Statistical Analysis to Support CSI-Based Sensing Methods

TL;DR

This paper provides an in-depth statistical analysis of Wi-Fi Channel State Information (CSI) to enhance ambient sensing techniques, supporting environment classification and movement recognition without solely relying on machine learning.

Contribution

It advances CSI analysis by characterizing its statistical properties, offering a mathematical foundation for environment and movement recognition in wireless sensing.

Findings

CSI traces exhibit distinct statistical behaviors in controlled environments

The analysis supports the development of dedicated algorithms for sensing tasks

Provides insights that could improve environment classification accuracy

Abstract

Building upon the foundational work of the Bachelor's Degree Thesis titled "Analysis and Characterization of Wi-Fi Channel State Information'', this thesis significantly advances the research by conducting an in-depth analysis of CSIs, offering new insights that extend well beyond the original study. The goal of this work is to broaden the mathematical and statistical representation of a wireless channel through the study of CSI behavior and evolution over time and frequency. CSI provides a high-level description of the behavior of a signal propagating from a transmitter to a receiver, thereby representing the structure of the environment where the signal propagates. This knowledge can be used to perform ambvient sensing, a technique that extracts relevant information about the surroundings of the receiver from the properties of the received signal, which are affected by interactions with the surfaces of the objects within the analyzed environment. Ambient sensing already plays an essential role in new wireless networks such as 5G and Beyond 5G; its use Joint Communication and Sensing applications and for the optimization of signal propagation through beamforming could also enable the implementation of efficient cooperative ambient sensing in vehicular networks, facilitating Cooperative Perception and, consequently, increasing road safety. Due to the lack of research on CSI characterization, this study aims to begin analyzing the structure of CSI traces collected in a controlled experimental environment and to describe their statistical properties. The results of such characterization could provide mathematical support for environment classification and movement recognition tasks that are currently performed only through Machine Learning techniques, introducing instead efficient, dedicated algorithms.