A comparison of ranging and localization techniques in indoor, urban, and tunnel environments

(1)

Institutionen för systemteknik

Department of Electrical Engineering

Examensarbete

A Comparison of ranging and localization

techniques in indoor, urban and tunnel

environments

Examensarbete utfört i Kommunikationssystem vid Tekniska högskolan i Linköping

av

Sasit Chuasomboon

LiTH-ISY-EX--13/4695--SE

Linköping 2013

Department of Electrical Engineering Linköpings tekniska högskola

Linköpings universitet Linköpings universitet

(2)

(3)

A Comparison of ranging and localization

techniques in indoor, urban and tunnel

environments

Examensarbete utfört i Kommunikationssystem

vid Tekniska högskolan i Linköping

av

Sasit Chuasomboon

LiTH-ISY-EX--13/4695--SE

Handledare: Vladimir Savic

isy,Linkopings Universitet

Examinator: Danyo Danev

isy, Linkopings Universitet

(4)

(5)

Avdelning, Institution

Division, Department

Division of Communication Systems Department of Electrical Engineering Linköpings universitet

SE-581 83 Linköping, Sweden

Datum Date 2013-06-14 Språk Language Svenska/Swedish Engelska/English Rapporttyp Report category Licentiatavhandling Examensarbete C-uppsats D-uppsats Övrig rapport

URL för elektronisk version http://www.commsys.isy.liu.se http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-ZZZZ ISBN — ISRN LiTH-ISY-EX--13/4695--SE

Serietitel och serienummer

Title of series, numbering

ISSN

—

Titel

Title

Svensk titel

A Comparison of ranging and localization techniques in indoor, urban and tunnel environments Författare Author Sasit Chuasomboon Sammanfattning Abstract

Localization in wireless network sensors is an attractive research area nowadays. It is widely used in many applications e.g., indoor/outdoor asset tracking, intrusion detection, search-and-rescue, road traffic monitoring, and water quality monitor-ing. An accuracy and robustness to noise are important issues for localization which is needed to study and research to find the best solution.

This thesis compares a ranging and localization techniques in indoor, urban and tunnel through a high performance ray-tracing simulator, Wireless InSite R_.

Rang-ing techniques are based on two standard distance related measurement schemes e.g., RSS and TOA. A linearized least squares technique with reference node se-lection approach is chosen to estimate unknown nodes positions. Indoor and ur-ban area are built-in floor plan and terrain available in simulator program, while tunnel is designed. In general, localization accuracy suffers from multipath and NLOS condition. This thesis also observes characteristic of them from ray-tracing method perspective. Firstly, important simulation parameters such as number of reflections/diffractions, types of waveform, and types of antenna are analyzed on each environments. Then, the models for distance estimation based on RSS and TOA measurements are created using measurements in simulated environments. The thesis proposes four scenarios for distance estimation model. They are line-of-sight (LOS), non-line-of-line-of-sight (NLOS), combination of LOS and NLOS, and NLOS with obstacle. All four scenarios models are derived along with model error dis-tribution to observe characteristic of noise due to multipath and NLOS condition. Finally, the localization using only LOS condition measurements, is tested on each environment and compared results in term of accuracy.

Nyckelord

Keywords Localization, Time of Arrival, Received Signal Strength, Distance estimation, Tun-nel, Indoor, Urban, Least square, Ray-tracing, Shooting and bouncing ray tracing method, Trilateration, Multilateration

(6)

(7)

Abstract

Localization in wireless network sensors is an attractive research area nowadays. It is widely used in many applications e.g., indoor/outdoor asset tracking, intrusion detection, search-and-rescue, road traffic monitoring, and water quality monitor-ing. An accuracy and robustness to noise are important issues for localization which is needed to study and research to find the best solution.

This thesis compares a ranging and localization techniques in indoor, urban and tunnel through a high performance ray-tracing simulator, Wireless InSite R_.

Rang-ing techniques are based on two standard distance related measurement schemes e.g., RSS and TOA. A linearized least squares technique with reference node se-lection approach is chosen to estimate unknown nodes positions. Indoor and ur-ban area are built-in floor plan and terrain available in simulator program, while tunnel is designed. In general, localization accuracy suffers from multipath and NLOS condition. This thesis also observes characteristic of them from ray-tracing method perspective. Firstly, important simulation parameters such as number of reflections/diffractions, types of waveform, and types of antenna are analyzed on each environments. Then, the models for distance estimation based on RSS and TOA measurements are created using measurements in simulated environments. The thesis proposes four scenarios for distance estimation model. They are line-of-sight (LOS), non-line-of-line-of-sight (NLOS), combination of LOS and NLOS, and NLOS with obstacle. All four scenarios models are derived along with model error dis-tribution to observe characteristic of noise due to multipath and NLOS condition. Finally, the localization using only LOS condition measurements, is tested on each environment and compared results in term of accuracy.

Sammanfattning

Svenskt abstract kan man placera här.

(8)

(9)

Acknowledgments

I firstly would like to show my sincerely thankful to Royal Thai Air Force who provides me a great opportunity and financial support for studying master degree in Sweden. And I would like to express my gratitude to former and present exam-iner, Dr. Karipidis Eleftherios and Dr. Danyo Danev, for their valuable guidances and many helpful suggestions. I must not forget to thank, Dr. Vladimir Savic, my supervisor, who helped me a lot from the beginning to the end of the thesis. His expertise in localization was the key which brought me to this complete version of the thesis.

Secondly, I would like to thank professors and PhD students in communication system for helping me went through all course works in the past two years. It was tough but good experience to be a master student here.

Thanks to all of my friends, my younger brother and sister for listening and cheering me up when I got into trouble and feel bad.

Finally, I would like dedicate all of my works to my parents whom I love the most in my life.

(10)

(11)

Abbreviations and Terms

RSS Received Signal Strength

TOA Time of Arrival

AOA Angle of Arrival

LOS Line of sight

NLOS Non line of sight

SBR Shooting and Boucing ray

Tx Transmitter

Rx Receiver

BS Base station

MS Mobile station

PDP Power Delay Profile

CIR Complex Impulse Response

AOA Angle of Arrival

CRLB Cramer-Rao Lower Bound

LSS Linearized Least Squares

LOS-OBS Line of sight with obstacle HF High frequency (3-30 MHz)

UHF Ultra-high frequency (0.3-3 GHz)

GTD Geometrical theory of diffraction

GO Geometrical optics

NTUST National Taiwan University of Science and Technology

(16)

2 Contents

List of Figures

2.1 An example of localization system . . . 11

2.2 Trilateration concept . . . 15

3.1 Spherical coordinate system [1] . . . 19

3.2 Top view of indoor area floor plan [2] . . . 20

3.3 3-Dimensions view of indoor area floor plan [2] . . . 21

3.4 Rosslyn city top view terrain [2] . . . 21

3.5 3-Dimensions view of Rosslyn city [2] . . . 22

3.6 Tunnel Floor plan . . . 22

3.7 Tunnel wall and ceiling sample (left), Underground ore mining tun-nel in Kiruna, Sweden (right) [3, 4] . . . 23

3.8 3 Dimensional illustration of designed tunnel . . . 23

4.1 Positions of Tx and 5 Rx routes of LOS scenario in indoor area . . 26

4.2 Received power (dBm) vs log-distance plot (left) and a distribution of multipath noise (right) from LOS scenario in indoor area . . . . 26

4.3 Positions of 2 Tx points and 8 Rx routes of NLOS in indoor area . 27 4.4 Received power (dBm) vs log-distance plot (left) and a distribution of multipath noise (right) from NLOS scenario in indoor area . . . 28

4.5 Received power (dBm) vs log-distance plot (left) and a distribution of multipath noise (right) from combination of LOS and NLOS scenario in indoor area . . . 29

4.6 Positions Tx point, Rx route and Obstacle in indoor area . . . 29

4.7 Received power vs log-distance plot and a distribution of error due to blockage and multipath noise (right) from NLOS with obstacle scenario in indoor area . . . 30

4.8 Positions of Tx and 5 Rx routes of LOS in urban area . . . 32

4.9 Received power (dBm) vs log-distance plot (left) and a distribution of multipath noise (right) from LOS scenario in urban area . . . . 33

4.10 Positions of Tx and 5 Rx routes of NLOS in urban area . . . 33

4.11 Received power (dBm) vs log-distance plot (left) and a distribution of multipath noise (right) from NLOS scenario in urban area . . . 34

4.12 Received power (dBm) vs log-distance plot (left) and a distribu-tion of multipath noise (right) from combinadistribu-tion of LOS/NLOS scenario in urban area . . . 35

4.13 Positions of Tx, Rx route and Obstacles in urban area . . . 36

4.14 Received power (dBm) vs log-distance plot (left) and a distribution of multipath noise (right) from LOS with obstacle scenario in urban area . . . 36

4.15 Positions of Tx, Rx routes of LOS in tunnel . . . 37

4.16 Received power (dBm) vs log-distance plot (left) and a distribution of multipath noise (right) from LOS scenario in tunnel . . . . 38

(17)

Contents 3

4.18 Received power (dBm) vs log-distance plot (left) and a distribution

of multipath noise (right) from NLOS scenario in tunnel . . . 39

4.19 Received power (dBm) vs log-distance plot (left) and a distribu-tion of multipath noise (right) from combinadistribu-tion of LOS/NLOS scenario in tunnel . . . 40

4.20 Positions of Tx, Rx routes of NLOS with obstacle in tunnel . . . . 40

4.21 Received power (dBm) vs log-distance plot (left) and a distribution of error (right) from NLOS with obstacle scenario in tunnel . . . . 41

5.1 True distance model vs measured samples (left) and distribution of error (right) from LOS scenario in indoor area . . . . 45

5.2 True distance model vs measured samples (left) and distribution of Nnlos (right) from NLOS scenario in indoor area . . . . 46

5.3 True distance model vs measured samples (left) and distribution of Nnlos (right) from combination of LOS/NLOS scenario in indoor area . . . 47

5.4 Measured samples vs true distance model and distribution of Nnlos with obstacle in indoor . . . 47

5.5 True distance model vs measured samples (left) and distribution of error (right) from LOS scenario in urban area . . . . 48

5.6 True distance model vs measured samples (left) and distribution of error (right) from NLOS scenario in urban area . . . 49

5.7 True distance model vs measured samples (left) and distribution of Nnlos(right) from combination of LOS/NLOS scenario in urban area . . . 50

5.8 True distance model vs measured samples (left) and distribution of error (right) from LOS with obstacle scenario in urban area . . 50

5.9 Tunnel Line of sight measured samples vs true distance model and histogram plots . . . 51

5.10 Tunnel Non line of sight measured samples vs true distance model and histogram plots . . . 52

5.11 Tunnel combination of LOS and NLOS measured samples vs true distance model and histogram plots . . . 52

5.12 Measured samples vs true distance model and distribution of Nnlos with obstacle in tunnel . . . 53

6.1 Indoor anchor nodes (Tx) layout and coordinates . . . 56

6.2 Indoor non-anchor nodes (Rx) . . . 56

6.3 Indoor localization via RSS measurement error plot . . . . 57

6.4 Urban anchor nodes (Tx) layout and coordinates . . . 58

6.5 Urban non-anchor nodes (Rx) . . . 58

6.6 Urban localization via RSS measurement error plot . . . 59

6.7 Tunnel anchor nodes (Tx) layout and coordinates . . . 60

6.8 Tunnel non-anchor nodes (Rx) layout and coordinates . . . 60

(18)

4 Contents

6.10 Position of non-anchor nodes vs distance error (m)(color) in indoor

area . . . 62

6.11 Position of non-anchor nodes vs distance error (m)(color) in urban area . . . 63

6.12 Position of non-anchor nodes vs distance error (m)(color) in tunnel 64 A.1 Position of Tx and Rx in Indoor area . . . 74

A.2 Comparison plot of number of reflection equal to 1,3,5 and 7 in indoor area . . . 74

A.3 Comparison plot of numbers of diffraction equal to 1, 2, 3 and 4 in indoor area . . . 75

A.4 Position of Tx and Rx route in urban area . . . 76

A.5 Comparison plot of numbers of reflection equal to 2, 4, 6 and 8 in urban area . . . 76

A.6 Comparison plot of numbers of diffraction equal to 1, 2, 3 and 4 in urban area . . . 77

A.7 Position of Tx and Rx in Tunnel . . . 77

A.8 Comparison plot of numbers of reflection equal to 1, 3, 5 and 7 in tunnel . . . 78

A.9 Comparison plot of numbers of reflection equal to 1, 2 and 3 in tunnel 79 A.10 Position of Tx point and circle of receivers in indoor area . . . 79

A.11 Comparison pathloss vs radiated angle plot of 5 types of antenna in indoor area . . . 80

A.12 Comparison plot of 5 types of antenna in urban area . . . 81

A.13 Position of Tx point and circle of receivers in tunnel . . . 81

A.14 Comparison plot of 7 types of antenna in tunnel . . . 82

A.15 Position of Tx point and Rx point in indoor area . . . 83

A.16 Comparison PDP plot of 6 waveform in indoor area . . . 83

A.17 Comparison CIR plot of 6 waveform in indoor area . . . 84

A.18 Position of Tx point and Rx point in urban area . . . 84

A.19 Comparison PDP plot of 6 waveform in urban area . . . 85

A.20 Comparison CIR plot of 6 waveform in indoor area . . . 85

A.21 Position of Tx point and Rx point in tunnel . . . 86

A.22 Comparison PDP plot of 6 waveform in tunnel . . . 86

A.23 Comparison CIR plot of 6 waveform in tunnel . . . 87

C.1 Root mean square error vs threshold plot (left) and Number of measurable non-anchor nodes vs threshold plot (right)of indoor area 92 C.2 Root mean square error vs threshold plot (left) and Number of measurable non-anchor nodes vs threshold plot (right) of urban area 93 C.3 Root mean square error vs threshold plot (left) and Number of measurable non-anchor nodes vs threshold plot (right) of tunnel . . 93 C.4 Root mean square error vs threshold plot (left) and Number of

measurable non-anchor nodes vs threshold plot (right)of indoor area 94 C.5 Root mean square error vs threshold plot (left) and Number of

(19)

Contents 5

C.6 Root mean square error vs threshold plot (left) and Number of measurable non-anchor nodes vs threshold plot (right)of indoor area 95

List of Tables

4.1 Indoor distance estimation model via RSS measurement results

summary . . . 31

4.2 Urban : Distance estimation model via RSS measurement results summary . . . 37

4.3 Tunnel : Distance estimation model via RSS measurement results summary . . . 42

5.1 Distance estimation model via TOA measurement results summary 48 5.2 Urban: Distance estimation model via TOA measurement results summary . . . 51

5.3 Tunnel: Distance estimation model via TOA measurements results summary . . . 53

6.1 Localization test results summary . . . 64

B.1 List of common parameters of RSS measurement in indoor area . . 89

B.2 List of common parameters of RSS measurement in urban area . . 90

(20)

(21)

Chapter 1

Introduction

This chapter provides the motivations behind this thesis. In addition, goal of the thesis, methodology, and thesis outline are briefly described.

1.1 Background and Motivation

If we have a discussion about a positioning or localization, most of the people usu-ally think about Global Positioning System (GPS). GPS could be a good method if we are in open wide outdoor area. However, GPS has limitation. If any mobile devices are located in an indoor area, GPS signals are unable to reach or very weak signal is received. Then it is not possible to have an accurate localization in the indoor area. Even in an outdoor area, in some situations such as the day that there are a lot of cloud in the sky which can degrade GPS signal board-casted from space or heavy raining day. To overcome a limitation of a core localization system such as GPS, wireless sensor networks or cellular network system are proposed to develop higher accuracy and availability of user location.

Moreover, nowadays localization techniques become very attractive research in-terest. These techniques provide opportunities for monitoring and controlling the environments by utilizing available sensors of existing network systems (wireless networks and cellular networks). There are many useful commercial applications of location techniques such as indoor/outdoor asset tracking [5–7] , intrusion de-tection [8–10], search-and-rescue [11,12], road traffic monitoring [13–15], and water quality monitoring [16, 17]. In a specific environment such as tunnel, there is also a research in localization/tracking in mine tunnels [18]. Moreover in military ap-plicants, e.g., battlefield surveillance [19] and target tracking [20], are also use a capability of sensor network for localization. In order to obtain high position-ing accuracy and robustness, we need to study different localization techniques in different environments to find the best solutions for our future needs and applica-tions.

In general, localization usually suffers from noises due to multipath propaga-tion and NLOS condipropaga-tion. Many researches apply mathematics model [21, 22], e.g., Gaussian, and exponential distribution, to specify these noises and test

(22)

8 Introduction

formance of localization techniques which may provide analysis results different from actual values. Moreover, on-site experiment may be the best option to have accurate actual data. However, the experiments usually consume a lot of time, and are typically very expensive due to equipment cost. And also human error could occur during measurement. To present another option for analyzing localization techniques more efficiently, we introduce ray-tracing simulation software, Wireless InSite R _{[23], which can simulate an electromagnetic wave propagation in building}

or terrain and provide useful raw data, e.g., RSS and TOA, for localization. We can also create our own specific environment and define type of materials containing in environment. By using raw data obtaining from simulation, we expect to discover new interesting results on performance of localization algorithms in more specific environment, e.g.,tunnel, which we cannot obtain from mathematical model.

1.2 Goal of the thesis

The main goal of this thesis is to compare localization algorithms (RSS and TOA) in different environments (indoor, urban, and tunnel). However, before we success the main goal, we derive distance estimation model from 4 scenarios which are LOS, NLOS, combination of LOS and NLOS, and NLOS with obstacle. Then we discuss and compare their results and behavior of RSS and TOA.

1.3 Methodology

In this thesis, we design and simulate different environments (indoor office, urban area and tunnels) by using a ray-tracing simulation software, Wireless InSite R _[23].

Then, the data is exported to MATLAB R_{, which is used to create a models for}

dis-tance estimation (using RSS/TOA measurements) [21, 24], and apply localization algorithms [21, 24, 25].

This thesis is divided into 4 phases as following :

Phase 1 : Understand state-of-the-art and learn how to use Wireless InSite R

Phase 2 : Design environments, find proper simulation parameters and obtain the measurements (RSS and TOA) in different environments (indoor office, urban area, and tunnel) using Wireless InSite R

Phase 3 : Create the models for distance estimation for the environments from phase 2 and consider 4 scenarios which are LOS, NLOS, combination of LOS and NLOS, and NLOS with obstacle.

Phase 4 : Localize targets in all environments from phase 2 using the model from phase 3 and localization techniques from state-of-the-art

1.4 Thesis Outline

In chapter 2, ranging and localization technique in wireless networks, mathematic models that lead to localization are briefly described. It explains how two linear models, e.g., distance estimation via RSS measurement and via TOA measurement,

(23)

1.4 Thesis Outline 9

are defined. Moreover, we introduce a localization algorithm used in this thesis, which is multilateration. And non-linear equations of the algorithm are solved by linearized least squares (LLS) method.

In chapter 3, a ray-tracing simulation program, Wireless InSite, is introduced. Important radio propagation theories related to e.g., multipath propagation and SBR method, are described. Materials and three environments design specifica-tions are also described in details.

Chapter 4 and 5 show the measurement results of RSS and TOA measurements in LOS, NLOS, combination of LOS and NLOS, and NLOS with obstacle. The results are the list of model coefficients derived by least squares fitting algorithm, a comparison plot of linear model and measurement samples, and a histogram plot of error distribution.

In chapter 6, we use the model parameters from chapter 4 and 5 to perform lo-calization in three environments. Then we compare lolo-calization accuracy between RSS and TOA based techniques.

(24)

(25)

Chapter 2

Ranging and Localization

Techniques in Wireless

Sensor Networks

Primarily, a purpose of wireless networks is to connect available network devices an the area and allow users to have voice and data communications through networks. Alternatively, wireless networks can be utilized for localization purpose as well, due to widespread distributed wireless nodes.

Figure 2.1. An example of localization system

In wireless sensor networks, nodes on known position called anchors are in-stalled at fixed positions with known coordinates. Unknown location nodes called

non-anchor are sensors that need to be known their locations [24]. This thesis

assumes that anchor nodes measure informations such as RSS and TOA from re-ceived signals transmitted from non-anchor nodes. These informations are sent to data fusion center as shown in Figure 2.1 and location of non-anchor nodes can be estimated by using proper localization algorithm. In following sections, measure-ment and localization techniques which are used in this thesis are described.

(26)

12 Ranging and Localization Techniques in Wireless Sensor Networks

2.1 Measurement techniques

In general measurement techniques can be classified into three categories [21, 26] : direction-based measurements, distance-based measurements and profiling tech-nique . Direction related measurement commonly refers to angle of arrival (AOA) technique. AOA related approach is described well in [27]. Profiling technique such as RSS fingerprint technique is the technique that the system priori con-structs a map of behavior of RSS in the coverage area. Later, based on this RSS map the location of unknown node is located. RADAR system [28] proposed by Bahl and Padmanabhan is an example of the system that operates by recording and processing signal strength information at available anchor nodes in the area to determine user location. And also T.Roos et al. propose a statistic signal power model based localization scheme in [29]. For distance related measurements, TOA, RSS, time-difference-of-arrival (TDOA) [30] are measurements that are commonly applied. This thesis mainly focuses on two standard distance-based measurements : TOA and RSS.

2.1.1 Distance-based RSS measurement

This technique utilizes a received signal strength indicator (RSSI) which is a stan-dard feature in common wireless network devices [31]. It is a low cost technique because it requires no additional hardware such as sensors which means that this technique could have lower cost and complexity than others in system implemen-tation [24].

In free space, a received power, Pr(d) when d is distance between Tx and Rx,

is attenuated relating to Friis equation [32].

Pr(d) =

PtGtGrλ2

4π2_d2 (2.1)

where Pt is transmitted power (Watt). Gt and Gr are transmitter and receiver

antenna gain, respectively. λ is transmitted signal wavelength.

However, eq.(2.1) is considered too optimistic. In reality, we need to consider effects of multipath propagation which is main problem in RSS measurement. Different signal amplitudes and phases arriving at receiver cause additive or de-structive function of frequency, called frequency-selective fading. Nevertheless, this fading could be avoided by transmitting signals in wide frequency band which significantly larger than coherence bandwidth of the channel. Then, measured received signal power in wide band is equivalent to sum of all multi-paths arriving at the receiver [24]. With this assumption and eq.(2.1), average received signal power ¯Pr(d) [dBm] can be modeled as following equation.

¯ Pr(d) = P0− 10nplog10 d d0 + X (2.2)

where P0[dBm] is known reference power at reference distance d0. npis path loss

(27)

2.1 Measurement techniques 13

which can be derived from difference between measures received signal and average received signal.

From eq.(2.2) a maximum likelihood estimation of distance between transmit-ter i and receiver j, ˆdi,j , can be written as

ˆ

di,j= d0· 10 −Pi,j −P0_10np

(2.3) where Pi,jis a measured received power [dBm] between transmitter i and receiver

j.

2.1.2 Distance-based TOA measurement

In general, TOA can be categorized into one-way and round-trip propagation time based measurement. According to simulator, we select a former approach. This thesis defines that TOA is a measurement of time when the first path of transmit-ted signal arrived at a receiver. TOA can be model as transmission time plus a time delay induced by noise. Therefore, TOA model can be written as following eq.(2.4).

τi,j=

( _d i,j

c + Nlos for LOS di,j

c + Nnlos for NLOS

(2.4)

where di,j is a distance between transmitter i and receiver j, c is a speed of light

3 × 108_{m/s, N}

los is LOS noise. Nnlosis a NLOS noise. However, other bias caused

by non-synchronization, limited bandwidth and processing time equipment, are not considered in this thesis.

Then, an distance estimation can be found from ˆ

di,j = cτi,j (2.5)

An error of distance estimation model is simply defined as,

e = ˆdi,j− di,j (2.6)

Then, mean value and standard deviation of e are

µe= 1 K K X i=1 ei (2.7) σe= v u u t 1 K K X i=1 (ei− µe)2 (2.8)

(28)

14 Ranging and Localization Techniques in Wireless Sensor Networks

2.2 Localization techniques

After estimated distance between anchor and non-anchor node is calculated from both RSS and TOA measurements, we apply a multilateration method to find a non-anchor node position. The multilateration method requires at least three fixed position nodes, and distances between them and an unknown position node. We firstly approach multilateration from trilateration. According to trilatera-tion, when there is no noise and NLOS bias, all circles intersect at the same point. However, because of existence of noise and NLOS bias in practice, all circles do not intersect at the same point. These equations become non-linear. We need a linearization technique to solve this problem.

For LOS scenario, Ismail et al. [22] propose five different non-linear and lin-earized least-squares (LLS) techniques to solve this non-linear problem. By com-paring to Cramer-Rao lower bound (CRLB), the LLS with reference node selection and LLS utilizing the covariance matrix [33] have the best performance among five candidates. However, the thesis use the former LLS because in practice it is difficult to have an exact variance from measurement and considering covariance matrix in LLS is too optimistic estimation for LOS scenario in TOA measurement. Let (xj, yj) donates non-anchor node j coordinate in a Cartesian coordinate

system. Let (xi, yi) donates coordinate of anchor node i. In this thesis z-axis

is ignored for simplicity. In trilateration, we have three anchor nodes and one unknown node. The distance between anchor and non-anchor nodes are estimated based on RSS or TOA measurement. LLS with reference node selection algorithm begins with choosing an anchor node that has smallest distance to non-anchor. This node is defined as the first anchor node (x1, y1) and set as an origin of coordinate system. The remaining two nodes are also chosen in ascending order. Figure 2.2 shows how to manage reference node selection algorithm. Then, we can estimate (xj, yj) by solving eq.(2.9).

d2_1,j = x2_j+ y2_j (2.9a)

d2_2,j = (x2− xj)2+ (y2− yj)2 (2.9b)

d2_3,j = (x3− xj)2+ (y3− yj)2 (2.9c)

where d1,j < d2,j < d3,j [25].

We can solve equation 2.9 by using LSS method. First, subtracting 2.9b from 2.9a and 2.9c from 2.9a provide

d2_2,j− d2 1,j = x 2 2+ y 2 2− 2x2xj− 2y2yj d2_3,j− d11,j = x 2 3+ y 2 3− 2x3xj− 2y3yj

Arranging above two equations into matrix as following x2 y2 x3 y3  xj yj =1 2 K2 2 − d22,j+ d21,j K32− d23,j+ d 2 1,j (2.10) where K2 i = x2i + yi2, i th anchor node.

(29)

2.2 Localization techniques 15

Figure 2.2. Trilateration concept

Equation 2.10 can be written as

H ˆx = b (2.11) where H =x2 y2 x3 y3 , ˆx =xj yj and b = 1₂K 2 2− d22,j+ d21,j K2 3− d23,j+ d21,j

In case of multilateration, there are more that 3 anchor nodes, H and b can be written as, H =      x2 y2 x3 y3 x4 y4 .. . ...      and b =1₂      K2 2− d22,j+ d21,j K2 3− d23,j+ d21,j K2 4− d24,j+ d21,j .. .      Finally, ˆx =xj yj

is derived by using LS method as following equation [34]. ˆ

x = (HTH)−1HTb (2.12)

We notice that in order to avoid divergent of LS method, at least three non-collinear anchor nodes are required.

(30)

(31)

Chapter 3

Simulation Platform

Multipath is a major problem in localization. There are many options to study multipath propagation such as on-site measurement, theoretical model, and tracing method. As mentioned in introduction chapter, the thesis uses a ray-tracing simulator program to generate signal propagations in study environments. This chapter firstly defines multipath channel response for the thesis. Then ray-tracing theory is described together with shooting and bouncing (SBR) method which is a common type of ray-tracing is introduced. In order to have a good performance ray tracing result, Wireless InSite program is introduced to serve all purposes of this thesis. All test and data collection are performed through this simulator program. Then, definition of material which plays important role in ray-tracing is explained. Finally, we describe our three environment design specifications. And how three environments are created in the simulator program.

3.1 Multipath propagation

In this simulation, we assume that transmitter (Tx) and receiver (Rx) are in fixed positions. Because of obstacles, transmitted radio-waves are reflected, diffracted and scattered along paths between them. As a consequence, radio-waves arrive at Rx from different directions, and their amplitudes are attenuated. A multipath channel h(t) can be written as a sum of all arrival path [35].

h(t) =

L

X

i=1

αiδ(t − τi) (3.1)

where αi and τi are the i-path amplitude and path delay of multipath channel.

In multipath, there are 3 propagation mechanisms that can occur and have a large impact to the propagation paths. They are briefly described in following sections.

(32)

18 Simulation Platform

3.1.1 Reflection

Reflection occurs in a situation that electromagnetic wave intersects with object which its dimension is significantly larger than signal wavelength [32]. Refrac-tion also occurs together with reflecRefrac-tion in some circumstances depending on how unique of material properties i.e., some parts of light are reflected away from water surface and some are refracted into water. Mathematically, coefficients of reflec-tion and refracreflec-tion are defined. They represent as funcreflec-tions of material properties of the object.

3.1.2 Diffraction

Diffraction occurs when there is obstacle, with surface with sharp edges, located between Tx and Rx path [32]. At the edge of the obstacle, another waves are produced throughout the surface and even behind the obstacle. This make signals tend to bend around the surface and be able to reach behind an object even in NLOS condition.

3.1.3 Scattering

Scattering occurs when the medium in which electromagnetic waves propagate through, has dimension smaller than the wavelength. Due to irregularity in the medium i.g., rough surfaces or small objects, scattered waves are produced. Com-monly, scattering can be neglected in propagation model because its magnitude decrease rapidly with distance [35].

3.2 Ray tracing method

Ray tracing method is derived from an asymptotic high frequency method based on geometrical optics (GO) proposed by Kouyoumjian [36]. From transmitting antenna, transmitted waves can be modeled as GO ray tubes shooting out to all direction from the origin based on far-field antenna pattern. Further, these ray tubes are incident on objects and lead to reflected and refracted rays depending on cross section of objects.

A SBR method is one of ray tracing algorithm that is the most common used. Rays are shot out from Tx normally in high frequency, then reach the surface of obstacles (e.g.,wall,window,door,building) in an area (e.g.,indoor office, urban area) where reflection, refraction and diffraction are occurred due to Snell’s law and electromagnetic wave propagation. The reflection, refraction and diffraction rays are generated further at each intersection with obstacles. The tracing of rays continues until the simulation reaches its limit bounded by number of reflections defined by user. Further informations relating to SBR method can be found in [37]. Generally, SBR method is combined with geometrical theory of diffraction (GTD) [38] method. At high frequency, the edge of an object is taken into account. The GTD adds paths of diffraction to ray-tracing at the edge of build or the corner of the wall.

(33)

3.3 Materials 19

3.3 Materials

In ray tracing method, properties of material play important role in simulation since all features or obstacles in area are ultimately composed of many types of material. According to Snell’s law, properties of material influence a refractive index which indicates how incident rays go through and be refracted to another direction or totally reflected out of the surface. Moreover, some types of material absorb signal energy.

3.4 Wireless InSite

R

Wireless InSite R _{is an electromagnetic modeling tool for predicting the effects of}

buildings and terrain on the propagation of electromagnetic waves. It predicts how the locations of transmitters and receivers within different environments affect re-ceived signal. It models the physical characteristics of the rough terrain and urban building features, performs the electromagnetic calculations, and then evaluates the signal propagation characteristics [23].

A performance of the simulator is shown in its user’s manual guide [2]. In each tutorial, they compare varieties of plot between simulated results and actual mea-surement results which are obviously similar to each other in 4 different scenarios. This could be one example that indicates a performance of the simulator.

Figure 3.1. Spherical coordinate system [1]

Wireless InSite R_{, can simulate signal propagation in full 3D model referring to}

the spherical coordinate system, shown in Figure 3.1. It can provide SBR method for ray-tracing. The SBR method can construct ray paths up to 30 total reflections which has a computation time Tc proportional to following equation 3.2 [1].

Tc=

(NR+ NT+ 1)!

NR!NT!

(34)

where NR is the number of reflections and NT is the number of transmissions. Tc

also depends on the number of diffractions. If no diffraction is requested, Tc will

proportional to number of facets in area. However, by increasing the number of diffractions, Tc will be raised exponentially, i.e.,the number of diffraction equals

to one, Tc will proportional to number of facets squared [1].

The program characterizes properties of material in simulation by referring to roughness, thickness, permittivity, conductivity and reflection coefficient. By defining specific value of each material properties, the simulator has a list of many types of material. However, types of material which are primarily used in this thesis, are dielectric half-space, layered dielectric, perfect electrical conductor, and PEC backed layer. Any further informations relating to materials can be found in [1].

3.5 Environments

In this thesis, we analyze three different environments : indoor area, urban area, and tunnel. For simplicity, we use built-in floor plan and terrain files in simulator for indoor area and urban area environment, respectively. Tunnel is the one that we design and create platform by ourselves. Each environment details are described in following subsections.

3.5.1 Indoor area

This indoor area design is an existing floor plan filename ’IndoorFloorPlan.flp’ used in tutorial lesson [2]. The floor plan is completely created by Remcom Inc. referring to real floor plan of the building at National Taiwan University of Sci-ence and Technology (NTUST) [1, 37]. We are interested in structure of the floor plan which contains of several types of material as shown in Figure 3.2 which are normally used in indoor office.

(35)

3.5 Environments 21

Figure 3.3. 3-Dimensions view of indoor area floor plan [2]

3.5.2 Urban Area

Similar to indoor area, we also use available terrain and building data, ’Ross-lyn_Flat_Terrain.ter’ and ’rosslyn_va.city’, respectively [2]. According to Figure 3.4, these data are created based on city plan of Rosslyn city, Virginia by Remcom Inc. All buildings materials are defined as concrete. We can see 3-Dimensions view of Rosslyn city from Figure 3.5. Later, all urban area experiments are performed on Lynn street located in the middle of the city.

Figure 3.4. Rosslyn city top view terrain [2]

3.5.3 Tunnel

The tunnel design is inspired by many mine tunnels, e.g.,ore mine tunnel in Kiruna, Sweden, and roadway tunnels as shown in Figure 3.7. The floor plan of tunnel is designed in ’s’ shape as shown in Figure 3.6. We create tunnel concrete wall/ceiling

(36)

Figure 3.5. 3-Dimensions view of Rosslyn city [2]

and add at least three irregularities to their inner surface as shown in Figure 3.7. Then we complete the tunnel by copying and connecting each sample together as shown in Figure 3.8.

(37)

3.5 Environments 23

Figure 3.7. Tunnel wall and ceiling sample (left), Underground ore mining tunnel in

Kiruna, Sweden (right) [3, 4]

(38)

(39)

Chapter 4

Distance Estimation Model

via RSS measurement tests

As mentioned before, there are 3 different environments needed to be compared. In each environment, we propose 4 scenarios which are LOS, NLOS, combination of LOS/NLOS, and NLOS with obstacle. In each scenario, the position of Tx and Rx are firstly set. Next, the proper parameters of the simulator are defined and then the measurement is made by running calculation in the simulation. The data from simulation, received signal strength, are exported as log files to MATLAB R_.

Then, LS fitting algorithm is applied to the data and constants such as path-loss coefficient np and reference power P0 related to eq.(2.2) are derived. Along with these parameters, an error of the model is also presented in term of standard deviation σx. The results are shown as a plot between RSS and log-distance and

distribution of multipath noise.

Note that before the measurement is taken, simulation parameters analysis is proceed to have the most proper parameters which are suitable for each environ-ment. The parameter analysis is well described in appendix A. We must notice that in further section of this chapter, to avoid confusing, we sometimes mention about Tx and Rx which refer to anchor node and non-anchor node, respectively.

4.1 Indoor results and discussion

Simulation parameters are used in this environment are listed on table B.1 in appendix B.

4.1.1 Indoor: LOS

We assume that wide band signal (fc=2.4GHz with 500MHz bandwidth) is

trans-mitted from one Tx point through the whole area. Rx routes are represented that we have measurements of received signal power in the area for a constant distance spacing. Since coordinate of each measurement point and a coordinate of Tx are

(40)

26 Distance Estimation Model via RSS measurement tests

known to the simulator, a true distance between Tx and Rx point can be calculated as euclidean distance. For simplicity, an z-axis is ignored.

As shown in Figure 4.1, this scenario has one fixed Tx point and 5 routes of receivers in different directions.

Figure 4.1. Positions of Tx and 5 Rx routes of LOS scenario in indoor area

Figure 4.2. Received power (dBm) vs log-distance plot (left) and a distribution of

multipath noise (right) from LOS scenario in indoor area

When the measurements are done by simulator, received power (dBm) vs log-distance is plotted as shown in Figure 4.2 and LS fitting algorithm is applied to the data to derive a linear model. On the left plot, dots are all measured samples from 5 Rx routes. The line is a linear model which is a result from LS fitting.

(41)

4.1 Indoor results and discussion 27

We choose to have linear model because its simplicity. Linear model is not unique option and some nonlinear model could be better. This will discuss later in future work.

From the result, the measured samples tend to concentrate around the model. However, because of multipath, there are fluctuations in RSS when distance in-creases. On the right plot, a histogram plot shows a distribution of error (dB) between measurement values and the model. The plot shows that the model fits the measurement samples quite well, because the sample of error around zero has the highest peak. From these informations, we can conclude that in LOS scenario, the RSS tends to be stable if Rx is nearby Tx and RSS becomes unstable if Rx is moved away from Tx. According to the model, we have np equals to 1.7191, P0 equals to -3.8185 and standard deviation of the model σx equals to 3.5076. An

distribution of multipath noise is roughly similar to log-normal distribution.

4.1.2 Indoor: NLOS

In this scenario, between Tx point and Rx point must be NLOS condition, so Tx and Rx routes are placed at location that have an obstacle between them. Moreover, because the floor plan shape is in L-shape, Tx point are increased to 2 and Rx routes are increased to 8 locating in two main area in order to capture a characteristic of an area as much as possible. According to Figure 4.3, on the left side of the area, Rx routes 1 to 5 measure RSS transmitted from Tx point 1. On the right side, Rx routes 4 to 6 measure RSS transmitted from Tx point 2. Two of Tx points are activated one by one in order to avoid interference of transmitted signal from multiple sources.

Figure 4.3. Positions of 2 Tx points and 8 Rx routes of NLOS in indoor area

Accord to Figure 4.4, the RSS vs log-distance are obviously more severe than LOS scenario. A linear model decays faster than LOS. Scattered measured samples result in high errors linear model which can be observed from a histogram plot. This result also show that in NLOS scenario, a localization by RSS measurement could provide a low accuracy positioning. From the model, we have np equals to

(42)

multipath noise (right) from NLOS scenario in indoor area

5.7884. The distribution of multipath noise does not match to any probability distribution.

4.1.3 Indoor: Combination of LOS and NLOS

This is the case that we ignore to separate between LOS and NLOS scenario and derive a linear model from both scenarios. According to Figure 4.5, this model provides more errors than previous LOS and NLOS scenarios due to σx equals

to 9.9808. For other constants, we have np equals to 1.8237 and P0 equals to -11.8302. The distribution of multipath noise is hardly model as mathematical model. From these results, we can conclude it is not possible to have a precise distance estimation when measurements from LOS ans NLOS are mixed together.

(43)

multipath noise (right) from combination of LOS and NLOS scenario in indoor area

4.1.4 Indoor: NLOS with Obstacle

The purpose of this scenario is to observe how different types of material effect RSS. As shown in Figure 4.6, we use one of Rx route from LOS scenario and place an obstacle plate, which has dimension 2m×2m×1.3m, between them. Six types of material (brick, concrete, glass, metal, water, and wood) are chosen from available material list in the simulator. Then six measurements are proceed by changing type of material every time. The plot results are shown in Figure 4.7 and constants obtained from each linear model are shown in table 4.1.5.

(44)

Figure 4.7. Received power vs log-distance plot and a distribution of error due to

blockage and multipath noise (right) from NLOS with obstacle scenario in indoor area

According to the results of 6 types of material from table 4.1.5, when metal and fresh water block LOS path between Tx and Rx, reference powers P0are relatively low comparing to others. It is obvious that in case of metal, the signal is not able to penetrate through it since the simulator defines metal as perfect electrical conductor [1]. Instead of passing through the material, the signal bends over the obstacle or reflect on environment surfaces to reach Rx. In case of fresh water, it is also obvious that the signal with high frequency is not able to go through water. Thus, with the same phenomenon that also happens to metal, the signal that reaches the Rx losses its energy by diffraction and reflection. It finally results in low reference power and highly erroneous model. However, in case of other 4 materials, the simulator defines them as a layered dielectric [1]. This means that some parts of signal are able to pass through these materials and reach receivers which are nearby the Tx point. This is why other materials have higher reference power and lower model error.

(45)

4.1.5 Indoor: Results summary

All parameters for indoor area are concluded in the table below. We have all mean of x equal to zero, because least-square fitting minimize overall error as much as possible.

Scenario np P0 (dBm) µx σx

LOS 1.7191 -3.8185 0 3.5076

NLOS 2.0940 -18.2534 0 5.7884

LOS+NLOS 1.8237 -11.8302 0 9.9808 NLOS with Obstacle

Metal 1.3840 -26.8401 0 4.8909 Water 1.3564 -27.8571 0 4.8854 Brick 1.6034 -5.7950 0 4.4132 Concrete 1.7145 -8.6387 0 3.6524 Glass 1.6001 -4.8606 0 4.1923 Wood 1.6718 -4.9200 0 4.6518

(46)

4.2 Urban results and discussions

Common simulation parameters which are used in this environment are listed on table B.2 in appendix B. From eq.(2.2), in urban we specifically assume that reference distance d0equals to 10 m.

4.2.1 Urban: LOS

The Tx point is placed at the end of Lynn street and there are 4 Rx routes placed on the whole street as shown in Figure 4.8.

Figure 4.8. Positions of Tx and 5 Rx routes of LOS in urban area

Measurement results are shown in Figure 4.9. On the left, a plot is similar to LOS scenario, but the difference is that when distance increases, more fluctuations are occurred in RSS. This is because the measurements are made in wider area than indoor. The model does not fit an environment well as we can also observe a model error distribution from histogram on the right plot. According to linear model, we have np equals to 1.7609, P0 equals to -1.5229 and standard deviation of the model σx equals to 4.0127. The distribution of error is slightly different

(47)

4.2 Urban results and discussions 33

multipath noise (right) from LOS scenario in urban area

4.2.2 Urban: NLOS

The concept of this scenario is the same as NLOS scenario in indoor area. Tx point is placed at behind the corner of the building at the end of the street as shown in Figure 4.10. Then, 5 of Rx routes are created through three different streets.

By referring to Figure 4.11, RSS vs log-distance plot significantly decays faster,

np=4.6257, than LOS when the distance increases. Moreover, a fluctuation of RSS

tends to be more severe than NLOS in indoor. This leads to a high error with σx

equals to 8.2864. We also found that P0equals to -5.4303.

(48)

multipath noise (right) from NLOS scenario in urban area

4.2.3 Urban: Combination of LOS and NLOS

This scenario is identical as previous one from indoor. However, according to Figure 4.12, this model provides very severe errors than previous scenarios due to

σx equal to 18.8547. This could lead to worse localization accuracy when we use

this model for localization. We also found that np equals to 1.8237 and P0 equals to -11.8302. A distribution of multipath noise is mixed between two distributions. One is from LOS and another is from NLOS.

(49)

4.2 Urban results and discussions 35

multipath noise (right) from combination of LOS/NLOS scenario in urban area

4.2.4 Urban: LOS with obstacle

This scenario is different from the one in indoor. There is not totally no line-of-sight between Tx and Rx, but we assume the situation that if there are vehicles in the street as obstacle, how these vehicles affect RSS. 45 Objects assumed as vehicle with dimension 2m×4m×1.5m are placed at the upper, middle and lower part of the street. The lay out of Tx and Rx is identical to the one from LOS scenario as shown in Figure 4.13.

According to Figure 4.14, because there is still LOS between Tx and Rx, the RSS results are not severe comparing to NLOS scenario. However, obstacles that are located in the area disturb ray paths between Tx and Rx and cause a fluctuation in RSS. From the model, we have np equals to 1.5909, P0 equals to -2.9912 and standard deviation of the model σxequals to 4.7467.

(50)

Figure 4.13. Positions of Tx, Rx route and Obstacles in urban area

(51)

4.3 Tunnel results and discussions 37

4.2.5 Urban: Results summary

All parameters for urban area are concluded in the table below.

LOS 1.7609 -1.5229 0 4.0127

NLOS 4.6257 -5.4303 0 8.2864

Combination of LOS/NLOS 2.2288 -7.8175 0 18.8547

Obstacle 1.5909 -2.9912 0 4.7467

Table 4.2. Urban : Distance estimation model via RSS measurement results summary

4.3 Tunnel results and discussions

Common parameters of this scenario are listed on B.3 in appendix B. We define reference distance d0 equals to 1 m in this scenario.

4.3.1 Tunnel: LOS

As shown in figure 4.15, floor plan of a tunnel is in ’s’ shape. In order to cover all area as possible, we separate the whole area into 3 different area. In each area, there are one Tx point and 3 Rx routes. Thus, we have 3 Tx points and 9 Rx routes in total.

Figure 4.15. Positions of Tx, Rx routes of LOS in tunnel

In tunnel, because of it shape and irregularities, the results are different from the same LOS in other environments. As shown in Figure 4.16, even the Rx is nearby the Tx, it also receives an unstable RSS. Moreover, a fluctuation becomes worse when Rx is away from Rx. An error due to multipath noise worse than LOS in indoor, but the error is still less than the one in urban area. From least square fitting, np=1.4833, P0=0.8548, and σx=3.9301.

(52)

multipath noise (right) from LOS scenario in tunnel

4.3.2 Tunnel: NLOS

With the same concept as previous NLOS scenario, Tx point is placed to avoid LOS path. There are one Tx point and 3 Rx routes in this scenario as shown in Figure 4.17.

Figure 4.17. Positions of Tx, Rx routes of NLOS in tunnel

Figure 4.18 shows that a shape of the tunnel has an influence on the RSS. As we can observe from the left plot, from the beginning, the RSS gradually reduces

(53)

multipath noise (right) from NLOS scenario in tunnel

when distance increases. Unexpectedly, when log-distance become around 1 to 1.5, the RSS suddenly decays faster than previous trend. If we compare this result to the tunnel shape, the phenomenon is occurred when the shape changes from one direction to another. This make the linear model cannot fit to the measurement well and introduce standard deviation σx equals to 5.8864. We also found that

np=2.0614 and P0=-12.3597.

4.3.3 Tunnel: Combination of LOS and NLOS

This scenario is identical to previous one from indoor and urban. However, accord-ing to Figure 4.12, this model provides severe errors than indoor but still better than the one in urban due to σx equal to 9.7440. For other constants, we have np

(54)

multipath noise (right) from combination of LOS/NLOS scenario in tunnel

4.3.4 Tunnel: NLOS with obstacle

The purpose of this scenario is similar to the one in indoor area. As shown in figure 4.6, we also use one of Rx route from LOS scenario and place an obstacle plate, which has dimension 2m×2m×1.3m, between them. Six types of material which are brick, dry sand, glass, metal, fresh water and wood are chosen from available material list in the simulator. Then six measurements are proceed by changing type of material every time. The plot results are shown in figure ?? to

?? and constants obtained from each linear model are shown in table 4.3.5.

(55)

According to the results of 6 types of material from table 4.3.5, when metal, dry sand and fresh water become an obstacle, reference powers P0 are relatively low comparing to others. It is the same reasons as indoor area. The only difference is when dry sand is used. In case of dry sand, the simulator defines it as dielectric half-space [1]. Due to its permittivity and conductivity, the signal is able to pass through and reflect at its surface. According to Figure ?? and np, the signal power

decays very fast when distance increases comparing to other materials. This is because that the some signal power is absorbed by sand relatively large comparing to other material.

(56)

4.3.5 Tunnel: Results summary

All results from all scenarios of urban area are concluded in the table below.

LOS 1.4833 0.8548 0 3.9301

NLOS 2.0614 -12.3597 0 5.8864

LOS+NLOS 1.4310 -4.9502 0 9.7440

NLOS with Obstacle

Metal 2.2627 -19.7568 0 5.7250 Sand 3.1012 -15.1819 0 6.1699 Water 3.0257 -15.6226 0 5.3240 Brick 1.8430 2.1658 0 4.8014 Glass 2.0379 5.1252 0 5.4168 Wood 1.8267 2.6629 0 4.8972

Table 4.3. Tunnel : Distance estimation model via RSS measurement results summary

4.4 Summary and comparison

For LOS scenario, indoor area has the lowest error (σx) following by tunnel and

urban has the highest error. According to histogram plots, multipath noise dis-tribution of indoor is nearly log-normal disdis-tribution but other two disdis-tributions are slightly different. This means that in short range, RSS tends to be reliable for distance estimation and also variance of the model can be lowered since the distribution of multipath is known. However, according to the result from urban area, RSS is unstable in large area surrounded with buildings. Multipaths bounce more often in long range and cause more fluctuations in distant RSS.

In case of NLOS scenario, urban has the highest error. Tunnel and indoor have errors closed to each other. From linear model plot, RSS decays faster than the one in LOS. An error distribution of NLOS is difficult to specify to any probabil-ity distribution. Moreover, it is obvious that RSS is unstable even the distance between Tx and Rx is small according to the measured samples of all plots. This means that NLOS must be treated carefully due to its high variance in error and unreliable relation between RSS and distance. In addition, we must notice that it is not a good idea to include NLOS in LOS for distance model due to NLOS and LOS have different error distribution from each other. It is better to detect NLOS and compensate its error, separately from LOS.

For obstacle, material properties of obstacle have an effect to distance esti-mation erroneous. Dielectric half-space [1] material which has high permittivity and low conductivity such as fresh water, does not allow high frequency to pass through its medium but reflect them away from itself. This is interesting because 57 percent of human body with averaging weight (70 kg) containing of water [39]. Thus, in the room or particular area filled with a group of people has an unavoid-able effect to signal propagation. Another dielectric half-space material is dry

(57)

4.4 Summary and comparison 43

sand using as a test object in tunnel. Dry sand provides result relatively closed to water. However, sand has lower permittivity than water so the some signals can go through its medium and attenuated. For other material, layered dielectric is one type of material that is mostly used in this thesis. Glass, wood, concrete and brick are examples of this type of material. They have low or moderate per-mittivity and low conductivity which allow signal to pass through their mediums meanwhile transmitted signal power is attenuated. This type of material does not has a significant effect to signal comparing to dielectric half-space. The last type of material that we need to mention is metal. The simulator defines metal as a perfect electrical conductor. It totally reflects all incoming signals away from its surface. Metal is used as test object in three environments. In all three environments metal causes high fluctuation in received signal in all three environments.

(58)

(59)

Chapter 5

Distance Estimation Model

via TOA measurement tests

The system and environments setting are similar to RSS measurement except that TOA data mentioned in section 2.4 are collected instead. In the following three sections, we do not describe in details because all set up in each environment are exactly the same as RSS, but only the results are discussed.

5.1 Indoor results and discussion

5.1.1 Indoor: LOS

Figure 5.1. True distance model vs measured samples (left) and distribution of error

(right) from LOS scenario in indoor area

(60)

46 Distance Estimation Model via TOA measurement tests

According to Figure 5.1, the histogram plot on the right shows that Nlos is

practically zero. This means that in ray-tracing simulation, LOS in TOA mea-surement always gives true distance.

5.1.2 Indoor: NLOS

Figure 5.2. True distance model vs measured samples (left) and distribution of Nnlos

(right) from NLOS scenario in indoor area

Due to Figure 5.2, it is obvious that NLOS introduces an error in distance estimation. NLOS condition causes a delay in signal propagation before reaching Rx which we can observe from a plot between TOA and distance on the left. Comparing to a true distance model plot (line), most of all measured samples (dots) are above a true distance model. A mean of Nnlos equals to 18.197 ns and

a standard deviation equals to 8.6212 ns.

5.1.3 Indoor: Combination of LOS and NLOS

The result from Figure 5.3 shows that in this scenario, noise is mainly donated by NLOS since LOS provides true distance model. However, because all samples from both scenarios are combined together, the mean of Nnlosis reduced to 7.668

ns but standard deviation is increased to 10.994 ns. Similar to RSS, we found that it is difficult to achieve an accuracy distance estimation, if we have LOS and NLOS in the same situation.