Model- and Deep Learning-Based Bandwidth and Carrier Frequency Allocation in Distributed Radar Networks

Optimum allocation of bandwidth and carrier frequency in a network of distributed radar nodes is an important non-trivial research problem. In this paper, we propose both model- and deep learning-based joint bandwidth and carrier frequency allocation algorithms for a network consisting of a central coordinator and distributed radar nodes, each operating in a monostatic mode. With an objective of enabling poor performing radar nodes, that observe low target signal-to-noise-interference ratio (SINR) values, benefit from distributed collaboration, we propose model-based max-min approach, in which we maximize the minimum of the SINRs observed by all nodes, under total bandwidth and individual node's range resolution (RR) constraints. This optimization is non-convex, but we solve it efficiently utilizing an explicit relationship between bandwidth and carrier frequencies, and the fact that each node's SINR is a monotonically decreasing function of bandwidth and carrier frequency allocated to the node. We propose two iterative optimization methods that employ successive convex approximation with a) semidefinite programming (SDP) and b) geometric programming (GP) problem formulations. Computer simulations show the performance of the proposed methods under different RR requirements, which significantly outperform the equal bandwidth allocation (EBWA) method and enable poor performing nodes to enhance their individual SINRs significantly. The solutions of this model-based optimization and target locations are then used, respectively, as labels and input, to train a bidirectional long short-term memory (LSTM) network. The trained network can significantly reduce the online run-time complexity of the bandwidth and carrier frequency allocation in distributed radar networks.