Features Relevant to Short-Term Wireless Channel Utilization Prediction

  • Author / Creator
    Kazemian, Sepehr
  • Wireless channel utilization prediction is useful in a number of applications, such as the recently proposed modalities of LTE networks allowing them to use unlicensed bands (LTE-U), otherwise used by Wi-Fi devices. Wireless utilization, as we are also able to also confirm, exhibits non-stationary behavior. The presented research provides an overall prediction strategy that can be implemented at the network edge. While the legacy view of "busy" hours vs. "non-busy" hours is still relevant, we approach the modeling with a finer definition for this busy/non-busy distinction. We split the utilization time series into intervals, each of them approximated as a stationary process modeled as a Markov chain. Each of those micro-models captures the short-term behavior and is characterized by its steady state distribution. The steady state distributions are used to define similarity among intervals in terms of their short-term behavior, i.e., the micro-models become a "library" of prior behaviors. We use a shallow neural network that combines features that express the similarity to a set of prior intervals, together with features arising from the time series using an auto-regressive model following the Box-Jenkins method, alongside features capturing straightforward step-to-step (lag one) transitions. The shallow network allow us to interpret the relative importance of the various features. It allows us to glean from the weights assigned why naive models (predicting next what has just been observed) are quite potent, and especially, and unsurprisingly, for non-busy hours. Moreover, we evaluate our prediction setup over "coarsened" utilization ranges since, for most applications, the granularity of prediction need not be fine as long as it describes distinct utilization regimes (e.g., "idle", "lightly loaded", "moderately loaded", "heavily loaded", "(almost) saturated"). The evaluation is carried out by predicting the utilization of Wi-Fi channels. Specifically for Wi-Fi channels, the architecture of our prediction platform exploits the utilization self-reporting performed by Access Points in Beacon frames. An extensive data collection experiment was designed and carried out, forming the real world data over which our prediction scheme is evaluated.

  • Subjects / Keywords
  • Graduation date
    Fall 2020
  • Type of Item
  • Degree
    Master of Science
  • DOI
  • License
    Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.