REPOSITORY > RESULTS

Abstract

Characterization of the indoor radio environment (RE) is a prerequisite for advances in the
design and optimization of next-generation indoor wireless networks and for the construction
of a digital twin of the building. The need for comprehensive and accurate indoor
characterization will be evident in the future hyper-connected mixed real-virtual world,
where emerging applications in various areas such as wireless communications, spatial understanding,
localization, automation, mediated reality, etc., will be based on environmental
awareness. Novel, parsimonious, and intelligent environment awareness methodologies
that do not rely on specialized infrastructure and manual intervention are needed.
The dissertation addresses the problem of indoor environment characterization using
state-of-the-art wireless technologies and machine learning (ML) approaches. A novel
methodology for identifying indoor RE properties based on channel state information (CSI)
using ML approaches is proposed, formalized, and evaluated. The methodology is based
on two assumptions: the received signal conveys a RE signature, and the RE signature can
be estimated by analysing the wireless link.
The procedure for constructing the RE identification model from RE signatures is
streamlined into a framework. The framework specifies the RE signature, RE signature
acquisition, feature selection from CSI, CSI processing and storage, ML task, and ML
workflow. A large data set of CSI data acquired using ultra wideband (UWB) technology
in the microwave frequency band and annotated with environmental properties is built.
The data set contains channel impulse response (CIR) from a large number of rooms with
different sizes and materials for the surfaces and different positions of radio nodes.
The experiments presented in the dissertation provide an evaluation of the proposed
methodology for identifying surface materials from CIR, acquired with ray tracing method
in plain indoor environments and a comparative analysis along three main aspects: room
size, CIR acquisition strategy, and learning method. The ability of the models to generalize
to CIR acquisition strategies and room sizes not considered in the training process is
evaluated. The results show that the methodology can be applied to identify the material
of a single wall as well as the material of all surfaces in plain indoor environments. The
impact of radio nodes’ position and room size on the model performance is also confirmed.
The methodology is one of the main contributions of the dissertation. The material
identification presented in the dissertation is an initial example of indoor RE using the
proposed methodology. The methodology can be considered separately or as part of a
larger methodology to create an accurate and detailed digital twin of the building. One
of its main values lies in its extensibility to different CSI properties, CSI estimation methods,
indoor environments, and indoor characterization tasks. The methodology provides
the foundation (or environmental context) needed to develop state-of-the-art methods for
environmentally aware indoor wireless communications. Its importance is emphasized in
the era of next-generation communications, where a detailed description of the propagation
environment is a prerequisite for improving communications performance to meet the
requirements of emerging applications.