Date of Award

12-2019

Degree Name

Doctor of Philosophy

Department

Electrical and Computer Engineering

First Advisor

Dr. Ikhlas M. Abdel-Qader

Second Advisor

Dr. Janos Grantner

Third Advisor

Dr. Azim Houshyar

Keywords

Multiple SSID Framework, RSS fingerprint, indoor positioning systems

Abstract

Location-Based Indoor positioning systems significance stems from the bloom of recent applications in various fields such as in tracking services for an elder or a patient within large living communities, mobile robot localization, and several other security applications. Currently, Global Positioning Systems (GPS) are the most widely used location-sensing technique. However, satellite-based GPS signals require line of sight (LOS) to work correctly, which is something cannot be achieved inside buildings. Fortunately, wireless LAN can be employed in indoor positioning systems (IPS), and since all large buildings such as malls, hospitals, airports, schools, and museums have hundreds of Wi-Fi access points, it can provide accurate IPS without any additional infrastructure. Of special significance, the Wi-Fi fingerprinting-based techniques that offer a much less complex when compared to other methods such as the angle of arrival (AOA) and time difference of arrival (TDOA). Wi-Fi fingerprinting-based techniques use the received signal strength (RSS) to build radio maps. However, RSS value is a function of the distance between the Mobile System (MS) and Access Point (AP), which varies due to the multipath propagation phenomenon and human body blockage. Furthermore, fingerprinting approaches have several disadvantages such as labor, computational cost, and diversity (in signals and environment). In this dissertation, a novel approach that uses Multiple Service Set Identifiers (MSSID) to tackle these challenges is presented. MSSID means each AP can be configured to transmit N signals instead of one signal, to serve different clients’ categories simultaneously. IPS MSSID-based framework using three different realizations is proposed, implemented, and verified inside the College of Engineering and Applied Sciences (CEAS) building at Western Michigan University.

First, a MSSID Probabilistic Neural Network (PNN)-based multi-classifier is proposed with a spatial voting scheme as a tool to determine the location of the user. Spatial voting is designed specifically to tackle the negative impact of multi-path propagation. The performance of the proposed system compared to some of the conventional methods such as K-Nearest Neighbors (K-NN) and multi-class support vector machine (SVM). Experimental results show that spatial voting of three PNN classifiers can significantly mitigate the adverse effects of RSS variation. The precision of the proposed system for PNN and K-NN at distance error of 2m is 90% and 85%, respectively. As a comparison to the proposed systems, the precision of the traditional techniques for PNN and K-NN is 82% and 78%, respectively. In addition, the experimental results show that the average distance error for PNN-based proposed system is less than 0.73 m when the length of AP (L) is 18. Furthermore, the distance error of proposed system shows high stability where it has lowest standard deviation as compared with other traditional techniques.

Second, an MSSID-based adaptive K-nearest neighbor (K-NN) is proposed to tackle the challenges associated with static K-NN based-systems. The K-nearest neighbors (KNN) is selected for its significant performance with ease of realization. However, the static nature of K-NN, that is, in using a constant number of the nearest neighbors, leads to a serious shortcoming in its accuracy. In addition, the nature of the RSS-IPS challenges such as fading due to the multipath of electromagnetic waves inside buildings would mislead the solution of nearest neighbors. These reasons often result in lower perform than expected because of the increase in the distant neighbors’ biasing error. In this part, we address these challenges by proposing a new method based on multiple services set identifiers (MSSID) to select adaptively the appropriate nearest neighbors, and reject undesired ones. The ensemble technique is utilized to enhance the performance by combining the outputs of three adaptive K-NN estimators. The experimental results demonstrate the superiority of the adaptive K-NN based-proposed system over static K-NN. The results show that the precision of the proposed system for the adaptive K-NN at distance error of 2m is 73%, and the average distance error is less than 1.3 m. As a comparison to the proposed systems, the precision of the traditional K-NN at distance error of 2m is 61%, and the average distance error is 1.85 m.

Third, an MSSID- based particle swarm optimization (PSO) system is proposed. PSO technique is designed to select the most informative APs at each clustered area and combined with the K-means clustering method to confine location of the user into a smaller area and thus enhance positioning accuracy. WLAN-fingerprint based methods require recording RSS data of the surrounding APs, which results in including much more than the needed number of APs. Therefore, eliminating redundant or non-informative APs not only reduces the computational cost but also improves performance accuracy. At each cluster, PSO is applied to select the best joint combination of APs decided by the minimum mean of distance error. The results show that the proposed system outperforms other commonly proposed selection methods such as random, strongest APs, and Fisher criterion. Moreover, with reduction of 68% AP vector’s length (L=11), the results report that the proposed system achieves a positioning accuracy of 0.85 m over 3000 m2, with an accumulative density function (CDF) of 88% with a distance error of 2 m.

The use of the multiple SSID technique supports IPS classifiers and produces higher precision than with single SSID. The proposed algorithms show a notable improvement over its counterpart with single SSID along with the distance error and reduction of RSS-vector’s length.

Access Setting

Dissertation-Open Access

Download

Share

COinS