Enabling Wireless Technologies for Green Pervasive Computing

EURASIP Journal on Wireless Communications and Networking

Guest Editors: Naveen Chilamkurti, Sherali Zeadally, Abbas Jamalipour, and Sajal k. Das Enabling Wireless Technologies for Green Pervasive Computing EURASIP Journal on Wireless Communications and Networking

Enabling Wireless Technologies for Green Pervasive Computing

This is a special issue published in volume 2009 of “EURASIP Journal on Wireless Communications and Networking.” All articles are open access articles distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Editor-in-Chief Luc Vandendorpe, Universite´ catholique de Louvain, Belgium

Associate Editors

Thushara Abhayapala, Australia Zabih F. Ghassemlooy, UK Marc Moonen, Belgium Mohamed H. Ahmed, Canada Christian Hartmann, Germany Eric Moulines, France Farid Ahmed, USA Stefan Kaiser, Germany Sayandev Mukherjee, USA Carles Anton-Haro,´ Spain George K. Karagiannidis, Greece Kameswara Rao Namuduri, USA Anthony C. Boucouvalas, Greece Chi Chung Ko, Singapore AmiyaNayak,Canada Lin Cai, Canada Visa Koivunen, Finland Claude Oestges, Belgium Yuh-Shyan Chen, Taiwan Nicholas Kolokotronis, Greece A. Pandharipande, The Netherlands Pascal Chevalier, France Richard Kozick, USA Phillip Regalia, France Chia-Chin Chong, South Korea Sangarapillai Lambotharan, UK A. Lee Swindlehurst, USA Soura Dasgupta, USA Vincent Lau, Hong Kong George S. Tombras, Greece Ibrahim Develi, Turkey DavidI.Laurenson,UK Lang Tong, USA Petar M. Djuric,´ USA Tho Le-Ngoc, Canada Athanasios Vasilakos, Greece Mischa Dohler, Spain Wei Li, USA Ping Wang, Canada Abraham O. Fapojuwo, Canada Tongtong Li, USA Weidong Xiang, USA Michael Gastpar, USA Zhiqiang Liu, USA Xueshi Yang, USA Alex B. Gershman, Germany Steve McLaughlin, UK Lawrence Yeung, Hong Kong Wolfgang Gerstacker, Germany Sudip Misra, India Dongmei Zhao, Canada David Gesbert, France Ingrid Moerman, Belgium Weihua Zhuang, Canada Contents

Enabling Wireless Technologies for Green Pervasive Computing, Naveen Chilamkurti, Sherali Zeadally, Abbas Jamalipour, and Sajal K. Das Volume 2009, Article ID 230912, 2 pages

A Reputation System for Traﬃc Safety Event on Vehicular Ad Hoc Networks, Nai-Wei Lo and Hsiao-Chien Tsai Volume 2009, Article ID 125348, 10 pages

A Cross-Layer Routing Design for Multi-Interface Wireless Mesh Networks, Tzu-Chieh Tsai and Sung-Ta Tsai Volume 2009, Article ID 208524, 8 pages

Intelligent Decision-Making System with Green Pervasive Computing for Renewable Energy Business in Electricity Markets on Smart Grid, Dong-Joo Kang, Jong Hyuk Park, and Sang-Soo Yeo Volume 2009, Article ID 247483, 12 pages

GRAdient Cost Establishment (GRACE) for an Energy-Aware Routing in Wireless Sensor Networks, Noor M. Khan, Zubair Khalid, and Ghufran Ahmed Volume 2009, Article ID 275694, 15 pages

Achievable Throughput-Based MAC Layer Handoﬀ in IEEE 802.11 Wireless Local Area Networks, SungHoon Seo, JooSeok Song, Haitao Wu, and Yongguang Zhang Volume 2009, Article ID 467315, 15 pages

On PHY and MAC Performance in Body Sensor Networks, Sana Ullah, Henry Higgins, S. M. Riazul Islam, Pervez Khan, and Kyung Sup Kwak Volume 2009, Article ID 479512, 7 pages

Problem Solving of Low Data Throughput on Mobile Devices by Artefacts Prebuﬀering, Ondrej Krejcar Volume 2009, Article ID 802523, 8 pages

A Potential Transmitter Architecture for Future Generation Green Wireless Base Station, Vandana Bassoo, Kevin Tom, A. K. Mustafa, Ellie Cijvat, Henrik Sjoland, and Mike Faulkner Volume 2009, Article ID 821846, 8 pages

Modeling Energy Consumption of Dual-Hop Relay Based MAC Protocols in Ad Hoc Networks, Rizwan Ahmad, Fu-Chun Zheng, and Micheal Drieberg Volume 2009, Article ID 968323, 11 pages Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 230912, 2 pages doi:10.1155/2009/230912

Editorial Enabling Wireless Technologies for Green Pervasive Computing

Naveen Chilamkurti,1 Sherali Zeadally,2 Abbas Jamalipour,3 and Sajal K. Das4

1 Department of Computer Science and Computer Engineering, La Trobe University, Melbourne, Victoria 3086, Australia 2 Department of Computer Science and Information Technology, University of the District of Columbia 4200, Connecticut Avenue, N.W. Washington, DC 20008, USA 3 School of Electrical and Information Engineering, University of Sydney, NSW 2006, Australia 4 Department of Computer Science and Engineering, University Texas at Arlington, P.O. Box 19015, Arlington, TX 76019, USA

Correspondence should be addressed to Naveen Chilamkurti, [email protected]

Received 31 December 2009; Accepted 31 December 2009

Copyright © 2009 Naveen Chilamkurti et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Wireless pervasive computing is a rapidly growing area that In “A reputation system for trafficsafetyeventon has attracted significant attention in recent years because vehicular ad hoc networks”, Lo and Tsai propose an event- of its ubiquitous deployment throughout our society. It based reputation system that avoids the dissemination of is widely accepted that pervasive computing is having a false traffic warning messages. The major benefits of such tremendous impact on the way we live our life, conduct a system include improving traffic safety and reducing businesses, communicate with each other, and many other driving time because in the later case false warning messages areas of life. Most of the technology enablers that make lead to unnecessary driving and increases the amount of wireless pervasive computing feasible have been deployed fuel consumed. In “A cross-layer routing design for multi- with little consideration to the energy they consume during interface wireless mesh networks”, Tsai and Tsai address their operations. A recent study [1] found that if the current the issue of transmission power control and routing path cost of power consumption continues to increase it may selection for wireless mesh networks. They propose a cross- reachtoapointwhereitwillcostmoremoneytopower layer routing protocol called M2iRi2 that takes into account than to purchase information technology equipment such both the transmission power and intra/interflow interference as servers. It is therefore imperative that new standards, as routing metrics to improve network throughput and end- protocols, and approaches are explored to minimize power to-end delay. consumption of various components (such as network inter- There has been a growing interest in renewable energy face cards, switches, routers, wireless/mobile devices, laptops, sources recently. Various types of renewable energy gen- and desktops) that constitute the IT infrastructure [2]. To erators are being designed and deployed to maximize the address the challenges of efficient energy consumption for production of renewable energy from natural sources. In wireless pervasive computing, this special issue focuses not “Intelligent decision-making system for renewable energy only on the area of green pervasive computing but also on business in e-commerce based electricity markets on smart- emerging technologies that are used to support sustainability grid”, Kang et al. propose an Intelligent Decision Making and minimize carbon emission. This issue highlights some System (IDMS) that is designed to address a major deficiency of the latest recent research results achieved for the various of current renewable generators namely, their sporadic technologies (such as wireless sensor networks, Smart Grid, output which is often difficult to predict and control. IDMS wireless local area networks, etc) currently being deployed can be deployed to improve operation of intelligent power and used in wireless pervasive computing and presents systemssuchasSmartGrids. innovative solutions that can lead to a more sustainable, In their paper, “GRAdient cost establishment (GRACE) greener computing environment. for an energy-aware routing in wireless sensor networks”, 2 EURASIP Journal on Wireless Communications and Networking

Khan et al. investigate the design of a dynamic energy-aware compatible with existing networks. In “Modeling energy routing protocol for wireless sensor networks. The proposed consumption of dual-hop relay based MAC protocol n routing protocol requires low power and communication ad hoc networks”, Ahmad et al. present an analytical bandwidth under dynamic network conditions and prolongs energy consumption model for dual-hop relay based MAC the lifetime of the underlying network. Their simulation protocols. This model can predict energy consumption in results demonstrate that the proposed routing protocol ideal environment and with transmission errors. It is shown achieves the desired performance objectives. that using a relay dual-hop model, a better throughput and In “Achievable throughput-based MAC layer handoff in energy efficiency are achieved. IEEE 802.11 wireless local area networks”, Seo et al. propose We would like to take this opportunity to thank the a Media Access Control (MAC) layer handoff mechanism Editor-in-Chief of EURASIP Journal on Wireless Commu- for IEEE 802.11 wireless local area networks to improve nications and Networking, Professor. Luc Vandendorpe and the performance of bandwidth greedy applications. By all staff at Hindawi Publishing Corporation for their support exploiting the Transient Frame Capture technique as an “on- during the preparation of this Special issue. We express our the-fly” approach, an optimal Access Point is selected using gratitude to all the anonymous reviewers who devoted much maximum achievable throughput as metric instead of the of their precious time reviewing all the papers submitted traditional signal strength approach. The major benefit of to this special issue. Their timely reviews greatly helped us such an approach is that it can be deployed without requiring select the best papers included in this issue. We also thank all any changes to Access Points since only the client needs authors who contributed to this special issue. software modifications. The fairness of the proposed handoff Finally, we hope you will enjoy reading this selection of scheme is also demonstrated and validated with extensive papers and you will find this issue informative and helpful in simulations. keeping yourselves up-to-date in the field of green pervasive Body Senor Networks (BSNs) are becoming increasingly computing. important for sporting activities and other healthcare systems. In “on PHY and MAC performance in body sensor Naveen Chilamkurti networks”, Ullah et al. present an empirical investigation Sherali Zeadally on the performance of body implant communication using Abbas Jamalipour Radio Frequency (RF) technology. They used a model Sajal K. Das mimicking electrical properties of the basic body tissue and observed best performance at 3cm depth inside the liquid References model. They also studied performance of low-power MAC protocols for an on-body sensor network using simulations. [1] “Report to Congress on Server and Data Center Energy ffi The mobile devices such as PDA devices and Smart E ciency. Public Law 109–431,” ENERGY STAR Program, Phones are commonly used with Internet connection. As U.S. Environmental Protection Agency, 2007, http://www these devices have low-performance components due to very .energystar.gov/ia/partners/prod development/downloads/EPA Datacenter Report Congress Final1.pdf. limited space, this can cause limited connection problems [2] N. Chilamkurti, S. Zeadally, and F. Mentiplay, “Green net- when connected to an online system with large artefacts working for major components of information communication data files. Ondrej Krejcar in “Problem solving of low data technology systems,” EURASIP Journal on Wireless Communi- throughput on mobile devices by artefacts prebuffering” cations and Networking, vol. 2009, Article ID 656785, 7 pages, uses a model of data Prebuffering to achieve high data 2009. throughput. Using a real-time setup built on purpose for this experiment, he proved that accessing prebuffered data on a mobile device can significantly improve response time needed to view large multimedia data. Fourth generation wireless systems require higher transmission rates which in turn need more power. As the world is preparing to reduce power consumption, mobile operators are looking at ways to reduce their operating costs and carbon footprint. In “A potential transmitter architecture for future generation green wireless base station”, Bassoo et al. propose a potential architecture design for future Green wireless base station. The all-digital transmitter architecture uses a combination of envelop elimination and restoration (EER) and pulse width modulation (PWM) modulation. The performance of this model shows that 57% efficiency can be obtained for an OFDM signal limited to 8-dB peak to average power ratio. Next Generation Networks (NGN) are expected to provide high throughput, low latency, and better quality of service. These NGNs are also required to be backward Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 125348, 10 pages doi:10.1155/2009/125348

Research Article A Reputation System for Trafﬁc Safety Event on Vehicular Ad Hoc Networks

Nai-Wei Lo and Hsiao-Chien Tsai

Department of Information Management, National Taiwan University of Science and Technology, No. 43, Section 4, Keelung Road., Taipei 106, Taiwan

Correspondence should be addressed to Hsiao-Chien Tsai, [email protected]

Received 28 February 2009; Accepted 15 September 2009

Recommended by Naveen Chilamkurti

Traffic safety applications on vehicular ad hoc networks (VANETs) have drawn a lot of attention in recent years with their promising functions on car accident reduction, real-time traffic information support, and enhancement of comfortable driving experience on roadways. However, an inaccurate traffic warning message will impact drivers’ decisions, waste drivers’ time and fuel in their vehicles, and even invoke serious car accidents. Toenable eco-friendly driving VANET environments, that is, to save fuel and time in this context, we proposed an event-based reputation system to prevent the spread of false traffic warning messages. In this system, a dynamic reputation evaluation mechanism is introduced to determine whether an incoming traffic message is significant and trustworthy to the driver. The proposed system is characterized and evaluated through experimental simulations. The simulation results show that, with a proper reputation adaptation mechanism and appropriate threshold settings, our proposed system can effectively prevent false messages spread on various VANET environments.

Copyright © 2009 N.-W. Lo and H.-C. Tsai. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction protocols and systems [8, 9, 11, 12] have been proposed to ensure message authentication and message integrity. On the There are 1.2 million people killed and as many as 50 million other hand, to determine whether the trafficeventreported people injured in traffic accidents each year [1]. In order by a warning message is really occurred, voting schemes [10] to preserve people’s lives, trafficsafetyapplications[2]on and data-centric trust establishment mechanism [13]have vehicular ad hoc networks [3] have been developed in recent been proposed recently to evaluate the trustworthiness of the years by broadcasting real-time warning messages [4](e.g., message content. car accident, traffic jam, obstacle detection, etc.) through In previously published works, generally vehicles are vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) assumedtobeabletodetecttraffic events along the road communication channels from one vehicle (or base station) all the time. However, this simple assumption may not be to other vehicles in order to notify drivers to avoid awful practical in a real world. First of all, some types of traffic trafficsituationsinadvance[5–7]. events (e.g., traffic jam) usually change their status such as Traffic safety applications enhance the safety of drivers location, size, or intensity over time [4]. In consequence, on the road. However, a false traffic warning message, that is, an inaccurate warning message may be broadcast if the the message with inaccurate traffic information, will impact corresponding traffic safety application does not consider drivers’ behaviors and increase the occurrence possibility the dynamics of event status. Secondly, sensors used to of traffic accidents. A malicious attacker can create bogus detect traffic events on a vehicle may have different levels of traffic warning messages and cause intelligent collisions [8]. detection capabilities, which are dependent on correspond- In addition, false warning messages can waste drivers’ time ing manufacture specifications. When vehicles encounter and fuel of vehicles [9, 10]. To prevent false traffic warning the same traffic event, those who only equipped with less messages spread on VANET, various secure communication powerful sensors may not be able to detect the event as 2 EURASIP Journal on Wireless Communications and Networking others do. In addition, the detection ratio of traffic event is Picconi et al. [15] proposed a solution to validate an affected by vehicle mobility. As data collections on sensors aggregated message with probabilistic signature checking are performed between each sampling period of time, there mechanism. The proposed scheme is used to verify vehicle- exists the possibility that a vehicle cannot sense or record an related information such as the current speed and geographic encountered traffic event during its high-speed movement. location, not traffic events occurred along the road. In In order to filter out inaccurate messages caused by the addition, a malicious vehicle may be able to circumvent the dynamics of traffic event and vehicles with different detection checking scheme if its false messages are far less than all capabilities on embedded sensors, and false messages spread transmitted messages in a VANET. by malicious attackers in VANET, an event-based reputation In general, it is difficult for a vehicle to determine the system is introduced in this paper. Our design concept plausibility of a reported traffic event solely. In [16]Raya is to determine whether a traffic event exists and how et al. applied message aggregation and group communica- long it lasts through distributed vehicle observations. The tion to validate a reported traffic event. The main idea status of a traffic event is stored and managed in each is to provide a vehicle more evidence about a reported vehicle which has encountered it or is aware of it from traffic event by collecting and analyzing multiple incoming received messages. A trafficeventwillbebroadcastbya messages from different vehicles. The main challenge of this vehicle through message transmission only if this event has paper is how to dynamically form and maintain a vehicle accumulated enough reputation credits on event intensity group with the characteristic of high mobility. The concept and event reliability in this vehicle. We evaluate and analyze of message aggregation is also adopted by Ostermaier et al. the performance of the proposed system by performing in [10]. The authors proposed four voting schemes on local network simulation experiments. The simulation results danger warning service. Their simulation results showed that reveal that the event-based reputation system is applicable one of the four schemes, called majority of freshest votes to most VANET environments and can successfully fil- with a threshold, sounds promising. However, the dynamics ter out false traffic warning messages. Consequently, our of traffic events and the differences of sensor capabilities reputation system can improve the safety of drivers on may cause some sensors to collect inaccurate information the road. when vehicles pass the same event location. In consequence, The rest of this paper is organized as follows. In Section 2, it is hard for voting vehicles to achieve an agreement on related work is discussed. In Section 3, we describe the areportedtraffic event and to further evaluate the event system model on which our reputation system is based. correspondingly based on the voting scheme. The proposed event-based reputation system is introduced Maya et al. [13] proposed a data-centric trust establish- in Section 4. The results and analyses of simulation experiment framework and applied it to the traffic safety appli- ments for the proposed reputation system are presented in cation in VANET. The novel concept in [13]istoevaluate Section 5. Finally, we give the conclusion in Section 6. the trustiness of sensed data or received messages rather than the trust of individual vehicle. However, the authors did not 2. Related Work consider the effect introduced by the dynamics of traffic events. A vehicle may not detect an occurred trafficevent The fraud message problem of trafficsafetyapplication or may collect imprecise data due to its sensor limitation on VANET has been studied extensively. Various secure when passing the occurrence location of this trafficevent; communication protocols have been proposed to provide consequently, for a vehicle, the evaluation result on the message authentication and integrity [8, 9, 11, 12]. In the trustiness of generated data (or received messages) regarding following, we review the development progress on reputation to the observed (or reported) trafficeventmaynotbefully evaluation scheme based on recently published research accurate and trustworthy. works [10, 13–16]. In summary, if we consider a practical VANET environ- Golle et al. [14] proposed a general approach to evaluate ment, inaccurate or imprecise traffic information caused by the validity of message data generated in VANET. In their dynamics of traffic events, differences of sensor capabilities, scheme, every vehicle builds a model for VANET environ- and interference of vehicle mobility will be generated and ment in which specific rules and statistical properties are aggregated to a reputation (or trust establishment) system implemented to validate message data received from other almost inevitably. Under such situations, related trust evalu- vehicles. The same concept for trustworthiness evaluation is ationsystemsandframeworksfrompreviousresearchworks also adopted later in [11, 17]. Golle et al. assumed that a node cannot function properly and effectively since aggregated (vehicle) always trusts the data generated from its own on- imprecise messages will produce false alarms to trafficsafety board sensors. In consequence, errors from sensor-generated applications. In Sections 3 and 4, we propose an event-based data, caused by malfunctioned sensors, dynamics of traffic reputation system to provide accurate and reliable traffic events (e.g. the speed of a vehicle is too fast for its sensors information to vehicle drivers and resist the false alarm effect to detect surrounding environment and gather meaningful from fraud messages spread in the network at the same time. or error-free data), and data manipulation from a malicious attacker (vehicle), were not considered in their system model. 3. Model of Reputation System As their system model requires offline construction and parameter calibration, system flexibility and scalability may 3.1. Network Model. Traditional traffic safety applica- become an issue. tions collect traffic related information with roadside EURASIP Journal on Wireless Communications and Networking 3 infrastructure and transmit traffic information to traffic operation centers through wired network. Because the cost Wireless Sensors Traffic information for deployment and management is relatively high, tradi- interface tional traffic safety applications are only deployed in certain areas. In brief, the traditional solution is not economic and eco-friendl, and cannot provide traffic information for drivers’ safety effectively and pervasively. As a VANET Event-based reputation system (ERS) does not require high-cost infrastructure and centralized traffic operation center to collect trafficevents,aVANET is more economic than traditional wired network solution. Event reputation value collection Furthermore, in a VANET environment, traffic information is collected and distributed by each vehicle; therefore, real- Event Event Event confidence list table time and effective traffic information can be broadcast in a management collection driver-concerned local area quickly and pervasively. Thus, we Reputation value adopt VANETs as our network environment. As the proposed adaptation module event-based reputation system will be implemented in the application layer of OSI (Open System Interconnection) network architecture, the proposed system is independent from lower OSI layers. Actually, the system can leverage Light Speaker Monitor User interface novel wireless technologies (e.g., WiMAX, IEEE 802.11p) to improve its overall performance as new wireless technologies or standards provide longer transmission range, larger Figure 1: System architecture of the proposed event-based reputa- bandwidth, and better mechanisms (e.g., routing schemes). tion system.

3.2. Models of Vehicle and Its Traffic Safety Application. We assume that each vehicle equips with a positioning device, 4. Event-Based Reputation System such as GPS (Global Positioning System). Multiple sensors with various data collection capabilities are installed in Our event-based reputation system (ERS) is enlightened by every vehicle. The details of data collection techniques of the cooperation enforcement schemes proposed in mobile sensors are beyond the scope of this paper. Vehicle mobility ad hoc networks [18], where nodes collaboratively observe and device specification make the event detection capability neighbors and broadcast warnings if misbehaved nodes were among similar sensors different with each other. In terms discovered. The system architecture is illustrated in Figure 1. of vehicle mobility, as traffic-related data collection with sensors is not performed in real time, it is possible for an 4.1. System Overview. ERS is composed of three interfaces, on-board sensor to overlook or miss the event signal when four functionalities, and one repository for table storage. the speed of the vehicle is over a certain sensor threshold. Traffic information comes either from received messages On the other hand, a sensor can detect the same event many via wireless interface or from on-board sensors. The event times when the vehicle is moving slowly. In terms of device table in ERS stores all received and derived trafficevent specification, the event detection capability of a sensor is information including event identity, type of trafficevent, mainly dependent on its manufacture specification. When occurrence timestamp, event location, message transmission vehicles encounter the same traffic event, vehicles with better range, event reputation value, and event confidence list. In sensors can easily detect the event but the others cannot. the event table, each record entry stores a distinct traffic When the value of an event data gathered by a sensor event. Event reputation value defines the intensity degree is over the predefined safety threshold, the information is of a traffic event and its initial value is always set to zero. sent to the traffic safety application in the vehicle. Based on A simple algorithm is adopted to compute the value of event the evaluation results from the proposed reputation system, reputation for a specific traffic event: (1) every time the given the traffic safety application will determine to broadcast vehicle’s ERS detects this event with its on-board sensors, the traffic warning messages to neighboring vehicles or not. value is increased by one; (2) when the given ERS receives a The transmission distance of a broadcast message depends traffic warning message from another vehicle, the ERS adds on the type of traffic event or the configuration of the the event reputation value in the received message into the traffic safety application. The neighbors that received the field of event reputation value at the same event record in warning messages can autonomously determine how to the event table or creates a new event record in the event react based on their own traffic safety application and table. Event confidence value indicates the reliability extent preconfigured policies. We assume that the type definition of a traffic event and the value is the number of distinct and granularity of a traffic event is properly defined and vehicles whose messages, regarding to the same trafficevent, agreed among various traffic safety applications in advance. have been received by the given vehicle’s ERS. In addition, Traffic event information with slight difference (below a the definition of event confidence list is a string list of predefined threshold), such as observed timestamp, will be the identities of distinct vehicles which encounter the same viewed as the same trafficevent. traffic event. When a given vehicle encounters a trafficevent 4 EURASIP Journal on Wireless Communications and Networking and detects it, the given ERS will append its vehicle’s identity message revocation scheme, the reputation value adaptation into the event confidence list field at the corresponding mechanism is introduced in ERS. event entry. Similarly, when a given vehicle receives a traffic The reputation value adaptation mechanism utilizes two warning message, the content of event confidence list in functions to control the corresponding event reputation the message will be appended in the event confidence list value of a detected event during the event’s lifetime so that field at the corresponding event entry. In an event record, the event status (resolved or not) is reflected by its reputation event identity represents the identity of trafficevent.Type value. The first function is the reputation value suppression of traffic event implies the predefined event type of this function which sets the event reputation value of an event event. Occurrence timestamp and event location indicate the record as the event reputation threshold if the reputation time and location when a traffic event is detected by a vehi- value of this event record is greater than the predefined cle. Message transmission range represents the predefined reputation threshold. Reputation value suppression function transmission distance in hop count for the traffic warning helps ERS to control the maximum value of reputation message. measurement. The four functions supported in the ERS are event The second function is the reputation value degradation management, reputation value adaptation module, event function which is used to decrease the event reputation value reputation value collection, and event confidence list col- of an event record in the event table according to the length of lection. We will introduce the first two functions in the event lifetime. As time passes, the existence possibility of an next subsection. For the two collection functions, we have unresolved traffic event decreases very quickly. For each event briefly illustrated how these functions work as previously record in the event table, a distinct software timer starting stated in this subsection. Here we want to introduce two with the predefined time period Td is invoked to trigger the important thresholds used in ERS, that is, event reputation reputation value degradation function automatically when threshold and event confidence threshold. Event reputation the timer is expired. The updated event reputation value threshold is used to set up the barrier for event intensity. of an event record is calculated by the reputation value If the event reputation value of a traffic event is higher degradation function. Equation (1) indicates the reputation than the predefined event reputation threshold, then the degradation formula in which Ru represents the updated intensity of this event is sufficiently strong enough to indicate reputation value, Rp means the previous reputation value the continuous existence of this event. Otherwise, the event before the timer expired, D( ) is a preselected degradation might not still exist anymore, even though it did occur function to control the degradation speed of an event sometime before. Event confidence threshold is used to set up reputation value, and Nte indicates the total number of timer the bottom line for event reliability. If the event confidence expiration times for an event record since it has been updated value of a traffic event is higher than the predefined last time. Notice that for an event record the ERS resets event confidence threshold, then it indicates that there the value of corresponding Nte to zero when the ERS has were sufficient amounts of vehicles that encountered the received the same event message later from others or detected same traffic event and the occurrence plausibility of this the same event by itself. When the event reputation value event is much more reliable. By properly setting these of an event record decreases to zero, the ERS will remove thresholds and other configurable system parameters, the the corresponding traffic warning notification on the user ERS can provide accurate and reliable traffic information interface and the event entry in the event table: to vehicle drivers. If a given ERS detects the event reputation value and the event confidence value of a traffic Ru = Rp − D(Nte). (1) event is over the corresponding event reputation threshold and event confidence threshold, which indicate that the In general, these two functions in the reputation adap- traffic event really exists and is still there, the ERS will tation mechanism, that is, the algorithm for reputation send this event information through the user interface value accumulation and the degradation function D()for to notify the driver and at the same time broadcast reputation decrease, can be flexibly defined and constructed atraffic warning message with current event reputation based on practical VANET environments in real world. value and the corresponding confidence list to nearby vehicles. 4.3. Configuration of Event Reputation Threshold and Event Confidence Threshold. Configuration of event reputation 4.2. Traffic Event Management. As the status of a traffic threshold and event confidence threshold in an ERS are event changes dynamically and the detection capabilities of dependent on the sensor capability of a vehicle and the type sensors in various kinds of vehicles are different, a vehicle characteristics of a traffic event. In general, there are some not detecting new traffic event at a specific location and time design criteria and guidelines to help vehicle manufacturers does not imply that there is no event occurred now or before. or drivers determine these two thresholds. For example, Therefore, some trafficsafetyapplications[10, 13]actively when instant notification of event occurrence is more impor- send traffic revocation messages to inform other vehicles tant than event reliability and event continuity in situations when an event is resolved. However, this mechanism might such as emergency braking event and speed decrease event, provide wrong event information to other vehicles if the both thresholds should be set to a lower value. On the sending vehicle of the original revocation message misjudges contrary, if event reliability and event continuity are more the event status. In order to eliminate the weakness of event important than instant notification of event occurrence in EURASIP Journal on Wireless Communications and Networking 5

(0, 1200) (1200, 1200)

Moving direction 100 m

V1 E V 1 5 Event

V2 V3

V4 (700, 600) 200 m 100 m

Figure 2: A vehicle (V1) encounters a traﬃc event (E1)and transmits the traﬃc warning message to other vehicles.

100 m 200 m 100 m situations such as vehicle accident event and trafficjamevent, (0, 0) (1200, 0) both thresholds should be set to a higher value. Therefore, Figure 3: The street map used in our simulations. The location we suggest that different pairs of event reputation threshold coordinate of the marked traffic event is at (700, 600). and event confidence threshold should be preconfigured in an ERS based on various event types and sensor capability of vehicle. each vehicle. Suppose that when V2 passes the location of E1, its sensors detect E1 8 times. Then V2 updates the reputation 4.4. An Illustrated Example. We adopt a simple example to value of this event to 9 (i.e., 1 + 8 = 9) and adds its identity illustrate the operation flow of the ERS in this subsection. to the confidence list of this event [V1, V2] in the event Assume that all vehicles have ERS installed and configured record. As the event reputation value of E1 in V2 is greater with the event reputation threshold, the event confidence than the preconfigured reputation threshold, the reputation threshold, and the message transmission range (in hop suppression function in the ERS is invoked to reset the count) been set as 8, 2, and 3, respectively. reputationvalueto8.Now,inV2 the event reputation value ffi As shown in Figure 2, there is a tra ceventE1 on a road. and the number of vehicle identities in the event confidence Assume that the vehicle V1 passes the location of event E1 and list for the event E1 have both reached the reputation the sensors on V1 have detected E1 3 times along the path. In threshold and the confidence threshold. Therefore, the ERS ffi consequence, the ERS in V1 stores this tra c information, in V2 will send the information of this reliable trafficevent v1 = sets the reputation value of this event as 3 (i.e., R 3), E1 through the user interface to notify its driver and then and inserts its vehicle identity V1 into the event confidence broadcast this traffic warning message with the reputation v2 list in the corresponding event entry. Next, V1 generates a value R = 8 and the confidence list [V1, V2]toneighbor ffi new tra c warning message for the event E1 that includes vehicles. Vehicles that receive this traffic warning message ffi v1 = the tra c information, the reputation value R 3, and the from V2 will repeat the same operation process of V2 as ffi confident list [V1]. Then V1 broadcasts the tra c warning described previously. message to its neighbors. Assume vehicles V2, V3, V4,and ffi V5 have received this tra c warning message. All four of 5. System Evaluation them will record this event E1 and store the corresponding traffic information, the event reputation value (Rv2 = Rv3 = Network simulator ns-2 [19] is used to evaluate system Rv4 = Rv5 = Rv1 = 3), and the event confidence list (each performance of the proposed event-based reputation system vehicle is [V1] in this case) into their individual event tables; (ERS). IEEE 802.11b DCF is adopted for the MAC layer however, the ERS systems in these four vehicles will not setting in our simulations. Omnidirectional antenna with notify their drivers this incoming traffic information and also 250-meter transmission range is assumed. The simulation not forward it, even though the message transmission range scenario is set in a grid-typed street map. As shown in of this event does not reach to zero (i.e., 3 − 1 = 2), because Figure 3, the map is constructed by 5 × 5 street blocks both the event reputation value and the event confidence and the size of each block is 200 square meters. For each value of this event do not reach the preconfigured thresholds. simulation 100 vehicle nodes are generated and randomly Assume that vehicles V2, V3,andV4 keep moving toward placed on roads in the scenario map. The traffic event is the location of event E1 after receiving the warning message assumed to be at location coordinate (700, 600). To reflect from V1.BeforeV2 encounters E1, the event reputation the dynamic status of a traffic event, the simulating event will values of event E1 in V2, V3, V4,andV5 all decrease to 1 duo occur at the 100th second and be resolved at the 400th second to the execution of event reputation degradation function in based on our simulation settings. The simulation time in 6 EURASIP Journal on Wireless Communications and Networking each run is 700 seconds. Each measured result (point) in the 20 following diagrams is an average number obtained from 500 18 replications of simulation runs. 16 We develop a new vehicle mobility model called random intersection, which is inspired by the trafficsignmodel 14 proposed in [20], to simulate the dynamic status of a 12 vehicle driving around in an urban area. In the beginning 10 each vehicle is randomly assigned a moving speed between 8 10 km/h and Smax km/h with a randomly determined driving 6 direction from its location, where Smax is the maximal moving speed predefined in the simulation environment. Average reputation value 4 ffi In our scenario map, all road intersections have tra c 2 lights. When a vehicle approaches a road intersection, it will 0 encounter a traffic light. The probability for a vehicle to stop 0 100 200 300 400 500 600 700 at a traffic light is set to 50%. The duration of a red light Simulation time (s) is randomly decided between 0 and 40 seconds. To simulate traffic delay situation at intersections, a vehicle always stops 20 km/h 80 km/h for 2 seconds at an intersection. Note that this time duration 40 km/h 100 km/h is independent with traffic light signals. Once the time 60 km/h duration for a vehicle to stop at an intersection is expired, Figure 4: Average accumulation speed of event reputation value to the vehicle randomly reselects its moving speed within the vehicles under different vehicle mobilities. preconfigured speed range and its next moving direction. Note that the speed legends in the following simulation figures all indicate the maximal moving speed of a vehicle. The sampling interval of on-board sensors in a vehicle is value in an ERS is faster when vehicle mobility is low in a set to one second and event detection distance is set to 16 VANET. As the sampling interval of on-board sensors in a meters in total; that is, sensors installed at the head and the vehicle is set as one second, vehicles passing the event with a rear of a vehicle can both detect events occurred in front of low speed such as 20 km/h can detect the event many times them less than 8 meters away. The parameter setting for on- in general. Contrarily, when vehicles pass the event at a high board sensors makes the event detection capability of each speed such as 100 km/h, their on-board sensors may not be vehicle depending on its moving speed. For ERS settings, able to react in time and detect the event. Consequently, the time period to trigger the reputation value degradation the corresponding accumulation speed of event reputation function is set to 15 seconds (i.e., Td = 15). value becomes slower. The accumulation speed of average event confidence value to vehicles under different vehicle 5.1. Effect of Vehicle Mobility and TrafficDensity. In VANET mobilities is shown in Figure 5. Contrary to the simulation environments, high vehicle mobility situation and low results on event reputation value, the increment of the event traffic density situation are main performance challenges confidence value in an ERS is faster when vehicles move for application systems. To evaluate the applicability of ERS at a high-speed. As vehicles move faster, the event will be under high vehicle mobility and low traffic density situations, encountered by those vehicles in a shorter time period; in we analyze the average accumulation speed for vehicles on consequence, the identity of each vehicle will be added to the event reputation value and event confidence value under event confidence list field of the corresponding event record different vehicle mobility and traffic density. Here we define in its event table. When vehicle speed varies from 60 km/h the average event reputation value as the average of the two to 100 km/h, the increment of average event confidence largest event reputation values among all vehicles at a specific value is not proportional to the increase of vehicle speed. simulation timestamp. A similar definition for the average This is because when a vehicle moves faster, the traffic event confidence value is applied. The reason is that in a lights are encountered sooner. A high speed vehicle takes VANET the vehicle with the highest reputation value and much more portion of its driving time to wait for traffic confidence value of an occurred event will be the first node lights. to broadcast the traffic warning message to others. As the event will be resolved at the 400th second based For this part of simulation experiments, we intentionally on our simulation settings, it is reasonable that the average disable the reputation value suppression function and the event reputation value to vehicles decreases linearly starting message forwarding module in the ERS. The reputation from 400 seconds. The linear decrease is caused by the setting degradation function is set as a constant (i.e., D(Nte) = of the reputation value degradation function which is set as 1). These settings simplify our experimental environment, a constant (D(Nte) = 1) in this experiment. The ERS in a reduce the amount of output data, and allow us to concen- vehicle will delete the corresponding event confidence list trate on effect analysis. when the event reputation value becomes zero. Therefore, Figure 4 shows the accumulation speed of average event the decrement trend of average event confidence value in reputation value to vehicles under different vehicle mobil- Figure 5 is similar to the decrement trend of average event ities. It is obvious that the increment of event reputation reputation value in Figure 4. EURASIP Journal on Wireless Communications and Networking 7

10 20 9 18 8 16 7 14 6 12 5 10 4 8 3 6 Average conﬁdence value 2 Average conﬁdence value 4 1 2 0 0 0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700 Simulation time (s) Simulation time (s)

20 km/h 80 km/h 4.5vehicle/km 12.5vehicle/km 40 km/h 100 km/h 6vehicle/km 20.8vehicle/km 60 km/h 8.3vehicle/km Figure 5: Average accumulation speed of event confidence value to Figure 7: Average accumulation speed of event confidence value to vehicles under different vehicle mobilities. vehicles under different trafficdensities.

70 value of vehicles located nearby the trafficeventareaccu- mulated fast. In brief, we show that ERS is very sensitive 60 and effective to high traffic density environments. Under our simulation environment configuration, the accumulation 50 speeds for both event reputation value and event confidence value are much slower in low traffic density situations 40 compared with the speeds in high traffic density cases. 30 In practical situations, the accumulation speeds for both ERS parameters under low traffic density environments are 20 affected by other variable factors such as the trafficevent Average reputation value duration, the physical range (extent) of the trafficevent, 10 the detection capability of on-board sensors in a vehicle, the message transmission range of wireless interface in a 0 0 100 200 300 400 500 600 700 vehicle, and the moving speed of a vehicle. Based on the Simulation time (s) design logic, the ERS requires more reliable or accountable information from other vehicles and its senor components to 4.5vehicle/km 12.5vehicle/km derive correct and precise warning information. Therefore, 6vehicle/km 20.8vehicle/km in general it will take more time for ERS to react in a low 8.3vehicle/km traffic density environment. To get better performance in low Figure 6: Average accumulation speed of event reputation value to traffic density environments, the ERS can associate with high vehicles under different trafficdensities. event resolution sensors, utilize more efficient protocols in lower OSI layer such as IEEE 802.11p standard (WAVE), and extend the wireless transmission range of the vehicle with more powerful wireless signal amplifier. Toevaluate the effect of traffic density to ERS, we perform another set of simulation experiments by only varying the 5.2. Effect of Degradation Function. In this subsection we size of street map between 3 × 3 blocks and 7 × 7 blocks. As want to explore the effect caused by the degradation function the total number of vehicles is the same as before (i.e., 100 D( ) and learn how to select a proper degradation function vehicles), the traffic density in the network varies between for ERS. As shown in Figure 6, after the event is resolved at 4.5 vehicles/km and 20.8 vehicles/km. Figures 6 and 7 show the 400th second, the average reputation value decreases very that the accumulation speeds of average event reputation slow, where the degradation function is set as a constant (i.e., value and average event confidence value raise significantly D(Nte) = 1). To explore the effect of degradation function when the traffic density increases. The reason is that a to the decrease speed of event reputation value, we execute lot of traffic warning messages are generated from vehicles another experiment by setting the degradation function to which have encountered the traffic event; consequently, the Fibonacci number function D(Nte) = Fibonacci(Nte)and2- N corresponding event reputation value and event confidence based exponent function D(Nte) = 2 te ,whereFibonacci(Nte) 8 EURASIP Journal on Wireless Communications and Networking

60 60

50 50

40 40

30 30

20 20 Average reputation value Average reputation value 10 10

0 0 0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700 Simulation time (s) Simulation time (s)

20.8vehicle/km 6vehicle/km 20.8vehicle/km 6vehicle/km 12.5vehicle/km 4.5vehicle/km 12.5vehicle/km 4.5vehicle/km 8.3 vehile/km 8.3vehicle/km Figure 8: Fibonacci number function is adopted as the degradation Figure 9: 2-based exponent function is adopted as the degradation = Nte function, D(Nte) Fibonacci(Nte). function, D(Nte) = 2 .

8 indicates the corresponding value of Fibonacci Sequence in 7 the index Nte. The simulation results for Fibonacci number function 6 and 2-based exponent function are shown in Figures 8 and 9, respectively. It is obvious that both nonlinear degra- 5 dation functions provide much better decrease speed on 4 average event reputation value after the event is resolved in comparison with linear degradation function. In addition, 3 both functions do not aﬀect the accumulation speed on

Average reputation value 2 average event reputation value much while the event exists. Therefore, based on our simulation results, to improve the 1 ERS performance a nonlinear degradation function should 0 be considered instead of a linear one when installing and 0 100 200 300 400 500 600 700 conﬁguring an ERS. Simulation time (s)

ff ffi Real event, 20 km/h Real event, 80 km/h 5.3. E ect of False Tra c Warning Message. To explore the Real event, 40 km/h Real event, 100 km/h effectiveness of ERS against false message flooding attack, Real event, 60 km/h False event, 60 km/h we perform the third set of simulation experiments in this subsection. The message transmission range field in Figure 10: The comparison of average reputation value between a a warning message is set to 3 hops in length. The event real event and a false event. reputation threshold and event confidence threshold is set to 9 and 4 in the ERS, respectively. Reputation value adaptation mechanism in the ERS is fully activated in this experiment. messages to ERS. In Figure 10, the average reputation value During simulation executions, there is a randomly selected ofarealtraffic event accumulates rapidly in all kinds of vehicle node to broadcast traffic warning messages with vehicle mobility environments when the event exists. On inaccurate content every 20 seconds. The content of these the contrary, the average reputation value of a false traffic false traffic warning messages is generated randomly. A event oscillates between zero and one in all kinds of vehicle vehicle will broadcast a traffic warning message for an event mobility environments. For clearness and simplicity, we only when the corresponding event intensity and event reliability show the average reputation value of a false event with the have reached the reputation and confidence thresholds maximal vehicle speed set as 60 km/h in Figure 10.Once defined in its ERS system. the event reputation value and event confidence value of a A vehicle trusting the content of received warning real event in a vehicle reach the reputation threshold and messages and notifying its driver the false event is defined as the confidence threshold, the corresponding traffic warning a message-affected vehicle. The average number of message- message will be broadcast up to 3 hops away. Figure 11 affected vehicles is adopted to measure the influence of false shows the number of vehicles affected by a real traffic EURASIP Journal on Wireless Communications and Networking 9

80 spread to the network and the system with its conﬁguration ﬂexibility is applicable to most VANET environments. 70 60 References

ected vehicles 50 ﬃ

ff [1] World Health Organization, WorldReportonRoadTra c 40 Injury Prevention, WHO, Geneva, Switzerland, 2004. [2] J. Luo and J. P. Hubaux, “A survey of inter-vehicle communi- 30 cation,” Tech. Rep. IC/2004/24, EPFL, Lausanne, Switzerland, 2004. 20 [3] J. J. Blum, A. Eskandarian, and L. J. Huffman, “Challenges of intervehicle ad hoc networks,” IEEE Transactions on Intelligent Average number10 of a Transportation Systems, vol. 5, no. 4, pp. 347–351, 2004. 0 [4] F. Dotzer,¨ M. Strassberger, and T. Kosch, “Classification for 0 100 200 300 400 500 600 700 traffic related inter-vehicle messaging,” in Proceedings of the Simulation time (s) 5th IEEE International Conference on ITS Telecommunications (ITST ’07), Brest, France, June 2005. Real event ,20 km/h Real event,80 km/h [5] T. Nadeem, S. Dashtinezhad, C. Liao, and L. Iftode, “Traf- Real event, 40 km/h Real event, 100 km/h ficView: traffic data dissemination using car-to-car commu- Real event, 60 km/h False event, 60 km/h nication,” ACM SIGMOBILE Mobile Computing and Commu- Figure 11: The comparison of the number of affected vehicles nications Review, vol. 8, no. 3, pp. 6–19, 2004. between a real event and a false event. [6] W. Chen and S. Cai, “Ad hoc peer-to-peer network architecture for vehicle safety communications,” IEEE Communica- tions Magazine, vol. 43, no. 4, pp. 100–107, 2005. [7] S. Biswas, R. Tatchikou, and F. Dion, “Vehicle-to-vehicle event increases very fast. On the other hand, the false wireless communication protocols for enhancing highway ffi messages generated from a malicious node does not affect the tra csafety,”IEEE Communications Magazine,vol.44,no.1, judgments of other vehicles at all, since their sensors do not pp. 74–82, 2006. [8] P. Papadimitratos, L. Buttyan, T. Holczer, et al., “Secure detect the fraud traffic event on the road and consequently vehicular communication systems: design and architecture,” their ERS systems do not accumulate the event reputation IEEE Communications Magazine, vol. 46, no. 11, pp. 100–109, value and event confidence value for the false event. 2008. In Figure 10, the average reputation value for the real [9] F. Kargl, P. Papadimitratos, L. Buttyan, et al., “Secure vehicular event is always under the event reputation threshold (which communication systems: implementation, performance, and is 9) while at the same time the average number of research challenges,” IEEE Communications Magazine, vol. 46, affected vehicles in Figure 11 keeps increasing steadily during no. 11, pp. 110–118, 2008. the event’s lifetime. This is because the reputation value [10] B. Ostermaier, F. Dotzer,¨ and M. Strassberger, “Enhancing the suppression function in the ERS is activated to control the security of local danger warnings in VANETs—a simulative maximal reputation value stored in an event record. analysis of voting schemes,” in Proceedings of the 2nd Inter- In summary, the simulation results show that our national Conference on Availability, Reliability and Security proposed event-based reputation system can dynamically (ARES ’07), pp. 422–431, Pheonix, Ariz, USA, April 2007. collect event information, determine the plausibility and [11] M. Raya, P. Papadimitratos, I. Aad, D. Jungels, and J.- P. Hubaux, “Eviction of misbehaving and faulty nodes in timeliness of an event, and broadcast accurate and reliable ffi vehicular networks,” IEEE Journal on Selected Areas in Com- tra c warning messages in most VANET environments. munications, vol. 25, no. 8, pp. 1557–1568, 2007. [12] C. Laurendeau and M. Barbeau, “Threats to security in 6. Conclusion DSRC/WAVE,” in Proceedings of 5th International Conference on Ad-Hoc Networks & Wireless, vol. 4104 of Lecture Notes in Traffic safety applications on vehicular ad hoc networks Computer Science, pp. 266–279, Ottawa, Canada, August 2006. have attracted significant attention in recent years as they [13] M. Raya, P. Papadimitratos, V. D. Gligor, and J.-P. Hubaux, improve driving quality, drivers’ comfort, and drivers’ safety. “On data-centric trust establishment in ephemeral ad hoc To enable the massive usage of traffic safety application, it is networks,” in Proceedings of the 27th IEEE Conference on necessary to prevent false traffic warning alarms spread on Computer Communications (INFOCOM ’08), pp. 1238–1246, VANETs which will strongly affect drivers’ behaviors and put April 2008. drivers and passengers in danger. To eliminate the concern [14] P. Golle, D. Greene, and J. Staddon, “Detecting and cor- recting malicious data in VANETs,” in Proceedings of the 1st on traffic message plausibility, we propose the event-based ACM International Workshop on Vehicular Ad Hoc Networks reputation system (ERS) which utilizes cooperative event (VANET ’04), pp. 29–37, Philadelphia, Pa, USA, October 2004. observation mechanism and reputation adaptation scheme [15] F. Picconi, N. Ravi, M. Gruteser, and L. Iftode, “Probabilistic along with event confidence threshold and event reputation validation of aggregated data in vehicular ad-hoc networks,” threshold to evaluate the event intensity and event reliability in Proceedings of the 3rd ACM International Workshop on at the same time. Experimental simulations show that the Vehicular Ad Hoc Networks (VANET ’06), pp. 76–85, Los proposed system can prevent false traffic warning messages Angeles, Calif, USA, 2006. 10 EURASIP Journal on Wireless Communications and Networking

[16] M. Raya, A. Aziz, and J.-P. Hubaux, “Eﬃcient secure aggregation in VANETs,” in Proceedings of the 3rd ACM International Workshop on Vehicular Ad Hoc Networks (VANET ’06), pp. 67– 75, Los Angeles, Calif, USA, 2006. [17] N.-W. Lo and H.-C. Tsai, “Illusion attack on VANET applications—a message plausibility problem,” in Proceedings of the 2nd IEEE Workshop on Automotive Networking and Applications (AutoNet ’07), pp. 1–8, Washington, DC, USA, November 2007. [18] G. F. Marias, P. Georgiadis, D. Flitzanis, and K. Mandalas, “Cooperation enforcement schemes for MANETs: a survey,” Wireless Communications and Mobile Computing, vol. 6, no. 3, pp. 319–332, 2006. [19] K. Fall and K. Varadhan, “The ns-2 manual,” the VINT Project, April 2002, http://www.isi.edu/nsnam/ns/doc. [20] A. Mahajan, N. Potnis, K. Gopalan, and A. I. A. Wang, “Evaluation of mobility models for vehicular ad-hoc network simulations,” in Proceedings of the IEEE International Workshop on Next Generation Wireless Networks (WoNGeN ’06),Banga- lore, India, December 2006. Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 208524, 8 pages doi:10.1155/2009/208524

Research Article A Cross-Layer Routing Design for Multi-Interface Wireless Mesh Networks

Tzu-Chieh Tsai and Sung-Ta Tsai

Department of Computer Science, National Chengchi University, Taipei, Taiwan

Correspondence should be addressed to Tzu-Chieh Tsai, [email protected]

Received 15 April 2009; Accepted 7 August 2009

Recommended by Naveen Chilamkurti

In recent years, Wireless Mesh Networks (WMNs) technologies have received significant attentions. WMNs not only accede to the advantages of ad hoc networks but also provide hierarchical multi-interface architecture. Transmission power control and routing path selections are critical issues in the past researches of multihop networks. Variable transmission power levels lead to different network connectivity and interference. Further, routing path selections among different radio interfaces will also produce different intra-/interflow interference. These features tightly affect the network performance. Most of the related works on the routing protocol design do not consider transmission power control and multi-interface environment simultaneously. In this paper, we proposed a cross-layer routing protocol called M2iRi2 which coordinates transmission power control and intra- /interflow interference considerations as routing metrics. Each radio interface calculates the potential tolerable-added transmission interference in the physical layer. When the route discovery starts, the M2iRi2 will adopt the appropriate power level to evaluate each interface quality along paths. The simulation results demonstrate that our design can enhance both network throughput and end-to-end delay.

Copyright © 2009 T.-C. Tsai and S.-T. Tsai. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction a node that can improve the throughput capacity [3]. This feature enables nodes to transmit and receive simultaneously, Wireless Mesh Networks (WMNs) have the characterisitcs of hence nodes can use nonoverlapping channels to transmit low deployment cost, easy maintenance, and reliable service and receive at the same time via different interfaces. WMNs coverage technologies to form robustness networks. The technologies accede to the advantages of ad hoc networks. task group “s” (TGs) of IEEE 802.11 develops a flexible Traditional ad hoc network routing protocols may not be and extensible standard for wireless mesh networks based suitalbe for WMNs since they do not fully consider the on the original IEEE 802.11. However, IEEE 802.11 TGs features of WMNs such as multi-interface. In IEEE 802.11s, adopts two main proposals—SEE-Mesh (Intel) and Wi-Mesh it presents the prototype of default path selection protocol- (Nortel) intending to specify a framework for WLAN Mesh HWMP (Hybrid Wireless Mesh Protocol) and routing networking [1]. In WMNs, nodes are comprised of mesh metric-airtime cost. The implementation details can be routers and mesh clients [2]. Mesh routers form a wireless based on user demands. Several routing protocol designs for backbone of WMNs, which provide multi-hop connectivities WMNs [4, 5] focus on single layer of network protocol stacks between mesh clients and mesh gateways that have wired and do not consider coordinating with different protocol connectivity with Internet (Figure 1). layers. Specifically, in the physical layer, the transmission WMNs are dynamically self-organized and self- power level decides the signal strength and determines the configured, with the nodes in the network automatically neighbor nodes which can hear the packet. This thus affects establishing and maintaining mesh connectivity among the network layer to select the forwarding nodes at the themselves and compatible with conventional Wi-Fi clients. route discovery. The transmission power also causes the Multi-interface WMNs provide multiple radio interfaces of interference that affect the link quality among nodes. The 2 EURASIP Journal on Wireless Communications and Networking appropriate transmission power level selection can improve network performance [6–8]. Traditional transmission power control problems in wireless ad hoc networks mainly focus Internet on reducing energy consumption. Some researches address power selection problems but still use minimum hop-count Mesh as the routing metric. Power control indeed impacts multiple gateway network protocol layers. Transmission power control tightly affects network performance [9]. The theoretical studies [10] have demonstrated that transmission power control Mesh Mesh backbone/backhaul can improve wireless network capacity. The result in [11] routers presents that the need to design future protocols is based on variable-range power control, not on common-range transmission power control. The higher transmission power increases network connectivity and gives lower end-to- ffi Mesh end delay in the low tra c load with slight interference. clients However, the higher transmission power will create high Conventional clients interference when concurrent transmissions in the vicinity are increased. This will decrease the spatial reusability. In this high loading case, using lower transmission power will result in lower interference, and thus increase the throughput. The Wired connection Wireless connection motivation of our cross-layer routing protocol development is inspired from the above features we observed. Figure 1: Wireless Mesh Networks. We previously proposed the MiRii (Multi-Interface Rout- ing with Intra/Inter-flow Interference) [5] routing protocol that measures intra-/interflow interference and applies to 2.1. Transmission Power Control. Previous transmission routing path selection in the network layer. The multi- power control schemes for ad hoc networks have focused on interface feature is utilized by considering the channel throughput improvement or power consumption. We do not diversity of the routing path. By contrast with AODV [12], consider the power consumption in our work becasue WMN ETX [13], and WCETT [4] (will be introduced later in backbone does not have power consumption issues. In [10], Section 2.2), the simulation results demonstrate that our the author shows that reducing the transmission power can MiRii routing protocol can improve packet delivery ratio increase the carrying capacity of the network. The work in and decrease end-to-end delay. To further improve the [11] concludes that variable-range transmission power can performance, the routing protocol has to work together improve the overall network performance. Some researches with the physical layer. In this paper, we propose our address the power-controlled problem on the network layer cross-layer routing protocol, namely, M2iRi2 (Multi-power, [6, 7]. The COMPOW protocol [6] relies on the DSDV Multi-interface Routing with Intra/Inter-flow Interference), routing protocol to discover the smallest common power that incorporates MiRii routing protocol with perflow level at which the entire network is still connected. However, transmission power control. M2iRi2 routing protocol jointly it suffers when some nodes can only use high power to be coordinates the transmission power at each trafficflowand connected. In [7], the proposed CLUSTERPOW protocol route selection among multi-interface nodes. The protocol performs the routing protocol several times with different interplays between the network and the physical layers. It power levels at each run and independently builds a routing aims to select appropriate transmission power to reduce the table at each power level. A node consults the routing table noise interference more efficiently when the traffic loading in with the lowest power to forward packets to the next hop. the network is increased. This consumes too much network resource. The rest of the paper is organized as follows. Section 2 Several researches [8, 14, 15] introduce the interference reviews the past work related to link quality routing, load- tolerance in their transmission power control architecture. balancing routing, and the transmission power control on The interference tolerance represents how much interference the physical layer. Section 3 describes our cross-layer routing a node can allow the potential transmission of its neighbors. protocol in detail. Section 4 presents the simulation result Nodes transmit the packets with the power level that does not and analysis. Finally, Section 5 concludes the paper with a disturb the ongoing receptions of its neighbors. In [8], the summary and proposes the future work. authors proposed a power controlled multiple access wireless MAC protocol (PCMA) within the collision avoidance framework. In PCMA, each receiver sends busy tone pulses 2. Related Work to announce its interference tolerance. If the trasmitter has data to send to the receiver, it will determine its power We first introduce some related work to our cross-layer bound according to the interference tolerance declared by the routing protocol M2iRi2, namely the transmission power receiver’s busy tone. It then sends Request-Power-To-Send control and routing metrics in WMNs, including our (RPTS) with appropriate power level setting accordingly. previous MiRii routing protocol. The receiver will calculate its tolerable power level and send EURASIP Journal on Wireless Communications and Networking 3 acceptable-power-to-send (APTS) back to the sender. In takes both link quality (ETT) and channel diversity (Xj ) into other words, the protocol design is based on CSMA/CA considerations. Thus, WCETT combines these two features and modifies the RTS/CTS to RPTS/APTS (Request-Power- by taking their weighted average as follows: To-Send/Acceptable-Power-To-Send) to support potential n interfering transmissions to transmit concurrently rather WCETT = 1 − β ∗ ETTi + β ∗ max Xj . (1) 1≤ j≤k than to silence it. However, PCMA uses the additional i=1 separate control channel to send the busy tone pulses. PCDC (Power Controlled Dual Channel) [14] also uses dual In fact, the WCETT metric takes “intraflow” (means channels for data and control packets. A single channel the same flow, but between different hops) interference solution for transmission power control (POWMAC) is used into consideration, but does not capture “interflow” (means in [15]. In POWMAC, the interference tolerance is inserted between different neighboring flows) interference. In our into the CTS packet and an additional control packet DTS previous work [5], we propose a new routing metric, MiRii, (Decide-To-Send) is used by transmitter to confirm the that considers both intraflow and interflow interference transmission. Furthermore, the DTS is utilized to inform in the multi-interface WMNs. In order to capture the the neighbors of transmitter about the power level that the interflow interference, we calculate the nodal activity and transmitter will use for its data transmission. The neighbors intertraffic flow interference. We introduce Little’s Result of the transmitter can determine whether or not they can into the routing metric design that makes the interflow receive the data packets form other nodes simultaneously interference unit cost compatible with WCETT. To this end, through DTS. we assume node k and node k’s neighboring nodes as a From the comparisons of [8, 14, 15], we finally adopt closed system and the Activity Time (AT) of node k is shown the interference tolerance concept to be integrated into in (2). Also, Nk and λk are the average queue length and our cross-layer routing protocol design. We will facilitate average packet receiving rate of node k,respectively.The concurrent interference-tolerable transmissions in the same sum of average queue length of k’s neighboring nodes is vicinity of the receiver to enhance the WMN backbone the second parameter of numerator, and the sum of average capacity. Furthermore, we piggyback the interference toler- packet receiving rate of k’s neighboring nodes is the second ance information in probe packets which are integrated in parameter of denominator in (2). The Activity Time (ATk) the network layer routing protocol. It is not using a separate regarding to k is the total average queue length divided by control channel to alert the neighboring nodes. total average packet receiving rate of the system: m k Nk + nb= N = 1 nb . AT k λ m λk (2) 2.2. Routing Metrics in WMNs. Most of the routing protocols k + nb=1 nb use “hop count” as the routing metric. The minimum hop- count routing is not suitable for wireless networks because Therefore, by combining both intraflow and interflow inter- of dynamic wireless link quality characteristics. The work in ferences, the MiRii routing cost is defined as (3) [4, 13] proposes new routing metrics considering the link n quality dynamics. MiRii proposed in [5] further considers MiRii=α ETTi +β ∗ max chanETT j +γ AT k, 1≤ j≤k the intra-/interflow interferences in multi-interface routing i=1 k∈path & k =/ src,dst path selections. (3) Theworkin[13] proposed the concept of the expected transmission count (ETX) as the routing metric. ETX is where α, β,and γare the constant weights subject to α + calculated by measuring the delivery ratios for probe packets β + γ = 1. Also, ETTi means the total link quality con- in bidirectional transmissions of each link. It predicts the siderations and end-to-end delay over an n-hops path. The number of data transmissions required to send a packet chanETTi is the sum of transmission times of hops using and get a successful acknowledgment. Therefore, the ETX channel j, and represents the channel-diversity in multiradio accounts for interference among the successive links of a WMNs. Finally, AT k is the interflow interference, and path. Although ETX does well in single-radio wireless ad represents the load-balanced routing cost. hoc network, it does not perform well in multiradio and multichannel wireless mesh networks. Reference [4] presents 3. Cross-Layer Routing Protocol Design a new routing metric for multiradio, multihop wireless mesh networks, called WCETT (Weighted Cumulative Expected In the WMNs, the traffic loading changes dynamically due Transmission Time). WCETT assigns weights to individual to leaving or entering the network of traffic flows. From the links based on the Expected Transmission Time (ETT) of above observation, when the traffic loading is low, the traffic a packet over the link. As a result, the WCETT of a route flows should select the higher transmission power to enhance with n hops can be the sum of the ETTs of all hops along the throughput and reduce delay. As the traffic loading the path. Further, WCETT assumes that the network has a is increased, the high transmission power imposes more total of k channels in an n-hop path. However, Xj is the interference that may disturb the ongoing transmission. The sum of transmission time of hops that uses channel j along new traffic flow needs to choose the lower power level the path. The total path throughput will be dominated by to transmit the data to alleviate interference. Further, the the bottleneck channel, which has the highest Xj .WCETT packet transmission at each hop on the routing path suffers 4 EURASIP Journal on Wireless Communications and Networking

Pi means the transmit power for node i and PiGij is the Network layer received power at node j,whereGij is the propagation gain for the direct transmission from node i to j. Also, Nj is Link quality measurement MiRii routing the thermal noise at node j,and l =/ i PlGlj is the sum of protocol interferences that transmit concurrently with node i. iTolerance info exchange However, iTolerance is deﬁned as the interference tolerance that a receiver node can tolerate a new joining neigh- TxPower level selection boring interference without destroying its existing ongoing receptions. So, the radio interface at node j, regarding to the ﬂow from node i, can allow its iTolerance as follows: ⎛ ⎞ P G i ij ⎝ ⎠ iTolerancej = − PlGlj + Nj . (5) SINR Threshold iTolerance calculation Physical layer l =/ i

In the protocol design, we measure the actual propagation Figure 2: The cross-layer routing protocol architecture. gain based on the received power of the probe packet at the receiver side. Nodes can locally measure their interference tolerance according to the sum of the strengths of all the propagation, handling, and queuing delays. When the traffic interfering signals. Each radio interface on the node will loading is low, the queuing day may be insignificant. It advertise its interference tolerance to its one hop neighbors. is better to use high transmission power to reduce hop In the network layer, each node updates its neighbors’ count and also reduce the handling delay. However, under iTolerance through the probe packet. When the route high traffic loading, the low transmission power reduces discovery starts, the node looks up its neighbors’ iTolerance the queuing delay because the queue length will grow due and chooses the appropriate TxPower (transmission power) to more neighbors’ interference or collisions. Therefore, level to send RREQ or forward RREQ messages. From the it is a good policy that we should adapt the appropriate (6), the routing protocol will select the highest TxPower transmission power level according to the surrounding when the iTolerance constraints can be satisfied: interference constraints. This is one of the basic motivations of our work. max TxPower level | TxPower level (6) 3.1. Overview of Protocol. The scheme of the proposed iTolerancel 2 2 ≤ minneighbor l . M iRi cross-layer routing protocol is shown in Figure 2. Gil The routing protocol is based on AODV [12]. We modify it to support MiRii routing metric and transmission power We also apply for the MiRii routing metric which is level selection on a perflow basis. The network layer coor- referred in (3)inourM2iRi2.Hence,eachtraffic flow select dinates with the physical layer to choose the appropriate the appropriate transmission power to send its route request transmission power level and to find a routing path with messages and use the MiRii routing metric to choose a better link quality. The proposed M2iRi2 routing protocol is routing path that has the smallest MiRii cost. In M2iRi2, operated among mesh routers. The mesh clients can access we use the probe packet to measure the link quality and the network by directly connecting to mesh routers. The piggyback the iTolerance information with the neighbors. protocol does not consider dynamic channel assignment for simplicity at this point. In the physical layer of the 3.3. Perflow-Based Transmission Power Control and Routing. model, there are several discrete transmission power levels The original AODV is a destination-based routing protocol, for each NIC (network interface card). The function of and suffers the route flapping problem. For example, we “iTolerance Calculation” calculates the interference tolerance assume that both flows 1 and 2 route through node A to of each ongoing receiving NIC at the node. The “Link Quality the same destination node B. Some time later, the route Measurement” function measures the ETT and AT that are entry of A’s routing table to destination B changes for used in the MiRii routing protocol. All the measurement and somereason.Itwillaffect both flow 1 and switch their information exchanges are through probe packets that are routing paths simultaneously. The destination-based routing broadcasted proactively by each NIC. protocol cannot balance the traffic loading. In this case, perflow routing will be a good choice to solve the problem. 3.2. iTolerance Calculation. Suppose that a packet transmis- In order to achieve the idea of perflow-based transmission sion from node i to node j is a successful reception if the power control and routing, the routing table should keep received SINR (Signal-to-Interference Noise Ratio) is above records for not only the destination (Dst) of the route but a certain threshold: also flow id (Fid) of pertraffic flow. Further, the routing table records each traffic flows TxPower level that it uses to reach PiGij = ≥ next hop. Hence, each interface on the node looks up the SINRi P G N SINR Threshold, (4) l =/ i l lj + j routing table according to the parameters (Dst, Fid) and EURASIP Journal on Wireless Communications and Networking 5

Type J R G D U Reserved Hop count Flow1 Flow2 RREQ ID Destination IP address 012 34 Destination sequence number Originator IP address Flow3 Flow4 Originator sequence number 567 89 Σ ETTlink Figure 4: A simple network topology. Σ chanETTlink

Σ ATnode Table 1: Delay-Throughput with different power levels. Fid TxPower Numberofflows 1234 MiRii-30 mW delay (ms) 14.0 37.9 102.3 157.7 Figure 3:TheformatofRREQpacket. MiRii-30 mW throughput (Kbps) 511.1 884.7 1119 1284 MiRii-100 mW delay (ms) 11.8 16.8 141.5 210.6 MiRii-100 mW throughput (Kbps) 510.2 1004 1057 931 adjusts the transmission power level for this trafficflowto forward to the next hop. The data transmission is based on CSMA/CA and the (Constant Bit Rate) during the ON period of an equally ON- iTolerance value at the radio interface changes dynamically. OFF model and the packet sizes are 1000 Bytes. So, we still transmit the data packet by using the lowest We first look at a simple topology (see Figure 4)which transmission power level even if there is no power level clearly demonstrates the benefits of using the appropriate satisfied the iTolerance constraints. Notice that here, for transmission power level at different interference environ- simplicity and fair performance comparisons later, we do ments. The flow-based MiRii routing protocol is introduced not apply call admission control to reject any new flows. to evaluate these two cases of power levels. The dash The Route Request (RREQ) packet format is illustrated in line in 4 denotes the connectivity using 30 mW power, Figure 3. The fields of ( ETT , chanETT , AT ) link link node while the communication range is double farther if using are used to calculate the MiRii value. The fields of (Fid, 100 mW power. Table 1 shows the throughput and end-to- TxPower) are utilized to achieve the per-flow transmission end delay with different traffic flow numbers. Each traffic power control. flow transmits with data rate 512 KBits/s. “MiRii-30 mW” In reality, the channel fading and interference change indicates that we fix transmission power at 30 mW in the dynamically. It is difficult to calculate exact tolerable power entire network. The average end-to-end delay is defined as level or link quality. There are some papers [16, 17] that the time of packet from leaving the source to successful proposed different approaches to deal with these. However, receiving at the destination. It includes the buffering time in concern with complexity of the algortihm, we are not before the routing path discovery, the queuing time, the dealing with the fading channel problem here. Instead, in delay of retransmission at MAC layer, and propagation our design, we use the “moving average” estimation to delay. When the number of traffic flow is one, the traffic find the path in the sense of “statistically approximation”. loading is low and interference is slight. We can utilize high Thus, it may combat the slow fading but not fast fading transmission power level (100 mW) to reduce end-to-end channels. Our proposed M2iRi2 can find the path with delay since it can travel through small hop counts. When the considering both existing and tolerable adding intereference. numbers of flows are increased, the interference among radio In this way, the throughput is increased with reducing the interfaces is increased. MiRii-30 mW can perform well since failed transmissions due to suffering too much intereference. radio interfaces with lower transmission power level reduce Therefore, energy consumption from the system view is the interference generating to its neighbors. The MiRii- reduced, and efficiency is increased. 30 mW has lower end-to-end delay and higher throughput than MiRii-100 mW. 4. Simulation Results and Analysis Now we consider a 4 × 4uniformtopologyina 500 m × 500 m region. Each node locates 80 meters apart. In this section, we evaluate the throughput and delay for In Figure 5, the light color (red) bars represent the high M2iRi2 using NS-2 and contrast it with the flow-based traffic loading with data rate 1 Mbits/s and the dark color MiRii and AODV routing protocol. The radio propagation (blue) bars represent the low traffic loading with data rate model adopts Two-Ray Ground model in NS-2. Each node is 512 Kbits/s. We consider the traffic flows in the WMNs equipped with two NICs. The off-the-shelf Cisco Aironet 350 randomly start and terminate. We let the CBR trafficflow series client adapters or access points allow different transmit randomly on/off but keep the number of active flows in power setting for one of 1, 5, 20, 30, 50, and 100 mW. We the network to be five in average. The numbers of traffic adopt the 30 mW and 100 mW in our NS-2 simulation. The flows and traffic pattern are the same in both cases. Figure 5 SINR threshold is setting to 6.02 dB and the noise floor shows that all the routing protocol can operate well in the at each node is −120 dBm. The traffic flow type is CBR low traffic loading. Because the trafficflowsarerandomly 6 EURASIP Journal on Wireless Communications and Networking

Traffic type: CBR, average flow number:5 flows 5000 4500 4000 11 3500 3000 1 2500 10 2000 1500

Throughput (Kbps) 1000 500 3 0 5 14 MiRi Mirii Mirii AODV AODV 30 mW 100 mW 30 mW 100 mW 0 15

CBR: 512 Kbits/s CBR: 1 Mbits/s 4 6 8 9 Figure 5: Throughput of different traffic load in a uniform network topology. 2 12 Traffic type: CBR,average flow number: 5 flows 350 7 13 300 Figure 7: A random network topology. 250 200 150 Traffic type: CBR, average flow number: 5 flows 5000 100 Throughput (Kbps) 50 4000 0 3000 MiRi Mirii Mirii AODV AODV 30 mW 100 mW 30 mW 100 mW 2000

CBR: 512 Kbits/s Throughput (Kbps) 1000 CBR: 1 Mbits/s 0 ff ffi MiRi Mirii Mirii AODV AODV Figure 6: Average end-to-end delay of di erent tra cloadina 30 mW 100 mW 30 mW 100 mW uniform network topology. CBR: 512 Kbits/s 1 Mbits/s ff on/o and choose the source-destination pair randomly. Figure 8: Throughput of different traffic load in a random network The destination-based routing protocol also increases RREQ topology. broadcast times and increases the packets waiting in the buffer before the routing path establishment. The flow-based routing protocol has the delay better than destination-based routing protocol. Even if the flow-based routing protocol about 7% and 14% at high data rate. Figure 9 shows the needs to broadcast RREQ packets for each flow, it can average end-to-end delay in the high and low data rate. The discover better routing path than destination-based routing delay of M2iRi2 is further lower than MiRii-100 mW and is protocol. When the data rate increases to 1 Mbits/s, the better than MiRii-30 mW about 28% at high trafficdatarate. transmission power fixed at 30 mW has throughput better The simulation results indicate that our proposed cross-layer than 100 mW, and the end-to-end delay has the same result. routing protocol utilizes the advantages of different power In this case, M2iRi2 have the throughput similar to MiRii- levels in different network environments and performs well 30 mW, and improve the throughput 13% contrasting with by controlling the transmit power efficiently for perflow per MiRii-100 mW. In Figure 6, the results of average end-to-end hop transmission. delay of M2iRi2 are decreased by 30% and 48% contrasting Finally, we simulate a WMN with gateways, we choose with MiRii-30 mW and MiRii-100 mW, respectively. two nodes in Figure 7 to play the roles of mesh gateways. The We next simulate our protocol on a topology that nodes transmissions send the data packets to mesh gateways instead are randomly placed in a 500 m × 500 m area (Figure 7). The of random Source-Destination pairs. The trafficpatterns simulation parameters are the same as the uniform topology. tightly affect the performance of the routing protocol. The throughputs are almost similar in these routing pro- Figures 10 and 11 show the simulation results in this case. tocols at low traffic loading as we observe in the uniform We observe that M2iRi2 still has better throughput and end- topology case. However, from Figure 8, the throughput of to-end delay than MiRii-30 mW and MiRii-100 mW when M2iRi2 is better than MiRii-30 mW and MiRii-100 mW the traffic data rate is 1 Mbits/s. All the routing protocols EURASIP Journal on Wireless Communications and Networking 7

Traffic type: CBR, average flow number: 5 flows also operate well when the flow data rate is 512 Kbits/s. The 350 destination-based AODV routing protocol might have the 300 lower end-to-end delay depending on whether the traffic 250 flows have the same destination (gateway) or not, which will 200 reduce the route discovery time. The results also indicate that 150 our M2iRi2 routing protocol operates well when the traffic 100 are all going towards gateways in the WMN. 50 0

Average end-to-end delay (msec) MiRi Mirii Mirii AODV AODV 5. Conclusion and Future Work 30 mW 100 mW 30 mW 100 mW In the paper, we proposed M2iRi2 routing protocol for CBR: 512 Kbits/s multi-interface WMNs. The main purpose is to coordinate CBR: 1 Mbits/s the physical layer and the network layer for a cross-layer routing protocol development. Previous researches show that Figure 9: Average end-to-end delay of different trafficloadina variable transmit power level control can improve network random network topology. performance but they still use minimum hop counts as the routing metric. We introduce the iTolerance to constrain the transmit power level and incorporate it to the route discovery. The MiRii routing metric is utilized to evaluate the routing path with consideration of both intraflow and Traffic type: CBR, average flow number: 5 flows interflow interferences. Furthermore, the power control is 5000 designed on perflow, perhop basis. We thoroughly observe 2 2 4000 the performance of M iRi at different traffic loadings. When the traffic loading is high, the newly trafficflow 3000 chooses the appropriate transmission power level along less interference path to transmit the data packets in order not to 2000 create intolerable interference to the existing transmissions. Through the simulation results, we have demonstrated that Throughput (Kbps) 1000 our M2iRi2 routing protocol can enhance both network 0 throughput and end-to-end delay. MiRi Mirii Mirii AODV AODV 2 2 30 mW 100 mW 30 mW 100 mW In the current version of M iRi routing protocol, the traffic flow selects the lowest power level even if it would violate the interference tolerance constraint. In the future, CBR: 512 Kbits/s 2 2 ffi 1 Mbits/s we may incorporate M iRi with the tra c flow admission control and extends the M2iRi2 to more stability and even better performance. Figure 10: Throughput of different trafficloadinarandom network topology with gateway. Acknowledgment This work is granted by NSC-97-2221-E-004-004-MY2.

Traffic type: CBR, average flow number: 5 flows 100 References

80 [1] H. Aoki, N. Chari, L. Chu, et al., “802.11 TGs simple efficient extensible mesh (SEE-Mesh) proposal,” IEEE802.11 document 60 05/0562r0, 2005. [2]I.F.AkyildizandX.Wang,“Asurveyonwirelessmesh 40 networks,” IEEE Communications Magazine,vol.43,no.9,pp. 20 22–30, 2005. [3] A. Adya, P. Bahl, J. Padhye, A. Wolman, and L. Zhou, “A multiradio unification protocol for IEEE 802.11 wireless networks,” Average end-to-end delay (msec) 0 MiRi Mirii Mirii AODV AODV in Proceedings of the 1st IEEE International Conference on 30 mW 100 mW 30 mW 100 mW Broadband Networks (BroadNets ’04), pp. 344–354, October 2004. CBR: 512 Kbits/s [4] J. Padhye, R. Draves, and B. Zill, “Routing in multi-radio, CBR: 1 Mbits/s multi-hop wireless mesh networks,” in Proceedings of the 10th Annual International Conference on Mobile Computing and Networking (MOBICOM ’04), pp. 114–128, Philadelphia, Pa, Figure 11: Average end-to-end delay of different tra0-40pt USA, 2004. 8 EURASIP Journal on Wireless Communications and Networking

[5] T.-C. Tsai and T.-F. Liu, “Multi-interface routing with intra/inter-ﬂow interference (MiRii) considerations in wireless mesh networks,” in Proceedings of the 3rd Asia-Paciﬁc Symposium on Queueing Theory and Network Applications, 2008. [6] S. Narayanaswamy, V. Kawadia, R. S. Sreenivas, and P. R. Kumar, “Power control in ad-hoc networks: theory, architecture, algorithm and implementation of the COMPOW protocol,” in Proceedings of the European Wireless Conference (EW ’02), 2002. [7] V. Kawadia and P. R. Kumar, “Principles and protocols for power control in wireless ad hoc networks,” IEEE Journal on Selected Areas in Communications, vol. 23, no. 1, pp. 76–88, 2005. [8]J.P.Monks,V.Bharghavan,andW.W.Hwu,“Apowercon- trolled multiple access protocol for wireless packet networks,” in Proceedings of the 20th Annual Joint Conference on the IEEE Computer and Communications Societies (INFOCOM ’01), vol. 1, pp. 219–228, 2001. [9] M. Krunz, A. Muqattash, and S.-J. Lee, “Transmission power control in wireless ad hoc networks: challenges, solutions, and open issues,” IEEE Network, vol. 18, no. 5, pp. 8–14, 2004. [10] P. Gupta and P. R. Kumar, “The capacity of wireless networks,” IEEE Transactions on Information Theory,vol.46,no.2,pp. 388–404, 2000. [11] J. Gomex and A. T. Campbell, “Variable-range transmission power control in wireless ad hoc networks,” IEEE Transactions on Mobile Computing, vol. 6, no. 1, pp. 87–99, 2007. [12] C. Perkins, E. Belding-Royer, and S. Das, “Ad hoc on-demand distance vector (AODV) routing,” IETF RFC 3561, 2003. [13] D. S. J. de Couto, D. Aguayo, J. Bicket, and R. Morris, “A high- throughput path metric for multi-hop wireless routing,” in Proceedings of the Annual International Conference on Mobile Computing and Networking (MOBICOM ’03), pp. 134–146, San Diego, Calif, USA, 2003. [14] A. Muqattash and M. Krunz, “Power controlled dual channel (PCDC) medium access protocol for wireless ad hoc networks,” in Proceedings of the Annual Joint Conference on the IEEE Computer and Communications Societies (INFO- COM ’03), pp. 470–480, 2003. [15] A. Muqattash and M. Krunz, “POWMAC: a single-channel power-control protocol for throughput enhancement in wireless ad hoc networks,” IEEE Journal on Selected Areas in Communications, vol. 23, no. 5, pp. 1067–1084, 2005. [16] L. Xiao, M. Johansson, and S. P. Boyd, “Simultaneous routing and resurce allocation via dual decomposition,” IEEE Transactions on Communications, vol. 52, no. 7, pp. 1136– 1144, 2004. [17] M. P. Anastasopoulos, A. D. Panagopoulos, and P. G. Cottis, “A distributed routing protocol for providing QoS in wireless mesh networks operating above 10 GHz,” Wireless Communi- cations and Mobile Computing, vol. 8, no. 10, pp. 1233–1245, 2008. Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 247483, 12 pages doi:10.1155/2009/247483

Research Article Intelligent Decision-Making System with Green Pervasive Computing for Renewable Energy Business in Electricity Markets on Smart Grid

Dong-Joo Kang,1 Jong Hyuk Park,2 and Sang-Soo Yeo (EURASIP Member)3

1 Korea Electrotechnology Research Institute, Uiwang, Gyeonggi, 437-808, South Korea 2 Department of Computer Science and Engineering, Seoul National University of Technology, South Korea 3 Division of Computer Engineering, Mokwon University, 302-729, South Korea

Correspondence should be addressed to Sang-Soo Yeo, [email protected]

Received 6 April 2009; Accepted 8 June 2009

Recommended by Naveen Chilamkurti

This paper is about the intelligent decision-making system for the smart grid based electricity market which requires distributed decision making on the competitive environments composed of many players and components. It is very important to consider the renewable energy and emission problem which are expected to be monitored by wireless communication networks. It is very diﬃcult to predict renewable energy outputs and emission prices over time horizon, so it could be helpful to catch up those data on real time basis using many diﬀerent kinds of communication infrastructures. On this backgrounds this paper provides an algorithm to make an optimal decision considering above factors.

Copyright © 2009 Dong-Joo Kang et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction has reduced costs until they are competitive with those of conventional power [2]. Since they took a small capacity Renewable generators have been increasing in generation of generation in the beginning, there has been no serious sector driven by government driven policies and economic consideration on impacts to the power system. As the portion incentives to each utility. Renewable Portfolio Standard of renewable energy shows a rapid increase recently, it (RPS) is a good example of forced encouragement on is expected to have a big influence on system operation introducing renewable energy sources by governments. RPS and business activities in the near future. Therefore it is is a state policy mandating a state to generate a percent required for utilities to build a management algorithm for of its electricity from renewable source [1]. On economic dealing with decision-making problems composed of the incentives, they are given to utilities by global environment changing variables such as renewable energy and emission regulations such as emission constraints and economic trading. And the algorithm is run based on the information efficiency improvements on renewable generation facilities acquired from the communication network. This study by technology developments. Generators are forced to deals with the idea of an intelligent decision-making system introduce renewable energy replacing the conventional fossil for the renewable issues under recent energy management fueled generators with renewable generators for mitigating system environments on the aspects of physical system and emission constraints with reference to the Kyoto protocol. institutional scheme changes. There are two ways to fulfill emission constraints. One is to purchase emission rights from those who pollute less or in trading markets. The other option is to produce the emission 2. SCADA System to Smart Grid credits themselves through (Clean Development Mechanism CDM) business opportunities including the construction of The electric power system is largely composed of two kinds renewable energy sources. Rapid progress in this technology of infrastructures which are the electric power supply system 2 EURASIP Journal on Wireless Communications and Networking

Competition

Vertically integrated Genco 1 Genco 2··· GencoN Unbundlin Generation Transco 1 Transco 2 Transmission g

Distribution Distco 1 Distco 2··· DistcoN

Figure 1: Evolution from SCADA system to Smart Grid.

Corporate network Other Serial protocol over networks TCP/IP network

Firewall Gateway

Local area or private network MODBUS/TCPUS/TCP DNP oveovo r TCP/IP Communication links (common networks)

DNP or MODBUS on serial network

Gateway

Figure 2: Smart grid vision [3].

120 Long-term Investment planning Short-term Operation scheduling 100

80 Electricity market price 60 Production cost Resource assessment Emission trading Feed-in tariff 40

20 Renewable energy sources 0 Figure 3: Decision-making process on renewable energy. 1 3 5 7 9 11131517192123

A B C and the information infrastructure to control the supply Figure 4: Examples of hourly market price trajectories. process. The communication networks of energy systems are being integrated into other communication networks and expanded to wide areas for the increasing requirements SCADA system is to introduce the end customers into the of intelligent automation and control while it is currently interaction in the network, which is illustrated in Figure 1. conﬁned to the SCADA system that monitors and controls As the SCADA system evolves it is expected to be the power system. One of the big diﬀerences from the current connected to satellites, sensor networks, the Internet, and EURASIP Journal on Wireless Communications and Networking 3

so forth, in near future. This trend comes to the vision of smart grid as shown in Figure 2 [3]. Many new concepts and paradigms are introduced into the smart grid such as renewable generator operation, demand response, microgrid, sensor networks, and so forth. Renewable generators are not able to be centrally controlled as the way of the conventional fossil fueled generators. Demand response is a newly introducing concept for making the markets in electric power industry. Microgrid is an independent local power system with a small scale of generators including renewable energy sources. Sensor networks are installed for many purposes such as monitoring system status or faults, Figure 5: User interface example of market simulator. assessing renewable resources, and so forth. All these factors are diﬃcult to be dealt with in central dispatch systems; therefore it is required to transfer from centrally controlled )

2 ECX CFI futures contracts: price and volume 14 35 structure distributed control scheme. 12 30 10 25 8 20 3. Decision-Making Problem 6 15 It is considered ﬁve variables such as renewable energy 4 10 2 5 production cost, potential renewable resource assessment, 0 0 Price per tonne (EUR) electricity market price, emission trading, and government

Volume (million tonnes CO policy like (feed-in tariff FIT) for the decision-making of utilities on the investment planning and operation scheduling of renewable energy sources. An example process of decision- 2006-02-06 2006-03-20 2006-05-03 2006-06-15 2006-07-27 2006-09-07 2006-10-19 2006-11-30 15-01-2007 26-02-2007 10-04-2007 22-05-2007 03-07-2007 14-08-2007 25/09/2007 06-11-2007 18-12-2007 01-02-2008 making on renewable energy is illustrated in Figure 3 based Total volume on the variables stated above. Dec08 Sett Decision-making is no longer a uniquely human func- Figure 6: ECX CFI futures contracts. tion in complex systems. Indeed, the speed and complexity of many system processes often preclude the human from decision and control functions [4]. Especially, the problems Price in electric power industry deal with lots of data and variables which comprises a huge scale of problem with thousands Market price FIT of variables so that it is inevitable to be supported by computerized tool on the decision-making process. Sup- Mitigated FIT porting decision-making requires understanding of both the processes involved and the provision of a computer-based system that supports these processes and allows them to be Time carried out more effectively [5]. Renewable energy business also requires several decision-making processes shown in FIT applied period Mitigated FIT or market price Figure 1 considering lots of strategic variables connected applied period with electric power system as stated above. This study applies the intelligent decision-making process to the renewable Figure 7: Feed-in tariff application. energy business.

Operation data 4. External Variables on Renewable Energy Prodution cost Monitoring by SCADA system Renewable energy business has several external variables Market price Intelligent affecting the decision-making process as shown in Figure 3. Emission credit cost decision-making In addition, renewable generators using wind power have system RPS more uncertainties than conventional fossil fueled generators Feed-in tariff Alternative 1 on output characteristics. There are several critical variables which are critically considered in guiding decisions on which Best option Alternative 2 . . energy resources are appropriate for given conditions. The . representative variables are energy and economic efficien- Alternative n cies, energy market price paid off for providing energy with renewable resources. Economic efficiency is about Figure 8: Intelligent decision-making system. getting the benefit as much as possible from the same 4 EURASIP Journal on Wireless Communications and Networking

Central SCADA server

Region: controlol areaea

Regional SCADAA server (1)(1) Regional SCADA RTURTU server (2) RTURTU RTU RTURTU RTU RTU RTU

Regional SCADA server (3) RTU RTU RTU RTU RTU

Figure 9: SCADA system conﬁguration.

Company-wide to guarantee renewable capacity developers a minimum price SCADA and power system access rights to promote the development Sub-SCADA 1 Communication of renewable energy.

4.1. Electricity Market Prices. Electricity market price is going to be the most critical factor affecting the profitability of renewable energy providers in near future while the excessive production cost above the market price is recovered by Wind-farm 1 transient policy supports like FIT. Electricity price changes Sub-SCADA 2 Ownership on real-time basis so it is necessarily required to introduce Sub-SCADA 3 stochastic models to reflect the uncertainty of future prices. Figure 4 shows an example of simulation on electricity market prices of hourly basis. These market prices could be forecasted by market simulators provided by commercial vendors. There are Wind-farm 2 several popular ones like GE’s MAPS, Henwood’s PROSYM, Wind-farm 3 Drayton Analytics’ PLEXOS, CRA’s CeMOS, and so forth. Academic and noncommercial versions have been made Figure 10: Wind farm SCADA system. for research purposes using general computerized tools like MATLAB or GAMS which are also used for making commercial tools. productive resources or the same benefit with the least cost while energy efficiency indicates a narrower concept 4.2. Production Costs. Production cost is another critical of getting the greatest benefit from energy resources. In factor with market price for determining the profit of renew- market environments resource allocation is guided by the able energy producers. The production cost of renewable signals of production costs and market prices. Renewable energy is much higher than fossil fuel, but the gap is energy is mainly transformed into electricity; therefore the being fast narrowed as the renewable technology advances, electricity market price is one of important factors affecting and the environmental costs are added to fossil fueled the production and allocation of renewable energy. On generators. So it is very important to assess the exact current situation the production cost of renewable energy production cost reflecting related polices and environmental is much more expensive than fossil fueled energy so it cost components like emission costs. is very difficult to expect a fair competition between two different energy sources. For mitigating the difference on 4.3. Resource Assessment. It is a main difficulty in develop- competitive capabilities originated from production costs, ing renewable energy sources to assess the exact resource governments generally support the renewable energy sources quantity of renewable energy. For example, wind energy with policy level measures such as RPS and RPS. RPS was is very variable according to time and site which also already described above, and feed-in tariff is a pricing scheme increases the difficulties on operation and the uncertainties EURASIP Journal on Wireless Communications and Networking 5

Meteorological sensor (data acquisition)

RTU RTU

Offshore wind-farm Onshore wind-farm

Coaxial/optical cable Signal processor Radio Private network/internet communication

Radio repeater Ethernet hub/switch Workstation

Wind-farm SCADA server

Figure 11: Wind farm SCADA system.

Cost = capital cost + O & M cost + opportunity cost Intelligent decision N-th decision making system Feedback Objective (N+1)-th decision Analysis on results Max profit (k, c, t) = revenue − cost function Data acquisition Wind-farm SCADA Revenue = (market price or FIT) × supplied energy quantity + emission cost savings Figure 12: Interaction between IDMS and SCADA system. External variables or constraints • Available capacity of renewable energy facilities • Lead time for building renewable capacity in business. The resource assessment results become a basis • RPS (renewable portfolio standards) for determining whether to build wind capacity or not • FIT (feed-in tariff) applied period ff because the resource quantity directly a ects the energy and Figure 13: Profit function formulization. economic efficiency at the same time.

4.4. Emission Trading. Since the Kyoto protocol in effect, year. Emission credits also traded in the market with the the emission constraint has been another cost component variable market price. Figure 6 shows the changing volumes to electric power generating companies. Therefore the com- and prices of (Carbon Financial Instrument CFI) futures panies should make their investment plan on generation contracts in the (European Climate Exchange ECX) market capacity considering emissions because the emission cost over the time horizon. increases the production costs of fossil fueled generating units. There are two options for generation companies to 4.5. Feed-in Tariff. Feed-in tariffs(FITs)aimtosupport fulfill emission constraints. One is to build renewable energy the market development of renewable energy technologies, facilities or work on CDM projects, which is usually long- specifically for electricity generation. FITs put a legal obliga- term based. The other option is to buy emission credits tion on utilities and energy companies to purchase electricity from the emission trading markets, which is available on from renewable energy producers at a favorable price per short-term basis when emission obligations are given at the unit, and this price is usually guaranteed over a certain period 6 EURASIP Journal on Wireless Communications and Networking

Electricity market price RPS (historical data) Investment options Expected mean of Renewable capacity future market price

Production cost FIT applied? (assessment results) No Yes Market price applied

Profit = (FIT–PC) × RC × CF Profit = (MP–PC) × RC × CF

1st alternative’s profit = (PP–PC) × RC × CF • FIT: Feed-in tariff • PC: Production cost 2nd alternative’s profit = (PP–PC) × RC × CF • MP: Market price • RC: Renewable capacity • CF: Capacity factor nth alternative’s profit = (PP–PC) × RC × CF • PP: Payoff price

Selecting the best alternative

Figure 14: Decision-making algorithm for renewable capacity.

Fossile-fueled Market price RPS Thermal generator options available capacity generation quantity Renewable options KPX IDMS Fuel mix Best renewable option Emission costs EPSIS settlement price load forecast

Cost/benefit analysis of Cost/benefit analysis of Figure 17: Periodic market data acquisition from EPSIS. renewable energy source thermal generators Comparison finding out the solution. Strategic variables are defined in the mathematical formulation and the operation data acquired Final option selected from communication channels as input data. Figure 15: Selection between renewable and fossil fuel. 5.1. Intelligent Decision-Making System. Renewable energy business has several strategic variables for generation com- Capacity investment, fuel Long-term planning panies to maximize the profit and minimize the risk mix, maintenance, etc considering the external variables. According to current elec- Economic dispatch, tricity market rules, renewable generators are not centrally Short-term operation unit commitment, etc dispatched as nuclear and fossil fueled generators so it is assumed that there is no strategic variable on operation Figure 16: Interaction between long-term and short-term Prob- mode. Considering investment problems there are several lems. strategic variables such as the kinds of renewable energies (solar, wind, geothermal, etc.), the capacity of energy source, the investment time, and so forth. For simplification these of time [6]. Therefore it is required to consider the cost- variables are integrated into an alternative function like benefit analysis and assess the related risks because FITs do fn (k, c, t). So it is considered as the decision-making not last beyond that period. The application period and problem to decide which one is the best option among given price level depend on the government policies which are very alternatives. Figure 8 is an example of concept design on uncertain variables. the intelligent decision-making system for renewable energy investments. 5. Strategic Variables on Decision Making 5.2. Acquisition of Data through SCADA system. (Supervisory It is required to define the strategic variables for modeling Control and Data Acquisition SCADA) is a system operation any decision-making tool and specify the data used for with coded signals over communication channels so as to EURASIP Journal on Wireless Communications and Networking 7

250

200

150 (Won) 100

0 1 9 5 9 1 5 9 1 8 8 8 05 49 93 8 8 37 8 45 8 8 8 221 441 661 88 133 4 529 749 969 177 265 309 353 397 573 617 705 793 8 925 1321 1541 1761 19 2201 2421 1101 1013 1145 11 1233 1365 1409 1453 15 1629 1673 1717 1 1 1 2025 2069 2113 2245 22 2333 1057 1277 1497 1937 2157 2377 (h) Figure 18: Electricity market clearing prices.

Average Price jump by market price fuel cost escalation 82

Price increase Price increase with inflation 75 based on inflation and load growth 72 69 Price drop by a fall of fuel cost Production cost (Won/kWh) Time 2009 2010 2011 2012 2013 2014 2015 2016 2017 Year Figure 19: Long-term average price variation. Figure 21: Expected production costs of wind energy.

204 199 192 550 187 500 172 166 470 165 161 445 145 Production cost (Won/kWh) Average market price (Won/kWh) 2009 2010 2011 2012 2013 2014 2015 2016 2017 2009 2010 2011 2012 2013 2014 2015 2016 2017 Year Year Figure 20: Market price scenario. Figure 22:Expectedproductioncostsofsolarenergy. provide control of (Remote Terminal Unit RTU) equipment Terminal Units RTUs), communication links connecting [7]. Recently Intelligent Electronic Device (IED) which is two terminal parts as shown in Figure 9. Communication control unit having communication function with master links consist of several kinds of channels on the aspect station is replacing the role of RTU. SCADA system has of physical media, protocols, topologies, and so forth. Or been used for remote measurement and control on the all those channels are mixed or interconnected. Originally critical infrastructures such as electric power, gas, and oil SCADA network was a private network exclusive to other as well as modern industrial facilities such as chemical networks or the Internet, but it is getting integrated into the factories, manufacturing facilities [8]. SCADA system is Internet for more advanced control functions and economic largely composed of three parts of SCADA server, (Remote eﬃciencies. 8 EURASIP Journal on Wireless Communications and Networking

180 1000 162 160 900 860 140 800 689 120 700 600 100 91 500 460 80 400 (Won/kWh) 60 300 39 43 Emission (g/kWh) 40 200 20 100 9 30 11 0 0 Nuclear Coal Oil LNG Nuclear Coal LNG Oil Solar Wind Figure 23: Generation costs of conventional thermal generators. Figure 27: Emission quantities from diﬀerent energy sources.

10000 9000 108 8000 7000 107 6000 (Won/kWh)

ﬀ 5000 105 4000 (Won/kWh) 3000

Feed-in tari 2000 2009 2010 2011 2012 2013 2014 2015 2016 2017 1000 Year 0 2009 2010 2011 2012 2013 2014 2015 2016 2017 Figure 24: Feed-in tariﬀs applied to wind energy. Oil Wind LNG Solar Coal Nuclear Figure 28: Emission costs on energy sources. 720

691

(Won/kWh) Operation data are collected from the (Supervisory ff 636 Control and Data Acquisition SCADA) system also for renewable energy sources. Wind power is the representative one among renewable energy sources. Assuming the Feed-in tari wind power, the company-wide SCADA system consists of 2009 2010 2011 2012 2013 2014 2015 2016 2017 several sub-SCADA system monitoring wind farms locally Year distributed over wide areas as shown in Figure 10. Operation data acquired from SCADA system are used Figure 25: Feed-in tariffs applied to solar energy. for the resource assessments and operational characteristics which are references for future investment decisions on new renewable facilities. Renewable generators are dispersed over wide area, so various channels are used for the 43,000 42,000 communication. Onshore wind farms are usually connected 40,000 41,500 39,000 with wire communication methods using coaxial or optical 29,000 cables, while offshore wind farms communicate with the SCADA server on radio channels. 22,000 19,000 Thedataacquiredfromremotewindfarmsareused 16,500 for intelligent decision-making system in Figure 8 and also reviewed to analyze if the decisions made by the system are appropriate. If the decision does not match the operation Emission credit price (Won/ton) 2009 2010 2011 2012 2013 2014 2015 2016 2017 data then the decision would be corrected for the next period. Year Figure 12 shows the interaction process between Intelligent Decision-making System (IDMS) and Wind farm SCADA Figure 26: Variable emission credit prices. system. EURASIP Journal on Wireless Communications and Networking 9

250 200 180 200 160 140 150 120 100 100 80 60 50 40 (Won/kWh) 20 0 2009 2010 2011 2012 2013 2014 2015 2016 2017 0 1 3 5 7 9 11131517192123 −50 Renewable −100 Load Figure 31: Forecasted load and renewable output. Nuclear Oil Coal Solar LNG Wind This decision-making algorithm could be modeled with Figure 29: Proﬁtability on each energy resource. lots of numerical and heuristic methods. And it could be reinforced with other mathematical model like game theory and (artiﬁcial intelligence AI) based approaches. Intermittent Complementary output Targeted output output 7. Problem Solving on IDMS and Case Study

Wind energy Traditionally, simulation of business process is used to Power support strategic decision-making. In this case, simulation pool is used as a tool to analyze long-term effects of certain Thermal plants decisions. Simulation is rarely used for management control Controlled and operation control, because building a simulation model output takes too much time to evaluate short effects [9]. However the short-term operation and constraints are critical also Forecasted Solar energy load on long-term strategic decision-making problems in the electric power system because the electricity market should Figure 30: Renewable thermal coordinated operation. be operated on the physical system. Therefore it is required to consider long-term and short-term problems as the components interacting with each other in a problem. In this section it is exampled that long-term and short- 6. Decision-Making Algorithm term problems are related to renewable energy sources in It is required to make a mathematical formulation for mod- electricity market. Economic aspects of renewable source eling of intelligent decision-making system as a program- investments are dealt with in the first long-term problem based tool. Optimization models are generally used for section, and operational issues of renewable generator like solving decision-making problems under given resources and coordinated dispatch with other thermal plants in the second conditions. The optimization model is generally composed short-term problem section. of an objective function and multiple constraints. In this kind of business related model it is commonly used to maximize 7.1. Long-term Investment Problem. Electricity market prices the profitability of the company for the objective function. are published by (Korea Power Exchange KPX) on its Considering the profit maximization, the objective function webpage, and those historical data used for forecasting future of renewable energy business described in Figures 3 and 8 market prices. These data could be retrieved from (Electric could be formulated as follows. Power Statistics Information System EPSIS) periodically as An example is illustrated in Figure 14, which is about input data for IDMS. the decision-making algorithm for choosing the best option There are lots of methods and tools for forecasting among candidate alternatives based on the optimization the electricity market prices. The most simple and popular model in Figure 13 to maximize the profitability under given ones are statistics based function provided by spreadsheet conditions. programs, which mostly uses historical data. As more Once the best renewable option selected it is compared specialized tools, there are several computerized tools for with the fossil fueled generator option on the aspect of electricity market simulation provided by commercial ven- cost/benefit analysis as shown in Figure 15. The one giving dors. Figure 18 shows the electricity market prices recently more benefit is selected as the final option. published by KPX (Korea Power Exchange) for one month 10 EURASIP Journal on Wireless Communications and Networking

Energy management system SCADA server

ICCP

ICCP DNP DNP/TCP/IP DNP MTUs (regional SCADA servers)

ICCP TCP/IP

MOUBUS, Harris, FIELDBUS, DNP (→IEC61850) Generating stations

RTUs (each at a substation) IED

Figure 32: Data exchange between SCADA and EMS.

160 Feed-in tariffs are temporal measures for supporting the 140 introduction of renewable energy sources until they have the 120 economic competitiveness compared to conventional fossil 100 fueled generators. The purchasing prices are applied to wind 80 and solar energy resources in feed-in tariffs at the level of 60 Figures 24 and 25. The price for wind is discounted 2% every 40 3 years and solar for 4%. 20 Feed-in tariffs are uncertain variables determined by 0 government policies because policy-related variables are very 1 3 5 7 9 11131517192123 hard to forecast. Therefore it is required to build various Figure 33: Net load offset between load and renewable output. scenarios on feed-in tariffs to minimize the risks by applying the wrong payoff price to the profitability estimation of renewable energy sources. Emission costs are changing on real-time basis correlated with the price of emission credits according to the balancing of March in 2009 [10]. The average price for the month is condition between supply and demand in emission trading 145[Won/kWh]. markets. The annually averaged prices are used for this case One month is not a long-term in electricity market and study for simplification. And the prices are multiplied by a it shows any increasing or decreasing trend in Figure 18. multiplier (0.02) reflecting the transient status of emission The average price could be used for future investment plan costs applied to generation costs. by reflecting the load growth and the inflation rate as a Theemissionquantity[g]fromeachenergysourceper simple scenario that there is no change on fuel costs and unit electricity [kWh] production is illustrated in Figure 25. fuel mix ratio. When there is capacity investment or fuel cost Through the emission credit price [Won/ton] in variation, the long term market prices could go down as well Figure 26 and the emission quantity [g/kWh] in Figure 27, as up. the emission cost can be recalculated as the unit [Won/kWh] A scenario about the market price is assumed as shown in Figure 28. in Figure 18 for the case study based on the concepts given in Considering all the data till now decision-making system Figures 18 and 19. The product costs of wind power energy based on the algorithm in Figures 14 and 15 gives the are variable dependent on the wind resource quantity and the profitability result of each generation source over time load factors of the wind generators, so it is quite difficult to horizon. This result is based on the assumption given in quantify the unit cost per unit energy (kWh). However it is the beginning, so the result could be different by another required to do economic assessments. assumption. However it is expected to be similar to the trend Production costs are applied like in Figures 21 and 22 in in which nuclear and renewable energy sources have good this case study, which are similar with the current cost levels profitability. of renewable energy production in Korea. As the technology Nuclear, solar energy and wind energy show good advances the unit cost of renewable energy production is profitability compared to other energy resources in Figure 29 expected to decrease year by year. Figure 23 shows the and that trend will be stronger as the emission cost loads generation costs of conventional thermal generators. are heavier by increasing the multiplier to the value more EURASIP Journal on Wireless Communications and Networking 11

SCAD A

SCADA communication network DTE Thermal plant 2

RTU RTU Residential load

RTU Wind-farm 1

Power system RTU Industrial load Local microgrid RTU Thermal plant 1

RTU Solar-farm 1 RTU RTU Commercial load Wind-farm 2 Data exchange for coordination on wireless networks Metrological sensor

Figure 34: Data exchange for coordination on wireless networks.

than 0.02. Through the automation of this decision-making module in (Energy Management System EMS) of KPX and process, energy companies are expected to assess multiple published at the homepage. alternatives on renewable investment plans. Historical data acquired from SCADA system are used for load forecast. SCADA system retrieves data from the RTU installed in each substation. SCADA system and EMS are 7.2. Short-term Operation Problem. Renewable energy also connected with each other’s data exchange based on the sources as wind power and photovoltaics are intermittent (InterControl Center Communication Protocol ICCP). in production and therefore not always available in the Considering the forecasted load and expected renewable power supply, when needed. This of course can imply that outputs, EMS issues the dispatch order to thermal generators conventional power capacity is to be available to compensate to meet the net load subtracted by the renewable output from for the missing production from renewable plants [11]. It the originally forecasted load. In the case of Figure 31 the net is necessary to coordinate the renewable energy operation load is calculated as shown in Figure 33. with other thermal plants both on economic aspects and The net load is supplied by thermal generators consid- system aspects because thermal plants have the most ﬂexible ering the uncertainties and up/down ramp rates of thermal ramp rate compared to other energy sources. This has also generators. There are two uncertain variables on this process. been studied in hydrothermal coordination for a long time. One is forecasting errors in load forecast, and the other is in Traditionally, hydrothermal coordination is formulated renewable output. Therefore it is required for system oper- as a cost minimization problem, that is, to minimize ator or generation companies to be updated with real-time the total system cost (usually, the thermal production information and thereby follow the net load variation. The cost) [12]. Renewable energy sources have intermittent information is mainly collected from SCADA system, but it supply characteristics more diﬃcult to control compared is recommended to acquire from many other metrological to hydroenergy sources. As technology evolves renewable sensor devices for predicting renewable resources. Wireless energy sources have been being more controllable. But it is sensor networks seem very appropriate for this purpose on not yet enough, so they are also required to be supported by the aspect of power supply. Energy conservation plays a thermal plants. crucial role in wireless sensor networks since such networks It is assumed that load data and renewable outputs are are designed to be placed in hostile and nonaccessible forecasted as in Figure 30. Load data is forecasted by a areas. While battery-driven sensors will run out of battery 12 EURASIP Journal on Wireless Communications and Networking sooner or later, the use of renewable energy sources such [5] M. A. Hersh, “Sustainable decision making: the role of as solar or gravitation may extend the lifetime of a sensor decision support systems,” IEEE Transactions on Systems, Man [13]. Interaction and coordination between generators could and Cybernetics Part C, vol. 29, no. 3, pp. 395–408, 1999. mitigate the uncertainties caused by intermittent property [6] http://www.e-parl.net. of renewable energy sources based on information acquired [7] D.-J. Kang and H.-M. Kim, “A proposal for key policy of periodically from the existing SCADA system and wireless symmetric encryption application to cyber security of KEPCO networks. SCADA network,” in Proceedings of International Conference The data collected from various network routes are on Future Generation Communication and Networking (FGCN analyzed and used for IDMS for optimal decision-making on ’07), vol. 2, pp. 609–613, Jeju, Island, December 2007. short-term operational problems. [8] G. Clarke, D. Reynders, and E. Wright, Practical Modern SCADA Protocols, Newnes, Oxford, UK, 2004. [9]H.A.ReijersandW.M.P.vanderAalst,“Short-term 8. Conclusion simulation: bridging the gap between operational control and strategic decision making,” in Proceedings of the IASTED Renewable generators are increasing in electric power International Conference Modeling and Simulation (MS ’99), industry with many incentives on green technologies and pp. 417–421, May 1999. environmental regulations. This process is globally driven [10] http://www.kpx.or.kr. by governments, which causes many uncertainties on the [11] P. E. Morthorst, “Assessment and optimization of renewable aspects of physical system operation and electricity market support schemes in the European electricity market,” Work business activities. In addition, electricity markets are inter- Package 6 Interactions with Other Policies and Markets, RisΦ connected with other business area like emission trading Laboratory, 2006. markets, and thereby the system is getting complicated as [12] Z. Yu, F. T. Sparrow, and D. Nderitu, “Long-term hydrother- time goes on. Furthermore the system is required to be mal scheduling using composite thermal and composite hydro operated on real-time basis, so it is almost impossible for representations,” IEE Proceedings: Generation, Transmission humans to make a decision on the huge scale of problem and Distribution, vol. 145, no. 2, pp. 210–216, 1998. at every moment. This paper tries to propose a concept of [13] T. Voigt, H. Ritter, and J. Schiller, “Utilizing renewable an intelligent decision-making system to solve that kind of energy in cluster-based sensor networks,” in Proceedings of problem with renewable investment and operational issues the 1st Swedish National Computer Networking Workshop, instead of humans. For that purpose it is needed to have Arlandastad, Sweden, September 2003. its own intelligent decision-making algorithm independent of humans. The last section introduced the application concept and example of IDMS to the long-term decision- making problem related to investment issues and short- term one about operational issues, respectively. It is expected to deal with more detail algorithms for modeling IDMS such as game theory, genetic algorithm, and so forth. The IDMS could be helped by distributed sensor networks for collecting information and monitoring the recent status on the aspect of physical system infrastructures. And the battery problem for the sensor networks is solved by renewable energy sources dispersed over wide areas, which comes to the concept of green pervasive computing.

References

[1] R. Singh and Y. R. Sood, “Policies for promotion of renewable energy sources for restructured power sector,” in Proceedings of the 3rd International Conference on Electric Utility Deregulation and Restructuring and Power Technologies, pp. 1–5, Nanjuing, China, April 2008. [2] D. C. Quarton, “The evolution of wind turbine design analysis—a twenty-year progress review,” in Wind Energy, vol. 1, pp. 5–24, John Wiley & Sons, New York, NY, USA, 1998. [3] http://www.dailykos.com/story/2008/11/10/154815/14/322/ 658688. [4] R. J. Martel and J. J. Sudano, “Information optimization for decision making,” in Proceedings of the IEEE National Aerospace and Electronics Conference (NAECON ’97), vol. 1, pp. 454–461, Dayton, Ohio, USA, July 1997. Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 275694, 15 pages doi:10.1155/2009/275694

Research Article GRAdient Cost Establishment (GRACE) for an Energy-Aware Routing in Wireless Sensor Networks

Noor M. Khan,1 Zubair Khalid,2 and Ghufran Ahmed1

1 Department of Electronic Engineering, Mohammad Ali Jinnah University, Islamabad 44000, Pakistan 2 Faculty of Electronic Engineering, GIK Institute of Engineering Sciences and Technology, Topi 23640, Pakistan

Correspondence should be addressed to Ghufran Ahmed, [email protected]

Received 14 March 2009; Revised 27 September 2009; Accepted 8 October 2009

Recommended by Naveen Chilamkurti

In Wireless Sensor Network (WSN), the nodes have limitations in terms of energy-constraint, unreliable links, and frequent topology change. In this paper we propose an energy-aware routing protocol, that outperforms the existing ones with an enhanced network lifetime and more reliable data delivery. Major issues in the design of a routing strategy in wireless sensor networks are to make eﬃcient use of energy and to increase reliability in data delivery. The proposed approach reduces both energy consumption and communication-bandwidth requirements and prolongs the lifetime of the wireless sensor network. Using both analysis and extensive simulations, we show that the proposed dynamic routing helps achieve the desired system performance under dynamically changing network conditions. The proposed algorithm is compared with one of the best existing routing algorithms, GRAB. Moreover, a modiﬁcation in GRAB is proposed which not only improves its performance but also prolongs its lifetime.

Copyright © 2009 Noor M. Khan et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction networks can monitor natural habitats, remote ecosystems, endangered species, and emergency situations. 1.1. Overview. Advances in sensor technology, low-power In addition to sending the information to the sink, sensor electronics, and low-power radio frequency (RF) design have nodes also perform complex computations for decision enabled the development of small, relatively inexpensive making within the network, either individually or in local and low-power sensors, called microsensors, which can be clusters [6]. A major energy consumer in WSN is radio com- wirelessly connected [1–3] to form a wireless sensor network munication [3]. A comparison of the cost of computations to (WSN). The sensor nodes (or simply nodes) are usually that of communication by Pottie and Kaiser [3] reveals that deployed randomly and densely in hostile environment. 3000 instructions can be executed for the same cost as the Depending on the environment, it may or may not be feasible transmission of one bit over 100 m. An unlimited quantity to harness energy from ambient sources, such as solar power of data is generated by the physical world, but wireless [4]. telecommunication infrastructure is ﬁnite. This leads to a Sensor nodes collaborate to observe the surroundings burden on communication systems, computer networks, and send the collected information back to the sink (a node and human resources, which can be drastically reduced if responsible for collecting such information) in the case of raw data are processed at the source and the decisions any abnormal event. conveyed [5]. Hence by performing the computations inside WSNs ﬁnd their applications in many diverse indoor the network, communication payload may be reduced thus and outdoor areas including medicine, security, factory prolonging the network lifetime [6]. automation, environmental monitoring, and condition- The wired networks, unlike wireless sensor networks, are based maintenance [5]. In indoor settings, WSNs are already not limited by energy, node failure, and lack of a centralized being used for condition-based maintenance of complex controller. It is, therefore, easier to design and model a equipment in factories. In outdoor environment, these real-time wired network system. However, due to inherent 2 EURASIP Journal on Wireless Communications and Networking

from source to the destination over which the data are Satellite transmitted. This scheme, however, results in substantial energy overhead, suffers from cache pollution, and does not consider time constraint nature of the packets. Certain schemes like [20] require both GPS and GIS capability to Control centre find out the best route. Use of GPS-capable nodes is not recommended in sensor networks due to two reasons: firstly, it is too expensive in terms of power consumption to be used Sensor node Sensor field in power-aware networks. Secondly, it is subjected to failure when sensor nodes are deployed within some buildings, shades, tunnels, or caves [18]. In another real-time communication protocol, SPEED Mobile sink [21] achieves the goal of forwarding the packets closer (gateway) to the destination and takes into account the presence of hot regions and congestion at forwarding nodes into its Event area routing strategy. However, it does not take into account the energy of the forwarding nodes in order to balance the Figure 1: Wireless Sensor Networks. node energy utilization. Furthermore, the selection of region for forwarding data does not dynamically depend on the deadlines of the packets. SPEED also offers low reliability problems of multihop wireless sensor networks, the design since it does not transmit any redundant data packets and of a routing protocol, which is not only Quality of Service uses a single route for data delivery. Meanwhile several other (QoS) and energy aware [7] but also supports real-time strategies were also proposed to choose an optimal path for communication, is a challenging problem. Applications also real-time communication like minimal-load routing [22], ff set di erent delay requirements for the design of a routing minimal hop-routing, shortest-distance path routing [23], protocol in WSNs. For instance, in surveillance applications, and so forth, but these strategies do not specifically support authorities need to be notified sooner about high-speed the stateless architecture and the energy constraints of the motor vehicles than slow-moving pedestrians. To support sensor networks. such applications, a real-time communication protocol must Power Aware Chain (PAC) [24]protocolachievesa adapt its behavior based on packet deadlines. Hence, this relatively better network lifetime and is fault tolerant. It is implies that due to resource constraints of WSN platforms, a also scalable and does not require geographic information WSN protocol should introduce minimal overhead in terms to build routing chains. However it is highly complex and of communication and energy consumption (Figure 1). involves too many control overheads which in turn enhances its memory requirements in densely populated networks. 1.2. Literature Survey. A general data collection problem in PAC assumes that all nodes are capable of reaching the sink a given sensor network refers to the problem of routing the node which may not be possible in randomly deployed sensor data collected by the sensor nodes to the sink as efficiently as nodes. possible keeping in view the awareness of time and energy. Proactive Routing Protocol (PROC) [25] is another However, most of the conventional routing protocols do example of computationally expensive protocol and is used not consider time deadlines, energy, or congestion at the especially for real-time applications. Since it involves very forwarding nodes while routing a packet to its destination high control overhead and requires high memory, its per- [7]. Therefore, no single routing protocol performs well in formance thus degrades like SPEED in densely populated a complex real-world environment. If the impact of the networks. above-mentioned characteristics is also added to the routing Efficient And Reliable (EAR) [26] routing protocol also protocol designing problem, the situation is more intensified. uses proactive approach to build routes and thus is suited In order to address these challenging issues, efforts have for real-time applications. It routes the data reliably but dies been made by the researchers around the globe. One such out comparatively quicker due to energy depletion of the effort is to study the impact of energy utilization on the nodes around the hub (the node that collects the data from performance of WSN [8–11]. Several algorithms that lead the network and forwards it to the base station). It also to optimal connectivity topologies for power conservation needs global identifiers which may not be feasible for large have been proposed [12–17]. Later on these efforts were networks. extended for more rigorous solutions. Flooding information In GRAB [27], authors have focused on the problem of [7, 18] through the network was considered a common delivering messages from any sensor nodes to an interested way of ensuring real-time packet delivery. Nevertheless, this client along a minimum-cost path in a large sensor network. technique has extremely poor forwarding efficiency and Authors have presented a novel backoff-based cost field setup results in lot of redundant transmissions, increased energy algorithm that searches for the optimal costs of all nodes to consumption, and thus decreased network lifetime. the sink with one single message overhead at each node. Once A comparatively better approach had already been sug- the field is established, the message, carrying dynamic cost gested in [19], where a set of disjoint paths is maintained information, flows along the minimum cost path in the cost EURASIP Journal on Wireless Communications and Networking 3

field. Each intermediate node forwards the message only if As discussed earlier, a lot of work has been done in it finds itself to be on the optimal path, based on dynamic addressing the above issues in WSNs. However every listed cost states. The design does not require an intermediate node piece of work either discusses only one issue from the above to maintain forwarding path states explicitly. It needs a few two issues and ignores the other one completely or gives simple operations and has an ability to scale itself to any lesser importance to one or both of them. Our research network size. thus finds its directions to the theoretical underpinnings In [28, 29], Local Update-based Routing Protocol and design principles for an energy-efficient routing strategy (LURP) and Sensor Networks With Mobile Access (SeNMA) that can ensure sustainable higher throughput in WSN with protocol have been presented for WSNs with mobile sinks, prolonged lifetime. In addition, the aim of this work is to find respectively. In LURP, as the sink node moves, it only broad- a dynamic way to maintain an efficient routing structure with casts its location information within a local area rather than minimal overhead. broadcasting among the entire network. The node presents Organization of the rest of the paper is as follows. in that local area, communicating their data to the sink Section 2 discusses the proposed strategy, GRACE, in detail. dissipating lesser energy as compared to communicating the Section 3 presents various modes of operation involved same data from a distant location. This scheme also decreases in updating procedure of status information in routing the probability of collisions in wireless transmission. One tables of sensor nodes. Section 4 presents simulation results major drawback of this protocol is that the sink broadcasta considering various performance metrics, which are usually its location information to the entire network, whenever it used to evaluate the performance of routing strategy in goes outside the destination area. So if the network is large, a wireless sensor network. Section 5 proposes a modified the sink has to broadcast its location information to all of the and improved version of GRAB protocol. Finally, Section 6 sensor nodes in the entire network, which takes a lot of time concludes the paper and discusses the future work. and consumes a large portion of the available bandwidth. In SeNMA, an airplane acts as a mobile sink, which is not a 2. Proposed Routing Strategy—GRAdient Cost practical approach. The reason is that the sensor nodes have resource constraints like limited energy and low transmitting Field Establishment (GRACE) ability. However, a ground vehicle as a mobile sink is a The drawbacks and shortcomings of the routing strategies practical approach in many WSN applications [30]. discussed in Section 1.2 were properly dealt with imple- Chen et al. [31] have recently proposed a routing menting better broadcast routing approaches. The resulting protocol, named STEER (Spatial-Temporal relation-based improved routing strategy thus presents good results and ffi Energy-E cient Reliable routing protocol) which uses a outperforms the previous routing approaches published in distributed framework for routing data from source to the literature so far. sink. In traditional approaches, a path is usually established before the data transmitted. This degrades the performance of a routing protocol that does not work in a highly dynamic 2.1. GRACE System Model environment. In a dynamic environment, usually the path 2.1.1. Model Assumptions. We randomly deploy a large (or set of links, or next hop nodes) chosen at an earlier time number of sensor nodes in a monitoring area, which sense may not work well during data transmissions after a while. In the data and send it to the control center via stationary sink. STEER, a packet is broadcast first and the node closest to the We make the following assumptions in the present study. sink among all those neighbors that receive the packet will be chosen as the next hop relay nodes in a distributed manner. (i) To simplify the energy analysis, the time for sending However this approach is not bandwidth-efficient as a node a certain amount of data is assumed to be the same as broadcasts the data to each of its neighbors and thus uses the time for receiving the same amount of data. most of the bandwidth. (ii) The distance from the different nodes to the sink is From the above discussion, it can be concluded that the ignored as we are dealing with the number of hops main problems in using conventional protocols are [32] the instead of propagation delay which is usually based following: on the physical distance from source to the sink. (i) the size of processor and required memory are too (iii) All sensor nodes are assumed to be homogeneous; large; therefore the energy consumption for sensing is the (ii) the bandwidth required is too high; same to each sensor node. (iii) the protocols are not energy usage aware. 2.1.2. Stochastic Model. As we know that the radio pattern These problems lead to an interesting debate on the fun- is largely random, there are certain other factors which damental limits of wireless sensor network. The debate starts are also random; but once we pick a particular value of with the basic question of what the maximum sustainable a parameter for an experiment, it becomes deterministic. throughput and the maximum lifetime of a network are. For example, the value of transmission power can be a The answers to these and similar other questions are of great uniformly distributed random variable and can be varied importance to both the theoretical and practical aspects of from [max, min], but in order to start an experiment we wireless sensor networking research. pick a particular power value. This value remains constant till 4 EURASIP Journal on Wireless Communications and Networking

= = the end of the experiment. Hence, for an entire process, the X(tnS) X(tn) Xn value of transmission power can be selected randomly from its domain; therefore the process is called as random process or stochastic process. We can apply same procedure to the weather conditions and other environmental factors. After completing X(t, S1) t theexperimentsatdifferent parameter values, the entire process becomes a random process and we can apply statistical techniques on it. Figure 2 shows a set of index X(t, S2) t random variables which combine to form a whole random process. It is also called a set of samples or a set of sample paths or realization. Here we take different data X(t, S3) t samples X(t, S1), X(t, S2), ..., X(t, Sn)fromeachofdifferent sensor nodes S1, S2, S3, ..., Sn after a specific time interval t1, t2, t3, ..., tn. The collection of data points from different X(t, S4) t sensor nodes at each time tn is represented by a random variable Xn as shown in Figure 2. Associated with each of these random variables is a probability mass function X(t, S5) t (pmf) or a probability density function (pdf). Therefore if there are n index random variables: x1, x2, x3, ..., xn, then for each random variable xn, there is an associated pdf fXn(x). In addition, there is a joint pdf corresponding to X(t, Sn) t all of these pdfs. In other words, in order to represent the entire random process which consists of a set of index random variables x1, x2, x3, ..., xn, we should have a joint pdf f(x1,x2,x3,...,xn) which can represent or characterize the entire Time: t1 t2 t3 ··· tn random process. We can get this joint pdf by summing up RV: x1 x2 x3 ··· xn ··· each of these individual pdfs. PDF: fx1 (x) fx2 (x) fx3 (x) fxn (x) The joint probability density function is given by Figure 2: Random Process. = fx1,x2,x3,...,xn fx(t)(x). (1) The mean, variance, autocorrelation, autocovariance, and We are dealing with an event-based WSN system where the correlation coefficient values of the Random Process (RP) sensor nodes activate whenever an event occurs. These events can be obtained from (2), (3), (4), (5), and (6), respectively: occur according to a random process with a rate denoted as λ. Hence we collect the data X each time an event occurs. (i) Mean: Let X(t) be the total data collected till time t, as shown in +∞ Figure 3: mx(t) = fx(t)xdx,(2) −∞ n X(t) = x(i). (7) (ii) Variance: i=0 +∞ The probability that the total data collected till time t, X(t), = − 2 var[x(t)] x mx(t) fx(t)xdx, equal to j is given by −∞ (3) ∞ 2 2 j var[x(t)] = E x (t) − E [x(t)], P X(t) = j = P X(t) = = n P[N(t) = x]. (8) n=0 N(t) (iii) Auto Correlation: Here Xn is a poison process, and therefore ∞ Rx(t ,t ) = E[x(t1)x(t2)] = E[x1x2],(4) j 1 2 j n − P X(t) = = n = exp n. (9) N(t) j! (iv) Auto Covariance: n=0 Hence, (8)becomes = − Cx(t1,t2) Rx(t1,t2) mx(t1)mx(t2),(5) ∞ j n n − (λt) − P X(t) = j = exp n exp λt. (10) (v) Correlation Coefficient: n=0 j! n!

= Cx(t1,t2) ρx(t1,t1) . (6) 2.1.3. GRACE Parameters. Each sensor node is deﬁned by C C x(t1,t1) x(t2,t2) a infovalue pair. These infovalue pairs have already been EURASIP Journal on Wireless Communications and Networking 5

Total number of events: N(t) = n Sink 123 4··· n IL,j−sink IL,L−sink

I − j L,k sink L 0 t IL,j−k k IL,k−L x0 x1 x2 x3 ··· xn

(λt)n PMF = e−λt IL,i−k n! IL,i− j IL,i−L Figure 3: Poisson Process.

discussed in our previous work [33] and are discussed here i again brieﬂy. Figure 4: Cost Field Establishment. Energy of Node, IE,i. In order to increase the lifetime of WSN, low-energy nodes are avoided in routing. This is achieved by maintaining the following attribute for each (i) Let C be the cost of the path which heads to the node: i-Sink sink from the ith node. 0 Pi (ii) Let Cij be the cost of the path which heads to the sink IE,i = , (11) Pi via jth node from the ith node.

0 (iii) Let Ai be the advertisement packet broadcasted by ith where Pi is the remaining battery power and Pi is the starting node to its immediate neighbors. battery power. From the above formula, we can conclude that we should avoid those paths which contain nodes having The cost field propagation is better understandable by high value of IE,i. an example. As shown in Figure 4,nodesj, k,andl are the Link Cost, IL. The proposed strategy uses link costs that immediate neighbors of the ith node. We can define the cost reflect the communication energy consumption rates at the fields and advertisement packets as follows, two end nodes. The aim of the strategy is to maximize the A = C + I , lifetime of the network by carefully defining link cost as a j j-Sink E,j function of receiving and transmission power using that link. Ak = Ck-Sink + IE,k, The transmission-value is set initially same for all the nodes. = The link cost between nodes u and v can be measured as Al Cl-Sink + IE,l, follows: = Cij Aj + IL,i− j , (13) = Pt,u IL,u−v , (12) Cik = Ak + IL,i−k, Pr,v C = A + I − , where P is the transmission power of node u and P is il l L,i l t,u r,v the received power of node v. For convenience in use, we will Ci-Sink = min Cij, Cik, Cil . represent IL,u−v as IL from now onward. Intuitively, a link that has high value of IL means Initially Cnode-Sink is set to infinite for all the nodes that there exist more chances of packet drop and more in the sensor field. The sink initiates the setup phase by transmission energy would be required to overcome the broadcasting the advertisement packet containing the cost hindrances of the path. So we can conclude that we should ASink = 0 to all of its immediate neighbors. When a node avoid such links that have higher values of IL. receives the advertisement message with the cost, it stores the cost in its routing table. Then it calculates the link 2.2. Phases of GRACE cost IL,node-Sink, as described in (12).Thus,anode’srouting table contains cost C received from each of its immediate 2.2.1. Setup Phase Algorithm. Most of the WSNs routing neighbors along with the neighbors’ id. Now, the receiving strategies are data-centric. In data-centric strategies, sink node (say i) picks the smallest C value from its routing table, sends interest packets to the area in the sensor field where adds its own IE,i cost in it, and broadcasts this final value Ai it wants to collect the data. However in our strategy, which to all of its immediate neighbors. Also, the receiving node is more generalized as compared to the above mentioned considers the smallest value node as the relay node to send approach, the sink initiates the setup phase for the entire data back to the sink. The similar algorithm is running on WSN. In the setup phase, a cost propagates throughout other nodes and this process continues till the last node of the the sensor field. This cost field is established using the sensor field. Once the setup phase is completed, the steady- advertisement packet. state phase is performed to find the best path. 6 EURASIP Journal on Wireless Communications and Networking

Sink performance of any routing strategy depends on the use of D any particular mode. In this section, we present the behavior 3 4 of our proposed routing strategy under the operation of these 1 modes. These modes of operation are given as follows J 2 2 1 B 1 C (1) Single Setup (SS) Alone Mode, 1 1 E 4 (2) Unicast Acknowledgement Mode, G 1 F (3) Broadcast Acknowledgement Mode, 1 I 2 3 (4) Correction Mode (starting from the sink), H (5) Correction Mode (starting from the intermediate Figure 5: Example Scenario. node). The setup phase will be run at start and information 2.2.2. Steady-State Phase Algorithm. After the completion update will be made according to the operation of these of the setup phase, the source node sends the data to that modes. The plots showing the behavior of these modes on particular node which has the smallest cost C value in its the performance of the network would consequently be used routing table. The receiver then forwards the data to that for choosing the best mode of operation for the information node having the smallest cost C value in its routing table and update procedure. the same process continues till the data reach to the sink. In order to update the status information of sensor nodes, we 3.1. Single Setup (SS) Alone Mode. Inthismodeofoperation, ff propose di erent modes of operations that will be discussed the setup phase runs only once at the startup. Thus later on in detail in Section 3. using this mode, there is no mechanism to update the status information of sensor nodes. This leads to the continuous 2.3. An Example Scenario of the Proposed Strategy. The setup usage of a routing path till any of the node in the path dies. and steady-state phases can be better understandable if We take a network, deployed in an area of 50 m × 50 m as an we take an example. Let us take an example network as example to illustrate various modes of operations. shown in Figure 5. The energy levels and the link costs are calculated using (11)and(12), respectively. First the SINK node broadcasts the advertisement message to nodes 3.2. Unicast Acknowledgement Mode. Since every node has cost factors of its neighbor nodes, it selects node for routing B, D,andJ. This advertisement message contains the cost data that has minimum cost. Later on, this cost factor is A = 0. Nodes B, D,andJ receive the message, calculate Sink updated in such a way that the receiving node sends an their respective link costs I , I ,andI ,and L,B-Sink L,D-Sink L,J-Sink acknowledgement to the sender whenever it receives the then add their link costs to A to form C , C , Sink B-Sink D-Sink data. This acknowledgement comprises of one extra byte, and CJ-Sink,respectively.NodesB, D,andJ store these information in their routing tables, as shown in Table 1.After showing the current minimum cost factor of the receiver a certain period of time, which depends on these costs as node. Thus, the updates propagate in the sensor field by sending acknowledgments for the received data. Figure 6 discussed in [27], the nodes select the minimum cost C x-Sink shows the Unicast Acknowledgement Mode. from their routing tables, add their own energy cost IE in it, and broadcast it to all of their immediate neighbors (In the figure node B broadcasts its advertisement AB to nodes A, C, 3.3. Broadcast Acknowledgement Mode. One major drawback and E.NodeD broadcasts its advertisement AD to nodes of the acknowledgement phase is that only the sender A, C,andG.NodeJ broadcasts its advertisement AJ to nodes knows about the updated status information of the receiving A and I). The same procedure also runs at nodes G, C, E, node. In order to prevent from it, the receiving node can and I. This process goes on one after the other according to broadcast the acknowledgement along with its updated their intervals, till the last node of the sensor field establishes status information to all of its immediate neighbors. In its routing table. After the setup phase, steady-state phase this way, a node can inform all of its neighbors about its begins. We take node H as a source node. Now node H looks updated status information. Figure 7 shows the Broadcast for the node in its routing table which has the smallest cost Acknowledgement Mode. C. In our case, it is node F;sonodeH sends the data to node F. Same decisions for forwarding data are made on other 3.4. Correction Mode (Starting from the Sink). Whenever nodes. In this way data reach the sink with minimal routing a node sends data packet to another node, it keeps the overhead. packet ID in its buffer. Similarly, every node gets a list of all the packet IDs it receives. Whenever a packet reaches 3. Modes of Operation for Updating the sink, sink sends the acknowledgment to the node from Status Information which it receives the packet. That node then broadcasts the acknowledgement containing its updated status information We propose various modes of operation for updating to all of its neighbors along with data packet IDs. The packet status information of the sensor nodes in the WSNs. The ID will help recognize the corresponding node among the EURASIP Journal on Wireless Communications and Networking 7

Table 1: Energy Levels of Nodes at some time after the deployment of the Network.

ID ABCDEFGHI J

IE 02345678910

Table 2: Cost Fields. ith Node Neighbor jth Node Aj IL,i− j Cij Ci-Sink IE,i Ai Sink 0 1 1 BC82101 23 E 9110 C 8210 D Sink 0 4 4 4 4 8 G 16 1 17 E 9110 J Sink 0 3 3 3 10 13 I 26 4 30 D 8210 CB325 5 38 F 15 1 16 J 13 1 14 EB314 4 59 F 15 1 16 E 9110 FC819 9 615 H 22 1 23 H 22 2 24 G 9716 D 819 I 26 3 29 HG16 2 18 16 8 22 F 15 1 16 H 22 3 25 I 17 9 26 J 13 4 17 neighbors which took part in carrying that packet. This Table 3: Parametric values used in Simulations. process will continue till the source node, which originated Parameters Value the data packet, get the corrected cost of the path used in carrying its data. Storing packet IDs gives an extra burden to Number of nodes 250 the node memory. In order to minimize this burden, node Initial energy 100 J will use a speciﬁed memory for packet ID storing on FIFO Communication Range 10 m basis. Consequently, in case of congestion in a particular Sensor ﬁeld size 50 × 50 m2 region of the network, node will lose the packet ID from its Data rate 40 kbps memory and hence will stop broadcasting for not allowing Simulation Time 1500 units an increase in the congestion. Figures 8(a) and 8(b) show the Correction Mode (Starting from the sink).

3.5. Correction Mode (Starting from the Intermediate Node). 4. Results and Discussion Sometimes the packet is lost or dropped at some intermediate node. In this case the correction mode will not be 4.1. Simulation Setup. To investigate the performance and initiated as the packet is not reached at the sink. Therefore the scalability of the proposed protocol, we generate a sensor there must be a mechanism which initiates the correction network comprising of 100 nodes and carry out extensive operation at any intermediate node, so that the updated simulations in Matlab 6.0 in order to validate the proposed cost field is propagated along the entire path. Correction routing strategy under different modes of operation. Our operation starting from the intermediate node is a solution sensor field’s dimension is 0.0025 Kilometer Square. The for it. Figures 9(a) and 9(b) show the Correction mode numerical values chosen for our simulations can be seen in (Starting from the intermediate node). Table 3. 8 EURASIP Journal on Wireless Communications and Networking

50 50 Sink Sink 45 45 3 3 40 40 1 2 7 1 2 7 35 5 35 5 4 6 20 4 6 20 30 8 30 8 25 10 9 25 10 9 20 13 20 13 Distance (m) 19 12 Distance (m) 19 12 15 18 15 18 14 14 10 16 15 17 10 16 15 17

5 5 11 11 Source Source 0 0 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 Distance (m) Distance (m)

Data packet Data packet Unicast acknowledgment (a) Figure 6: Unicast Acknowledgment Mode. 50 Sink 45 3 50 40 Sink 1 2 7 45 3 35 5 40 1 4 6 20 2 7 30 8 35 5 25 10 9 4 6 20 30 8 20 13

Distance (m) 12 25 9 19 10 15 18 14 20 13 10 16 Distance (m) 19 12 15 17 15 18 5 16 14 11 Source 10 15 17 0 0 5 10 15 20 25 30 35 40 45 50 5 11 Source Distance (m) 0 0 5 10 15 20 25 30 35 40 45 50 B.C. from sink Distance (m) (b) Data packet B.C. from node Figure 8: Correction Mode (Starting from the sink). (a) Data Packets. (b) Acknowledgment Packets. Figure 7: Broadcast (B.C) Acknowledgment Mode.

4.2. Performance Metrics. A set of performance metrics is 4.2.1. Network Lifetime (in Terms of Node Failures, f ). It used for evaluating the performance of the proposed strategy. shows how much time the network will alive. In Figure 10, One point that should be kept in mind is the degree number of alive nodes is plotted against simulation time of goodness or badness of the results. It is clear that it units. It can be seen that the correction mode from depends on the working life of the network. A network intermediate node has the lowest working life while the having only one established path from the source to the broadcast acknowledgement mode has the highest working sink is much better than the network that has got large lifetime, thus keeping a large number of nodes alive with number of disconnected nodes scattered in the field. This high data rate and reliable data delivery. The reason of this takes us to the strategy that utilizes the network nodes on a difference in results is that setup phase with the broadcast uniform balanced manner. Another criterion that promises acknowledgement uses the nodes evenly in terms of energy the reliability and useability of the network is preventing utilization, while the other approaches like GRAB [27]donot the nodes from dying till a large number of nodes die out ensure a balance utilization of nodes. collectively. The collective death of a large number of nodes In Figure 11, we draw a bar graphs of the node failure, will ensure a reliable data delivery and network operation for f (in percentage) versus time elapsed. It is also clear a specified time. This time would thus give us a prediction from that when first node dies, single setup with unicast about the safe operation of the network. The use of network acknowledgement mode has longer time elapsed, while the beyond this time would make its operation unreliable and single setup mode and GRAB [27] have the lowest time unpredictable. The figures show the result obtained under elapsed. This is due to the fact that in case of single various scenarios and modes of operation. setup mode, which is based upon the initial nodes’ status EURASIP Journal on Wireless Communications and Networking 9

50 250 Sink 45 3 Intermediate node

40 f 200 1 2 7 35 5 4 20 30 6 8 150 25 10 9 20 13 100

Distance (m) 12 15 19 18

14 Number of alive nodes, 10 16 15 17 50 5 11 Source 0 0 0 5 10 15 20 25 30 35 40 45 50 0 500 1000 1500 Distance (m) Time t (units)

Data packet Single setup (SS) SS with unicast acknowledgement (a) SS with broadcast acknowledgement 50 SS with correction from sink Sink SS with correction from intermediate node 45 3 SS with hybrid correction + acknowledgement 40 Intermediate node 1 2 7 GRAB, event based setup initialization (Ye et al.) 35 5 4 20 30 6 8 Figure 10: Network Lifetime: SS Alone, SS with Unicast, SS with 25 9 Broadcast, SS with Correction from Sink, SS with Correction from 10 Intermediate Node, SS with Hybrid Mode and GRAB, an event- 20 13

Distance (m) 12 based setup initialization (Ye et al. [27]). 19 15 18 16 14 10 17 15 5 11 Source 0 0 5 10 15 20 25 30 35 40 45 50 600 Distance (m)

Broadcast acknowledgment 500 (b) 400 Figure 9: Correction Mode (Starting from the intermediate node). (a) Data Packets. (b) Acknowledgment Packets. 300

200 information, it continuously uses a path till any of the nodes Network time elapsed in the path dies. While in case of GRAB [27], the setup phase will not run till the occurrence of any event. 100

0 4.2.2. Network Energy Left, e. It shows the amount of energy 110203040 left, e, in the alive nodes whether connected or disconnected Node failure, f (% age) in the network with the passage of time. Figure 12 shows plots of the network energy versus simulation time. From the Single setup (SS) ﬁgure, it is clear that use of single setup mode outperforms SS with unicast acknowledgement the others if energy consumption is considered. This is due SS with broadcast acknowledgement to the fact that the setup phase runs only at the startup and SS with correction from sink no acknowledgment and correction is done at later times. SS with correction from intermediate node SS with hybrid correction + acknowledgement Although this mode is good in the energy consumption sense GRAB, event based setup initialization (Ye et al.) but as a result of not using acknowledgement and correction, it loses data reliability as compared to other nodes. Figure 11: Node Failure in Percentage: SS Alone, SS with Unicast, SS with Broadcast, SS with Correction from Sink, SS with Correc- 4.2.3. Data Reliability, μ. It shows the success ratio of the data tion from Intermediate Node, SS with Hybrid Mode and GRAB, an packets, that is, the number of data packets received by the event-based setup initialization (Ye et al. [27]). 10 EURASIP Journal on Wireless Communications and Networking

100 100

90 90

80 80 (%) e μ 70 70

60 60

50 50 Network energy, 40

Age packet delivered, 40

30 30

20 20 0 500 1000 1500 0 500 1000 1500 Time t (units) Time t (units)

Single setup (SS) Single setup (SS) SS with unicast acknowledgement SS with unicast acknowledgement SS with broadcast acknowledgement SS with broadcast acknowledgement SS with correction from sink SS with correction from sink SS with correction from intermediate node SS with correction from intermediate node SS with hybrid correction + acknowledgement SS with hybrid correction + acknowledgement GRAB, event based setup initialization (Ye et al.) GRAB, event based setup initialization (Ye et al.)

Figure 12: Network Energy Left: SS Alone, SS with Unicast, SS with Figure 13: Data Delivery in Percentage: SS Alone, SS with Broadcast, SS with Correction from Sink, SS with Correction from Unicast, SS with Broadcast, SS with Correction from Sink, SS with Intermediate Node, SS with Hybrid Mode and GRAB, an event- Correction from Intermediate Node, SS with Hybrid Mode and based setup initialization (Ye et al. [27]). GRAB, an event-based setup initialization (Ye et al. [27]). sink out of the total number of data packets generated by the source. In Figure 13 one aspect of data reliability comparison is shown, where the plots represent the percentage data 100 delivery with respect to simulation time. It is clear from the 90

ﬁgure that the hybrid approach and the single setup with (%) broadcast acknowledgement have high data reliability. This 80 is due to the fact that the status information of the sensor 70 nodes is updated frequently, in these modes of operation. interval based 60 Another aspect of data reliability comparison is shown in μ Figure 14, where the plots show interval-based data delivered 50 to the sink after a speciﬁed time interval (e.g., after each 100 40 seconds in our case); we note down the number of data pack- 30 ets received at the sink. It can be noted from the plots that initially the single setup with broadcast acknowledgement 20 mode has the highest percentage of delivered packets to the Age packet delivered, 10 sink but cannot keep its pace at later times and degrades its 0 performance due to bulk node failures. 0 500 1000 1500 Discussing the last aspect of data-delivery performance Time t (units) comparison, the packet received by the sink have been plotted against the packets sent by the source. Figure 15 shows Single setup (SS) SS with unicast acknowledgement that the single setup with broadcast acknowledgement mode SS with broadcast acknowledgement has large number of packets received. The reason is obvious SS with correction from sink that in the single setup with broadcast acknowledgement SS with correction from intermediate node mode status information of the sensor nodes is updated SS with hybrid correction + acknowledgement frequently and thus nodes are evenly utilized. GRAB, event based setup initialization (Ye et al.)

Figure 14: Interval-based Data Delivery in Percentage: SS Alone, 4.2.4. Collective Performance Metric, β = ( f × μ × e). SS with Unicast, SS with Broadcast, SS with Correction from Sink, The Collective Performance Metric, β,canbeusedto SS with Correction from Intermediate Node, SS with Hybrid Mode reﬂect the network energy left, reliability, and the node and GRAB, an event-based setup initialization (Ye et al. [27]). EURASIP Journal on Wireless Communications and Networking 11

500 1 450 0.9 400 0.8 350 0.7

300 e 0.6 × μ

250 × 0.5 f

200 = 0.4 β Packet received 150 0.3 100 0.2 50 0.1 0 0 0 500 1000 1500 0 500 1000 1500 Packet sent Time, t (units)

Figure 15: Packet send versus Packet Received: SS Alone, SS with Figure 16: Collective Performance Metric, β: SS Alone, SS with Unicast, SS with Broadcast, SS with Correction from Sink, SS with Unicast, SS with Broadcast, SS with Correction from Sink, SS with Correction from Intermediate Node, SS with Hybrid Mode and Correction from Intermediate Node, SS with Hybrid Mode and GRAB, an event-based setup initialization (Ye et al. [27]). GRAB, an event-based setup initialization (Ye et al. [27]).

failures (Figure 16). It is clear from Figure 23 that the hybrid of the performance of the modified GRAB with that of approach and the single setup with broadcast acknowledge- the simple GRAB [27] can be well visualized simply by ment have high value of this metric. This is due to the fact observing how node failures and packet losses affect the that the status information of the sensor nodes is updated network lifetime and packet delivery in a wireless sensor frequently. network. Figure 17 shows the resulting pattern of alive nodes after 5. Modified GRAB the introduction of broadcast acknowledgement strategy in GRAB. It is clear from the figure that a large number of nodes GRAB [27] strategy discussed in Section 1 is based on die out almost at the same time in modified GRAB, whereas dynamic cost information which is used to find an opti- simple GRAB exhibits worst performance as some nodes that mal path from source to the sink. It ensures a robust remain alive in simple GRAB find themselves disconnected data delivery using unreliable sensor nodes. However, each from the rest of the network nodes. The reason is that in packet is forwarded over multiple paths, which increases simple GRAB, nodes die out at a constant interval of time the probability of data delivery of packets to the sink on due to which existence of some alive nodes does not prevent one hand, but results in high bandwidth consumption, and the network to fall into a not-connected state. Figure 18 increased redundancy and more interference, on the other also shows that the modified GRAB with the inclusion of hand. Although these demerits are the results of a tradeoff broadcast acknowledgement strategy keeps the network alive for reliable data delivery, yet a slight modification in the for a relatively longer period of time. Figure 19 elaborates GRAB strategy can also assure its energy-efficient use with no that simple GRAB leaves behind a large amount of energy significant loss in data delivery. The proposed modification which remains unutilized till the death of the whole network. is to decrease the number of broadcast messages. This can A good routing strategy should residue as little amount of be done either through setup phase with acknowledgement energy as possible. GRAB [27] can also be modified with or with the introduction of setup phase with correction. the inclusion of hybrid strategy, that is, a combination of Both approaches have been discussed in Section 3.Ithas the correction and acknowledgement strategies, discussed been observed there that these approaches in the routing in Section 3. Figure 20 shows a significant improvement in strategy result in less bandwidth consumption and efficient the delivery of successful packets when GRAB is used with energy utilization. Figures 17, 18, 19, 20, 21, 22,and23 the hybrid of correction and acknowledgement strategies. show the resulting improvements in GRAB [27] with the When the time exceeds 1400 units, the percentage of packets introduction of the proposed modifications. A comparison delivery goes down to its minimum value in simple GRAB 12 EURASIP Journal on Wireless Communications and Networking

250 100

90 200

f 80 e 150 70 60

100 50 Network energy,

Number of alive nodes, 40 50 30

0 20 0 500 1000 1500 0 500 1000 1500 Time t (units) Time t (units)

GRAB, event based setup initialization (Ye et al.) GRAB, event based setup initialization (Ye et al.) GRAB with unicast acknowledgement GRAB with unicast acknowledgement GRAB with broadcast acknowledgement GRAB with broadcast acknowledgement GRAB with correction from sink GRAB with correction from sink GRAB with correction from intermediate node GRAB with correction from intermediate node GRAB with hybrid correction + acknowledgement GRAB with hybrid correction + acknowledgement

Figure 17: Network Lifetime: GRAB, an event-based setup initial- Figure 19: Network Energy Left: GRAB, an event-based setup ization (Ye et al. [27]), GRAB with Unicast, GRAB with Broadcast, initialization (Ye et al. [27]), GRAB with Unicast, GRAB with GRAB with Correction from Sink, GRAB with Correction from Broadcast, GRAB with Correction from Sink, GRAB with Correc- Intermediate Node, and GRAB with Hybrid Mode. tion from Intermediate Node, and GRAB with Hybrid Mode.

600 100

500 90

(%) 80

400 μ

70 300 60 200

Network time elapsed 50 Age packet delivered, 100 40

0 30 110203040 0 500 1000 1500 Node failure, f (% age) Time t (units)

Figure 18: Node Failure in percentage: GRAB, an event-based Figure 20: Data delivery in Percentage: GRAB, an event-based setup initialization (Ye et al. [27]), GRAB with Unicast, GRAB setup initialization (Ye et al. [27]), GRAB with Unicast, GRAB with Broadcast, GRAB with Correction from Sink, GRAB with with Broadcast, GRAB with Correction from Sink, GRAB with Correction from Intermediate Node, and GRAB with Hybrid Mode. Correction from Intermediate Node, and GRAB with Hybrid Mode. EURASIP Journal on Wireless Communications and Networking 13

100 600 90

(%) 500 80 70 400

interval based 60 μ 50 300 40 Packet received 200 30 20 100 Age packet delivered, 10 0 0 0 500 1000 1500 0 500 1000 1500 Time t (units) Packet sent

Figure 21: Interval-based Data delivery in Percentage: GRAB, an Figure 22: Packet Send Vs. Packet Received: GRAB, an event- event-based setup initialization (Ye et al. [27]), GRAB with Unicast, based setup initialization (Ye et al. [27]), GRAB with Unicast, GRAB with Broadcast, GRAB with Correction from Sink, GRAB GRAB with Broadcast, GRAB with Correction from Sink, GRAB with Correction from Intermediate Node, and GRAB with Hybrid with Correction from Intermediate Node, and GRAB with Hybrid Mode. Mode. but GRAB with hybrid correction and acknowledgement strategy still keeps a higher rate. This is because the updated 1 cost information is propagated along the entire path even in the case of packet loss at some intermediate nodes. Thus 0.9 the packets that reach their destinations successfully rise 0.8 percentage of packet delivery. 0.7 Figure 21 shows interval-based data delivered to the e 0.6 sink after a specified time interval (e.g., after each 100 × μ seconds in our case). After noting down the number of data × 0.5 packets received at the sink, it can be seen that GRAB with f = 0.4 broadcast acknowledgement results in the highest percentage β of interval based packets delivered to the sink. In Figure 22, 0.3 GRAB with hybrid correction and acknowledgement strategy 0.2 results in the highest ratio of packets received to packets sent, while GRAB with correction from sink results in the 0.1 lowest ratio of packets received to packets sent. Performance 0 comparison on the basis of combined metric has been given 0 500 1000 1500 in Figure 23. It shows the impact of all of the considered Time, t (units) performance parameters. It is clear from Figure 23 that GRAB, event based setup initialization (Ye et al.) GRAB with hybrid mode and GRAB with broadcast mode GRAB with unicast acknowledgement show the highest performance among all other modification GRAB with broadcast acknowledgement modes of the GRAB. GRAB with correction from sink GRAB with correction from intermediate node GRAB with hybrid correction + acknowledgement 6. Conclusions and Future Work Figure 23: Collective Performance Metric: GRAB, an event-based In this paper, we have proposed an energy-aware routing setup initialization (Ye et al. [27]), GRAB with Unicast, GRAB strategy based on GRAdient Cost Establishment (GRACE) with Broadcast, GRAB with Correction from Sink, GRAB with for Wireless Sensor Networks. The proposed routing strategy Correction from Intermediate Node, and GRAB with Hybrid outperforms the existing ones with an enhanced network Mode. 14 EURASIP Journal on Wireless Communications and Networking lifetimeandmorereliabledatadelivery.Asetupmechanism a lifetime constraint,” IEEE Transactions on Mobile Computing, governing the GRACE scheme has also been discussed in vol. 4, no. 1, pp. 4–15, 2005. detail. Various modes of operation for updating status [12] D. J. Chmielewski, T. Palmer, and V. Manousiouthakis, “On information of the sensor nodes have been indicated. More- the theory of optimal sensor placement,” AIChE Journal, vol. over, some performance metrics have been set to evaluate 48, no. 5, pp. 1001–1012, 2002. the performance of WSNs. A comparison of the proposed [13] C. Zhou and B. Krishnamachari, “Localized topology genera- strategy, GRACE, with a well-known event-based cost field tion mechanisms for wireless sensor networks,” in Proceedings establishment scheme, GRAB [27], has been given which of IEEE Global Telecommunications Conference (GLOBECOM ’03), vol. 3, pp. 1269–1273, San Francisco, Calif, USA, showsabetterperformanceofGRACEoverGRAB.Some December 2003. modifications have also been suggested in the GRAB scheme, [14] S. Ghiasi, A. Srivastava, X. Yang, and M. Sarrafzadeh, “Optimal which improve the performance of GRAB with respect to energy aware clustering in sensor networks,” Sensors, vol. 2, no. ffi bandwidth e ciency and network life time. The proposed 7, pp. 258–269, 2002. research work can be extended to a cost-based globally [15] V. Rodoplu and T. H. Meng, “Minimum energy mobile gradient setup mechanism which reduces the number of wireless networks,” IEEE Journal on Selected Areas in Commu- broadcast messages made by the sensor nodes during cost nications, vol. 17, no. 8, pp. 1333–1344, 1999. field establishment procedure. Since the density of nodes [16] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, affects the number of broadcast messages, therefore the “Energy-efficient communication protocol for WSN,” in Pro- proposed strategy GRACE can be modified in accordance ceedings of the 33rd Hawaii International Conference on System with the density of nodes in the vicinity of sink in order to Sciences, Wailea Maui, Hawaii, USA, January 2000. improve the lifetime of the sensor’ network. [17] J.-H. Chang and L. Tassiulas, “Energy conserving routing in wireless ad-hoc networks,” in Proceedings of the 19th Annual Joint Conference of the IEEE Computer and Communications References Societies (INFOCOM ’00), vol. 1, pp. 22–31, Tel Aviv, Israel, 2000. [1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, [18] M. Saleem, “Routing mchanisms in wireless sensor networks: “A survey on sensor networks,” IEEE Communications Maga- survey and analysis,” Tech. Rep., Centre for Advance Studies in zine, vol. 40, no. 8, pp. 102–114, 2002. Engineering, Islamabad, Pakistan, December 2005. [2] J. Agre and L. Clare, “An integrated architecture for coopera- [19] S. J. Lee and M. Gerla, “Split multipath routing with tive sensing networks,” IEEE Computer, vol. 33, no. 5, pp. 106– maximally disjoint paths in ad hoc networks,” Tech. Rep., 108, 2000. University of California, Oakland, Calif, USA, 2000. [3] G. J. Pottie and W. J. Kaiser, “Wireless integrated network [20] U. Bilstrup and P. A. Wiberg, “Routing protocol for wireless sensors,” Communications of the ACM, vol. 43, no. 5, pp. 51– real-time multihop networks,” in Proceedings of the 11th 58, 2000. Euromicro Conference on Real-Time Systems (ECRTS ’99), [4] D. Ganesan, A. Cerpa, W. Ye, Y. Yu, J. Zhao, and D. Estrin, York, UK, June 1999. “Networking issues in wireless sensor networks,” Journal of [21] T. He, J. Stankovic, C. Lu, and T. Abdelzaher, “Speed: a Parallel and Distributed Computing, vol. 64, no. 7, pp. 799–814, real time routing protocol for sensor networks,” Tech. Rep. 2004. CS-2002-09, University of Virginia, Charlottesville, Va, USA, [5] G. Pottie, “Wireless integrated network sensors (wins): the March 2002. web gets physical,” The Bridge, vol. 31, no. 4, pp. 22–27, 2001. [22] J. R. A. Shaikh and K. Shin, “Dynamics of quality-of-service [6] S. Park, A. Savvides, and M. B. Srivastava, “Simulating routing with inaccurate link-state information,” Tech. Rep. networks of wireless sensors,” in Proceedings of the 33rd Winter CSE-TR-350-97, Department of Electrical Engineering and Simulation Conference, vol. 2, pp. 1330–1338, Arlington, Va, Computer Science, University of Michigan, Ann Arbor, Mish, USA, December 2001. USA, November 1997. [7] A. Mahapatra, K. Anand, and D. P. Agrawal, “QoS and energy [23] Q. Ma and P. Steenkiste, “On path selection for traffic with aware routing for real-time traffic in wireless sensor networks,” bandwidth guarantees,” in Proceedings of the International Computer Communications, vol. 29, no. 4, pp. 437–445, 2006. Conference on Network Protocols (ICNP ’97), pp. 191–202, [8]S.SlijepcevicandM.Potkonjak,“Powerefficient organization Atlanta, Ga, USA, October 1997. of wireless sensor networks,” in Proceedings of IEEE Interna- [24] M. L. Pham, D. Kim, Y. Doh, and S. Yoo, “Power aware tional Conference on Communications (ICC ’01), vol. 2, pp. chain routing protocol for continuous data dissemination in 472–476, Helsinki, Finland, June 2001. wireless sensor networks,” in Proceedings of the 1st Interna- [9] B. Krishnamachari and F. Ordońez,˜ “Analysis of energy- tional Conference on Intelligent Sensors, Sensor Networks and efficient, fair routing in wireless sensor networks through Information Processing (ISSNIP ’04), Melbourne, Australia, non-linear optimization,” in Proceedings of the 58th IEEE December 2004. Vehicular Technology Conference (VTC ’03), vol. 5, pp. 2844– [25]D.F.Macedo,L.H.A.Correia,A.L.dosSantos,A.A. 2848, Orlando, Fla, USA, October 2003. F. Loureiro, and J. M. S. Nogueira, “A pro-active routing [10] A. Trigoni, Y. Yao, A. Demers, J. Gehrke, and R. Rajaraman, protocol for continuous data dissemination in wireless sensor “Wave scheduling: energy-efficient data dissemination for networks,” in Proceedings of the 10th IEEE Symposium on sensor networks,” in Proceedings of the International Workshop Computers and Communications (ISCC ’05), pp. 361–366, on Data Management for Sensor Networks in Conjunction with Murcia, Spain, June 2005. VLDB (DMSN ’04), Toronto, Canada, August 2004. [26] L. Keong, L. Huan, and P. Yi, “An efficient and reliable routing [11] V. P. Mhatre, C. Rosenberg, D. Kofman, R. Mazumdar, and N. protocol for wireless sensor networks,” in Proceedings of the 6th Shroff, “A minimum cost heterogeneous sensor network with IEEE International Symposium on a World of Wireless Mobile EURASIP Journal on Wireless Communications and Networking 15

and Multimedia Networks (WoWMoM ’05), vol. 2, pp. 512– 516, Los Alamitos, Calif, USA, June, 2005. [27] F. Ye, G. Zhong, S. Lu, and L. Zhang, “GRAdient broadcast: a robust data delivery protocol for large scale sensor networks,” ACM Wireless Networks, vol. 11, no. 3, pp. 285–298, 2005. [28] G. Wang, T. Wang, W. Jia, M. Guo, H. H. Chen, and M. Guiazani, “Local update-based routing protocol in wireless sensor networks with mobile sinks,” in Proceedings of IEEE International Conference on Communications, pp. 3094–3099, Glasgow, UK, June 2007. [29]G.Mergen,Q.Zhao,andL.Tong,“Sensornetworkswith mobile access: energy and capacity considerations,” IEEE Transactions on Communications, vol. 54, no. 11, pp. 2033– 2044, 2006. [30] Z. Khalid, G. Ahmed, and N. M. Khan, “Impact of the mobile speed on the performance of wireless sensor networks,” in Proceedings of Cyber-Tech Conference, Karachi, Pakistan, July 2007. [31] M. Chen, T. Kwon, S. Mao, and V. C. M. Leung, “Spatial- temporal relation-based energy-eﬃcient reliable routing protocol in wireless sensor networks,” International Journal of Sensor Networks, vol. 5, no. 3, pp. 129–141, 2009. [32] J. E. Tateson and I. W. Marshall, “An adaptive routing mechanism for ad hoc wireless sensor networks,” in Proceedings of London Communication Symposium (LCS ’03), London, UK, September 2003. [33] Z. Khalid, G. Ahmed, and N. M. Khan, “A real-time energy- aware routing strategy for wireless sensor networks,” in Proceedings of Asia-Paciﬁc Conference on Communications (APCC ’07), pp. 381–384, Bangkok, Thailand, October 2007. Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 467315, 15 pages doi:10.1155/2009/467315

Research Article Achievable Throughput-Based MAC Layer Handoff in IEEE 802.11 Wireless Local Area Networks

SungHoon Seo,1 JooSeok Song,1 Haitao Wu,2 and Yongguang Zhang2

1 Department of Computer Science, Yonsei University, Seoul 120-749, South Korea 2 Wireless and Networking Group, Microsoft Research Asia, Beijing 100190, China

Correspondence should be addressed to SungHoon Seo, [email protected]

Received 27 March 2009; Accepted 10 June 2009

Recommended by Naveen Chilamkurti

We propose a MAC layer handoff mechanism for IEEE 802.11 Wireless Local Area Networks (WLAN) to give benefit to bandwidth- greedy applications at STAs. The proposed mechanism determines an optimal AP with the maximum achievable throughput rather than the best signal condition by estimating the AP’s bandwidth with a new on-the-fly measurement method, Transient Frame Capture (TFC), and predicting the actual throughput could be achieved at STAs. Since the TFC is employed based on the promiscuous mode of WLAN NIC, STAs can avoid the service degradation through the current associated AP. In addition, the proposed mechanism is a client-only solution which does not require any modification of network protocol on APs. To evaluate the performance of the proposed mechanism, we develop an analytic model to estimate reliable and accurate bandwidth of the AP and demonstrate through testbed measurement with various experimental study methods. We also validate the fairness of the proposed mechanism through simulation studies.

Copyright © 2009 SungHoon Seo et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction differ in accordance with AP configuration setting. Also, each AP can be configured with a different channel; thus As wireless networking grows in popularity, various radio adjacent APs with orthogonal frequencies (e.g., 1, 6, and access technologies have been developed to provide bet- 11 in 802.11b) are recommended to avoid interchannel ter environment for user data service. Most of all, IEEE interference which causes the disruption of signal quality and 802.11 Wireless Local Area Network (WLAN) is one of channel utilization [1]. the dominant wireless technologies to support high-speed Due to the nature of 802.11, an STA can associate with network access nowadays. The WLAN basically forms an only an AP at a time through a channel assigned on the infrastructure with two network components, Access Point AP; thus at the same time the STA cannot listen to any (AP) and Station (STA). An AP is generally distributed at a signal from APs operated on the other channels. In order to fixed location, and the WLAN infrastructure connects STAs listen to signals from other channel APs, STAs should switch to a wired network via the AP within their communication their channel, but it may cause the blocking of on-going range. AP’s signal range is denoted by Basic Service Set (BSS) communication through their current associated AP. Even if or hotspot which generally provides coverage within a few STAs can listen to beacon frames from other APs operated ten-meter radius. on the same channel, it is limited only when their listen In large scale wireless networks, multiple APs are densely period and the APs’ beacon interval are exactly matched. This deployed, and their hotspot ranges are overlapped in the is because the 802.11 STAs repeat to change their Network vicinity of one another (e.g., campus, building, and airport Interface Card (NIC) mode in sleeping and listening to lounge) with different types of physical (PHY) standard beacon frame for Power Saving. and channel frequency. Each PHY standard provides various When the signal condition from the current associated channel modulation rate (e.g., 1, 2, 5.5, 11 Mbps for 802.11b AP becomes poor to communicate, STAs should discover and 6, 12, 24 Mbps for 802.11a); thus the performance may other APs and continue the communication by performing 2 EURASIP Journal on Wireless Communications and Networking aMAClayerhandoff.Forthediscovery,STAsperformactive Section 5 presents the analytic model to estimate the achiev- scanning by broadcasting a special management frame, that able throughput. In Section 6, we show the evaluation of the is, Probe Request, to every channel supported by their NIC. proposed mechanism through experiment and simulation AnSTAtriggerstheactive scanning when the Received Signal studies, and Section 7 concludes this paper. Strength Index (RSSI) of the current associated AP is below the predefined threshold (usually about −90 dBm), and the STA builds the list of the AP available to itself. Then, the STA 2.RelatedWorkandMotivation performs handoff to an AP whose signal condition is better ff than the current associated AP, mainly based on the RSSI as The IEEE 802.11 MAC layer hando procedure is split into in [2, 3]. However, using the RSSI as a criterion to perform trigger, discovery, AP selection, and commitment (Through- ff handoff is not good enough because the RSSI itself does not out this paper, the MAC layer hando is alternatively used ff ff mean the AP’s capability information. for the term “layer 2 hando ” or “L2 hando ”). The most Therefore we propose a MAC layer handoff mechanism of previous researches [2–4] are based on the RSSI measured for IEEE 802.11 WLAN by using AP’s capability information from current associated AP as a criterion not only to trigger ff as a handoff criterion, especially an achievable throughput hando but also to select optimal AP. After an STA triggers ff from APs. To estimate the achievable throughput, we devise hando , it discovers neighbor APs and channels available to a new method, namely, Transient Frame Capture (TFC). The itself with active scanning to all channels supported by its TFC works with the promiscuous mode of WLAN NIC so WLAN NIC which causes the major portion of the entire ff that STAs can keep their connections through the current hando latency. Even if authors of [3, 5] proposed solutions ffi associated AP without service degradation. The proposed to reduce the latency, they have limitations of a di culty to ff handoff mechanism allows STAs to determine an optimal AP modify already deployed AP software and ine ective cost to whose bandwidth satisfies the requirement of applications equip additional scanning purpose NIC at the STA. The AP at the STAs, thus gives the most benefit to the STAs when selection procedure is also based on the RSSI so that STAs ff performing the handoff. We develop an analytic model and perform hando to an AP with the maximum RSSI. Wu et demonstrate through testbed measurements with various al. [4] proposed an RSSI-based AP selection mechanism to ff experimental study methods to show the effects on reliability reduce the hando latency and to avoid service degradation ffi and accuracy of the throughput estimation. Furthermore, of VoIP tra c. However, RSSI itself does not indicate the we perform simulation studies to validate the proposed AP’s capability (e.g., achievable bandwidth); thus the STA ff mechanism in regard to the fairness of APs. Especially, this may su er the severe degradation of on-going service after ff paper contributes in the following four aspects. performing the hando to a highly loaded AP. Bandwidth estimation has been a hot research topic (1) We provide a client-only solution for the achievable and mainly addressed by using packet dispersion [6]. The throughput-based handoff mechanism so that it packet dispersion was originally designed to estimate end- does not require any modifications or changes on to-end bandwidth on wired network environment where AP’s protocol and configuration. That is, it works cross traffic exists along with the intermediate nodes in the with any existing setup of already deployed WLAN routing path. However, the packet dispersion over- or under- infrastructure. estimates the bandwidth on the wireless network environ- (2) We devise a new method to estimate the actual ment; thus a few research [7–12] has been investigated to bandwidth capacity as well as the achievable through- estimate accurate bandwidth for the wireless environment. put from neighbor APs without service degradation References [7, 8] provided solutions to estimate the sat- through the current associated AP. urated and the potential bandwidth on AP by analyzing the distribution of packet delay and beacon frames. In (3) From a view point of AP deployment, the trafficload [9], Li et al. attempted to use the packet dispersion in on multiple APs should be fairly distributed. The the 802.11 WLAN by analyzing the channel access time. proposed handoff mechanism enables STAs to select Also, as a passive manner, [10–12] presented solutions to the most bandwidth-beneficial AP. This also gives an estimate bandwidth on AP by analyzing channel occupation advantage of balancing the load on the different types probability. However, these methods mainly focused on the of APs. bandwidth measurement itself by actively sending probes to (4) Our implementation and experimental studies are the AP or passively receiving beacons from the AP (one-way the first attempt to address AP’s throughput measure- measurement); thus they are not applicable methods as a ment only from the STA side. Also, the measurement client-only solution which limits the protocol changes at APs. estimates near the boundary of the actual throughput Most recently, Kandula et al. [13] proposed a client- in the 802.11 environments. only solution to maximize user throughput based on the available bandwidth measurement by switching channel The rest of this paper is organized as follows. Section 2 between multiple APs. To increase the user throughput, introduces background on MAC layer handoff and band- the solution virtually maintains multiple IP flows mapped width estimation. In Section 3, we describe the proposed with WLAN NIC’s duplicated MAC addresses. However, it handoff mechanism which is the basis of achievable through- cannot maintain a single flow (e.g., UDP-based application) put. Section 4 provides details of TFC algorithm, and separately through multiple APs because the throughput gain EURASIP Journal on Wireless Communications and Networking 3 depends on the number of flows. Moreover, STAs should conducted under the promiscuous mode operation of NIC; it always maintain connections and monitor actual packets thus affects no interference to other contending STAs within through multiple APs to measure available bandwidth. It the same channel BSS. With the result of the TFC, STAs means that the solution may degrade the entire channel uti- can perform handoff to an optimal AP which guarantees the lization since STAs should be fully connected to the multiple maximum achievable throughput to the STAs. APs whether they are used for communication or not. 3. The Proposed MAC 2.1. Problem Statement—The Motivation. As mentioned Layer Handoff Mechanism earlier, most of L2 handoff mechanisms addressed RSSI as ahandoff criterion but the RSSI itself does not indicate the In this section we describe the details of the proposed MAC ff actual capability of APs. If an STA has the knowledge of AP’s layer hando mechanism which addresses the achievable ff capability information (i.e., achievable throughput after the throughput as a hando criterion rather than the RSSI. A STA handoff to the AP), it can help the STA to determine newly devised TFC method enables STAs to estimate the a better AP which provides higher throughput to the STA. bandwidth capacity and to predict the achievable throughput Even if IEEE 802.11e [14] provides a capability information, of neighbor APs. Since no guarantee STAs will be able the number of STA associated with the AP,this information is to achieve similar performance due to asymmetric fading, ff not enough to estimate the AP’s current bandwidth occupied we further investigate how wireless condition a ects the by active STAs. New radio resource measurements for WLAN predicted achievable throughput according to the link quality are defined in IEEE 802.11k [15], and how meaningful data such as RSSI and Frame Error Rate (FER). ff can be collected through the measurements is discussed in We summarize the procedure for our hando mechanism [16]. The 802.11k enables STAs to request measurements as follows. (e.g., channel occupation rate) from other STAs (or APs), but (1) The proposed handoff is triggered. it requires the protocol modification of both STAs and APs. Furthermore, measurement frames either on the operating (2) Build a BSS list for neighbor APs available to the STA. or nonoperating channel affect the on-going trafficthus We assume that this step can be actively performed by they may increase the signaling overhead which causes the channel scanning as in [4]. interruption of data services. (3) Capture 802.11 frames on the BSS of neighbor APs Figure 1 illustrates a scenario that an STA moves across (appeared in the BSS list) by utilizing Transient the overlapped hotspots, BSS1 and BSS2, with two APs, Frame Capture. ff where each hotspot is configured with a di erent channel (4) Estimate the achievable throughput from each of the number (1 and 149). In the BSS1, the STA has associated with APs by analyzing the captured frame information. current AP (cAP) which supports 802.11b. The STA’s RSSI − (5) Select an optimal AP with the maximum achievable from the cAP is very high ( 45 dBm), but the bandwidth ff loaded on the cAP is relatively higher because other n STAs throughput and perform hando to the AP. are activated through the cAP in the BSS1 (e.g., 4 STAs, from STA 1 to STA 4, each of these individually occupies 3.1. AP Selection Algorithm. STAs should select a target about 1 Mbps bandwidth on the cAP). On the other hand, in AP before they perform a handoff. We use an achievable the BSS2, a neighbor AP (nAP) supports 802.11a. Relatively throughput as a metric to determine an optimal target AP lower RSSI of the nAP is acceptable for the STA to associate, among neighbor APs. The AP selection for the proposed but the traffic load on the nAP is lower than that on the cAP handoff mechanism is conducted with an algorithm as (<1 Mbps). If the STA associates with the nAP even in lower follows. Once an STA finds neighbor APs with a scan method RSSI, it is beneficial for the STA to achieve higher bandwidth asintroducedin[4], it builds a BSS list for every neighbor through the nAP. AP. Let U denoteasetofeveryneighborAPintheBSS ={ } In this sense, using the RSSI as a handoff criterion list, and it is given by U AP1,AP2, ...,APN where the ∈ ≤ ≤ in the conventional MAC layer handoff mechanism is not STA finds N APs, thereby APi U(1 i N). Then the good enough to give more benefit to bandwidth-greedy STA performs the TFC and collects information about the applications (such as FTP, P2P file sharing, and e-mail) achievable throughput (ai) and RSSI (si) on every AP in the which require bandwidth as high as possible. We therefore set U. The APi is assumed to have the maximum achievable take the achievable throughput from APs available to STAs throughput which is determined by into account the main criterion of the proposed handoff arg max ai ∈ i∀ j : aj ≤ ai , (1) mechanism. By utilizing newly devised method, Transient i Frame Capture (TFC), STAs not only estimate bandwidth capacity but also predict the achievable throughput from the and then the STA performs handoff to the APi. target AP (as denoted by nAP in Figure 1). Since the TFC is When there exists more than one AP with the same performed in a very short time with fast channel switching, maximum achievable throughput using (1), the AP selection STAs do not suffer from the service degradation through the algorithm employe the RSSI as another metric. Let U denote cAP even occurring retransmissions caused by frame loss and a set of APs, U ={AP1,AP2, ...,APM},whereM APs are delayed ACK transmission. Moreover, the TFC is passively determined with the same achievable throughput, thereby 4 EURASIP Journal on Wireless Communications and Networking

802.11b with CH# 1 loaded BW>4 Mbps RSSI to STA = - 45 dBm 802.11a with CH# 149 BSS1 loaded BW<1 Mbps RSSI to STA = - 60 dBm

STA 1 cAP

STA 2 Overlapped nAP

hotspot area BSS2 .

STA n STA a STA b Movement STA

Figure 1: A scenario for MAC layer handoﬀ within overlapped hotspot area.

APi ∈ U(1 ≤ i ≤ M ≤ N). As similar to (1), an optimal AP Associated with cAP on channel number X One TFC procedure (for nAP 1) is determined by arg maxi si, and the STA ﬁnally performs to the APi which has the maximum ai as well as si. Build BSS list for nAP by active scanning 1. Switch channel to X (nAP1: X, nAP2: X ) 4. Transient Frame Capture 2. Change NIC to promiscuous mode TFC start We mentioned that the proposed handoﬀ mechanism utilizes 3. Capture all frames on channel X Capture the Transient Frame Capture (TFC) not only to estimate the period 4. Frame filtering (w/nAP’s BSSID) bandwidth capacity of neighbor APs but also to predict the TFC end achievable throughput from the neighbor APs. Utilizing the 5. Change back to STA mode TFC has several advantages as follows. (1) To the best of our TFC (nAP 2)

knowledge, there exists no approach to passively measure 6. Switch back to original channel X . the AP’s bandwidth capacity and achievable throughput TFC without any AP protocol change, and thus it can be (nAP 1) easily applied to the any existing 802.11 NIC. (2) The . TFC works with switching the NIC’s operation status to a STA performs handoff to a nAP with maximum achievable throughput promiscuous mode during very short period, and thus it does not affect the current data service in use (The most Figure 2: Transient frame capture. of commercial IEEE 802.11 WLAN NIC supports to use the promiscuous operation by both kernel and user level API). (3) Measured information by utilizing the TFC can be used set of frames whose sender or receiver address field in MAC for estimating the achievable throughput from the neighbor header matches to the nAP’s BSS Identification (BSSID). APs and properly reflects wireless network environment As soon as a CP expires, the TFC ends with changing the which dynamically varies according to the link condition. STA’s mode back to the original (infrastructure mode) and Figure 2 shows an example when an STA periodically switching the channel back to the original for the cAP (X → performs the TFC to nAPs belonging to the BSS list which X). Since the TFC is conducted by fast channel switching is collected by active scanning; for example, nAP1 and within operating and nonoperating channels, STAs in range nAP2 work on channel number X and X,respectively. of several neighbor APs can obtain individual information Each TFC procedure continues a certain time duration, of the APs even in a different channel. For neighbor APs Capture Period (CP). To minimize the service degradation in a same channel, STAs can collect the information by of activated connection through cAP, the length of the CP performing one TFC to the channel. should be as short as possible, but it affects the reliability By utilizing the TFC, STAs can obtain several infor- of throughput estimation. The impact of the CP will be mation, such as (sub)type, length, and Traffic Indication discussed in Section 6.1. Map (TIM) fields from the MAC header of the captured The detail procedure of a TFC is described as follows. frames. These pieces of information play an important role Once an STA starts a TFC, it switches the channel of its NIC to infer the number of active STA which currently receives or to the target channel of the nAP (X → X)andchanges transmits frames via the nAP, not the number of associated to the promiscuous mode to capture frames on the target STA as in [14]. The number of active STA involved in channel. During a CP, the STA captures all WLAN frames receiving downlink frame from AP can be easily inferred by and builds the nAP specific information based on a filtered counting the receiver address field in downlink data frames. EURASIP Journal on Wireless Communications and Networking 5

Table 1: Parameter values for the analysis of throughput estima- as the CP timer expires, the capture module restores to the tion. original context information and ﬁnishes the TFC procedure. Parameter Values Descriptions ρ 20 μsec A slot time 5. Achievable Throughput Estimation SIFS 10 μsec Short Interframe Space This section provides an analytic model to estimate the DIFS 50 μsec Distributed IFS current bandwidth loaded on a target AP and to predict ACK TIMEOUT 300 μsec ACK timeout the achievable throughput which is expected after the STA LPHY 128 bits PHY header length associates with the AP. In addition, we also investigate LMAC 272 bits MAC header length the achievable throughput taking into account the rate LACK 112 bits + LPHY ACK frame length discounted according to the wireless link condition, that is, L variable Data length RSSI and FER. Symbols for the analysis are explained in Table 1, and they will be used throughout this paper.

However, a certain STA is activated but currently staying in 5.1. Bandwidth Capacity Estimation. Let n denote the num- power saving mode. We thus additionally address the TIM ber of active STA which is contending in an AP’s BSS, and field in Beacon frames as to infer the number of receiving τ is the probability that an STA transmits in a given time STA. Since the TIM includes a set of association ID of the STA slot. For a certain time slot, Pi, Ps,andPc are the probability whose downlink trafficisnowbuffered at the AP, counting 1 that the channel is idle, the transmission is successful set bit denotes the number of active STA in receiving. because only one STA tries transmission, and the collision is On the other hand, inferring the number of active STA occurred when more than two STAs simultaneously transmit, involved in transmitting uplink frame to AP differs from respectively, which are given by that in receiving downlink frame because the STA cannot n P = (1 − τ) , capture every frame on the target channel (X )becauseof i − following reasons. The first reason is that APs and STAs may P = n × τ(1 − τ)n 1, (2) drop frames if their internal buffer overflows. Fortunately, it s is ignorable since we only focus our throughput estimation Pc = 1 − Pi − Ps. on the transmission rate of frames actually leaved from the APs or STAs. The other reason is that an STA is not in Let LPAYLOAD and LUPPER denote the length of a frame the propagation range of other STAs as known as hidden (payload) and upper layer protocol headers (i.e., IP and = − terminal. As an example, in Figure 1, the propagation range UDP), where L LPAYLOAD LUPPER. The average time of nAP and STA a is reachable to the STA but that of STA associated with one successful transmission, Ts,andwith b is not. It means that, by utilizing the TFC, the STA can collision, Tc,aregivenby capture only frames propagated from the nAP and the STA T = T + T + T + T + SIFS + DIFS, a, whereas it is impossible to capture any frame transmitted s PAYLOAD PHY MAC ACK from the STA b. Therefore, we use the receiver address field Tc = TPAYLOAD + TPHY + TMAC +ACK TIMEOUT + DIFS, in ACK frames to infer the number of active STA involved in (3) transmitting uplink frame to the AP. where TPAYLOAD, TACK, TPHY,andTMAC are the average time associated with the transmission of a payload, an ACK frame, 4.1. Implementation Issues. We implement a real-system a PHY header, and a MAC header, respectively. These can be testbed and demonstrate the TFC to estimate the bandwidth easily obtained by dividing L , L , L ,andL capacity and the achievable throughput from APs. The PAYLOAD ACK PHY MAC into the channel rate (CR) of the AP, respectively. key part of the testbed implementation is the basis of the Based on (2)and(3), channel idle ratio (R )and kernel level miniport driver for NIC in Realtek-8185 chipset idle channel busy ratio (R ) can be expressed by under Microsoft Windows Vista’s Network Driver Interface busy Specification (NDIS) architecture. × = Pi ρ Figure 3(a) shows the overall architecture of the testbed Ridle , Pi × ρ + Ps × Ts + Pc × Tc where TFC functionalities are implemented as a capture (4) module in the miniport driver. By calling the special function Rbusy = 1 − Ridle. (DeviceIo-Control) from user application, the capture mod- × × ule starts the TFC procedure. While the TFC is performed, On substituting L Ps for Pi ρ in (4) we obtain the target every frame captured on the specific channel is stored in AP’s bandwidth, B, which is given by ff Net Bu er List (NBL), and then the user application refers L × P B = s . (5) the captured frame by reading the address of the NBL as P × ρ + P × T + P × T in Figure 3(b). Whenever the capture module performs a i s s c c TFC procedure, it starts a timer for the Capture Period (CP By assuming that all data length is equal to L, Rbusy can be timer) and saves the current context information such as the derived from a function of n and τ. With the number of channel number and the operation mode of the NIC. As soon DATA frame (NDATA) andACK frame (NACK)measuredby 6 EURASIP Journal on Wireless Communications and Networking

Capture module in User application miniportdriver

User application DeviceIoControl MPDeviceIoControl (CH#, CP) User level Save context information Start Kernel level CP timer Start_Capture ReadFile MPCaptureNBL Empty face

r Capture module Circular Read frame from queue Received Miniportdriver buffer address NDIS inte frame

MPCaptureTimerCallback

CP timer 802.11 NIC Restore context information expiry

(a) (b)

Figure 3: (a) Overall architecture of testbed implementation. (b) Work flows between user application and the capture module in the miniport driver. the TFC, we can obtain the channel time associated with one to n + 1 when a new STA performs handoff and continues its successful transmission, Ts, for downlink and uplink traffic. transmission through the nAP. Thus we can expect the per- Thus, for a CP, the busy ratio is given by Rbusy = ((NACK + STA bandwidth in the nAP with n +1STAsasBn+1 = B(n + NDATA) · Ts +(NDATA − NACK) · Tc)/CP where NDATA − NACK 1)/(n+1). By assuming that every STA transmits (or receives) denotes the number of unacked data which is retransmitted its individual traffic in same data rate within the nAP’s range, n during the channel collision, Tc. In addition, n is also inferred S(Rbusy) is the busy ratio for the maximum peak of Bn,where n = by the TFC as mentioned in Section 4. Based on the obtained (d/dRbusy)Bn 0. It means that the throughput of the nAP n = n Rbusy and n,wecandefineτ as a nonlinear algebraic equation. with n-STA is saturated when Rbusy S(Rbusy). Finally, an Generally, the nonlinear algebraic equation can be exactly STA’s achievable throughput, A, from an nAP (i.e., the nAP solvable through numerical method (e.g., Newton-Raphson with n +1STAs,butactuallyn STAs are associated with the method). Therefore, the AP’s bandwidth (B)canbemade nAP) is given by perfectly obtainable by using the (5). Figures 4(a)–4(f) are plots of B by using (5)asafunction ⎧ ⎪ ≤ = ⎪(Bn+1, Bn] Bn Bn+1, of Rbusy in [0, 1] for the L 500 and 1000 Bytes. Each ⎨⎪ analysis is computed by MATLAB programming when n is A = 0, B B > B ∧ Rn ≤ S Rn+1 , ⎪ n+1 n n+1 busy busy 1, 5, and 10, and CR is 1, 2, 5.5, and 11 Mbps. These results ⎪ ⎩ ∧ n n+1 show that, for n = 1, the B has been increasing steadily as [0, Bn+1) Bn > Bn+1 Rbusy >S Rbusy , the Rbusy increases, regardless of the L and the CR. On the (6) other hand, for n = 5 and 10, the B has shown a linear increase until it reaches a local maximum, which denotes a saturated throughput, and decreases considerably as the R where Bn is the maximum per-STA bandwidth from the busy n = n increases. nAP with n-STA when Rbusy S(Rbusy). According to Rbusy, the A has different ranges as follows. For Bn ≤ Bn+1, the Rbusy increases when the n becomes n + 1 since individual 5.2. Achievable Throughput Prediction. As we have seen bandwidth occupied by each STA is same as Bn+1. On the in Section 5.1, STAs can estimate the bandwidth capacity n+1 other hand, for Bn > Bn+1, it is hard to estimate Rbusy by currently loaded on the target AP by utilizing the TFC. using the TFC. Thus we choose zero as the lower bound of However, the AP’s bandwidth capacity does not indicate n n+1 the A. When Rbusy >S(Rbusy), the A may be less than Bn+1 the throughput which is achievable after the STAs perform because the achievable throughput decreases as the busy ratio handoff to and associate with the AP. Therefore, we present increases, while the A may be less than or equal to Bn+1 for how to predict the achievable throughput of APs based on n ≤ n+1 Rbusy S(Rbusy). Figure 5 depicts the analysis result for the the TFC we are addressing. = n achievable throughput prediction when n 3. By using (4)and(5), we newly define B(n)andRbusy as the current bandwidth loaded on an nAP and its busy ratio, respectively, when the nAP has n active STAs. Then per-STA 5.3. Rate Discount of Achievable Throughput. As an STA bandwidth in the nAP is given by Bn = B(n)/n as a function moves away from an AP, the signal from the AP reaches n of Rbusy. Suppose that the number of active STA may increase the STA with reduced power so that the lower RSSI is EURASIP Journal on Wireless Communications and Networking 7

101 101 101

100 100 100

− − 10−1 10 1 10 1 Bandwidth (Mbps) Bandwidth (Mbps) Bandwidth (Mbps)

− − 10−2 10 2 10 2 00.5100.5100.51

Rbusy Rbusy Rbusy (a) L = 500 B and n = 1 (b) L = 500 B and n = 5 (c) L = 500 B and n = 10 101 101 101

100 100 100

10−1 10−1 10−1 Bandwidth (Mbps) Bandwidth (Mbps) Bandwidth (Mbps)

10−2 10−2 10−2 00.5100.5100.51

Rbusy Rbusy Rbusy

CR = 1Mbps CR = 1Mbps CR = 1Mbps CR = 2Mbps CR = 2Mbps CR = 2Mbps CR = 5.5 Mbps CR = 5.5 Mbps CR = 5.5 Mbps CR = 11 Mbps CR = 11 Mbps CR = 11 Mbps (d) L = 1000 B and n = 1 (e) L = 1000 B and n = 5 (f) L = 1000 B and n = 10

Figure 4: Numerical analysis results of bandwidth estimation.

measured at the STA. Even if an AP transmits a certain rate frame retransmission ratio (ReTX) for CR in 1, 5.5, and of data frames to an STA, the STA is likely to miss several 11 Mbps and average RSSI with respect to the distance frames because of frame loss or bit error occurrence in a between an STA and an 802.11b AP, from 10 m to 70 m at poor wireless link condition. Typically, the lower RSSI is intervals of 7 meters. We generate 100 Kbps downlink traffic measured, and the STA suffers from the higher Bit Error Rate with 500 B length UDP datagram. The ReTX is calculated (BER), causing the degradation of the achievable throughput as # of retransmitted frame/# of received frame where the obtained from the AP. Therefore, the achievable throughput retransmitted frame is distinguished by Retry bit in 802.11 should be discounted according to the BER, and we call header. The result shows that the frame retransmission it rate discount. However, to the best of our knowledge, rarely occurs until 60 m (CR = 1), 50 m (CR = 5.5), there exists no method to obtain the BER directly from the and 40 m (CR = 11). After that, the frame retransmission 802.11 NIC [17]. We thus present three alternative methods ratio significantly increases, while the average RSSI gradually to obtain the discounted rate without the basis of the BER decreases as the distance increases. It means that we cannot measurement. determine the RSSI where the retransmission begins to increase regardless of the AP’s channel rate. Even if the 5.3.1. Frame Retransmission versus RSSI. In 802.11, data number of retransmitted frame is a good decision criterion frame loss or error initiates retransmission of the frame for WLAN handoff [18], it is not applicable to obtain the to provide reliable communications. As RSSI between STA discounted rate in our handoff mechanism since the STA and AP decreases, the number of frame retransmission cannot measure the number of frame retransmission without may increase. Figure 6 shows the experiment result of associating with the AP. 8 EURASIP Journal on Wireless Communications and Networking

10 −45 1.6 Bn ≤ Bn+1 Bn > Bn+1 8 −55 n ≤ n+1 1.4 Rbusy S(Rbusy) 6 −65 n n+1 4 −75 1.2 Rbusy >S(Rbusy) 2 −85 −

Throughput (Kbps) 0 95

1 Average RSSI (dBm) 0 1020304050607080 0.8 Distance (m) (Bn+1, Bn] ) +1 n

0.6 (D(Bn+1), Bn] B 1M 6M [0, Bandwidth (Mbps) 2M 12 M

0.4 ))

+1 5.5 M 24 M n [0, B ] B n+1 ( 11 M

0.2 D [0, D(Bn+1)] [0, (a) R = 10 K 0 00.10.20.30.40.50.60.70.80.91 1000 −45 Channel busy ratio ( ) Rbusy 800 −55 n+1 600 −65 Bn S(Rbusy) 400 − Bn+1 Bn+1 75 − D(Bn+1) D(Bn+1) 200 85 − Throughput (Kbps) 0 95 Figure 5: Numerical analysis of achievable throughput estimation 0 1020304050607080 Average RSSI (dBm) /w and /wo rate discount for n = 3 and FER = 0.1. Distance (m)

RSSI (11b) 1 RSSI (11a) −40 0.8 (b) R = 1000 K −50 −60 0.6 Figure 7: Throughput and RSSI versus distance.

−70 0.4 −80 0.2 ﬃ Average RSSI (dBm) generates tra c destined to the STA with a ﬁxed rate (R)in −90

Frame retransmission ratio 10 and 1000 Kbps. We use 1 KB length UDP datagram for the 0 10 20 30 40 50 60 70 traffic generation. Distance between STA and AP (m) The result shows that, for all R, the STA achieves less throughput as the distance increases. Furthermore, as R Average RSSI ReTX (CR = 5.5 M) increases, the discounted rate is also increases regardless of = = ReTX (CR 1M) ReTX (CR 11 M) CR. Remarkably, we can observe that the location where Figure 6: RSSI and frame retransmission ratio. the throughput is dramatically decreased is similar as 68, 60, 44, and 28 m for CR = 1, 2, 5.5, and 11 Mbps (802.11b), and 70 m for CR = 6, 12, and 24 Mbps (802.11a). 5.3.2. Throughput versus RSSI. When an AP transmits data From these results, we believe that the discounted rate strongly depends on the RSSI and CR. Therefore, when the frames to an STA at a constant rate, the receiving rate at the ff STA should be also constant. However, the receiving rate is predicted achievable throughput of di erent APs is same, the determined by FER (regard it as related to BER); it thus varies comparison of the APs’ RSSI is a useful metric to determine according to signal conditions. BER is determined by Signal abetterAP. to Interference and Noise Ratio (SINR) where the signal is denoted by RSSI, but the noise cannot be obtained from the 5.3.3. FER Measurement with Probe Frame. Usually the received signal. Since we are not intended to calculate exact number of errors in a sequence of bits is modeled by rate value, the RSSI is still useful to deduce the discounted a binomial distribution; thus FER can be expressed as LDATA +LACK rate. FER = 1 − (1 − BER) where LDATA is a DATA Figure 7 illustrates the experiment result of throughput frame length [19]. Noting that the STA cannot send DATA and RSSI degradation as the distance between an STA and an frame to the not-yet-associated AP, we measure the FER AP increases where the AP is located at the start of an 80 m by sending/receiving Probe Request/Response management corridor whose width and height are 2 and 3 m, respectively. frames instead of DATA/ACK frames. Since 802.11’s We plot the STA’s throughput and RSSI for CR = 1, 2, 5.5, contention mechanism for both management and DATA and 11 Mbps for 802.11b on channel 13 and CR = 6, 12, and frames is same before being sent, the FER measurement 24 Mbps for 802.11a on channel 44 as the STA moves away with probe frames is acceptable. Let LP denote the length from the AP and toward the end of the corridor at intervals of a pair of Probe Request and Response frame. (The of 2 meters until it reaches 80 m. During each experiment, a IEEE 802.11 standard specifies that the Probe Request PC is directly connected to the AP in an Ethernet link and frame is broadcasted, but for the FER measurement, we EURASIP Journal on Wireless Communications and Networking 9

10.1.2.10 10.1.1.10 10.1.1.20 10.1.1.30 10.1.1.40 10.1.1.50 n0 n1 n2 n3 n4 n5 Windows PCs (XP) Gigabit Ethernet

AP 802.11b channel #1 AP 802.11b nAP cAP channel #11 10.1.1.100 10.1.2.100 Wireless 10.1.1.1 10.1.1.2 network s1 s2 monitor (NetMon) Windows laptops (XP) 802.11b NIC STA 10.1.2.1 Windows laptop (Vista) 802.11b NIC s3 s4 s5 10.1.1.3 10.1.1.4 10.1.1.5

Figure 8: Experiment environment for throughput measurement with TFC. used a unicast address as the destination address field a neighbor 802.11b AP (nAP) on channel number 11 which is of the Probe Request frame.) Then the probability of orthogonal to that of the cAP. The nAP is a target to measure successful transmission for a pair of Probe Request and the achievable throughput by utilizing the TFC while the STA Response frames without error is given by (1 − BER)LP , is connected via the cAP.The cAP and the nAP is deployed by and it can be easily obtained by regarding the FER as using Belkin wireless b/g router and D-Link DWL-8200AP, 1−(# of received Probe Response/#ofsentProbeRequest). respectively. The only modification is applied at the STA by In addition, transmission may fail due to collision when installing implemented miniport driver. the channel is congested. The probability of collisions In order to generate the cross traffic on the APs, we occurred by other active STA can be expressed by use Windows XP powered 6 PCs labeled from n0ton5 n−1 n−1 (1 − Ps − Pi) = (Pc) as introduced in [20]to and 5 laptops labeled from s1tos5asinFigure 8. While increase the bandwidth accuracy. Hence, the rate discounted the n0 is connected directly to the cAP and generates the per-STA bandwidth achievable from the nAP with n-STA, traffic destined to the STA, other PCs (n1 ∼ n5) are directly ffi D(Bn), is given by connected to the nAP and generate the tra c destined to the corresponding laptops (s1 ∼ s5). Additionally, we locate a PC n−1 LP D(Bn) = Bn × (Pc) × (1 − BER) . (7) with a tool provided by [21], namely, NetMon, on near by the nAP.The NetMon is to capture every frame transmitted from As an example, in Figure 5, we plot the range of A with the nAP, thus works independently of others. To simplify, we rate discount by applying (7)forFER = 0.1 (black-solid assume that every PC generates their traffic with fixed-length error bar) when n = 3. Obviously, the range of A with rate UDP datagram, and the direction of the traffic is downlink. discount differs from that of A without rate discount (gray- For the experiment of the traffic in uplink direction, we could obtain similar results as the downlink traffic experiment. dashed error bar). The lower bound for Bn ≤ Bn+1 and ∧ n n+1 the upper bound for (Bn > Bn+1) (Rbusy >S(Rbusy)) are diminished in D(B ) since the throughput is affected by n+1 6.1. Impact of Capture Period (CP). In regards to the BER. On the other hand, for (B > B )∧(Rn ≤ S(Rn+1 )), n n+1 busy busy throughput measurement, finding an optimal CP plays an the upper bound is reduced to D(Bn+1). important role to make the TFC procedure do not disrupt the active session via the associated cAP. We thus do an 6. Experimental Studies experiment to find the optimal CP which minimizes the data loss of the current active session. The n0 sends the This section provides the experiment of the proposed MAC traffic of 1000-Byte length UDP datagram generated with layer handoff mechanism and the TFC. Figure 8 shows our 20 milliseconds interval (= 400 Kbps), and we check the experiment environment as follows. An STA works with a sequence number of each datagram. (We implement a new Windows Vista powered laptop equipping Netgear JWAG511 traffic generation application that the sequence number is WLAN NIC and is associated with an 802.11b AP (cAP) appeared in the data part of each UDP datagram.) As a result, on channel number 1. On the other hand, there exists we observe that no data loss is examined when CP ≤ 200 10 EURASIP Journal on Wireless Communications and Networking

5000

4000

3000

2000

Bandwidth (Kbps) 1000

0 500 B 1000 B 500 B 1000 B 500 B 1000 B 500 B 1000 B Channel rate = 1M Channel rate = 2M Channel rate = 5.5 M Channel rate = 11 M (a) CP = 200

5000

4000

3000

2000

Bandwidth (Kbps) 1000

0 500 B 1000 B 500 B 1000 B 500 B 1000 B 500 B 1000 B Channel rate = 1M Channel rate = 2M Channel rate = 5.5 M Channel rate = 11 M

R (50 K) TFC R (500 K) Avg-TFC R (2500 K) TXnAP R (5000 K) (b) CP = 300

Figure 9: Case 1: comparison between estimated bandwidth with the TFC and AP’s actual transmission rate (TXnAP)forn = 5. milliseconds, while for CP = 300 milliseconds, the result cAP buffers data destined to the STA and informs it via TIM averaged over 10 experiments shows that 1.8 datagrams are at beacon frame by next listen interval. As soon as the STA lost during a TFC procedure. However, if the CP ≥ 400 switches back to the original channel on the cAP, it sends PS- milliseconds, the number of datagram loss is significantly POLL frame to the cAP and then receives the buffered data increased in average 3.4 and 5.7 for CP = 400 and 500 from the cAP. milliseconds, respectively. We confirmed that the datagram is lost since the STA 6.2. Evaluation. We evaluate the performance of the TFC cannot receive frames sent from the cAP while the STA is on (1) reliable and (2) accurate estimation of AP’s band- in the promiscuous mode for the TFC procedure. When the width capacity by studying experiments in various traffic cAP does not receive ACK for a sent frame, it sends the frame environments. Also, we show that the prediction of the again until exceeding the retransmission limit in RetryLimit achievable throughput, which is the basis of the estimated where the RetryLimit is usually set by 7, but it is dependent bandwidth capacity, well matches the actual throughput to the NIC manufacturer. After the number of retransmission from the AP even applying (3) rate discount based on the exceeds the RetryLimit, the cAP drops the frame and tries to FER measurement. send the other frame in its buffer. In the rest of experiments, Each of these evaluation cases are performed under we thus use two CPs of 200 and 300 milliseconds to improve individual experiment scenario. During each experiment the reliability of data transmissions via the cAP during the scenario, we apply different n’s; thus, according to the n, n TFC proceeds. PCs send UDP datagram in L = 500 and 1000 B destined It is worth noting that the selection of CP duration is to the corresponding n laptops with the rate in 10, 100, 500, a huge problem since the heuristic value of the CP may and 1000 Kbps to generate cross traffic on the nAP. Also, we not fit other network setups. We thus address a method to vary the nAP’s CR in 1, 2, 5.5, and 11 Mbps and the CP avoid the service degradation of data connection through for the TFC in 200 and 300 milliseconds for various traffic the associated cAP. Whenever an STA performs a TFC to the environments. other channel for nAPs, it employs power saving technique as follows.: Before the STA switches its channel to a target AP’s 6.2.1. Case 1—Reliable Bandwidth Estimation. Figures 9(a) channel, it sends a null frame to the cAP, which is to enter and 9(b) are plots of the estimated bandwidth loaded on the into the power saving mode. During a CP for the TFC, the nAP (TFC) as a function of cross traffic when five other STAs EURASIP Journal on Wireless Communications and Networking 11

3 3

2.5 2.5

2 2

1.5 1.5 Bandwidth (Mbps) 1 Bandwidth (Mbps) 1

0.5 0.5

0 0 10 100 500 1000 10 100 500 1000 Traﬃc generation rate (R)(Kbps) Traﬃc generation rate (R)(Kbps) (a) CP = 200, L = 500 (b) CP = 200, L = 1000

3 3

2.5 2.5

2 2

1.5 1.5

Bandwidth (Mbps) 1 Bandwidth (Mbps) 1

0.5 0.5

0 0 10 100 500 1000 10 100 500 1000 Traﬃc generation rate (R)(Kbps) Traﬃc generation rate (R)(Kbps)

R (30 K) CR = 1M R (30 K) CR = 1M R (300 K) CR = 2M R (300 K) CR = 2M R (1500 K) CR = 5.5 M R (1500 K) CR = 5.5 M = R (3000 K) CR 11 M R (3000 K) CR = 11 M (c) CP = 300, L = 500 (d) CP = 300, L = 1000

Figure 10: Case 2: estimated bandwidth in average with the TFC for n = 3.

receive downlink traffic from the nAP (n = 5) and CR = 1, As a result, we can obtain that the reference bandwidth 2, 5.5, and 11 Mbps for CP = 200 and 300 milliseconds, differs from the TXnAP. When CR is lower than the reference respectively. We generate 10, 100, 500, and 1000 Kbps of cross bandwidth (i.e., R(2500 K) and R(5000 K) for CR ≤ 2 Mbps), trafficfromeachoffivePCs(n1 ∼ n5). For the nAP, this the actual bandwidth on the nAP is lower than the reference cross traffic is denoted as reference bandwidth in R(50 K), bandwidth since the nAP cannot transmit all trafficflowed R(500 K), R(2500 K), and R(5000 K), respectively. For each from PCs (n1 ∼ n5) to laptops (s1 ∼ s5) with the traffic scenario, the nAP’s bandwidth is estimated by utilizing configured channel rate. It means that comparing the the TFC at the STA. The estimated bandwidth averaged over estimated bandwidth with the TXnAP is more reliable. Most 5 TFC trials (Avg-TFC) with standard error is compared of the result indicates that the estimated bandwidth with the with the reference bandwidth. We also compare the esti- TFC is well matched to the TXnAP rather than the reference mated bandwidth (TFC) with the nAP’s actual transmission bandwidth. Moreover, the standard error of 5 TFC trials is rate (TXnAP) obtained by an independent NetMon (see distributed within the reliable range of the actual bandwidth Figure 8). loaded on the nAP. 12 EURASIP Journal on Wireless Communications and Networking

0.9 0.9 0.9

0.8 0.8 0.8

0.7 0.7 0.7

0.6 0.6 0.6

0.5 0.5 0.5

0.4 0.4 0.4

0.3 0.3 0.3 Achievable throughput (Mbps) Achievable throughput (Mbps) Achievable throughput (Mbps) 0.2 0.2 0.2

0.1 0.1 0.1

0 0 0 0.25 0.50.75 0.98 0.25 0.50.75 0.98 0.25 0.50.75 0.98

Channel busy ratio (Rbusy) Channel busy ratio (Rbusy) Channel busy ratio (Rbusy)

Actual Bn+1 Actual Bn+1 Actual Bn+1 D(Bn+1) D(Bn+1) D(Bn+1) Range A Range A Range A Mean A Mean A Mean A (a) FER = 0.1 (b) FER = 0.25 (c) FER = 0.5

Figure 11: Case 3: achievable throughput estimation for n = 3&CR= 5.5Mbps.

Remarkably, in the case of R(500 K) for CR = 11 Mbps of the estimated bandwidth for R = 500 Kbps are higher and L = 500 B, the TFC measurement underestimates than thos for R = 1000 Kbps when CR ≤ 2Mbps and the bandwidth on the nAP. Because the nAP receives five L = 500, because the CR is lower than the traffic generation types 1 Mbps traffic with short L from PCs (n1 ∼ n5), it rate. The other reason why these results happen is that the heavily increases the nAP’s transmission rate, and the STA UDP datagram length of the cross trafficaffects the nAP’s cannot capture all the frame transmitted from the nAP. This processing overhead where the shorter L makes data frames problem can be solved if the STA uses large enough CP for on the nAP be generated with the more frequent interval. utilizing the TFC procedure, but the large CP may degrade Thus the accurate bandwidth estimation should take into the service through the current associated cAP as mentioned account the current channel rate set on both the nAP and in Section 6.1. the STA. The rest of the results show that the bandwidth As an additional observation, the TXnAP is higher than estimation with the TFC matches in higher accuracy when the reference bandwidth when the CR = 11 Mbps and L = the CR is greater than the reference bandwidth. 1000 B for both CP = 200 and 300 milliseconds. This is caused by the nAP’s retransmission of data frames when the 6.2.3. Case 3—Achievable Throughput Prediction with Rate nAP does not receive corresponding ACK frame before the Discount. In Cases 1 and 2, we evaluated how well measured ACK TIMEOUT expires. actual nAP’s bandwidth capacity by utilizing the TFC. In Case 3, we investigate the prediction of the achievable 6.2.2. Case 2—Accuracy of Bandwidth Estimation. Figures throughput based on the nAP’s bandwidth estimated by 10(a), 10(b), 10(c),and10(d) are plots of the estimated the TFC. Figures 11(a), 11(b),and11(c) are plots of the bandwidth in average of 5 TFC trials as a function of traffic achievable throughput predicted by (6) as a function of generation rate (R) when CR = 1, 2, 5.5, and 11 Mbps for Rbusy for FER = 0.1, 0.25, and 0.5, respectively, when the (CP = 200, L = 500), (CP = 200, L = 1000), (CP = 300, CR = 5.5 Mbps on the nAP. In this experiment, we perform L = 500), and (CP = 300, L = 1000), respectively. In the TFC for n = 3 on the nAP; thus it is to predict the this experiment, three PCs (n1, n2, and n3) generate UDP achievable throughput (range A) from the nAP for n = 4 datagram with the R of 10, 100, 500, and 1000 Kbps for cross which is expected after the STA associates with the nAP. To traffic on the nAP (n = 3); thus the reference bandwidth simplify the experiment, we generate cross trafficinfixed to be compared with the estimated bandwidth is R(30 K), length (L = 1000 B) and adjust Rbusy in 0.25, 0.5, 0.75, and R(300 K), R(1500 K), and R(3000 K), respectively. Most of 0.98 by varying the individual traffic generation rate in PCs. the result shows that the estimated bandwidth increases as We first measure the nAP’s bandwidth capacity (Bn) with the higher traffic is loaded on the nAP. However, several results TFC when three PCs (n1 ∼ n3) generate traffic destined to EURASIP Journal on Wireless Communications and Networking 13

Table 2: Simulation parameters.

(0,0)Simulation region (300,0) Parameter Value × 2 (300 × 300 m2) Simulation region 300 300 m (50,50) (150,50) (250,50) The number of AP 9 AP’s channel rate 11 and 54 Mbps (Fixed) AP’s transmission range 100 m radius The number of STA (N)50∼ 450 Required bandwidth by a STA 500 ∼ 1000 Kbps

(50,150) (150,150) (250,150) 6.3. Fairness. The proposed handoff mechanism also fairly balances the traffic loaded on multiple APs in regard to their bandwidth capacity. With a simple simulation in C programming, we evaluate the fairness of traffic load dis- (50,250) (150,250) (250,250) tributed on the APs by comparing our proposed mechanism, namely Best Bandwidth Fit (BBF), with two conventional (0,300) (300,300) AP selection mechanisms [22], Strongest Signal First (SSF) and Least Load First (LLF). The SSF and the LLF triggers STAs to perform handoff to an AP with the strongest signal strength and the lowest traffic load, respectively. In the BBF, 802.11a AP (54 Mbps) on the other hand, STAs perform handoff to an AP with the 802.11b AP (11 Mbps) most bandwidth capacity obtained by TFC so that it takes into account both the achievable throughput and the signal Figure 12: Simulation environment. strength from the AP. We use simulation parameters as in Table 2. As shown 800 in Figure 12, within the simulation region, we deploy 9 APs in the fixed location with different channel rates (five 802.11a and four 802.11b APs) and set all their channels 600 which do not interfere one another. By setting that the AP’s propagation range is 100 m, every STA can sense the carrier from at least one AP up to four overlapped APs. We vary 400 the number of STAs (N) from 50 to 450 where each STA locates in random location within the simulation region and requires constant bandwidth chosen from 500 to 1000 Kbps

Throughput per a STA200 (Kbps) (average 750 Kbps). Each simulation is performed 10 times 50 100 150 200 250 300 350 400 450 and obtained the result in average. Number of STA (N)

SSF 6.3.1. Throughput per an STA. Figure 13 shows the average LLF achieved throughput per an STA for SSF, LLF, and BBF mech- BBF anisms as N increases. When N>150, SSF and LLF dramatically decrease the throughput per an STA since they force Figure 13: Average throughput per an STA. the STAs to select an AP according to only the signal strength and the amount of loaded traffic. On the other hand, BBF the corresponding three laptops (s1 ∼ s3). Then we make an mechanism does not fluctuate the throughput per an STA STA associate with the nAP for the nAP in n = 4andmeasure since it fairly distributes the bandwidth capacity on the APs. the actual throughput (actual Bn+1) obtained from the nAP when n0 sends UDP traffic to the STA through the nAP. 6.3.2. AP’s Bandwidth Utilization. Figure 14 depicts the For the same Rbusy, the achievable throughput decreases average bandwidth utilization on APs as a function of N. as FER increases since it is affected by the quality of channel When N>150 which denotes that 802.11b APs are saturated, condition. Remarkably, comparing the rate discounted per- STAs in LLF mechanism are likely to select 802.11b APs STA bandwidth obtained by (7)(D(Bn+1)) with the actual since the loaded traffic on the 802.11b APs (≤11 Mbps) is Bn+1 leads to a similar bound. Also, every actual Bn+1 is less than that on the 802.11a APs (≤54 Mbps) so that the appeared within the range of A. Especially, for Rbusy ≤ 0.5, we utilization remains in about 60%. On the other hand, the can observe that the actual Bn+1 is closely distributed around average utilization of APs in SSF and BBF mechanism shows the mean A. In contrast, for Rbusy > 0.5, the actual Bn+1 is a linear growth. It means that APs in different PHY types also appeared within the range of A, but it is distributed in fully utilize their bandwidth capacity as the number of STA much more closer to the upper bound of the A. increases. 14 EURASIP Journal on Wireless Communications and Networking

100 service degradation of the active connection through the current associated AP. Based on the numerical analysis and 75 experimental studies, we showed that the estimation result of analytical model reasonably well matches the empirical one in terms of reliable and accurate bandwidth capacity as 50 well as rate discounted achievable throughput from neighbor APs. The proposed handoﬀ mechanism also achieves a better

AP utilization (%) 25 fairness by balancing the traffic load on the densely deployed APs. Moreover, our mechanism requires no changes in AP 0 protocols; thus it is easily applicable to any IEEE 802.11 50 100 150 200 250 300 350 400 450 WLAN NIC-based STA. Number of STA (N) As a future work, we will study a further model for throughput estimation taking into account the dynamic SSF LLF length of L which was assumed as a fixed length in this paper. BBF In addition, we assumed that APs use fixed channel rate, but the APs are often set with automatic fallback algorithm to Figure 14: Average bandwidth loaded on an AP. dynamically adjust the rate against the distance between STAs and the APs. Thus the heterogeneity of channel rate in APs 1 should be considered to design the estimation model.

0.8 Acknowledgments 0.6 This work was in part supported by a grant from Microsoft 0.4 Research Asia. This work was also partially supported by the cient of variation Korea Science and Engineering Foundation (KOSEF) grant ﬃ 0.2 funded by the Korea government (MEST) (2009-0076476). Coe

0 50 100 150 200 250 300 350 400 450 References Number of STA (N) [1] “IEEE Std 802.11. Wireless LAN Medium Access Control SSF (MAC) and Physical Layer (PHY) Specifications,” 1997. LLF [2] V. Mhatre and K. Papagiannaki, “Using smart triggers for BBF improved user performance in 802.11 wireless networks,” in Proceedings of the 4th International Conference on Mobile ffi Figure 15: CV of AP’s tra cload. Systems, Applications and Services (MobiSys ’06), pp. 246–259, Uppsala, Sweden, June 2006. [3] I. Ramani and S. Savage, “SyncScan: practical fast handoff 6.3.3. Coefficient of Variation. In order to show the fairness for 802.11 infrastructure networks,” in Proceedings of the of traffic distribution on APs, we obtained the coefficient of 24th Annual Joint Conference of the IEEE Computer and variation (CV) in regard to the traffic loaded on the APs. Communications Societies (INFOCOM ’05), vol. 1, pp. 675– The CV is calculated by CV = σ/μ where σ and μ are the 684, Miami, Fla, USA, March 2005. standard deviation and the mean of the loaded trafficforall [4] H. Wu, K. Tan, Y. Zhang, and Q. Zhang, “Proactive scan: ff APs. Figure 15 is a plot of the CV as a function of the number fast hando with smart triggers for 802.11 wireless LAN,” in Proceedings of the 26th IEEE International Conference on of STA. As N increases, the CV of every mechanism is Computer Communications (INFOCOM ’07), pp. 749–757, gradually reduced. The reducing slope of LLF mechanism is Anchorage, Alaska, USA, May 2007. slight while those of SSF and BBF mechanisms are drastically [5] V. Brik, A. Mishra, and S. Banerjee, “Eliminating handoff reduced. After the 802.11b APs are saturated (N>150), the latencies in 802.11 WLANs using multiple radios: applica- CV of BBF mechanism is less than that of SSF mechanism. It tions, experience, and evaluation,” in Proceedings of the ACM is obvious that BBF mechanism can more fairly distribute the SIGCOMM Internet Measurement Conference (IMC ’05),pp. traffic load on densely deployed APs than other mechanisms. 299–304, 2005. [6] C. Dovrolis, P. Ramanathan, and D. Moore, “Packet- dispersion techniques and a capacity-estimation methodol- 7. Conclusion ogy,” IEEE/ACM Transactions on Networking,vol.12,no.6,pp. ff 963–977, 2004. In this paper, we proposed a MAC layer hando mechanism [7] K. Lakshminarayanan, V. N. Padmanabhan, and J. Padhye, for IEEE 802.11 WLAN to determine an optimal AP “Bandwidth estimation in broadband access networks,” in with the maximum achievable throughput rather than the Proceedings of the ACM SIGCOMM Internet Measurement highest RSSI. The proposed handoff mechanism performs Conference (IMC ’04), pp. 314–321, 2004. Transient Frame Capture (TFC) to estimate the neighbor [8] S. Vasudevan, K. Papagiannaki, C. Diot, J. Kurose, and D. AP’s bandwidth capacity and achievable throughput without Towsley, “Facilitating access point selection in IEEE 802.11 EURASIP Journal on Wireless Communications and Networking 15

wireless networks,” in Proceedings of the ACM SIGCOMM Internet Measurement Conference (IMC ’05), p. 26, 2005. [9] M. Li, M. Claypool, and R. Kinicki, “Packet dispersion in IEEE 802.11 wireless networks,” in Proceedings of Conference on Local Computer Networks (LCN ’06), pp. 721–729, 2006. [10] A. Bazzi, M. Diolaiti, C. Gambetti, and G. Pasolini, “WLAN call admission control strategies for voice traffic over integrated 3G/WLAN networks,” in Proceedings of the 3rd IEEE Consumer Communications and Networking Conference (CCNC ’06), vol. 2, pp. 1234–1238, 2006. [11] D. Pong and T. Moors, “Call admission control for IEEE 802.11 contention access mechanism,” in Proceedings of IEEE Global Telecommunications Conference (GLOBECOM ’03), vol. 1, pp. 174–178, San Francisco, Calif, USA, December 2003. [12] Z. Kong, D. H. K. Tsang, and B. Bensaou, “Measurement assisted model-based call admission control for IEEE 802.11e WLAN contention-based channel access,” in Proceedings of the 13th IEEE Workshop on Local and Metropolitan Area Networks (LANMAN ’04), pp. 55–60, April 2004. [13] S. Kandula, K. C.-J. Lin, T. Badirkhanli, and D. Katabi, “FatVAP: aggregating AP backhaul capacity to maximize throughput,” in Proceedings of the 5th USENIX Symposium on Networked Systems Design and Implementation (NSDI ’08),pp. 89–104, San Francisco, Calif, USA, April 2008. [14] “IEEE Std 802.11e. Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, Amendment 8: Medium Access Control (MAC) Quality of Service Enhance- ments,” 2005. [15] “IEEE Std 802.11k. Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, Amendment 1: Radio Resource Measurement of Wireless LANs,” 2008. [16] S. Mangold and L. Berlemann, “IEEE 802.11k: Improving confidence in radio resource measurements,” in Proceedings of the 16th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC ’05), vol. 2, pp. 1009– 1013, Berlin, Germany, September 2005. [17] A. F. Conceicao, J. Li, D. A. Florencio, and F. Kon, “Is IEEE 802.11 ready for VoIP?” in Proceedings of the 8th IEEE Workshop on Multimedia Signal Processing (MMSP ’06),pp. 108–113, October 2007. [18] K. Tsukamoto, T. Yamaguchi, S. Kashihara, and Y. Oie, “Experimental evaluation of decision criteria for WLAN handover: signal strength and frame retransmission,” IEICE Transactions on Communications, vol. E90-B, no. 12, pp. 3579– 3590, 2007. [19] Y. Y. Kim and S.-Q. Li, “Modeling multipath fading channel dynamics for packet data performance analysis,” Wireless Networks, vol. 6, no. 6, pp. 481–492, 2000. [20]D.-J.Deng,C.-H.Ke,H.-H.Chen,andY.-M.Huang,“Con- tention window optimization for IEEE 802.11 DCF access control,” IEEE Transactions on Wireless Communications, vol. 7, no. 12, pp. 5129–5135, 2008. [21] Microsoft, “Network Monitor 3.1,” http://blogs.technet.com/ netmon. [22] Y. Bejerano, S.-J. Han, and L. Li, “Fairness and load balancing in wireless LANs using association control,” IEEE/ACM Trans- actions on Networking, vol. 15, no. 3, pp. 560–573, 2007. Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 479512, 7 pages doi:10.1155/2009/479512

Research Article On PHY and MAC Performance in Body Sensor Networks

Sana Ullah,1 Henry Higgins,2 S. M. Riazul Islam,1 Pervez Khan,1 and Kyung Sup Kwak1

1 Graduate School of Telecommunication Engineering, Inha University, 253 Yonghyun-Dong, Nam-Gu 402-751, Incheon, South Korea 2 Microelectronics Division, Zarlink Semiconductor Company, Castlegate Business Park, Portskewett, Caldicot NP26 5YW, UK

Correspondence should be addressed to Sana Ullah, [email protected]

Received 26 January 2009; Accepted 14 May 2009

Recommended by Naveen Chilamkurti

This paper presents an empirical investigation on the performance of body implant communication using radio frequency (RF) technology. In body implant communication, the electrical properties of the body influence the signal propagation in several ways. We use a Perspex body model (30 cm diameter, 80 cm height and 0.5 cm thickness) filled with a liquid that mimics the electrical properties of the basic body tissues. This model is used to observe the effects of body tissue on the RF communication. We observe best performance at 3cm depth inside the liquid. We further present a simulation study of several low-power MAC protocols for an on-body sensor network and discuss the derived results. Also, the traditional preamble-based TMDA protocol is extended towards a beacon-based TDMA protocol in order to avoid preamble collision and to ensure low-power communication.

Copyright © 2009 Sana Ullah et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction applications include monitoring and program changes for pacemakers and implantable cardiac defibrillators, control of Body Sensor Networks (BSNs) are becoming increasingly bladder function, and restoration of limb movement. These important for sporting activities, unobtrusive healthcare applications may require continuous or occasional one- or systems, and members of military services. They are con- two-way transmission. Some applications require a battery sidered as a key technology to prevent the occurrence of where the current drain must be low, so as not to reduce the myocardial infarction, monitor series of events or any other working life of the implant function. life critical condition, and are used for interactive gaming The development of an unobtrusive ambulatory BSN and entertainment applications. Traditionally, many body induces a number of issues and challenges such as interoper- functions were rarely monitored and separated by a consid- ability, scalability, Quality of Service (QoS), and low-power erable period of time. Holter monitors were used to collect communication protocols. A number of ongoing projects cardio rhythm disturbances for offline processing but they such as CodeBlue, MobiHealth, and iSIM have contributed were not used to provide real-time feedback [1]. For instance, to establish a proactive and unobtrusive BSN system [3–5]. transient abnormalities are sometimes hard to capture, for A system architecture presented in [6]performsrealtime example, many cardiac diseases are episodic such as transient analysis of sensor’s data, provides real time feedback to the surges in blood pressure, paroxysmal arrhythmias or induced user, and forwards the user’s information to a telemedicine episodes of myocardial ischemia and their timing cannot be server. UbiMon aims to develop a smart and affordable predicted [2]. BSNs allow continuous monitoring of patients health care system [7]. MIT Media Lab is developing MIThril under natural physiological states without constraining their that gives a complete insight of human-machine interface [8] normal activities. They are used to develop a smart and HIT focuses on quality interfaces and innovative wearable affordable health care system and can be a part of diagnostic computers [9]. IEEE 802.15.6 aims to provide power-efficient procedure, maintenance of chronic condition, and super- in-body and on-body wireless communication standards vised recovery from a surgical procedure. In-body sensor for medical and nonmedical applications [10]. NASA is networks are used to restore control over paralyzed limbs, developing a wearable physiological monitoring system for enable bladder and bowel muscle control, and maintain reg- astronauts called LifeGuard system [11]. ETRI focuses on the ular heart rhythm as well as many other functions. In-body development of a low-power MAC protocol for a BSN [12]. 2 EURASIP Journal on Wireless Communications and Networking

In this paper, we study the performance of radio Table 1: Body electrical properties [13]. frequency (RF) communication to an implant and present Muscle Fat a simulation study of several low-power MAC protocols Frequency −1 −1 for an on-body sensor network. The rest of the paper is (εr ) ρ(S.m ) Z0(Ω)(εr ) ρ(S.m ) Z0(Ω) categorized into four sections. Section 2 presents a discussion 100 66.2 0.73 31.6 12.7 0.07 92.4 on antenna design for an in-body sensor network. Section 3 400 58 0.82 43.7 11.6 0.08 108 investigates the performance of RF communication between 900 56 0.97 48.2 11.3 0.11 111 an implanted device and a base station. Section 4 provides a simulation study of several low-power MAC protocols for an on-body sensor network. This section also discusses the Planar Inverted-F Antenna (PIFA), loaded PIFA, the bow tie, potential issues and challenges in the development of in- spiral and trailing wire. These antennas have properties that body and on-body MAC protocols. Section 5 concludes our make them better suited for certain applications. work. 2.4. Impedance Measurement. The impedances of the patch 2. Antenna Design for an In-Body and dipole are affected considerably by the surrounded body Sensor Network tissue. The doctor determines the position of the implant within the body. It may move within the body after fitting. The band designated for in-body communication is Medical Each body has a different shape with different proportions of Implant Communication System (MICS) band, and is fat and muscle that may change with time. This means that around 403 MHz. Its wavelength in space is 744 mm, so a half a definite measurement of the antenna impedance is of little wave dipole is 372 mm. Clearly it is not possible to include an value. Measuring it immersed in a body phantom and makes antenna of such dimensions in a human body [13]. These an approximation of impedance liquid [14]. Using this constraints make the available size much smaller than the impedance, the antenna-matching network can be designed optimum. with the provision of software controlled trimming as can be The electrical properties of the body affect the prop- done with variable capacitors integrated into the transceiver. agation in several ways. First, the high dielectric constant The trimming routine should be run on each power up or at increases the electrical length of E-field antennas such as regular intervals to maintain optimum performance. a dipole. Second, body tissue, such as muscle, is partly conductive and absorbs some of the signal but it also acts as a parasitic radiator. This is significant when the physical 3. In-Body RF Communication antenna is much smaller than the optimum size. Typical The requirements of RF communication for on-body and dielectric constant (εr ), conductivity (ρ), and characteristic in-body sensor networks are different due to their cor- impedance Z0(Ω) properties of muscle and fat are shown in responding channel characteristics. In an on-body sensor Table 1. network, signals often propagate across the body surface. This propagation may be a combination of surface waves, 2.1. Dipole Antenna. For a dipole of length 10 mm, at creeping waves, diffracted waves, scattered waves, and free 403 MHz, the radiation resistance is 45 mΩ in air. The space propagation depending on the antenna position [15]. electrical length of the dipole is increased when surrounded In an in-body sensor network, the signals propagate inside by a material of high dielectric constant such as the body. the human body where the electrical properties of a body affect the signal propagation. All existing formulas to design 2.2. Loop Antenna. For a loop of 10 mm diameter, the area free-air communication are used for on-body communica- is 78.5 mm2. This results in a radiation resistance of 626 μΩ. tion systems. However, it is very difficult to calculate the However, the loop acts as a magnetic dipole that produces performance of in-body communication systems [16]. To more intense magnetic field than that of a dipole. The loop is compound the design challenges, the location of the implant of use within the body as the magnetic field is less affected by is also variable. During surgery the implant is placed in the the body tissue compared to a dipole or patch and it can be best position to perform its primary function, with a little more readily integrated into existing structures. consideration for the wireless performance. In-body RF communication uses MICS band that has ff 2.3. Patch Antenna. A patch antenna can be integrated amaximumEective Radiated Power (ERP) of 25 μW − into the surface of an implant. Without requiring much ( 16 dBm) in air. The Industrial Scientific and Medical additional volume, the ideal patch has dimensions as shown (ISM, 2.4–2.5 GHz) band is used to transmit a wakeup signal in Figure 1 and acts as a λ/2 parallel-plate transmission line to an implant with a power of 100 mW (+20 dBm). Once with impedance inversely proportional to the width. a wakeup signal is received at the implant, it powers up its The radiation occurs at the edges of the patch, as shown circuit as given in Figure 3. in Figure 2. For in-body use, a full size patch is not an option. An electrically small patch has a low real-valued impedance 3.1. Results. It is possible to simulate the performance of and therefore impaired performance compared to the ideal RF implant using 3D simulation software but this is time one. There are several other options for antenna such as consuming and is not valuable. We use a Perspex body model EURASIP Journal on Wireless Communications and Networking 3

6 2 L<λ/ Transmit 5 PLL 4 Sleep lock 200 nA Start crystal Feed point 3 oscillator, (position aﬀects W<λ calibrate impedance) 2 and memory Wake up check 1

Figure 1: Patch antenna plan view. 0 0 5 10 15 20 Time (ms)

Patch Air or other Propagation Figure 3: Implant wakeup sequence and current consumption. medium from edge

−80

−85

Ground Shorting pin Dielectric −90 ERP Feed plane (option) substrate point −95

Figure 2: Patch antenna side view. −100 0 5 10 15 20 Depth in liquid (cm)

filled with a liquid that mimics the electrical properties of Figure 4: ERP versus Depth in liquid. the basic body tissue. The liquid contains water, sodium chloride, sugar, and Hydroxyl Ethyl Cellulose (HEC), which mimics muscle or brain tissue for the frequency range from Table 2: Body tissue recipes [17]. 100 MHz to 1 GHz as given in Table 2. The Perspex body is %ofweight %ofweight defined in standard ETSI [17]. It is a 76 cm high and has a Ingredient 30 cm diameter. The Perspex tank that we use has a 30 cm (100 MHz to 1 GHz) (1.5 MHz to 2.5 GHz) diameter, an 80 cm height, and a 0.5 cm wall thickness. Water 52.4 45.3 Figure 4 shows the ERP from an implant immersed in a Sugar 45.0 54.3 tank of body phantom liquid. The implant is transmitting Salt (NaCl) 1.5 0.0 a Continuous Wave (CW) signal, where the measurement HEC 1.1 0.4 is performed with a log periodic antenna and a spectrum analyzer. The environment is an anechoic chamber with a tank and a log periodic antenna separated by 3 m. Using the antenna parameters and the measured signal power, the (RSSI) of the base-station is recorded. RSSI is a relative ERP is calculated. Clearly, the ERP increases from a 1 cm measure of signal strength with each point equivalent to depth to a maximum between 2 cm and 7 cm, thereafter approximately 2.5 dB. As with the signal level measurement, it decreases. The gradual increase is due to the simulated the RSSI increases from the initial value, then decreases with body acting as a parasitic antenna. The implant patch is very depth as illustrated in Figure 5. small compared to the air wavelength and its performance In Figure 6, data is exchanged between the implant is improved by contact with tissue—holding it in a hand and the base station. When data is exchanged between the improves the measured signal strength by about 10 dB over implant and the base-station, error correction is used to performance in air. There are possibilities, that is, the liquid ensure that reliable data is obtained. If an error is detected acts as a parasitic antenna and also attenuates the signal. The then it is corrected by invoking an Error Correction Code reduction in signal level with depth is expected as the liquid (ECC). The infrequent ECC invocation shows better link absorbs the signal. quality. As with the signal level and RSSI, the figure further The implant is immersed into a tank of body phantom shows an improvement in the link at a depth between liquid at various depths. The base-station antenna is a 3 cm and 5 cm. We conclude that the implant reveals best dipole with a distance to the tank of 3 m. With the implant performance at a depth of 3 cm and not close to the skin transmitting a CW signal, the Remote Signal Level Indication surface. 4 EURASIP Journal on Wireless Communications and Networking

4. MAC Protocol for BSNs 10

MAC protocols are classiﬁed into contention-based and 8 TDMA-based protocols. In contention-based protocols, nodes contend for the channel using CSMA mechanism. 6

If the channel is busy, the node defers its transmission RSSI until the channel becomes idle. These protocols are scalable 4 with no strict time synchronization constraint. However, 2 they incur significant protocol overhead. In TDMA-based protocols, the channel is divided into time slots of fixed 0 duration. These slots are assigned to the nodes and each node 0246810 transmits during its own slot period. These protocols are Depth in liquid (cm) energy conserving protocols. Because the duty cycle of radio Figure 5: RSSI versus Depth in liquid. is reduced and there is no contention, idle listening, and overhearing problem but these protocols require frequent synchronization. 10 Li and Tan proposed a novel TDMA protocol for an on- body sensor network that exploits the biosignal features to 8 perform TDMA synchronization and improves the energy 6 efficiency [18]. Other protocols like WASP, CICADA, and BSN-MAC are proposed in [19–21]. The performance of 4 a nonbeacon IEEE 802.15.4 is investigated in [22], where ECC average the authors considered low upload/download rates, mostly 2 per hour. Furthermore, the data transmission is based on 0 periodic intervals that limit the performance to certain 0246810 applications. There is no reliable support for on-demand and Depth in liquid (cm) emergency traffic. The BSN traffic requires sophisticated low-power tech- Figure 6: ECC invocation versus Depth in liquid. niques to ensure safe and reliable operations. Existing MAC protocols such as SMAC [23], TMAC [24], IEEE 802.15.4 [25], and WiseMAC [26] give limited answers to the Table 3: Coexistence test results between IEEE 802.15.4 and heterogeneous traffic. The in-body nodes do not urge syn- microwave oven. chronized and periodic wakeup patterns due to unpredicted Packet success rate Microwave status medical events. Medical data usually needs high priority and Mean Std. reliability than nonmedical data. In case of emergency events, ON 96.85% 3.22% the nodes should access the channel in less than one second OFF 100% 0% [27]. The IEEE 802.15.4 can be considered for certain on- body applications but it does not achieve the required power level of in-body nodes. For critical and noncritical medical Table 4: Packet delivery ratio and power (in mW). traffic, the IEEE 802.15.4 has several power consumption and IEEE IEEE QoS issues [28–31]. Also, this standard operates in 2.4 GHz IEEE Sensor nodes 802.11e 802.11e 802.15.4 band, which allows the possibilities for interference from (AC BE) (AC VO) other devices such as IEEE 802.11 and microwave. Table 3 Wave-form 100% 100% 100% shows the effects of microwave oven on the XBee remote Packet delivery ratio module [32]. When the microwave oven is ON, the packet Parameter 99.77% 100% 100% Wave-form 1.82 4.01 3.57 success rate and the standard deviation are degraded to Power (mW) 96.85% and 3.22%, respectively. However, there is no loss Parameter 0.26 2.88 2.77 when the XBee modules are taken 2 meters away from the microwave oven. Dave et al. studied the energy efficiency and QoS provides reliable solution for an on-body sensor network but performance of IEEE 802.15.4 and IEEE 802.11e [33]MAC it has several limitations for an in-body sensor network. The protocols under two generic applications: a wave-form real main reason is that the path loss inside the human body due time stream and a real-time parameter measurement stream to tissue heating is much higher than in the free space. The [34]. Table 4 shows the packet delivery ratio and the Power in-body nodes cannot perform Clear Channel Assessment (in mW) for both applications. The AC BE and AC VO (CCA) in a favorable way. Zhen et al. analyzed the perfor- represent the access categories voice and best-effort in the mance of CCA by in-body and on-body nodes [35]. Figure 7 IEEE 802.11e. shows that for a given −85 dBm CCA threshold, the on-body In a beacon-enabled IEEE 802.15.4, nodes use slotted nodes cannot see the activity of in-body nodes when they are CSMA/CA to contend for the channel. The use of CSMA/CA away at 3 m distance from the surface of the body. EURASIP Journal on Wireless Communications and Networking 5

−50

−60 37.05 −70 C)

◦ 37.04 (dBm) −

r 80

owe −90 37.03 p −100

CCA 37.02 −110 Temperature ( 37.01 −120 12345678 Free space distance (meters) 37 2 4 6 8 10 12 14 On-body Sleep time In-body CCA threshold Aloha CSMA/CA Figure 7: CCA in on-body and in-body sensor networks. Figure 8: Saturated temperature using aloha and CSMA/CA.

The in-body nodes (MAC) should also consider the thermal influence caused by the electromagnetic wave expo- 102 sure and circuit heat. Nagamine and Kohno discussed the thermal influence of the in-body nodes using different MAC protocols in [36]. Figure 8 shows the temperature of a node 1 when ALOHA and CSMA/CA are used. 10

4.1. Simulation Environment. We present the performance analysis of Preamble-Based TDMA (PB-TDMA) [37], 100 beacon-enabled IEEE 802.15.4, and S-MAC protocols for an on-body sensor network using NS-2 [38]. In case of Packet delivery ratio (%) PB-TDMA and S-MAC, the wireless physical parameters − 10 1 are considered according to low-power Nordic nRF2401 −25 −20 −15 −10 −5 transceiver [39]. This radio transceiver operates in the Transmission power (dBm) 2.4–2.5 GHz band with an optimum transmission power of −5 dBm. However, in case of IEEE 802.15.4, Chipcon PB-TDMA S-MAC CC2420 radio interface is considered [40]. We use the 802.15.4 shadowing propagation model throughout the simulations. The parameters in the shadowing propagation model are Figure 9: Packet delivery ratio. adjusted according to [41]. We consider 6 nodes ﬁrmly placed on the human body. The nodes are connected to the coordinator in a star topology. The initial node energy is 5 5 Joules. The data rate of the nodes is heterogeneous. The simulation area is 1 × 1 meter and each node generates 4.98 Constant Bit Rate (CBR) traﬃc. The packet size is 134 bytes. 4.96 The transport agent is User Datagram Protocol (UDP). For the performance analysis of IEEE 802.15.4, we use part of the 4.94 results discussed in [42]. 4.92

4.2. Results. In Figure 9, we present the packet delivery 4.9 ratio for diﬀerent transmission powers. In a beacon-enabled Residual energy (Joules) mode, the packet delivery ratio of IEEE 802.15.4 for all 4.88 transmission powers is almost 100% with tolerable power − 4.86 consumption. PB-TDMA gives 90% value for 5 dBm, while 100 101 102 103 S-MAC gives only 5% value. Simulation time (seconds) Figure 10 considers PB-TDMA protocol to show the residual energy at ECG node for diﬀerent transmission 5dBm powers. There is a minor change in the residual energy 10 dBm for three transmission powers. This further concludes that 20 dBm reducing the transmission power does not ensure low-power Figure 10: Residual energy at ECG node. 6 EURASIP Journal on Wireless Communications and Networking

101 sensor network. We also discussed the potential issues and challenges in the development of a novel low-power MAC protocol for a BSN. 100

Acknowledgment 10−1 This research was supported by the The Ministry of Knowl-

− edge Economy (MKE), Korea, under the Information Tech- 10 2 nology Research Center (ITRC) support program supervised Energy consumption (Joules) by the Institute for Information Technology Advancement 10−3 (IITA) (IITA-2009-C1090-0902-0019). 100 101 102 103 Packets(s) References TDMA with preamble TDMA with beacon [1] December 2008, http://www.nyp.org/health/electrocardio- gram-stresstest-holter.html. Figure 11: Power consumption of TDMA protocol with a preamble and a beacon. [2] B. Lo and G. Z. Yang, “Key technical challenges and current implementations of body sensor networks,” in Proceedings of the 2nd International Workshop on Body Sensor Networks (BSN ’05), pp. 1–5, April 2005. ffi communication unless supported by an e cient power [3] November 2008, http://fiji.eecs.harvard.edu/CodeBlue. management scheme. [4] January 2009, http://www.mobihealth.org. Generally, PB-TDMA protocol uses a preamble for data [5] July 2008, http://www.cs.uoregon.edu/research/wearables/ slot allocation. The preamble contains a dedicated subslot for index.html. each node. These subslots are used to activate the destination [6] E. Jovanov, A. Milenkovic, C. Otto, and P. C. De Groen, node by broadcasting the destination node ID of an outgoing “A wireless body area network of intelligent motion sensors packet. This leads the high traffic nodes (in case, many for computer assisted physical rehabilitation,” Journal of nodes activate their destination nodes) towards a preamble NeuroEngineering and Rehabilitation, vol. 2, no. 6, 2005. collision. We propose a beacon-based TDMA protocol that [7] March 2008, http://www.ubimon.net. provides a solution to avoid preamble contention by using [8] Febuary 2008, http://www.media.mit.edu/wearables/mithril. a beacon (based on IEEE 802.15.4) instead of a preamble. [9] April 2008, http://www.hitl.washington.edu. The beacon frame is controlled and broadcasted by the [10] December 2008, http://www.ieee802.org/15/pub/TG6.html. coordinator and is mainly used for synchronization and [11] January 2009, http://lifeguard.stanford.edu. resource allocation purposes. Figure 11 shows the energy [12] http://www.etri.re.kr/eng. consumption of a TDMA protocol with a preamble and a [13] G.-Z. Yang, Body Sensor Networks,Springer,NewYork,NY, beacon for a 256 bytes packet size. Unlike preamble which USA, 2006. is used by the nodes to broadcast destination ID, coordinator [14] J. Wojclk, et al., “Tissue Recipe Calibration Requirements, broadcasts the beacon frames and hence, avoids collisions. SSI/DRB-TP-D01-003,” Spectrum Sciences. The figure also shows that a proper coordination and con- [15] P. Hall, “Antennas challenges for body centric communica- trolling mechanism (beacon-based TDMA protocol) at the tions,” in Proceedings of the International Workshop on Antenna Technology: Small and Smart Antennas Metamaterials and coordinator ensures low-power communication compared Applications, 2007. with an improper coordination (preamble-based TDMA [16] S. Ullah, H. Higgins, M. Siddiqui, and K. S. Kwak, “A study protocol) mechanism. of implanted and wearable BSN,” vol. 4953 of Lecture Notes in Computer Science, pp. 464–473, 2008. 5. Conclusions [17] Electromagnetic Compatibility and Radio Spectrum Matters (ERM), “Electromagnetic Compatibility (EMC); standard for This paper studied the possibilities of RF communication to radio equipment and services; Part 27; specific conditions for a device implanted under the human skin. We used a Perspex Ultra Low Power Medical Implants (ULP-AMI) and related tank of a 30 cm diameter, an 80 cm height, and a 0.5 cm wall peripheral devices(ULP-AMI-P),”Tech. Rep. ETSI EN301 489- thickness for empirical investigation. The tank was filled with 27, V1.1.1, ETSI, Cedex, France, March 2003. a liquid that mimicked the electrical properties of the human [18] H. Li and J. Tan, “Heartbeat driven medium access control body at 400 MHz. The liquid acted as a parasitic antenna and for body sensor networks,” in Proceedings of the 1st ACM SIG- MOBILE International Workshop on Systems and Networking also attenuated the signal. We concluded that the gradual Support for Healthcare and Assisted Living Environments,pp. increase in ERP is due to the liquid acted as a parasitic 25–30, 2007. antenna. Furthermore, the signal increased to an optimum as [19] B. Braem, B. Latre, I. Moerman, C. Blondia, and P. Demeester, we immersed the implant deeper into the tank. We observed “The wireless autonomous spanning tree protocol for mul- best performance at 3 cm depth inside the liquid and not tihop wireless body area networks,” in Proceedings of the 1st close to the skin surface. We further provided a simulation International Workshop on Personalized Networks,SanJose, study of several low-power MAC protocols for an on-body Calif, USA, 2006. EURASIP Journal on Wireless Communications and Networking 7

[20] B. Latre, B. Braem, I. Moerman, C. Blondia, E. Reusens, W. [35] B. Zhen, H.-B. Li, and R. Kohno, “IEEE body area networks Joseph, and P. Demeester, “A low-delay protocol for multihop and medical implant communications,” in Proceedings of the wireless body area networks,” in Proceedings of the 4th Annual ICST 3rd International Conference on Body Area Networks, International Conference on Mobile and Ubiquitous Systems: Tempe, Ariz, USA, 2008. Computing, Networking and Services (MobiQuitous ’07),pp.1– [36] S. Nagamine and R. Kohno, “Design of communication model 8, 2007. suitable for implanted body area networks,” in Proceedings of [21] H. Li and J. Tan, “Medium access control for body sensor the 3rd International Symposium on Medical Information and networks,” in Proceedings of the 16th International Conference Communication Technology (ISMICT ’09), Montreal, Canada, on Computer Communications and Networks (ICCCN ’07),pp. February 2009. 210–215, Honolulu, Hawaii, USA, August 2007. [37] S. Ullah, R. Islam, A. Nessa, Y. Zhong, and K. S. Kwak, [22] N. F. Timmons and W. G. Scanlon, “Analysis of the perfor- “Performance analysis of a preamble based TDMA protocol mance of IEEE 802.15.4 for medical sensor body area network- forwirelessbodyareanetwork,”Journal of Communications ing,” in Proceedings of the 1st Annual IEEE Communications Software and Systems, vol. 4, no. 3, 2008. Society Conference on Sensor and Ad Hoc Communications and [38] March 2008, http://www.isi.edu/nsnam/ns. Networks (SECON ’04), Santa Clara, Calif, USA, October 2004. [39] July 2008, http://www.sparkfun.com/datasheets/RF/nRF2401- [23] W. Ye, J. Heidemann, and D. Estrin, “Medium access control rev1 1.pdf. with coordinated adaptive sleeping for wireless sensor net- [40] July 2008, http://www.moteiv.com. works,” IEEE/ACM Transactions on Networking,vol.12,no.3, [41] A. Fort, C. Desset, et al., “Ultra wide-band body area channel pp. 493–506, 2004. model,” in Proceedings of IEEE International Conference on [24] T. Van Dam and K. Langendoen, “An adaptive energy-efficient Communications (ICC ’05), Seoul, Korea, May 2005. MAC protocol for wireless sensor networks,” in Proceedings [42] W. Lars and S. Sana, “Architecture concept of a wireless body of the 1st International Conference on Embedded Networked area sensor network for health monitoring of elderly people,” Sensor Systems (SenSys ’03), pp. 171–180, Los Angeles, Calif, in Proceedingsof the 4th IEEE Consumer Communications and USA, November 2003. Networking Conference (CCNC ’07),LasVegas,Nev,USA, [25] IEEE Std.802.15.4, “Wireless medium access control (MAC) January 2007. and physical layer (PHY) specifications for low data rate wireless personal area networks (WPAN),” 2006. [26] A. El-Hoiydi and J.-D. Decotignie, “WiseMAC: an ultra low power MAC protocol for the downlink of infrastructure wireless sensor networks,” in Proceedings of the 9th IEEE Symposium on Computers and Communication (ISCC ’04),pp. 244–251, Alexandria, Egypt, 2004. [27] Technical Requirement Document, IEEE 802.15.6, January 2009. [28] A. Sikora and V. Groza, “Coexistence of IEEE 802.15.4 with other systems in the 2.4 GHz ISM-band,” in Proceedings of IEEE Instrumentation and Measurement Technology Conference (IMTC ’05), Ottawa, Canada, May 2005. [29] N. Golmie, D. Cypher, and O. Rebala, “Performance analysis of low rate wireless technologies for medical applications,” Computer Communications, vol. 28, no. 10, pp. 1266–1275, 2005. [30] N. Chevrollier, N. Montavont, and N. Golmie, “Handovers and interference mitigation in healthcare environments,” in Proceedings of IEEE Military Communications Conference (MILCOM ’05), Atlantic City, NJ, USA, October 2005. [31] I. Howitt and J. Gutierrez, “IEEE 802.15.4 low rate—wireless personal area network coexistence issues,” in Proceedings of IEEE Wireless Communications and Networking Conference (WCNC ’03), pp. 1481–1486, 2003. [32] C. Chen and C. Pomalaza-Raez,´ “Monitoring human move- ments at home using wearable wireless sensors,” in Proceedings of the 3rd International Symposium on Medical Information and Communication Technology (ISMICT ’09), Montreal, Canada, February 2009. [33] IEEE 802.11e Std, “Amendment to Part 11: “Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications”,” “Medium Access Control Quality of Services Enhancements” November 2005. [34] D. Cavalcanti, R. Schmitt, and A. Soomro, “Performance analysis of 802.15.4 and 802.11e for body sensor network applications,” in Proceedings of the 4th International Workshop on Wearable and Implantable Body Sensor Networks (BSN ’07), 2007. Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 802523, 8 pages doi:10.1155/2009/802523

Research Article Problem Solving of Low Data Throughput on Mobile Devices by Artefacts Prebuffering

Ondrej Krejcar

Department of Measurement and Control, Centre for Applied Cybernetics, Faculty of Electrical Engineering and Computer Science, VSB Technical University of Ostrava, 17 Listopadu 15, 70833 Ostrava Poruba, Czech Republic

Correspondence should be addressed to Ondrej Krejcar, [email protected]

Received 29 March 2009; Revised 29 July 2009; Accepted 11 November 2009

Recommended by Naveen Chilamkurti

The paper deals with a problem of low data throughput on wirelessly connected mobile devices and a possibility to solve this problem by prebuffering of selected artefacts. The basics are in determining the problem parts of a mobile device and solve the problem by a model of data prebuffering-based system enhancement for locating and tracking users inside the buildings. The framework uses a WiFi network infrastructure to allow the mobile device determine its indoor position. User location is used for data prebuffering and for pushing information from a server to PDAs. All server data are saved as artefacts with its indoor position information. Accessing prebuffered data on a mobile device can significantly improve a response time needed to view large multimedia data. The solution was tested on a facility management information system built on purpose with a testing collection of about hundred large size artefacts.

Copyright © 2009 Ondrej Krejcar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction The goal is to complete the data networking capabilities of RF wireless LANs [1] with accurate user location and The mobile wireless devices (laptops, PDA devices, Smart tracking capabilities for user needed data prebuffering. This phones, etc.) are commonly used with internet connection property is used as an information base for an extension of which is available almost everywhere and anytime these days. existing information systems or to create a special new one. The connection speed of the two most common standards An information about location is used to determine GPRS and WiFi varies from hundreds of kilobits to several both an actual and future position of a user [2]. A number megabits per second. In case of corporate information sys- of experiments with the information system have been tems or some other types of facility management, zoological performed and their results suggest that determination of or botanical gardens, libraries, or museums information the location should be focused on. The following sections systems, the WiFi infrastructure network is often used to describe also the conceptual and technical details about connect mobile device clients to a server. Unfortunately, the Predictive Data Push Technology Framework (PDPT). theoretical maximum connection speed is only achievable on laptops where high-performance components are used (in comparison to mobile devices). Other mobile devices 2. The Problem of Low Data Throughput like family PDAs or Smart phones have low-performance components due to a very limited space. The limited Why can we not use the classical model of user’s requests and connection speed represents a problem for online systems server’s response for large data files? It is because some large using large artefacts data files. It is not possible to preload amounts of data (artefacts) are impossible to download to these artefacts before the mobile device is used in remote PDA device and to be displayed in relatively short time. access state. This problem was found as a very important Each data artefact has to go through an identical way. point. The rest of this paper specifies the problem and Starting at the server database, it follows the WiFi Access suggests a possible solution. Point (AP) and finally reaches a PDA display (Figure 1). 2 EURASIP Journal on Wireless Communications and Networking

PDA mobile device Table 1: Data transfer tests—PDA is connected through WiFi infrastructure—use of internal FLASH ROM memory. WiFi SD antenna card CPU PDA device Memories Data size (MB) Athena Universal Blueangel Roadster WiFi Flash ROM Transfer Speed (kB/s) adapter 10 347 123 160 106 RAM 20 344 121 157 79 Air 30 314 123 58 43

Server Table 2: Data transfer tests—SPB Benchmark software—Internal ﬂash memory of PDA devices.

WiFi AP Display PDA device Type of test Athena Universal Blueangel Roadster Figure 1: A data communication way—from server database to Transfer Speed (kB/s) PDA display. Write 1 MB 2268 667 359 133 Read 1 MB 20400 3659 2180 30395 × There are several important components: Write 100 10 kB 298 229 128 27 Read 100 × 10 kB 1880 1637 1199 737 (1) an Ethernet network (Server to WiFi AP), (2) WiFi Access Points, (3) wireless communication between WiFi AP and WiFi real transfer rate (687 kB/s versus 160 kB/s) is due to the low antenna of the PDA device, performance components of mobile devices. (4) a PDA WiFi antenna, In theory, it is possible to find the worst component of all (5) a PDA WiFi adapter, components and try to improve it. With this idea in mind, components 5 to 8 have been tested. It is not possible to (6) a PDA CPU, test the WiFi adapter directly. The only possible way is to (7) a PDA memory, update a driver. The hardware of the WiFi adapter was found (8) a graphical unit—display data to user. as identical in most cases of classical PDA devices from HP, HTC, or Acer companies. State-of-the-art drivers were found For large amount data artefacts, the most important for all test devices showing us exactly the same transfer rates. parts are those listed under 5 to 8. The first and the Unfortunately, no easier solution to improve this component second components have a relatively high throughput when is known at this stage. compared to PDA device components and do not require A CPU component is of course one of the most testing. The Wireless communication between WiFi AP and important parts at all. The power of the CPU can be easily PDA WiFi antenna cannot be improved. benchmarked. All tested devices have an Intel XScale CPU. The fourth component is a WiFi antenna of the PDA. The Blueangel is equipped with PXA 263 with 400 MHz, This antenna has a standard of 2 dB gain and cannot be Universal has 520 MHz PXA 270, and Athena and Roadster improved due to the absence of a connector to an external have 624 MHz PXA 270. Let us try to compare the best antenna. It is thus not necessary to test the antenna. devices—Athena and Roadster with same CPU unit. A single A number of tests with several types of PDA devices CPU has entirely different transfer rates. That is why the (HTC Athena, HTC Universal, HTC Blueangel and HTC speed of CPU is not so important on this occasion. Roadster) have been undertaken. These PDA devices were The PDA memory is a very important part on the other connected through CISCO WiFi APs (Signal Strength quality side. There is a large room for improvement of every PDA ≥80%) to an FTP server. The FTP server holds 3 types of because of an SD Card slot which is present in most cases. large artefacts (files) which were downloaded to the internal The test was performed with standard SD Cards (Table 3). PDA memory. In the test, Athena outperformed other devices (Table 2). The maximal theoretical transfer rates (max 687 kB/s for A writing operation of 1 MB file was achieved at a speed of IEEE 802,11b standard [3]) have not been achieved (Table 1). 2268 kB/s. The remaining devices only achieve about 25% or The maximum transfer rate 350 kB/s has been obtained by less of this speed which is insufficient. A reading operation of HTC Athena, but this is not a standard PDA device. Athena 1 MB file provides two errors in case of Athena and Roadster. is a mininotebook with Windows Mobile 6 operating system. The testing was sufficient because of extremely high and All other standard mobile devices have reached only about quick RAM memory. On the other hand, Universal and 25% of this speed (max 160 kB/s). Significant part of transfer Blueangel provide good and relevant data. The 100 × 10 kB speed is taken by the protocol cost on physical layer (app. test has been undertaken for comparative purposes only. The 30–40%), but the rest of difference between theoretical and main objective is to focus on large data artefacts. EURASIP Journal on Wireless Communications and Networking 3

Table 3: Data transfer tests—SPB Benchmark software—SD Cards. SD Card Type of test Kingston 1 GB 50× Kingston 2 GB 50× Kingston 2 GB 120× Pretec 2 GB 133× Transfer Speed (kB/s) Write 1 MB 523 414 929 717 Read 1 MB 1008 1101 1219 1006 Write 100 × 10 kB 73 144 263 48 Read 100 × 10 kB 756 899 1012 822

The best tested SD Card (Table 3) is a Kingston 2 GB Table 4: Data files displaying tests for 500 kB files (13 iterations). 120× (120× means the card is 120 times faster than standard UniversalBlueangelRoadster single speed CD ROM (150 kB/s)). When compared to an Data type internal flash ROM, the writing speed of internal flash ROM Data file displaying time (s) is lower the reading speed remains 2 or 3 times higher. Jpeg 0,88 1,27 4,5 The problem of low data throughput begins exactly AutoCAD 45,1 56,8 81,6 in writing speed of internal flash. All data files are first Word 19,8 29,9 40,3 transferred to cache (use of Microsoft explorer or Opera browser) which is relatively slow. The use of an SD card could speed up the transfer. The other problem particularly for large data artefacts is Table 5: Application starting times for selected data types (13 the free internal memory. All PDA devices have a very limited iterations). space and usually operate with a memory allowing for 20 to 50 MB of free space including Athena device. Therefore the Application Universal Blueangel Roadster use of SD Cards would be a solution. Application Start Speed (s) MS Internet Explorer 2,3 3,9 5,4 PPC CAD Viewer 1,3 2,4 3,1 2.1. Maximum Application Response Time. Nielsen [4]spec- MS Word Mobile 5,4 6,7 8,2 ified the time delay of application response to user request to 10 seconds [5]. “During this time the user was focused on the application and was willing to wait for an answer.” Nielsen’s book [4] was published in 1994, but it is a basic literature for this phenomenon. Another interesting source [6] suggests Thetimerequiredtoopena500kBfileissummarizedin that “decreases in performance and behavioural intentions Table 4. The data file open delay for both AutoCAD and MS begin to flatten when the delays extend to 4 seconds or longer, Word files is significantly longer than the basic Jpeg data type. and attitudes flatten when the delays extend to 8 seconds or To avoid any doubt, the application start time was measured longer.” Based on these sources the maximum application too (Table 5). The results are acceptable as the delays in all response time is set to 10 seconds. cases remain below the 10 seconds limit. During the set maximum response time, the requested data must be downloaded and showed to user on display (in Unfortunately, the displaying time delays of files with case of remote request to server’s data). The time period of 10 nonbasic types are unacceptable. A basic data format must seconds is used to calculate the maximum possible data size be used to display files by PDA natively (BMP, JPG, GIF) of a file transferred from server to client (during this period). without any additional striking time consumption. The To achieve the best transfer speed 160 kB/s, the calculated file solution is a conversion from any format to these native size is 1600 kB. formats (for PDA devices). In case of sound and video Thenextstepistodefineanaverageartefactsize.The formats, it can also be recommended to use basic data format network architecture building plan is currently used as a (wav, mp3, wmv, and mpg). In case of sample database, the sample database, which contains 100 files of an average display time of artefact is only about a half second per 500 kB size of 470 kB. During the 10-second period, the client artefact. This short time delay is not considered to 10 seconds application can download 2 to 3 files (depending on the response limitation. If other file types are used, the delay for actual connection capabilities). presentation of file must be included. The second problem is the extremely long delay in The end result of several real tests and subsequent displaying files in certain original file types (e.g., AutoCAD calculations give a definition of artefact size as an average in case of vector graphic or MS Office in general cases). value of 500 kB. The buffer size may differ from 50 to 100 MB An AutoCAD file type is used in most cases of facility in case of 100 to 200 artefacts. management of modern building [7]. In such cases the In order to provide the reader with more information, the mobile user needs to view a selected building scheme next chapter describes how a position can be obtained from (building area plan, gas line plan, etc.) immediately. wireless networks background. 4 EURASIP Journal on Wireless Communications and Networking

setting up a WiFi connection. If the mobile device knows the PDPT server PDPT client WiFi signal position of the stationary device, it also knows that its own Location strength Location position is within a 100-meter range of this location provider. processing sensor The location accuracy can be improved by triangulation of two or several visible WiFi APs [12, 13]. The PDA Artifacts PDPT Artifact client will support the application in automatically retrieving buffering managing core location information from nearby location providers, and in interacting with the server. Naturally, this principle can be applied to other wireless technologies. The application SQL server SQL server CE is now implemented in C# using the MS Visual Studio .NET 2005 with .NET compact framework and a special Figure 2: PDPT Framework architecture. Measured WiFi SS goes OpenNETCF library enhancement. The information about through the Location sensor to Location processing where the user the basic concept and technologies of user localization can position, current track, predicted track, and predicted user position be found in [7]. are computed. PDPT Core makes a selection of artefacts to be The current and predicted user positions are used ff prebu ered to mobile SQL Server CE. for the PDPT framework to make decisions as to which data artefacts are needed in the PDA memory. The data prebuffering increases the primary dataflow from WiFi AP 3. Predictive Data Push Technology Framework (server side) to PDA (client side). These techniques form the basis of the predictive data push technology (PDPT). In most cases the low software level cache is used [8]or PDPTs push the data from an information server to the residing of chips on system desk is recommended [9] the client’s PDA on the basis of the user’s location and to improve the performance of a system when operating user’s future predicted location. The prebuffered data will with multimedia content. Such techniques are not allowed be helpful when the user comes to the location which was on existing mobile device where the operation system exists. predicted by PDPT Framework. The benefit of the PDPT Only a software solution added on top of the OS can achieve consists in the reduction of time needed to display a desired the objective. information requested by a user command on the PDA. This A combination of a predicted user position with pre- delay may vary from a few seconds to a number of minutes. buffering of data associated with physical locations bears It depends on two aspects. many advantages in increased throughput of mobile devices. The first aspect is the quality of wireless WiFi connection An interesting solution (Microsoft US patent [9]) in this used by the client PDA. A theoretical speed of WiFi field needs to know all information (AP location, SS power, connection is maximum 687 kB/s. However, the test suggests etc.) of all wireless base stations in mobile device before the a speed of only 43–160 kB/s (depending on file size and PDA localization process can be started (see the Location Manager device) (Table 1). module [10]). Moreover, the Moving Direction Estimator The second aspect is the size of copied data. The current module is also situated in a mobile device application. application records just one set of WiFi signal strength (SS) These two facts present limitations to changing wireless values at a time (by Locator unit in PDPT Client). From this base stations structure or to computing power consumption. set of values the actual user position is determined by the Another solution (HP US patent [11]) represents a similar PDPT Server side. PDPT Core responds to a location change concept. A Location Determination and Path Guide modules by selecting the artefact to load to PDPT Client buffer. The are situated in mobile device side too. data transfer speed is to a large extent influenced by the size The key difference between [10, 11] and PDPT solution of these artefacts. For larger artefact size the speed decreases. is that the location processing, track prediction, and cache content management are situated at server side (Figure 2). This fact allows for managing many important parameters 3.1. PDPT Framework Data Artifact Manager. The PDPT (e.g., AP info changes, position determination mechanism Server SQL database manages the information (e.g., data tuning, artefacts selection evaluation tuning, etc.) online at about Ethernet hardware such as Ethernet switch UTP aPDPTServer. socket, CAT5 cable lead, etc.) in the context of their location The created PDPT Framework is based on a model in building environment. This contextual information is the of location-aware enhancement. This concept enables to same as location information about user track. The PDPT increase the real dataflow from wireless AP (server side) to Core controls data, which are copied from the server to PDA (client side). The fact that the throughput (Table 1)is the PDA client by context information (position info). Each low on wireless connected mobile devices is very important database artefacts must be saved in the database along the with regards to the idea of using a prebuffered data for position information to which it belongs. increasing transfer speed through WiFi connection on PDA During the process of creating of a PDPT Framework the mobile devices. new software application called “Data Artefacts Manager” The general principle of localization states that if a WiFi- was developed (Figure 3). The manager is also described in enabled mobile device is close to such a stationary device— [14]. This application manages the artefacts in WLA database base station, it may “ask” the provider’s location position by (localization oriented database). The user can set the priority, EURASIP Journal on Wireless Communications and Networking 5

PDPT client

Start/stop PDPT User input (start/stop)

No PDPT active? Artefacts pushing to Yes PDA buffer Sending of PDA buffer image to PDPT core Figure 3: PDPT Framework Data Artefact Manager—this software substitutes a live connection to information system if it does not exist. PDPT server web service location, and other metadata of the artefact. This manager Static/dynamic area definition substitutes the online conversion mechanism, which can No Actual position of PDA transform the real online information system data to WLA Buffer PDA = predicted buffer database data artefacts during the test phase of the project. This manager can be also used in case of offline version of Yes PDPT Framework usage. Predicted position of The Manager allows to the administrator to create a PDA calculation new artefacts from multimedia files (image, video, sound, Static/dynamic enhanced area etc.) and edit or delete the existing artefacts. The left side definition of the screen contains the text field of artefact metadata as No Position of PDA Buffer PDA = predicted buffer a position in 3D space. This position is determined by the artefact size (in case of building plan) or by binding of the artefact to some part of a building in 3D space. It is possible Yes to take the 3D axis from a building plan by a GIS software like Quantum GIS or by own implementation [15, 16]. The Wait 1–10 seconds central part represents a multimedia file and the right side contains the buttons to create, edit, or delete the artefact. The lower part of the application screen shows the actual artefacts Figure 4: PDPT Framework—UML Design of flow diagram. The artefacts collections which belong to the actual user position are in WLA database located on MS SQL Server. copied to PDA buffer firstly. The predicted position and new artefacts collection is used if all artefacts are already loaded to PDA 3.2. The PDPT Framework Design. The PDPT framework buffer. The definition of the Static or Dynamic area depends on the design is based on the most commonly used server-client specific case of artefacts size, area size of where the PDPT framework architecture. To process data the server has an online is equipped. connection to the information system. Technology data are continually saved to SQL Server database [17]. A part of this database (desired by user location or his demand) is replicated online to client’s PDA, where it is 3.3. PDPT Client. The PDPT Client is a Windows Mobile visualized on the screen. The user’s PDA has a location sensor 6.1-based application. The PDPT Client was developed for component, which is continuously sending the information testing and tuning the PDPT Core. This client realizes a about nearby AP’s intensity to the framework kernel. The classic client to server side and an extension by PDPT and kernel processes this information and makes a decision as Locator module. to which part of MS SQL Server database will be replicated Figure 5 shows a screenshot from the mobile client. The (pushed) to client’s MS SQL Server CE database. The kernel figure shows the typical view of the data presentation from decisions constitute the most important part of the whole MS SQL CE database to the user (in this case the Ethernet framework as the kernel must continually compute the plan of the current area). Each process running in a PDPT position of the user and track and predict the user’s future Client is measured in a millisecond resolution to provide a movement. After making this prediction, the appropriate feedback from a real situation. The time window is in the data (part of MS SQL Server database) are prebuffered to upper right side of the screen (Figure 5). The user can select the client’s database for the future possible requirements an artefact from prebuffered artefacts to view. Unfortunately, (Figure 4). The PDPT framework server is created as a if the requested artefact does not exist in the PDA memory Microsoft web service to act as bridge between the MS SQL buffer, the online connection to the server must be used to Server and PDPT PDA Clients. select and download them online. 6 EURASIP Journal on Wireless Communications and Networking

Figure 5: PDPT Client. View of prebuffered artifacts (building Figure 7: PDPT Client. On PDPT tab is a summary of prebuffered computer network plans) from Microsoft SQL Server CE database. artefacts history and a part and full time of prebuffering. The upper part of screen is only for testing the functionality.

Figure 6: PDPT Client. On Locator tab is possible to set an interval for scanning of WiFi network neighbour. Locator time displays a real interval between a sending of neighbour WiFi AP summary and Figure 8: PDPT Client. The DB tab allows managing all necessary the reply from server. things from creation of database to Compact or Shrink.

Tabs Locator (Figure 6) and PDPT (Figure 7)presentsa this position can be used by the PDPT Core to prebuffer the way to tune the settings of PDPT Framework. In the first case data to the client device. (Figure 5), the user must turn on Locator check-box which The middle section of the PDPT tab (Figure 7) shows means that the continual measurement of WiFi signals of logging info about the prebuffering process. The right side nearby APs (time of these operations is measured in Locator shows the time of artefact loading (part time) and the full Time text window) will start. time of prebuffering. The info about nearby APs is sent to the PDPT Server which responds with a number of recognized APs in the 3.4. PDPT Client—Microsoft SQL Server CE Database. ADB database (Locator AP ret. Text window). In the presented manager for managing a database file on the PDA device was case, the 7 APs are in user neighbourhood, but only 2 APs are created (Figure 8). The first combo box menu on this tab recognized by the PDPT Server database (info about 2 APs is deals with IP address settings of the PDPT Server. DB Buffer in WLA database). The scanning interval is set to 2 seconds size follows on the second combo box. This size is important and finally the text “PDPT Server localization OK” means for maximum space taking by prebuffering database on that the user PDA was localized in an environment and that selected data media. Data medium can be selected on DB EURASIP Journal on Wireless Communications and Networking 7

Table 6: PDPT Framework—Final Data transfer tests. Execution Move Quality of Test no. C Time (min:s) speed (m/s) prebuﬀering (%) 1 1 : 40 1,32 25 2 1 : 48 1,22 18,18 B 3 3 : 37 0,61 75

4 3 : 24 0,65 36,36 A 5 5 : 12 0,42 97 6 5 : 26 0,40 54,55

Storage combo box. To check if a database exists, the SQLCE DB Exist button must be pressed. For example the db is ready D F means that the database file exists in a selected location. If such db file does not exist, the execution of SQL CE DB Delete & Create must be done. This button can be used for recreating of db file. Compact and shrink of DB file means two options for E manual database compression. The time in millisecond is GP measured in a text box in between the two buttons. Both of these mechanisms are used in prebuffering cycles when the large artefact is deleted from database table to release space of deleted artefact. The database file has occupied space of deleted artefact by default, because the standard operation of NK (m) delete order does not include this technique. This is due to 02550 recovery possibilities in Microsoft SQL Server CE databases.

Figure 9: VSB Technical University of Ostrava—University Cam- 4. Experiments pus Map. For a mobile device to determine its own position, it must have a WiFi adapter still alive. This fact provides a limitation of using of mobile devices. The complex test with several A special Biotelemetry system for patient monitoring types of batteries is described in [18]. is under development at our department. In this complex A number of indoor experiments were achieved with the system the wide network of remote sensors is used to PDPT framework using the PDPT Client application. The collect data. This system proved to be a useful platform for main result of the use of the PDPT framework is a reduction prebuffering the large data-artefacts [19, 20]. Localization of data transfer speed. The tests focused on the real use of module of PDPT framework is suitable for home security the developed PDPT Framework and its main impact on system [21]. For any kind of emergency cases, the special increased data transfer. wireless network MANET can be suitable improvement of A realization of tests consists of a user movement from PDPT solution to avoid any problems in case the signal of a sample location NK to C at a predefined direction. See preferred WiFi network is missing [22]. the university campus map (Figure 9) where the tests were realized. For the purposes of the test, five mobile devices were selected with different hardware and software capabilities. Six 5. Conclusions types of test batches were executed in the test environment. Each test was between two points of the testing environment The problem of low transfer rates in mobile devices was (building NK and C) with 132 meter distance. Every other presented. Some suggestions have been put forward (e.g., to test was in reversed direction. Five iterations (five devices use a high-performance SD Cards for large data amount to used) were performed in each batch. get a higher transfer rate). The low transfer rates problem Results (Table 6) provide a good level of usability when was considered also in the context of a maximum response user is moving slowly (less than 0,5 m/s). This is caused by time for user requests. a low number of visible WiFi APs in the test environment, The PDPT Framework was described as one of the where for 60% of total time only 1 AP was visible, 20% 2 possible solutions. The indoor location of a mobile user is visible, and 5% 3 or more visible WiFi APs. 15% of time obtained through an infrastructure of WiFi APs. This mech- represents a time without any WiFi connections. The values anism measures the quality of the link of nearby location of prebuffering quality achieved in such case are very good. provider APs to determine the actual user position. User 8 EURASIP Journal on Wireless Communications and Networking location is used in the core of server application of the PDPT networks,” in Personal Wireless Communications, vol. 245 of framework to data prebuffering and pushing information IFIP International Federation for Information Processing,pp. from the server to the user’s PDA. Data prebuffering is 423–432, Springer, New York, NY, USA, 2007. the most important technique to reduce the time from a [13]P.Brida,N.Majer,J.Duha,andP.Cepel,“AnovelAoA user request to system response. The experiments show that positioning solution for wireless ad hoc networks based on the location determination mechanism accurately and with six-port technology,” in Wireless and Mobile Networking, vol. 308 of IFIP International Federation for Information Processing, sufficient quality determines the actual location of the user pp. 208–219, Springer, New York, NY, USA, 2009. in most cases. Minor inaccuracies do not impact significantly [14] O. Krejcar and J. Cernohorsky, “Database prebuffering as a on the PDPT Core decision making. The framework was way to create a mobile control and information system with evaluated in a real use experiment. better response time,” in Proceedings of the 8th International Conference on Computational Science (ICCS ’08), vol. 5101 Acknowledgment of Lecture Notes in Computer Science, pp. 489–498, Krakow,´ Poland, June 2008. This research has been carried out under the financial sup- [15] J. Horak, J. Unucka, J. Stromsky, V. Marsik, and A. Orlik, port of the research grant “Centre for Applied Cybernetics,” “TRANSCAT DSS architecture and modelling services,” Con- Ministry of Education of the Czech Republic under Project trol and Cybernetics, vol. 35, no. 1, pp. 47–71, 2006. [16] J. Horak, A. Orlik, and J. Stromsky, “Web services for 1M0567. distributed and interoperable hydro-information systems,” Hydrology and Earth System Sciences, vol. 12, no. 2, pp. 635– References 644, 2008. [17] M. Jewett, S. Lasker, and S. Swigart, “SQL server everywhere: [1] F. Evennou and F. Marx, “Advanced integration of WiFi and just another database? Developer focused from start to finish,” inertial navigation systems for indoor mobile positioning,” DR DOBBS Journal, vol. 31, no. 12, 2006. EURASIP Journal on Applied Signal Processing, vol. 2006, [18] O. Krejcar, “PDPT framework—building information system Article ID 86706, 11 pages, 2006. with wireless connected mobile devices,” in Proceedings of [2]V.M.Olivera,J.M.C.Plaza,andO.S.Serrano,“WiFi the 3rd International Conference on Informatics in Control, localization methods for autonomous robots,” Robotica, vol. Automation and Robotics (ICINCO ’06), pp. 162–167, Setubal, 24, no. 4, pp. 455–461, 2006. Portugal, August 2006. [3] A. Kostuch, K. Gierłowski, and J. Wozniak, “Performance [19] M. Penhaker, M. Cerny, L. Martinak, J. Spisak, and A. analysis of multicast video streaming in IEEE 802.11 b/g/n Valkova, “HomeCare—smart embedded biotelemetry sys- testbed environment,” in Wireless and Mobile Networking, tem,” in Proceedings of the World Congress on Medical Physics vol. 308 of IFIP Advances in Information and Communication and Biomedical Engineering (WC ’06), vol. 14, pp. 711–714, Technology, pp. 92–105, Springer, New York, NY, USA, 2009. Seoul, South Korea, August-September 2006. [4] J. Nielsen, Usability Engineering, Morgan Kaufmann, San [20] M. Cerny and M. Penhaker, “Biotelemetry,” in Proceedings of Francisco, Calif, USA, 1994. the 14th Nordic-Baltic Conference on Biomedical Engineering [5] M. Haklay and A. Zafiri, “Usability engineering for GIS: and Medical Physics (NBC ’08), vol. 20, pp. 405–408, Riga, learning from a screenshot,” The Cartographic Journal, vol. 45, Latvia, June 2008. no. 2, pp. 87–97, 2008. [21] V. Kasik, “FPGA based security system with remote control [6]D.F.Galletta,R.M.Henry,S.McCoy,andP.Polak,“When functions,” in Proceedings of the 5th IFAC Workshop on the wait isn’t so bad: the interacting effects of website delay, Programmable Devices and Systems (PDS ’01), pp. 277–280, familiarity, and breadth,” Information Systems Research, vol. Gliwice, Poland, November 2001. 17, no. 1, pp. 20–37, 2006. [22] T. A. Ramrekha and Ch. Politis, “An adaptive QoS routing [7] O. Krejcar, “User localization for intelligent crisis manage- solution for MANET based multimedia communications in ment,” in Proceedings of the 3rd IFIP Conference on Artificial emergency cases,” in Proceedings of the 1st ICST International Intelligence Applications and Innovation (AIAI ’06), pp. 221– Conference on Mobile Lightweight Wireless Systems (MOBI- 227, Athens, Greece, June 2006. LIGHT ’09), vol. 13 of Lecture Notes of the Institute for [8] A. Asaduzzaman, I. Mahgoub, P. Sanigepalli, H. Kalva, R. Computer Sciences, Social Informatics and Telecommunications Shankar, and B. Furht, “Cache optimization for mobile Engineering, pp. 74–84, Athens, Greece, May 2009. devices running multimedia applications,” in Proceedings of the IEEE 6th International Symposium on Multimedia Software Engineering (ISMSE ’04), pp. 499–505, December 2004. [9] S. Rosner, M. Mcclain, and E. Gershon, “System and method for improved memory performance in a mobile device,” US patent, Spansion LLC, 20060095622, 2006. [10] G. P. Brasche, R. Fesl, W. Manousek, and I. W. Salmre, “Location-based caching for mobile devices,” US patent, Microsoft Corporation, Redmond, Wash, USA, 20070219708, 2007. [11] R. F. Squibbs, “Cache management in a mobile device,” US patent, Hewlett-Packard Development Company, L.P., 20040030832, 2004. [12] P. Brida, J. Duha, and M. Krasnovsky, “On the accuracy of weighted proximity based localization in wireless sensor Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 821846, 8 pages doi:10.1155/2009/821846

Research Article A Potential Transmitter Architecture for Future Generation Green Wireless Base Station

Vandana Bassoo,1 Kevin Tom,1 A. K. Mustafa,1 Ellie Cijvat,2 Henrik Sjoland,2 and Mike Faulkner1

1 Centre for Telecommunications and Micro-Electronics, Footscray Park Campus, Victoria University, P.O. Box 14428, 8001 Melbourne, Australia 2 Department of Electrical and Information Technology, Lund University, Sweden

Correspondence should be addressed to Vandana Bassoo, [email protected]

Received 30 March 2009; Accepted 14 August 2009

Recommended by Naveen Chilamkurti

Current radio frequency power amplifiers in 3G base stations have very high power consumption leading to a hefty cost and negative environmental impact. In this paper, we propose a potential architecture design for future wireless base station. Issues associated with components of the architecture are investigated. The all-digital transmitter architecture uses a combination of envelope elimination and restoration (EER) and pulse width modulation (PWM)/pulse position modulation (PPM) modulation. The performance of this architecture is predicted from the measured output power and efficiency curves of a GaN amplifier. 57% efficiency is obtained for an OFDM signal limited to 8 dB peak to average power ratio. The PWM/PPM drive signal is generated using the improved Cartesian sigma delta techniques. It is shown that an RF oversampling by a factor of four meets the WLAN spectral mask, and WCDMA specification is met by an RF oversampling of sixteen.

Copyright © 2009 Vandana Bassoo et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction technology. The goal here is to replace the power connection (or diesel generator in developing countries) with solar The existing third generation network deployed around the cells. This is only practical if the overall basestation power world is not sufficient to meet the needs of new upcoming consumption can be reduced to realistic levels. At present, bandwidth intensive applications. It is expected that by the radio frequency power amplifier (PA) is the largest power 2012, two-thirds of the estimated 1.8 billion broadband usage component, accounting for approximately 30%–40% users will be connected through a wireless device [1]. of a 3G wireless base station’s total consumption. High A fourth generation wireless system will be required to power usage leads to higher costs and an unacceptable transmit higher bit rates to more users in a more flexible environmental impact. way. This will lead to higher transmission bandwidths and Efficient multicarrier schemes such as orthogonal fre- greater transmission powers. Transmission efficiencies must quency division multiplexing (OFDM) are required to therefore increase if the equipment is to be housed in cater for the increasing transmission bandwidths. OFDM existing shelters, using existing cooling (air-conditioning) signals require conventional linear PAs. A tradeoff between and existing power sources. efficiency and linearity always exists in PA design. A typical Mobile operators around the world are becoming feed-forward class AB PA has efficiency in the range of 10% increasingly interested to reduce their operating costs and with modulated signals. Alternatively, a class E switched carbon footprint. Developing countries are particularly mode PA has the potential to attain drain efficiencies of up susceptible because of the low income of their subscribers. to 45% when operated with 8 dB peak to average Rayleigh Therefore, base station manufacturers around the world enveloped signal [2]. Switch mode power amplifiers (SMPAs) are looking for solutions to develop “greener” and efficient are highly nonlinear and attain maximum efficiency if driven 2 EURASIP Journal on Wireless Communications and Networking by a pulse waveform. The challenge is on the modulation component. Section 2 compares the efficiency of a GaN stage to generate the appropriate drive signal which is amplifier operating in two modes; the first mode uses an optimised for efficiency and linearity. This has lead to a approximate PWM gate drive signal and the second mode renewed interest in architectures such as linear amplification uses a high level EER drive on the Vdd supply line. Section 3 with nonlinear components (LINCs), envelope elimination introduces a new technique for bandwidth reduction of the and restoration (EER), and pulse width modulation (PWM) polar envelope drive signal and then predicts the efficiency [3–5]. This paper considers a combined EER and PWM based on the amplifier measurements. Section 4 discusses the amplifying scheme. use of (ΣΔ) to generate a suitable amplifier input pulse train The continuing improvement in silicon technology is with phase and amplitude information encoded in its pulse enabling digital circuits to operate at higher frequencies width and pulse position. Section 5 concludes by discussing suggesting that direct digital generation of the amplifier’s the viability of the new architecture. RF drive signal is now possible. Figure 1(top) shows the traditional wireless base station architecture and a proposed architecture (Figure 1(bottom))forafuturegenerationof 2. Switch Mode Power Amplifier wireless base stations using SMPAs (e.g., classes D and E). In this section, we predict from measurements an amplifier’s The new architecture eliminates most analog components in performance in two operating modes. The first mode uses the driver circuit. These include digital to analog converters a PWM [6] drive signal and the second mode uses the (DACs), reconstruction filters, the local oscillator, and the more conventional polar EER drive signals. The operating quadrature modulator. The new all-digital architecture can frequency of the power amplifier is designed to be 400 MHz. thus be easily integrated on chip. A discrete GaN HEMT device is chosen because it has good The EER-driven PA architecture, Figure 1(middle), has gain at high power and a lower drain-source capacitance ffi a theoretical e ciency of 100% and operates by splitting than competing LDMOS devices which should make it more the signal into polar components (envelope and phase). The suitable for high efficiency in switch-mode operation [7]. envelope, u, controls the SMPA supply and the phase, s, controls the pulse position modulator of the input pulses which form the RF drive signal. This is a constant envelope 2.1. PA with PWM Drive and EER Drive. A simple way modulated signal that keeps the amplifier operating in its to obtain PWM operation is to use a comparator and a most efficient saturated mode. However, this structure suffers triangular drive signal with the reference input controlled by from bandwidth expansion when used with OFDM signals. the envelope modulation (Figure 3). In this work the PWM OFDM is a noise-like signal with a random path in the signal is directly generated in the output power device to I and Q plane, any near zero crossings cause large dips avoid the need for special wideband input matching circuits. in the envelope signal and a large rate of phase change The amplifier input is overdriven with the phase modulated (instantaneous frequency) for the phase drive signal. The sinusoidal carrier signal (Vp), and the gate bias (Vbias)is coordinate transform of the OFDM signal from Cartesian controlled by the envelope component of the modulated to polar therefore results in envelope and phase components signal. Since the device threshold voltage (VT )isfixed, ff ff that have wider bandwidths than that of the input signal. this has the e ect of varying the on/o duty cycle of the Unfortunately, the bandwidth of the polar envelope compo- amplifier, as illustrated in Figure 4. The PWM like signal nent cannot be too high since sharp dips in the magnitude at the drain of the device has a slightly reduced slew rate cannot easily be reproduced by the switch mode DC- because of the limited gain of the device. Also, the pulse DC converter. We therefore propose a technique to reduce width is no longer linear with respect to the envelope signal, = − the envelope’s bandwidth expansion without distorting the ΔV (VT Vbias), because of the nontriangular drive signal. output signal (Figure 1(bottom)). Bandwidth reduction is The latter indicates the need for linearisation. The output possible by allowing some amplitude variation in the phase tuning network provides filtering of the pulsed drain signal signal, s, driving the amplifier. The amplitude variation can and performs the impedance transformation from the load be impregnated onto the phase modulated pulse position impedance to the drain of the device. The output and input networks were optimised for a range of pulse widths by modulation (PPM) signal, sΣΔ, by applying pulse width modulation (PWM) on the PPM pulses of the input RF drive simulation using the harmonic balance method of ADS [6]. signal (Figure 2). This maintains the all-digital architecture. The same amplifier configured as an EER PA structure is If synchronous digital circuits are to be used the pulse edges shown in Figure 5. Here the low frequency envelope signal must occur on the timing grid of the digital clock. This limits modulates the PA supply (Vdd). the selection of possible pulse positions and pulse widths leading to time quantisation. Sigma delta (ΣΔ)converters 2.2. Measurement Results. The full amplifier including can generate the two state PWM/PPM signal, while shaping matching networks is implemented using surface mount the time quantisation noise away from the band of interest. components on a standard FR4 PCB, with double sided In this paper, we measure the output power and efficiency copper layers as illustrated in Figure 6. The inductor at the of a GaN single stage amplifier and from these results drain of the device is implemented using a transmission we then predict the performance of the amplifier in the line. The device is a CREE CGH40010 discrete GaN HEMT new architecture. We show there is a tradeoff between device suitable for high output power (10 W), high efficiency amplifier efficiency and the bandwidth of the envelope (60%), and high frequency (6 GHz) operation. The threshold EURASIP Journal on Wireless Communications and Networking 3

LPF I DAC Digital IQ Filter Linear Q predistortion mod PA DAC LPF

Tx Lo

Switch mode DC-DC modulator u I Cartesian to s PPM SMPA BPF Q polar

Switch mode DC-DC modulator u I Signal s ΣΔ drive sΣΔ SMPA BPF Q conditioning PWM/PPM

Figure 1: Traditional (top), EER (middle), and proposed (bottom) transmitter architectures.

Waveform after PPM Envelope −

Reference (a) waveform

Waveform Envelope after PWM

Figure 2: Demonstrating PWM (amplitude modulation) and PPM (phase modulation) concept. Edges occur on the digital timing grid. Phase

(b) voltage is −2.5 V, and the nominal supply voltage is 28 V [7]. Figure 3: PWM generation: (a) ideal, (b) PA schematic. The carrier frequency was 395 MHz and a high-power input signal was used (23 dBm), since the PA was to operate in switch-mode. VT Vp In Figure 7(a), the efficiency and output power as a ΔV function of Vbias are shown. The figure shows that the useful Vbias range of gate bias voltage is roughly 4 V, from −5Vto−1V. This results in an output power variation from 19 dBm to 39.6 dBm, and a drain efficiency variation from 6% to 61%. The loss of efficiency at low output powers is consistent with PWM operation, where slew-rate losses are essentially On constant whatever the pulse width (output power). Off The EER architecture was not assembled fully, but measurements were made on the PA varying the supply Figure 4: Operating principle of the power amplifier. The difference voltage (Vdd) and keeping the gate bias voltage constant. The ΔV between Vbias and VT varies, depending on Vbias and the efficiency and output power are plotted in Figure 7(b). Vdd resultant PWM drain signal is illustrated. 4 EURASIP Journal on Wireless Communications and Networking

70 Envelope 60

Phase 50 Eﬃciency ciency (%)

ﬃ 40

Figure 5: EER architecture. 30 Pout

(dBm) and e 20 out P 10

0 −5 −4.5 −4 −3.5 −3 −2.5 −2 −1.5 −1

Vgs (Volts) Transmission line (a) Vbias 70 RF output 60

ﬃ 50 E ciency ciency (%)

RF ﬃ 40 input 30

Pout Figure 6: The PCB with the transmission line inductance and I/O (dBm) and e 20 ports indicated. out P 10 scales the output signal, for example, doubling the supply 0 voltage will increase the amplifier’s output by 6 dB. 51015202530 Comparing the efficiency performance of both the PWM Vdd (Volts) and EER PA architectures, it can be noticed that with (b) the PWM structure the efficiency holds good only over a Figure 7: Measurement results. Output power and drain efficiency small dynamic range, within less than 3 dB of peak output for (a) PWM-like operation and (b) EER operation. power. Therefore, to maintain high efficiency the amplitude variation needs to be small. On the other hand, the EER structure maintains efficiencies above 60% over a wide dynamic range (18 dB). Figure 8 restates the measured results high. Therefore, a suitable bandwidth limitation scheme in a way that gives a more insightful comparison. The plot ffi needs to be applied to the envelope component in order to shows the e ciency versus normalised output power for maintain the supply modulator efficiency. The next section both architectures. The power is normalised to the amplifier’s ff = describes the trade-o between envelope bandwidth and peak output power; in this case 39.6 dB for Vdd 30 V. The amplifier efficiency. dotted PWM curve is for measurement data with Vdd = 10 V. It has the same basic shape as the Vdd = 30 V curve, except for a small efficiency scaling due to the improvement 3. Bandwidth Limitation in efficiency at lower Vdd’s (Figure 7(b)). The similarity between the normalised Vdd curves will be used later in the In the EER architecture, Figure 1(middle) generates the paper to calculate the expected efficiency of amplifiers using envelop signal for the DC to DC converter and a constant a combination of PWM and EER modulation. amplitude phase modulated signal for the amplifier input Although, the EER architecture gives better efficiency RF drive. The coordinate transform from Cartesian to polar with modulated signals it still requires a supply modulator, results in signal components that have wider bandwidth than commonly implemented as a class-S amplifier. The switching that of the original input signal. We propose a technique frequency of the supply modulator must be considerably to reduce the bandwidth expansion of the envelope polar larger than the envelope bandwidth to minimise distortion; drive signal by using both EER and PWM/PPM modula- this increases the switching losses when the bandwidth is tion as shown in Figure 9. The amplifier effectively scales EURASIP Journal on Wireless Communications and Networking 5

70 vectormagnitude(EVM);butatalevelof−38 dB, this is unlikely to cause a problem. When compensation is applied 60 to the input signal, s, its bandwidth reduces (Figure 10 dotted), and the blowout in the RF spectrum is repaired, 50 as is the EVM. The new architecture leads to envelope and input bandwidths being reduced to 28% of the original 40 EER bandwidths, when measured at a −50 dB threshold. Even though the RF signal suﬀers no EVM or adjacent ciency (%) 30

ffi channel interference (ACI), when the envelope bandwidth E is limited, the amplifier efficiency is compromised. We use 20 the measured curves of Figure 8 to predict the amplifier efficiency when it passes an OFDM signal. This noise-like 10 signal has a wide dynamic range and in the simulations amplifier clipping occurs when R + dc < |x|,orif|s| > 1. 0 Low DC values increase clipping but are best for efficiency. 00.10.20.30.40.50.60.70.80.91 For each bandwidth limit we choose the DC value to give the Normalised P out same clipping energy as a normal OFDM signal clipped to

Figure 8: Efficiency against power output. PWM curves are for Vdd 8 dB PAPR (peak to average power ratio). The clipping noise = 30 V (solid) and for Vdd = 10 V (dashed) and EER (solid). for each simulation is therefore the same. Figure 11 shows the predicted efficiency versus envelope bandwidth plots for OFDM modulation. When there is no envelope bandwidth limitation the amplifier structure works (multiplies) the PWM/PPM output by the instantaneous as EER and the efficiency is at 62%. However, when the Vdd. bandwidth limitation is at 0 Hz, that is, the envelope has The process first limits the envelope bandwidth, R,ina only a fixed dc value, the efficiency is at 28% which is the low pass filter and then adds a DC offset to stop clipping efficiency of the PWM amplifier by itself. A further 10% at the amplifier. The u = R + dc signal acts as the efficiency can be gained with the envelope filtered to 0.25 supply modulator for the amplifier. The phase information channel bandwidths. The optimal point lies at the knee of is obtained by dividing the input signal with u.Here, the curve at 0.75 channel bandwidths which yields a 57% s = x/(R + dc) is the drive signal for the PA. The ΣΔ efficiency. operation converts the magnitude/phase information of s into an equivalent PWM/PPM pulse driver signal sΣΔ for the amplifier. In the case of no bandwidth limitation (i.e., 4. Pulse Train Generation Using LPF bandwidth = ∞ Hz), the input drive, s, is a constant Sigma Delta Modulation Technique magnitude signal containing phase information only. The architecture then works as EER. On the other hand, if there Figure 12 shows the block diagram of an architecture is full bandwidth limitation (LPF bandwidth = 0 Hz), u proposed in [8]whichwerefertoasCartesianΣΔ.It contains only a fixed DC value and s is the original input can generate a PWM/PPM pulse train with the appropri- signal, x. Limiting the envelope bandwidth and allowing ate phase and amplitude information, while having full some envelope variation into the phase drive signal, s,result compatibility with synchronous digital circuit design. It in reduced bandwidth expansion for both envelope and consists of two first-order lowpass sigma delta modulators phase drive signals. The input drive, s, is scaled multiplied (MOD 1 [9]), amplitude and a phase quantisers, and a by the envelope signal, u, within the amplifier to recreate, x, polar to “PWM/PPM” block. The Cartesian signals pass at the amplifier output. through ΣΔ filters, after which they are converted to polar Figure 10 shows the spectra of the OFDM RF signal, [R, θ] for quantisation in blocks QR and Qθ. The quantised x, and its polar drive components s and R. The effect of signals [R, θ] are then reconverted to Cartesian before being envelope bandwidth limitation is shown with and without fedbacktotheΣΔ filters. ΣΔs shape the quantisation compensation of the phase drive signal. The thin solid noise away from the signal band. The previous reported lines show the RF and polar components, when there is no structures perform ΣΔ filtering on the polar signal [10], bandwidth limitation. This is the EER condition and clearly while this work gives superior performance because the shows the bandwidth expansion of the envelope and input ΣΔ filtering is on the Cartesian signal where there is no signals. bandwidth expansion. The equations from [11]areusedto The dashed line, x, shows the blow out of the RF spec- decide the amplitude threshold levels as the quantisation is trum when the envelope (R, shown dotted) is bandwidth performed in the polar domain. In this case, the amplitude is limited, to one OFDM channel without applying the corre- quantised into (n/2 + 1) levels corresponding to pulse widths sponding compensation. Adjacent channel power exceeding (0, 2/n,4/n,6/n···(n/2)/n)(1/fc)(fc = carrier frequency) −40 dBc will exceed the spectrum transmission mask for and the phase is quantised into n phase increments from zero most standards (WLAN@ −40 dB and UMTS@−50 dB). The to 2π. This quantisation process requires the system digital OFDM inband signal is also affected by a build up in error clock to oversample fc by a factor of n ( fclock= nfc). 6 EURASIP Journal on Wireless Communications and Networking

R u = R + dc x Cartesian to BW limit + polar clk ∗ OFDM sΣΔ x = u s

s x/(R + dc) ΣΔ

Figure 9: The new transmitter architecture. The ΣΔ clk operates at the carrier frequency and gives an amplitude and phase output for each RF cycle.

0 A simulated OFDM signal was used to test the system described. In the spectral domain, the noise in the adjacent −10 channels needs to be below an acceptable level. The out-of- s band distortion products are measured by calculating the −20 R adjacent channel power (ACP) ratio, deﬁned as the noise power in the adjacent channel over signal power. Figure 13 −30

Power (dB) x shows a spectrum plot of the Cartesian sigma delta at an − −40 input power of 7 dB. ACP1 indicates the first adjacent channel and ACP2 indicates the second adjacent channel. −50 The shaping effect of the ΣΔ filters can be observed from the 00.511.522.53gradual increases in noise power as we move away from the Relative BW band of interest. The output band-pass filter will limit any BW =∞ BW = 1 unwanted noise far away from the desired band. EER RF out (x) RF out (BW limited envelope) Since the oversampling factor, n, has a direct impact on EER envelope (R) BW limited envelope (R) possible carrier frequencies, it is crucial to keep n as low EER phase (s) Compensated phase (s) as possible. Figure 14 shows a plot of ACP versus n.The Figure 10: Spectra of the input drive signal, s, envelope signal, R, appropriate n can be chosen depending on the spectrum andRFsignal,x. mask of the standard. The Cartesian ΣΔ clearly outperforms the polar ΣΔ, leading to a lower n requirement for the same ACP. The results shown here for n = 4 are reasonable for the WLAN standard (ACP < −40 dB). However a higher oversampling rate of n = 16 is required to meet the tougher −50 dB WCDMA spectrum mask, limiting fc to about 40 200 MHz using current digital technology ( fclock = 3.2 GHz). Clock rates must further improve before carrier frequencies in the cell phone bands can be handled. ciency (%) ffi E 30 5. Conclusion ffi 00.511.5 Ahighe ciency all-digital transmitter architecture that Envelope BW (channels) uses a combination of EER and PWM/PPM modulation is described. We predicted the performance of this architecture Figure 11: Average efficiency versus envelope bandwidth for an from the measured output power and efficiency curves of a OFDM signal. GaN amplifier. An OFDM signal limited to 8 dB PAPR will give 57% efficiency which compares favourably to the peak amplifier efficiency of 64% with a continuous wave signal (Figure 7). The “polar to PWM/PPM” block is responsible for A number of challenges still exist before a high efficiency upconverting the output of the quantisers to RF. The all-digital transmitter can be realised at high carrier frequen- quantised phase, θ, determines the pulse position and the cies. From the amplifier perspective parasitic capacitances quantised amplitude, R, determines the pulse duration [8]. will always limit the efficiency of the PWM/PPM mode, The output of the “polar to PWM/PPM” block is a pulse since the wide bandwidths of switching waveforms make train to be fed to the switch mode PA and band-pass filtered resonating out unwanted capacitances somewhat difficult. to eliminate quantisation noise and out-of-band distortion The situation should improve at lower carrier frequencies. products. The EER mode has excellent efficiency for most amplifier EURASIP Journal on Wireless Communications and Networking 7

1/(nfc)

fs = 2 fc

I R I R ΣΔ ﬁlter QR Polar C P to to to PWM/ P C Q θ PPM ΣΔ ﬁlter Qθ SMPA BPF@ fc

θ Q Figure 12: Block diagram of cartesian sigma delta.

0 classes, and the amplifier described here is no different. The major limitation of EER is the envelope bandwidth. −10 We have not considered the DC/DC converter in this work, but we have shown that most of the EER efficiency −20 can be conserved if the envelope bandwidth is limited to no less than 0.75 channel bandwidths. The ACI and ACP1 ACP2 −30 EVM distortion that this causes in the RF output can be removed by varying the amplitude (in this case using PWM) of the phase modulated input drive signal. Further −40 Average spectrum reduction in the envelope bandwidth requires the PWM waveform to have high efficiency over a greater dynamic − 50 range. The PWM/PPM amplifier drive signal can be directly −60 850 900 950 1000 1050 1100 1150 1200 formed from the I and Q digital baseband signals using ΣΔ Frequency techniques to mitigate time quantisation of the pulse edges. The improved performance of the Cartesian filtering in the Figure 13: Spectrum plot of Cartesian sigma delta, n = 16, fc = ΣΔ enables coarser time quantisation of the RF carrier signal. 1024 MHz, OFDM channel bandwidth = 20 MHz. Even so, the digital clock must oversample the RF carrier by a factor of four to meet the WLAN spectral mask and by at least a factor of sixteen to meet the more demanding WCDMA specification. Even though the CREE amplifier is −20 good to 6 GHz [7], it is the digital modulating circuits that limit the maximum carrier frequency. − 25 Polar ΣΔ −30 References

−35 [1] Ericsson, “Sustainable energy use in mobile communications,” August 2007, http://www.ericsson.com/campaign/sustainable −40 mobile communications/downloads/sustainable energy.pdf. ACP (dB) [2]K.Tom,M.Faulkner,andT.Lejon,“Performanceanalysis −45 of pulse width modulated RF class-E power amplifier,” in Proceedings of the IEEE 63rd Vehicular Technology Conference −50 (VTC ’06), vol. 4, pp. 1807–1811, Melbourne, Australia, May Cartesian ΣΔ 2006. −55 [3] A. Diet, C. Berland, M. Villegas, and G. Baudoin, “EER 0 5 10 15 20 25 30 architecture specifications for OFDM transmitter using a class Oversampling factor, n E amplifier,” IEEE Microwave and Wireless Components Letters, vol. 14, no. 8, pp. 389–391, 2004. ACP1 [4] C. Berland, I. Hibon, J. F. Bercher, et al., “A transmitter ACP2 architecture for nonconstant envelope modulation,” IEEE Figure 14: Plot of ACP (dB) against oversampling rate, n. The first Transactions on Circuits and Systems II, vol. 53, no. 1, pp. 13– and second adjacent channels are shown. 17, 2006. 8 EURASIP Journal on Wireless Communications and Networking

[5] F. Wang, D. F. Kimball, J. D. Popp, et al., “An improved power- added efficiency 19-dBm hybrid envelope elimination and restoration power amplifier for 802.11g WLAN applications,” IEEE Transactions on Microwave Theory and Techniques, vol. 54, no. 12, pp. 4086–4099, 2006. [6] E. Cijvat, K. Tom, M. Faulkner, and H. Sjoland, “A GaN HEMT power amplifier with variable gate bias for envelope and phase signals,” in Proceedings of the 25th IEEE Norchip Conference (NORCHIP ’07), pp. 1–4, Aalborg, Denmark, November 2007. [7] Cree Inc., “CGH40010 product specification,” http://www.cree .com/products/pdf/CGH40010.pdf. [8] V. Bassoo and M. Faulkner, “Sigma-delta digital drive signals for switch-mode power amplifiers,” Electronics Letters, vol. 44, no. 22, pp. 1299–1300, 2008. [9] R. Schreier and G. C. Temes, Understanding Delta-Sigma Data Converters, Wiley-IEEE Press, New York, NY, USA, 2004. [10] J. Keyzer, R. Uang, Y. Sugiyama, M. Iwamoto, I. Galton, and P. M. Asbeck, “Generation of RF pulsewidth modulated microwave signals using delta-sigma modulation,” in Proceed- ings of the IEEE MTT-S International Microwave Symposium Digest, vol. 1, pp. 397–400, Seattle, Wash, USA, June 2002. [11] P. Wagh and P. Midya, “High-efficiency switched-mode RF power amplifier,” in Proceedings of the 42nd Midwest Sym- posium on Circuits and Systems, vol. 2, pp. 1044–1047, Las Cruces, NM, USA, 1999. Hindawi Publishing Corporation EURASIP Journal on Wireless Communications and Networking Volume 2009, Article ID 968323, 11 pages doi:10.1155/2009/968323

Research Article Modeling Energy Consumption of Dual-Hop Relay Based MAC Protocols in Ad Hoc Networks

Rizwan Ahmad,1 Fu-Chun Zheng,2 and Micheal Drieberg1

1 Centre for Telecommunications and Micro-Electronics, School of Engineering and Science, Victoria University, Melbourne, Victoria 8001, Australia 2 School of Systems Engineering, University of Reading, Whiteknights Reading, RG6 6AY, UK

Correspondence should be addressed to Rizwan Ahmad, [email protected]

Received 16 April 2009; Revised 21 September 2009; Accepted 25 November 2009

Recommended by Naveen Chilamkurti

Given that the next and current generation networks will coexist for a considerable period of time, it is important to improve the performance of existing networks. One such improvement recently proposed is to enhance the throughput of ad hoc networks by using dual-hop relay-based transmission schemes. Since in ad hoc networks throughput is normally related to their energy consumption, it is important to examine the impact of using relay-based transmissions on energy consumption. In this paper, we present an analytical energy consumption model for dual-hop relay-based medium access control (MAC) protocols. Based on the recently reported relay-enabled Distributed Coordination Function (rDCF), we have shown the eﬃcacy of the proposed analytical model. This is a generalized model and can be used to predict energy consumption in saturated relay-based ad hoc networks. This model can predict energy consumption in ideal environment and with transmission errors. It is shown that using a relay results in not only better throughput but also better energy eﬃciency.

Copyright © 2009 Rizwan Ahmad et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction which ensures a better quality of life and environment. For developed and developing countries which have made much Next generation networks (NGN) (or 4G, as they are better investment on existing networks and due to high number known) are to provide voice, data and multimedia access of users, it may take longer to completely migrate to next to users on an “anytime” and “anywhere” basis. This vision generation networks. However for underdeveloped countries of 4G forms the requirement to achieve high throughput, who still lacks infrastructure the adaptation may be easy but high energy efficiency and low latency to provide quality there are cost constraints for them as well. of service (QoS) and efficient utilization of the scarce Existing networks have many issues which when inte- bandwidth. Therefore 4G networks are expected to result in grated into the new networks will greatly influence the overall a better quality of life and environment. Another important performance. Therefore, much research currently is focused requirement for next generation networks is backward on improving performance of existing networks. This can compatibility with existing networks. In today’s world, due be achieved in numerous ways such as efficient algorithms, to reasons such as cost and lack of infrastructure, it may protocol modification, new protocols, and so forth. One not be possible for all to adapt new networks at once and such example of issues is from the existing 802.11 networks make the existing ones obsolete. This may happen eventually where the performance of the whole system degrades greatly with time; however, we expect to have a hybrid of the new once low data rate nodes become dominant. A solution to and existing networks for quite some time. It is of high address this issue comes from the advent of cooperative importance to have smooth interaction between existing communication [1, 2] in the form of relay based MAC networks and next generation networks. This intermediate protocols. This results in intermediate data rates to provide period where both networks will coexist is an important step higher throughput and capability to fight against the varying towards complete transition to next generation networks channel conditions. At MAC layer, cooperation can be 2 EURASIP Journal on Wireless Communications and Networking

to IEEE 802.11 DCF. Use of relay requires justification both from throughput and from energy perspectives. In addition Source Destination to this, another concern of importance is the impact of relay ffi (a) nodes on the energy e ciency as relay nodes will utilize their own energy reserve to help other nodes. The main contribution of this paper is the generalized analytical energy model for relay based MAC protocols in ideal channel, channel with transmission errors and the Relay decomposition of energy. In this paper we have used relay- enabled Distributed Coordination Function (rDCF) [6]as a case study to show the efficacy of the proposed energy Source Destination model. Therefore it is a matter of high importance for (b) MAC protocol developers to have an idea on the energy consumption while still in design phase. We have shown Figure 1: (a) Slow Single hop direct transmission, (b) Fast Dual- ff hop transmission via relay. results of energy consumption with di erent number of nodes and rate combinations for relay links. Furthermore we have shown the impact of variable packet length (expected incorporated by replacing slow single-hop transmission by payload) on energy consumption. Decomposition of energy fast dual-hop transmissions. This means that the source, after for various operations is also shown and will help in the acquiring the medium, transmits to a relay first at a higher design of energy efficient MAC protocols. This is particularly rate and the relay will then transmit to destination as shown useful for devising energy saving mechanisms and policies in Figure 1(b). This solution, although appropriate for the for existing and new protocols. The energy model will benefit throughput, has emerged in an era when the awareness to “go the application of relay based MAC protocols (e.g., rDCF) in green” is widely discussed. This has triggered a debate on the energy critical areas such as sensor networks and integration energy consumption which has now gained importance in with next generation networks. the minds of the MAC protocol developers. It is important to reiterate the fact that before the ultimate phase of complete convergence to 4G networks, a hybrid phase will dominate. 2. Energy Model Relay based MAC protocols are in its infancy and most of the In [7] authors have used Markov chain model of [8] current literature deals with the throughput improvement to show the energy consumption of relay based ad hoc gained by using these. There is only limited reporting in the networks. In this paper, we have proposed an energy model literature dealing with the energy issues in relation to relay to accommodate relay based transmissions by using an based MAC protocols. improved model proposed by Wu et al. [9]. This is a more Carvalho et al. [3] model node’s energy consumption accurate model for saturation throughput, which requires in a single-hop IEEE 802.11 ad hoc network as shown in the throughput model of [8] to incorporate a finite retry Figure 1(a). Carvalho calculated the average service time limit. We further improve it by considering transmission ofapackettransmittedinasaturatedadhocnetwork. errors. The following challenges have been addressed in the Results show that passive modes (idle, overhear, receive) paper: (1) how to incorporate relay nodes, (2) treatment of dominate the energy consumption and transmission of relay node, (3) how relay nodes differ from other nodes in large payloads is more advantageous. However, this model energy consumption behavior, and (4) impact on energy in treats receiving and idle state in the same way and gives the presence of transmission errors. no consideration for channel condition. Wang et al. [4] proposed a model for energy efficiency in IEEE 802.11 DCF and tried to maximize energy efficiency based on packet 2.1. System Model. We consider a wireless network of n size and contention window. They have considered channel nodes based on IEEE 802.11 MAC that can support multiple errors on the data packet only, which is not reflective of transmission rates and supports relay based transmissions. the real situation. Ergen and Varaiya presented a model The wireless medium is shared among multiple contend- in [5] for decomposition of energy consumption in IEEE ing nodes, that is, a single physical channel is available for 802.11. They derived the formula for the amount of energy wireless transmission. We make use of the control packets to consumed by a node in order to transmit 1 MB of data in solve the hidden terminal problem and to improve the system anetworkwithn nodes in ideal channel conditions. This performance. Another assumption in this model is that the model can differentiate receiving and idle states. However collision can only take place at the initializing control packet. none of the above models are suitable for the relay based For the modeling of energy, we assume saturated network MAC protocols and they require significant modifications for where nodes always have packets to transmit. In addition to the later situation. this we assume that there is always a relay available to help. In this paper, we propose a generalized model for energy All nodes are capable of being relays for other nodes. The consumption and address the energy concerns of using relay relay nodes simply forward the packets and reduce the overall based MAC protocols, as it is important to examine the transmission time via dual-hop transmission at higher data impact of using a relay on energy consumptions compared rates. They are not required to contest for access as once a EURASIP Journal on Wireless Communications and Networking 3 source node acquires the medium the transmission is done via the relay. These assumptions lead to the following [7–9] Transmit equations: ⎧ +1 ⎪ 2 1 − 2p 1 − pm ⎪ ⎪ m+1 m+1 ⎪ CWmin 1 − 2p 1 − p + 1 − 2p 1 − p ⎪ ⎪ ⎪ m ≤ m ⎨ Receive Listen = − − m+1 τ ⎪ ⎛ 2 1 2p 1 p ⎞ ⎪ ⎪ − m +1 − − − m+1 ⎪ CWmin 1 (2p) (1 p)+(1 2p)(1 p ) ⎪ ⎝ ⎠ Figure 2: Physical States. ⎪ ⎪ +CW 2m pm +1 1−2p 1−pm −m ⎩⎪ min ( )( ) m>m (1) we consider a saturated environment for this analysis, we n−1 p = 1 − (1 − τ) (1 − Pe),(2)only consider three physical states: transmit, receive and listen (idle/overhearing) as shown in Figure 2.Activenodes = − − n Ptr 1 (1 τ) ,(3)transmit, receive and listen during a transmission whereas − non active nodes only listen to transmission. Next, we use nτ(1 − τ)n 1 Ps = . (4) this system model to derive the energy analysis in ideal Ptr channel and in channel with transmission errors. Based on Here, (1) shows the probability τ that a node transmits the physical states we define operations for active and non in a randomly chosen slot depends on the conditional active nodes involved. probability of packet failure p,whereCWmin is the minimum contention windows, m is the maximum backoff stage and 2.2. Energy Analysis for Ideal Channels. For energy con- m is the retry limit. Equation (2) gives the packet failure sumption in ideal channel, we know that the packet failure −1 probability in terms of collision (i.e., (1 − τ )n ) and packet is only due to collision and there are no transmission errors (Pe), where n is the total number of nodes. This errors. For the case of ideal channel Pe is zero. Based equation differs from [7–9] where probability of failure is on the above concept of active and non active nodes we only due to collisions. From this equation we can see that the have three available states, that is, transmit, receive and packet failure is due to collision, transmission errors or both. listen (idle/overhearing). We further define operations within Here the probability of having failure due to both is almost the three states: (a) successful transmission; (b) successful negligible. Equations (1)and(2) are a nonlinear system reception; (c) overhearing (reception of packets intended for which can be solved numerically to find p and τ.Equation other stations); (d) idle listening (when the channel is idle); (3)givesPtr which is the probability that there is at least one (e) unsuccessful (colliding) transmissions; and (f) reception transmission in the considered slot time. In expression (4), of collisions. The probabilities of different operations in an Ps is the probability of a successful transmission. The total ideal channel are classified as follows: energy in joules consumed by Node l to successfully transmit ⎡ ⎤ ⎡ ⎤ and receive 1 MB of data can then be defined as r s t u rx = ⎣ ⎦ ⎣ ⎦ Js ρrx Tcontrol + Tdata + ρtx Tcontrol + Tdata E Energy consumed in one slot = = = = J(n) = , (5) i 0 j 0 k 0 l 0 E[Data transmitted and received in one slot] + ρ [(r + s + t + u − 1)(T + δ) + (T + δ)], σ SIFS DIFS P P (1 − P ) E MB transmitted by l in one slot = s tr e E[P ]. n (7) (6) ⎡ ⎤ ⎡ ⎤ r s t u tx = ⎣ ⎦ ⎣ ⎦ In (5) J(n) is the energy consumed in Joules/MB. This Js ρrx Tcontrol + Tdata + ρtx Tcontrol + Tdata = = is the ratio of expected energy (in Joules) consumed by i 0 j 0 k=0 l=0 Node l in one slot to the expected data (in MB) successfully + ρσ [(r + s + t + u − 1)(TSIFS + δ) + (TDIFS + δ)], transmitted and received by node l in one slot. Equation (6) (8) gives the expression for expected data (in MB) transmitted and received by a node l in one slot, where E[P] is the tx = ∗ Jc ρtxTcontrol + ρσ (δ + TEIFS),(9) packet size in MB. In (5) slot refers to a transmission slot rx = ∗ and successful transmission includes transmission by source, Jc ρrxTcontrol + ρσ (δ + TEIFS), (10) forwarding by relay and reception by destination. = Nodes in the network are classified as active (source, relay Jσ ρσ σ, (11) and destination) and non active (all other nodes listening) tx J = ρtxTe + ρσ (δ + TEIFS), (12) nodes. Energy consumed in each kind of slot is the product e rx = of slot duration and power consumption in that slot. As Je ρrxTe + ρσ (δ + TEIFS), (13) 4 EURASIP Journal on Wireless Communications and Networking E energy consumed by l in one slot and extended inter-frame times. δ is the propagation delay and σ is the slot time. In (7)and(8), reception and = − tx − (n−1) − (1 Ptr)Jσ + τpJc + τ(1 τ) (1 Pe) transmission of multiple packets is shown. Equation (7)gives a generalized equation for determining Jxr(r)andJxr(l), × rx rx − − (n−1) s s Js (l) + Js (r) + (n 3)τ(1 τ) which are probabilities of successful reception of packets × − rx ∼ − by relay and destination (which are active nodes). Equation (1 Pe)Js ( l) + (1 τ) (7) consists of the sum of energy consumed in receiving, − × 1 − (1 − τ)(n 1)(1 − P ) − (n − 1)τ transmitting and listening. Energy consumed in each of these e states is the product of slot duration and respective power. ( −2) Here the slot duration in transmitting and receiving of the ×(1 − τ) n Jrx + τ 1 − p Jtx + τP Jtx c s e e control and data packets is the sum of their times. Where r − and s are total number of control and data packets received. + τ(1 − τ)(n 1)P Jrx(l) + Jrx(r) e e e Similarly t and u are total number of control and data packets (n−1) transmitted. The sum of r, s, t and u is the total number of + (n − 3)τ(1 − τ) P Jrx(∼ l). e e control and data packets in a protocol. The same expression (14) rx ∼ is used to determine Js ( l), where no transmission of rx packets is involved. In (8) successful transmission of a packet Js (l) is the probability of successful reception of packet ∗ ( −1) by an active node (source) is given. In (9)and(10), T destined for Node l, and is equal to τ(1 − τ) n (1 − control is the time for collision of control packet (initiated from P ); Jrx(∼ l) is the probability of successful reception of e s source to relay or destination) and Jtx are Jrx the probabilities packet not destined for Node l, and is equal to (n − c c − of transmission and reception of collided packets. Equation − (n 1) − rx 3) τ(1 τ) (1 Pe); Js (r) is the probability of successful (11) shows the listening (idle) state as a product of idle slot reception of packet destined for relay r, and is equal to and idle power. Equations (7)–(11) are for the ideal case − (n−1) − tx τ(1 τ) (1 Pe); Js is the probability of successful where there are no errors and are the same as in [7]. This set transmission of a packet by Node l, and is equal to τ(1 − of equations represent a generic model and is used to show rx p); Jc is the probability of reception of a collided packet, performance of relay based MAC protocols and can easily (n−1) and is equal to (1 − τ). [1 − (1 − τ) (1 − Pe) − be adapted to cater for 802.11 a/b/g physical layers, with the − − (n−2) tx (n 1)τ(1 τ) ]; Jc is the probability of collision on parameters changed appropriately. transmission of a packet by Node l, and is equal to τp(1−Pe); is the probability of idle slots, and is equal to (1 − ). Jσ Ptr 2.3. Energy Analysis for Channels with Transmission Errors. In rx( ) is successful reception of packets destined for node Js l this section we take into consideration the impact of trans- provided that there is a transmission free from collision and l mission errors on the energy consumption. Unlike collision error. Similarly rx( ) is the successful reception by relay with Js r which occurs at the first control packet, transmission errors the same conditions. It is true as the relay is not involved in can occur at any packet. Therefore it is important to take into the contention process. rx(∼ ) is the successful reception by Js l consideration that even with successful reception of one or all non active overhearing nodes. The term ( − 3) ensures n more packets involved in the transmission a failure can still that only non active nodes are considered. tx is the successful Js take place due to one of the following packets being in error. transmission of a packet provided there is a transmission For energy analysis of channel with transmission errors without any failure. tx is the transmission where there is Jc we add more operations to those defined earlier in no error and failure is due to collision. rx is the reception Jc Section 2.2. The additional operations due to errors are: of collided packet. is the probability that there is no Jσ (g) unsuccessful (error) transmissions and (h) reception transmission. of errors. The probabilities of additional operations are as The numerator in (5) is defined in expression (14). As follows. evident from the nature of relay based MAC protocols, we tx Je is the probability of transmission of a packet in error use control packets to coordinate relays which are followed − − (n 1) rx by the data and Acknowledgement (ACK) packets. As from by Node l, and is equal to τ(1 τ) Pe; Je (l) is the the operations defined earlier in this section we know that probability of reception of a packet in error by destined Node − (n−1) rx in a transmission slot we have active nodes and non active l, and is equal to τ(1 τ) Pe; Je (r) is the probability of nodes. In order to model this behaviour of transmitting reception of a packet in error by destined relay r, and is equal − (n−1) rx ∼ and receiving (active nodes) or receiving only (non active to τ(1 τ) Pe; Je ( l) is the probability of reception of nodes) control and data packets, we formulate (7)to(11) a packet in error not destined for Node l, and is equal to (n−1) to show the working of the protocol. To calculate the energy (n − 3)τ(1 − τ) Pe. consumed by nodes (active and non active) (7)to(11) shown These expressions together with those defined earlier will above are used in (14). For ideal scenario where we have give the energy consumption in the case of transmission tx no transmission errors, it is possible to simplify (14)by errors. Je is the transmission of a packet in error provided = rx rx rx ∼ substituting Pe 0. These equations are independent of there is no collision. Je (l), Je (r), and Je ( l)areproba- the protocol. Also, ρtx, ρrx and ρσ are the power consumed bilities of reception of a packet destined for Node l,relayr (in Watts) to transmit, receive and listen (idle/overhearing), and reception of packet not destined for Node l respectively. respectively. TSIFS, TDIFS and TEIFS are the short, distributed Reception of packet in error is conditioned on a transmission EURASIP Journal on Wireless Communications and Networking 5

− rx ∼ free from collision. The term (n 3) in Je ( l) ensures that In 802.11 DCF, if the CTS is not received at the source due onlynonactivenodesareconsidered.In(12)and(13), Te is to collision or channel error, the neighbour nodes of the the average time for a particular packet in error. In this case it source are blocked for the whole duration of transmission is showing the first packet in error. For energy consumption which reduces the bandwidth utilization. Unlike the standard tx in channel experiencing transmission errors, we define Je DCF, in rDCF if CTS/RCTS is not received, the neighbor rx and Je as the probabilities of transmission and reception of nodes are not blocked for the whole duration of transmis- tx packetsinerror.In(12) Je is the probability of transmitting sion. In the 802.11 DCF the source estimates the possible rx apacketinerror.Equation(13) is for determining Je (r), transmission rate and the duration, whereas in modified rx rx ∼ Je (l)and Je ( l), which are probabilities of reception carrier sensing scheme of rDCF, the source first calculates of packet (first packet) in error by relay, destination and (as all control packets are transmitted at base rate of 2 Mbps) overhearing nodes. For simplicity we have only shown (12) the duration of the RRTS and RCTS/CTS transmissions only. and (13) for the case if the first control packet is in error. The The destination based on the received RRTS1 and RRTS2, equations become more complex for following packets being decides in favour of the relay or to revert to the direct in error and are shown in the appendix. Finally the energy transmission. If the destination confirms to the source for a (J/MB) is calculated by using (6), (14)and(5). Equation direct transmission it transmits a CTS packet or else a RCTS (14) is the sum of the products of operations/states and their packet for the relay based transmission. The source extracts probabilities. the agreed transmission rates from RCTS and calculates the To this point, we have shown generalized equations duration of data packet and ACK. This information is made for the energy analysis in ideal channel and channel with available to all overhearing nodes via the RSH attached to the transmission errors. In the following section we will apply data packet. This prevents the unnecessary blocking of nodes the above energy analysis to a relay based MAC protocol. for the entire duration of transmission. The rDCF uses the same physical characteristics such as 3. Relay-Enabled Distributed transmission power and Receiver Signal Strength Indication (RSSI) as in IEEE 802.11 DCF. There is no power control and Coordination Function both control and data packets are transmitted at maximum This section briefly describes the relay-enabled distributed power. Relay transmission is intended to provide higher coordination function. The rDCF was originally proposed in throughput and reduced blocking time. [6], where relay is used to improve the system throughput All the nodes maintain a willing list based on the channel and reduce packet delay. In rDCF a high data rate dual hop quality between them and neighbouring nodes. The length path is used instead of a low data rate direct path between of the willing list is limited to 10 entries to reduce overheads. the source and destination as shown in Figure 1.TherDCF Nodes keep updating their willing lists with better links and is based on the IEEE 802.11 DCF, but has introduced the frequently broadcast their willing list to their neighbours. following modifications. The willing list contains an entry for the credit rating of each potential relay node. This rating improves with successful (i) Backward compatibility to 802.11 DCF (nonrelay relaying and degrades with inability to relay. mode) and requires only a firmware upgrade. (ii) Control packets transmitted at the base rate of 3.1. Throughput Analysis of rDCF. To analytically model 2Mbps. rDCF, the authors in [6] have used Bianchi’s model [8]. For the throughput calculation, saturated condition (i.e., every (iii) Modified carrier sensing scheme (shown in Figure 3). node always has data to transmit) is assumed. It further (iv) Introduction of Reservation Sub Header (RSH) assumes that the channel is ideal (i.e., there are no hidden (transmitted at 2 Mbps and used to broadcast dura- nodes and capture effect), and calculates the saturated tion information for the rest of packet) in DATA throughput for Relay-Request-To-Send (RRTS)/Relay-Clear- packets transmitted at higher rates from source to To-Send (RCTS) access. For rDCF, the equations for the relay. average times of channel sensed busy for collision and successful transmission, respectively, are (v) Frequent broadcasting of willing lists (potential relay rDCF = entries) between nodes. Tc TRRTS1 + TDIFS + δ, (vi) Relay selection based on a credit system. rDCF = Ts TRRTS1 + TRRTS2 + TRCTS + TACK + TDATA(L,R1) (15) Considering the fact that rDCF is backward compatible + TDATA( , 2) +5TSIFS +6δ + TDIFS, to 802.11 DCF and have the same backoff scheme, we can L R observe that the process of channel contention and time where Tc and Ts are the average times channel is sensed busy spent in contention for each node in rDCF is the same as during collision and successful transmission. Here RRTS in 802.11 DCF. and RCTS are control packets for coordinating relay-enabled The modified carrier sensing scheme (shown in Figure 3) transmission as shown in Figure 3.In(15), TRRTS1, TRRTS2, used in rDCF achieves better bandwidth utilization. A major TRCTS and TACK are the transmission times for RRTS1 (source advantage of this scheme compared to 802.11 DCF is that the to relay), RRTS2 (relay to destination), RCTS and ACK, nodes are blocked exactly for the data transmission duration. respectively. TSIFS and TDIFS are interframe times and δ is the 6 EURASIP Journal on Wireless Communications and Networking

SIFS SIFS SIFS SIFS SIFS

DATA(R1) Source RRTS1 Time

RRTS2 DATA(R2) Relay Time

Destination RCTS ACK Time

NAV(RRTS1/RRTS2) Busy medium (DATA) NAV(DATA) Others Time NAV(RCTS)

Figure 3: Carrier sensing scheme of rDCF.

RRTS1 = propagation delay. TDATA(L,R1) and TDATA(L,R2) are the times Te TRRTS1 + TEIFS + δ, for data packets of length L bytes at rates R1 and R2. RRTS2 = Te TRRTS1 + TRRTS2 + TSIFS +2δ + TEIFS, 3.2. Analysis of rDCF with Transmission Errors. Due to the RCTS = Te TRRTS1 + TRRTS2 + TRCTStimeout + TSIFS +3δ + TDIFS, nature of rDCF, we must consider all the links: the link DATA1 = between source and relay (with probability of bit errors Pb1 Te TRRTS1 + TRRTS2 + TRCTS + TDATA1 +3TSIFS and distance dsr ), the link between relay and destination (with probability of bit errors Pb2 and distance drd), and the +4δ + TEIFS, link between source and destination (with probability Pb and DATA2 T = TRRTS1 + TRRTS2 + TRCTS + TDATA1 + TDATA2, distance dsd). As a result, the probability of packet errors for e the rDCF protocol and overhead caused by packet errors can +4TSIFS +5δ + TEIFS be derived as (16)and(17). In (16), is the probability of packet errors, which is ACK = Pe Te TRRTS1 + TRRTS2 + TRCTS + TDATA1 + TDATA2 based on transmission of individual packets (control and data) involved in rDCF. Note that for packets following + TACKtimeout +4TSIFS +6δ + TDIFS. RRTS1 their probability of error is conditioned on successful (17) reception of previous packets. For RRTS2 in (16), probability of error is based on the successful reception of RRTS1. In the Expressions (18)to(27) are derived and simpliﬁed based same way total probability of error is based on the successful on (7)to(13)forrDCF: reception of all control and data packets. rx( ) = Probability of packet error is calculated based on the Js l ρrx TRRTS1 + TRRTS2 + TDATA(L,R1) + TDATA(L,R2) bit error probability of a particular link and length of that + ρtx(TRCTS + TACK) packet. In (17) we work out the average time spent in all the packets in error, respectively: + ρσ (5TSIFS + TDIFS +6δ), (18) RRTS1 = − − LRRTS1 rx = Pe 1 (1 Pb1) , Js (r) ρrx TRRTS1 + TRCTS + TACK + TDATA(L,R1) RRTS2 = − LRRTS1 − − LRRTS2 Pe (1 Pb1) 1 (1 Pb2) , + ρtx TRRTS2 + TDATA(L,R2) (19) RCTS = − LRRTS1 − LRRTS2 − − LRCTS Pe (1 Pb1) (1 Pb2) 1 (1 Pb) , + ρσ (5TSIFS + TDIFS +6δ),

DATA1 = − LRRTS1 − LRRTS2 − LRCTS Pe (1 Pb1) (1 Pb2) (1 Pb) rx(∼ ) = ( + + + Js l ρrx TRRTS1 TRRTS2 TRCTS TACK × − ( − )LDATA1 1 1 Pb1 , +TDATA(L,R1) + TDATA(L,R2) (20) DATA2 = − LRRTS1 − LRRTS2 − LRCTS Pe (1 Pb1) (1 Pb2) (1 Pb) + ρ (5T + T +6δ), σ SIFS DIFS × − LDATA1 − − LDATA2 (1 Pb1) 1 (1 Pb2) , tx = Js ρtx TRRTS1 + TDATA(L,R1) ACK = − LRRTS1 − LRRTS2 − LRCTS Pe (1 Pb1) (1 Pb2) (1 Pb) + ρrx TRRTS2 + TRCTS + TACK + TDATA(L,R2) (21) LDATA1 LDATA2 LACK × (1 − Pb1) (1 − Pb2) 1 − (1 − Pb) , + ρσ (5TSIFS + TDIFS +6δ), P = 1 − (1 − P )LRRTS1 (1 − P )LRRTS2 (1 − P )LRCTS e b1 b2 b rx = ∗ Jc ρrxTRRTS1 + ρσ (δ + TEIFS), (22) LDATA1 LDATA2 LACK × (1 − Pb1) (1 − Pb2) (1 − Pb) , tx = ∗ (16) Jc ρtxTRRTS1 + ρσ (δ + TEIFS), (23) EURASIP Journal on Wireless Communications and Networking 7

tx = RRTS1 Je (l) ρtxTe + ρσ (δ + TEIFS), (24) Table 1: rDCF parameters [6, 11]. rx = RRTS1 Physical characteristic IEEE 802.11 b DSSS Je (l) ρrxTe + ρσ (δ + TEIFS), (25) CWmin 32 rx = RRTS1 Je (r) ρrxTe + ρσ (δ + TEIFS), (26) m 5 rx ∼ = RRTS1 m 6 Je ( l) ρrxTe + ρσ (δ + TEIFS), (27) DIFS 50 μs where TACKtimeout = TACK + TSIFS and TRCTStimeout = TRCTS + SIFS 10 μs TSIFS. It is evident that for RRTS1 in error the time spent is EIFS DIFS + SIFS + ACK shortest and for ACK in error the time spent is the highest. Slot 20 μs From (18)to(27) we have derived further parameters MAC header 272 bits for rDCF. In rDCF, the total number of control and data PHY header 192 μs packets is six, which is the sum of r, s, t and u. This 160 bits/control rate + PHY information is substituted in (7)and(8)toderive(18)to RTS header (21) for rDCF. We make use of (7)toderive(18)–(20) 112 bits/control rate + PHY for rDCF. In (18)and(19), the probability of successful CTS header reception by relay and destination in rDCF is shown. In 112 bits/control rate + PHY (20), the probability of successful reception by overhearing ACK nodes in rDCF is shown. In the same way, (8)isusedfor header 256 bits/control rate + PHY the derivation of (21), showing the probability of successful RRTS1 transmission by the source in rDCF. Now to address collision header 260 bits/control rate + PHY in rDCF we make use of (9)and(10), to derive (22)and RRTS2 (23). Expressions (22)and(23) show the probability of header 120 bits/control rate + PHY reception and transmission of collided packets by destination RCTS and source nodes, respectively. Similarly, (24)to(27) shows header the probability of transmission (by source) and reception (by Data Rates 1, 2, 5.5, 11 Mbps relay, destination and overhearing nodes) of packets in error. Control Rate 2 Mbps Further we have shown the calculation for the Te for the Propagation Delay 1 μs RRTS1. Antenna height 1.5 meters In Section 4, we perform rigorous performance analysis Transmit Power 15 dBm to show energy consumption, impact of change in packet Loss 0 dB length, performance under transmission errors and decomposition of energy. Shadowing deviation 10 dB BPSK @ 1 Mpbs, QPSK @ Data Rates and Modulations 2 Mbps, CCK5.5 @ 5.5 Mbps, 4. Performance Analysis CCK11 @ 11 Mbps −94 dBm, −91 dBm, Receiver Sensitivity For performance evaluation, we assume that (1) each node −87 dBm, −82 dBm always has data to transmit and (2) a relay is available. The results in this section are for combinations of 11 and 5.5 Mbps, denoted by rDCF (R1,R2). A typical set of Figure 5 plots (6) and shows average payload per node parameters used for the evaluation are given in Table 1. transmitted and received in one slot and is the same for 802.11 DCF and rDCF rate combinations. Figure 6 plots (5) 4.1. Energy Consumption. In this section we will show the and shows the average energy consumed in transmitting and energy consumption of rDCF to analyze the effectiveness receiving 1 MB of data at packet length of 1000 Bytes in ideal of proposed model. We will use it to calculate the energy channel. Energy grows linearly with the increasing number consumption (J/MB) of rDCF. Equations (7)to(13)are of nodes. As seen all rate combinations of rDCF perform in modified according to the protocol and are shown for a similar fashion but rDCF (11, 11) achieves slightly higher respective operations in (18)to(27). Expression (11) is used savings. in existing form. Here ρtx, ρrx and ρσ are assigned 1.34 watts, As observed in this section, energy consumption grows 0.90 watts and 0.73 watts, respectively [10]. Here for the ideal linearly with node density. Therefore, it is important to case Pe = 0. analyze the performance of relay based schemes, to see Figure 4 plots expression (14) and shows average energy the impact on energy with change in packet length and consumed by each node in one slot of IEEE 802.11 and rDCF effectiveness of the proposed method. It is also important for different rate combinations. Data rates used for IEEE to observe the decomposition of energy to make efficient 802.11 is 2 Mbps and for rDCF combinations of 5.5 and utilization of energy. 11 Mbps. Packet length of 1000 Bytes, CWmin = 32, m = 5 The rDCF (11, 11) achieves maximum savings of 24.9% and m = 6 are used. Energy consumption per slot of 802.11 and 36.99% at 5 and 50 nodes, respectively, due to faster two and rDCF increases with the number of nodes. hops of 11 Mbps in ideal channel conditions. This is evident 8 EURASIP Journal on Wireless Communications and Networking

×10−4 30 2 25

20 1.5 in one slot] (Joules)

l 15

10 1

Energy consumption (Joules/MB) 5 [Energy consumed by

E 0 0 5 10 15 20 25 30 35 40 45 50 0.5 0 5 10 15 20 25 30 35 40 45 50 Number of nodes Number of nodes 802.11 DCF rDCF(5.5, 11) 802.11 DCF rDCF(5.5, 11) rDCF(5.5, 5.5) rDCF(11, 11) rDCF(5.5, 5.5) rDCF(11, 11) Figure 6: Energy consumed (J/MB) for 802.11 DCF and rDCF in Figure 4: Average energy (J) consumed in one slot. ideal channel.

×10−5 plots the average energy (J/MB) consumed in transmitting 5 and receiving 1 MB of data. The energy consumption grows linearly with number of nodes while the slope depends on 4.5 the packet size. It is interesting to see that the results are in 4 agreement with the ﬁndings of single hop 802.11 DCF, that is, it is still advantageous to transmit large payloads. This is 3.5 true even with the doubled overhead used due to relay based transmission. 3

2.5 4.3. Performance under Transmission Errors. For the perfor-

[MB transmitted] mance of rDCF (R1, R2) under transmission errors we con- E 2 siderbiterrorprobabilitiesofdifferent modulation schemes 1.5 used in IEEE 802.11b under Additive White Gaussian Noise (AWGN). The bit error probabilities for BPSK, QPSK, CCK 1 5.5 and CCK 11 can be easily obtained from [12]tocalculate 0.5 the corresponding packet error rate. In this paper we use 0 5 10 15 20 25 30 35 40 45 50 the two-ray ground reflection model and card specifications Number of nodes of ORINOCO11b in NS-2 [13]. The two-ray ground model consists of two parts: (1) Free space path loss for distances Figure 5: Average payload (MB) transmitted and received in one ff slot. less than the Friss cuto distance, dfriss and (2) The two-ray propagation loss for distances greater than dfriss.Thebiterror rates below were obtained using the two-ray ground model from the following results that this model helps in predicting where dfriss = 230 meters and antenna height ht and hr = 1.5 the energy consumption and it is encouraging to observe meters. that using a relay not only results in higher throughput We consider two scenarios: (a) for the symmetric (i.e., ffi but is energy e cient as well. As a consequence of these Pb1 equal to Pb2) and (b) the asymmetric link (i.e., Pb1 results we conclude: (1) relay based transmissions are energy not equal to Pb2) are shown in Figure 8. For the symmetric efficient and (2) relaying for others saves energy for the scenario we have placed relay exactly between source and −5 whole network. destination (i.e., dsd = 400 m @ 2 Mbps with Pb = 10 , −9 dsr = drd = 200 m @ 5.5 Mpbs with Pb1 = Pb2 = 3 × 10 ). 4.2. Impact of Change in Packet Length. Since rDCF (11, 11) The probability of errors for the direct link is 10−5 (which is is the most energy efficient under ideal channel conditions, equivalent to a packet error rate of 8% at a packet length of we use it for the performance analysis. We analyze the per- 1000 bytes). Error probabilities for relay links are worked out formance of rDCF (11, 11) with varying packet sizes of 100, relatively based on [13]. 500 and 1000 bytes in ideal channel conditions. The rest of For the asymmetric scenario, we have placed relay closer −5 the parameters used are the same as in Section 4.1. Figure 7 to source (i.e., dsd = 400 m @ 2 Mbps with Pb = 10 , EURASIP Journal on Wireless Communications and Networking 9

160 16

140 14

120 12

100 10 8 80

(Joules/MB) 6 60 4 40

Energy consumption (Joules/MB) 2 20 0 0 0102030405060 0 5 10 15 20 25 30 35 40 45 50 Number of nodes Number of nodes Idle 100 Reception by intended node 500 Reception by intended relay 1000 Reception by overhearing nodes Figure 7: Energy consumed (J/MB) for rDCF (11, 11) in ideal Reception of collided packet channel at diﬀerentpacketsizes. Successful transmission Transmission of collided packet

35 Figure 9: Decomposition of energy (J/MB) for rDCF (11, 11) in ideal channel. 30

25 5. Decomposition of Energy Consumed To show the decomposition of energy, rDCF (11, 11) in 20 the ideal case, a packet length 1000 bytes is used. From the decomposition of energy in Figure 9, we can observe the 15 energy consumed in various operations. The operations can be mainly classified as useful and overheads. 10 The useful operations are successful transmission (by source) and successful reception of packets (by relay and Energy consumption (Joules/MB) 5 destination). The overhead operations which waste energy are success- 0 ful reception of packets (overhearing nodes), reception of 0 5 10 15 20 25 30 35 40 45 50 collided packet, transmission of collided packet and staying Number of nodes idle. It is observed that the energy consumed in successful Symmetric transmission and reception of data by destination and relay Asymmetric is almost constant. Here it is interesting to see that most Ideal of the energy is consumed in listening/overhearing by other nodes. This increases with respect to the number of nodes. Figure 8: Energy consumed (J/MB) for rDCF in channel errors. In addition to this the energy consumed in receiving collided packet and staying idle also increases with the increase in number of nodes. −7 dsr = 160 m @ 11 Mpbs with Pb1 = 10 and drd = 270 m Further to this we can see from the Figure 10 (asymmet- −6 @5.5MpbswithPb2 = 7 × 10 ). ric scenario), that overhearing is related to both successful We can observe much higher energy consumption for the transmission and transmission in error. It is important to see rDCF (symmetric and asymmetric) in transmission errors as here that the energy consumption for transmission in error compared to rDCF (ideal). The energy consumption almost also increases with the number of nodes. In this the major doubles for both the symmetric and asymmetric cases, contributor is again overhearing. Overhearing of packet in whereas the symmetric and asymmetric scenarios results in error is an increasing function of number of nodes and it also similar energy consumption. The difference between the two increases with the average time spent in error (i.e., for RRTS1 scenarios is very small and is mainly due to the average time in error the energy consumed is minimal and for ACK in calculation for errors (shown in the Appendix) with different error energy consumption is high). Decomposition of energy rate combinations. in rDCF shows that it is possible to improve the performance 10 EURASIP Journal on Wireless Communications and Networking

Table 2: Energy consumption of rDCF. help in devising energy saving mechanisms/policies based on the energy consumption behavior and decomposition of Energy consumption (J/MB) 5 nodes 50 nodes energy. We have used rDCF as a case to show the eﬃcacy rDCF (11, 5.5) Ideal 1.75 16.89 of our proposed analytical model. This model also shows rDCF (11, 5.5) Errors 1.98 29.52 the decomposition of energy for relay based MAC which is of interest to protocol designers. Extensive performance analysis for the relay based MAC has also been provided. 35 We have analyzed energy consumption under the impact of varying packet sizes and rate combinations. We have shown 30 that transmission errors can greatly escalate the energy consumption as it will give rise to overhearing again. Possible 25 future work includes the extension of the above model to the unsaturated case and to obtain experimental results from real 20 life scenario.

15 (Joules/MB) Appendix 10 Equation for Packet in Error 5 Here we have shown the extended equations (by replacing 0 (24)to(27)by(A.1)to(A.4)) for the error averaged over all 0102030405060packets and used in Figures 8 and 10: Number of nodes tx = Idle Je ρtxTRRTS1 + ρσ (TEIFS + δ) + ρtxTRRTS1 + ρrxTRRTS2 Reception by intended node Reception by intended relay + ρσ (TEIFS + TSIFS +2δ) + ρtxTRRTS1 Reception by overhearing nodes Reception of collided packet + ρrxTRRTS2 + ρσ (TRCTStimeout + TDIFS + TSIFS +3δ) Successful transmission Transmission of collided packet + ρtx TRRTS1 + TDATA(L,R1) + ρrx(TRRTS2 + TRCTS) Transmission of packet in error Reception of errored packet by intended node + ρσ (TEIFS +4δ +3TSIFS) Reception of errored packet by intended relay Reception of errored packet by overhearing nodes + ρtx TRRTS1 + TDATA(L,R1) Figure 10: Decomposition of energy (J/MB) for rDCF (11, 5.5) in + ρrx TRRTS2 + TRCTS + TDATA(L,R2) channel with errors.

+ ρσ (TEIFS +4TSIFS +5δ) of this protocol by devising a policy which can reduce the + ρtx TRRTS1 + TDATA(L,R1) energy consumption by overhearing nodes. Energy consumption of rDCF is shown in Table 2.In + ρrx TRRTS2 + TRCTS + TDATA(L,R2) error rDCF consumes 11.6 % and 42.7% extra energy at 5 and 50 nodes, respectively. This analysis allows us to + ρσ (TACKtimeout + TDIFS +4TSIFS +6δ), design energy eﬃcient protocols by predicting the energy (A.1) consumption. Finally, it can be used for the prediction of rx = energy consumption and will beneﬁt the design of MAC Je (l) ρrxTRRTS1 + ρσ (TEIFS + δ) + ρrx(TRRTS1 + TRRTS2) protocols for energy critical environments. + ρσ (TEIFS + TSIFS +2δ) + ρrx(TRRTS1 + TRRTS2)

6. Conclusions + ρσ (TRCTStimeout + TDIFS + TSIFS +3δ) We have presented a general analytical energy model for + ρrx TRRTS1 + TDATA(L,R1) + TRRTS2 + TDATA(L,R2) relay based MAC protocols. This model assumes saturated environment and collision of first control packet only. This + ρtxTRCTS + ρσ (TEIFS +5δ +4TSIFS) model is applicable to both ideal channel and transmission errors. This model can be used to illustrate energy consump- + ρrx TRRTS1 + TRRTS2 + TDATA(L,R1) + TDATA(L,R2) tion of any relay based MAC protocol with modification in accordance to the protocol flow and to cater for any physical + ρtxTRCTS + ρσ (TACKtimeout + TDIFS +4TSIFS +6δ), layer with change in parameters. Further this model will (A.2) EURASIP Journal on Wireless Communications and Networking 11

rx = Je (r) ρrxTRRTS1 + ρσ (TEIFS + δ) + ρrxTRRTS1 + ρtxTRRTS2 International Symposium on Wireless and Pervasive Computing (ISWPC ’09), Melbourne, Australia, February 2009. + ρσ (TEIFS + TSIFS +2δ) + ρrxTRRTS1 + ρtxTRRTS2 [8] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordination function,” IEEE Journal on Selected + ρσ (TRCTStimeout + TDIFS + TSIFS +3δ) Areas in Communications, vol. 18, pp. 535–547, 2000. [9] H. Wu, Y. Peng, K. Long, S. Cheng, and J. Ma, “Performance + ρrx TRRTS1 + TDATA(L,R1) + TRCTS of reliable transport protocol over IEEE 802.11 wireless LAN: analysis and enhancement,” in Proceedings of the Annual + ρtx TRRTS2 + TDATA(L,R2) Joint Conference of the IEEE Computer and Communications Societies (INFOCOM ’02), vol. 2, pp. 599–607, 2002. + ρσ (TEIFS +5δ +4TSIFS) [10] L. M. Feeney and M. Nilsson, “Investigating the energy consumption of a wireless network interface in an ad hoc + ρrx TRRTS1 + TDATA(L,R1) + TRCTS networking environment,” in Proceedings of the Annual Joint Conference of the IEEE Computer and Communications Soci- + ρtx TRRTS2 + TDATA(L,R2) eties (INFOCOM ’01), vol. 3, pp. 1548–1557, April 2001. [11] IEEE 802.11b, Part II, “Wireless LAN Medium Access Control + ρσ (TACKtimeout + TDIFS +4TSIFS +6δ), (MAC) and Physical Layer (PHY) Spec: High Speed Physical (A.3) layer Extension in the 2.4 GHz Band,” supplement to IEEE 802.11 Standard, 1999. [12] J. del Prado Pavon and S. Choi, “Link adaptation strategy rx ∼ = Je ( l) ρrxTRRTS1 + ρσ (TEIFS + δ) + ρrx(TRRTS1 + TRRTS2) for IEEE 802.11 WLAN via received signal strength measurement,” in Proceedings of the IEEE International Conference on + ρσ (TEIFS + TSIFS +2δ) + ρrxTRRTS1 + ρrxTRRTS2 Communications (ICC ’03), vol. 2, pp. 1108–1113, Anchorage, Alaska, USA, May 2003. + ρσ (TRCTStimeout + TDIFS + TSIFS +3δ) [13] http://www.isi.edu/nsnam/ns/. + ρrx TRRTS1 + TDATA(L,R1) + TRRTS2 + TDATA(L,R2)

+ ρrxTRCTS + ρσ (TEIFS +5δ +4TSIFS) + ρrx TRRTS1 + TRRTS2 + TDATA(L,R1) + TDATA(L,R2)

+ ρrxTRCTS +ρσ (TACKtimeout +TDIFS +4TSIFS +6δ). (A.4)

References

[1] A. Nosratinia, T. E. Hunter, and A. Hedayat, “Cooperative communication in wireless networks,” IEEE Communications Magazine, vol. 42, no. 10, pp. 74–80, 2004. [2] H. Zheng, Y. Zhu, C. Shen, and X. Wang, “On the eﬀectiveness of cooperative diversity in ad hoc networks: a MAC layer study,” in Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP ’05), vol. 3, pp. 509–512, March 2005. [3] M. M. Carvalho, C. B. Margi, K. Obraczka, and J. J. Garcia- Luna-Aceves, “Modeling energy consumption in single-hop IEEE 802.11 ad hoc networks,” in Proceedings of the Interna- tional Conference on Computer Communications and Networks (ICCCN ’04), pp. 367–372, 2004. [4] X. Wang, J. Yin, and D. P. Agarwal, “Analysis and optimization of the energy eﬃciency in the 82.11DCF,” Mobile Networks and Applications, vol. 11, pp. 279–286, 2006. [5] M. Ergen and P. Varaiya, “Decomposition of energy consumption in IEEE 802.11,” in Proceedings of the IEEE International Conference on Communications (ICC ’07), pp. 403–408, June 2007. [6]H.ZhuandG.Cao,“rDCF:arelay-enabledmediumaccess control protocol for wireless ad hoc networks,” IEEE Transac- tions on Mobile Computing, vol. 5, no. 9, pp. 1201–1214, 2006. [7] R. Ahmad, F.-C. Zheng, M. Drieberg, S. Olafsson, and M. Fitch, “Modelling energy consumption of relay-enabled MAC protocols in ad hoc networks,” in Proceedings of the 4th