and overhead associated with centralized coordination methods.
• To enhance the performance of LTE-U and WiFi coexistence using machine learning.

Main Methods:

• Implementation of Experience Replay (ER) and Reward selective Experience Replay (RER) Q-learning algorithms.
• Development of a WiFi saturation sensing model for LTE-U traffic demand estimation.
• Performance comparison with existing rule-based and standard Q-learning coexistence schemes.

Main Results:

• The Reward selective Experience Replay (RER) Q-learning scheme demonstrates faster convergence compared to the ER scheme.
• RER Q-learning achieved a 19.1% and 5.2% enhancement in aggregated throughput over rule-based and Q-learning schemes, respectively.
• RER Q-learning provided a 16.4% and 10.9% improvement in fairness compared to rule-based and Q-learning schemes, respectively.

Conclusions:

• Proposed ER and RER Q-learning techniques offer effective solutions for uncoordinated LTE-U and WiFi coexistence.
• The RER Q-learning scheme significantly outperforms existing methods in terms of convergence speed, throughput, and fairness.
• Machine learning, particularly RER Q-learning, presents a promising approach for optimizing shared spectrum utilization.