Online Learning for Provable Timely and Reliable Deliveries in Highly Dynamic Wireless Networks
Abstract
Future naval missions will be increasingly relying on the joint operations among vastly different entities, including warships, unmanned aerial/surface vehicles, and marines. To ensure the success of missions and the safety of soldiers, it is imperative to build wireless networks that are capable of delivering safety-critical messages, such as monitoring data, situational awareness, and tactical commands, in real-time. Motivated by this observation, the objective of this research is to design network algorithms that are both provably timely and reliable in delivering and adaptive to mobility and changing environments.The research plans of this projectare based on two recent lines of work that show great promise in quantifying and optimizing delay and reliability on a per-packet basis. The first line of work studies the second-order performance, namely, the mean and the temporal variance, of packet deliveries in wireless networks. Similar to how Central Limit Theorem provides a much richer characterization than the Law of Large Numbers, the analysis on second-order performance leads to important insights on understanding and optimizing network behaviors on the fast time-scale. Another line of work develops and analyzes online network algorithms that make routing and scheduling decisions without knowledge about future user mobility and packet generations. It specifically focuses on quantifying the fundamental relations between the amount of bandwidth and the achievable performance guarantees. The plans can be summarized in the following three thrusts:Thrust 1: Optimal network algorithms for timely and reliable deliveries through second-order analysis and design: This thrust will establish the second-order capacity region for dynamicwireless networks with fading channels and stochastic packet arrivals. Specifically, this thrust seeks to model each stochastic component by its mean and temporal variance, and then characterize the fundamental limits of the achievable mean and temporal variance of the packet delivery process for each wireless client. Such characterization will then be used to precisely quantify the timeliness and reliability of packet deliveries.Thrust 2: Online algorithms with the optimal performance guarantees under worst-case conditions: This thrust will study the challenging problem of providing provable timelinessand reliability guarantees for any mobility and packet arrival patterns. The focus will be onquantifying and optimizing competitive ratios and regrets of online network algorithms andon establishing the fundamental relations between the amount of bandwidth and competitiveratios and regrets.Thrust 3: Online learning of network algorithms: Most network algorithms need knowledgeabout some system parameters, such as the number of flows and their respective channelconditions, to compute the optimal routing and scheduling policy. Indynamic systems, thealgorithms need to re-compute the solutions every time the parameters change. This thrust willdevelop online learning algorithms that quickly adapt to any system changes.The outcomes of the proposed research efforts will result in a rich suiteof network algorithms that provide timely and reliable deliveries in a wide range of scenarios. They will also be able to quickly adapt to changing system environments without sacrificing timeliness and reliability. Such network algorithms will open up many opportunities for future naval applications that leverage communication networks for real-time sensing, monitoring, and coordination.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 05, 2021
- Source ID
- N000142112385
Entities
People
- I-Hong Hou
Organizations
- Office of Naval Research
- Texas Engineering Experiment Station
- United States Navy