Approximately Optimal Radar Resource Management for Multi-Sensor Multi-Target Tracking

Radar Resource Management in a multi-sensor multi-target scenario is considered. A dynamic resource balancing algorithm is proposed which optimizes target task parameters assuming an underlying partially observable Markov decision process (POMDP). By applying stochastic optimization methods, such as policy rollout, the POMDP is solved non-myopically. The approximately optimal approach is formulated assuming a central processor. Subsequently, a distributed implementation is introduced that converges to the same results as given by the centralized implementation and requires less computational resources. The performance of the proposed approach for both centralized and distributed implementation is demonstrated through dynamic tracking scenarios.

I. Introduction

Recent developments in multi-function radars (MFR) led to an increasingly flexible usage of these systems. Phased array antennas, digital beamforming, and waveform agility led to new degrees of freedom. As a consequence, modern MFR systems are capable of adjusting automatically to new operational scenarios and tasks during runtime. The control of such automatic adaptation is often called radar resource management (RRM) and is often considered within the framework of cognitive radar (e.g. [1], [2]). By combining the measurements of multiple connected radar sensors placed at different locations, the accuracy, resolution and coverage of the radar system can be improved. An overview of sensor and data fusion algorithms can be found in [3], [4]. Possible applications of such a sensor fusion are for instance weather or space observation, where the information of multiple separate sensors is combined for more accurate results. Instead of optimizing each sensor individually, it would be beneficial to optimize the resources jointly as only this would lead to a globally optimal solution. Possible applications for such an RRM approach can be found, e.g., in automotive scenarios, as well as maritime or air surveillance scenarios. This paper is based on the results of the master’s thesis in [5] and focuses on developing an approximately optimal solution approach for a multi-target tracking scenario assuming a sensor network. Although we illustrate our approach in a radar scenario, it is generally applicable to other kinds of adaptive sensors, as long as a limited resource has to be allocated.