Environment Mapping Using Wireless Channel State Information and Deep Learning

Abstract

Situational awareness is becoming ubiquitous with modern communication technologies. Traditionally, user requirements for location based services have evolved from being driven only by government legislation for improved emergency response, to that of social and market based pressures including social-networking, geo-marketing and augmented reality. Not only do these applications necessitate improved localisation, they must also include awareness of the surrounding environment. Furthermore, recent years have seen a marked explosion in Artificial Intelligence (AI), machine learning and deep learning. Such tools enable inference to be drawn from complex data when traditional mathematical techniques cannot. This thesis explores the use of AI and wireless Channel State Information (CSI) for obtaining situational awareness through environment mapping. Several deep learning algorithms are developed that permit estimation of interior room dimensions using wireless Channel Impulse Responses (CIRs). Moreover the physical effects that bandwidth and received signal power incur on a CIR, and hence on room dimension prediction, is investigated yielding results with greater than 90% accuracy. This demonstrates that existing consumer wireless systems with similar physical constraints, should be capable of accurately estimating room dimensions given multiple CIR measurements from multiple receiver locations.