Structured Learning for Wireless Scheduling in Naval Networks: Theory to Prototyping
Abstract
Approved for Public ReleaseFuture naval operations will be increasingly reliant on the integration of artificial intelligence (AI) and unmanned systems with existing force. For example, a recent Department of Navy publication has described scenarios where a swarmof unmanned vehicles monitor all domains of the battlespace and provide the sensed data to an AI at the tactical edge, who, in turn, process the information to obtain timely and precise decisions and situational awareness for warfighters.Communication networks will be a key to enable the integration of AI and manned/unmanned systems. Reliable and timely communications are needed to coordinatemanned and unmanned systems, to gather information for AI to process, and to deliver control decisions and situational awareness towarfighters. The objective of this project is to develop wireless scheduling policies to provide tailored service for data flows whose application behaviors and mobility patterns are not known a priori. Our main insight is to combine machine learning and analytical network optimization theory to obtain efficient learning algorithms. In addition to theoretical analysis, we will also build a prototype and evaluate the performance of learning algorithms. Our research efforts can be summarized in the following three thrusts:Thrust 1: Learning to Schedule for Goal-oriented Utility Maximization. This thrust will develop wireless scheduling algorithms that focus on achieving the goals of each flow, rather than optimizing traditional QoS metrics. We will model this problem as a restless bandit problem. Leveraging on various decomposing techniques in existing network optimization literature, we will develop learning-by-decomposition (LeaD) algorithms that can drastically reduce the dimension and hence improve the learning efficiency.Thrust 2: Learning to Schedule under Mobility Management Protocols. This thrust studies extending LeaD algorithms in the first thrust to be compatible with mobility management protocols. Most existing mobility management protocols employ a separate control channel to convey scheduling decisions to each wireless client. Such a mechanism limits the granularity of scheduling decisions. We will study joint optimization of scheduling in the data channel and the control channel. We will also propose and optimize a new mobility management protocol with less overhead.Thrust 3: Integration, Implementation, and Evaluation. This thrust will implement all research outcomes of this project on a software-defined wireless network testbed and evaluate their performance. The objective of this thrust is two-fold: First, we will adopt a hardware-software separation architecture to enable the real-time execution of learning algorithms on the resource-constrained software-defined radios. Second, we will conduct comprehensive experiments and emulations to evaluate the computation overheads and the network performance of learning algorithms.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 09, 2024
- Source ID
- N000142412615
Entities
People
- I-Hong Hou
Organizations
- Office of Naval Research
- Texas Engineering Experiment Station
- United States Navy