Multi-Objective Optimization of Energy Saving and Throughput in Heterogeneous Networks Using Deep Reinforcement...

Area of Science:

• Wireless communication networks
• Green networking and energy efficiency
• Optimization techniques

Background:

• Mobile service providers deploy small cells for enhanced wireless networking using mmWave frequencies.
• Denser small-cell heterogeneous networks (HetNets) face significant power consumption challenges.
• Green networking is crucial for reducing CO2 emissions and combating climate change.

Purpose of the Study:

• To investigate the feasibility of deep reinforcement learning (DRL) for energy-efficient routing and throughput maximization in HetNets.
• To develop a dual-objective optimization model addressing both energy consumption and user throughput.
• To explore a novel application of DRL in a previously unaddressed wireless networking problem.

Main Methods:

• Formulated a mixed integer linear programming (MILP) dual-objective optimization model.
• Proposed a proximal policy (PPO)-based multi-objective deep reinforcement learning algorithm.
• Utilized an actor-critic model within an optimistic linear support framework for iterative solution searching.

Main Results:

• The proposed DRL algorithm demonstrated comparable performance to CPLEX in achieving energy savings and maximizing user throughput.
• Experimental results validate the effectiveness of the PPO-based approach for dual-objective optimization in wireless networks.
• The algorithm successfully balances energy efficiency with user throughput in dense HetNets.

Conclusions:

• Deep reinforcement learning, specifically the PPO algorithm, is a viable and effective approach for energy-efficient routing and throughput maximization in wireless HetNets.
• The developed algorithm offers a near-optimal online solution with fast inference times.
• This research contributes to the advancement of green networking solutions for future wireless infrastructures.