Point-ConNet: Integrated Power Allocation and Target Assignment for Efficient Multi-Target Tracking in Distributed Radar Networks

Abstract

The technique of using radar or other sensing devices to simultaneously monitor and track the positions of several moving objects is known as multiple target tracking or MTT. The challenge is to efficiently follow many targets in distributed radar networks by optimizing combined power allocation, resource allocation, and radar target assignment. This publication proposes an efficient deep-learning strategy for Joint Power Resource Allocation and Radar Target Assignment (JPRA-RTA) in distributed radar networks with ground-based transmitters and aerial receivers. The method has two phases: (i) Point-wise Activations Steerable Convolutional Neural Networks (Point-ConNet) with Narwhal Optimizer for joint power allocation and radar target assignment, and (ii) Hybrid Snow Geese Alpine Skiing Optimization (Hyb-SGASO), for resource allocation. First, the JPRA-RTA problem is converted into a regression problem solvable by Point-ConNet with the Narwhal Optimizer, enhancing computational efficiency, accuracy, fast convergence, and scalability. After optimizing power and radar target assignment, Hyb-SGASO is used to optimize remaining resources like communication bandwidth, ensuring balanced and efficient resource use. Thus, the proposed Point-ConNet- Hyb-SGASO method is implemented in Python and the method's performance is evaluated using metrics such as power consumption, spectral capacity, spectral efficiency, energy efficiency, tracking frame accuracy, cumulative distribution function (CDF), Mean Square Error (MSE), and Root Mean Square Error (RMSE), demonstrating significant improvements over traditional approaches. Thus, the proposed Point-Connet-Hyb-SGASO approach has achieved 18.76%, 23.04%, 28.06%, and 17.67% lower NMSE, 33.78%, 31.09%, 28.76%, and 24.89% higher energy efficiency compared with other conventional approaches like SDP-LHS-IPSOTS, ITPRS-LADMM, BCRLB-PSO, and SCA-RWO methods respectively.