A mechanism design approach to bandwidth allocation in tactical data networks

Ankur Mour, Purdue University

Abstract

The defense sector is undergoing a phase of rapid technological advancement, in the pursuit of its goal of information superiority. This goal depends on a large network of complex interconnected systems - sensors, weapons, soldiers - linked through a maze of heterogeneous networks. The sheer scale and size of these networks prompt behaviors that go beyond conglomerations of systems or `system-of-systems'. The lack of a central locus and disjointed, competing interests among large clusters of systems makes this characteristic of an Ultra Large Scale (ULS) system. These traits of ULS systems challenge and undermine the fundamental assumptions of today's software and system engineering approaches. In the absence of a centralized controller it is likely that system users may behave opportunistically to meet their local mission requirements, rather than the objectives of the system as a whole. In these settings, methods and tools based on economics and game theory (like Mechanism Design) are likely to play an important role in achieving globally optimal behavior, when the participants behave selfishly. Against this background, this thesis explores the potential of using computational mechanisms to govern the behavior of ultra-large-scale systems and achieve an optimal allocation of constrained computational resources Our research focusses on improving the quality and accuracy of the common operating picture through the efficient allocation of bandwidth in tactical data networks among self-interested actors, who may resort to strategic behavior dictated by self-interest. This research problem presents the kind of challenges we anticipate when we have to deal with ULS systems and, by addressing this problem, we hope to develop a methodology which will be applicable for ULS system of the future. We build upon the previous works which investigate the application of auction-based mechanism design to dynamic, performance-critical and resource-constrained systems of interest to the defense community. In this thesis, we consider a scenario where a number of military platforms have been tasked with the goal of detecting and tracking targets. The sensors onboard a military platform have a partial and inaccurate view of the operating picture and need to make use of data transmitted from neighboring sensors in order to improve the accuracy of their own measurements. The communication takes place over tactical data networks with scarce bandwidth. The problem is compounded by the possibility that the local goals of military platforms might not be aligned with the global system goal. Such a scenario might occur in multi-flag, multi-platform military exercises, where the military commanders of each platform are more concerned with the well-being of their own platform over others. Therefore there is a need to design a mechanism that efficiently allocates the flow of data within the network to ensure that the resulting global performance maximizes the information gain of the entire system, despite the self-interested actions of the individual actors. We propose a two-stage mechanism based on modified strictly-proper scoring rules, with unknown costs, whereby multiple sensor platforms can provide estimates of limited precisions and the center does not have to rely on knowledge of the actual outcome when calculating payments. In particular, our work emphasizes the importance of applying robust optimization techniques to deal with the uncertainty in the operating environment. We apply our robust optimization - based scoring rules algorithm to an agent-based model framework of the combat tactical data network, and analyze the results obtained. Through the work we hope to demonstrate how mechanism design, perched at the intersection of game theory and microeconomics, is aptly suited to address one set of challenges of the ULS system paradigm - challenges not amenable to traditional system engineering approaches.

Degree

M.S.A.A.

Advisors

DeLaurentis, Purdue University.

Subject Area

Economics|Aerospace engineering|Military studies|Systems science

COinS