Managing Energy Consumption of Devices with Multiconnectivity by Deep Learning and Software-Defined Networking

Area of Science:

• Computer Science
• Electrical Engineering
• Telecommunications

Background:

• Multiconnectivity enables simultaneous connections to multiple radio access technologies (5G, 4G LTE, WiFi), crucial for meeting escalating mobile data demands.
• Multipath TCP (MPTCP) facilitates reliable data transmission over these diverse links, but increases energy consumption in battery-powered devices.
• Managing energy efficiency in multihomed wireless devices is a significant challenge for current and future mobile networks.

Purpose of the Study:

• To develop and evaluate an energy management strategy for multiconnected devices utilizing MPTCP.
• To leverage Software-Defined Networking (SDN) and Deep Neural Networks (DNNs) for optimizing energy consumption.
• To enhance network throughput performance alongside energy savings.

Main Methods:

• Implementation of two lightweight algorithms on an SDN controller for managing multiconnectivity.
• Utilizing a hardware testbed with dual-homed wireless nodes connected to WiFi and cellular networks.
• Employing a DNN trained on diverse network scenarios to refine network connection decisions.

Main Results:

• The proposed SDN and DNN-based approach significantly reduces device energy consumption.
• The method achieves improved network throughput performance compared to single-path TCP and standard MPTCP algorithms (Cubic, BALIA).
• Experimental validation demonstrates the effectiveness of the algorithms in real-world dual-homed network environments.

Conclusions:

• SDN and DNN integration offers an effective solution for managing energy consumption in multiconnected devices.
• The developed algorithms provide a practical method for optimizing resource utilization in 5G and future wireless networks.
• This approach balances the need for high performance with the critical requirement of energy efficiency for mobile devices.