Graph Representation Learning and Optimization for Spherical Emission Source Microscopy System

doi:10.1109/TASE.2024.3375024

Graph Representation Learning and Optimization for Spherical Emission Source Microscopy System

2025-01-01

6

PRE

AI

DOI：10.1109/TASE.2024.3375024

原文链接

原文求助

摘要

En 中文

Emission source microscopy (ESM) technique can be utilized for the localization of electromagnetic interference (EMI) sources in electronic systems, its performance greatly depends on the scanner accuracy and back-propagation method. In this paper, we introduce a novel spherical ESM system driven by 6-DOF manipulator, and investigate the back-propagation based on sphere wave expansion and robot control strategy. For spherical scanning aperture, we fuse the robot kinematics model and measurement constraints, and propose solving the optimal scanning grid with nonlinear programming method. For manipulator control, we present a graph-based learning framework (Gash-LKH) that combines sparse graph neural network with Lin-Kernighan heuristic (LKH) solver. This framework adopts the gated single-head attention module and parallel sparse graph feature abstracting channels, it can produce high-qualified edge candidate set that help subsequent LKH solver generate optimal scanning path with lower memory cost and less computation. Extensive experiments are conducted to validate the performance of Gash-LKH and Spherical ESM system, the results have demonstrated the feasibility and superiority of spherical ESM system in providing accurate microscopy and localization in EMI measurement. Note to Practitioners-The motivation of this paper is to develop an automated ESM system that realizes the spherical aperture scanning and pattern reconstruction of radiation source. Since adoption the back-propagation method based on sphere wave expansion, this system is supposed to achieve better microscopy performance and lower truncation error than other scanner. In this paper, we employ 6-DOF manipulator as scanner, and propose a complete spherical aperture generation method that produces the discrete and even scanning grid based on any source, frequency band and measurement constraints. Furthermore, we propose an end-to-end learning framework, Gash-LKH, to solve the optimal scanning path for given scanning aperture. The achieved accuracy and time-consumption of Gash-LKH is satisfactory for solving large-scaled and high-density scanning path planning. Note that the entire framework can be trained through random 3D-TSP instance dataset, and can be transferred to handle various radiation sources operating in microwave band. We have demonstrated the feasibility of proposed methodology and system through the experiments using Elfin-5 manipulator and benchmark sources. The results have offered the possibilities of achieving satisfactory localization and characterization in EMI measurement.

Keyword:

Manipulator control

emission source microscopy

spherical synthetic scanning

sparse graph neural network

traveling salesman problem

Graph Representation Learning and Optimization for Spherical Emission Source Microscopy System

摘要

期刊

机构

引用论文