Machine learning based cognitive radar resource management

A modern radar may be designed to perform multiple functions, such as surveillance, tracking, and fire control. Each function requires the radar to execute a number of transmit-receive tasks. This raises the problem of assigning radar resources, such as the time, frequency and energy budget, to different tasks. Specifically, a radar resource management (RRM) module makes decisions on parameter selection, prioritization, and scheduling of such tasks. RRM becomes especially challenging in overload situations, where some tasks may need to be delayed or even dropped. Such task scheduling is an NP-hard problem. Furthermore, with multichannel, e.g., multi-frequency, radars becoming increasingly viable, multiple tasks can be executed simultaneously. While this greatly enhances the ability to execute tasks, it also complicates task scheduling, now on multiple timelines. In previous work, we had developed the branch-and-bound (B&B) method to solve the NP-hard problem, an approach with exponential computational complexity. In this work, we use the results of the B&B method to train a machine-learning based scheduler. Essentially, we propose to speed up the B&B method by estimating the value of the nodes of the search tree using a neural network. Our results show that the use of neural networks in conjunction with the B&B method results in a close-to-optimal solution while significantly reducing the computational complexity.