category: literaturenote citekey: hasanintroduction2012 title: Introduction authors: "Hasan, Syed Faraz; Siddique, Nazmul; Chakraborty, Shyam" year: 2012 date: 2012-00-00 2012 doi: 10.1007/978-1-4614-3272-2_1 url: "https://doi.org/10.1007/978-1-4614-3272-2_1" zotero_key: G8GCPVFP zotero_storage: NR23927X collections: transpordi_planeerimine folder: 001_artiklid firstAuthor: "Hasan, Syed Faraz"

status: converted

Syed Faraz Hasan Nazmul Siddique Shyam Chakraborty

Intelligent Transport Systems

802.11-based Roadside-to-Vehicle Communications

Intelligent Transport Systems

Syed Faraz Hasan • Nazmul Siddique Shyam Chakraborty

Intelligent Transport Systems

802.11-based Roadside-to-Vehicle Communications

Syed Faraz Hasan College of Information and Communication Engineering Sungkyunkwan University 300 Cheoncheon-Dong, Jangan-gu Suwon 440-746, Gyeonggi-Do Republic of Korea

Nazmul Siddique School of Computing and Intelligent Systems University of Ulster Northland Road Londonderry BT48 7JL UK

Shyam Chakraborty Akselinpolku 7 I 73 02230 Espoo Finland

ISBN 978-1-4614-3271-5 ISBN 978-1-4614-3272-2 (eBook) DOI 10.1007/978-1-4614-3272-2 Springer New York Heidelberg Dordrecht London

Library of Congress Control Number: 2012941283

-Springer Science+Business Media New York 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection with reviews or scholarly analysis or material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of this publication or parts thereof is permitted only under the provisions of the Copyright Law of the Publisher's location, in its current version, and permission for use must always be obtained from Springer. Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law.

The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.

While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

To Prof. S. M. Iqbal and Mrs. Yasmeen Iqbal, and Nida, Khadeeja and Waqar —Faraz

To Kaniz, Oyndrilla, Opala, and Orla —Nazmul

To Titia, Gupi and Joanna —Shyam

Foreword

This book is concerned with 802.11-based wireless networks and their use in emerging vehicular mobility domains. The book is generally concerned with information exchange between vehicles and 802.11-based roadside infrastructure, and lays particular focus on two key research directions: mathematical modeling and handover latency in vehicular communication. The major contributions of this book are focused on the mathematical modeling of vehicular communications an important and relatively less investigated research area. Since the models developed are based on experimental data, they are capable of portraying real picture of 802.11-based R2V communications. Moreover, the issue of large handover latency has been addressed. This is one of the main hindrances in the widespread use of WLANs from vehicles. In addition to delineating a comprehensive delay analysis of handover, this book introduces a scheme to reduce the delay in the scanning phase of handovers. I sincerely hope that the book will be helpful for graduate, doctoral, and post doctoral researchers.

Prof. Dong-Ryeol Shin, Dean College of Information and Communication Engineering Sungkyunkwan University Suwon Republic of Korea

Preface

The European Union directive 2010/40/EU defines the Intelligent Transport System (ITS) as a system with advanced applications which, without embodying intelligence as such, aims to improve transport management by increasing coordination and flow of information between on-road vehicles. It has also been envisaged that the application of information and communication technologies (ICT) in the transport sector has a key role in improving efficiency, safety, public security, and freight mobility management of a transportation system. Keeping in view the contribution made by ICT in realizing ITS, the authors have explored the communication mechanism between vehicles to enable various infotainment services.

This book is a result of 3 years of doctoral research focusing on one important aspect of Intelligent Transportation System: the roadside to vehicle (R2V) communications. This research work specifically explores the use of IEEE 802.11 technologies in the R2V scenario. The outcome of this research is also highly relevant in those situations where roadside 802.11 APs are specifically deployed for R2V communications.

Ever since their introduction in the late 1990s, WLAN APs have seen massive deployment across most modern cities. The idea of using already available indoor APs in outdoor vehicular environments has recently come under scrutiny. Although WLAN APs exist in large numbers alongside roads in most developed cities, the placement of these APs is highly unplanned. These WLAN APs cannot support continuous connectivity over a large mobility domain due to their unplanned deployment. Consequently, the network services offered by WLANs are disrupted when used in vehicular environments. While various research works are focused on disruption tolerant networking, this book develops and explores a completely new research direction—developing stochastic models of disruption in 802.11 WLANs and their implications on the vehicular communication systems. Second, a vehicle spends very little time within the AP footprint, most of which is consumed in handing over to the AP. Because of the small outdoor coverage region of the AP and the high speed of the vehicle, handovers inevitably occur frequently. The delay in handing over to these APs is termed as handover latency. x Preface

In order to achieve smoother transitions between the APs, the handover latency must be reduced. This book analyzes the latency by measuring various delays incurred during handovers.

This book is written for both mature and early stage researchers including postgraduate and doctoral students. Researchers from other fields interested in vehicular communications can also find this book interesting and informative. The detailed discussion on the prevailing research trends provided here will be useful for both doctoral and post doctoral researchers. This book would also be helpful in courses related to wireless networking since it covers the application of stochastic tools in analyzing communication networks in considerable detail.

The book comprises eight chapters. The basic concepts pertinent to IEEE 802.11 networks, vehicular communications, and challenges associated with 802.11-based vehicular communications have been discussed in Chap. 1. Chapter 2 provides a detailed review of previous research done in vehicular communications. More specifically, the works pertinent to disruption tolerant networking and handover latency have been reviewed. It also introduces some recent IEEE standards that are relevant in vehicular communication. Chapter 3 discusses the measurement results on parameters such as the signal strength and the data rates supported by the indoor APs in vehicular environments. A scheme for monitoring traffic congestion using the existing WLAN infrastructure has also been presented in this chapter.

Chapters 4–6 focus upon the analytical modeling of the disruption tolerant vehicular networks. Chapter 4 provides a 2-state and, next, a 3-state Markov model to compute the long-term error rate as a means of representing disruption. Chapter 5 modifies this model to hidden Markov model and introduces the probabilistic measures of disruption. Chapter 6 contains the application of the proposed model to quantify the benefits of using inter-operator roaming. Chapter 7 discusses the issues related with handovers in the vehicular context. Latency evaluations are provided at the beginning of the chapter followed by a description of the proposed channel scanning scheme to reduce scanning phase delay. The concluding remarks and future works are supplied in Chap. 8 while references and appendices are given at the end of this book.

Acknowledgments

This book is based on Syed Faraz Hasan's PhD thesis submitted to the University of Ulster.

The authors extend their gratitude to Dr. Allen Sturat (Cardiff University) and Dr. Tom Lunney (University of Ulster) for their insightful reviews on this writing. The authors are also grateful to Dr. Swades De (Indian Institute of Technology) and Dr. Rashid Mehmood (Swansea University) for their comments and support. The authors acknowledge the support received from Dr. Syed Wasim Raza, Dr. Pawan Goyal, Vaibhav Gandhi, and Jennifer Hyndman on various personal and professional matters.

The authors are thankful to their family and friends whose personal support has been instrumental in making this book possible.

1		Introduction	1
	1.1	802.11 Wireless LANs	2
	1.2	Expanding the Mobility Domain of WLANs.	4
		1.2.1 Vehicular Communications	4
		1.2.2 V2V and R2V Communications	5
		1.2.3 Wireless Technologies for Vehicular Communications	7
	1.3	Challenges in 802.11-Based Vehicular Communications.	9
	1.4	Summary	11
2		Vehicular Communication: Issues and Standards	13
	2.1	Disruption Tolerant Networking	13
		2.1.1 Systems and Architectures	14
		2.1.2 New and Modified Protocols	17
		2.1.3 Prediction-Based Techniques	18
	2.2	Handover Latency in Wireless Networks	20
		2.2.1 Detection, Search, and Probing Delay	21
		2.2.2 Authentication and Address Allocation Delay.	23
		2.2.3 Handovers in Vehicular Context	24
	2.3	IEEE Standards for Vehicular Communication	27
		2.3.1 Wireless Access in Vehicular Environments—802.11p	27
		2.3.2 Fast Transition—802.11r	28
		2.3.3 High Throughput—802.11n	30
	2.4	Summary	31
3		Evaluation of WLAN Parameters in Vehicular Setup	33
	3.1	Parameters of Interest	34
	3.2	Measurement and Analysis	35
		3.2.1 Signal Strength	35
		3.2.2 Data Rates	38
		3.2.3 Correlation Between Data Rates and RSS	42

xiv Contents

	3.3	Application: Traffic Congestion Monitoring	44
		3.3.1 Extended MULE Concept	45
		3.3.2 Roadside Infrastructure	47
		3.3.3 Communication Mechanism	50
	3.4	Summary	54
4		Markov Model for R2V Communications	57
	4.1	Markov Models	58
		4.1.1 Fundamentals of Markov Chains.	58
		4.1.2 Markov Process in R2V Communications	60
	4.2	Estimating the Transition Probability	62
		4.2.1 Data Collection.	62
		4.2.2 Probability Distribution of Dataset	64
		4.2.3 Calculating Transition Probability	67
		4.2.4 Long-Term Error Rate	69
	4.3	3-State Markov Model	70
	4.4	Toward Hidden Markov Model	72
	4.5	Summary	73
5		Measuring Disruption in R2V Communications	75
	5.1	Hidden Markov Models	75
	5.2	HMM Representation of R2V Communication	76
		5.2.1 Model Structure	77
		5.2.2 Estimating Model Parameters	79
		5.2.3 Model Generality, Limitations, and Need.	81
	5.3	Observation Sequence of HMM	84
	5.4	Probabilistic Measures of Disruption	88
		5.4.1 Forward Algorithm	88
		5.4.2 State Probability	90
		5.4.3 Encounter Probability	92
	5.5	Traffic Pattern Analysis	93
		5.5.1 Drive Tests	94
		5.5.2 Variation in Disruption with Traffic Patterns	96
	5.6	Summary	98

6		Inter-ISP Roaming for Vehicular Communications	99
	6.1	Intra- and Inter-ISP Roaming	99
	6.2	Wireless Internet Service Provider Roaming	102
		6.2.1 WISPr Architecture	103
	6.3	Modifications in HMM	105
		6.3.1 Effectiveness of WISPr	107
	6.4	Summary	109

Contents xv

7		Handover	Latency: Evaluation and Reduction	111
	7.1	Handovers	in WLANs	111
	7.2		Experiments and Observations	112
		7.2.1	Measurement Setup	113
		7.2.2	Observations in Vehicular Environments	113
	7.3	Latency	Analysis	116
		7.3.1	DHCP Delay	116
		7.3.2	EAP Delay.	119
		7.3.3	Scanning Delay.	121
	7.4	Reducing	Scanning Phase Delay	121
		7.4.1	Scanning Orthogonal Channels	122
		7.4.2	AP Performance on Orthogonal Channels	125
	7.5	Summary		127

8	Future		Directions and Research Ideas.	129
	8.1	Future	Directions.	130
		8.1.1	Network Convergence	131
		8.1.2	Location Invariant Models	131
		8.1.3	Data Handling in Vehicular Sensor Networks.	133
	Appendix	A:	Backward Algorithm.	135
	Appendix	B:	EAP Authentication Mechanism	137
	Appendix	C:	Software Tools	139
	References			141
	Index			149

Abbreviations

3GPP 3rd Generation Partnership Project

AP Access point A2M All to minimum

ACK Acknowledge (packet)

AMPS Advanced Mobile Phone Services AMV Automatic mobile vehicle ARP Address resolution protocol

BSS Basic service set

CDMA Code Division Multiple Access CEPS Center for European Policy Studies

CTP Cabernet transfer protocol

DHCP Dynamic host configuration protocol

DSL Digital subscriber line

DSRC Direct short range communication DSSS Direct sequence spread spectrum DTN Disruption tolerant networking EAP Extensible authentication protocol

EAPOL EAP Over LAN

EDGE Enhanced data rates for GSM Evolution

ESS Extended service set

FT Fast transition

GSM Global system for mobile communication

GPRS General packet radio service HAPS History based AP selection

HMAC Hash message authentication code

HMM Hidden Markov model

HSDPA High speed downlink packet access

HT High throughput

IANA Internet assigned number agency ICMP Internet control messaging protocol

ICT Information and communication technology

xviii Abbreviations

IEEE Institute of Electrical and Electronic Engineers

IP Internet protocol

IPN IP network

ISM Industrial, scientific, medical (band) ISP Internet service provider

ITS Intelligent Transportation Systems

LAN Local Area Network

LoS Line of sight

LTER Long term error rate MAC Medium access control MANE Mobile Adhoc NETwork MAR Mobile access router

MIMO Multiple input multiple output

MN Mobile node

MRP Markov Renewal Process

MULE Mobile ubiquitous LAN extensions

NAK Negative ACK

NIC Network interface card NMT Nordic Mobile Telephone

OBU On board unit OFDM Orthogonal frequency division multiplexing

OSA Open systems authentication

PEAP Protected EAP

PEN Private enterprise number

PHY Physical (layer)

PKI Public key infrastructure PL Packet loss PMK Pairwise master key

PRMA Packet reservation multiple access

PTK Pairwise transient key QoS Quality of service R2V Roadside-to-vehicle

RADIUS Remote authentication dial In user service

RSS Received signal strength RSU RoadSide unit

RTT Round trip time

SAPS Scan-based AP selection SKA Shared key authentication SNR Signal to noise ratio SOHO Small office home office SSID Service Set Identifier SSL Secured socket layer

TCP Transmission control protocol

UMTS Universal mobile telecommunication system

V2V Vehicle-to-vehicle

Abbreviations xix

VAC Vehicular address configuration VANET Vehicular Adhoc NETwork VoIP Voice over internet protocol

WAVE Wireless access in vehicular environments

WBSS WAVE BSS

WEP Wired equivalent protocol WHO World Health Organization

WiMAX Wireless Interoperability for Microwave Access

WISPr Wireless ISP roaming

WLAN Wireless local area network WNIC Wireless network identity card WSN Wireless sensor Network

Chapter 1 Introduction

Within 20 years of the introduction of electronic computers, the need to have a connection between them emerged. The early foundations of the modern day computer network dates back to the 1960s when large universities and research labs wanted to share information between their computers. Consequently, Ethernet was developed in the 1970s to interconnect local computers via cables and wires. Ethernet, standardized as IEEE 802.3, provides a framework for wiring, and protocols for signaling between computers that are not geographically far off. A network comprising computers connected with wires is known as the Local Area Network (LAN). As of today, LANs not only provide local information sharing, but can also be connected to a router (or hub) to access the external networks. The Internet itself is an interconnection of LANs that allows information sharing over a global scale (Bodden 2008).

Soon after the popularity and success of wired LANs, interest in getting wireless services began to mount. The overwhelming interest in wireless network services led to the introduction of many wireless technologies. The main reason for the popularity of wireless services was not high data rates, instead, the provision of mobility has been the primary cause of the popularity of wireless networks and technologies. The wireless links can provide coverage in areas where it is difficult to lay cables and wires (Stallings 2008). Due to the increasing demands of mobility and freedom from wires, wireless products and services rapidly became popular in both domestic and commercial sectors. The wireless technologies penetrated into the market through two main directions. The first is with the evolution of cellular systems which primarily supported voice services but are now providing data services as well. On the other hand, Wireless Local Area Networks (WLANs) were introduced as a wire-free version of the conventional LANs. The cellular networks were meant to provide services in both indoor and outdoor environments while WLANs were meant for data services in the indoor applications only (Hassan et al. 2009b).

Over the last few years, wireless networks have seen massive development with introduction of innovative technologies. Various wireless networks and technologies have been launched each of which targets a specific application. In terms of 2 1 Introduction

coverage, wireless networks can be classified as those supporting long range communication and those that suffice for communication needs over a shorter domain. In the long range category, cellular networks became popular and different underlying technologies such as Global System for Mobile communication (GSM), Universal Mobile Telecommunications System (UMTS), Enhanced Data for Global Evolution (EDGE), etc., gained popularity (Kwok and Lau 2007). WiMAX has also been introduced as a communication service for far-fetched areas where laying cables and wires is difficult. Like long range communications, communication over shorter range also witnessed innovation with the advent of technologies such as Bluetooth, Zigbee, and 802.11 Wireless LANs, etc (Garroppo et al. 2011). IEEE standardized WLANs as 802.11a/b/g networks. The first in the 802.11 series was IEEE 802.11b that was introduced in the late 1990s. After its success, IEEE standardized a range of 802.11 variations that cater for different purposes. While 802.11 networks inherently suffice for short range indoor communications, this book explores its use in highly mobile outdoor environments.

1.1 802.11 Wireless LANs

WLANs and LANs serve the same purpose, that is, information sharing, except that the former offers network services without interconnecting the devices with cables and wires. The devices in WLAN can communicate with each other wirelessly and hence are not fixed at one particular location. The legacy 802.11 standards operate on 2.4 GHz (802.11b/g) and 5 GHz (802.11a) frequency bands, and can support different data rates ranging from 6 to 54Mbps. The advantage of using 802.11a networks over 802.11b/g is that the frequency spectrum they operate on is rarely used. On the other hand, the spectrum used by 802.11b/g is already in use by several other devices (Zhu et al. 2004). With regard to the transmission technologies, 802.11a and 802.11g use OFDM (orthogonal frequency division multiplexing) while 802.11b uses DSSS (direct sequence spread spectrum). 802.11g networks theoretically offer 54Mbps data rates and are backward compatible with the 802.11b networks. A comparatively newer standard, IEEE 802.11n, offers data rates as high as 600Mbps. A more detailed account of this standard has been given in Sect. 2.3.3.

The 802.11 networks have two modes of operation. One is the ad hoc mode, in which the mobile nodes in close vicinity connect with each other wirelessly and employ routing protocols to communicate with each other. The second mode is called the infrastructure mode (shown in Fig. 1.1), in which the mobile nodes communicate with each other via a central station called the Access Point (AP). The 802.11 AP together with the mobile nodes within its footprint constitute a Basic Service Set (BSS). A set of BSSes connected to the Internet is called the Extended Service Set (ESS) (Bhola 2002). Note that the end-users connect wirelessly to the AP, which itself has wired connections to the external network. In order to connect to WLAN, mobile nodes must have its Service Set Identifier (SSID). SSIDs are periodically transmitted by the WLAN AP in the beacon messages. A mobile node can gain

Fig. 1.1 Extended service set in 802.11 infrastructure networks

network access by listening to the beacon signals and extracting SSID information therefrom. Some vendors also give an option of disabling SSID broadcast to ensure security (Oppenheimer 2004). However, with the advent of new and more robust authentication mechanisms, SSID-based authentication is rarely employed.

A single WLAN AP deployed inside a building can allow multiple users to connect to the Internet without having to stay at a fixed location at all times. Such wireless access to the Internet and other network services increases the work force productivity by 35%, allowing easy communications among the co-workers, efficient maintenance of schedules, and quick access to emails (Reinward 2007). Ever since their introduction in the late 1990s, 802.11 networks have seen massive deployment across the cities throughout most of the developed world. Their popularity and growth was so rapid that by the summer of 2002, the number of 802.11 networks ranged between 15 and 18 million (Kanellos and Charny 2002). Schmidt and Townsend (2002) have pointed out that by the end of 2002, Wi-Fi1 connectivity would be available in most of the universities and large corporations. A recent study has suggested that the number of domestic WLAN APs is more than 14 million in the US alone (Bychkovsky et al. 2006). Hull et al. (2006) report encountering 32000 WLAN APs during their drive tests that lasted for 290 drive hours.

As of today, a WLAN AP can be traced in most of the businesses, offices, restaurants, airports, shopping malls, university campuses, and houses. WLANs have now become so ubiquitous that their use as a replacement of cellular service is being

¹ The terms 802.11 networks, WLANs, and Wi-Fi are used interchangeably for the rest of this book.

4 1 Introduction

considered. Several cellular companies have started offering paid Wi-Fi services before WLAN emerges as a wide area technology (Drucker and Angwin 2002).

1.2 Expanding the Mobility Domain of WLANs

The feature of mobility, combined with the provision of data rates that are much higher than the cellular systems, 802.11 WLANs, became extremely popular in the late 1990s and early 2000s. However, 802.11 WLANs are designed to support "restricted" mobility applications and therefore support network services only inside a building. Unlike the cellular systems, which provide network services over a larger geographical expanse by virtue of their planned base station deployment, WLANs were initially meant to cover smaller coverage regions. Therefore, WLANs inherently possess a limited mobility domain. This limitation of WLANs is now being addressed in response to the increased demand of 802.11 access over a larger area.

Presently, when aWLAN mobile node moves out of an indoor environment (home, office, etc), it has to switch over to another technology for the continued use of the network services. Instead of changing the wireless service, the idea being investigated is to use WLANs in the challenged environments. Recently, the concept of using WLANs from outdoors in high mobility vehicular environments has come under consideration. Since WLANs are capable of providing high data rates and because they are already available in large numbers, they can offer a cost-effective solution to allow communications between vehicles and between vehicles and the roadside infrastructure. Vehicular communication is an emerging research area in the Information and Communications Technology (ICT), that allows the use of safety application on roads and highways (Chisalita and Shahmehri 2004). In the following, an introduction to vehicular communication is given, which is followed by a brief discussion on the candidate wireless technologies for vehicular communications.

1.2.1 Vehicular Communications

Vehicular communication is concerned with enabling communication between vehicles, and between vehicles and roadside infrastructure to improve on-road safety. Thousands of fatalities and serious casualties in road accidents are reported every year across the world. Such accidents involve vehicles as well as pedestrians (David and Flach 2010). World Health Organization (WHO) and the World Bank predict that traffic injuries shall become the third biggest contributor to the burden of disease if necessary steps are not taken (Strom et al. 2010). It is expected that vehicular communications can play an effective role in reducing the traffic casualties consequently improving the transportation safety. This idea is not completely new because it existed in the form of "telematics" previously. However, the recent innovations in the lowcost communication technologies have substantially increased the research interest in this field (Bilchev et al. 2004). Housing communication devices within vehicles has now become commercially and technically viable as these devices become increasingly portable. In today's research world, vehicular communication is a hot issue that is being explored from different perspectives in several ongoing projects, such as the Intelligent Transportation Systems (ITS) (Joseph 2006a).

ITS is concerned with using information and communication technologies from vehicles for various purposes. Despite its advantages and widespread applications, ITS faces a variety of challenges in different countries of the world. Ezell (2010) has classified ITS applications into two types to better understand its associated challenges. The first set of applications include those ITS solutions that can be deployed independently. For example, different communities and councils can deploy roadside cameras on independent basis. The second set of applications involve those which depend on other systems. For example, roadside communication stations may be deployed independently; however, their presence shall be useless unless communication devices are installed in all vehicles of a city. If vehicles are not equipped with communication devices, they will not be able to get useful information from the roadside communication units. Lack of funding is another problem faced by largescale ITS deployment. Government transportation departments usually do not have enough funds to take ITS-related initiatives. On the technical side, ITS currently does not have sufficient standards which makes it difficult to integrate multiple ITS applications in a single system (Ezell 2010). Despite these challenges, various countries have successfully deployed ITS and are reaping its benefits in everyday life.

1.2.2 V2V and R2V Communications

The success of ITS applications depend significantly on the communication between vehicles. Like the classification of 802.11 as ad hoc and infrastructure networks, vehicular communication is also classified into two types. One is the Vehicle-to-Vehicle (V2V) communication that exploits the characteristics of the ad hoc networks. The vehicular nodes communicate with other nodes that are within their transmission range without requiring the services of a central entity. This is often referred to as the Vehicular Ad hoc Network (VANET) (Toor and Muhlethaler 2008), in which communication between the vehicular nodes is regulated by different routing algorithms and protocols (Wilke et al. 2009). One sample application of the V2V communication may be the transmission of a signal from a vehicle to others when it is about to change lanes on a freeway. This shall allow the neighboring vehicles to anticipate the lane change even in blind spots to reduce the risks of casualties.

Using cooperative communications between vehicles in a V2V scenario is getting increasingly popular. Vehicles traveling in a group are employing cooperative communication to improve road safety. Bauza et al. (2010) use cooperation among vehicles to detect traffic congestion on roads. Every vehicle traveling in a group estimates traffic conditions separately. These estimations are joined together to collectively decide whether the roads are congested. Similarly, neighboring vehicles in 6 1 Introduction

Fig. 1.2 R2V and V2V communications between RSUs and OBUs

a V2V network can avoid traffic casualties by exchanging collision warnings in a cooperative manner (Yang et al. 2004).

In 2009, General Motors (GM) introduced innovative V2V solutions that have the potential to make roads safer for commuters. The developed technology uses V2V transponder placed on a vehicle's roof, which can determine its own location and that of other vehicles in close vicinity. The device is capable of transmitting audible and visual warning messages in harsh driving conditions. The GM V2V device can detect sudden application of brakes by the vehicle ahead in order to warn the driver to slow down or change lane.

The other type of vehicular communication is the Roadside-to-Vehicle (R2V) communication which conforms to the principles of infrastructure networks. R2V communication is also often referred to as Vehicle-to-Infrastructure (V2I) communication. In R2V communications, the vehicles communicate with the roadside infrastructure, e.g., base stations and AP, to send (or receive) information. Figure 1.2 shows the V2V and R2V communication between OBUs (On Board Unit, the vehicle) and RSUs (Roadside Unit, roadside base station or AP). The sample applications of R2V communication may include regular advertisements from a gas station, indication of a free parking space in an airport, etc (Sichitiu and Kihl 2008). R2V communications may also be used to upload and download traffic information from a central server and can also support Internet services on the move. For the rest of this book, the primary focus remains on the R2V communications.

Miller (2008) has proposed an architecture that combines V2V and R2V communication in a single network. The so-called Vehicle-to-Vehicle-Infrastructure (V2V2I) network allows ad hoc communication between vehicles like in V2V network, and uses a 'Super Vehicle' to communicate with the roadside infrastructure. All vehicles in a network transmit data to the super vehicle, which in turns communicates the same to the roadside base station. Super vehicles in one V2V network can also communicate with the super vehicles of other networks. It has been shown in Miller (2008) that V2V2I architecture can reduce the bandwidth requirement of the roadside base station by a factor proportional to the number of vehicles in a network.

Note that V2V and R2V communications augment each other and are not separate networking paradigms. However, in order to provide a more focused account, this book deals with R2V communication scenarios and their associated issues.

1.2.3 Wireless Technologies for Vehicular Communications

In order to facilitate vehicular communications, the use of a suitable wireless technology is a must. Among the various wireless technologies, cellular, 802.16 WiMAX, and 802.11 WLANs have emerged as the three major candidates for use in vehicular communications. These technologies originated at different times with different aims and objectives. The cellular networks first emerged in 1981 in the form of Nordic Mobile Telephone (NMT) systems in Scandinavia. NMT was followed by the release of Advanced Mobile Phone Services (AMPS) in 1983 in the US. These cellular networks accommodated a large number of users with their planned base station deployment and frequency reuse mechanism. They were initially meant to provide voice services with mobility over a large geographical domain. The cellular networks face signal degradation problems when the outdoor base stations are accessed from the indoor environments. The cellular signals have to penetrate through the walls to reach an indoor mobile node. To tackle this problem, femtocells are recently introduced to improve system capacity by avoiding the penetration of signals through walls and buildings (Hasan et al. 2009b). On the other hand, Wireless Interoperability for Microwave Access (WiMAX) was introduced as IEEE 802.16 in 2001. WiMAX provides broadband network access with range as long as 30 miles. While cellular and WiMAX offered network services in the outdoor setups, WLANs were introduced to provide broadband network services in the Small Office/Home Office (SOHO) setups. WiMAX, WLAN, and cellular networks are all being considered for vehicular communications because of their unique features and advantages. However, they also possess various drawbacks because they are not designed to meet the requirements of vehicular communications. The suitability of these technologies in the vehicular context and their brief comparison in terms of data rates, cost, and deployment issues have been given in the following.

Cellular Networks

The cellular networks may be a reasonable choice for use in the vehicular context because the cellular base stations are already massively deployed across the cities and are already offering network services on the move. Santa et al. (2008) have shown the feasibility of using the cellular infrastructure in vehicular communications. The reported delay analysis suggests that HSPA technology over the European 8 1 Introduction

UMTS will further enhance the suitability of cellular systems for both V2V and R2V communications. Nevertheless, the cellular systems suffer from low data rates. For example, GSM EDGE and UMTS HSPDA theoretically offer 1 and 7.5Mbps data rates, respectively. Second, the cellular systems operate on licensed frequency spectrum which is purchased (or rented) by the cellular companies for dedicated use. The communication cost over the cellular frequencies may be comparatively higher than WLANs because they account for charges associated with using the dedicated spectrum. In summary, the cellular network, despite being almost ubiquitous, provides a low speed network connection with comparatively high communication costs.

802.16 Networks

Worldwide Interoperability for Microwave Access (WiMAX) is another candidate technology that is considered to support vehicular communications. Although it provides a larger coverage area than the WLANs, it is not as well deployed as WLANs and cellular networks. Therefore, dedicated WiMAX base stations must be deployed across the areas of the interest for enabling WiMAX-based vehicular communication. This certainly incurs significant deployment and labor cost. These heavy investments have also been recognized in the report published by the Center for European Policy Studies (CEPS) (Renda et al. 2009). Therefore, enabling vehicular communications by deploying WiMAX base stations shall have considerable economical constraints. On the other hand, WLANs (and cellular networks) incur no deployment costs because they are already available in large numbers in most of the developed world.

In addition to providing larger coverage area, WiMAX theoretically promises high data rates (75Mbps for 802.16 e) for fixed wireless communications but provides much lower data rates under mobile conditions. For instance, the data rates up to 10Mbps for 10 km Line-of-Sight (LoS) conditions have been reported by Ahmed and Habibi (2008). Intuitively, the data rates will further decrease with mobile non-LoS communication such as that expected in vehicular environments. Chou et al. (2009) have compared the achievable throughput from 802.11 WLAN and 802.16 WiMAX networks in the vehicular environments. The experiments which compared 802.11g with 802.16d reveal that the throughput from the former is much higher. Therefore, WLANs outplay WiMAX in terms for data rates.

802.11 Networks

There are two main reasons for preferring 802.11 WLANs in vehicular communications over the cellular and WiMAX networks. First, the WLAN APs are massively deployed across most of the developed cities of the world and hence provide reasonable infrastructure support. The already available WLAN infrastructure eliminates the need of heavy investments required for deploying the roadside infrastructure. Additionally, since they operate on free and unlicensed Industrial, Scientific, and Medical (ISM) frequency band, they do not incur additional cost of dedicated spectrum as is the case with the cellular systems. Second, WLANs support data rates that are much higher than WiMAX and cellular networks. WLANs can support fast exchange of information even at vehicular speeds (Tufail et al. 2008). With increasing interest in exploring 802.11 networks in vehicular environments, IEEE has standardized 802.11p WAVE to support information exchange among vehicles, and between vehicles and roadside infrastructure. 802.11p has been discussed in detail in Sect. 2.3.1.

Despite the apparent advantages of 802.11-based vehicular communications, there are some outstanding issues that must be addressed before vehicular communication can be realized using WLANs. The main research challenges addressed in this book are introduced in the following section.

1.3 Challenges in 802.11-Based Vehicular Communications

802.11 networks have several issues associated with them such as the rate adaptation, fair carrier access techniques, QoS provisions, interference and security, etc. The fact that WLANs are not meant to support outdoor communications further increases the challenges associated with the 802.11-based vehicular communications. V2V and R2V communication scenarios may have separate issues and challenges. However, as mentioned earlier, this book is concerned with R2V communication only; therefore, it explores the challenges in R2V communication environments. More specifically, the focus of this book is on addressing two important challenges that are pertinent to R2V communications, namely disruption and handover latency.

Disruption in wireless networks is used to mean interruption in communication services. Various sources of disruption exist in wireless networks most of which relate to the fading characteristics of the wireless channel. Fluctuations in the wireless channel and different sources of refraction and interference result in disrupted wireless services. Using 802.11 networks over a larger mobility domain introduces another kind of disruption that is specific to WLAN-based vehicular communications. This kind of disruption is different from the general interruption due to the changes in environment and wireless channel, and is primarily due to the unplanned deployment of WLAN APs. The unplanned deployment of APs leaves areas with no coverage in between two APs. Figure 1.3 shows a typical scenario in which a vehicle faces a disruption period between two connectivity periods. A vehicle faces periods of connectivity and dis-connectivity as it tries to access WLAN APs on the move. This phenomenon is termed as disruption.

One of the main challenges with WLAN-based vehicular access is reducing or tolerating this disruption. In fact, reducing irregularity in network services is a specialized area of research in wireless networking which is referred to as Disruption Tolerant Networking (DTN) (Farrell et al. 2006). The concept of DTN was initially introduced as delay tolerant networking with its main application in the deep space communications. The idea was to tolerate the elongated delays in the long distance 10 1 Introduction

Fig. 1.3 Disruption in WLAN-based vehicular communications due to the unplanned deployment of 802.11 APs

communications (Fall and Farrell 2008). The term disruption is also being used in the context of vehicular communication. Disruption in vehicular communication is the irregularity in network services received by a vehicular node due to unplanned deployment of roadside infrastructure (Eriksson et al. 2008).

In order to make the effective use of WLAN infrastructure on the move, the mobile node must be able to establish quick connections with the roadside APs. In 802.11 based vehicular communications, the mobile node leaves and enters the footprints of the APs very frequently. The process of connecting to a new AP after moving out of the footprint of the previously associated AP is called handover. The current handover procedure between a mobile node and an AP takes a considerable amount of time, which is often larger than the time a vehicle spends within the footprint of an AP. Table 1.1 shows the observations on the time spent by a vehicle within the footprint of an AP (Hasan et al. 2009a). The observations are drawn from the drive tests that are discussed in detail later in this book. It can be seen from Table 1.1 that the vehicle has to perform a handover after every 15 s. The delay in a process as frequent as this must be small. The latency associated with the increased number of handovers can also affect the QoS offered by the WLAN (Kwak et al. 2009). Therefore, the handover delay in the WLANs must be reduced for uninterrupted use of 802.11 services from vehicles.

Table 1.1 Encounter duration between a vehicle and an AP as observed in two different areas

	Area-1	Area-2
Mean encounter time (sec)	15.179	14.741
Median encounter time (sec)	7	3
Standard deviation (sec)	20.85	24.17

The WLAN access for complete mobility over a large geographical expanse also requires consideration of interoperator handovers, i.e., handovers to APs that belong to the unsubscribed (foreign) Internet Service Providers (ISPs). The ISPs, or simply operators, are the network service providers that connect the end-user AP to the Internet. Different APs deployed across the cities may belong to different ISPs. Generally, an end-user is subscribed to one single ISP, and may not access the APs that belong to other ISPs. For enabling WLANs to provide complete mobility, some sort of universal roaming must be enabled among different ISPs. If handovers to the foreign APs are prohibited, the overall connectivity of the vehicle shall become limited. In addition to reducing the handover latency, the use of inter-ISP handovers is important for 802.11-based vehicular communications.

1.4 Summary

WLAN APs have been massively deployed by the end-users ever since their introduction in the late 1990s. WLAN APs can now be traced in various commercial entities, for example shopping malls, restaurants, businesses, airports, etc, as well as in domestic buildings such as houses. Their heavy presence across most of the developed cities and their ability to support high data rates have motivated the researchers to analyze the performance ofWLANs in the vehicular environments. Vehicular communication can play a vital role in ensuring passenger safety on roads and highways by facilitating various applications such as traffic congestion monitoring, exchanging lane changing messages, warning about possible traffic hazards in advance, etc. Instead of deploying dedicated roadside infrastructure, already available WLAN APs can be used for vehicular communication. This book explores different techniques and challenges in applying these WLAN APs in R2V communications. The use of 802.11 networks from vehicles has two major limiting factors, namely disruption and handover latency, both of which have been discussed in the rest of this book.

Chapter 2 Vehicular Communication: Issues and Standards

Due to the recent advances in the wireless technology and widespread use of 802.11 networks, WLAN-based vehicular communication has attracted significant research attention. 802.11-based vehicular communication is a challenging research area with several associated issues. This book is concerned with two of these issues, namely disruption and handover delay. Recall from Sect. 1.2.2 that vehicular communication is classified as V2V and R2V communications. In V2V communications, since vehicles do not communicate with the roadside APs, unplanned deployment of APs is an irrelevant issue. Similarly, since vehicles in V2V scenario do not connect to the APs at all, the delay in handing over to APs does not concern the vehicles in a V2V network. Hence, disruption and handover issues are important considerations for R2V communications only and have little relevance to V2V communications. Therefore, for the rest of this book, the primary focus stays on 802.11-based R2V communications only.

While the basic concepts pertinent to R2V communication have been discussed in the previous chapter, this chapter explores the previous works done in addressing disruption and handover latency. First, the research issues in tolerating disruption have been discussed in the light of previous works. The adopted approaches have been divided into categories for a classified discussion. This is followed by a review of previous works pertinent to the second issue namely handover latency. The literature review on handover latency has also been divided into three categories. The research efforts focusing particularly on handovers in the vehicular context have also been highlighted. Toward the end of this chapter, three IEEE standards have been briefly discussed that can affect the performance of 802.11-based vehicular communications.

2.1 Disruption Tolerant Networking

802.11-based vehicular communication suffers from unplanned deployment issues. In case of cellular systems, an operator plans the deployment of the base stations so as to bring the maximum number of subscribers within its coverage region. In case of WLANs, on the other hand, every AP owner deploys WLAN for personal use only. WLAN deployment focuses on providing the network services only for a certain business and/or home use. AlthoughWLAN APs are massively deployed across many developed cities, their placement has not been preplanned. This unplanned placement of APs leaves uncovered regions in between the footprints of the APs when they are accessed from outdoors over a larger mobility domain. Consequently, unlike the cellular systems, the deployment of WLAN APs does not allow continuous network services on the move. As a mobile node attempts to use WLAN APs while moving at vehicular speeds, it encounters periods where a connection with an AP is possible and periods where no AP is available to make a connection with. The intermittency or irregularity in the network services offered by the WLAN in vehicular scenarios is termed as "disruption", which exists primarily due to the unplanned placement of the WLAN APs.

For enhanced end-user experience, this disruption must be reduced or tolerated. This section focuses on exploring various techniques adopted in the previous works to address disruption in vehicular communication. It categorizes the previous works done in the area of disruption tolerance as (i) those proposing new systems and architectures for disruption tolerance, (ii) those proposing new or modified protocols for reducing disruption, and (iii) those which are based on predicting the upcoming disruption. In the following sections, each of these categories is discussed in line with the previous works reported in the literature.

2.1.1 Systems and Architectures

The terms system and architecture are often used interchangeably. Both imply an interconnection of entities to perform a particular task. System may have different definitions in different contexts. In computer science, software system is a more popular term. A software system is concerned with the interconnection of software entities (codes and subroutines) that often communicate with each other. An architecture is concerned with the functional organization of its constituent units. It may also include consideration of principles and procedures involved in executing a task. The internet architecture, for example, conforms to different principles of the Internet Protocol (IP) suite. Here, some of the systems and architectures are discussed that tolerate disruption in vehicular communications.

Ott and Kutscher (2004a) have proposed a network architecture for using WLAN APs from vehicles (see Fig. 2.1). The proposed architecture requires all mobile nodes to communicate via the "drive thru" proxy placed between the mobile node and the internet. The proxy serves the mobile nodes when the direct connectivity from a roadside AP is available, otherwise, it buffers the requested contents until the connectivity is reinstated.

The main idea of drive thru internet is to conceal the frequent disruption periods from the rest of the network, and to manage or "tolerate" disconnectivity rather than offering continuous service. Since this architecture is not designed for supporting continuous services, the real-time applications such as VoIP cannot be used with

Fig. 2.1 Drive thru internet showing the intermediary proxy in between the mobile node and the network

Fig. 2.2 The vehicle saves its session in the proxy at point (a) and retrieves the same from the next AP at point (b)

satisfactory performance in this setup. In the drive thru architecture, the mobile node uses the standard protocols when communicating with the local hosts but uses an optimized protocol for communicating with the drive thru proxy (Ott and Kutscher 2004b). Using the specific protocols, the drive thru proxy saves the ongoing session when a vehicle exits the footprint of one AP. The same session is restored by contacting the proxy again via the new AP as shown in Fig. 2.2. Pegasus is another

Fig. 2.3 MAR allows the mobile nodes to connect to another wireless technology in the absence of WLAN APs

system which runs on the in situ Wi-Fi networks for providing wireless roaming (Frangiadakis et al. 2007). Pegasus houses a special cache to keep records of all assigned IP addresses which are continuously reused. The primary i.e., however, is similar to that proposed in (Ott and Kutscher 2004a), in which the overall connection is split into two: one between the mobile node and the manager proxy, and the other between the manager proxy and internet.

The same connection splitting mechanism is also used by Mancuso et al. (2004) to support streaming services in the disrupted networks. The proxy in this case downloads the required stream and delivers the contents to the mobile node whenever the connectivity is available. On top of this proxy system, A2M (All to Minimum) bandwidth sharing algorithm is applied for faster transfers between the proxy and the mobile node. Thedu is also a new system that prefetches the web search responses during the disruption periods and transmits the results to the mobile node during the connectivity periods (Balasubramanian et al. 2007). Using Thedu, the mobile nodes can download relevant web responses within 2.7 min using the unplanned network infrastructure. A system that exploits wireless diversity for improved end-user experience has been developed as MAR (Mobile Access Router) by Rodriguez et al. (2004). In essence, the idea is to switch to other wireless networks if the prevailing connection is facing disruptions. Figure 2.3 shows that a vehicle

Fig. 2.4 Exploiting base station diversity as envisaged by ViFi

switches to the cellular network as soon as the WLAN connectivity disappears. MAR supports several wide area interfaces such as GPRS, UMTS, CDMA, etc., to which the mobile node can connect for getting seemingly continuous network services. An innovative application of the roadside infrastructure for supporting a mobile sensor network has been given in (Hull et al. 2006). CarTel is a distributed system for sensing and communicating the sensed stimuli to a central portal via the wireless hot spots (Wi-Fi, Bluetooth, etc.).

2.1.2 New and Modified Protocols

In the context of communication networks, protocols define the rules and regulations of data transfer between two hosts. Transmission Control Protocol (TCP) is the most commonly used protocol for exchanging information over the internet. Since the conventional protocols (such as TCP) are not meant to support vehicular communications, several previous works have either modified the existing protocols or have proposed completely new ones.

The Cabernet Transport Protocol (CTP) (Eriksson et al. 2008), for example, has been introduced as a replacement for TCP for 802.11-based vehicular communications. While TCP reduces data rate as soon as it detects packet loss, CTP distinguishes between congestive and non-congestive packet losses and imposes rate adaptation only when packet loss occurs due to network congestion. Consequently, CTP achieves twice as high throughput as the TCP. Along with CTP, ViFi is another protocol that

Table 2.1 Flow table for ViFi

SRC transmits message M
If DST receives M, it broadcasts ACK
If AUX hears M but no ACK for a certain time t,
it relays M with probability P
If DST receives relayed M, it broadcasts M
If SRC does not receive ACK within a certain time t,

it retransmits M

minimizes disruption by using the base station diversity (Balasubramanian et al. 2008). The basic idea behind this protocol is that the frequent disruptions can be masked if a mobile node is associated with more than one base station at the same time. The main challenge in this regard is to develop coordination between the associated base stations. According to ViFi, when a source transmits a packet, it reaches other base stations (auxiliaries) along with the intended destination, as shown in Fig. 2.4. If the destination receives the packet, it broadcasts an ACK message. If an ACK is not heard from the destination, an auxiliary relays the same packet with a certain probability and broadcasts an ACK. Table 2.1 gives the brief flow of procedures in the ViFi protocol. The performance of ViFi is a close approximate of the scenario where the mobile node opportunistically communicates with all nearby base stations.

Subramanian et al. (2007) propose several changes in the link-level protocols to "survive" the disruption periods. The link-level enhancements are in conjunction with the loss-tolerant TCP (LT-TCP) protocol proposed for congestion detection. Predicting the vehicle's route to destination for enhanced routing between the vehicle and the roadside infrastructure has been addressed in (Leontiadis et al. 2010). The idea is to determine the vehicle's route in advance using the navigation system on board the vehicle, and to communicate with the AP closest to the vehicle's current location. The proposed two-way protocol facilitates both vehicle-to-AP and AP-to-vehicle communications.

While these protocols may be useful in maintaining continuous connectivity, they may require a great deal of hardware and software changes. The idea of switching between wireless technologies, for example, may require even more changes. Second, there is no known method to evaluate the performance improvements brought by the implementation of these architectures and protocols. The mathematical techniques given in Chap. 5 of this book can be used to assess the performance of different techniques that set out to reduce disruption. As an example, the improvements brought by WISPr have been analyzed in Chap. 6 using the developed stochastic models.

2.1.3 Prediction-Based Techniques

Reserving the network resources at the new AP helps in executing smooth handovers. Resource reservations require advance knowledge of the candidate APs that may be handed over to in near future. In other words, predicting the path of the mobile node

Fig. 2.5 Vehicular motion represented by state transitions in accordance with the changes in geographic coordinates

can improve the overall network performance significantly. As the name suggests, prediction-based disruption tolerance is concerned with predicting the connectivity conditions t seconds ahead of time. These forecasts can help in scheduling the network usage in a more efficient manner. Some of the works addressing predictionbased techniques are highlighted in the following.

Based on a discrete-time mobility model that is personal to every device, combined with the past network conditions, breadcrumbs predicts the near-term connectivity of the device (Nicholson and Noble 2008). It uses the state transition property of the Markov models to represent the motion of a vehicle as transitions from one state to another. Figure 2.5 shows that the state transition is recorded for different geographic coordinates of the moving vehicle. The location information of the vehicle is obtained by the GPS. This kind of prediction not only requires the knowledge of user mobility but also the performance of the AP encountered by the mobile node. Probably, the biggest issue with Breadcrumbs is estimating the performance of the next AP. For this purpose it runs a series of tests on the encountered AP after a usable connection is setup. Not only establishing a connection with an AP is time consuming but the tests with the AP further reduce the amount of usable time.

The concept of making the network resources available at locations which are expected to be visited by the user has been introduced by A-Ghazaleh and Alfa (2010). The Markov Renewal Process (MRP) has been used to make future predictions about the user's location by dividing the geographic area of interest into cells (A-Ghazaleh and Alfa 2008). The user's motion from one cell to another represents its motion along a route. The MRP predicts which cell the mobile node is likely to dwell at t seconds ahead of time. Mahajan et al. (2007) have used the location-specific performance to predict the regions where connectivity is poor. It has been shown that the long-term history of a particular location is stable enough to support predictions about the so-called "gray periods" over a trace. These kinds of history driven predictions have also been discussed in (Deshpande et al. 2009). It has been shown that the RF fingerprint (in the form of SNR) at a certain location from a particular AP can be a handy way of predicting the location of the mobile user. The SNR and location information allows offline calculation of when (and where) handoffs should be initiated to provide smoother services and reduced disruption. The historic tracks can also facilitate prefetching, which improves system performance by a factor of 2.5 (Deshpande et al. 2009). Lee and Hou (2006) have proposed a model that takes into account both steady-state and transient mobility behaviors of the WLAN users. Based on the association history of the users, the model not only forecasts about the user mobility but also predicts the prospective APs that the user may associate with in future. A 3D Markov model has been presented (Hassan and Hassan 2009) to evaluate the performance of a proxy housed inside a vehicle. This onboard proxy attempts to mask service interruptions for scenarios such as when a vehicle (in this case a train) passes through a tunnel.

While these works model vehicular communications for different purposes, in this book, 802.11-based R2V communication has been modeled to measure the available disruption. It has been discussed at length in Chap. 4 that R2V communication exhibits the feature of state transition. Therefore, Markov and hidden Markov model techniques have been adopted to measure disruption in 802.11-based vehicular communication.

2.2 Handover Latency in Wireless Networks

Handover occurs when a mobile node leaves the footprint of the previous AP and associates with a new one. Handover procedure itself involves several subprocesses, broadly classified as probing, authentication/association and allocation of unique identification. These individual handover phases have been shown in Fig. 2.6. The handover starts with the probing phase where the mobile node searches for a candidate AP. After finding an appropriate AP, the mobile node authenticates and associates with it by providing valid user credentials. Finally, the mobile node is assigned with a unique identifier (commonly an IP address) so that it becomes reachable over the network. For a fast handover, the delays associated with all these processes should be reduced.

In the vehicular context, the handovers occur very frequently. This is mainly because the WLAN APs cover a small outdoor region. When a vehicle passes by an indoor AP, it can get network services for a few seconds before it gets out of range. To continue using the network services, it has to handover to another AP. One way of reducing the number of handovers is to increase the signal strength of the APs, thus allowing the vehicle to stay within the AP footprint for a longer duration. This of course requires changes in the roadside infrastructure which is not the focus of this book. Instead of reducing the number of handovers, reducing the delay involved in performing the handover is more often addressed. Reducing handover delay is important in the vehicular context because it saves the vehicles from

Fig. 2.6 The handover process can be divided into three phases, namely, probing, authentication/association, and address allocation

spending a considerable portion of their already small connection time in executing the handovers.

In the following, the previous research efforts in reducing the delays incurred during handover have been reviewed.

2.2.1 Detection, Search, and Probing Delay

Velayos and Karlsson (2004) have divided the handover procedure into three phases, namely detection, search and execution. In the detection phase, the mobile node realizes the need for handover and the search phase is concerned with acquiring essential information for the handover. Their work has primarily focused on reducing the delay associated with these phases. In another approach, the detection and search phases have been omitted by introducing the preactive scan phase which scans the neighboring APs without terminating the prevailing connection (Manodham et al. 2005). The preactive scan phase becomes active once every 2 s and looks for the APs offering a better signal strength than the current one. Replacing the search and detection phases with the preactive scan phase results in handovers fast enough to support VoIP services. A similar approach has been adopted in (Mhatre and Papagiannaki 2006), in which long-term and short-term signal strength characteristics decide the handover. The mobile node continuously monitors the signal strength levels of the neighboring APs on same as well as overlapping channels. Instead of handing over to APs showing instantaneous increase in the signal strength, the proposed scheme initiates the handover when a decrease (or increase) in the trend is observed for the associated (or the next) AP.

SyncScan in (Ramani and Savage 2005) synchronizes the mobile node's listening periods with the beacon transmission periods of the APs. If this synchronization is achieved, the mobile node can scan channels without spending excessive amounts of time in the scanning phase. Figure 2.7 shows that the candidate AP transmits a beacon signal after every σ seconds starting from time t. The mobile node switches

Fig. 2.7 SyncScan synchronizes the mobile node's listening periods with the beacon transmission times of the candidate APs

to the candidate AP only during these times while still maintaining the usual communication session. In another attempt to reduce the probing delay, HaND (Chen and Qiao 2010) requires a mobile node to jump back to its present channel immediately after sending a probe request on another channel. The scheme can be effective if the mobile node knows in advance that it is about to disassociate from the present AP. The concept of interleaving the scanning phase with the ongoing session of the mobile node has been presented in (Wu et al. 2007). In this scheme, the mobile node keeps probing the nearby APs during its usual communication session with unnoticeable impact on the overall performance. It also addresses link asymmetry and explores the triggering mechanisms to initiate proactive scan and handover. In another approach, the handover is initiated when there are few (preferably no) packets to be delivered (Choi et al. 2010). The objective in this approach is to reduce packet loss during the handover.

Teng et al. (2009) have established thresholds for starting the prescan phase, and for initiating the handover. If the signal strength from an AP falls below the prescan threshold, the mobile node starts a background scan and selects three nearby APs as the prospective APs. As the AP signal strength becomes smaller, the mobile node hands over to one of these APs.

The use of multiple WNIC (Wireless Network Interface Card) for mobile nodes has been proposed in (Jin et al. 2009). The approach adopted here is that one of the channels is dedicated just for scanning purposes while the rest are used for normal communication purposes. Chen et al. (2009) propose the transmission of an authentication request to the APs listed in a smart list as soon as the handover is triggered. The smart list contains the candidate APs selected on the basis of their signal quality. This algorithm is referred to as the Mesh Scan algorithm.

Instead of searching for APs on all 802.11 channels, selective channel scanning calls for examining only those channels which meet a certain criteria. The Neighbourhood Graph in (Kim et al. 2004), for example, selects the channels to be scanned based on the graph topology. In the literature (Tommasi et al. 2006), a scan agent assigns weights to the channels which correspond to the probability of finding a suitable AP on a particular channel. These weights are made available to the mobile nodes which use them to select the next candidate AP. Selective channel scanning has also been used (Eriksson et al. 2008) and (Shin et al. 2004), in which mobile nodes prioritize a certain group of channels to scan. The main idea is to restrict the scan cycle to a smaller number of channels to avoid scanning the entire 802.11 spectrum.

In Sect. 7.4 of this book, a channel scanning scheme has been presented which reduces the scanning phase delay in the vehicular context.

2.2.2 Authentication and Address Allocation Delay

Security has always been an issue with the wireless networks. The vulnerabilities in the Wired Equivalent Protocol (WEP) require the use of more robust security solutions and protocols (Borisov et al. 2001). However, the robustness in the authentication mechanism should not incur extra delay in the handover procedure. Fathi et al. (2005) have evaluated the latencies in some basic authentication schemes such as Open Systems Authentication (OSA) and Shared Key Authentication (SKA). The fast handover method in (Ok et al. 2008) also operates over the OSA mechanism. However, more advanced and robust security solutions are now available for the wireless networks. For instance, the Extensible Authentication Protocol (EAP) and the 802.1X security framework (Chen and Wang 2005) are commonly employed in the recent 802.11 deployments. While EAP ensures network security, its tedious handshake mechanism increases the delay in the authentication phase. Xu et al. (2008) propose to reduce the number of handshakes in the EAP protocol to minimize the authentication delay. A similar reduction in the number of handshakes has also been proposed for the 802.11i networks in (Altunbasak and Owen 2004).

Address allocation refers to identifying the network nodes via unique tags or addresses. In a typical IP network, the task of address assignment is performed by the Dynamic Host Configuration Protocol (DHCP).1 Like all other phases, the delay incurred in this phase should also be kept to a minimum. Cardenas et al. (2008) have proposed to reduce the number of handshakes between a DHCP server and the mobile node in an attempt to reduce the DHCP delay. While the legacy DHCP requires four-way handshake between mobile node and server (see Fig. 2.8), their proposal calls for retaining only the first and last message (namely DHCP discover and DHCP Acknowledge). While this proposal can speed up address allocation, at the same time, a loss of discover or acknowledge packet can incur intolerable delays. For instance, if the mobile node fails to receive the acknowledge message from the DHCP server, it may keep on waiting for an acknowledgment in vain. Some extensions in the current DHCP protocol, such as reducing the number of handshakes in the protocol

¹ See Sects. 7.2.2 and 7.3.1 for detailed discussion on DHCP and its associated delay.

Fig. 2.8 Handshakes involved in the DHCP procedure

and detecting the subnet changes, etc., have been discussed in (Floris et al. 2003), while a seamless handoff process has been presented by Chen et al.(2007). It has been proposed in (Chen et al. 2007) that the mobile node uses previous AP until a new address from the new AP has been issued to it.

Section 7.3 of this book gives detailed measurement of authentication and address allocation delays. It also highlights the key delay contributors in the EAP-based authentication and DHCP-based address allocation methods.

2.2.3 Handovers in Vehicular Context

Smooth handovers are difficult to realize in the vehicular context because of their inherent latency and high frequency. Figure 2.9 shows that the small AP coverage leads to frequent handovers in the vehicular environments. Note that, the figure shows overlapping AP footprints which is not the case with 802.11-based vehicular communications. This section discusses the research efforts in allowing fast handovers in vehicular communications.

The advance information on neighboring APs and the path followed by a vehicle may speed up the handovers in the vehicular context (Kwak et al. 2009). Paik and Choi (2003) have studied the predictability of handovers in WLANs. The proposed method relies on the user's mobility pattern to predict the handover. The impact of velocity of the vehicle has been studied in (Emmelmann 2005), which leads to the conclusion

Fig. 2.9 Frequent handovers in vehicular environments

that signal strength based triggering mechanisms cannot be effective in high velocity scenarios. Giannoulis et al. (2008) have explored the disparity in the performance of APs and have used the per-AP scores in making the handover decisions. They report that different APs perform in a different way but their characteristics, although different from others, remain steady over a period. The per-AP scores may therefore be recorded and used repeatedly over a time period.

Apart from detecting and initiating the handovers, some works have focused on analyzing the authentication mechanism required for vehicular communications. Some guidelines about the authentication mechanisms have been covered in (Ott et al. 2005). While the private APs have a framework to authenticate the vehicles, the main challenge is with accessing and authenticating with the unsubscribed public APs. To tackle this problem, Luo and Henry (2003) argue that the customers that are not subscribed to the APs should be given an option of one-time payment for immediate access to the network. Once the payment is made, the APs will issue the user credentials and open its resources to the user in a secure way. Rapid symmetric decryption can be employed in an efficient manner for authenticating the vehicles using the hash message authentication code (HMAC) (Zhang et al. 2008a). The simulation results show that this scheme outperforms the conventional Public Key Infrastructure (PKI) and group signature schemes. Recall from Sect. 1.3.1 that the vehicles spend very little time within the AP footprint. Therefore, any authentication framework that causes intolerable delays will not be suitable in the vehicular context.

The address allocation issues in vehicular environments have also been discussed in a few works. Vehicular Address Configuration (VAC) has been introduced for VANETs in (Fazio et al. 2007), which uses a distributed DHCP mechanism for

Fig. 2.10 AMV changes its position to act as a relay between the vehicle and the AP

address allocation. According to this scheme, a vehicular node among a group of vehicles is selected as the leader. The leader serves the IP address requests for the nearby vehicles and acts as a DHCP server for them. The leaders can communicate with each other to maintain a list of already configured addresses. Note that, this scheme calls for allowing mobility for the DHCP server and proposes a new protocol (VAC) instead of using the legacy DHCP. In another approach (Mohandas and Liscano 2008), a centralized addressing scheme has been presented for the Vehicular Ad hoc Networks (VANETs). Instead of having multiple DHCP servers scattered across a certain geographic area, this scheme proposes to use one central DHCP server to assign the addresses. The roadside APs act as relays between mobile nodes and the central server to assist in obtaining an IP address. The simulation results show that this scheme leads to a smaller DHCP delay. Although the performance of this scheme has been simulated for VANETs, its use for R2V communications needs to be evaluated. Chao et al. (2010) proposes to merge the AP and the DHCP server as a single entity. Accordingly, the roadside APs will assign addresses to the mobile nodes instead of contacting the DHCP server every time an address request is received.

Note that, the aforementioned works focus on reducing the latency incurred by certain phases of the handover. On the contrary, QuickWiFi (Eriksson et al. 2008) addresses almost all phases of the handover procedure and makes modifications in almost all phases. Apart from proposing changes in the timeout and authentication/association mechanisms, it gives optimal channel scanning and connection loss detection techniques. It also analyzes the performance of individual DHCP phases in the vehicular context.

Reducing the frequency of handovers by controlling the coverage range of the APs may be another approach to reduce irregularity in the 802.11-based vehicular communication. Two range extension schemes have been given in (Amdouni and Filali 2009). The Scan-based AP Selection (SAPS) and History-based AP Selection (HAPS) work particularly well for an average density of APs across the road. Both SAPS and HAPS maintain a list of the scanned candidate APs based on a certain criteria. Once the connection with one AP is lost, the next AP is chosen from the candidate list without having to scan the channels again. The motivation is to connect to the AP closest to the mobile node so that its connectivity lasts for larger time duration. Another method of APs range extension is by using the Automatic Mobile Vehicle (AMV) as discussed in (Hu et al. 2009). As the mobile node begins to get out of range of the associated AP, the AMV positions itself such that it acts as a relay between the APs and the mobile node, as shown in Fig. 2.10.

2.3 IEEE Standards for Vehicular Communication

Various technologies have emerged over the years which meet the increasing demands of vehicular communications. IEEE has developed various standards that are applicable to vehicular communications. A brief discussion on some of the IEEE standards that may affect the network performance in the vehicular context is presented in this section. The standards IEEE 802.11r and 802.11n are not specifically designed to facilitate vehicular communications, however, their use in this context can be helpful. On the other hand, IEEE 802.11p has been standardized specifically for use in vehicular communications.

2.3.1 Wireless Access in Vehicular Environments—802.11p

Direct Short Range Communication (DSRC) aims at improving passenger safety by allowing communications between vehicles. The efforts on standardizing and implementing DSRC started in 1990s (Morgan 2010). As a result of these efforts, IEEE 802.11p has been standardized to facilitate R2V and V2V communication scenarios. The 802.11p standard (often referred to as WAVE) is aimed at enabling vehicles to exchange safety messages, such as lane-changing-warning messages and images, etc., for keeping the vehicles aware of the real-time traffic situation (Bilstrup et al. 2008).

IEEE 802.11p is basically a modified version of IEEE 802.11a. The main difference between 802.11p and 802.11a standards lies in the PHY and MAC layers. On the PHY layer, 802.11p uses a reduced channel bandwidth i.e. 10MHz in comparison with 20MHz. Using smaller bandwidth allows a larger guard band between adjacent channels to avoid Inter Symbol Interference (Stancil et al. 2007). On the MAC layer, 802.11p supports a new mode of operation in addition to ad hoc and infrastructure modes of 802.11a/b/g networks. This new mode of operation is termed as WAVE (Wireless Access in Vehicular Environments) mode. In emergency

Fig. 2.11 OBU-OBU and OBU-RSU communication scenario

Table 2.2 PHY and MAC layer differences between 802.11a and 802.11p

Layer	802.11a	802.11p
PHY	20MHz bandwidth, smaller guard band	10MHz bandwidth, larger guard band
MAC	Ad hoc and Infrastructure modes	WAVE mode in addition to legacy modes

situations, vehicular nodes operating in the WAVE mode can send and receive messages without associating with a BSS in the conventional manner (Jiang and Delgrossi 2008). The so-called WAVE BSS (WBSS) allows quick commencement of information exchange with an added advantage of very low overhead. The information exchange using WAVE may be between the vehicles or between the vehicles and the roadside infrastructure. Figure 2.11 shows the communication between OBUs and RSUs in a typical vehicular setup, while Table 2.2 tabulates the main differences between 802.11a and 802.11p standards.

Additionally, 802.11e-based priority schemes are also considered for DSRC communication. The objective is to allow service differentiation between safety and non-safety messages (Etemadi and Ashtiani 2011). 802.11e-based VANET and its broadcast services for safety related packets has been discussed in (Ma et al. 2009).

2.3.2 Fast Transition—802.11r

IEEE 802.11r, also called Fast BSS Transition (FT) standard, allows faster transition of mobile nodes from one AP to another within a single mobility domain. The main purpose of introducing 802.11r is to reduce handover latency and facilitate delay intolerant applications such as VoIP. IEEE 802.11r attains low reconnection times by skipping the authentication process at every reconnection to the AP belonging to the same subnet. By eradicating the need of going through the entire authentication procedure at every AP encounter, 802.11r can avoid considerably large authentication delays.

Fig. 2.12 Handover in the IEEE 802.11r FT context

The 802.11r handover procedure is shown in Fig. 2.12 (Hasan et al. 2011), and briefly explained as follows. The first AP to which a mobile node authenticates, referred to as R0 Key Holder (R0KH), saves the Pair-wise Master Key (PMK) generated from the authentication procedure. The rest of the APs of the subnet derive their unique session keys using the same PMK stored at the first AP. The session keys derived are used to encrypt messages exchanged between mobile node and AP. As the mobile node transits from R0KH to new AP, R0KH derives PMK-R1 and forwards it to the new AP (R1KH) (Clancy 2008). In other words, the master key is issued only once; the rest of the session keys are derived from the same master key by the encountered APs. Thus, a lengthy hand shake authentication process reduces to two round trip exchanges: first, in which mobile node expresses its desire of handing over to the new AP and second, in which new AP confirms delivery of the new authentication key.

Handover Keying (HOKEY) is another new protocol that incorporates the so-called re-authentication mechanism in EAP authentication protocol (Zheng and Sarikaya 2009). HOKEY adds two new messages to the conventional EAP protocol (see Sects. 7.1, 7.3.2 and Appendix-B) that allows commencement and completion of the re-authentication phase. HOKEY also derives subsequent session keys from the key derived in the initial authentication session (Clancy 2008). However, unlike 802.11r, HOKEY is specific to EAP authentication mechanism.

802.11n modes of operation

Mode	MIMO	Frame Structure
Legacy	Only while receiving	Legacy Structure
Mixed	Only with 802.11n stations	Legacy with 802.11b/g and HT with 802.11n
Greenfield	Yes	HT Structure

Table 2.4 Prospective IEEE standards for vehicular communications

Standard	Modification	Application in Vehicular Environment
802.11p WAVE	PHY & MAC layer changes and introduction of WAVE mode	Communications between OBUs and RSUs
802.11r FT	Skips authentication at every reconnection and supports prioritizing traffic classes	Allows very fast hand overs between APs
802.11n HT	Using MIMO and larger BW (40MHz)	Offers very high data rates

2.3.3 High Throughput—802.11n

IEEE 802.11n HT (High Throughput) has been introduced to provide data rates as high as 600Mbps to the WLAN users. 802.11n uses Multiple Input Multiple Output (MIMO) along with Orthogonal Frequency Division Multiplexing (OFDM) and a bandwidth of 40MHz (instead of 20MHz) in order to offer high data rates (Goth 2008). Frame aggregation is another method used in 802.11n to increase the throughput of the compliant devices. Since 802.11n networks use multiple antennae for reception and transmission, their transmission signal is not simple to comprehend for the legacy 802.11a/b/g stations. This is because the 802.11a/b/g standards use only one antenna. To prevent 802.11b/g and 802.11n stations from interfering with each other, 802.11n introduces protection schemes at PHY and MAC layers (Viswanathan 2009), which inform the stations about the periods when 802.11n stations are transmitting. The legacy stations stay silent during these durations and hence avoid collision and interference. 802.11n operates in three modes namely, legacy, mixed, and greenfield, in order to coexist with the 802.11b/g networks. Table 2.3 tabulates some characteristics of these modes (Hasan et al. 2011). With the popularity of 802.11n in domestic and commercial deployments, the vehicles are expected to attain higher throughputs, and hence an enhanced communication experience. Table 2.4 summarizes the key characteristics of the standards highlighted in this section.

Efforts on standardizing an improvement on 802.11n networks are already underway. Very High-Throughput 802.11ad is capable of supporting data rates as high as 7 Gbps (Goth 2011). However, since they operate on high frequency band of 60 GHz, they are meant for short range communication.

2.4 Summary 31

2.4 Summary

This chapter reviews relevant previous works that address disruption and handover latency issues. A categorized review is given by classifying different approaches adopted in these directions. This chapter highlights the research efforts in modifying and proposing different systems and protocols, in addition to the works that predict the future connectivity conditions of the mobile node. The works addressing the handover latency are categorized in terms of the individual phases of the handover, namely, scanning, authentication/association, and address allocation. Some IEEE standards that can significantly affect the performance of vehicular communications are also discussed toward the end of this chapter.

Chapter 3 Evaluation of WLAN Parameters in Vehicular Setup

Various WLAN parameters have specific significance with regard to their application domain. Parameters that affect WLAN performance in indoor environments may be different from those that are important in vehicular communication. Performance of WLAN in a particular application mainly depends on the proper evaluation of the associated parameters. The parameters such as the beacon frame interval, Short Inter Frame Space (SIFS), DCF Inter Frame Space (DIFS), Contention Window size, for example, are specific to the general purpose use of WLANs. While these parameters are also important for R2V communications, there are two PHY layer parameters that are important in any kind of wireless communications, namely signal strength and data rate. It is obvious that R2V communications shall be rendered ineffective without the availability of strong signal strength and high data rates on roads and highways. On the network layer, handover latency is an important consideration. A handover in vehicular communication must not add delay to communication services for an improved end-user experience.

In this chapter, an experimental assessment of signal strength and data rates is presented. These parameters are first defined and then evaluated in the vehicular set up.1 While these two parameters are significant for all applications that are envisaged for use in vehicular communication, there are other application-specific parameters that must be considered as well. For instance, real-time multimedia applications are getting increasingly popular these days. In addition to signal strength and data rates, network delay is a crucial parameter for such applications. Therefore, a detailed assessment of handover delay in vehicular communication has been presented in Chap. 7.

¹ A mobile node is in vehicular setup when it is placed outdoors, such as on the roads.

3.1 Parameters of Interest

The Received Signal Strength (RSS) is the signal power received by a mobile node from an AP. It is imperative to ensure that reasonable RSS is available to the mobile node for effective communication. This is especially crucial for the 802.11-based vehicular communications because the signals from the indoor APs have to penetrate through the walls to reach the vehicles. Several mathematical models have been proposed to estimate the RSS and signal loss, for example the log-distance path loss model (in Eq. (3.1)) and signal attenuation due to penetration (in Eq. (3.2)).

$$Pr(d) = P{ro} - 10\alpha \log(d) + X_{\sigma}$$ (3.1)

$$L{dB} = 32.5 + 20\log{10}(f) + 20\log_{10}(d) + (N_w * W)$$ (3.2)

where, *P*ro is the signal strength at a distance of 1 m from transmitter, α is the Path Loss Exponent, and X^σ represents the Gaussian random variable with zero mean and standard deviation of σ dB (Rappaport 1996). f is the carrier frequency, d is transmitter-to-receiver separation, N^w is the number of walls, and W is the wall loss factor.

Instead of using the mathematical models, this chapter covers the measurement of RSS in typical vehicular environments. Although the signal strength from an AP at a certain position may suffice for a particular Internet application, it may not be enough to support another application at the same location. Therefore, it is important to first evaluate the signal strength requirements of various applications and then to compare the same with the WLAN RSS levels available on the roads.

In addition to RSS, another factor that affects WLAN performance is its ability to support high data rates in the vehicular set up. A few works have addressed data rate evaluation, such as Ott and Kutscher (2005), Gass et al. (2006), Cottingham et al. (2007), etc. However, most data rate evaluations are conducted in non-urban settings such as in deserts and highways. In this chapter, data rate evaluation has been done in an actual urban setting. Such an evaluation results in the observations that have not been covered in the literature. It has been pointed out by Shin et al. (2004) that a better signal strength from the AP does not guarantee better data rates. This is because too many nodes may be connected to an AP having a higher signal strength, thus degrading its capability to support high data rates. On the other hand, Srinivasan and Levis (2006) maintain that RSSI can be a good measure of channel quality but only within a certain threshold, while Zhang et al. (2008b) have used the signal-to-noise ratio guidelines for data rate adaptation. These works indicate that the relation between data rates and RSS is still under investigation. After measuring RSS and data rate in the vehicular set up, a study analyzing the correlation between these two has also been presented.

Together with the observations on these parameters and the AP population (highlighted later), some comments on the feasibility of using the in situ WLAN infrastructure for real-time traffic congestion have been made. The "Extended MULE" concept presented later in this chapter shows that the sparsely locatedWLAN APs can still support on road safety applications. The Extended MULE concept can be seen as a sample application of WLANs in vehicular communications that requires no modifications in the existing WLAN infrastructure.

3.2 Measurement and Analysis

3.2.1 Signal Strength

The Internet applications in common use are classified as the Background, Interactive, Streaming, and Conversational by the third Generation Partnership Project (3GPP) (Chakraborty et al. 2007). The conversational applications include the realtime applications while the streaming applications are generally semi-real-time in which the user runs a video/audio stream while downloading it from a server. The interactive applications are Internet messaging applications, and the browsing and e-mail services come under the category of background applications. The main concern with regard to the use of WLANs in vehicular environments is to verify whether sufficient RSS is available on roads to support these applications. Since a commuter in a vehicle may choose to use any of these Internet applications, the minimum required RSS thresholds for all of them must be established. Note that RSS is conventionally expressed in decibel (dB, dBm, dBi, etc.) and bears a negative value. For ease of analysis, signal strength values can also be expressed in percentage as RSSI (RSS Intensity).

IEEE 802.11 standard has specified RSSI as an integer with a range from 0 to 255 signal levels. Different vendors specify their own range of measuring RSSI that extends from 0 to a certain maximum value (RSSI_MAX). RSSI can be expressed as percentage by dividing the observed RSSI level with the RSSI_MAX of the vendor card. For instance, if the RSSI_MAX of a certain card is 50 and the observed RSSI level is 30, then the RSSI in percentage comes out to be 60% (Tao et al. 2009). RSSI expressed as percentage offers an easier and more general way of analyzing signal strengths.

For evaluating the minimum RSS thresholds for 3GPP applications, these applications are executed one by one on a mobile node (laptop with Realtek 802.11g wireless card running Windows Vista) that is connected to an AP. In these tests, a Skype call represents the conversational traffic, a 300 kbps live video represents the streaming traffic, MSN chat session suffices for the interactive applications, and the web browsing session is used to represent the background applications. While running these applications one by one, the mobile node is moved away from the AP, thus reducing the RSS level. The motion of the mobile node away from the AP continues until an application fails; the corresponding RSS level is recorded as the minimum threshold for that particular application. The observed thresholds for all applications

Table 3.1 Minimum RSS thresholds for 3GPP applications (Hasan et al. 2009c)

Table 3.2 RSS observed in domestic and commercial areas (Hasan et al. 2009c)

Traffic class	RSS (in dBm)	RSS (in %)
Background	−78	30
Interactive	−69	55
Streaming	−60	80
Conversational	−57	85

	Domestic area	Commercial area
Mean RSS (%)	30.59	32.25
Median RSS (%)	30	32.5
Std deviation (%)	18.75	20.06

have been evaluated. Numerous tests on all 3GPP applications were performed in this manner but all of them have not been reported here. Since RSS thresholds were almost similar in all tests, the observations from one of the tests are recorded in Table 3.1. It can be seen from Table 3.1 that the RSS threshold for the interactive applications is the smallest while the same for conversational class is the highest. While the threshold for background applications is observed to be 55%. It was also observed that the browsing speed becomes very slow in the RSS range 30–55%.

Now that the RSS thresholds have been established for various Internet applications, they must be compared with the RSS levels offered by the indoor APs on roads. For the sake of this comparison, drive tests are performed in two areas, namely the domestic area and the commercial area. The domestic area is one in which the vehicle traverses through the residential buildings with few or no commercial entities. The commercial area, on the other hand, comprises entities such as businesses, shopping malls, etc. The drive tests are performed on a public bus that rides through the commercial and domestic areas. The 25-minute drive tests recorded encounters with 185 and 274 APs in the domestic and commercial areas, respectively, using Vistumbler (2011).

The peak RSS values recorded for first 100 APs in both areas are shown in Fig. 3.1, while these observations are tabulated in Table 3.2. The RSS sample is taken once every second throughout the time vehicle spends in AP's footprint. The maximum observed RSS was identified as the peak RSS for that particular AP. The overall mean RSS of an area is actually the mean of the peak RSS observed from the APs. As can be seen from Table 3.2, the mean RSS value for both areas is just above 30%. The RSS value of 30% is very low at which even the background class does not perform satisfactorily. Note that Table 3.2 enlists the peak RSS values only; the RSS observed at other time instants with these APs may have been even lower.

An interesting question is posed with regard to the RSS considerations in vehicular communications, "do these observations represent the RSS in all areas in a city?" In other words, can a generalization be made out of the observations recorded in these two areas? The signal strength received by the mobile node depends on several factors such as the power settings of the APs, their distance from the roads, etc. Since

Fig. 3.1 Peak RSS observed during encounters in domestic/commercial areas (Hasan et al. 2009c)

Table 3.3 RSS observed in a different area (Hasan et al. 2009c)

	RSS (%)
Mean	9.93
Median	7
Std deviation	11.10

these factors are different for different areas in which drive run is performed, the RSS observations are also expected to vary accordingly. A similar test is performed in another commercial area that lasted for 25 min. Following conclusion is drawn from the observations that are shown in Fig. 3.2 and tabulated in Table 3.3.

The number of APs encountered in this test drive is far less than that encountered in the previous two areas. From this observation alone, it follows that performance of R2V communication may vary from place to place. The statistics for peak RSS

Fig. 3.2 Peak RSS observed in another area (Hasan et al. 2009c)

obtained in this trace highlights that the mean RSS value is *<*10 %. It follows that to exploit Internet services from vehicles on roads and highways, an additional external antenna (such as one proposed by Eriksson et al. (2008)) can improve performance of these short lived connections.

3.2.2 Data Rates

Because WLANs are inherently designed for indoor use, their data rate evaluation is conventionally carried out in the indoor environments. The mathematical interpretations used for data rates also suffice for the indoor use only. The data rates estimated from Eq. (3.3) for 802.11g networks (Behzad 2002), for example, are used primarily for indoor applications only.

$$DR = sc * rate{sym} * rate{code} * t_{sym}$$ (3.3)

where, DR is the data rates, sc is the number of sub-carriers, ratesym is bits per symbol, ratecode is the code rate, and *t*sym is the symbol duration.

This section measures the WLAN data rates in the high mobility environments using the following experimental setup. Contrary to most previous investigations, these experiments are conducted in the realistic urban setting.

	NIC-1	NIC-2	NIC-3
Mean data rate (Mbps)	2.603	2.515	1.62
Maximum data rate (Mbps)	6.1717	4.7112	4.1371
Minimum data rate (Mbps)	0.0065	0	0.0187
Mean connection time (sec)	62.8	65.7	65.42

Table 3.4 Data rates obtained from various NICs in the low mobility scenario (Hasan et al. 2010d)

3.2.2.1 Experimental Setup

The experimental setup is such that an outdoor mobile node connects to an indoor 802.11g AP, which is in turn connected to a server. Since the main focus here is to evaluate the data rates, the amount of time required to connect to an AP is not addressed here. The mobile node is moved from one place to another along the road adjacent to the building that houses the AP, first at walking speed and then at typical vehicular speed by placing it inside a vehicle. The tests at walking speed are referred to as those conducted in the low mobility scenario, while those at vehicular speeds are said to be conducted in the high mobility scenario. The tests are performed on three laptops with different combinations of hardware and software. For the rest of this discussion, these laptops are referred to as the Network Interface Cards (NICs). NIC 1 is equipped with Intel driver, Intel chipset, and Windows Vista, NIC 2 has a Realtek driver, ATI chipset, and Windows Vista while NIC 3 has Intel driver, Intel chipset with Windows 7. The difference in hardware and software is deliberately chosen because any combination of drivers, chipsets, and operating systems can be used from the vehicles.

In order to report generalized results, 10 tests are performed in both low and high mobility scenarios. These 10 tests are performed on all 3 NICs in the low mobility setup and the best performing NIC is identified. The high mobility tests are performed only on the NIC that performs best in the low mobility setup. During both kinds of tests, the data rates are evaluated using IPerf (2011). The experimental setup is shown in Fig. 3.3. This evaluation requires the exchange of data from the mobile node (vehicle in Fig. 3.3) to the server via the indoor AP.

It must be noted that this is one small portion of the overall R2V communication in which mobile node encounters one AP housed inside a building. On a larger scale, several such encounters in quick successions are desired for getting continuous WLAN services on the move. The data rate observations recorded in the low and high mobility tests are analyzed in the following section.

3.2.2.2 Observations and Analysis

Ott and Kutscher (2004b) have classified the connection period between a vehicle and the AP into three phases, namely, entry phase, production phase, and the exit

Fig. 3.3 Setup for evaluating data rates in vehicular environments

phase. As the name suggests, the entry (or exit) phase is the time period during which the vehicle is entering (or leaving) the footprint of the AP. The production phase is when the vehicle is in the close vicinity of the AP, and hence gets high data rate values. Most of the data transfer takes place during the production phase. One production phase per every AP encounter has been reported in Ott and Kutscher (2004b). According to the observations from low mobility tests recorded in Fig. 3.4 (tabulated in Table 3.4), one AP encounter may give rise to two production phases as well. Figure 3.4 shows the mean data rate curve of an NIC and also highlights the two production phases observed during the tests. This trend of observing two production phases during a single AP encounter has also been observed for the other two NICs as well (see Fig. 3.5). The reason for observing two production phases is that the vehicle momentarily enters a region where the signal strength becomes very low. Due to the reduced RSS, there is a dip in the data rate curve which is evident in

Fig. 3.4 Data rates observed in low mobility vehicular setup

Fig. 3.5 Data rates observed for all three NICs (Hasan et al. 2010d)

Figs. 3.4 and 3.5. This region of low RSS is often termed as the shadowing period. The experiments in Ott and Kutscher (2004b) do not report the shadowing period because the experiments are performed on a freeway in a desert. There are fewer refracting obstacles on a freeway as compared to the urban setup. In the real urban setting, however, one AP may not always provide one continuous production phase.

The low mobility tests confirmed that NIC 1 performs comparatively better and hence it was used for further experimentation in the high mobility scenario. NIC 1 is placed in the car which is driven on exactly the same path where the walking tests had taken place. The observations from these tests may be segregated into two sets: one is the high performance set of tests in which reasonable data rates were observed, while the other is the low performance set of tests in which small values of data rates were observed. Note that the data rate tests conducted by placing the mobile

Fig. 3.6 Data rates in high mobility scenario, classified as low and high performing observations (Hasan et al. 2010d)

node inside vehicles are termed as high "mobility" tests. The high "performance" tests are a subset of these tests which have shown reasonable data rates at vehicular speeds. Although the range of vehicular speed used in the tests is not very large, still, intuition suggests that changes in the vehicle speeds and the ever changing wireless channel characteristics (Godara 2002) have caused this difference in performance.

The mean data rate in the high performance tests was 1.44Mbps whereas the maximum observed was 5Mbps. Note that the optimal data rate for vehicular communication is identified as 6Mbps in Jiang et al. (2007). Only four out of ten tests resulted in reasonable data rates, which are classified as the high performance tests. The rest of the six tests were the low performance tests in which the mean and maximum data rate values were found to be 0.33 and 1.38Mbps, respectively. Figure 3.6 shows the mean values of data rates observed for high performance and low performance tests, and the combined performance of the high mobility tests. It is obvious from this discussion that the observed data rate values are smaller than expected from the 802.11 networks. Hadallerp et al. (2007) address various issues that degrade the WLAN performance when accessed from vehicles. Keeping in view the RSS observations made in Sect. 3.2.1, the impairments in the RSS may be responsible for reduced data rates. In order to analyze the relation between RSS and data rates, a correlation study between the two has been presented in the following.

3.2.3 Correlation Between Data Rates and RSS

The measurement and analysis reported in the previous sections show that RSS and data rates are low in 802.11-based vehicular communications. This leads to the debate whether the two are interrelated. Studying the relationship between the RSS and data

Fig. 3.7 Data rates with increasing distance from AP, and decreasing RSS level (Hasan et al. 2011d)

rates is important because vehicles would normally handover to the APs with higher RSS and would anticipate higher data rates. Generally, the mobile node uses the RSS criteria to select the next AP for handovers. Because of this criteria, an AP with high RSS may have to serve more mobile nodes. Therefore, APs offering high signal strength may get too densely populated and consequently may not support high data rates.

An experiment is conducted on an indoor AP to test the variation in the data rates by varying the RSS in the vehicular environments. The RSS can be varied by simply moving the mobile node away from the AP. The data rates are recorded at different distances from the AP while moving the mobile node away from its footprint. Figure 3.7 shows the steady decrease in the RSS as recorded from the periodic address

Statistical measure	Observation	Remarks
Pearson correlation	0.423	Weak positive correlation
t-value	1.40	Insignificant correlation
Co-efficient of determination R2	0.179	Small dependence

Table 3.5 Relationship between the signal strength and the data rates

resolution protocol (ARP) messages received by the Microsoft Network Monitor running on the mobile node. Starting at an initial distance of approximately 17 ft from the foot of the building that houses the AP, the data rates are recorded at approximately 8.5 ft intervals from the initial position. Figure 3.7 also shows the measured data rates, RSS, and the locations at which these measurements are taken. The data rates are measured using IPerf by sending UDP datagrams continuously to the AP for 30 s from location x (x : 1 → 11). It can be seen that despite the fact that the RSS is constantly decreasing, the data rates face a sudden peak at step 7 in Fig. 3.7.

The statistical analysis tabulated in Table 3.5 shows that small positive correlation exists between data rates and RSS. The Pearson correlation is 42.3% which indicates a weak positive correlation. One of the main reasons for the small correlation between RSS and data rates is the presence of external interference (Raman et al. 2009). The t-test is often used to measure the t-value that quantifies the significance of correlation. For d f = 9 and p = 0*.*05, the critical value from the t table comes out to be tc = 1*.*83, while the calculated t = 1*.*40. For all cases when t < tc, the null hypothesis cannot be ignored. This infers that even this small correlation between data rate and RSS is not significant. Furthermore, it can be seen from the co-efficient of determination (*R*2) that only 17.94% of variation in the data rates can be attributed to RSS. In summary, a strong signal strength from the AP does not ensure high data rates and handovers to the APs with high RSS may not always support high data rates.

To this end, the measurement results on data rate and RSS have been presented. It has been shown that these two parameters are weakly correlated. Therefore, an AP providing low RSS level on the roads (which is often the case) does not necessarily support low data rates and vice versa. The next section uses these parameters along with other information to study the feasibility of using WLANs in traffic congestion monitoring on the urban roads.

3.3 Application: Traffic Congestion Monitoring

Road accidents and casualties have become a serious hazard in everyday life. Thousands of fatalities and serious casualties are reported every year across the world. Traffic congestion has also become a serious issue. Traffic congestion costs around £20 billion every year on the UK roads (Joseph 2006b), while the same in the US is 10 times higher Dot (as in (2006)). This calls for proposing and analyzing innovative methods to monitor traffic congestion on roads. This section analyzes the use of roadside WLAN APs for traffic congestion monitoring application.

Conventionally, the wireless sensor networks (WSN) spread across the area of interest are usually employed to support traffic congestion monitoring. Enabling low power application in the WSNs has been an interesting area of research for quite some time. Recently, the use of mobile ubiquitous LAN extensions (MULE) entities has been introduced to enable low power sensor nodes. MULEs are mobile nodes (such as vehicles) that can communicate with the fixed sensor nodes to receive information and convey the same to a nearby base station, by means of their mobility (Shah et al. 2003). MULE provides energy-efficient communication with the WSN because the sensor nodes only communicate with the MULEs over short distances and hence consume less energy. Further in the MULE approach, communicating a sensed stimulus to the base stations requires the resources of only one sensor node, which otherwise would require multihop communication over several intermediate sensor nodes. One issue with the MULE approach is that in order to monitor a large geographical expanse, considerable number of sensor nodes and base stations must be deployed across the area of interest. Therefore, the MULE concept faces heavy financial constraints. This concept can be revised (or extended) to propose a cost-effective scheme for monitoring traffic congestion on the urban roads using indoor WLAN APs. The "Extended MULE (X-MULE)" concept for traffic congestion monitoring has been proposed and discussed in the following.

3.3.1 Extended MULE Concept

The original MULE concept has been shown in Fig. 3.8a. X-MULE proposes the following changes in the MULE concept (see Fig. 3.8b).

Instead of deploying dedicated base stations, the already available WLAN infrastructure (series of roadside 802.11 APs) can be used to receive sensed stimuli from the sensor nodes. This can significantly save the deployment costs required to set up a dedicated infrastructure (Cho 2007).
Instead of placing sensors at fixed locations across the area of interest, the same can be placed on vehicles, such as the public buses. This adds the notion of mobility to the legacy WSNs that are generally fixed at a particular location.

The public buses are suitable vehicles to house the sensor nodes. This is because the buses traverse all the major areas of a city on frequent basis. In contrast to taxis and cars, buses cannot change their route if they face congestion, and hence are in a better position to monitor traffic variations. Also note that traffic congestion does not occur on all roads of a city. In fact, only specific areas which are more frequently visited are subject to congestion. While private cars and taxis try to avoid such routes, buses are often scheduled to run more frequently on the congested roads.

Fig. 3.8 MULE and Extended MULE approaches for traffic congestion monitoring (Hasan et al. 2011c)

Therefore, X-MULE proposes to place the sensor nodes on buses which traverse across the city at regular intervals, sense, and convey traffic information to the already available WLAN APs. Figure 3.8b shows that a bus (mounted with a sensor node) conveys the sensor information to the WLAN AP as soon as it enters its footprint.

3.3.1.1 Comparison with Other Works

X-MULE is a modified version of the MULE idea presented in (Shah et al. 2003). As pointed out earlier, while shah et al. (2003) propose to have the sensors laid across the area of interest, X-MULE proposes to place the sensors on the vehicles (such as buses). The advantage of placing the sensor nodes on buses is that the sensor nodes shall not have to wait for the MULEs to arrive and collect their sensed stimuli. Li et al. (2009) have also studied vehicular sensor networks by considering the use of taxis to monitor traffic. Since different roads and streets have different densities of taxis (Li et al. 2009), some mechanism of detecting multiple deliveries of the same information must be developed to make this scheme effective. While Li et al. (2009) have given two algorithms to detect congestion, in this work, the main concern is to determine whether the WLAN infrastructure can support traffic monitoring applications. The design and implementation of a mobile sensor system has also been covered by Hull et al. (2006) who exploit Bluetooth and WiFi networks. In X-MULE, however, the main focus in on the WLAN infrastructure only. WLAN footprint has been used in a neural network approach to detect vehicles in Caceres et al.(2009), while a technique to detect traffic congestion has been proposed in P-Atikom et al. (2006). This work focuses on the communication aspects of X-MULE; the techniques to detect congestion have not been discussed here.

3.3.1.2 X-MULE Issues

The communication aspect of X-MULE requires detailed consideration of the following issues.

• It is obvious from the previous discussion that reasonable roadside infrastructure is pivotal in supporting the X-MULE idea. Not only the population of APs, but their location is also significant. One AP located close to a road is better than several of them deployed far away from the road. In addition to the population and location of the APs, the data rates supported by the APs are also important. For instance, encountering 802.11g and 802.11n APs is more desirable for quick transmission of information.
A vehicle must stay connected with the APs during most of the time it spends on the roads; in other words, the ratio of connected time to the total drive time must be high. It is worth mentioning here that a high population of APs does not guarantee a large connection ratio. This is because the APs are classified as open and closed based on their authentication schemes. While open APs are accessible by the walk-up users, closed APs require proper user credentials before granting the network access.
Finally, the communication between the vehicles and AP for conveying congestion information must not affect the performance of thatWLAN AP. Note that the public WLAN APs that are used in X-MULE shall also be serving their respective owners. The use of these APs for traffic congestion monitoring should not degrade the QoS received by their users.

In the following sections, the aforementioned issues shall be addressed to comment on the feasibility of X-MULE concept.

3.3.2 Roadside Infrastructure

The series ofWLAN APs that are located alongside roads is referred to as the roadside infrastructure. In this section, an analysis on the roadside infrastructure in the form of number of APs, their radio type, and authentication scheme has been given. The objective is to determine whether reasonable WLAN APs are available to support X-MULE concept. This analysis is based on the observations taken from the drive tests carried out in the city of Derry, NI, from a chosen area to another. The drive tests run from Strand Road (SR) to Northland Road (NR) via Buncranna and Branch Roads and back through the same route. Followed route is shown in Fig. 3.9 with starting and ending points inscribed in boxes. While this route faces reasonable congestion during the rush hour, no congestion was observed during the tests. The vehicle speed was manually limited to 25–30 km/hr to represent the congested traffic patterns and 45–50 km/hr to represent normal traffic conditions.

The information on the AP encounters has been collected using Vistumbler2 (Vistumbler 2011). During the tests, the DHCP utility in the mobile node was switched off to avoid connections with any open unencrypted AP. Only the information on the availability of the APs was desired which was obtained without connecting to

² The mobile node was not equipped with the GPS utility during the tests and the information on the location of APs was not collected to ensure privacy.

Fig. 3.9 The route showing starting and ending location inscribed in boxes (Hasan et al. 2011c). The tests are performed in the City of Derry, Northern Ireland, UK

Table 3.6 APs encountered during the tests conducted in dense traffic conditions (Hasan et al. 2011c)

	SR to NR	NR to SR
Total number of APs	34	31
Number of Open APs	25	18
Number of 802.11b	0	1
Number of 802.11 g	31	26
Number of 802.11 n	3	4

them. Table 3.6 reports the observations of the drive tests conducted at low speed (representing dense traffic conditions), while Table 3.7 reports the same for the normal traffic conditions. For clarity and ease of analysis, the results are classified in two sets; one set of observations is for the drive test from SR to NR, while the other corresponds to the test from NR to SR. It is encouraging to note from the observations that reasonable infrastructure is available to support the X-MULE concept. The following contains a more detailed analysis of these observations.

Tables 3.6 and 3.7 have shown that on an average 30 APs remained available in both normal and dense drive tests. It can also be observed from the tables that

Table 3.7 APs encountered during the tests conducted in normal traffic conditions (Hasan et al. 2011c)

	SR to NR	NR to SR
Total number of APs	25	27
Number of Open APs	18	14
Number of 802.11b	1	0
Number of 802.11 g	19	27
Number of 802.11 n	5	0

Table 3.8 Encounter time statistics for dense and normal traffic conditions (Hasan et al. 2011c)

	Dense traffic	Normal traffic
Encounters <10 s	0	11
Mean encounter time (sec)	57.3	55.76
Median encounter time (sec)	57	57
Standard deviation (sec)	1.4	9.21
Overall connected (%)	60.22	55.76

66.15% of encountered APs in dense conditions and 61.5% of encountered APs in the normal conditions are open. Presently, the closed APs do not provide access to the walk-up users; however, the implementation of WISPr can even make these APs available to the vehicles. Also note from Tables 3.6 and 3.7 that 802.11g APs are more popular among the WLAN users. Even a small encounter of 10 s with an 802.11g AP can practically transfer up to 33.75MB of data (Geier 2002). Theoretically, the 57 (802.11g) APs recorded in Table 3.6 can transfer 3847.5MB of data in a 10 s encounter. 802.11n APs were also encountered during the drives. The encounters with these APs can increase the data transfer capacity by many folds.

3.3.2.1 Encounter Duration

Table 3.8 records the encounter time statistics for the test drives. The observed mean encounter time of more than 55 s is consistent with the observations recorded in Ott and Kutscher (2004a). Recall from Sect. 1.3.2 that the mean encounter time of 15 s was observed for a particular area. This difference in encounter times implies that different areas have different encounter time characteristics. This shall be highlighted again in Chaps. 4 and 5 while modeling disruption in R2V communications. Also, note from Table 3.8 that the mean and median encounter times for dense and normal tests are similar but their standard deviation is very different. High standard deviation is observed in the normal trace because 11 encounters lasted for *<*10 s (which were not recorded in the dense conditions). The importance of 10-second encounter time comes from the fact that the delay in handing over to the AP ranges between 8 and 10 s (Mishra et al. 2003). Therefore, only the encounters lasting more than 10 s shall be productive in R2V communications. Assuming the handover latency to be 10 s, the average usable time (timeusable) for vehicles in normal and dense conditions becomes 45.76 and 47.3 s, respectively. The average theoretical data transfer capacity (DTCavg) can therefore be calculated using Eq. (3.4).

$$DTC{avg} = DataRate * time{usable} * N$$ (3.4)

where theoretical DataRate is 54Mbps for 802.11 g APs and 600Mbps for 802.11 n APs, N is the number of APs encountered. Using the above relation, DTCavg for dense and normal traces are found to be 18199 and 14208MB, respectively, for 802.11g networks.

3.3.3 Communication Mechanism

This section discusses the information exchange between the vehicle and the roadside AP. The vehicle can make use of the ISM links to convey information regarding traffic congestion. It has been mentioned in Sect. 3.3.1 that the traffic-related communication must not degrade the performance of the public APs. The AP performance (δ) is mathematically expressed in Eq. (3.5). Intuitively, it depends on the end-user load χ, its intensity γ, and the signaling overhead of the congestion messages σ. The signaling overhead further depends on the frequency φ and the packet size θ of the congestion messages, as given in Eq. 3.6.

$$\delta = f(\chi, \gamma, \sigma) \tag{3.5}$$

$$\sigma = f(\phi, \theta) \tag{3.6}$$

It is assumed that the frequency of sending the Internet control messaging protocol (ICMP) messages is once every second (φ = 1*/s) while three different values of packet sizes (θ = 1.5,* 3, 4*.*5 KB) are examined. As the Internet users can use four different 3GPP applications, the load intensity γ can be further expanded as given in Eq. (3.7).

$$\gamma \epsilon { \gamma_1, \ \gamma_2, \ \gamma_3, \ \gamma_4 } \tag{3.7}$$

where γ¹ denotes the conversational applications, γ² represents streaming applications while γ³ and γ⁴ represent the interactive and background applications, respectively. Note that the bandwidth requirements of these applications decrease from γ¹ to γ4.

Based on the above parameters, the following experiments are performed to evaluate the AP performance when the vehicles use their ISM links for delivering congestion information to the roadside APs. In order to represent the scenario in which a vehicle sends congestion signals to an indoor AP as it serves another indoor node, one indoor and one outdoor mobile nodes are connected to the Internet via the same AP. The setup is shown in Fig. 3.10. The node that behaves as a general indoor Internet user during the experiments is referred to as Node A (a laptop running Windows Vista

Fig. 3.10 Experimental setup for checking the impact of ISM emissions on AP performance

with Realtek 802.11g driver). Since Node A can choose to run any 3GPP application, the tests are performed with the two most bandwidth demanding applications: conversational (γ1) and streaming (γ2). The other node in the setup is Node B (a laptop running Windows Vista with Intel(R) Wireless WiFi link 5100 driver), which represents the outdoor vehicle that conveys congestion information to the AP via the ICMP messages.

In the first experiment, Node A played an online 300 kbps video stream for 15 min through the AP, while Node B sent 3 KB ICMP messages continuously over the ISM links to the same AP. In the vehicular context, Node B floods Node A's AP with congestion information over the ISM band. The throughput for Node A is evaluated under this condition and compared with that achieved in the absence of congestion signaling. The throughputs calculated using the OnlineEye Pro are graphically shown in Fig. 3.11a and b. Note that measuring the performance of streaming application represents the downlink characteristics only. When a mobile node plays a stream, it receives information from the server without necessarily uploading anything. Therefore, a slight decrease in the downlink throughput (43.81 kbps with ISM emissions compared with 46.71 kbps without them as shown in Fig. 3.11) may be observed. Note that there are some glitches in the throughput diagram shown in Fig. 3.11b due to the congestion signaling. During the experiment, the mobile node continuously sent the ICMP messages to the AP (with φ = 1/s), which shall not be the actual case for reporting congestion. One ICMP signal from the vehicle to the AP would be enough to report congestion in normal environmental conditions. Therefore, it is safe to assume that these glitches shall not be present in the throughput diagrams of actual X-MULE application. Note that this evaluation involves one vehicle and one AP. More rigorous evaluation requires multiple vehicles communicating traffic information to the AP. Moreover, since X-MULE proposes to use public buses, a method of identifying different buses needs to be established as well.

For assessing the impact of congestion signaling on AP performance while it supports conversational application, the same procedure is repeated with Node A making a Skype call of around 10 min with a wired client (node-C in Fig. 3.10). Node B, like the previous experiment, floods the AP with the 3 KB ICMP messages.

Fig. 3.11 a/b: Throughput diagram for node-A playing streaming video while node-B is/is not sending ICMP messages (Hasan et al. 2011c). a without congestion signal, b with congestion signal

The observations from these tests infer that there is no significant difference between the amount of information sent (or received) per unit time; however, like the former experiment, ISM emissions do introduce glitches in the throughput diagrams which may lead to annoying results for the end-users. It should again be emphasized that the actual traffic monitoring application would not require so many signals sent over the ISM links and hence will not even lead to this small performance degradation. The throughput diagrams for Node A in the presence and absence of ISM emissions are shown in Fig. 3.12a and b, respectively.

Fig. 3.12 a/b: Throughput diagram for node-A making a skype call with node-C while node-B is/is not sending ICMP messages (Hasan et al. 2011c). a without congestion signal, b with congestion signal

The performance of X-MULE concept also depends on the available RSS on the roads. It has been shown earlier in this chapter that different network applications require different RSS levels for satisfactory performance. The following experiments determine the RSS levels that are required by a vehicle to convey a congestion signal to the AP. ICMP ping messages of three sizes are exchanged between the mobile node and the AP at different RSS levels to determine round trip times (RTT) and packet loss (PL). The focus on RTT is because of the fact that the delivery of congestion message is a time critical issue and should not incur large delays. Similarly, studying PL is important because smaller the PL, smaller the need to retransmit the same information repeatedly. While Eriksson et al. (2008) have used 1,500 bytes ICMP

Table 3.9	RTT (in msec)
	increasing with increasing
	packet size and decreasing
	RSS (Hasan et al. 2011c)

Table 3.10 Packet losses increase with increasing packet size and decreasing RSS (Hasan et al. 2011c)

RSS (%)	1.5KB	3 KB	4.5 KB
80–70	192.66	242.33	275.66
70–60	275.33	314.66	344.44
60–50	342.33	433.33	427.66
Average	270.01	316.99	349.25

RSS (%)	1.5 KB	3 KB	4.5 KB
80–70	2	2	1
70–60	2	2	4.33
60–50	3	4.66	6.66
Average	23.33	28.88	39.96

ECHO message for probing the APs, these tests investigate the performance of 1.5, 3, and 4.5 KB ICMP packets at varying RSS levels. Three tests are conducted for each packet size at three different RSS ranges, 50–60, 60–70 and 70–80%. In order to change the RSS, the distance between the AP and the mobile node was increased by moving the mobile node away from the AP. At different distances from the APs, the RSS fell in the pre-defined ranges where the RTT and PL were evaluated. Each test comprised of sending 10 ICMP messages of the said sizes to the AP.

Tables 3.9 and 3.10 quantify the inverse relation that exists between RTT and RSS, and between PL and RSS. It can be inferred from Table 3.9 that if the RSS from the roadside APs is low, delays may be introduced in transferring the congestion signal to the AP. Furthermore, Table 3.10 suggests that for better reliability (reduced PL), the size of the congestion signal must be kept small. As can be seen from Table 3.10, 39.96% of the packets were lost while experimenting with 4.5 KB ICMP packets at varying signal strengths.

3.4 Summary

This chapter gives the measurement and analysis of the data rates and RSS supported by an indoor AP in the vehicular setup. The RSS thresholds for the 3GPP applications are evaluated and compared with the signal strength available on the roads. Data rate evaluation in a typical urban setting is reported under low and high mobility scenarios. Based on the finding that both data rate and RSS levels are low, the correlation between these two parameters is analyzed. The correlation analysis suggests that the small RSS from most of the APs in the vehicular environments may not always infer a low data rate. Toward the end of this chapter, X-MULE concept for traffic congestion monitoring was presented and discussed. Various issues related to this idea such as the infrastructure support, connection times, RSS threshold required for 3.4 Summary 55

congestion signaling and the impact of these signals on individual AP performance are also discussed.

While the data rate and RSS considerations have been taken into account in this chapter, the issue of disruption has not been covered here. It has been pointed out earlier in Sects. 1.3.1 and 2.1 that the problem of disruption arises inherently because of the unplanned deployment of WLAN APs. The subsequent chapters focus on measuring disruption in the vehicular environments using mathematical modeling.

Chapter 4 Markov Model for R2V Communications

Mathematical models are used in various fields of science and engineering to represent a system in terms of mathematical equations. A system expressed in terms of mathematics allows detailed analytical evaluation of the same. In the context of this book, the main interest is in representing vehicular communications using mathematical models so that some means of analyzing disruption are developed. It has been discussed in Sects. 1.3 and 2.1 that disruption has been a major problem with WLAN-based vehicular communications due to the unplanned placement of 802.11 APs. While some efforts have been made that focus on tolerating disruption, this chapter (and the next one) models R2V communication setup and introduces a mathematical interpretation of disruption. The main motivation behind mathematically modeling disruption is that its quantitative analysis is imperative to assess the extent of tolerance required in a particular area.

Among the various modeling tools available, Markov models have been used overwhelmingly in modeling different communication systems. Stern et al. (1994) have used the Markov models to model the effects of short silent gaps during a telephonic conversation. The communication between two parties has been represented by the transitions of a communication device between a set of states. The state transitions depend on the device's talking and silent characteristics. The developed model is actually an enhanced version of the Brady model (Brady 1969) that models on–off speech patterns in a two-way conversation. The Brady's model is a six state model which incorporates several user behaviors during a conversation such as talk spurts and long silence gaps, etc; it ignores the silence gaps shorter than 200 msec. The state transition characteristics of the Markov model have been exploited in (Goodman and Wei 1989) to detect speech activity during a conversation. The model is used to analyze the performance of Packet Reservation Multiple Access (PRMA) by varying parameters such as channel rate, source rate, delay, etc. In each of these approaches, a process or entity changes its state based on certain events and probabilistic parameters. Note that the state transition addressed in these models does not relate with the physical motion of the node; rather, these works focus on modeling the changing behavior of a stationary node as state transitions of a Markov model.

This chapter shows that theMarkov models can be used to represent the connection between the roadside WLAN APs and the vehicles. The chapter starts with the basic introduction of Markov models followed by a discussion on how they can be used to model R2V communication scenarios.

4.1 Markov Models

4.1.1 Fundamentals of Markov Chains

A collection of discrete random variables is often termed as a discrete-time stochastic process X, such that X = {Xn, n ∈ N}, where N is a discrete index set and Xn is the state of the process at time instant n (Ross 2002). This stochastic process is referred to as a Markov chain if the probability of transition to the next state j from the present state i only depends on state i and not on the previous states. More specifically, in a Markov chain, the probability that a system is in state xn⁺¹ at time n + 1 depends only on the probability that it is in state xn at time n (Haykins 2009). This property is termed as the Markov property. If a system does not exhibit the Markov property explicitly, an implied representation of the same may also be constructed (Stewart 2009).

$$P(X{n+1} = x{n+1} | X_n = xn, X{n-1} = x_{n-1}, \dots, X_1 = x_1)$$

$$= P(X{n+1} = x{n+1} | X_n = x_n)$$ (4.1)

The probability of transition from state i to j, Pi j , is given in Eq. (4.2).

$$P(X_{n+1} = j | Xn = i) = P{ij}$$ (4.2)

Since Pi j is the conditional probability, it satisfies the conditions given in Eq. (4.3) and (4.4) (Ching and Ng 2006):

$$P_{ij} \ge 0 \ \lor \ i, j \tag{4.3}$$

$$\sum{j} P{ij} = 1 \lor i \tag{4.4}$$

Equation (4.4) suggests that the row elements of the transition probability matrix must add to 1.

An m-step transition between the states i and j infers that the system has to undergo m transitions between states i and j.

$$P{ij}^m = P(X{n+m} = x_j | X_n = x_i)$$ for $m = 1, 2, ...$ (4.5)

4.1 Markov Models 59

Fig. 4.1 Schematic of Modified Brady's model

Figure 4.1 shows the Markov model that represents the communication between nodes as state transitions (Tsankov et al. 2007). Note that direct transition between state 0 and state 2 is not possible in Fig. 4.1. The transitions between these states involves an m-step transition where m = 2. It can also be seen from Fig. 4.1 that the communication system is making transitions between the 3 states, ON, OFF, and 2ON. These transitions represent the underlying communication process between the two nodes. For instance, state-1 in Fig. 4.1 represents the scenario when one node is silently listening to the other node. States i and j are said to communicate with each other if they can be reached from each other. This communication is represented by i ↔ j. It follows that, if i ↔ j and j ↔ k, then i ↔ k. The Markov chains in which

Fig. 4.2 State transition phenomenon in R2V communication

it is possible to go to every state from every state are called Ergodic Markov chains (Grinstead and Snell 2007).

4.1.2 Markov Process in R2V Communications

As mentioned in the previous section, a Markov model represents a system that makes transitions between a set of states. The main interest in using Markov models in vehicular communications is that their characteristic of state transition can be used to depict a vehicle entering and exiting the AP footprints. The vehicle can be shown to make transitions between two states, usable (when the vehicle is under the AP footprint) and disconnected (when the vehicle is away from the AP footprint). When a vehicle establishes a connection with a WLAN AP, it is said to have transitioned to the usable state from the previous (disconnected) state. When it moves out of the AP footprint, it is considered to have transitioned to the disconnected state from the usable state. Figure 4.2 shows a vehicle transitioning between the two states while moving from one place to another.

While the state transition characteristic of Markov models is directly relevant to vehicular communication, it is not clear whether Markov property holds in this context. Markov property may not hold in certain cases, for example in areas where AP deployment is sparse making it unlikely to encounter two APs together. Nevertheless, this book assumes that the Markov property holds and gives a limited insight into model accuracy toward the end of Sect. 5.5.2. This assumption is in line with the modeling approach adopted in (Yen and Yang 2006). Note that this Markov model 4.1 Markov Models 61

Fig. 4.3 2-state Markov model representing R2V communication

serves as a foundation to develop a hidden Markov model (HMM) in the next chapter. The generality, limitations, and need of the model have been highlighted at the end of Sect. 5.2.3.

Figure 4.3 shows the proposed Markov model which represents this R2V communication. In essence, the motion of vehicle is modeled by a 2-state Markov model in which the mobile node makes transition to one of the two states based on certain transition probabilities. Figure 4.3 also shows all possible transition probabilities for the 2-state model. Note that Puu (or Pdd ) denotes the probability of re-entering the usable (or disconnected) state at t + σ while starting from the same state at t, where σ is the time period after which the model fires the state transition. The other two probabilities Pdu and Pud are called the crossover probabilities because the system "crosses" over to a different state with these probabilistic values. The crossover probability Pdu (or Pud ) denotes the probability of making transition from the disconnected state to the usable state (or from usable state to the disconnected state). The mathematical definitions of Pdu and Pud are given in Eqs. (4.6) and (4.7) respectively.

$$P(X_{n+1} = usable | Xn = dis) = P{du}$$ (4.6)

$$P(X_{n+1} = dis|Xn = usable) = P{ud}$$ (4.7)

Pdd , Puu, Pdu and Pud are conventionally represented in the matrix form as shown in Eq. (4.8), where PT is the state transition matrix.

$$PT = \begin{bmatrix} P{uu} & P{ud} \ P{du} & P_{dd} \end{bmatrix} \tag{4.8}$$

The proposed is an event-driven model that fires every second (σ = 1) based on the event of an AP hit. The model shows the mobile node as making transitions at σ = 1 second interval based on whether it is under AP footprint. One second interval is chosen because the software used in data collection records AP information after every single second. Upon encountering an AP, it makes a transition to the usable state otherwise to the disconnected state. If the mobile node enters an AP footprint at a certain time, the probability Puu shall increase every second until the periodic beacon messages from that AP cease to reach the mobile node. As soon as this happens, the mobile node makes transition to the disconnected state with probability Pud . The mobile node stays in the disconnected state until it receives a beacon signal from some other AP. Consequently, the probability Pdd keeps increasing every second until a new AP is encountered. Similarly, as long as the mobile node keeps listening to the beacon signals from an AP, it stays in the connected state and the probability Puu keeps increasing. The next section details the evaluation of transition probability PT = {Puu, Pud ; Pdu, Pdd } for the Markov model representing R2V communication.

4.2 Estimating the Transition Probability

4.2.1 Data Collection

The calculation of PT requires the use of probability and statistical tools on a given set of data. Data acquisition is key to proposing effective and accurate mathematical models. The data may be collected by simulation or via experiments. Simulations are less labor intensive and cost effective but do not guarantee the level of credibility that comes with the experimental data. In some cases, various organizations assume the role of data collection and make the acquired data available to the researchers for further analyses. These data may also be made available to the public databases for a more widespread use. However, if the data are not already available, the process of data collection precedes the probability calculations. This book sets out to measure disruption in different geographical areas. Disruption in a certain area depends on the AP placement in that area. Therefore, the information on APs in the areas of interest needs to be collected first. This information is collected by conducting drive tests, which is used to train the model. The resulting model is used (in Chap. 5) to measure disruption. This section uses the data obtained from the drive tests reported in Table 3.8 of Chap. 3. The same has been shown here in Table 4.1. Recall from Sect. 3.3.2 that the data on AP hits were collected by driving through a certain area at high and low vehicle speeds.

It can be seen from Table 4.1 that the mean encounter time is almost one minute at both traffic patterns. However, the standard deviation in the high speed tests is very large. The reason for observing a higher standard deviation is that several AP encounters in the high speed tests lasted for less than 1 second. Figure 4.4 shows the encounter times between the mobile nodes and the APs observed during the drive tests. It can be seen from Fig. 4.4a that between encounter number 26 to 42, 10 encounters did not even last for one second. These encounters are the main reason for observing a high standard deviation in the high speed dataset. No such observation was found for the low speed tests as can be seen in Fig. 4.4b. The steady-state model

Speed	25-30 km/hr	45-50 km/hr
Total No. of APs	65	52
No. of Open APs	43	32
Mean encounter time (sec)	57.3	55.76
Median encounter time (sec)	57	57
Standard deviation (sec)	1.41	9.21

Table 4.1 Statistics representing encounter times between a mobile node and an AP

Fig. 4.4 Fluctuations in the encounter times for high speed and low speed datasets a Encounter times at 45–50 km/hr b Encounter times at 25–30 km/hr

proposed in this chapter is based on the number of encounters and their duration. The impact of vehicle's speed on the model parameters is not addressed here. However, Sect. 5.5 describes the variations in the intermittency of 802.11-based vehicular communications with changing traffic patterns.

The following section details the method that is conventionally employed to calculate PT from the available data. It also shows that the assumptions made for probability calculations with regard to data distribution are not suitable for R2V communications.

4.2.2 Probability Distribution of Dataset

The transition probabilities are generally computed on the assumption that the given dataset is exponentially distributed. The works reported in (Stern et al. 1994) and (Tsankov et al. 2007) assume that the dwell time in the steady states of the Markov model are exponentially distributed. Therefore, the general convention is to use the exponential relations such as one given in Eq. (4.9) to compute the transition probabilities.

$$P_{s_1,s_2} = 1 - e^{-\tau/t_1} (4.9)$$

where, *Ps*1,s² is the transition probability from state 1 (*s*1) to another state (*s*2) when the system spends time t¹ in s1 with a slot duration τ (Goodmannd Wei 1989).

Before using Eq. (4.9) for calculating PT , a brief analysis is required to see whether the dataset obtained from the drive tests follows exponential distribution. Drawing a probability plot is a subjective way of determining whether a dataset follows a certain distribution. Kvam and Vidakovic (2007) have debated the feasibility of using probability plots in comparison with the analytical methods to examine the distribution of the dataset. The probability plot for the dwell times in the low speed test is sketched in Fig. 4.5. The exponential and Poisson fits are also plotted alongside the dataset. It can be seen from Fig. 4.5 that the probability plot deviates significantly from the exponential and Poisson's fits. Similar observations were also made for the dwell times observed during the high speed tests (Hasan et al. 2010a).

While Fig. 4.5 conforms to a smaller dataset containing 65 APs, Fig. 4.6 has been plotted to analyze a larger dataset. This dataset contains 785 APs and is formally introduced later in Sect. 5.5. Because Fig. 4.6 plots more samples it is comparatively smoother and provides a better graphical representation (Hasan et al. 2010a). It is interesting to note that the dataset approximates exponential distribution in the

Fig. 4.6 Probability plot for a comparatively larger dataset (later used in Sect. 5.5)

Table 4.2 Difference in the observed and calculated median values for the dataset

Median	High speed	Low speed
Observed	57	57
Calculated	38.64	39.71

beginning but shows deviation toward the ending tail of the plot. This has also been observed for a dataset that encountered 274 APs.

Equation (4.10) gives the relationship between mean and median of exponentially distributed data (Panik 2005).

$$Median = \frac{ln(2)}{\lambda} \tag{4.10}$$

Note that the median values calculated from the datasets (reported in Table 4.1) are quite different from those computed using Eq. (4.10). Table 4.2 shows the observed and calculated median values for the dataset. It can be seen that the property given in Eq. (4.10) does not hold true for the collected dataset.

In light of these observations, the exponential relations are not used in calculating the transition probabilities for the R2V communication model.

It may be argued that because the AP encounters during a drive can be seen as "events", the entire process may be modeled as a Poisson process. In order to verify this, Poisson fits are also shown in Fig. 4.5. The figure shows that the collected dataset does not follow the Poisson distribution as well. Additionally, note the following properties of Poisson processes that are not found in R2V communication setup.

First, in a Poisson process, the arrival of the events depends only on the length of the time interval for which the events are observed. Applying this characteristic to the R2V communication scenario infers that the longer the drive time the larger the number of observed AP hits. More specifically, Poisson processes have stationary increments (Ross 2002). In typical R2V communications, several other factors apart

from the length of the drive time affect the AP hits, such as the vehicle speed, AP population, the location of APs, etc. Therefore, the length of time interval is not the only factor affecting the arrival of events in R2V communications.

Second, the events in a Poisson process do not arrive in batches (Viniotis 1998). In the context of R2V communications, this property infers that the mobile node shall encounter only one AP at a particular time instant. On the contrary, the drive tests have recorded multiple AP hits at the same time instants. Figure 4.7a and 4.7b shows that several APs were encountered at the same time instants during the high and low speed drive tests respectively (Hasan et al. 2011c). Encountering and communicating with more than one APs at the same time is termed as AP diversity. Exploiting AP diversity results in improved end-user experience (Balasubramanian et al. 2008). It follows that the AP hits can occur in batches in the R2V setup, which is not the characteristic of a Poisson's process.

Finally, the interarrival time of the events in a Poisson process is exponentially distributed (Bernstein 1999). The probability plots of the interarrival times for the AP hits observed at low and high speeds have been shown in Fig. 4.8a and Fig. 4.8b respectively. As the name suggests, the interarrival time is the time interval between

Fig. 4.8 Interarrival times for low and high speed datasets a Inter-arrival times at 25–30 km/hr b Inter-arrival times at 45–50 km/hr

two successive AP hits. It can be seen from Fig. 4.8a and 4.8b that interarrival times are not exponentially distributed (Hasan et al. 2010a). As all three properties of a Poisson's process do not hold true for R2V communications, the Poisson's distribution shall not be used to compute transition probability.

4.2.3 Calculating Transition Probability

It has been shown in the previous section that the exponential relations cannot be used for the calculation of transition probability for R2V communications. Therefore, the fundamental definition of probability given in Eq. (4.2) is used to calculate transition probability PT . The probability Pi j , for instance, is calculated first by counting the

Fig. 4.9 Steady-state Markov models for low and high speed datasets a Model for low speed test b Model for high speed test

number of transitions made from state i to state j, and then dividing the number of these transitions with the total transitions made from state i (Jurafsky and Martin 2009). This value is the estimated probability of making transition from state i to state j. Note that $P_T$ is the transition matrix which is composed of state transition probabilities from one state to another, such that $PT = {P{ii}, P{ij}, P{ji}, P_{jj}}$ .

All parameters of the transition matrix for R2V communications are calculated offline in this manner from the obtained datasets. The resulting transition matrices for low speed $(P_T^{ls})$ and high speed $(P_T^{hs})$ tests are given in Eqs. (4.11) and (4.12) respectively.

$$P_T^{ls} = \begin{bmatrix} 0.983 & 0.016 \ 0.027 & 0.972 \end{bmatrix} \tag{4.11}$$

$$P_T^{hs} = \begin{bmatrix} 0.984 & 0.015 \ 0.031 & 0.969 \end{bmatrix} \tag{4.12}$$

Equation (4.11) and (4.12) represent the Markov models for one area at different vehicle speeds that have been shown in Fig. 4.9a and 4.9b. Similar Markov representations for two other areas, namely area-2 ( $ar_2$ ) and area-3 ( $ar_3$ ), are given in Eqs. (4.13) and (4.14) respectively. The comparison of state transition matrices of these three areas reveals that each area is expected to have its own transition matrix and hence defines a Markov model of its own. This is because the models are calculated from the datasets that vary from area to area. Consequently, the evaluated model parameters are also different for different areas.

Table 4.3 LTER values for

Table 4.3 LTER values for different areas		LTER (%)
	Area - 1	36.36
	Area - 2	63.99
	Area - 3	70.9

$$P_T^{ar_2} = \begin{bmatrix} 0.720 \ 0.279 \ 0.156 \ 0.843 \end{bmatrix} \tag{4.13}$$

$$P_T^{ar_3} = \begin{bmatrix} 0.592 & 0.407 \ 0.167 & 0.833 \end{bmatrix} \tag{4.14}$$

It can be seen from Fig. 4.9 that the crossover probabilities of both models is very small. That is to say, the model has a higher tendency to stay in its current state than to move into the other. The calculated transition probabilities can now be used to evaluate one single parameter that can highlight the inherent disruption available in a certain area.

4.2.4 Long-Term Error Rate

The Long Term Error Rate (LTER) is one parameter that can represent the amount of disruption in a certain area. LTER quantifies the percentage of information sent in error over a communication channel. In (Vergetis et al. 2006), LTER has been used as a measure of packet loss over a rapidly changing wireless channel. In this book, LTER implies the intermittency of WLAN-based network services faced by the vehicles. An x valued LTER infers that the (x ∗ 100)% network services in a certain area shall be disrupted. It can be calculated using Eq. (4.15) (Vergetis et al. 2006).

$$LTER = \frac{P{ud}}{1 - P{dd} + P_{ud}} \tag{4.15}$$

where, Pdd is the probability of entering the disconnected state from the disconnected state and Pud is the crossover probability for transition from usable to the disconnected state.

The calculation of LTER requires Pdd and Pud which can be obtained from the Markov model. Note from Fig. 4.9a that Pud = P(qn⁺¹ = d|qn = u)=0.016 and Pdd = P(qn⁺¹ = d|qn = d)=0.972. Using these values, the LTER for the low speed model comes out to be 0.3636. This value of LTER suggests that 36.36% of the transmissions made in the R2V communication in this area shall be received in error. In other words, disruption in this area amounts to more than 35%. Note that LTER is a function of model parameters Pud and Pdd . The parameters (PT = {Puu, Pud ; Pdu, Pdd }) are calculated using the information on AP population in different areas. To cut short, LTER depends on the AP population available in a certain area. Since the AP population varies from area to area, LTER also varies accordingly. This variation in LTER infers that differences in AP population and their distribution leads to different levels of disruption in R2V communications. Table 4.3 lists the LTER calculated for three different areas. LTER calculated for these areas can give a comparative evaluation of the available intermittency. For instance, it can be inferred from Table 4.3 that Area 1 can offer R2V communications with lesser disruption in comparison with the other two areas.

The models given in Fig. 4.9 rely only on one parameter, the transition probability PT , for quantifying disruption. These models can be improved by incorporating more characteristics of R2V communication. The following modifications can be made in this model to better represent the R2V communication scenario.

In its present form, this model does not take into account the authentication scheme followed by the encountered APs. It is argued later in Sect. 4.4 that this model can be extended to a hidden Markov model to incorporate this information.
Presently, the model assumes that the mobile node stays in one of the two states. However, the following section shows that two states do not represent the R2V communication setup completely. This model can be modified by adding a third state. The next section introduces the third state and finally presents a 3-state Markov model.

4.3 3-State Markov Model

This section is concerned with adding a third state to the proposed model. The model developed in the previous section consists of two states, usable and disconnected. In the usable state, the mobile node is within an AP footprint and gets ample time to connect to the AP. The disconnected state is one in which the mobile node is out of the AP footprint so the network usage is not possible. The connected state is added to the model as the third state in which, the mobile node enters the AP footprint but exits before it completes the handover procedure.

The handover procedure has been previously introduced in Sect. 1.3.2 as the process of connecting to a new AP after leaving the footprint of the previous AP. Recall that in the connected state, vehicle does not exchange any meaningful information because it exits the AP footprint before establishing a usable connection. Very often a mobile node exhibits this behavior because of large handover latencies. The undesired handover delay is often larger than the time a mobile node spends within the footprint of the AP, consequently, it leaves the AP footprint without completely connecting to it (Eriksson et al. 2008). Figure 4.10 1 shows one example scenario in

¹ For the rest of the discussion, usable state is represented as state-1, connected as state-2, and disconnected as state-3. These notations are also used interchangeably.

Fig. 4.10 Transitions of a mobile node between usable, connected, and disconnected states

which a mobile node remains in the footprint of AP-B for 8 s and gets out of range before the connection could be established. This state in which the mobile node fails to establish a connection because of high handover latency is referred to as state-2 or the Connected state. While calculating the transition probabilities, it is assumed that the mobile node is in the connected state whenever the encounter time with an AP is less than 10 s. This is because the mobile node requires around 10 s to connect to an AP (Mishra et al. 2003). In all encounters lasting less than 10 s, it is almost impossible for the mobile node to establish a usable connection with the AP.

Therefore, the complete Markov model represents a vehicle as making transitions between 3 states throughout the length of its commute based on AP encounters. The 3-state Markov model is shown in Fig. 4.11. Note that all states in the model communicate with each other and are directly accessible. This infers that the model does not involve m-step transition to reach any of the three states. This Markov model is therefore an ergodic Markov chain.

The model discussed so far only uses the AP hit information to trigger the state transitions in the model. However, the transition to the usable (or connected) state also depends on the authentication information of the encountered APs. To incorporate the information on authentication scheme of the AP in the model, it needs to be extended from Markov model to the hidden Markov model. In principle, the states of the model remain the same, however, observation symbols are added to extend it to a hidden Markov model. The next section discusses these extensions.

Fig. 4.11 3-state Markov model representation of R2V communications

4.4 Toward Hidden Markov Model

A hidden Markov model (HMM) comprises a hidden and an observable probabilistic process. In the HMM proposed in the next chapter, the three states usable, connected, and disconnected constitute the hidden part of the HMM. The state of the vehicle being in the usable or connected, for example, depends on the amount of time it spends within the AP footprint. Suppose an encounter starts at time t and finishes at time t + n. The time period n, which is available only after the encounter finishes, shall determine whether the encounter was usable or not. Therefore, the state of a vehicle remains "hidden" until the prevailing connection comes to an end. This is why it is reasonable to consider the three states as the three hidden states of the HMM.

The second probabilistic process of an HMM is directly observable. The HMM proposed in the next chapter uses the authentication information of the encountered APs as the observable process. The WLAN APs have been classified as open or closed based on their authentication scheme earlier in Sect. 3.3.2. A mobile node encountering an AP may still not transit to states 1 or 2 if the encountered AP is closed. This infers that a more realistic representation of R2V communication must take into account the authentication information of the AP, because, whether the mobile node makes transition to state 1 or state 2 depends on the authentication scheme followed by the AP. The steady-state model presented in this chapter can be extended to the hidden Markov model (HMM) by incorporating the authentication information of the AP as the observable symbols. The observable symbols along with hidden states yield an HMM representation of R2V communications. The notion of hidden states is relevant here because the state of the mobile node remains hidden until some information about the observation symbol is revealed. Hence, a mobile node transits to the state-1 (or state-2) only when it observes an open AP. Without the proper knowledge of observable symbols, it is impossible to know the present state of the mobile node. Based on the authentication information, three observation symbols have been used. These are open, which represents the encounter of a mobile node with an open AP, closed, which represents the encounter of a mobile node with a closed AP, and no AP which represents the case that no AP has been encountered.

The disconnected state and the No AP observation might seem synonymous terms for some readers; however, there is a distinct difference between the two. Being disconnected represents a state in which no AP is observed. In this sense, being disconnected is a consequence of observing no APs. Moreover, there were AP encounters experienced during the experiments that lasted for less than 1 second. Several AP encounters started at t = t¹ and finished at t = *t*1+1 second. In these cases, while the observation is not No AP, yet the state is disconnected because the encounter simply did not begin. The difference between disconnected state and no AP observation shall become clearer when we formally define the notion of observation sequence in the next chapter.

4.5 Summary

This chapter is concerned with mathematically modeling R2V communications using Markov models. The data collected from the drive tests were used to evaluate the model parameters. It has been shown that the dataset in 802.11-based R2V setup does not follow exponential distribution. The state transition matrix was therefore calculated using the fundamental probability definitions. Using the developed models, one single parameter that can reflect on the available disruption was introduced as LTER. To this end, R2V communication has been given a Markov model representation. The next chapter extends this model to HMM and uses the same to measure disruption in a more complete manner.

Chapter 5 Measuring Disruption in R2V Communications

Vehicular communication requires continuous connectivity between a vehicle and an 802.11 AP. Due to the random nature of AP deployment, there will always be some inherent disruption that eventually limits the use of 802.11-based vehicular communication. In order to prevent disruption from affecting vehicular communication experience, different disruption tolerant algorithms have been proposed in various previous works. However, to the best of authors' knowledge, no method of measuring disruption is available in the literature. Measuring disruption requires a complete mathematical representation of an R2V communication scenario. Although the previous chapter gives an elementary Markov model, it needs to incorporate more detail regarding R2V communication to better reflect the characteristics of this scenario. This chapter takes into account the authentication scheme of the APs and modifies the Markov model to a hidden Markov model (HMM). It has been shown in this chapter that the developed HMMs can be used to compare and analyze the connectivity patterns of two different geographical areas in terms of disruption. The discussion begins with a brief introduction to the HMM and its associated notations and terminologies. The notation described in the next section shall be used throughout the chapter (unless stated otherwise).

5.1 Hidden Markov Models

The HMMs are used in modeling various stochastic processes. As opposed to the deterministic processes, the stochastic processes are probabilistic in nature. There is always some uncertainty in the way a stochastic process may evolve in future. This characteristic makes them difficult to model and predict over a certain period of time. While HMM can be used in analyzing a variety of stochastic processes, their main application lies in speech and pattern recognition. A HMM has been defined by Rabiner (1989) as a double stochastic process, in which an underlying process is indirectly defined by another stochastic process. As an example of one possible application of HMM, the inference about the weather conditions can be made by observing a seaweed, instead of observing the weather itself. A soggy seaweed corresponds to a wet day while a dry one corresponds to a sunny day. Such a system that contains some hidden states (weather conditions) and some observable symbols (soggy or damp seaweed) can be modeled by using the HMMs.

A HMM (λ) is defined by the initial state distribution (π), the state transition probability (PT ), and the observation probability (Bj(k)) (Rabiner and Juang (1986)), collectively given as λ = (π, A, B). To comply with the conventional notation of the HMM, the rest of this chapter uses *A*ij as the state transition matrix (instead of PT ). The mathematical description of π, *A*ij, and Bj(k) have been given in Eqs. (5.1), (5.2), and (5.3), respectively.

$$\pi_i = P[q_1 = S_i], (5.1)$$

$$A{ij} = P[q{n+1} = j | q_n = i], \quad 1 \le i, \quad j \le N,$$ (5.2)

$$B_j(k) = P[O_n = k | q_n = j].$$ (5.3)

where, *A*ij represents the state transition matrix,

Bj(k) is the probability of observing symbol k from state j,

π is the initial state distribution,

qn denotes the state of the mobile node at time n,

qn⁺¹ is the next state of the mobile node, and

On is the current observation symbol (at time n).

Conventionally, a HMM is used to solve three problems, namely evaluation, decoding, and learning (Yang et al. (1997)). This book is concerned with the evaluation problem of the HMM which is generally solved using the Forward and Backward algorithm. The evaluation problem is concerned with the calculation of P(O|λ), where O is an observation sequence and λ is the HMM. In the subsequent sections, the R2V communication is first given a HMM representation, then the Forward algorithm is applied on the resulting HMM to evaluate the probabilistic measures of disruption.

5.2 HMM Representation of R2V Communication

This section estimates the parameters λ = {π, *A*ij, Bj(k)}, and comments on the generality and limitations of the model. Before proceeding with the calculations, the definitions of model parameters in the specific context of R2V communication have been given in the following section.

5.2.1 Model Structure

The formal definitions of the model parameters π, *A*ij, and Bj(k) have been given in the previous section. This section describes how these parameters are used to represent the characteristics of R2V communications.

Among the three model parameters, *A*ij has already been discussed and analyzed in the previous chapter (see Sect. 4.1.2). Recall that *A*ij represents the transition probability from one state to another, where the model comprises of 3 states, namely, usable (u), connected (c), and disconnected (d). The formal definitions and method of estimating *A*ij remains the same in this chapter. The structure of the state transition matrix *A*ij is given in Eq. (5.4).

$$A{ij} = \begin{bmatrix} P{uu} & P{uc} & P{ud} \ P{cu} & P{cc} & P{cd} \ P{du} & P{dc} & P{dd} \end{bmatrix},$$ (5.4)

where i and j denote the present and next state of the vehicle respectively.

For extending the Markov model given in Chap. 4 to a HMM, the parameters π and Bj(k) must be introduced and calculated. The initial state distribution (π) is defined as the probability of staying in a particular state at time t = 0. In case of R2V communication, the initial state distribution gives the most probable state of the mobile node at the beginning of a drive as shown in Eq. (5.5), where the vector element π^x is the probability of starting in state x. To elaborate this, a sample initial state distribution (πsam) has been given in Eq. (5.6). The πsam vector shows that the most probable state of a mobile node at t = 0 is "connected" with a probability of 75% and the least probable state is "usable" with a probability of 5%.

$$\pi = \left[ \pi^u \ \pi^c \ \pi^d \right], \tag{5.5}$$

$$\pi_{\text{sam}} = [0.05 \ 0.75 \ 0.2].$$ (5.6)

The observation probability matrix Bj(k) represents the probability of encountering k-type AP from state j, where k represents the authentication scheme of the encountered AP. For example, Bu(C) = 30 % shows that the probability of encountering a "closed" AP from the "usable" state is 30%. The structure of the observation probability matrix is shown in Eq. (5.7), where rows represent the three states, and the columns denote three observation symbols O (open), C (closed), and N (No AP). Note that N does not correspond to a specific authentication scheme. Instead, the N observation shows that the mobile node is away from the footprint of the (open and closed) APs. The structure of the envisaged HMM is given in Fig. 5.1 (Hansan et al. (2011a)).

$$B{j}(k) = \begin{bmatrix} P{u}(O) \ P{u}(C) \ P{u}(N) \ P{c}(O) \ P{c}(C) \ P{c}(N) \ P{d}(O) \ P{d}(C) \ P{d}(N) \end{bmatrix}. \tag{5.7}$$

Fig. 5.1 Structure of the proposed hidden Markov model

It can be seen from Fig. 5.1 that the three states are represented by circles while the transition probability from one state to another has been given on the branch joining the two states. The three observation symbols are represented by rectangles which have connections with the respective states.

The probability values given on the branch joining a state and an observation symbol is the observation probability of a symbol from that state. Note that the No AP symbol is joined only with the disconnected state and the remaining two symbols are joined only with the usable and connected states. For further illustration, it follows that:

$$P_d(O) = P_d(C) = P_u(N) = P_c(N) = 0,$$ (5.8)

$$P_d(N) = 1. (5.9)$$

Using these values, the observation matrix given in Eq. (5.7) modifies to:

$$B_j(k) = \begin{bmatrix} P_u(O) & P_u(C) & 0 \ P_c(O) & P_c(C) & 0 \ 0 & 0 & 1 \end{bmatrix}.$$ (5.10)

As can be seen from the observation matrix in Eq. (5.10), the first two entries in the last column and last row of the observation matrices are zero.

These 4 entries simply say that there is no possibility of being in the usable or connected state if No AP is observed (first two entries of the last column) and that

	Commercial	Domestic
Total No. of APs	196	274
No. of 802.11g APs	181	231
No. of Open APs	108	156
Drive time	24.5 min	25 min
Mean encounter time	14.92 s	14.74 s

Table 5.1 Data collected from drive runs in commercial and domestic areas

there is no possibility of being in the disconnected state if open or closed APs are encountered (first two entries of the last row).

$$B(i,3) = B(3,j) = 0$$ ; with, $i = j = {1, 2}.$ (5.11)

After defining the model parameters $\lambda = {\pi, A_{ij}, B_j(k)}$ in the context of R2V communication, their estimation is covered in the next section.

5.2.2 Estimating Model Parameters

In R2V communication, the mobile node has an equal likelihood of starting in any of the three states. The vehicle may or may not be within the footprint of an AP at the beginning of the drive. Therefore, it is safe to assume that all states have the same probability values in the initial state distribution:

$$\pi = [0.3333 \ 0.3333 \ 0.3333]. \tag{5.12}$$

The observations from fresh drive tests are used for the computation of $A_{ij}$ and $B_j(k)$ . The tests are conducted in two areas referred to as the commercial and domestic areas. These tests are performed in a similar manner as highlighted in Sect. 3.3.2. The observations from these drive tests are presented in Table 5.1. The probability values are computed using the fundamental probability definitions given in Eq. (5.13) and Eq. (5.14) (Jurafsky and Martin (2009)).

$$A_{ij} = \frac{\text{number of transitions from state } i \text{ to } j}{\text{number of transitions from state } i},$$ (5.13)

$$B_j(k) = \frac{\text{number of } k \text{ observations in state } j}{\text{number of observations in state } j},$$ (5.14)

$$B_u(O) = \frac{z1}{\sum{y=1}^3 z_x}. (5.15)$$

*A*ij has been calculated using the procedure highlighted in Sect. 4.2.3. A similar technique is used for calculating Bj(k) as well. To illustrate this with an example, Bu(O) can be calculated using Eq. (5.15), when the mobile node observes z¹ open APs, z² closed APs, and makes z³ No AP observations, all from the usable state. The remaining elements of state transition probability and observation probability matrices estimated from the data of two drive tests are given in Eqs. (5.16)–(5.19).

$$A_{ij}^{\text{dom}} = \begin{bmatrix} 0.6081 \ 0.1892 \ 0.2027 \ 0.1039 \ 0.5974 \ 0.2987 \ 0.1639 \ 0.1393 \ 0.6967 \end{bmatrix}, \tag{5.16}$$

$$B_j(k)^{\text{dom}} = \begin{bmatrix} 0.6214 & 0.3783 & 0\ 0.5640 & 0.4369 & 0\ 0 & 0 & 1 \end{bmatrix},$$ (5.17)

$$A_{ij}^{\text{com}} = \begin{bmatrix} 0.7015 \ 0.1493 \ 0.1493 \ 0.1765 \ 0.5588 \ 0.2647 \ 0.1186 \ 0.3390 \ 0.5424 \end{bmatrix}, \tag{5.18}$$

$$B_j(k)^{\text{com}} = \begin{bmatrix} 0.5821 & 0.4178 & 0 \ 0.5546 & 0.4452 & 0 \ 0 & 0 & 1 \end{bmatrix}.$$ (5.19)

Figure 5.2 shows only a selected portion of λ that represents πu, *A*ud, and Bu(O). *A*ud is defined as the probability of transition from the usable state to the disconnected state. It represents the probability that the vehicle shall move away from the AP footprint and get disconnected. On the other hand, Bu(O) is the probability of observing open AP from the usable state. Collectively, this portion of the model shows that the vehicle is within the footprint of an open AP at time n and moves out of the footprint at time n +σ, as shown in Fig. 5.3. The definitions of *A*ij and Bu(O) are given in Eqs. (5.20) and (5.21), respectively.

$$A_{\text{ud}} = P(qn = d | q{n-1} = u),$$ (5.20)

$$B_u(O) = P(O_n = \text{open}|q_n = \text{usable}), \tag{5.21}$$

where q denotes the state and On denotes the current observation of the mobile node. In summary, the state transitions in the proposed model depict the frequent entrances and exits of the vehicle from the AP footprints and the observation symbol signifies the authentication scheme of the encountered AP. This underlying concept of the model makes it applicable in all WLAN-based R2V communication scenarios. The same concept can be related with a few previous works as well. For instance, the frequent and difficult to predict periods of poor connectivity (called gray periods) introduced by Mahajan et al. (2007) may be seen as transition to the disconnected

Fig. 5.2 Describing initial state, transition, and observation probabilities

state in the context of this model. Instead of predicting the gray periods, however, this model quantifies the aggregate disruption within a particular drive. The proposed model is also comparable with Breadcrumbs (Nicholson and Noble (2008)) because it also uses the notion of state transition to represent vehicular motion. The state transition in (Nicholson and Noble (2008)) is fired when a vehicle enters/leaves a certain geographical area. In this model, however, the state transitions fire every $\sigma$ units based on event AP hit.

The developed HMMs for two areas, $\lambda^{\text{dom}} = {\pi, A_{\text{ij}}^{\text{dom}}, Bj(k)^{\text{dom}}}$ and $\lambda^{\text{com}} = {\pi, A{\text{ij}}^{\text{com}}, B_j(k)^{\text{com}}}$ , are based on experimental data and therefore can represent all 802.11-based communications in general. Nevertheless, it suffers from some limitations that may affect their generality. The following section explores the generality of the model along with its limitations.

5.2.3 Model Generality, Limitations, and Need

It may be argued that the model may become invalid upon the introduction of IEEE 802.11p and/or 802.11r standards. Since these standards allow quick connection with the AP, their introduction may render state-2 of the model useless. This shall inevitably invalidate the entire model. However, note that the connection establishment time between the AP and the mobile node cannot become negligible. Despite using the WAVE mode in 802.11p and fast reconnection mechanism in 802.11r, the probing and detection phase latencies will still be observed during the frequent handovers. The scanning phase in the 802.11p networks, for example, would require scanning 7 DSRC (Direct Short Range Communication) channels which incurs a certain nonzero latency. Therefore, even the use of 802.11p and 802.11r shall not

Fig. 5.3 Transition from the usable state to the disconnected state with probability $A_{\rm ud}$ at a firing rate $\sigma$

allow a usable connection with all encountered APs. In other words, the proposed model will stay valid even after the introduction of 802.11p or 802.11r. The second possibility that can render state-2 useless is if the vehicle always spends more than 10 s within the footprint of every encountered AP. This is not the case in the real world because several drive tests reported in this book have shown a number of encounters that lasted less than 10 s.

The main limitation of the model is that it cannot be generalized for use in all areas of the city in its present form. The model parameters, $A_{ij}^{\text{dom}}$ , $Bj(k)^{\text{dom}}$ , $A{ij}^{\text{com}}$ , $Bj(k)^{\text{com}}$ , essentially rely on information, such as the AP population, the location of APs, and the authentication schemes followed by the roadside APs, etc. Since all these factors vary from area to area, one single model cannot represent all areas in the city. Consequently, for analyzing R different areas of the city, different models $\lambda^r$ must be developed by using the procedure outlined previously, where r=1 to R. In spite of this limitation, note that only the transition probability $A{ij}$ and observation probability $B_j(k)$ matrices of the model would differ from area to area; the procedures outlined in the later sections to measure disruption remain generic and valid for all WI AN-based R2V communications

5.2.3.1 Online and Offline Calculations

It has been pointed out in Sect. 1.3.1 that disruption is one of the main challenges in 802.11-based vehicular communications. Since calculating or "measuring" disruption has not been rigorously addressed in the literature before, this book sets out to develop means to estimate the amount of disruption in a particular area. A naive approach toward measuring disruption is to calculate the periods for which a vehicle remains connected (or disconnected) while on the move. Measuring only encountered durations would not be sufficient because the encountered APs follow open and closed authentication schemes. Therefore, the information on encounter times and authentication schemes both must be considered to get a reasonable reflection on the available disruption. Note that both encountered durations and authentication information can be obtained from drive tests; however, their combined analysis is required to get meaningful results. In other words, entering and exiting the footprints of roadside APs and encountering open/closed APs are two probabilistic processes that need to be combined together for evaluating disruption. HMM is one tool that provides a framework to represent such a double stochastic process and is therefore used in this book.

Secondly, disruption is measured in terms of certain matrices that are derived by applying Forward algorithm on the model. For example, Sect. 5.4.3 uses P(O|λ) as a matrix to reflect on the available disruption. It is obvious that the calculation of such a matrix requires a HMM λ. HMM λ is composed of parameters {π, *A*ij, Bj(k)} which can be calculated either online or offline using the data pertinent to encounter duration and authentication information collected during drive tests. The online method keeps updating the model parameters {π, *A*ij, Bj(k)} based on the collected data as the vehicle moves from one place to another. The online method can measure disruption during the drive tests if the vehicle is able to repeatedly process the collected data. Online processing requires the application of forward algorithm on the collected data, which results in the matrices that reflect on the available disruption (discussed later in Sect. 5.4.1). It follows that the online calculation requires processing capabilities onboard the vehicle. Also note that the online method is computationally expensive because it requires updating, first the model parameters (*A*ij, Bj(k)) and then the results obtained from the Forward algorithm on a frequent basis. It has been mentioned in Sect. 4.1.2 that data is collected every second (σ = 1 s) during the drive tests. In order to calculate disruption in an online manner, the Forward algorithm shall have to be applied every second on the collected data. In the offline method, on the other hand, the data collection is performed during the drives only. The collected data can be analyzed after the drive test on a processor that need not be accommodated inside the vehicle. The online calculation, on one hand, gives rise to processing constraints (updating *A*ij, Bj(k), and executing Forward algorithm), and on the other hand has data storage limitations as well. The state transition matrix, for example, shall have to be stored onboard for online calculation. In the offline method, this can be calculated and stored at the end of the drive.

Finally, this book is concerned by estimating the overall disruption faced by a vehicle in a particular area over a given period of time. Even if disruption is calculated in the online manner, our interest is in analyzing the net disruption available in the overall drive. Such an aggregate value of disruption represents a complete drive that lasts for certain time period. The intermediate results on disruption can be obtained using the online method (if processing constraints are ignored) but are out of the scope of this book. Because of this reason, the computational complexity of the online method has not been discussed any further. In summary, we are interested in measuring disruption in fixed geographical areas, and therefore set out to use the offline method which yields results after the drive tests and does not require onboard processing and storage capabilities.

To this end, two HMMs for two areas of the city have been proposed. The generality and limitations of these models have also been discussed. The models λcom and λdom can now be used to measure disruption in vehicular communications. Measuring disruption using the HMM can be seen as solving the evaluation problem of the HMM. Solving the evaluation problem requires the application of Forward algorithm on the model λ for a given "observation sequence," O S^v of length v. Before proceeding with the evaluation problem, the following section gives an overview of the observation sequence of a HMM.

5.3 Observation Sequence of HMM

Rabiner (1989) has used the coin toss model to explain an observation sequence *O S*v. In that model, a coin toss experiment is conducted such that it is not directly visible to the observer. The person doing the experiment would not inform the observer about the actual actions involved in the experiment but would only tell the outcome of each coin toss trail. The observation in this case is a series of observations about an experiment that is not directly visible. An observation sequence from v tosses in such trails is expressed as:

$$OS_v = {O_1, O_2, O_3, ..., O_v},$$ (5.22)

$$OS_v = {H, T, T, H, ..., H},$$ (5.23)

where H and T denote heads and tails respectively.

In the context of R2V communications, an observation sequence is a set of observations made by the mobile node regarding the authentication schemes supported by the APs encountered during the drive tests. For example, *O S*v={O,C, O} is an observation sequence of length v = 3 which represents the back-to-back encounters of the mobile node with 3 APs following open, closed, and open authentication schemes, respectively. The observation sequence *O S*v={O,C, O} implies that the mobile node is experiencing continuous connectivity from the roadside APs, which is not the practical case in 802.11-based vehicular communications. In the practical case, the mobile node experiences frequent disruption periods identified by the No

Fig. 5.4 A scenario resulting in observation sequence {O, N,C}

Table 5.2 Selected events from the log of experimentation

No.	Event	Duration	Authentication
1	AP hit	12:06-12:13	Open
2	AP hit	12:19-12:26	Closed
3	AP hit	12:31-12:31	Open
4	AP hit	12:36-12:45	Closed

AP (N) observations between the O and C observations. Figure 5.4 shows a typical scenario which results in observing {O, N,C}. This observation sequence infers that the mobile node first encountered an open AP and then a closed AP with a disruption period in between (represented by N).

For the provision of continuous (or at least close to continuous) vehicular network services using WLANs, the number of N observations should be minimum. The observation N marks the beginning of the disruption period which grows every second until an O orC is observed. Therefore, the model depicts the disruption period of τ seconds as τ back-to-back N observations. The back-to-back N observations encountered during a drive are explained with the help of a log given in Table 5.2 and shown in Fig. 5.5. Table 5.2 and Fig. 5.5 represents a vehicle's commute during which it encounters 4 APs.

Figure 5.6 shows that the AP hit events are non-overlapping in time. The periods in between these events (e.g.: from 12:13 to 12:19 in Fig. 5.6) are the disruption periods. The observation of the mobile node during these periods is NoAP. The observation sequence for the events logged in Table 5.2 is {O, N,C, N, N,C}. It is interesting to note from Figs. 5.5 and 5.6 that event-3 does not even last for a single

Fig. 5.5 The drive during which the mobile node experiences the observation sequence {O, N,C, N, N,C}. The diagram also establishes that two consecutive No AP observations can be made during a drive

Fig. 5.6 Representation of the non-overlapping events recorded in Table 5.2

second. Although the mobile node has experienced an AP hit but because it did not last even for 1 s, it shall be treated as no AP observation N.

Two observation sequences having same observation symbols but placed at different positions infer different events. For example, the observation symbol {O, O,C} is not equivalent to {O,C, O} despite having the same observation symbols. To further elaborate this property, the probability of staying in the usable state for an

Fig. 5.7 State vector probability against the observation sequences having symbol N placed at different positions. The probability reduces as N assumes the right most position in the sequence

observation sequence O S^v of length v = 5 is calculated. This can be done by calculating the state vector1 given a λ and an *O S*v. From the calculated state vector, the probability of staying in the usable state given λ and O S^v is determined. The sequences (*O S*v) are such that they consist of four Os and one N observations arranged in different order. The probability values for all possible combinations given in Eq. (5.24) are calculated.

$$OS_v = {NOOOO, ONOOO, OONOO, OOONO, OOOON}.$$ (5.24)

The probability for NOOOO is calculated first, after which in all subsequent calculations, the observation N is shifted one place to the right in the *O S*v. The results are plotted in Fig. 5.7. Note that as N moves more and more toward the right most position, the probability of being in the usable state decreases. This is because the right most symbol of an observation sequence is the most recent observation made by the mobile node. The probability of being in a certain state will, therefore, depend more on the rightmost symbol of the observation sequence than others. This can be appreciated by noting the probability values given in Eqs. (5.25) and (5.26) for the commercial area HMM (λcom). Note that the probability of being in the usable state approaches 0 as soon as the most recent observation becomes N.

$$P(q_n = \text{usable}|O = OOOOO) = 63.55\%,$$ (5.25)

$$P(q_n = \text{usable}|O = OOOON) = 0.02\%.$$ (5.26)

This discussion on O S together with the description of HMM provides sufficient background to proceed with the calculation of probabilistic measures of disruption in the following section.

¹ The state vector probability is discussed at length in the next section.

5.4 Probabilistic Measures of Disruption

This section uses the HMM λ to solve for P(O|λ) given an observation sequence O S^v = *O*1, *O*2, *O*3, ..., O^v and the model λ = {π, *A*ij, Bj(k)}. It is conventional to use the Forward and Backward algorithm in the context of HMM, however, the Forward part of the algorithm is sufficient to solve the evaluation problem (Rabiner (1989)). Therefore, this section deduces the results from the Forward algorithm only and shows how the obtained results can reflect on disruption in R2V communications. In essence, the interest is in calculating the following two parameters:

The probability of staying in a certain state given a certain observation sequence O S^v = *O*1, *O*2, *O*3, ..., On at time n. This value is termed as the state probability which serves as the first measure of disruption.
The probability of observing a certain sequence given the model λ. This value is called the encounter probability which is used as a second measure of disruption.

Before proceeding with the formal definition and analyses of these measures, the following section gives the brief working of the Forward algorithm. For the sake of completeness, the Backward algorithm is covered in Appendix-A. Since the Backward algorithm is not used in this book, only a brief description of the same is provided in the appendix.

5.4.1 Forward Algorithm

The evaluation problem of the HMM can be addressed by computing the so-called forward and backward variables α and β, respectively (Koski (2001)):

$$P(O|\lambda) = \sum_{j=1}^{N} \alpha_n(j)\beta_n(j), \qquad (5.27)$$

where, O = {*O*1, *O*2, *O*3, ..., On} and state of the HMM is j.

The algorithm itself is well known in literature. The forward variables αn(j)represent the probability of generated observation sequence up to n (*O*1, *O*2, *O*3, ..., On) and the HMM λ staying in state j.

$$\alpha_n(j) = P(O = O_1, O_2, O_3, ..., O_n | q_n = j).$$ (5.28)

The probability of making transition from one state to another and observing a symbol is:

$$p{(0,1)} = \pi A{ij} O_1, \tag{5.29}$$

where π is the initial state distribution, *A*ij is the state transition probability matrix, and O¹ is the observation matrix for the first symbol. Note that O¹ does not follow the same structure as Bj(k) matrix. O¹ is a diagonal matrix in which the entries denote

Fig. 5.8 The Forward algorithm generates S P and E P given λ and O^v

the probability of observing symbol 1 from different states. For a 3 state model, O¹ becomes (FB-Algorithm (2011)):

$$O_1 = \begin{bmatrix} p_u(1) & 0 & 0 \ 0 & p_c(1) & 0 \ 0 & 0 & p_d(1) \end{bmatrix}, \tag{5.30}$$

where pu(1) is the probability of observing symbol 1 from state u.

For every new observation, the algorithm calculates the probability values, such that for *t*th observation:

$$p{(0,t)} = p{(0,t-1)} A_{ij} O_t. (5.31)$$

Every step yields a vector that represents the probability of transition to each state for the given observation sequence. The structure of this vector is similar to that of initial state distribution vector (π). Every entry in the vector corresponds to the probability of staying in a certain state. The *i*th entry of the vector determines:

$$p_{(0,t)}(i) = P(O_1, O_2, O_3, ..., O_t, X_t = x_i | \lambda).$$ (5.32)

The vector obtained for a given observation sequence is termed as the state vector and the probabilities of individual states given in the vector are called the state probabilities. The next section defines the state probability in detail and highlights its significance in measuring disruption.

The state vector given in Eq. (5.31) needs to be normalized so that its probability values add up to 1. For this purpose, the scaling factor c^s is introduced.

$$p{(0,t)} = c{s} p{0,t-1} A{ij} O_{t}. (5.33)$$

The product of the scaling factors at each step gives the probability of observing a sequence given λ (FB-Algorithm (2011)). For Eq. (5.33), product of c^s at each step is the probability of observing O = {*O*1, *O*2, *O*3, ..., Ot} given the model λ. Recall from the previous section that this value is termed as the encounter probability:

$$P(O_1, O_2, ..., Ot | \lambda) = \prod{i=1}^t c_s.$$ (5.34)

Table 5.3 State vector probabilities against selected observation sequences for the commercial HMM

O Sv	Calculated state vector	Most probable state
OOO	0.60276 0.39716	USABLE p = 60.27%
OON	0.00006 0	DISCONNECTED p = 100%
CCC	0 1 0.55013	USABLE p = 55.01%
	0.44977 0.00009

Figure 5.8 shows that the Forward Algorithm is applied on the observation sequence which yields the state probability and encounter probability. These two values are estimated and introduced as the probabilistic measures of disruption in the following sections.

5.4.2 State Probability

As mentioned in the previous section, the Forward algorithm generates a state vector which assigns probability values to each state. The probability of each state is termed as the state probability (S P). It assigns the highest value to the most probable state of the mobile node given an O S^v and λ. Predetermining the state of the mobile node can assist in various ways. For example, Breadcrumbs (Nicholson and Noble (2008)) is concerned by making predictions about the state of the mobile node to better schedule the network services.

Suppose that the state probability for a certain mobile node comes out to be S P = {0.3, 0.55, 0.15}for states *q*1, q² and q³ for a certain λ and *O S*v. The given S P vector infers that the most probable state of the mobile node is q² with a probabilistic value of 55%. In the context of R2V communication, the most probable state is connected given an O S^v and λ. Since λ corresponds to a particular area, S P also represent the state of the mobile node in certain areas. The comparison of S P vectors for two areas of the city, e.g. S P¹ = {0.8, 0.1, 0.1}for area-1 and S P² = {0.2, 0, 0.8} for area-2, can determine which area offers less disrupted services. Upon comparison of S P¹ and *S P*2, it can be said that area-1 offers less disrupted network services because the most probable state is usable with a probability of 80%. A similar analysis for more than one areas is also possible.

The state probabilities for the commercial area model (λcom) for different observation sequences of length v = 3 have been calculated using the Forward algorithm.

Table 5.4 State vector probabilities against selected observation sequences for the domestic HMM

$OS_v$	Calculated state vector	Most probable state
000	0.50439 0.49551 0.00008	USABLE p=50.43 %
OON	0	DISCONNECTED p=100%
	0
	1
CCC	0.38815	CONNECTED p=61.17%
	0.61171
	0.00013

The results are tabulated in Table 5.3. While there can be $v^3$ distinct combinations of an observation sequence of length v=3, only selected sequences are reported to make some meaningful comments. Similar computations on the domestic area model have also been done and the selected results have been recorded in Table 5.4. The rows of Tables 5.3 and 5.4 contain the observation sequence $(OS_v)$ , followed by the estimated state probability vectors (SP) and the conclusion about the most probable state of the mobile node. For example, for the observation sequence $OS_3 = {O, O, O}$ in Table 5.4, the calculated state vector is $SP = {0.50439, 0.49551, 0.00008}$ which infers that the mobile node shall be in the "usable" state with a probability of 50.43 %.

The observation sequence $OS_3 = {C, C, C}$ represents a mobile node's encounter with 3 consecutive closed APs in the commercial and domestic areas. For the commercial area, the most probable state given $OS_3$ is usable according to Table 5.3. On the other hand, for the same observation sequence $OS_3$ in the domestic model (see Table 5.4), the mobile node is in the connected state. Note that the same $OS_3$ results in different SP vectors for two different models. This observation suggests that the domestic area shall be inherently more disrupted as compared to the commercial area. Second, the performance difference between two areas can be mathematically expressed by noting the probability value of staying in the usable state given a particular observation sequence. It can be seen from Tables 5.3 and 5.4 that $(\lambda^{\text{com}})$ has a higher probability of staying in the usable state $(p_u^{\text{com}} = 0.602)$ than $\lambda^{\text{dom}}$ in which $p_u^{\text{dom}} = 0.550$ , for the same $OS_3 = {O, O, O}$ . It follows that the commercial area offers a higher probability of staying in the usable state $(p_u)$ , as compared to the domestic area. This mathematical comparison, $p_u^{\text{com}} > p_u^{\text{dom}}$ , suggests that the domestic area faces more disruption than the commercial area.

No. of APs	1	2	3	4	5	6	7
Commercial	0.3868	0.174	0.0799	0.0367	0.0169	0.0078	0.0036
Domestic	0.355	0.1573	0.069	0.0308	0.0134	0.0064	0.0026

Table 5.5 The probabilities of encountering i open APs in commercial and domestic areas

5.4.3 Encounter Probability

The probability of encountering i back-to-back open APs during a particular drive is termed as encounter probability $(EP_i)$ . The encounter probability $EP_i = 75\%$ for i = 5 infers that the mobile node shall encounter 5 back-to-back APs in a particular area with a probability of 75%. Such a measure can be effective because, for increasingly continuous network services, a mobile node must experience back-to-back AP encounters (Frangiadakis et al. (2007), Ott and Kutscher (2005)). Therefore, the higher the encounter probability, the lower is the amount of disruption in a particular area. Like SP, the comparison of two areas in terms of $EP_i$ can determine the least disrupted area.

$EP_i$ is calculated for the commercial and domestic areas, where i ranges from 1-7. $EP_i$ is evaluated from the scaling factors $(c_s)$ which normalize the forward variables at each step during the Forward algorithm. The probability of observing a given sequence is the product of these scaling factors (FB-Algorithm (2011)). In order to determine the encounter probability, the product of scaling factors are calculated for observation sequences for length v: 1-7 as shown in Eq. (5.35).

$$OS{v} = \begin{bmatrix} O{v=1} = O{1} \ O{v=2} = O{1}O{2} \ O{v=3} = O{1}O{2}O{3} \ \vdots \ O{v=7} = O{1}O{2} \dots O{7} \end{bmatrix}.$$ (5.35)

While Table 5.5 records the calculated encounter probabilities, Fig. 5.9 shows that $EP_i$ in 802.11-based R2V communications decrease exponentially as the number of consecutive AP encounters increase. This inverse relation is intuitive because the WLAN APs are not deployed to support continuous connectivity and hence the probability of encountering $i_1$ consecutive APs should always be less than encountering $i_2$ consecutive APs when $i_1 > i_2$ . Note that the curves for domestic and commercial areas shown in Fig. 5.9 converge with the increasing number of APs. Figure 5.9 shows, on one hand, that the mobile node in the commercial area has a slightly higher probability to encounter consecutive APs, while on the other hand, the disruption in terms of encounter probability becomes almost similar in both areas for higher number of back-to-back AP encounters.

^<sup>2

The symbol O in Eq. (5.35) denotes an observation symbol and not Open AP observation.

Fig. 5.9 Encounter probability for different numbers of consecutive APs

While this section has analyzed the performance of two geographical areas, the state vector and encounter probability can be useful in evaluating different disruption tolerant algorithms and schemes as well. If a certain algorithm can show high values for these parameters in comparison with another algorithm, it will have the ability to reduce disruption to a larger extent. Apart from comparing the DTN algorithms, these techniques can be useful in assessing different traffic patterns. The next section uses the encounter probability to measure disruption at two different traffic conditions in the same area.

5.5 Traffic Pattern Analysis

The impact of traffic patterns on vehicular communications has been addressed by other researchers. For instance, Ott and Kutscher (2004b) and, Gass and Diot (2010) report an inverse relation between the vehicle speed and the data transfer capacity. On the other hand, Cottingham et al. (2007) maintain that a higher vehicle speed leads to a larger data transfer. These works, however, do not address the changes in the intermittency of 802.11-based vehicular communications with varying traffic conditions. This section uses the techniques highlighted earlier in this chapter, to first develop two models for two different traffic patterns. Secondly, it measures encounter probability using each of these models to analyze the variations in disruption due to the changing traffic conditions.

		Number	Total delay (sec)
Normal	Traffic signals	3	59
	Bus stops	14	238
Dense	Traffic signals	3	50
	Bus stops	21	418

Table 5.6 Delay due to traffic signals and bus stops

Table 5.7 AP population encountered during the drive tests

	Normal	Dense
Total No. of APs	771	858
No. of 802.11g APs	416	451
No. of 802.11n APs	311	365
No. of 802.11a/b APs	44	42
No. of Open APs	377	417

5.5.1 Drive Tests

In order to estimate the model parameters two fresh drive tests in a new area are performed that represent two different traffic patterns. These tests are performed between two fixed locations and are referred to as the normal and dense tests. During both tests, a mobile node (Intel Centrino Advanced-N 6,230 wireless card on Intel Mobile Express Chipset running Windows 7) running Vistumbler is placed inside a public bus. The location of the mobile node inside the bus is same in both tests.

In the normal test, the bus faced lesser congestion on the roads and had to pick-up and drop-off lesser number of passengers on its route. The bus covered the same distance in about 23 min in the normal test while the same distance was covered in about 30 min in the dense test. This is because the bus had to travel with a slower speed because of relatively dense traffic conditions during this test. The delays incurred due to stoppages at the traffic signals and bus stops in both tests are given in Table 5.6. Note that the delay due to traffic signals and that due to the frequent stoppages at the bus stops is larger for the dense tests. Thus the normal and dense tests correspond to two different traffic conditions on the same route.

Table 5.7 summarizes the information on the encountered APs collected during the normal and the dense tests. It can be seen that a higher number of APs were encountered during these drives as compared to those reported in Chaps. 3 and 4. Note that more than 300 802.11n radio type APs were encountered during both tests. As highlighted earlier in Sect. 3.3.2, encountering considerable numbers of 802.11n APs is very encouraging for the 802.11-based vehicular communication because they support high data rates. Also note that about 49% of the encountered APs in the normal test and more than 48% of the APs encountered in the dense tests were open.

Fig. 5.10 Encounter durations for the normal and dense tests

All other APs employed WPA security protocols that are not reported separately in this book.

The encounter durations observed for the two drive tests are shown in Fig. 5.10. Note that Fig. 5.10 only shows 711 and 785 APs for the normal and dense tests respectively. This is because the rest of the APs were encountered more than once as the bus visited the same road segments twice on the return route. The repeated APs have been excluded from the analysis.

5.5.2 Variation in Disruption with Traffic Patterns

The methods outlined in Sect. 5.5.2 are used here to estimate the model parameters λ^r = {π, *A*ij, Bj(k)} such that r = r¹ for the normal tests and r = r² for the dense tests. Here, r¹ and r² denote different traffic patterns but not different areas. The initial state distribution π remains the same, while the state transition and observation matrices are calculated using Eqs. (5.13) and (5.14) respectively. The obtained model parameters are as follows.

$$\pi = [0.3333 \ 0.3333 \ 0.3333], \tag{5.36}$$

$$A_{ij}^{r_1} = \begin{bmatrix} 0.6471 & 0.1686 & 0.1843 \ 0.1256 & 0.5426 & 0.3318 \ 0.1180 & 0.1161 & 0.7660 \end{bmatrix},$$ (5.37)

$$A_{ij}^{r_2} = \begin{bmatrix} 0.6600 \ 0.2686 \ 0.0714 \ 0.2343 \ 0.6294 \ 0.1362 \ 0.1491 \ 0.1798 \ 0.6711 \end{bmatrix}, \tag{5.38}$$

$$B_j(k)^{r_1} = \begin{bmatrix} 0.5231 & 0.4766 & 0.00015 \ 0.4877 & 0.5120 & 0.00015 \ 0.00015 & 0.00015 & 0.9998 \end{bmatrix},$$ (5.39)

$$B_j(k)^{r_2} = \begin{bmatrix} 0.5184 & 0.4814 & 0.00015 \ 0.4815 & 0.5183 & 0.00015 \ 0.00015 & 0.00015 & 0.9998 \end{bmatrix}.$$ (5.40)

Recall that the Bj(k) elements Pd (O), Pd (C), Pu(N), Pc(N) were zero in all previous observation matrices (see Eqs. (5.17) and (5.19)). However, the application of Forward algorithm requires nonzero matrix values. Therefore, very small nonzero values have been assigned to these matrix elements for the sake of calculation. This holds true for all previous calculations concerning Bj(k) as well. Eqs. (5.39) and (5.40) give the observation matrices that have been used to calculate the encounter probabilities for λr¹ and λr² respectively.

It has been pointed out in Sect. 5.4.3 that disruption can be measured by estimating the value of E Pi for a particular drive test. The calculation of E Pi for λr¹ and λr² would therefore represent disruption in the normal and dense traffic patterns. The obtained results have been shown in Fig. 5.11. The differences in the values of E Pi shown in Fig. 5.11 show that the traffic patterns do affect the available disruption. Furthermore, higher value of E Pi for λr² infers that the drives done in the dense patterns shall be inherently less disruptive in nature. The primary reason for this is that the vehicle encounters more APs during the drives and spends more time within the AP footprint.

Fig. 5.11 Encounter probability for i consecutive APs, where i = 1 to 5

Validating this model and determining its accuracy are interesting future directions. Note that the previous works use Markov-based techniques to predict measurable quantities. For example, Nicholson and Noble (2008) forecast the grid location of a vehicle t seconds ahead of time. The accuracy of such a model can be determined by comparing the predicted location with the actual location. On the other hand, the models reported in this book estimate, for instance, the probability of encountering back-to-back APs in a certain area. Finding an effective way to determine the accuracy of such evaluations require rigorous research effort. A basic approach may be to count the consecutive encounters with n roadside APs from the drive test data. For ease of analysis, encounter probability $EP_2^{r1}$ and $EP_3^{r1}$ have been computed empirically from the data set without using the model and Forward algorithm. Number of encounters with n = 2 and n = 3 are counted and divided by the total number of encounters faced by the vehicle assuming that it does not exploit AP diversity. These empirical evaluations suggest that $EP_2^{r1} = 11.3\%$ (as opposed to about 10.97% in Fig. 5.11) and $EP_2^{r1} = 4.18\%$ (as opposed to about 4.4% in Fig. 5.11). This gives a general impression that the proposed model has reasonable accuracy. However, this gives some insight into the accuracy of model $\lambda^{r1}$ only. Similar tedious computations shall have to be performed on all data sets in order to comment on the accuracies of the previously developed models. Therefore, analyzing the accuracy of this model has been recognized as future work and is not discussed here any further.

Although this book is not concerned with reducing or tolerating disruption, it briefly explores the role of inter-operator roaming in reducing disruption. It can be argued that one method of reducing disruption is to open more and more APs to the vehicles. This shall intuitively increase the number of O observations in a certain drive. In order to make the closed APs accessible from the vehicles, a large scale roaming mechanism between network operators must be enabled. This would allow the vehicles to connect to all APs without having to register with each operator separately. WISPr offers one such architecture that allows inter-operator roaming.

The next chapter uses the techniques developed in this chapter to evaluate the benefits of using WISPr in R2V communications.

5.6 Summary

HMM representing 802.11-based R2V communications has been developed in this chapter, which has originally evolved from the Markov model presented in Chap. 4. The model parameters, λ = π, *A*ij, Bj(k) are calculated using the experimental data. In addition to outlining the calculation of this model, its generality and limitations are also discussed. The developed HMM can reflect on disruption in terms of two probabilistic measures, namely state probability (S P) and encounter probability (E P). These measures are estimated using the Forward algorithm given the model λ and an observation sequence *O S*v. A discussion on observation sequence has also been presented in this chapter.

The state vector determines the state of the mobile node in a particular area. It assigns probability values to each state with the highest probability states becoming the most probable ones. On the other hand, the encounter probability gives the probability of encountering a certain number of back-to-back APs in a particular area. The calculation of S P and E P, and their relevance in measuring disruption has been covered. The notion of encounter probability has been used to analyze the variation in disruption with the changing traffic conditions.

Chapter 6 Inter-ISP Roaming for Vehicular Communications

HMM $\lambda^r = {\pi, A_{ij}^r, B_i(k)^r}$ (where r denotes geographical regions such that r = 1to R) has been developed in the previous chapter. The developed model $\lambda$ gives mathematical representation of 802.11-based vehicular communications. It has been demonstrated that the model $\lambda$ can give a comparative evaluation of disruption in different areas in terms of state probability (SP) and encounter probability (EP). It has also been shown mathematically in the previous chapter that a high number of AP encounters corresponds to a larger EP and hence reduced disruption. The encountered APs are either open or closed. Recall from Sect. 3.3.1 that the open APs can be accessed by the walkup users but the closed APs require user credentials before allowing the network access. In order to increase the number of APs encountered during a drive (and hence increase EP), the closed APs should also be made available to the vehicles. This would simply increase the number of usable APs encountered during a drive. One way to make use of the closed APs is through the inter-ISP roaming. In a nutshell, inter-ISP roaming allows the mobile nodes to use the resources of the ISP they are not originally subscribed to. Therefore, inter-ISP roaming mechanism can open the previously closed APs to the vehicles, thus increasing the total number of usable AP encounters.

The need of mathematically evaluating the benefits of inter-ISP roaming in the R2V communication has been addressed in this chapter. The HMM developed in the previous chapter is modified to incorporate the characteristics of inter-ISP roaming. The modified model is then used to measure disruption in terms of $EP_i$ . A brief description of intra- and inter-ISP handovers and roaming is given in the following, which will help understanding the concepts used in this model.

6.1 Intra- and Inter-ISP Roaming

As the name indicates, Internet Service Providers (ISPs) are network operators that allow the user devices to connect to the Internet. In general, an end-user has to register with an ISP in order to use the Internet services. The ISPs offering wireless Internet

Fig. 6.1 The intra- and inter-ISP handovers occur frequently in vehicular communications

services are often termed as the Wireless ISPs (WISPs).1 The basic function of ISPs and WISPs is the same, except that WISP provides connectivity to a router (or an AP), which in turn provides wireless Internet services to the end-user devices. To deploy such wireless network, an end-user houses an AP, connects it to the ISP infrastructure on one end, and with the personal communication devices on the other. Since these APs have limited coverage region, the service provided by the ISPs is also limited to a particular geographical area. In order to use the APs over a larger mobility domain, the mobile nodes need to connect to the new APs as they are encountered on the move.

The process of connecting to new APs after moving out of the coverage region of the previous one has been introduced as handover in Sect. 1.3.2 Based on whether the new AP belongs to the same ISP or not, handovers are classified as the intra-ISP (or intra operator) and the inter-ISP (or inter operator) handovers. Figure 6.1 shows the inter-ISP handover (between AP II and AP III) and the intra-ISP handover (between AP I and AP II). These terms are discussed in the following for further clarity.

When a mobile node hands over to an AP that belongs to the same ISP as the previous AP, the handover is termed as the intra-AP handover. Since 802.11 APs are not used over a larger mobility domain, the 802.11 handovers are generally intra-ISP in nature. These are more commonly addressed for evaluating handover latencies and

¹ ISPs and WISPs are used interchangeably in this chapter.

Fig. 6.2 Inter-ISP roaming shall allow a more complete use of the 802.11 infrastructure by opening more APs to the vehicles

for proposing fast handoff mechanisms. Conventionally, an AP owner cannot access any other AP even if it belongs to the same ISP, without permission. The intra-ISP handovers allow the mobile nodes to connect to any AP that belongs to the same ISP. For instance, intra-ISP handover allows an end-user to use its neighbor's AP if both APs are registered with the same ISP.

The FON community (Asheralieva et al. 2009) is a practical example of intra-ISP handover scenario. The clients subscribed to the FON network are required to keep their APs open to other FON members. This way, any client having a FON subscription can use the network service even when it is away from its home network. However, to ensure secure transactions, the members are required to provide valid user credentials before the network access is granted. Unlike the conventional WLANs, the credentials assigned to the FON clients can allow network access via any AP as long as it belongs to the FON network. The financial incentives offered by different ISPs to their subscribers and the increased security measures encourage the members to open their resources to other members of the same club. Consequently, 802.11-based resource sharing is getting increasingly popular.

While intra-ISP handovers increase the network coverage for the mobile nodes, it still limits them to connect to APs that belong to one particular ISP. This limitation is not desirable in vehicular communications because a vehicle may encounter several APs belonging to different ISPs on the move. For example, if a vehicle encounters five APs during a drive, out of which only two belong to its subscribed ISP, the vehicle shall be allowed to use only 40% of the available infrastructure (see Fig. 6.2). This is a serious constraint with the 802.11-based R2V communications because it prevents the complete 802.11 infrastructure from being used in the extended mobility domain. Therefore, for smoother end-user experience, the inter-ISP roaming mechanisms must be explored.

Roaming refers to moving out of the home ISP coverage region while connecting to the network using some other ISP, often called the foreign ISP. Home ISP is one to which the end-user is subscribed to; all unsubscribed ISPs are termed as the foreign ISPs. The foreign ISP allows network usage to the roaming users if they have mutual understanding with the home ISP. This allows the mobile node to use the network services without having to register with a foreign ISP. Roaming is therefore an extension of the coverage area of an ISP by letting the users connect to the infrastructure that does not belong to the home ISP.

The inter-ISP roaming leads to inter-ISP handovers. The inter-ISP handovers occur when a mobile node switches between two APs belonging to two different ISPs. To allow such handovers, ISPs make mutual agreements to allow each others' subscribers to use each others' network infrastructure based on certain financial and technical terms and conditions (Reynolds 2003). This mutual resource sharing is beneficial for both because the ISPs can make financial benefits every time an unsubscribed user accesses their infrastructure, and the end-users can get continuous network services on the move without subscribing to more than one ISP.

The inter-ISP roaming requires a dedicated framework which can take care of the authentication, authorization, and accounting (AAA) issues for ISPs and their end-users (Basios 2005). Wireless Internet Service Provider roaming (WISPr) is one architecture that proposes detailed specifications for inter-ISP roaming in the 802.11 networks (Bing 2008). However, it is still in draft stages and has not been implemented yet. Recently, WISPr 2.0 update has been introduced by the Wireless Broadband Alliance (WBA) to the existing specifications. The upgrade allows seamless authentication between WiFi and other wireless technologies such as GSM, LTE, WiMAX, etc. While the new WISPr 2.0 update allows interoperability between different networking technologies, this chapter focuses only on WLAN-to-WLAN handovers.

The main motivation of analyzing WISPr is that it can increase the number of APs that are open to the end-users. This shall in turn increase the encounter probability and thus reduce disruption. The following sections first define WISPr, and then mathematically evaluate its role in improving encounter probability.

6.2 Wireless Internet Service Provider Roaming

WISPr allows more network operators (like FON) to come into agreement so that the overall wireless coverage is extended beyond the boundaries of a single ISP. The network architecture proposed in (Sun et al. 2005) makes a similar assumption that all operators shall come to an agreement to provide network services to each others' subscribers. WISPr can be seen as a higher level integration of ISPs to support inter-ISP roaming over a larger scale. If WISPr is implemented, all participating ISPs shall open their APs to all end-users that are registered with at least one of the participating ISPs. This increases the number of open APs encountered by the vehicle during a particular drive. Table 6.1 shows the distribution of open and closed APs encountered during the drive tests reported in Sect. 5.5.1 (see Table 5.7). It can be seen from Table 6.1 that WISPr allows the use of more than 50% of the infrastructure that is not accessible without inter-ISP roaming.

	Normal	Dense
Number of open APs	377	417
Number of closed APs	394	441
Age of closed APs (%)	51.10	51.39

Table 6.1 Distribution of open and closed APs in the normal and dense tests of Sect. 5.5.1

Fig. 6.3 Diagrammatic representation of WISPr architecture

It must be noted, however, that the service capacity of an AP may be a limiting factor in implementing WISPr in R2V communications (Zhang et al. 2007). Service capacity of an AP refers to its ability to handle a certain number of mobile nodes simultaneously without degrading the QoS. If the APs become open to all participating mobile nodes, they may have to serve more mobile nodes than usual. Some sort of scheduling algorithm may be required to prioritize network access if a WLAN AP gets too many mobile nodes within its footprint at the same time (Jiang and Vaidya 1999). The study of service capacity, although important, is beyond the scope of this book. However, this issue will come up again in Sect. 7.4.2 where AP performance is analyzed while it serves several indoor clients along with an outdoor mobile node.

6.2.1 WISPr Architecture

Some efforts in standardizing the architecture of WISPr have been made by the WiFi Alliance. An schematic of WISPr architecture has been shown in Fig. 6.3. Note that the individual entities in the WISPr architecture are connected through an AAA gateway which has not been shown in the figure. AAA gateway handles the AAA issues for the mobile nodes (Lee et al. 2002). The specifications of WISPr are flexible with regards to the use of intermediary shown in the architecture. If present, it liaisons

Fig. 6.4 The vehicle encounters APs that belong to different ISPs throughout its commute. WISPr allows the vehicle to connect to every encountered AP as long as the AP belongs to one of the participating ISPs

the handovers and roaming mechanisms for the operators. Figure 6.3 assumes that the home ISP of a mobile node covers a certain geographical region called area -h. As the mobile node moves out of area -h and enters a foreign area (area -f), it can still use the network services if the home and foreign ISPs are in agreement with each other. This should not give the impression that certain regions in a city are dedicated to certain ISPs. In fact, the APs are sparsely located throughout the city such that some APs belong to one ISP while others may belong to the others.

Figure 6.4 shows a typical scenario in which the vehicle encounters APs that belong to various ISPs spread across its commute. WISPr is concerned with binding these ISPs into agreements so that all these APs shall become accessible to a vehicle. According to the WiFi Alliance, WISPr has been assigned an Internet Assigned Numbers Authority (IANA) Private Enterprise Number (PEN) to enable roaming functionalities in the vendor devices. Some of the recommendations of the alliance are:

The networking components shall be WiFi and/or WiFi5 certified.
The authentication protocol throughout the roaming area shall be remote authentication dial in user service (RADIUS) (Feng 2009).

• New emerging technologies shall be considered in accordance with their potential relevance to WISPr.

The implementation of WISPr envisages the provision of 802.11 roaming which is similar in experience to that offered by the cellular networks. Consequently, the WiFi clients would join the clubs housing several ISPs and acquire the membership to use their resources. Such membership agreements are already underway at some places. In one such agreement, the member ISPs can authenticate all member mobile nodes and the mobile nodes are billed by the home ISP for the used service (Reynolds 2003). The trend of motivating small businesses, such as coffee shops and restaurants, to open their APs to a particular ISP is also getting popular (Tan and Bing 2003). The idea is to get hold of the available APs and make them available to the subscribers. This gives the sense of location-independent network services to the end-users. Using the already available infrastructure not only increases the coverage region of an ISP but also saves the extra deployment cost.

The subsequent sections show how the HMM developed in Chap. 5 can be modified to incorporate WISPr characteristics. The modified model is then used to evaluate the increase in encounter probability due to WISPr in the R2V context. Such an evaluation shall also reflect on the application of the developed HMM.

6.3 Modifications in HMM

The HMM λ = {π, Ai j, Bj(k)}, its calculation, use, and limitations have been discussed in detail in Chap. 5. This section modifies the model discussed in the previous chapter to represent the vehicular context in which WISPr has been implemented. Recall from the previous chapter that π represents the initial state distribution, Ai j is the state transition probability matrix, and Bj(k) is the observation probability matrix. The model in Chap. 5 (henceforth λold) comprises 3 states, namely usable, connected, and disconnected, and three observation symbols Open AP (O), Closed AP (C), and No AP (N). Since WISPr allows all encountered APs to become open, there is no need to have the closed AP observation symbol. Therefore, the modifications in the observation probability matrix of λold are desired to incorporate the WISPr characteristics. While λold uses 3 observation symbols, the model incorporating the characteristics of WISPr uses only 2 symbols, namely Open AP (O) and No AP (N).

The modified model λnew with the same number of states but only two observation symbols is shown in Fig. 6.5. The observable symbol from the usable and connected states is referred to as AP, 2 while the observable symbol from the disconnected state is No AP (Hasan et al. 2010c). This infers that:

$$P_u(AP) = P_c(AP) = P_d(No \ AP) = 1$$ (6.1)

² Symbol AP is used instead of open AP for the rest of the discussion.

Fig. 6.5 Modified HMM representation of R2V communication incorporating the characteristics of WISPr

Conversely,

$$P_u(No\ AP) = P_c(No\ AP) = P_d(AP) = 0$$ (6.2)

Note that only the number of observation symbols is reduced in λnew, the number of states remain the same. The initial state distributionπ and the state transition matrix Ai j are shown in Eqs. ( 6.3) and ( 6.4), respectively, while the modified observation matrix of the HMM has been given in Eq. ( 6.5).

$$\pi = \left[ \pi^u \ \pi^c \ \pi^d \right] \tag{6.3}$$

$$A{ij} = \begin{bmatrix} P{uu} & P{uc} & P{ud} \ P{cu} & P{cc} & P{cd} \ P{du} & P{dc} & P{dd} \end{bmatrix}$$ (6.4)

$$B{j}(k) = \begin{bmatrix} P{u}(AP) & P{u}(NAP) \ P{c}(AP) & P{c}(NAP) \ P{d}(AP) & P_{d}(NAP) \end{bmatrix}$$ (6.5)

The rows in the observation matrix represent the three states Usable, Connected, and Disconnected, and the columns represent two observation symbols AP and No AP. For further clarity, the probability of being in the disconnected state $(P(q_d))$ is given in Eq. (6.6), where $O_n$ represents the observation symbol at time n.

$$P(q_n = \text{Disconnected}) = \begin{bmatrix} 0 & \text{if } O_n = AP \ 1 & \text{if } O_n = NO & AP \end{bmatrix}$$ (6.6)

The following section uses this model $\lambda^r = {\pi, A_{ij}^r, B_j(k)}$ (where r = 1 to R) to comment on the effectiveness of WISPr in reducing disruption (in terms of encounter probability)

6.3.1 Effectiveness of WISPr

The encounter probability has been previously defined as the probability of encountering consecutive WLAN APs during a drive. The amount of disruption in R2V communications has an inverse relation with the encounter probability. Encountering consecutive APs would mean continuous network coverage and hence reduced disruption. While it is intuitive that the implementation of WISPr shall increase the encounter probability, the main interest is in expressing this gain mathematically. For this purpose, the encounter probability in the normal and dense tests (introduced in Sect. 5.5.1) is estimated using the model $\lambda^{\rm old}$ . The resulting encounter probability $EP_i^{\rm old}$ for i consecutive APs corresponds to the case when WISPr has not been implemented. In the second step, the encounter probability for the same tests is estimated using the modified HMM $\lambda^{\rm new}$ . The calculation of $EP_i^{\rm new}$ corresponds to the case when WISPr has been implemented. The difference in the two encounter probabilities shall mathematically describe the improvements brought by WISPr in R2V communications.

$$\pi = [0.333\ 0.333\ 0.333] \tag{6.7}$$

Using the the initial state distribution given in Eq. (6.7), the state transition matrices for the normal and dense tests given in Eqs. (6.8) and (6.9), respectively, and the observation matrix given in Eq. (6.10), the encounter probability $EP_i$ is calculated for r = nor, den and i = 1 to 5. The obtained results are shown in Fig. 6.6.

$$A_{ij}^{r=\text{nor}} = \begin{bmatrix} 0.6471 & 0.1686 & 0.1843 \ 0.1256 & 0.5426 & 0.3318 \ 0.1180 & 0.1161 & 0.7660 \end{bmatrix}$$ (6.8)

$$A_{ij}^{r=\text{den}} = \begin{bmatrix} 0.6600 & 0.2686 & 0.0714 \ 0.2343 & 0.6294 & 0.1362 \ 0.1491 & 0.1798 & 0.6711 \end{bmatrix}$$ (6.9)

Fig. 6.6 Encounter probabilities for the normal and dense tests with and without WISPr

$$B_j(k) = \begin{bmatrix} 1 & 0 \ 1 & 0 \ 0 & 1 \end{bmatrix} \tag{6.10}$$

Note that for $\lambda^{new}$ ,

$$B_j(k)^{r=\text{nor}} = B_j(k)^{r=\text{den}}$$ (6.11)

It can be seen from Fig. 6.6 that an increase of more than 25% is observed in the encounter probability for the normal test if WISPr is implemented. This is because more APs become accessible to the vehicle due to WISPr. The results given in Fig. 6.6 suggest that similar improvements are observed in the encounter probability for the dense test. Note that the values of absolute and percentage increase in encounter probability are different for the two tests as can be seen from Fig. 6.6. The percentage increase in $EP_i^{\rm den}$ is larger than that in $EP_i^{\rm nor}$ in the presence of WISPr. Second, note that the variations in $EP_i^{\rm den}$ with WISPr are comparatively smaller. It is interesting to note that the $EP_i^{\rm den}$ and $EP_i^{\rm nor}$ with WISPr have differences despite corresponding to the same geographical area. Since the initial state distribution and observation matrices in $\lambda^{\rm new}$ are same for both areas, it may be concluded that this difference is because of the state transition matrices of the two areas. A more rigorous investigation is required to assess the transition probability matrices in this context.

6.4 Summary

A modified HMM has been developed (based on the HMM developed in Chap. 5) to analyze the role of inter-ISP roaming in the vehicular context. A discussion on intraand inter-ISP roaming has been given first, which is followed by the introduction of a roaming architecture WISPr. It has been discussed that WISPr can increase the number of usable AP encounters and hence reduce disruption. This claim has been mathematically verified in this chapter using the modified model ( \lambda^{\text{new}} ).

Note that the intra- and inter-ISP handovers discussed in this chapter do not take into account the handover delay. Handover delay has been identified as an important issue in Sects. 1.3.2 and 2.2. Since the encounter duration between a vehicle and an AP is already very small, the delay in handing over to an AP should be as small as possible. The evaluation and reduction of the handover delay with specific regard to the R2V communication scenario has been reported in the next chapter.

Chapter 7 Handover Latency: Evaluation and Reduction

Handover has been introduced as the process of disassociating from the previous AP and establishing a connection with the new AP. In an ideal case, for getting continuous network services on the move, there should be no delay in handing over to an AP. This is because the network services remain suspended for the mobile node during the handover process. The amount of time in between initiating and completing a handover is called the handover delay.

The term handover latency is often used instead of handover delay. Latency is generally a packet-level terminology, which is the time difference between the moment a packet is sent and the moment it is received. Delay, on the other hand, usually refers to the undesired holdups in communication (due to queuing, congestion, packet loss, etc). This book uses handover latency and handover delay interchangeably.

7.1 Handovers in WLANs

In 802.11 networks, the mobile node starts searching for the new candidate AP when the need for handover is detected. The need of handover is generally detected when the RSS from the current AP falls below a certain threshold. The process of selecting a suitable AP to handover to is called scanning (or probing), which is the first phase of the handover process. Some works have also considered the detection phase as the first phase of the handover (Velayos and Karlsson 2004). Nevertheless, it is maintained in the literature that the scanning phase is the most time-consuming phase in the entire handover procedure (Mishra et al. 2003). Scanning can be of two types: active and passive. In active scanning, the mobile node sends request frames over all 802.11 channels and waits for the responses from the probed APs. Among the APs that respond to the mobile node's request, one AP is selected for handover based on certain selection criteria. In passive scanning, the mobile node spends a fixed amount of time (usually 100 ms) on every 802.11 channel, listening to the periodic beacon signals transmitted by the APs. This type of scanning is generally employed in most WLAN deployments.

In the second phase of the handover, the mobile node has to authenticate itself with the AP selected during the scanning phase. While WLAN APs can employ various authentication schemes such as OSA, PKA, SKA, etc, the extensible authentication protocol (EAP) is increasingly deployed in most commercial deployments (Chen and Wang 2005). Note that the authentication schemes used by the individual APs cannot be controlled. For example, if most APs deployed across an area of interest employ EAP for authentication purposes, this book requires to analyze EAP-based handover mechanisms. Because of the popularity of EAP authentication among the end-users, this chapter takes into account the handovers to the EAP-enabled WLAN APs. Most of the handover-related contributions reported in the literature do not take into account the EAP (or any other robust authentication mechanism). A brief description of the EAP authentication can be seen in Appendix B. After successful authentication, the mobile node gets associated with the AP by sharing its specifications, which is termed as association. The association process is often considered as a part of the authentication phase.

Finally, the mobile node attempts to get a unique address to communicate with the other nodes in the network. The process of acquiring a unique address is referred to as address acquisition/allocation. Addresses are generally assigned by the dynamic host configuration protocol (DHCP). These addresses are referred to as the IP addresses.

In summary, the overall handover procedure can be seen as the combination of various phases. The handover latency is therefore the sum of the delays incurred by n individual phases, discussed later in Sect. 7.2.2.

Handover Delay = $$\sum{i=1}^{n} \text{Delay}{i}$$ . (7.1)

This chapter focuses on reducing the handover delays in 802.11-based vehicular communications. It starts by measuring the delays in the handover procedure. The handover process is divided into its constituent phases and the delay in each of these phases is measured in the next section. Based on these measurements, the phases that contribute most to the handover delay are identified and discussed.

7.2 Experiments and Observations

This section covers the experimental measurement of the handover delay in the vehicular environments. Since this book focuses on using indoor APs from outdoors, such an investigation must precede the efforts in reducing the handover delay. The latency incurred in handing over in the vehicular setup may be different from that reported in the conventional delay analysis.

Fig. 7.1 Setup for handover latency evaluation in the vehicular environments

7.2.1 Measurement Setup

A mobile node (laptop with Windows Vista, Realtek driver, ATI chipset) placed outdoors is deliberately handed over to an indoor AP. The indoor AP is 802.11g radio type and is EAP enabled. The delays in the individual phases are evaluated by capturing live traffic using the Microsoft Network Monitor (Posey 2008). The mobile node is connected to a different AP on a different subnet at the beginning of the tests. The packet capture begins when it is handed over to the indoor AP. Hence the setup comprises two APs and a mobile node (running the network monitor) as shown in Fig. 7.1. The handover traffic is recorded in two different scenarios 10 times each. In the first scenario, the mobile node hands over while staying at a fixed outdoor location, and in the second, it moves from one place to another during the handover. The test area is an empty, unsheltered car park, adjacent to the building housing the AP. The signal strength recorded from the periodic beacon messages at the time of the tests ranged from −75 to −85 dBm.

7.2.2 Observations in Vehicular Environments

The indoor AP (to which the handover is made) employs open system authentication (OSA) in addition to EAP. The handover to such an AP comprises scanning phase, authentication phase (OSA and EAP) followed by association. The address allocation via the DHCP followed by the address resolution protocol (ARP) marks the successful completion of the handover. The entire handover procedure to an EAPenabled AP and the associated delays are shown in Fig. 7.2. Note that the handshake procedure shown for EAP is not complete and only serves to emphasize that the involved handshakes are between a mobile node, an AP, and remote authentication

Fig. 7.2 Handshake mechanism for handovers to EAP-enabled APs

Table 7.1 Delays measured in individual phases during a handover in stationary and low mobility scenarios (Hasan et al. 2010b)

Handover phase	Stationary	Low mobility
Scanning/probing (ms)	324.13	303.511
OSA (ms)	47.5	8.08
Association (ms)	5.4	10.18
EAP (ms)	974.94	Failed in 6 tests
DHCP	Completed only once	Never completed
ARP	Completed only once	Never completed

dial-in user service (RADIUS) server. RADIUS servers are commonly used to hold authorization information and to validate the network access for the end-users. When used with EAP, AAA protocol carries EAP messages between RADIUS server and the authenticator (Zheng and Sarikaya 2009). The DHCP handshakes have not been shown here because these shall be highlighted later in Sect. 7.3.1. Table 7.1 records the latency observed during each of these phases in stationary and low mobility conditions. Figure 7.3 gives a graphical representation of the delay measured from 10 observations each in the stationary and low mobility tests.

The OSA phase (very much like the association phase) does not cause considerable delays in both stationary and low mobility environments. Technically, OSA does not perform any authentication. The mobile node sends a MAC authentication frame to the AP, which is simply accepted to complete the OSA process

Fig. 7.3 Contribution of delay in the overall handover latency under stationary and low mobility conditions (Hasan et al. 2010b) a Handover latency under stationary conditions b Handover latency under low mobility conditions

(Chen and Guizani 2006). Since OSA does not involve rigorous handshakes, it does not add significantly to the handover delay. Note that the OSA delay in the low mobility environments is smaller than the stationary environments. One possible reason for this is that the mobile node was moved toward the AP during the low mobility tests and hence it might have received a better signal strength while approaching the building. In the low mobility tests, the EAP authentication phase failed several times and had to restart. This re-initiation of EAP due to frequent failures incurred additional delays that have not been shown in Fig. 7.3b. EAP delay is discussed in Sect. 7.3.2.

The ARP phase cannot be considered as a major delay contributor because its performance depends on the DHCP phase. ARP translates the IP addresses into the MAC addresses of the communicating devices. In a typical local network, ARP informs the sender about the MAC address of the receiver given its IP address. Therefore, completion of ARP requires a valid IP address, which is obtained during the DHCP phase. If the DHCP phase does not complete successfully, the ARP phase will also remain incomplete. Note that Fig. 7.3 also does not report the DHCP delay. The reason for this is that the DHCP procedure completed only once in all 20 tests (10 each in stationary and low mobility).

From the above discussion, it follows that three phases dominate all others in terms of contributing toward the handover delay. Therefore, the overall handover delay mainly depends on these three phases. The following section discusses these phases in greater detail.

$$Handover Delay = D_d + EAP_d + S_d$$ (7.2)

where, Dd is the delay in the DHCP phase, EAP^d is the delay in EAP authentication while Sd is the scanning delay. The observations on these phases are further analyzed in the following section to make some meaningful comments.

7.3 Latency Analysis

The observations reported in the previous section have shown that three phases, namely DHCP, EAP, and scanning are the largest delay contributors. The following sections discuss these phases separately.

7.3.1 DHCP Delay

The simulation results reported in (A-Helali et al. 2009) suggest that 17% of the handover delay can be attributed to the DHCP protocol for a traffic load of 80%. In the low mobility vehicular environments, the performance of DHCP is even worse. In the light of observations recorded in the previous section, the DHCP completed successfully only once in total 20 tests performed at stationary and low mobility conditions. This observation alone calls for drastic changes in the DHCP to allow faster address assignment in the vehicular context.

Fig. 7.4 The mobile node requests to use the previous IP address, which is denied using the NAK packet by the AP. The DHCP procedure starts when the NAK packet is issued by the AP

Legacy DHCP

The DHCP follows a four-way handshake mechanism to allocate an IP address to a mobile node. The procedure starts when the mobile node sends a broadcast message, DHCP discover (DISC), to request an IP address from the DHCP server. Upon receiving DISC, the DHCP server checks for the available addresses and offers one of them to the mobile node by sending a DHCP offer (OFFER) message. The mobile node accepts the offer by formally requesting the grant of this address using the DHCP request (REQ) message. The DHCP server acknowledges the successful assignment of the address with an ACK packet (Shinder et al. 2008). The DHCP handshake procedure has been shown in Fig. 2.8 in Sect. 2.2.2.

The lease for the assigned IP address is given for a fixed time period. After the time for which the lease was obtained finishes, the IP address stands invalid. The mobile node generally renews its lease with the DHCP sever before its present lease expires (Blank 2004). While the conventional DHCP procedure is thought to comprise only these processes (DISC, OFFER, REQ, ACK), the later part of this section reveals that the primary cause of DHCP delays are the processes that are often not considered during the DHCP evaluations and analyses. In particular, the NAK and INFORM packets add significantly to the DHCP delay. These packets are often not included in the DHCP-related analyses and hence remain largely unexplored.

The NAK packet informs the mobile node that the previously held IP address must be released and a new one must be acquired. When a mobile node hands over to a new AP, it tries to reuse the previously associated address on the next connection that it makes with another WLAN AP. If the next AP is on the same subnet as the previous AP, the mobile node may use the same IP address and the entire procedure of acquiring an address may be avoided. On the other hand, if the new AP resides on a different subnet, the DHCP procedure starts afresh and a new IP address is granted using the DHCP handshake mechanism. The new AP advises the mobile node that the previously held address is not usable by issuing an NAK packet. This has been graphically shown in Fig. 7.4.

A web browser on the mobile node generally issues an INFORM packet to get the address of ISA firewall (Shinder et al. 2005). The ISA server caches and keeps track of the network traffic for allowing faster access to the network. This way, the web browser automatically connects to the internet through the ISA firewall in a more secure way. The INFORM message may not be the part of the DHCP itself but is generally available and affects the handover delays.

Assessing Individual DHCP Processes

Note that the observations recorded in Table 7.1 report the overall performance of the DHCP in the vehicular setup. Additional tests are performed in both indoor and vehicular setups to analyze the performance of the individual DHCP phases (DISC, OFFER, REQ, ACK). The same tests also reflect on the performance of NAK and INFORM packets during the DHCP process. The test setups and observations are as follows.

In the indoor setup, the mobile node hands over to the AP from inside the building when it is in close proximity of the AP. This represents the usual use case scenario of the WLANs. The second set of tests is conducted in the vehicular setup. The mobile node is handed over to an indoor AP while it is placed approximately 32 ft away from the foot of the building housing the AP. The packet capture has been done using the Windows Network Monitor for both indoor and vehicular setups. The observations from the indoor tests serve as the base line values for comparison with the outdoor tests.

Figure 7.5 shows one log that is obtained during the handover in the indoor setup. The log shows that at the beginning of the handover, the mobile node sends a REQ packet, which is a request to re-use the previously held IP address. Since the mobile node was previously connected to an AP on a different subnet, the new AP rejects this request by issuing a NAK packet. This causes a delay of 2.15 s in sending the DISC packet, which should have been sent immediately. The rest of the packet exchange including DISC, OFFER, REQ, and ACK takes around 0.453 s. While the legacy DHCP procedure finishes in 0.453 s, the delay associated with the INFORM packet is 5.678 s and that due to NAK is 2.15 s. This delay needs to be reduced for faster IP address allocation.

Figure 7.6 shows the log of the handover in the vehicular setup during which the DHCP was able to assign an IP address to the mobile node. All connection attempts were not successful. Figure 7.6 reports the log of one of the successful address assignments. Although the address allocation was successful, the entire procedure took almost 15 s. This delay is intolerable in the vehicular environments where the mobile nodes only spend 10–15 s within the footprint of an AP. It is interesting to note that the time taken by the DISC, OFFER, REQ, and ACK handshakes was 2.87 s and the delay due to NAK was about 5.85 s. The NAK mechanism may be discarded because it is unlikely that a vehicle connects to several APs on the same subnet consecutively. The INFORM packet added a further delay of 7.26 s in the handover procedure. The observations for indoor and vehicular setups are summarized in Table 7.2. These delays have been shown in Fig. 7.7.

It can be seen that the performance of DHCP is poor in vehicular setup. The NAK and INFORM packets are largely responsible for the delay in the overall procedure. Since the reduction in the DHCP delay requires consideration of address allocation issues, this chapter does not address it any further.

Frame Number	Time Offset	Source	Destination	Description
130	0.811202	0.0.0.0	255.255.255.255	DHCP:Request, MsqType = REQUEST, TransactionID = 0x539E9813
148	0.842402	193.61.191.153	255.255.255.255	DHCP:Reply, MsqType = NAK, TransactionID = 0x539E9813
314	2.152804	0.0.0.0	255.255.255.255	DHCP:Request, MsqType = DISCOVER, TransactionID = 0x740AB32
344	2.277604	193.61.191.153	255.255.255.255	DHCP:Reply, MsgType = OFFER, TransactionID = 0x740AB329
349	2.293204	0.0.0.0	255.255.255.255	DHCP:Request, MsqType = REQUEST, TransactionID = 0x740AB329
402	2.605205	193.61.191.153	255.255.255.255	DHCP:Reply, MsgType = ACK, TransactionID = 0x740AB329
1228	8.283615	193.61.191.68	255.255.255.255	DHCP:Request, MsqType = INFORM, TransactionID = 0x8B33A87C
1243	8.314815	193.61.191.153	193.61.191.68	DHCP:Reply, MsgType = ACK, TransactionID = 0x8833A87C

Fig. 7.5 Log obtained while handing over to an AP from indoors (Hasan et al. 2011d)

Frame Number	Time Offset	Source	Destination	Description
63	0.312000	0.0.0.0	255.255.255.255	DHCP:Request, MsgType = REQUEST, TransactionID = 0xDC7A9DA7
891	4.524008	0.0.0.0	255.255.255.255	DHCP:Request, MsqType = REQUEST, TransactionID = 0xDC7A9DA7
909	4.617608	193.61.178.201	255.255.255.255	DHCP:Reply, MsqType = NAK, TransactionID = 0xDC7A9DA7
910	4.617608	193.61.178.201	255.255.255.255	DHCP:Reply, MsqType = NAK, TransactionID = 0xDC7A9DA7
1158	5.850010	0.0.0.0	255.255.255.255	DHCP:Request, MsgType = DISCOVER, TransactionID = 0xD395FC8
1322	6.567611	193.61.178.201	255.255.255.255	DHCP:Reply, MsqType = OFFER, TransactionID = 0xD395FC89
1323	6.567611	0.0.0.0	255.255.255.255	DHCP:Request, MsqType = REQUEST, TransactionID = 0xD395FC89
1757	8.720415	193.61.178.201	255.255.255.255	DHCP:Reply, MsqType = ACK, TransactionID = 0x81F85553
1758	8.720415	193.61.178.201	255.255.255.255	DHCP:Reply, MsqType = ACK, TransactionID = 0x81F85553
2161	10.732818	0.0.0.0	255.255.255.255	DHCP:Request, MsqType = REQUEST, TransactionID = 0xD395FC89
2170	10.764018	193.61.178.201	255.255.255.255	DHCP:Reply, MsgType = ACK, TransactionID = 0xD395FC89
2171	10.764018	193.61.178.201	255.255.255.255	DHCP:Reply, MsgType = ACK, TransactionID = 0xD395FC89
3260	15.958828	193.61.178.91	255, 255, 255, 255	DHCP:Request, MsqType = INFORM, TransactionID = 0x3A2B63E8

Fig. 7.6 Log of the handover that successfully issued an IP address in the vehicular environments (Hasan et al. 2011d)

Table 7.2 The delay contributions (in sec) by different phases of the DHCP procedure (Hasan et al. 2011d)

Setup	DISC-to-ACK	NAK	INFORM	Total delay
Indoor	0.453	2.15	5.678	8.3
Vehicular	2.8704	5.85	7.26	15.9

7.3.2 EAP Delay

It has been mentioned before that the scanning phase is the most significant contributor in the handover latency (Kim et al. 2006; Mishra et al. 2003). However, note from Table 7.1 that the scanning phase delay is almost one-third of the total delay caused by the EAP authentication phase. This infers that if robust authentication schemes are employed, the authentication delay becomes the largest delay contributor.

The large delay due to EAP is not unexpected because it comprises of several handshake processes between the applicant, authenticator, and the authentication server. The handshake mechanisms involved in the EAP process are classified as EAP

Fig. 7.7 The delay incurred by individual DHCP phases during the handover

Table 7.3 Delays (in msec) contributed by individual EAP processes during handovers in stationary and low mobility scenarios (Hasan et al. 2010b)

	Stationary	Low mobility
EAP identity	219.1	159.28
SSL	144.31	134.72
PEAP	99.6	102.61
EAPOL exchange	511.93	291.88
Total	974.94	Failed in 6 tests

identity exchange, SSL handshake, PEAP handshake, and EAPOL key exchange (see Appendix B for details). The overall EAP authentication delay (EAPd ) therefore comprises the EAP identity delay (EIDd ), SSL delay (SSLd ), PEAP delay (PPd ), and EAPOL delay (ELd ), as given in Eq. (7.3).

$$EAP_d = EID_d + SSL_d + PP_d + EL_d$$ (7.3)

Table 7.3 gives the delay distribution incurred by individual EAP phases in the stationary and low mobility environments. Under the stationary conditions, the EAP authentication completes in around 1 s on an average. While this delay is not too significant in the stationary conditions, it is the low mobility scenario where the performance of EAP becomes poor. Not only do the individual EAP phases incur significant delays in the low mobility tests, but they also fail to even complete on several occasions. Every time EAP authentication failed at any stage, it had to restart the entire process afresh after a large timeout. Table 7.4 suggests, for example, that the PEAP phase failed in test number 2 and 5 incurring additional delay of 18 and 36.394 s, respectively. According to Table 7.4, the EAP authentication phase failed in 6 out of 10 tests in the low mobility environments incurring additional delays.

The comparison of Tables 7.3 and 7.4 concludes that even the inclusion of slight mobility can incur large delays in the authentication phase of the handover. Hence, the delay reduction in the authentication phase requires considerable attention with specific regards to WLAN-based vehicular communications. However, any reduction in the authentication delay must not compromise the robustness of the security

	Test number	Delay incurred (s)
EAP identity	10	18.6
SSL	7	36
PEAP	2, 5	18, 36.394
EAPOL exchange	3, 6	21.62, 2.1

Table 7.4 Different EAP phases failing to complete the handover during the low mobility tests (Hasan et al. 2010b)

scheme. The security provision for wireless networks is a separate area of research and is therefore not considered for investigation any further.

7.3.3 Scanning Delay

The time consumed in searching for a suitable AP to initiate the handover is referred to as the scanning delay. During the handover tests discussed in the previous section, the average scanning delay is more than 300 ms in both stationary and low mobility conditions. The scanning phase does not show significant variations in the low mobility and stationary conditions. However, it exhibits high standard deviations of 458.79 ms in stationary tests and 434.868 ms in the low mobility tests. This variation is observed because two tests done in the stationary conditions and one under low mobility conditions exhibited larger scanning delays than the other tests. In one of the low mobility tests, the scanning delay alone was about 1,500 ms.

It has been pointed out in Sects. 7.3.1 and 7.3.2 that the reduction in authentication and address allocation phases respectively require changes in the protocols, which is beyond the scope of this chapter. However, the next section shows that the scanning phase delay can be reduced without making too many modifications in the hardware and software.

7.4 Reducing Scanning Phase Delay

Most 802.11 deployments use passive scanning, which requires the mobile node to spend around 100 ms on every channel of the 802.11 frequency band. Figure 7.8 shows the 802.11 spectrum that comprises of 14 channels, of which the 14th channel is used in Japan only. In passive scanning, the mobile node listens to the available channels one by one, waiting to hear a beacon signal from the candidate APs. Therefore, one complete scan cycle would ideally require 1,300 ms (for scanning 13 channels). If the mobile node fails to find an AP after scanning all channels, it generally restarts scanning from the first channel and keeps scanning until an appropriate AP is discovered. One way to reduce scanning phase latency is to reduce the

Fig. 7.8 802.11b/g spectra divided into 14 channels (Hasan et al. 2011d)

number of channels to be scanned. Scanning only a limited number of channels that are selected based on a certain criteria is called "selective channel scanning" (Shin et al. 2004). In the following, a selective channel scanning scheme has been proposed in which the criteria of scanning a channel is "orthogonality".

7.4.1 Scanning Orthogonal Channels

In the 802.11b/g spectrum, the channels 1, 6, and 11 are the orthogonal channels, i.e., they are nonoverlapping (Hossain and Leung 2008). The radios tuned to these channels can be operated in close vicinity of each other with minimum interference. While the conventional scanning method requires the mobile node to scan all channels one by one, scanning only three channels can significantly reduce the scanning phase delay.

This idea is particularly beneficial in the vehicular context because the vehicles cannot afford to waste too much time in searching for an AP. Therefore, reducing the scan cycle from 13 channels (scanning time 1,300 ms) to 3 channels (300 ms) can save 1,000 ms of scanning time per every scan cycle. Equation (7.4) shows that a particular channel shall be scanned only if f(x) = 1, which is true for x = 1, 6, 11, where x is the channel number such that $x = {1, 2, 3, \ldots, 12, 13}$ , and f(x) decides whether x is to be scanned. If the mobile node fails to find an appropriate AP at the end of the scan cycle, it will have to continuously scan only three channels instead of all 13.

$$f(x) = \begin{cases} 1 & \text{if } x = 1, 6, 11 \ 0 & \text{for all } x \end{cases}$$ (7.4)

Using the orthogonal channels in this context has three basic advantages.

First, very few modifications are required on the client side. Unlike the selective channel scanning which requires dedicated processing power to determine which channels to scan against a certain criteria, scanning orthogonal channels shall not require dedicated processing power and memory space. Hence, it is computationally inexpensive and requires very few changes in the hardware and/or software. Besides, the mobile node often does not have enough time to execute an algorithm to select a channel for scanning. The connection time in vehicular communications is already very small and cannot be wasted in running time-consuming channel

Fig. 7.9 Population of active 802.11 APs on different channels as observed in area-1 (Hasan et al. 2011d)

Table 7.5 Percentage occupancy of the orthogonal channels in Area-1, 2, and 3 (Hasan et al. 2011d)

	APs on ch-1 (%)	APs on ch-6 (%)	APs on ch-11 (%)
Area-1	37.02	32.09	30.61
Area-2	18.98	67.08	13.92
Area-3	36.60	42.85	20.05

selection algorithms. Second, the orthogonal channels are nonoverlapping and hence offer interference-free communications. This may help in reducing the packet losses due to collisions and therefore may reduce the need for retransmissions.

Finally, and probably most significantly, most of the available WLAN infrastructure is tuned to operate on the orthogonal channels. Figure 7.9 shows the observations from one area in the UK which is referred to as area-1 in this chapter. It is obvious that most of the already available APs are operating on the orthogonal channels. A similar trend was observed in the drive tests conducted in the other areas of the UK. In fact, some drive tests performed in China also suggest the same phenomenon (Hasan et al. 2011d).

Figure 7.10a, b show the population of APs operating on different 802.11 channels in two areas in China (referred to as area-2 and area-3). Based on the data collected from the drive tests, it is obvious that channels 1, 6, and 11 accommodate a significant majority of the roadside APs. Further according to the observations, after encountering an AP operating on a non-orthogonal channel, 7.46 APs in area-2 and 6.41 APs in area-3 (on average) are successively encountered on the orthogonal channels. This shows that most APs are inherently tuned to operate on the orthogonal channels.

Table 7.5 tabulates the percentage of AP population residing on the orthogonal channels in area-1, 2, and 3. It can be inferred from Table 7.5 that scanning only orthogonal channels shall miss only a small population of the roadside APs while

Fig. 7.10 Population of active 802.11 APs on different channels as observed in area-2 and area-3 (Hasan et al. 2011d)

saving reasonable scanning time. Because it is intuitive that scanning fewer channels shall require smaller amount of time, this book does not cover the actual calculation of the time required to scan the orthogonal channels. Instead, this book investigates the performance of APs operating on the orthogonal channels in supporting vehicular communications. The performance of APs operating on the orthogonal channels can be affected by at least two factors. One is the ability of an AP to support high data rates while similar APs operate on the same channel in its close vicinity. The second issue is the service capacity of an AP. These issues have been discussed in the following.

Fig. 7.11 Number of APs operating on channel 11 at the time of conducting the data rate tests (Hasan et al. 2011d)

7.4.2 AP Performance on Orthogonal Channels

WLAN deployments often suffer from interference not only from other WLANs but also from other devices operating on the 802.11 band. Interference results in various problems, for instance, low throughput, fluctuating connectivity and packet loss (Raghavendra et al. 2010). The interference-free nature of channels 1, 6, and 11 motivates the WLAN users to tune their APs on the orthogonal channels. Since the idea is to use interference-free orthogonal channels, the vehicles shall anticipate high data rates from the indoor APs. The concern here is that too many APs operating in close vicinity on the same channels should still be able to provide adequate levels of data rates in the vehicular environments. In other words, if n number of APs are operating in close vicinity of each other all on channel x, they should still be able to support reasonable data rates. This calls for data rate evaluation for the APs operating on a densely populated orthogonal channel. The data rate tests are performed on an AP operating on channel-11 (one of the orthogonal channels) on which several other APs were also operating at the time of the tests. Figure 7.11 shows the number of APs operating on channel-11 from the 25 scans recorded using Net Surveyor at the time of conducting the data rate tests.

Apart from the high density of APs on the orthogonal channels, another limiting factor affecting AP performance is the service capacity of an AP. The service capacity of an AP refers to the number of mobile nodes it can serve simultaneously. Note that the main idea is to use the already available roadside APs, which may be in use by other users for different Internet applications. Due to their preoccupation in supporting other services, the APs may not be able to provide high data rates for the vehicular nodes. In the following data rate evaluation, the WLAN AP (operating on a dense orthogonal channel) is loaded with three other mobile nodes running different Internet applications to find the realistic values of achievable data rates from outdoors.

Fig. 7.12 Experimental set up for data rate evaluation (Hasan et al. 2011d)

Experimental set up for the data rate evaluation is shown in Fig. 7.12, which shows 3 mobile nodes connected to an AP. The figure also shows the outdoor test node which evaluates the data rates using IPerf. All three nodes have Intel wireless cards. Nodes A and B are running Windows Vista while Node C uses Windows 7. The test node is a laptop running Windows Vista with a Realtek wireless driver. In order to examine the performance of a loaded AP, the three indoor nodes use the AP for supporting high bandwidth Internet applications. Node A uses the AP to play a live streaming video; node B and node C are making Skype calls with two wired stations, also shown in Fig. 7.12. Note that the AP being examined is operating on a densely populated orthogonal channel and serving other indoor applications at the same time. During the tests, the test node sends 1470 byte datagrams at 6Mbps (data rate specified by Direct Short Range Communications) every second to the AP while moving away from it. Although these tests represent the low mobility environments, the results can give a valuable insight into the WLAN-based R2V communications. The same process was repeated 5 times to obtain a statistical average.

The data rates evaluated during the 5 tests by the test node in low mobility vehicular scenario are shown in Fig. 7.13. The mean data rate curve (with the thick line) is also shown in the figure. The decreasing trend of the data rates is consistent with the fact that the test node is moved out of the AP's footprint during the tests. Instead of plotting distance, Fig. 7.13 plots time on the x-axis. Note that both are directly proportional because the distance between the AP and the mobile node increases with time. The data rates observed during the first 14 s are around 5Mbps which are considerably high for delivering traffic updates and safety information on the move. Note that although the data rate curves for 4 tests are reasonably smooth, one of the tests has a very fluctuating nature. This performance degradation may be attributed to the ever changing wireless channel characteristics. These fluctuations have been observed in the previous data rate evaluations as well.

Fig. 7.13 Observed data rates from the 5 tests and the mean curve (Hasan et al. 2011d)

In summary, the APs operating on the orthogonal channels can still perform well despite heavy AP population on the same channel and heavy load supported by that AP. Therefore, scanning only orthogonal channels can reduce scanning phase delay without affecting the end user experience.

7.5 Summary

This chapter starts with the basic introduction to handovers and its associated problems that must be dealt with while using 802.11 networks from vehicles. The handover latency issues associated with WLAN-based vehicular communications have been discussed. A detailed measurement and analysis of handover delay has been presented which identifies the DHCP procedure, EAP authentication, and scanning phase as the largest delay contributors. While any reduction in the authentication or address allocation delays requires considerable changes in the protocols, this chapter focuses on reducing the scanning phase delay without requiring significant changes in the client side.

The scanning phase delay is reduced by limiting the scanning cycle only to three orthogonal channels (instead of all 13 channels), thus reducing the time required to find an AP. The main motivation of scanning only orthogonal channels is that most of the APs are tuned to orthogonal channels. Therefore, there is no point in investing time for scanning other channels. The subsequent performance evaluation reveals that a loaded AP operating on a densely populated orthogonal channel can still support reasonable data rates.

Chapter 8 Future Directions and Research Ideas

The WLAN APs are being analyzed for use in vehicular communications due to their heavy presence alongside roads. However, these APs are deployed in an unplanned manner thus cannot support continuous connectivity. As a consequence of this random deployment, the vehicles face periods of connectivity and disconnectivity as they try to access the roadside WLAN infrastructure on the move. This intermittency in 802.11-based vehicular communications is termed as "disruption". Disruption is the major obstacle in the wide spread use of 802.11-based R2V communications. While disruption cannot be completely eradicated without extra deployment of APs, various attempts have been made to reduce or tolerate it. Some of the common approaches adopted to tolerate disruption are:

By developing new systems and network architectures,
By developing and/or modifying communication protocols, and
By predicting the future connectivity state of the vehicle.

While various previous works are available in literature that address these methods, very little work has been done in modelling disruption in the vehicular context. This book focuses on measuring disruption in R2V communication scenario. It develops a hidden Markov model that represents 802.11-based vehicular communications and gives a quantitative measure of disruption. The main motivation is to have a mathematical interpretation of disruption in vehicular communications. The developed model has been used to study the impact of traffic patterns on disruption. It was noted the dense traffic patterns offered less disrupted network services. The developed model has also been used to analyze the improvements brought by inter-operator roaming. It was shown that WISPr can increase $EP_l$ in vehicular communications thus reducing disruption.

The second issue addressed in this book is concerned with the handover latency in 802.11 networks. The handover delay adds to disruption because it suspends network services during the time a vehicle attempts to connect to a roadside AP. Since the encounter time between the vehicle and the AP is already very small, large handover delays cannot be tolerated. This book has given detailed measurement and analysis

of handover latency in the outdoor environments, and has also introduced a channel scanning scheme to reduce the handover delay. The proposed idea is such that it suits the rapidly changing vehicular context.

The contribution of this book has mainly been in three directions:

Proposing a scheme for traffic congestion monitoring,
· Developing mathematical models to measure disruption, and
Reducing handover latency in vehicular communications.

The research work reported in this book can be extended in each of these directions. Some ideas pertinent to each of these directions are discussed in the following.

The Extended MULE concept reported in Chap. 3 can be extended by developing effective algorithms to "detect" traffic congestion. Note that this book has focused on the communication issues only, and significant room for research exists in determining when a vehicle faces traffic congestion. The proposed X-MULE concept makes use of the sparsely located WLAN APs in R2V communication setup. Due to the sparse nature of AP deployment, there would be time periods where connectivity shall remain suspended. Buffering techniques need to be explored that can proxy a congestion signal in the absence of connectivity, and convey the same to the nearest AP as soon as one becomes available.

The models developed in Chaps. 4 and 5 can be modified to incorporate more information regarding the R2V scenario. Currently, it uses the information on encounter duration and authentication scheme of the encountered APs as two layers of the hidden Markov model. More layers corresponding to other important parameters such as the available signal strength may be added for improved performance. It has been mentioned in Sect. 5.5.2 that determining accuracy of this model also requires research effort. While limited evaluation of accuracy has been given, more rigorous evaluation of the same may be taken up as future research.

Chap. 7 of this book has addressed the handover latency issues in vehicular communication. A channel scanning strategy has been presented that reduces the number of channels to be scanned. Based on the foundation knowledge of 802.11 networks, it follows that the proposed scheme can reduce channel scanning time. However, actual delay evaluation has not been performed. Tuning a mobile node to scan only 3 channels and experimentally assessing the viability of this scheme shall also improve this book significantly.

8.1 Future Directions

In this final section of this book, three interesting research directions have been highlighted that possess considerable room for further research. The first is concerned with improving the generality of the developed model while the second deals with network convergence.

8.1 Future Directions 131

8.1.1 Network Convergence

It has been pointed out in Chap. 6 that for extending the mobility domain of WLANs, the operators must come into agreement to allow each others' users to access each others' resources. Apart from sharing resources that conform to the same wireless technology, the concept of using different types of radio technologies is also getting popular. At present, a WiFi user would generally connect to the WiFi infrastructure only. The emerging idea trend is to allow a WiFi node to use all available infrastructure regardless of the communication technology in use. A WiFi node is envisaged to have network access via infrastructure of WiMAX or cellular networks. This phenomenon is called Network Convergence. Network convergence allows a mobile node to access which ever network is readily available if the parent network faces disruption. Different works have been done in this area. The WLAN/cellular interoperability, for example, to provide network services to African rural areas has been studied by Chatelain and Wyk (2007). If WLAN/Cellular roaming is allowed, the irregularity in WLAN-based services can be significantly reduced by switching to the cellular services in the event of disruption. The Unlicensed Mobile Access (UMA) allows merging the WLAN (and bluetooth) with the cellular systems in order to provide continuous network services over an extended mobility domain. Panken et al. (2007) have discussed the integration of WLAN with 3G and WiMAX networks. The IP Radio Access Network (IP RAN) is yet another example of network convergence that not only supports a variety of network protocols but also allows flexibility in selecting an appropriate access network (Ivanek (2007)).

Network convergence is still an open area with various interesting research directions. One of these directions is allowing inter-technology handovers. The handovers in the convergent network environments must be quick and seamless to the end user. Facilitating the real time applications, such as VoIP, in these environments is another big challenge. The VoIP sessions can tolerate a delay of no more than 50 ms without affecting the user experience. This calls for a significant reduction in the cross-technology handover latency to realize an all IP Network (IPN) that facilitates all applications. Moreover, developing standard methods and practices to allow network convergence is another challenge. IEEE is standardizing 802.11u that addresses Media Independent Handover (MIH) to allow access to heterogeneous networks (Kim et al. (2009)). MIH itself provides significant room for future research activities.

8.1.2 Location Invariant Models

The Markov and hidden Markov models (λ = {π, Ai j, Bj(k)}) developed in Chaps. 4 and 5 are based on the data collected from specific geographical regions. Consequently, the calculated model parameters conform to specific areas only. It follows that the models developed in this book suffer from generality problem because the

Fig. 8.1 Measuring disruption in a real time manner

model developed from the data of one area cannot be used to analyze another area. In other words, the model parameters π, Ai j, Bj(k) need to be re-estimated for use in other areas. Re-estimating these parameters require fresh drive tests in the areas of interest, which may be tedious and often inconvenient. To increase their application domain, the models developed in Chap. 5 must be modified to exhibit adaptive behaviour.

A state-space neural network (SSNN) is a recurrent neural network that updates model parameters after observing one input-output pattern. The SSNN trained for one dataset can be quickly trained for other datasets without having to re-compute all model parameters repeatedly. A SSNN has been introduced for adaptive predictionbased systems in (Lint et al. (2005)). In essence, the SSNN is designed for travel time prediction on the freeways. In travel time prediction application, a particular route is broken down into segments each of which contains traffic sensors. The SSNN takes the measurements from these sensors as inputs from all sections of the route and evaluates the mean travel time for the vehicles on that route. This approach is called incremental learning approach which is different from the conventional batch learning approach. The parameters in the later are computed offline, while the incremental learning allows online parameter update.

The effectiveness of the hidden Markov model presented in Chap. 5 can be increased by updating the model parameters in real time. For example, Nicholson and Noble (2008) have developed a model that modifies as the vehicle visits different geographical locations. This concept may be used for the developed model by updating it at every AP hit in an online manner. Recall from Sect. 5.2.3 that model parameters can be updated using offline and online methods. Figure 8.1 shows that the model (housed inside the vehicle) keeps changing in an online manner as the vehicle traverses from one place to another. As a consequence of this online updating, the resulting E Pi values also keep changing. Thus, instead of estimating disruption offline after the drive tests, a real time interpretation of the same can be obtained.

8.1 Future Directions 133

8.1.3 Data Handling in Vehicular Sensor Networks

Vehicular Sensor Networks (VSN) are getting considerable research attention due to their involvement in road safety applications. A network of vehicles connected with each other can share useful and time critical information to avoid on-road accidents and casualties. This book has introduced Extended MULE concept in Sect. 3.3.1 that supports traffic congestion monitoring by using wireless sensors mounted on the public buses.

Vehicular sensor nodes collect a huge pile of information about vehicle speed, acceleration, etc. For example, in an experimental study that started in 2004, one million sets of second-by-second data from 400,000 vehicle trips were acquired by the on board sensors (Guizzo (2004)). It follows that handling large quantities of data acquired by the vehicular sensors must be dealt with in an efficient manner. Compressed sensing has been introduced as a technique that reconstructs information content based on a limited set of observed data (Donoho (2006)). The use of compressive sensing techniques in a VSN has been reported in (Yu et al. (2010)) that monitors urban environment. Similar works in efficient data acquisition for traffic monitoring require further investigation.

Appendix A Backward Algorithm

Recall from Fig. 5.8 in Chap. 5 that forward algorithm was used to estimate the SP and EP for the observation sequence $O = O_1, O_2, \dots, O_r$ .

The backward algorithm is concerned with estimating the future observation sequence from a given state. The backward algorithm computes $P(O_{v+1:t}|Xv)$ , such that the overall observation sequence is $O{k=t} = O_1, O_2, \dots, Ov, \dots, O{t-1}, O_t$ .

Therefore, the backward variables are expressed as:

$$\betan(j) = P(O{v+1}, \dots, O_t | X_v = j)$$ (A.1)

Collectively, the forward and backward variables are used to compute the so-called posterior probability (Xu and Gogarten 2008):

$$\gamma_n(i) = \frac{\alpha_n(i)\beta_n(i)}{P(S)} \tag{A.2}$$

Figure A.1 shows the application of the forward and backward algorithm on an observation sequence of length k=t.

Fig. A.1 The forward and backward algorithm applied to an observation sequence

Appendix B EAP Authentication Mechanism

Instead of conventional security mechanisms, EAP uses port-based authentication. A port can be seen as a point of attachment to the network, which opens only when the mobile node successfully completes the authentication procedure. The portbased mechanism is a general approach which may also be used in conjunction with authentication mechanisms other than EAP. However, its combination with EAP is most commonly employed.

The EAP authentication mechanism comprises the following elements:

Supplicant: this is the mobile node seeking WLAN services. It has to provide valid user credentials to complete authentication.
Authentication Server: this is an external server (e.g. RADIUS) which verifies user credentials against a database before allowing access to the network.
Authenticator: this is the WLAN AP which would serve the supplicant. Authenticator plays an intermediary role in the authentication process between the supplicant and the Authentication Server.

The procedure begins when the supplicant identifies itself with the authenticator and requests to initiate the EAP procedure. The authenticator identifies the supplicant and sends the response packet. After receiving the response and initiating the EAP, the supplicant starts the Secured Socket Layer (SSL) handshake procedure. SSL was introduced by Netscape communications for securing information exchange over the World Wide Web. In the EAP, it ensures a secured transfer of supplicant and server certificates. On top of the SSL layer, Protected EAP (PEAP) provides another security layer by encapsulating the messages between the supplicant and the server.

A successful EAP session generates Pairwise Master Key (PMK) for both the supplicant and the server. PMK is also delivered to the authenticator via a secure layer, and is used in all subsequent communications. The same PMK is used to generate another key called the Pairwise Transient Key (PTK). The difference between PMK and PTK is that PMK acts as a pass phrase for the entire user

Fig. B.1 EAP authentication mechanism (Hasan et al. 2010b). a The four processes involved in the EAP authentication mechanism. b The key exchange involved in the EAPoL process of the EAP authentication mechanism

session while the PTK is used for encrypting the data exchange between the supplicant and the authenticator. To derive PTK from PMK, a 4-way handshake between the authenticator and supplicant is executed. This handshake is collectively referred to as the EAP over LAN (EAPoL) key exchange. The EAP procedure is therefore classified into four processes, EAP identity, SSL, PEAP, and EAPoL, as shown in Fig. B.1a. The key exchange involved during the EAPoL phase is shown in Fig. B.1b.

Appendix C Software Tools

This appendix briefly introduces three tools (IPerf, Vistumbler, Network monitor) that have been extensively used in this book. These tools are independent of each other and suffice for different purposes.

IPerf

IPerf is a command line tool that measures network throughput by generating and transmitting TCP/UDP packets over the concerned network. In a typical setup, IPerf sends these packets from source node to the destination and reports the achievable throughput. It is also capable of reporting other parameters such as packet loss, delay, and jitter. IPerf can be used on both wired and wireless networks. Refer to Schroder (2008) for more information.

IPerf has been used in Sects. 3.2.2, 3.2.3 and 7.4.2 for measuring data rates in both stationary and vehicular setups.

Vistumbler

Vistumbler is Windows Vista compatible version of Netstumbler, which is commonly used in war driving tests. Vistumbler listens to the periodic beacon messages transmitted by the roadside APs and collects information on APs' signal strength, authentication scheme, encryption, radio type, population, etc. It also reports the time of receiving first and last beacon message from a particular AP. This time difference is referred to as the encounter time in this book (see Sect. 3.3.2). Vistumbler is meant for recording WLAN AP information only. It also supports GPS logs, however, positioning information has not been used in this work.

Fig. C.1 Vistumbler user interface

This book uses Vistumbler in Sect. 3.2.1 for measuring signal strengths of roadside APs. It is used in Sects. 3.3.2 and 5.5.1 for recording population, encounter durations, radio types, and authentication information of the roadside APs. Section 7.4.1 also uses vistumbler to report AP population in three different geographical areas.

Windows Network Monitor

Windows Network Monitor (WNM) provides enhanced functionalities for packet capture in wired and wireless networks. In this book, WNM has been used to capture and analyze packet exchange that occurs during handovers in 802.11 networks. In addition to the detailed information on the captured packets, WNM reports the time at which a certain packet is received. Time difference between two consecutive packets is interpreted as the time delay (or time offset in WNM environment). It also reports the signal strength with which a certain packet is captured. Refer to Tulloch et al. (2009) for more information.

WNM is used in Sect. 3.2.3 for recording signal strength operating in parallel with IPerf that measures data rates. It has been used in Sect. 7.3 for evaluating delay in different phases of handover.

Others

In addition to IPerf, Vistumbler, and WNM, Sect. 3.3.3 uses Online Eye Pro for measuring throughput while Sect. 7.4.2 uses Net Surveyor to report the number of APs operating on a particular channel.

A-Ghazaleh H, Alfa AS (2008) Mobility prediction and spatial-temporal traffic estimation in wireless networks. In: IEEE vehicular technology conference, Singapore, 11–14 May 2008, pp 2203–2207
A-Ghazaleh H, Alfa AS (2010) Application of mobility prediction in wireless networks using Markov renewal theory. IEEE Trans Veh Technol 59(2):46–57
A-Helali A-A, Mahmoud A, A-Kharobi T, Sheltami T (2009) Analysis of handoff delay components for mobile-IP based 3GPP UMTS/WLAN interworking architecture. In: International conference on advanced information networking and applications workshop, Bradford, UK, 26–29 May 2009, pp 798–803
Ahmed I, Habibi D (2008) A novel mobile WiMAX solution for higher throughput. In: IEEE conference on networks, New Delhi, India, 12–14 Dec 2008, pp 1–5
Altunbasak H, Owen H (2004) Alternative pair-wise key exchange protocols for robust security networks (IEEE 802.11i) in wireless LANs. In: IEEE southeast conference, Tennessee, USA, 26–28 March 2004, pp 3–9
Amdouni I, Filali F (2009) Intelligent strategies of access point selection for vehicle to infrastructure opportunistic communications. In: IEEE vehicular networking conference, Tokyo, Japan, 28–30 Oct 2009, pp 1–8
Asheralieva A, Erke TJ, Kilkki K (2009) Traffic characterization and service performance in FON network. In: First international conference on future information networks, Beijing, China, 14–17 Nov 2009, pp 285–291
Balasubramanian A, Mahajan R, Venkataramani A, Levine BN, Zahorjan J (2008) Interactive WiFi connectivity for moving vehicles. In: ACM SIGCOMM, Seattle, USA, 17–22 Aug 2008, pp 427–438
Balasubramanian A, Zhou Y, Croft WB, Levine BN, Venlataramani A (2007) Web search from a bus. In: ACM CHANTS, Montreal, Canada, 14 Sept 2007, pp 59–66
Basios C (2005) Defining architecture and key issues towards WLAN roaming. In: 8th international conference on telecommunications, Zagreb, Croatia, 15–17 June 2005, pp 225–230
Bauza R, Gozalvez J, Soriano J (2010) Road taffic congestion detection through cooperative vehicle-to-vehicle communications. IEEE workshop on user mobility and vehicular networks, pp 606–612
Behzad A (2002) Wireless LAN radios: system definition to transistor design. Wiley-IEEE, New York
Bernstein R (1999) Schaum's outline of theory and problems of elements of statistics II: inferential statistics. McGraw Hill, New York
Bhola J (2002) Wireless LANs demystified. McGraw-Hill Professional, New York

Bilchev G, Marston D, Hristov N, Peytchev E, Wall N (2004) Traffimatics—intelligent co-operative vehicle highway systems. BT Technol J 22(3):73–83

Bilstrup K, Uhlemann E, Strom EG, Bilstrup U (2008) Evaluation of the IEEE 802.11p MAC method for vehicle-to-vehicle communication. In: IEEE vehicular technology conference, Calgary, 21–24 Sept 2008, pp 1–5
Bing B (2008) Emerging technologies in wireless LANs: theory, design and deployment. Cambridge University Press, Cambridge
Blank AG (2004) TCP/IP foundations. Wiley, Hoboken
Bodden V (2008) Internet. The Creative Company, Mankato
Borisov N, Goldberg I, Wagner D (2001) Intercepting mobile communications: the insecurity of 802.11. In: International conference on mobile computing and networking, Rome, Italy, 16–21 July 2001, pp 180–189
Brady PT (1969) A model for generating on-off speech patterns in two-way conversation. Bell Syst Technol J 48(7):2445–2472
Bychkovsky V, Hull B, Miu A, Balakrishnan H, Madden S (2006) A measurement study of vehicular internet access using in situ WiFi networks. In: 12th ACM MobiCom, California, USA, 24–29 Sept 2006
Caceres M, Sottile F, Spirito MA (2009) WLAN-based real time vehicle locating system. In: IEEE vehicular technology conference, Barcelona, Spain, 26–29 Apr 2009, pp 1–5
Cardenas LR, Boutabia M, Afifi H (2008) Infrastructure-based approach for fast and seamless Handovers. In: 3rd international conference on digital communications, Bucharest, Romania, 29 June–5 July 2008, pp 105–109
Chakraborty S, Frankkila T, Peisa J, Synnergren P (2007) IMS multimedia telephony over cellular systems. Wiley, New York
Chao S-J, Zhang J-M, Tuan C-C (2010) Hierarchical IP distribution mechanism for VANET. In: International conference on ubiquitous and future networks, Jeju, South Korea, 16 June 2010, 349–354
Chatelain D, Wyk BJ (2007) WLAN over GSM for IP access in African rural areas. IET Electron Lett 43(5):61–62
Chen H-H, Guizani M (2006) Next generation wireless systems and networks. Wiley, London (illustrated)
Chen J-C, Wang Y-P (2005) Extensible authentication protocol (EAP) and IEEE 802.1x: tutorial and empirical experience. IEEE Commun Mag 43(12):26-32
Chen J-J, Tseng Y-C, Lee H-W (2007) A seamless handoff mechanism for DHCP-based IEEE 802.11 WLANs. IEEE Commun Lett 11(8):665–667
Chen X, Qiao D (2010) HaND: fast handoff with Null Dwell time for IEEE 802.11 networks. IEEE INFOCOM, San Diego, USA, 15–19 Mar 2010, pp 1–9
Chen Y, Kowalik K, Davis M (2009) MeshScan: performance of passive handoff and active handoff. In: International conference on wireless communications and signal processing, Nanjing, China, 13–15 Nov 2009, pp 1–5
Ching WK, Ng MK (2006) Markov chains: model, algorithms and applications. Birkhauser, Basel Chisalita I, Shahmehri N (2004) Vehicular communication—a candidate technology for traffic safety. In: IEEE International conference on systems, man and cybernetics, Linkoping, Sweden, 10–13 Oct 2004, pp 3903–3908
Cho Y (2007) Estimating velocity fields on a freeway from low-resolution videos. IEEE Trans Intell Transp Syst 7(4):463–469
Choi H-H., Song O, Park Y-K, Lee J-R (2010) Performance evaluation of opportunistic vertical handover considering on–off characteristics of VoIP traffic. IEEE Trans Veh Technol 59(6):3115–3121
Chou C-M, Li C-Y, Chien W-M, Lan K-C (2009) A feasibility study on vehicle-to-infrastructure communication: WiFi vs. WiMAX. In: 10th international conference on mobile data management: systems, services and middleware, Taiwan, 18–20 May 2009
Clancy T (2008a) Secure handover in enterprise WLANs: CAPWAP, HOKEY and IEEE. IEEE Wireless Commun 15(5):80–85

Clancy TC (2008b) Secure handover in enterprise WLANs: CAPWAP, HOKEY and IEEE. IEEE Wireless Commun 15:80–85

Cottingham DN, Wassell IJ, Harle RK (2007) Performance of IEEE 802.11a in vehicular contexts. In: IEEE vehicular technology conference, Dublin, Ireland, 22–25 Apr 2007, pp 854–858
David K, Flach A (2010) Car-2-x and pedestrian safety. IEEE Veh Technol Mag 5:70–76
Deshpande P, Kashyap A, Sung C, Das SR (2009) Predictive methods for improved vehicular WiFi access. In: ACM MobiSys, Krakow, Poland, 22–25 June 2009, pp 263–276
Donoho DL (2006) Compressed sensing. IEEE Trans Inf Theory 52(4):1289–1306
DoT (2006) US department of transportation. Available at: http://ops.fhwa.dot.gov
Drucker J, Angwin J (2002) Wi-Fi gives cell carriers static. Wall Street J, 1–5
Emmelmann M (2005) Influence of velocity on the handover delay associated with a radio-signalmeasurement-based handover decision. In: IEEE vehicular technology conference, Texas, USA, 25–28 Sept 2005, pp 2282–2286
Eriksson J, Balakrishnan H, Madden S (2008) Cabernet: vehicular content delivery using WiFi. In: 14th ACM MobiCom, California, USA, 14–19 Sept 2008, pp 199–210
Etemadi N, Ashtiani F (2011) Throughput analysis of IEEE 802.11-based vehicular ad hoc networks. IET Commun 5(14):1954–1963
Ezell S (2010) Explaining international it application leadership: intelligent transportation systems. Information technology and innovation foundation
Fall K, Farrell S (2008) DTN: an architectural retrospective. IEEE J. Sel. Areas Commun. 26(5):828–836
Farrell S, Cahill V, Geraghty D, Humphreys I, McDonald P (2006) When TCP breaks: delay- and disruption-tolerant networking. IEEE Internet Comput 10(4):72–77
Fathi H, Kobara K, Chakraborty SS, Imai H, Prasad R (2005) On the impact of security on latency in WLAN 802.11b. In: IEEE GLOBECOM, St. Louis, 28 Nov–2 Dec 2005, pp 1752–1756
Fazio M, Palazzi CE, Das S, Gerla M (2007) Facilitating real-time applications in VANETs through fast address auto-configuration. 3rd IEEE CCNC international workshop on networking issues in multimedia entertainment, Nevada, USA, 10–12 Jan 2007
FB-Algorithm (2011) Forward-backward Algorithm. Wikipedia online: http://en.wikipedia.org/ wiki/Forward-backward\_algorithm. Accessed 1 Apr 2011
Feng J (2009) Analysis, implementation and extensions of RADIUS protocol. In: International conference on networking and digital society, Cancun, Mexico, 1–7 Feb 2009, pp 154–157
Floris A, Tosetti L, Veltri L (2003) Solutions for mobility support in DHCP-based environments. In: IEEE international conference on communications, Alaska, USA, 11–15 May 2003, pp 1043–1047
Frangiadakis N, Kuklov D, Roussopoulos N (2007) PEGASUS: 802.11 connectivity at high speed. IEEE GLOBECOM workshops, Washington DC, 26–30 Nov 2007
Garroppo RG, Gazzarrini L, Giordano S, Tavanti L (2011) Experimental assessment of the coexistance of Wi-Fi, ZigBee, and Bluetooth devices. In: IEEE international symposium on world of wireless, mobile and multimedia networks, pp 1–9
Gass R, Diot C (2010) Eliminating backhaul bottlenecks for opportunistically encountered Wi-Fi hotspots. In: IEEE vehicular technology conference, Washington, USA, 6–7 Apr 2010, pp 1–5
Gass R, Scott J, Diot C (2006) Measurements of in-motion 802.11 networking. 7th IEEE workshop on mobile computing systems and applications, Washington, USA, 6–7 Apr 2006, pp 69–74
Geier J (2002) The BIG question: 802.11a or 802.11b? Wi-Fi planet online articles
Giannoulis A, Fiore M, Knightly EW (2008) Supporting vehicular mobility in urban multi-hop wireless networks. In: ACM MobiSys, CO, USA, 10–13 June 2008, pp 54–66
Godara LC (2002) Handbook of antennas in wireless communications. The electrical engineering and applied signal processing series, CRC Press, Boca Raton
Goodman DJ, Wei SX (1989) Factors affecting the bandwidth efficiency of packet reservation multiple access. In: IEEE vehicular technology conference, San Francisco, USA, 1–3 May 1989, pp 292–299

Goth G (2008) New Wi-Fi technology racing past standards process. IEEE Distrib Syst Online 9(10)

Goth G (2011) Wi-Fi making big news waves. IEEE Internet Comput 15(5):7–10
Grinstead CM, Snell JL (2007) Introduction to probability. AMS Bookstore, Providence
Guizzo E (2004) Netwok of traffic spies built into cars in Atlanta. IEEE Spectr Online
Hadallerp D, Keshav S, Brecht T, Agarwal S (2007) Vehicular Oopportunistic communication under the microscope. In: ACM MobiSys, San Juan, Puerto Rico, 11–14 June 2007, pp 206–219
Hasan SF, Ding X, Siddique NH, Chakraborty S (2011a) Measuring disruption in vehicular communication. IEEE Trans Veh Technol 60(1):148–159
Hasan SF, Siddique NH, Chakraborty S (2011b) Developments and constraints in 802.11-based vehicular communication. J Wireless Pers Commun, Springer (under review)
Hasan SF, Siddique NH, Chakraborty S (2011c) Extended MULE concept for traffic congestion monitoring. J Wireless Pers Commun, Springer
Hasan SF, Siddique NH, Chakraborty S (2011d) Scanning and address allocation delays in vehicular communications. J Wireless Pers Commun, Springer
Hasan SF, Siddique NH, Chakraborty S (2009a) Disruption model for net-on-roads. In: International conference on applications of digital information and web technologies, pp 282–287
Hasan SF, Siddique NH, Chakraborty S (2009b) Femtocell versus WiFi—a survey and comparison of architecture and performance. In: International conference on wireless VITAE, pp 916–920
Hasan SF, Siddique NH, Chakraborty S (2009c) Impact of RSS on the performance of 3GPP applications in a net-on-roads connection. 17th Telecommunications forum, pp 266–269
Hasan SF, Siddique NH, Chakraborty S (2010a) HMM-based modelling of roadside-to-vehicle WLAN communications. 2nd International workshop on communication technologies for vehicles, pp 427–431
Hasan SF, Siddique NH, Chakraborty S (2010b) On evaluating the latency in handing over to EAP-enabled WLAN APs from outdoors. In: 7th IEEE/IET international symposium on communication systems, networks and digital signal processing, pp 278–282
Hasan SF, Siddique NH, Chakraborty S (2010c) On the effectiveness of WISPr in roadside-tovehicle communications. IEEE Commun Lett 14(9):818–820
Hasan SF, Siddique NH, Chakraborty S (2010d) WLAN datarates achievable from roads in low and high mobility environments. In: IEEE international communications conference workshops, pp 1–5
Hassan MB, Hassan M (2009) A Markov chain model of streaming proxy for disconnecting vehicular networks. IEEE vehicular technology conference, pp 1–5
Haykins SS (2009) Neural networks and learning machines. Prentice Hall, New York
Hossain E, Leung KK (2008) Wireless mesh networks: architectures and protocols. Springer, Heidelberg (illustrated)
Hu H, Wang F, Wang F, Jia W, Tang G (2009) Automatic mobile vehicle for adaptive real-time communication relay. In: IEEE international conference on distributed computing systems workshops, Genova, Italy, 22–26 June 2009, pp 32–37
Hull B, Bychkovsky V, Zhang Y, Chen K, Goraczko M, Miu A, Shih E, Balakrishnan H, Madden S (2006) CarTel: a distributed mobile sensor computing system. In: ACM SenSys, Colorado, USA, Oct/Nov, pp 125–138
IPerf (2011) IPerf Tutorial. Available at: http://openmaniak.com/iperf.php. Accessed 1 Apr 2011 Ivanek F (2007) Convergence and competition on the way toward 4G: where are we going. In: IEEE radio and wireless symposium, California, USA, 9–11 Jan 2007, pp 265–268
Jiang D, Chen Q, Delgrossi L (2007) Optimal data rate selection for vehicle safety communications. ACM VANET workshop on vehicular internetworking, California, USA, 15 Sept 2007, pp 30–38
Jiang, D., Delgrossi, L., 2008. IEEE 802.11p: Towards an International Standard for Wireless Access in Vehicular Environments. IEEE Vehicular Technology Conference, Singapore, 11–14 May, 2036–2040.

Jiang S, Vaidya N (1999) Scheduling data broadcast to 'impatient' users. In: 1st International ACM workshop on data engineering for wireless and mobile access, Seattle, USA, Aug, pp 52–59

Jin S, Choi M, Choi S (2009) Multiple WNIC-based handoff in IEEE 802.11 WLANs. IEEE Commun Lett 13(10):752–754
Joseph AD (2006a) Intelligent transportation systems. IEEE Pervasive Comput 5(4):63–67
Joseph AD (2006b) Works in progress: intelligent transportation systems. IEEE Pervasive Comput 5(4):63–67
Jurafsky D, Martin JH (2009) Speech and language processing: an introduction to natural language processing, computational linguistics and speech recognition. Prentice Hall series in artificial intelligence, Prentice Hall, Upper Saddle River
Kanellos M, Charny B (2002) PCs to be intel's wedge into wireless. CNet, Available at: http://news.cnet.com/2100-1040-957472.html
Kim H-S, Park S-H, Park C-S, Kim J-W, Ko S-J (2004) Selective channel scanning for fast handoff in wireless LAN using neighbour graph. In: International technical conference on CSCC, Miyagi-Pref, Japan, 6–8 July 2004
Kim M, Moon T-W, Cho S-J (2009) A study on IEEE 802.21 MIH frameworks in heterogeneous wireless networks. In: 11th International conference on advanced communication technology, Gangwon-Do, South Korea, 15–18 Feb 2009, 242–246
Kim S, Choi S, Park S, Lee J, Kim S (2006) An empirical measurements-based analysis of public WLAN handoff operations. In: First international conference on communication systems software and middleware, New Delhi, India, 8–12 Jan 2006, pp 1–6
Koski T (2001) Hidden Markov models for bioinformatics. Springer, Netherlands
Kvam PH, Vidakovic B (2007) Nonparamteric statistics with applications to science and engineering. Wiley series in probability and statistics, Wiley Interscience, Hoboken
Kwak D, Mo J, Kang M (2009) Investigation of handoffs for IEEE 802.11 networks in vehicular environment. In: International conference on ubiquitous and future networks, Hong Kong, China, 7–9 June 2009, 89–94
Kwok Y, Lau V (2007) Wireless internet and mobile computing: interoperability and performance. Wiley-IEEE Press, Hoboken
Lee BG, Kim HG, Park KH (2002) An AAA application protocol design and service for secure wireless internet gateway roaming. Lecture Notes in Computer Science, Springer-Verlag, London, pp 123–135
Lee J-K, Hou JC (2006) Modeling steady-state and transient behaviours of user mobility: formulation, analysis and application. In: ACM MobiHoc, Florence, Italy, 22–25 May 2006, pp 85–96
Leontiadis I, Costa P, Mascolo C (2010) Extending access point connectivity through opportunistic routing in vehicular networks. In: IEEE INFOCOM, San Diego, USA, 15–19 Mar 2010, pp 1–5
Li X, Shu W, Li M (2009) Performance evaluation of vehicle-based mobile sensor networks for traffic monitoring. IEEE Trans Veh Technol 58(4):1647–1653
Lint JWC, Hoogendoorn SP, Zuylen HJ (2005) Accurate freeway travel time prediction with state-space neural networks under missing data. Transp Res Part C: Emerg Technol 13(5/ 6):347–369
Luo H, Henry P (2003) A secure public wireless LAN access technique that supports walk-up users. In: IEEE GLOBECOM, San Francisco, USA, 1–5 Dec 2003, pp 1415–1419
Ma X, Chen X, Refai HH (2009) Performance and reliability of DSRC vehicular safety communication: a formal analysis. EURASIP J Wireless Commun Network, 1–13
Mahajan R, Zahorjan J, Zill B (2007) Understanding WiFi-based connectivity from moving vehicles. In: ACM SIGCOMM conference on internet measurement, Kyoto, Japan, 27–31 Aug 2007, pp 321–326
Mancuso V, Gambardella M, Bianchi G (2004) Improved support for streaming services in vehicular networks. In: IEEE ICC, New York, 20–24 June 2004, pp 4362–4366

Manodham T, Loyola L, Atoche G, Hayasaka M, Miki T (2005) A novel handover scheme for reducing latency in WLANs. In: Asia-Pacific conference on communications, 5 Oct 2005, pp 1141–1144

Mhatre V, Papagiannaki K (2006) Using smart triggers for improved user performance in 802.11 wireless networks. In: ACM MobiSys, Uppsala, Sweden, 19–22 June 2006, pp 246–259
Miller J (2008) vehicle-to-vehicle-to-infrastructure (V2V2I) intelligent transportation system architecture. In: IEEE intelligent vehicles symposium, pp 715–720
Mishra A, Shin M, Arbaugh W (2003) An empirical analysis of the IEEE 802.11 MAC layer handoff process. SIGCOMM Comput Commun Rev 33:93–102
Mohandas BK, Liscano R (2008) IP address configuration in VANET using centralized DHCP. In: IEEE conference on local computer networks, Montreal, Canada, 14–17 Oct 2008, pp 608–613
Morgan YL (2010) Notes on DSRC & WAVE standards suite: its architecture, design, and characteristics. IEEE Commun Surv Tutorials 12(4):504–518
Nicholson AJ, Noble BD (2008) Breadcrumbs: forecasting mobile connectivity. In: ACM MobiCom, California, USA, 14–19 Sept 2008, pp 46–57
Ok J, Morales P, Morikawa H (2008) AuthScan: enabling fast handoff across already deployed IEEE 802.11 wireless networks. In: IEEE international symposium on personal indoor mobile radio communications, Cannes, France, 15–18 Sept 2008, pp 1–5
Oppenheimer P (2004) Top-down network design. Cisco Press, Indianapolis
Ott J, Kutscher D (2004a) The ''Drive-thru'' architecture: WLAN-based internet access on the road. In: IEEE vehicular technology conference, Bremen, Germany, 17–19 May 2004, pp 2615–2622
Ott J, Kutscher D (2004b) Drive-thru internet: IEEE 802.11b for ''Automobile'' users. In: IEEE INFOCOM, Hong Kong, 7–11 May 2004, pp 362–373
Ott J, Kutscher D (2005) A disconnection-tolerant transport for Drive-thru internet environments. In: IEEE INFOCOM, Miami, USA, 13–17 Mar 2005, pp 1849–1862
Ott J, Kutscher D, Koch M (2005) Towards automated authentication for mobile users in WLAN hot-spots. In: IEEE vehicular technology conference, Texas, USA, 25–28 Sept 2005, pp 1232–1241
P-Atikom W, Pongpaibool P, Thajchayapong S (2006) Estimating road traffic congestion using vehicle velocity. In: International conference on ITS telecommunications, Chengdu, China, 21–23 June 2006, pp 1001–1004
Paik EK, Choi Y (2003) Prediction-based fast handoff for mobile WLANs. In: International conference on telecommunications, Zagreb, Croatia, 11–13 June 2003, pp 748–753
Panik MJ (2005) Advanced statistics from an elementary point of view. Academic Press, Frisco Panken F, Hoekstra G, Barankanira D, Francis C, Schwendener R, Grondalen O, Jaatun MG (2007) Extending 3G/WiMAX network services through residential access capability. IEEE Commun Mag 45:62–69
Posey B (2008) The real MCTS/MCITP Exam 70-649: upgrading your MCSE on Windows server 2003 to Windows server 2008 prep kit. Syngress, Burlington (Illustrated)
Rabiner LR, Juang BH (1986) An introduction to Hidden Markov models. IEEE ASSP Mag, 3(1):4–16
Rabiner RL (1989) A tutorial on Hidden Markov Models and selected applications in speech processing. Proc IEEE 77(2):257–286
Raghavendra R, Belding EM, Papagiannaki K, Almeroh KC (2010) Unwanted link layer traffic in large 802.11 wireless networks. IEEE Trans Mobile Comput 9(9):1212–1225
Raman B, Chebrolu K, Gokhale D, Sen S (2009) On the feasibility of the link abstraction in wireless mesh networks. IEEE/ACM Trans Networking 17(2):528–541
Ramani I, Savage S (2005) SyncScan: practical fast handoff for 802.11 infrastructure networks. In: IEEE INFOCOM, Miami, USA, 13–17 Mar 2005, pp 675–684
Rappaport T (1996) Wireless communications—principles and practice. Prentice Hall, Upper Saddle River

Reinward CC (2007) Municipal broadband—the evolution of next generation wireless networks. In: IEEE radio and wireless symposium, Long Beach, CA, 9–11 Jan 2007, pp 273–276

Renda A, Guerin S, Arbak E (2009) EU-turkey assessment negotiations. Center for European Policy Studies (CEPS)
Reynolds J (2003) Going Wi-Fi: a practical guide to planning and building an 802.11 network. Focal Press, Boston (illustrated edition)
Rodriguez P, Chakravorty R, Chesterfield J (2004) MAR: a commuter router infrastructure for the mobile internet. In: ACM MobiSys, Boston, USA, 6–9 June, pp 217–230
Ross SM (2002) Probability models for computer science. Academic Press, New York
Santa J, Moragon A, G-Skarmeta AF (2008) Experimental evaluation of a novel vehicular communication paradigm based on cellular networks, In: IEEE intelligent vehicles symposium, Murcia, 4–6 June 2008, pp 193–203
Schmidt T, Townsend A (2002) Why Wi-Fi wants to be free. Commun ACM 46(5):47–52
Schroder C (2008) Linux networking cookbook. O'Reilly Media, Sebastopol
Shah R, Roy S, Jain S, Brunette W (2003) Data mules: modelling a three-tier architecture for sparse sensor networks. 1st IEEE international workshop on sensor network protocols and applications, Alaska, USA, 11 May 2003, pp 30–41
Shin S, Rawat AS, Schulzrinne H (2004) Reducing MAC layer handoff latency in IEEE 802.11 wireless LANs. In: ACM MobiWac, Philadelphia, USA, 26 Sept–1 Oct 2004, pp 19–26
Shinder TW, Shinder DL, Grasdal M (2005) Configuring ISA server 2004. Syngress, Burlington Shinder TW, Shinder DL, Grasdal M (2008) Ip address passing for VANETs. In: IEEE international conference on pervasive computing and communications, Hong Kong, 17–21 Mar 2008, pp 70–79
Sichitiu ML, Kihl M (2008) Inter-vehicle communication systems: a survey. IEEE Commun Surv Tutorials 10(2):88–105
Srinivasan K, Levis P (2006) RSSI is under appreciated. 3rd Workshop on embedded networked sensors, Massachusetts, USA, 30–31 May 2006
Stallings W (2008) IEEE 802.11: moving closer to practical wireless LANs. IT Prof 3(3):17–23 Stancil D, Cheng L, Henty B, Bai F (2007) Performance of 802.11p waveforms over the vehicleto-vehicle channel at 5.9 GHz. IEEE 802.11 task group p report
Stern HP, Mahmoud SA, Wong K-K (1994) A model for generating on-off patterns in conversational speech, including short silence gaps and the effects of interaction between parties. IEEE Trans Veh Technol 43(4):1094–1100
Stewart WJ (2009) Probability, markov chains, queues, and simulation: the mathematical basis of performance modeling. Princeton University Press, Princeton
Strom E, Hartenstien H, Santi P, Wiesbeck W (2010) Vehicular communications: ubiquitous networks for sustainable mobility. Proc IEEE, 1111–1112
Subramanian V, Ramakrishnan KK, Kalyanaraman S (2007) Disruption-tolerant link-level mechanisms for extreme wireless network environments. In: 2nd IEEE/Create-Net/ICST international conference on communication system software and middleware (COMSWARE), Bangalore, India, 7–12 Jan 2007, pp 1–10
Sun YS, Pan Y-C, Chen MC (2005) Fast and secure universal roaming service for mobile internet. In: IEEE Globecom, Hawaii, Nov/Dec, pp 2896–2901
Tan T, Bing B (2003) The world-wide Wi-Fi: technological trends and business strategies. Wiley Inter-Science Publication, New York
Tao K, Li J, Sampalli S (2009) Detection of spoofed mac addresses in 802.11 wireless networks. E-business and telecommunications, Springer, Heidelberg
Teng J, Xu C, Jia W, Xuan D (2009) D-Scan: enabling fast and smooth handoffs in AP-dense 802.11 wireless networks. In: IEEE INFOCOM, Rio de Janeiro, Brazil, 19–25 Apr 2009, pp 2616–2620
Tommasi F, Molendini S, Tricco A (2006) Experience-driven selective scan for 802.11 networks. In: International conference on software in telecommunications and computer networks, Dubrovnik, Croatia, 1 Aug 2006, pp 137–141

Toor Y, Muhlethaler P (2008) Vehicle Ad Hoc networks: applications and related technical issues. IEEE Commun Surv Tutorials 10(3):74–85

Tsankov B, Pachamanov R, Pachamanov D (2007) Modified brady voice traffic model for WLAN and WMAN. IEEE Electron Lett 43(23):1295–1297
Tufail A, Fraser M, Hammad A, Hyung KK, Yoo S-W (2008) An empirical study to analyze the feasibility of WiFi for VANETs. In: 12th international conference on computer supported cooperative work in design, Suwon, 16–18 Apr 2008, pp 553–558
Tulloch M, Northrup T, Honeycutt J (2009) Windows 7 resource kit. O'Reilly Media Inc, Sebastopol
Velayos H, Karlsson G (2004) Techniques to reduce the IEEE 802.11b handoff time. In: IEEE ICC, Paris, France, 20–24 June 2004, pp 3844–3848
Vergetis E, Pierce E, Blanco M, Guerin R (2006) Packet level diversity—from theory to practice: an 802.11-based experimental investigation. In: 12th annual international conference on mobile computing and networking, Los Angeles, USA, 24–29 Sept 2006, pp 62–73
Viniotis Y (1998) Probability and random processes for electrical engineers. McGraw-Hill international edition, Boston
Vistumbler (2011) Vistumbler. Available at: http://www.vistumbler.net. Accessed 1 Apr 2011 Viswanathan S (2009) Backward compatibility with legacy devices and operating modes. 802.11n PHY Layer—a tutorial, network systems design line
Wilke TL, Tientrakool P, Maxemchuk NF (2009) A survey of inter-vehicle communication protocols and their applications. IEEE Commun Surv Tutorials 11(2):3–20
Wu H, Tan K, Zhang Y, Zhang Q (2007) Proactive scan: fast handoff with smart triggers for 802.11 wireless LAN. In: IEEE INFOCOM, Alaska, USA, 6–12 May 2007, pp 749–757
Xu Q, Wan C, Hu A (2008) The performance analysis of Fast EAP re-authentication protocol. In: International symposium on computer science and computational technology, Shanghai, China, 20–22 Dec 2008, pp 99–103
Xu Y, Gogarten JP (2008) Computational methods for understanding bacterial and archaeal genomes. Imperial College Press, London
Yang J, Xu Y, Chen C (1997) Human action learning via Hidden Markov model. IEEE Trans Syst Man Cybern 27(1):34–44
Yang X, Liu J, Zhao F, Vaidya NH (2004) A vehicle-to-vehicle communication protocol for cooperative collision warning. In: International conference on mobile and ubiquitous systems: networking and services
Yen L-H, Yang C-C (2006) Mobility profiling using Markov chains for tree-based object tracking in wireless sensor networks. In: IEEE international conference on sensor networks, ubiquitous, and trustworthy computing, pp 220–225
Yu X, Zhao H, Zhang L, Wu S, Krishnamachari B, Li VK (2010) Cooperative sensing and compression in vehicular sensor networks for urban monitoring. In: IEEE international communications conference
Zhang C, Lin X, Lu R, Ho P-H (2008a) RAISE: an efficient RSU-aided message authentication scheme in vehicular communication networks. In: IEEE ICC, Beijing, China, 19–23 May 2008, pp 1451–1457
Zhang J, Tan K, Zhao J, Wu H, Zhang Y (2008b) A practical SNR-guided rate adaptation. In: IEEE INFOCOM, California, USA, 13–18 Apr 2008, pp 146–150
Zhang Y, Zhao J, Cao G (2007) On scheduling vehicle-roadside data access. In: 4th ACM international workshop on vehicular and Ad Hoc networks, Montreal, Canada, 10 Sept 2007, pp 9–18
Zheng X, Sarikaya B (2009) Handover keying and its uses. IEEE Network 23:27–34
Zhu H, Li M, Chlamtac I, Prabhakaran B (2004) A survey of quality of service in IEEE 802.11 networks. IEEE Wireless Commun 11(4):6–14

Index

3GPP Applications, 35, 50, 51, 54	E
3-state Markov model, 61, 70, 71	EAP, 23, 24, 29, 112, 113, 116, 119, 120, 128,
802.11n, 2, 27, 30	137, 138
802.11p, 9, 27, 28, 81, 82	Encounter durations, 49, 83, 95, 109, 130
802.11r, 27–29, 81, 82	Encounter probability, 88–90, 92, 93, 97, 98,
	102, 107, 109
	Experimentation, 41, 85
A	Exponential distribution, 64, 73
Address allocation delay, 23, 24, 127	Extended Service Set, 2
AP footprints, 24, 60, 80
AP Population, 34, 66, 70, 82, 94, 123, 127, 140
Authentication delay, 23, 28, 119, 120	F
	FON network, 101
	Forward algorithm, 83, 84, 88, 90, 92, 97,
B	98, 135
Brady's model, 57, 59	Future directions, 97, 129


C	H
Cabernet Transport Protocol, 17	Handover latency, 9, 11, 13, 28, 31, 33, 49,
Cellular Networks, 1, 2, 7–9, 105, 131	111, 112, 119, 127, 129, 130
	Handovers, 10, 20, 21, 24–26, 43, 81, 100,
	102, 127
D	Hidden Markov model, 20, 61, 70–72, 75,
Data rate evaluation, 34, 38, 54, 125, 126	130–132
Data rates, 2, 4, 8, 12, 30, 33, 34, 38, 39,
41–44, 47, 54, 94, 124–126, 128
DHCP, 23, 24, 26, 47, 112–114, 116–118, 127	I
Direct Short Range Communication, 27, 81, 126	Initial state distribution, 76, 77, 79, 88, 89, 96,
Disruption, 10, 12, 13–15, 17, 18–20, 31, 55,	105, 109
62, 69, 70, 73, 75, 82–84, 88, 90, 93,	Intelligent Transportation Systems, 5
96, 98, 99, 102, 107, 109, 127, 131, 132
	Inter-ISP roaming, 99, 101, 102, 109
Disruption Tolerant Networking, 9, 13, 93	Internet Service Provider, 11, 99
Drive thru Internet, 14	Intra-ISP handover, 100, 101

150 Index

L Location invariance, 131 Long Term Error Rate, 69	R Received Signal Strength, 34 Roadside communication stations, 5 Roaming, 11, 97, 99, 101, 102, 104
M Markov models, 19, 57, 58, 60, 68 Measuring disruption, 75, 83, 84, 89, 98, 129 MULE Concept, 45 N Network Convergence, 130, 131	S Scanning phase delay, 23, 119, 121, 127 State probability, 88, 89, 90, 98 State transition probability, 76, 80, 88, 105 State-space Neural Network, 132 Stochastic models, 18
O Observation probability matrix, 76, 77, 105 Observation sequence HMM, 84 Online and Offline calculations, 83	T Traffic Congestion monitoring, 11, 44–46, 54, 130, 133 Traffic pattern analysis, 93
Orthogonal channels, 122–125, 127, 128 Outdoor experimentation, 41	V
Outdoor measurements, 132	V2V Communication, 5, 6, 13, 27 V2V2I Communication, 6 Vehicular Sensor Networks, 46
P Pegasus, 15, 16
Performance evaluation, 127 Poisson distribution, 65 Probability distributions, 64 Probability plots, 64, 66	W WiMAX, 2, 7, 8, 102, 131, 133 Wireless Local Area Networks, 1 WISPr, 18, 49, 97, 98, 102–109, 129
Q

Quick WiFi, 26

hasanintroduction2012.md 346 KB Histórico Raw

status: converted

Intelligent Transport Systems

Intelligent Transport Systems

Intelligent Transport Systems

-Springer Science+Business Media New York 2013

Foreword

Preface

Acknowledgments

Contents

Abbreviations

Chapter 1 Introduction

1.1 802.11 Wireless LANs

1.2 Expanding the Mobility Domain of WLANs

1.2.1 Vehicular Communications

1.2.2 V2V and R2V Communications

1.2.3 Wireless Technologies for Vehicular Communications

Cellular Networks

802.16 Networks

802.11 Networks

1.3 Challenges in 802.11-Based Vehicular Communications

1.4 Summary

Chapter 2 Vehicular Communication: Issues and Standards

2.1 Disruption Tolerant Networking

2.1.1 Systems and Architectures

2.1.2 New and Modified Protocols

2.1.3 Prediction-Based Techniques

2.2 Handover Latency in Wireless Networks

2.2.1 Detection, Search, and Probing Delay

2.2.2 Authentication and Address Allocation Delay

2.2.3 Handovers in Vehicular Context

2.3 IEEE Standards for Vehicular Communication

2.3.1 Wireless Access in Vehicular Environments—802.11p

2.3.2 Fast Transition—802.11r

2.3.3 High Throughput—802.11n

2.4 Summary

Chapter 3 Evaluation of WLAN Parameters in Vehicular Setup

3.1 Parameters of Interest

3.2 Measurement and Analysis

3.2.1 Signal Strength

3.2.2 Data Rates

3.2.2.1 Experimental Setup

3.2.2.2 Observations and Analysis

3.2.3 Correlation Between Data Rates and RSS

3.3 Application: Traffic Congestion Monitoring

3.3.1 Extended MULE Concept

3.3.1.1 Comparison with Other Works

3.3.1.2 X-MULE Issues

3.3.2 Roadside Infrastructure

3.3.2.1 Encounter Duration

3.3.3 Communication Mechanism

3.4 Summary

Chapter 4 Markov Model for R2V Communications

4.1 Markov Models

4.1.1 Fundamentals of Markov Chains

4.1.2 Markov Process in R2V Communications

4.2 Estimating the Transition Probability

4.2.1 Data Collection

4.2.2 Probability Distribution of Dataset

4.2.3 Calculating Transition Probability

4.2.4 Long-Term Error Rate

4.3 3-State Markov Model

4.4 Toward Hidden Markov Model

4.5 Summary

Chapter 5 Measuring Disruption in R2V Communications

5.1 Hidden Markov Models

5.2 HMM Representation of R2V Communication

5.2.1 Model Structure

5.2.2 Estimating Model Parameters

5.2.3 Model Generality, Limitations, and Need

5.2.3.1 Online and Offline Calculations

5.3 Observation Sequence of HMM

5.4 Probabilistic Measures of Disruption

5.4.1 Forward Algorithm

5.4.2 State Probability

5.4.3 Encounter Probability

5.5 Traffic Pattern Analysis

5.5.1 Drive Tests

5.5.2 Variation in Disruption with Traffic Patterns

5.6 Summary

hasanintroduction2012.md 346 KB

Histórico Raw