Investigating performance improvements in wireless networks using probabilistic graphical models

Date
2016-12
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: Probabilistic Graphical Models (PGMs) have proven to be a powerful and effective tool for predicting the behaviour of probabilistic systems. Their applicability for improving the performance of wireless networks, where most strategies are probabilistically founded, is therefore worth exploring. PGMs can infer states and conditions within the network and allow protocols to act accordingly. However, as this implies decision-making under uncertainty, investigating the application of PGMs for this purpose would have merit. In this work, we create an effective method for making decisions under uncertainty by expanding the current theory of strong junction trees to allow for loopy decision cluster graphs. However, similarly to the behaviour of loopy cluster graphs, this method also leads to imprecise probabilities and utilities, and sub-optimal decision strategies. We created 3 PGM-augmented Round Robin Medium Access Control (MAC) protocols by using different PGMs to determine which slave node the master node should poll next. This resulted in reduced latency for packets during unequal traffic loads. Furthermore, we created a PGM-augmented Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) MAC protocol by using a PGM in order to estimate the number of contending nodes and allowing the node to change the length of its contention window accordingly. This resulted in an efficient and fair network protocol irrespective of the number of nodes in the network.
AFRIKAANSE OPSOMMING: Probabilistiese Grafiese Modelle (PGM’e) het reeds bewys dat dit ‘n kragtige en doeltreffende manier bied om die gedrag van probabilistiese stelsels te voorspel. Dit blyk dus aantreklik om hul toepassing in radionetwerke, waar strategië probabilisites van aard is, te ondersoek. PGM’e kan die toestande en omstandighede van die netwerk afskat en protokolle kan daarvolgens optree. Aangesien dit egter impliseer dat daar besluite tydens onsekerheid geneem moet word, is die ondersoek om PGM’e te gebruik vir hierdie doel ook van belang. In hierdie werk skep ons ‘n effektiewe metode om besluite tydens onsekerheid te neem deur die huidige teorie van sterk aansluitingsbome uit te brei om beslissingkluster- grafieke met lusse moontlik te maak. Soortgelyk aan kluster-grafieke, lewer hierdie metode egter onakkurate waarskynlikhede, nut-waardes en sub-optimale beslissing-strategië. Ons het 3 PGM-uitgebreide ‘Round Robin’ medium toegangsbeheer-protokolle geskep deur verskillende PGM’e te gebruik om te bepaal watter slaaf-node die meester-node volgende moet ondervra. Hierdie lei tot verkorte transmissievertragings van pakkies tydens ongelyke netwerkladings. Verder het ons ‘n PGM-uitgebreide ‘CSMA/CA’ medium toegangsbeheer-protokol geskep deur ‘n PGM te gebruik om af te skat hoeveel nodusse aan die kontensie wil deelneem en die lengte van die kontensie-venster daarvolgens aan te pas. Hierdie lei tot ‘n meer doeltreffende en billike netwerk, ongeag die aantal nodusse in die netwerk.
Description
Thesis (MScEng)--Stellenbosch University, 2016.
Keywords
Graphical modeling (Statistics), Probabilistic database systems, Wireless personal area networks, UCTD
Citation