Applications of information channels to physics-informed neural networks for WiFi signal propagation simulation at the edge of the industrial internet of things

摘要

The ubiquitous presence of data driven technologies that move information from the edge of the Industrial Internet of Things (IIoT) to the cloud for advanced computation and back to the edge for action are pushing wireless connections to the limit. Under these conditions optimizing WIFI Received Signal Strength Intensity (RSSI) can improve data management, computational workflows, and geolocation accuracy while reducing energy consumption in order to minimize charging and computational resource requirements at the edge. Ensuring connectivity for these mission critical processes will require detailed knowledge (either measured or simulated) of the state of the electromagnetic fields in advanced manufacturing scenarios. Simulation has the advantage of developing more scalable solutions to this characterization problem but comes at a very high computational cost that may not be possible on edge devices with limited computational resources. In order to reduce the time and resource cost of achieving real time simulations with low computing specification edge devices, we propose creating a novel method that exploits the notion of information channels to create efficient Convolutional Neural Networks (CNNs) capable of determining the RSSI given a completely new geometry (never used in training) where objects or obstacles (walls, machines, tables, etc.) and their respective location, size and reflectivity indices, along with the antenna location are completely random. (c) 2021 Elsevier B.V. All rights reserved.