Dissertation/Thesis Abstract

The author has requested that access to this graduate work be delayed until 2029-04-12. After this date, this graduate work will be available on an open access basis.
WLAN Access Points RSSI-Fingerprinting and Compressive Sensing for Indoor Positioning Systems
by Al-Moukhles, Hussein Nasser Wazeer, Ph.D., Western Michigan University, 2018, 131; 13877013
Abstract (Summary)

Efforts to improve the performance accuracy of Indoor Positioning Systems (IPS) have been increasing substantially and it has been a very fierce competition among scientific and enterprise entities. This dissertation focuses on using existing wireless network infrastructure and addresses the utilization of advanced mathematical algorithms to achieve the sought higher positioning accuracy. The ability to formulate the positioning problem as a sparse system furnishes the impetus for investigating the promising Compressive Sensing (CS) theory for IPS. Therefore, this dissertation aims to design a Received Signal Strength Indicator (RSSI) Fingerprinting-based IPS while utilizing existing Wireless Local-Area Networks (WLANs) within a CS framework. The positioning accuracy of the proposed IPS-framework has been improved by employing CS-based classification and hybrid cluster-matching techniques.

A WLAN-APs selection scheme based on Probability of Detection (POD) that is, in turn, based on the AP frequency of detection at the different RPs, was proposed. The impact of POD, the Time Variance, and the Random APs selection schemes were verified through a comparison between positioning accuracy of CS-based IPS and that of k-NN-based IPSs. Experimental results show that the Random and POD schemes outperform the other two approaches when the CS- and k-NN-based frameworks are used, respectively. Multiple-Fingerprints (MFPs) technique, whereby each RP is represented by a set of RSSI-vectors forming multiple fingerprints was also proposed. MFPs are generated from the collected RSSI time-samples by means of random combinations and FPs partitioning approaches. The performance of CS-, Probabilistic Neural Networks (PNN)-, and k-NN-based IPSs are evaluated by using the proposed MFPs technique. Results show that the positioning performance of IPS with the MFPs technique outperforms that for the typical single FP-based systems.

IPS based on WLAN-RSSI Fingerprinting and CS-based classification was achieved using Fuzzy C-Means (FCM) clustering and a FCM-PNN hybrid cluster-matching approach based on the U-membership matrix of the FCM clustering approach and PNN. To evaluate the performance of the proposed framework, a comparison with k-NN-, PNN-, and typical CS-based IPSs approaches was conducted. The performance evaluation for the proposed IPS techniques has been based on using 1776 generated RSSI-vectors from within the building of the College of Engineering and Applied Sciences at Western Michigan University. This testing Radio Map (RM) has been constructed from the RSSI time-samples of both RPs and TPs by random combinations. The figure-of-merit has been selected as the Euclidian distance among the actual and estimated positions. Positioning performance versus the number of selected Aps was also investigated using the Root Mean Square Error (RMSE). The experimental results show that the proposed IPS framework outperforms k-NN-, PNN-, and typical CS-based ones with accuracy of 0.8 m at 36 APs. For the same number of selected APs, the accuracies for k-NN-, PNN-, and typical CS-based frameworks are 2.06 m, 1.47 m, and 1.12 m, respectively.

Indexing (document details)
Advisor: Abdel Qader, Ikhlas
Commitee: Abdel-Qader, Ikhlas, Grantner, Janos, Houshyer, Azim
School: Western Michigan University
Department: Engineering and Applied Science
School Location: United States -- Michigan
Source: DAI-B 80/08(E), Dissertation Abstracts International
Source Type: DISSERTATION
Subjects: Engineering, Electrical engineering
Keywords: Cluster matching, Compressive sensing, FCM custering, Fingerprinting, Indoor positioning, WLAN RSSI
Publication Number: 13877013
ISBN: 9781392054536
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest