A Greedy Policy for Fleet-Level Radar Resource Management Jinxin Zhao, Jinwoo Seok, Jhanani Selvakumar, Ricardo Bencatel, Pierre Kabamba and Anouck Girard Department of Aerospace Engineering University of Michigan, Ann Arbor, Michigan 48109-2140 Email:{jinxinzh, sjinu, jhanani, bencatel, kabamba, anouck}@umich.edu Abstract— A model for a phased-array radar in the context of a defensive Naval system has been developed using hierarchical finite state machines. This model is used to study resource management for a fleet of ships equipped with one radar unit each. This fleet is under attack from aerial enemy agents. A control algorithm formulated on this specific model is used to dynamically generate a greedy policy for radar operation. The defense system is evaluated in its capability to acquire and track those aerial threats, in two configurations -centralized and decentralized (independent). The performance is quantified on the basis of time taken to establish the threat trajectory, time to compute the policy and the number of threats the system fails to track. A comparison of performance in the two configurations is provided. I. INTRODUCTION A. Motivation The problem of radar resource management is significant to modern military systems. The era of fast and intelli- gent warfare necessitates use of complex radar systems with sophisticated detection and classification algorithms. A constraining factor is power utilization, particularly in offshore defense systems such as Naval vessels. In order to obtain a solution to the radar resource management problem that is feasible and matches performance standards, a radar model from the ONR has been adapted and an optimization procedure applied to it. The model is presented in this paper along with the control algorithm which generates a greedy policy for resource management. B. Mission Overview Consider a geographical area containing two forces: en- emy and friendly. The enemy force is composed of agents that pose a threat to the friendly force. The friendly force is composed of radar platforms, which are equipped to detect and track threats. The goals for the friendly force are to detect all the threats within the area, and to track all the detected threats. A practical radar model has been built and a control algorithm designed to fulfill these goals. Jinxin Zhao, Jinwoo Seok, and Jhanani Selvakumar are graduate students, and Ricardo Bencatel is a post-doctoral research fellow, at the Department of Aerospace Engineering, University of Michigan, Ann Arbor, Michigan 48109-2140. Pierre Kabamba and Anouck Girard are faculty at the Department of Aerospace Engineering, University of Michigan, Ann Arbor, Michigan 48109-2140. C. Literature Review A number of radar models exist at present. They corre- spond to different types of radars or are developed with spe- cific radar applications in consideration. One popular method of modeling radars is by Markovian processes. Visnevski et al [1] use a generalized semi-Markov process to model the emitter of a multi-function radar in application to electronic warfare. Watson and Blair [2] employ a Markov chain to mix state estimates from multiple models in their Interacting Multiple Model (IMM) algorithm. Ghosh et al [3] attempt to provide an integrated framework to optimize Quality of Service (QoS) and perform dynamic scheduling for radar systems with multiple resource constraints. A real-time dwell scheduling algorithm for a multifunction phased array radar using scheduling gain is presented by Ting et al [4]. It uses a combination of heuristics and scheduling gain. All of these papers consider a single radar unit capable of performing multiple functions simultaneously. In previous work [5], a method of building Dynamic Finite State Machines and solving the optimal policy using a Dynamic Programming method was presented. In this paper we provide a radar model based on a hierarchical network of Finite State Machines. The principal idea of the model is that a phased array radar is an antenna that can transmit waveforms in any direction. The inputs to the model are radar pointing orientations and the amount of energy allocated to each orientation in an interval of time. The outputs are the parameters of the threats including position, velocity and heading. This radar model is subject to a scheduling control algorithm that, under power constraints, is capable of providing specific schedules for the radar, as required by the external environment. D. Original Contributions The original contributions of this paper are as follows: 1. A radar model that uses F inite State Machines: The use of abstract models of computation such as finite state machines makes the model amenable to studying radar configurations for fleets. 2. A control algorithm that provides a greedy policy for operation: The control algorithm uses a step-by-step maximization algorithm applied to the finite state machine of the radar model. The output of the controller is a decision for radar orientation control and energy consumption at a particular time.