Which Wireless Technology for Industrial Wireless Sensor Networks ...

AQ3

AQ4

AQ5

AQ6

AQ7

AQ8

AQ9

AQ10

1

4

Which Wireless Technology for Industrial Wireless Sensor Networks? The Development of OCARI Technology

5 6

Khaldoun Al Agha, Senior Member, IEEE, Marc-Henry Bertin, Tuan Dang, Member, IEEE, Alexandre Guitton, Member, IEEE, Pascale Minet, Thierry Val, and Jean-Baptiste Viollet

1

2

3

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Abstract—In this paper, we present an industrial development of a wireless sensor network technology called OCARI: Optimization of Communication for Ad hoc Reliable Industrial networks. It targets applications in harsh environments such as power plants and warships. OCARI is a wireless-communication technology that supports mesh topology and power-aware ad hoc routing protocol aimed at maximizing the network lifetime. It is based on IEEE 802.15.4 physical layer with deterministic Media Access Control layer for time-constrained communication. During the nontime-constrained communication period, its ad hoc routing strategy uses an energy-aware optimized-link state-routing proactive protocol. An OCARI application layer (APL) is based on ZigBee application support sublayer and APL primitives and profiles to provide maximum compatibility with ZigBee applications. To fully assess this technology, extensive tests are done in industrial facilities at Electricité De France R&D as well as at Direction des Constructions Navales Services. Our objective is then to promote this specification as an open standard of industrial wireless technology.

26 27 28 29 30 31

Index Terms—Ad hoc network, bit-error rate (BER), deterministic Media Access Control (MAC), IEEE 802.15.4, interference model, ISA100.11a, power-aware routing strategies, power plants, signal-to-interference-plus-noise ratio (SINR), warships, wirelessHART, wireless sensors network (WSN), WSN middleware, ZigBee.

I. I NTRODUCTION

Manuscript received July 20, 2008; revised June 30, 2009. This work was supported in part by the French National Research Agency (Agence Nationale de la Recherche) under Grant ANR-06-TCOM-025. K. Al Agha is with LRI, Bat 490, Université Paris-Sud11, 91405 Orsay, France (e-mail: [email protected]). M.-H. Bertin is with Telit Communications S.p.A.—EMEA, Trieste, Italy (e-mail: [email protected]). T. Dang is with Electricité De France (EDF) Research and Development, 78401 Chatou, France (e-mail: [email protected]). A. Guitton is with Clermont University/LIMOS CNRS, Complexe Scientifique des Cézeaux, 63173 Aubière, France (e-mail: [email protected]). P. Minet is with French National Institute for Research in Computer Science and Control (INRIA) Rocquencourt, 78153 Le Chesnay, France (e-mail: [email protected]). T. Val is with the Laboratory of Technology and System Engineering of Toulouse (LATTIS), IUT de Blagnac, 31703 Blagnac, France (e-mail: [email protected]). J.-B. Viollet is with Direction des Constructions Navales Services (DCNS), Département Ingénierie, 56311 Lorient, France (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TIE.2009.2027253

32

W

IRELESS communication represents a major industrial 33 stake in the coming years. It offers numerous usages 34 and helps industry save operating costs as well as improving 35 operational efficiency. In the recent years, WiFi (IEEE 802.11- 36 WLANs) and Bluetooth technologies (IEEE 802.15-WPANs) 37 have known tremendous development and have penetrated 38 small office and home office as well as large enterprise office. 39 These general-public wireless technologies may find their lim- 40 ited usage in industrial installations because of harsh environ- 41 ments, electromagnetic compatibility and interference issues, 42 safety and information technology (IT) security constraints, and 43 battery autonomy. Some of these issues have been addressed 44 by addenda to existing standards. For example, IEEE 802.11i 45 addresses the IT security, and IEEE 802.11e deals with WiFi 46 multimedia (WMM) quality of service (QoS) and WMM Power 47 Save. However, these specifications target consumer market 48 and do not take into account industrial needs in constrained 49 environment. 50 Applications of wireless sensors network (WSN) technology 51 in industrial environments such as power plants and warships 52 typically require the following characteristics. 53 At the physical (PHY) and Media Access Control (MAC) 54 layers: 55

IE E Pr E oo f

AQ1 AQ2

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO. 10, OCTOBER 2009

1) Robust radio transmission (low bit error rate (BER) [1]) regarding radio interferences (measured as signal-tointerference-plus-noise ratio (SINR) [2], [3]); 2) Low power consumption along with power-management capability to maximize battery autonomy; 3) Compatibility with electromagnetic constraints (e.g., low equivalent isotropically radiated power ≤ 10 mW); 4) Deterministic MACs.

61 62 63

At the network layer:

64

56 57 58 59 60

1) Network topology flexibility: star, tree, mesh topologies; 65 2) Network scalability: ability to deal with large network 66 topology and high density of network nodes; 67 3) Transparency for application layer (APL): self- 68 organizing and autoconfigurable network parameter (net- 69 work address, network path, router node selection, . . .); 70 4) Support of energy-aware routing protocol; 71 5) Support of mobility; 72

0278-0046/$26.00 © 2009 IEEE

2


II. R EVIEW OF E XISTING I NDUSTRIAL W IRELESS -C OMMUNICATION S TANDARDS

130 131 132

Wireless mesh networking has emerged in the recent years 133 as a promising design paradigm for next-generation wireless- 134 communication networks with interesting characteristics such 135 as self-organizing and autoconfigurable topology, and ad hoc 136 routing concept. These properties promise substantial benefits 137 in terms of operating and maintenance costs of the communica- 138 tion infrastructure in industrial installations. They also ease the 139 development of “killer applications” such as condition monitor- 140 ing or condition-based maintenance (CBM) that requires flex- 141 ible and cost-effective sensor networks. Wireless technologies 142 help engineers achieve these objectives. However, most of the 143 existing general-public wireless-communication technologies 144 do not take into account the industrial requirements. There 145 exists proprietary radio-communication technologies for indus- 146 trial use (e.g., Wavenis), but the benefits of interoperability (and 147 thus, cost) are lost from multivendors solutions. Developing and 148 promoting industrial wireless-communication standards help 149 industrial end users preserve the expected benefits of wireless 150 technologies. We propose to review the state of the art of current 151 industrial wireless networking standards. 152

IE E Pr E oo f

6) Support of authentication of network node and antiintrusion (to the network) mechanisms. 75 At the APL: 76 1) Support of application profiles (e.g.: sensors, actuators, 77 time-constrained); 78 2) Support of different application communication modes: 79 request/reply, publish/subscribe (event-based notifica80 tion), and periodic/programmable notification; 81 3) Support of IEC61804/electronic device description lan82 guage (EDDL) for diagnosis and maintenance purposes; 83 4) Support of authentication mechanisms. 84 In response to these industrial needs and challenges, there are 85 some working groups such as the wireless industrial networking 86 alliance (WINA), the ZigBee Alliance, WirelessHART [4] from 87 HART communication foundation (HCF), and ISA100 that 88 have tried to define and establish industrial wireless-technology 89 standards for different application domains. Currently, only 90 ZigBee has commercially available products as this Alliance 91 was formed very soon in the end of 2004. These specifica92 tions are all based on IEEE 802.15.4, which provides a good 93 foundation for building ad hoc mesh network. However, IEEE 94 802.15.4 does not specify a standard way or algorithm to fully 95 optimize power consumption in the MAC layer along with 96 a corresponding routing schema. It is up to the application 97 designer to elaborate its own strategy. Deterministic MAC layer 98 [5] is also absent from this standard. 99 In this paper, we describe the Optimization of Communica100 tion for Ad hoc Reliable Industrial networks (OCARI) project 101 in which we try to develop a wireless sensor communication 102 module running an industrial ad hoc mesh networking protocol. 103 It is based on IEEE 802.15.4 PHY layer and satisfies the 104 following criteria in harsh environment: 105 1) Deterministic MAC layer for time-constrained communi106 cation; 107 2) Optimized energy-consumption routing strategy for max108 imum network lifetime within the nontime-constrained 109 communication period; 110 3) Support of human walking-speed mobility for some par111 ticular network nodes (sinks). 112 The project is funded by the French National Research 113 Agency (Agence Nationale de la Recherche). It started at the 114 end of 2006 and gathers partners (see [6] for more details) 115 from industries (EDF/project leader, DCNS, and Telit), as 116 well as university laboratories and research institutes (LIMOS, 117 LATTIS, LRI, and INRIA). EDF and DCNS provide require118 ments and use cases in power industry and warship applications. 119 Telit, a high-technology company in wireless communication, 120 industrializes the prototype. LIMOS and LATTIS university 121 laboratories develop and implement OCARI medium-access 122 methods. INRIA and LRI research institutes work on optimized 123 energy-consumption routing strategy based on optimized-link 124 state-routing (OLSR) proactive protocol [7], [8]. 125 This paper is organized as follows. Section II presents a 126 review of existing wireless-communication standards which 127 may be used in industrial environments. Section III shows 128 the technical aspects of OCARI and details the technological 129 choices. Finally, future works are detailed in Section IV. 73

74

A. IEEE 802.15.4-2003-Based ZigBee Specifications

153

ZigBee is a low-rate wireless personal-area network (PAN) 154 standard for embedded communication system with very low 155 power consumption. It proposes a lightweight1 protocol stack 156 for applications which require low data rates (up to 250 kb/s) 157 and low latency. ZigBee is designed to interconnect au- 158 tonomous sensors and actuators to control units. Battery life- 159 times last from a few months to many years as a result of 160 system power-saving modes, battery-optimized network param- 161 eters, and application configurations. ZigBee is based on IEEE 162 802.15.4-2003 PHY (868 MHz/915 MHz or 2.4 GHz) and 163 MAC layers over which it specifies its network layer (NWK) 164 and APL (Fig. 1). 165 The responsibilities of the ZigBee NWK include mecha- 166 nisms used to join and leave a network, to apply security to 167 frames, and to route frames to their intended destinations. In ad- 168 dition, the NWK is in charge of the discovery and maintenance 169 of routes between devices. This is achieved by discovering one- 170 hop neighbors and storing relevant neighbor information. In a 171 star topology, the network is controlled by one single device 172 called the ZigBee coordinator. In mesh and tree topologies, the 173 ZigBee coordinator is responsible for starting a new network, 174 when appropriate, and assigning addresses to newly associated 175 devices, but the network may be extended through the use 176 of ZigBee routers. In tree networks, routers transfer data and 177 control messages through the network using a hierarchical 178 routing strategy. Tree networks may employ beacon-oriented2 179 communication as described in the IEEE 802.15.4-2003 180

1 Full implementation of the protocol stack takes less than 32 KB of memory and up to 64 KB for the network coordinator which requires extra RAM for the node devices database and for transaction and pairing tables. 2 Since the new release of ZigBee specification in 2008, this feature is abandoned for saving space in the MAC layer.

AQ11

Fig. 1.

IE E Pr E oo f

AL AGHA et al.: WHICH TECHNOLOGY FOR WIRELESS SENSOR NETWORK? DEVELOPMENT OF OCARI

ZigBee stack architecture [10].

specification. Concerning ZigBee routing strategy, it is a mixed mechanism composed of a simplified version of ad hoc onDemand Distance Vector (AODV) [9], [12] and tree routing intended to extend the coverage of the network beyond the coverage of each network node. The ZigBee APL consists of the application support 187 sublayer (APS), the ZigBee device object (ZDO) containing 188 the ZDO management plane, and the manufacturer-defined 189 application objects. The responsibilities of the APS include 190 maintaining tables for binding, which is the ability to match 191 two devices together based on their services and their needs, 192 and forwarding application messages between bound devices. 193 The responsibilities of the ZDO include defining the role of the 194 device within the network (ZigBee Coordinator, Router, or End 195 device), discovering devices on the network and determining 196 which application services they provide, initiating and/or 197 responding to binding requests, and establishing a secure 198 relationship between network devices. More details about 199 NWK and APL may be found in [10]. 200 IEEE 802.15.4 specifies two medium-access methods: unco201 ordinated mode and coordinated mode (beacon-enabled mode) 202 in which a coordinator (called the PAN-coordinator) regularly 203 sends beacons to synchronize the network nodes. In coordinated 204 mode, the PAN coordinator does not need to listen all the time 205 to the communication channel. However, ZigBee has chosen 206 the use of the uncoordinated mode that requires the ZigBee 207 coordinator to listen permanently to the channel and, thus, 208 wastes the coordinator battery. Energy is saved in ZigBee by 209 allowing a very low power consumption for the end devices in

181 182 183 184 185 186

3

“doze” mode (lower than 10 μA), and letting them switch to 210 normal operating mode in less than 300 μs. 211 In spite of these characteristics of ZigBee, this standard 212 does not satisfy the very constrained battery autonomy of the 213 wireless sensor networks in application such as environmental 214 monitoring in large industrial facility. Indeed, as ZigBee does 215 not retain coordinated mode, only ZigBee End Devices (IEEE 216 802.15.4 Reduce Function Devices) can be put to a doze mode. 217 ZigBee Routers and Coordinators need to be always awakened 218 (active mode) in order to listen to the communication channel. 219 Crossbow and Telecom Italia have submitted an extension, 220 called low-power active router protocol for adoption in ZigBee 221 Pro 2009. It concerns the NWK and proposes some improve- 222 ments such as periodic listening to reduce radio duty cycle 223 and wake-up message that the receiver periodically sniffs for 224 a wake-up signal and then waits for data, otherwise, it sleeps 225 at the end of the wake-up period [13]. Another limitation of 226 ZigBee is that it does not directly support device mobility. 227 1) AODV only discovers the route on demand and the only 228 used QoS is the instantaneous radio link, thus, route repair 229 is done on error. The complete route-discovery process 230 can take a significant time (up to 10 s). This is not useable 231 in unstable topology in which network nodes regular- 232 ly move. 233 2) In ZigBee 2006 specification, when the link to the parent 234 node fails, a reassociation is required, and a new network 235 address is attributed to the concerned child node depend- 236 ing on its position in the network tree. This does not 237

AQ12

4


241 242 243

work for sleeping End Devices which might not receive the broadcast message signaling the new network-address renumbering. This limitation has been removed by the ZigBee Pro specification since its release in January 2008 [11]: The network nodes keep their existing addresses, but new routes have to be rediscovered.

244 245 246

In spite of these properties, ZigBee does not offer support for sink mobility [14] in which data-collector points travel through the nodes of a wireless sensor network.

247

B. IEEE 802.15.4-2006-Based ISA100.11a Specification

238 239 240

ISA100.11a is part of ISA100, a family of standards of wireless systems for industrial automation, which results from converging efforts of defining industrial wireless standards from different organizations and alliances such as WINA, 252 NCCR-MICS, NSF-Program on Sensors and Sensor Networks 253 and HCF. ISA100.11a Working Group3 aims at defining all 254 specifications including security and management for wireless 255 devices serving various application classes [15]. The focus 256 here is to address performance needs for periodic monitoring 257 and process control where latencies on the order of 100 ms 258 can be tolerated with optional behavior for shorter latency. 259 ISA100.11a specification is still under development. This re260 view is based on published presentations and on current works 261 from the ISA100.11a WG. 262 ISA100.11a key features aim at responding to the following 263 requirements: 264 1) The ability to serve process-industry applications without 265 excluding factory automation; 266 2) In-plant and near-plant use; 267 3) Technology to address different traffic class; 268 4) A single APL providing both native and tunneling proto269 col capability for broad usability; 270 5) The addressing of 2.4 GHz IEEE 802.15.4-2006 PHY 271 layer devices; 272 6) A comprehensive coexistence strategy with channel hop273 ping to support coexistence (with IEEE 802.11) and 274 increase reliability; 275 7) Simple, flexible, and scalable security addressing ma276 jor industrial threats leveraging IEEE 802.15.4-2006 277 security; 278 8) Field-device meshing and star capability. 279 ISA100.11a stack architecture has more layers than the Zig280 Bee one. It lets the device-management application process 281 (DMAP) directly access (using service access point) to the 282 Data-Link Layer, the Network Layer, the Transport Layer, 283 and the Application Sublayer in order to manage the de284 vice and its communication services. The DMAP is a spe285 cial type of user application process that provides a basis 286 for building system-management-configuration application and 287 communication-monitoring application. Either a distributed or 288 centralized system manager is supported in ISA100.11a. 289 ISA100.11a addresses failed communications using 290 frequency- and slotted-hopping architecture by adding a MAC 248

IE E Pr E oo f

249 250 251

Extension Shim [16] to IEEE 802.15.4 MAC, whereas a slow 291 frequency-hopping schema (frequency agility) is adopted in 292 ZigBee-2007 MAC, which is designed primarily for operation 293 in a fixed-channel network. 294 Two main classes of devices are defined in ISA100.11a: 295 Field Devices and Backbone Devices. A Field-Device class can 296 have devices with (i.e., IEEE 802.15.4 Full Function Device) 297 and without routing capability (i.e., IEEE 802.15.4 Reduced 298 Function Device). For example, a handheld device is considered 299 as a nonrouting field device. The typical mode of operation of 300 an SP100.11a handheld device is to attach to a full-function 301 device and to communicate data or monitor network traffic. 302 Roaming of handheld device is not supported by ISA100.11a. 303 Backbone devices are full-function devices which are contin- 304 uously powered, whereas field devices have limited battery 305 power (without routing capability) or moderate power (with 306 routing capability). 307 Network time and time-synchronization information for de- 308 vices on the network are provided by the system manager, a 309 particular backbone device, which acts as clock source. 310 Routing in ISA100.11a is based on graphs using a directed 311 list of links that connect devices. The links associated with each 312 graph are configured by the system-management function on a 313 centralized or decentralized basis. A single network instance 314 may have multiple graphs, some of which may overlap. Each 315 device may have multiple graphs going through it, even to the 316 same neighbors. Each device data-link layer service has one 317 route associated with it. 318 It is difficult to make some critical analysis of this evolving 319 specification. However, one may consider the technical diffi- 320 culties to implement the full stack architecture on low cost 321 hardware. In terms of software architecture, ISA100.11a WG 322 has produced a relatively complete functional specification 323 for system management that is almost absent from ZigBee 324 standard. Coexistence with WiFi is also taken into account 325 in ISA100.11a, and that is really appreciated in industrial 326 environment. Currently, there is no optimized routing strategy 327 for maximizing the lifetime of field network. 328

3 Composed

of industrial end users and technology providers.

C. IEEE 802.15.4-2006-Based WirelessHART Specification

329

WirelessHART is specified in HART protocol specification 330 revision 7. The HART protocol is a digital-communication 331 technology that is designed for process measurement and con- 332 trol devices. 333 WirelessHART is an optional HART PHY, data link and net- 334 work layers. More details about WirelessHART can be found in 335 [17]. Compared with ZigBee and ISA100.11a, WirelessHART 336 has adopted time-division multiple access as an access method 337 to the communication medium in order to offer an equivalent 338 of token passing procedure in wired HART. This gives some 339 deterministic behavior and guaranteed bandwidth. 340 WirelessHART does not specify, however, any energy-aware 341 ad hoc routing strategy in its network layer. 342 Before describing our OCARI project, let us summarize the 343 important features of ZigBee, WirelessHART, and ISA100.11a 344 specifications, as shown in Table I. 345

AQ13

AQ14


TABLE I ZigBee, WirelessHART, AND ISA100.11a P ROTOCOL L AYERS C HARACTERISTICS

2) Condition-based maintenance of mechanical and electri- 367 cal components in power plant as well as in warship; 368 3) Environmental monitoring in and around power plant; 369 4) Structure monitoring of hydroelectric dams. 370 In the following paragraphs, we will describe the main 371 specification points of OCARI. 372 A. OCARI Network Topology

III. OCARI S PECIFICATION

Regarding the aformentioned features of ZigBee, WirelessHART, and ISA100.11a, we have focused our work on improving the ZigBee standard by developing a complementary 350 industrial specification called OCARI, rather than creating a 351 new one from scratch. It aims at responding to the following 352 requirements which are particularly important in power gener353 ation industry and in warship construction and maintenance:

347 348 349

360 361 362

1) Support of deterministic MAC layer (at least inside a cell, see Fig. 2) for time-constrained communication; 2) Support of optimized energy-consumption routing strategy in order to maximize the network lifetime within the nontime-constrained communication period; 3) Support of human walking-speed mobility for some particular network nodes (sinks); 4) Support of HART application layer APL (and EDDLIEC61804).

363 364

The development of OCARI targets the following industrial applications:

365

1) Real-time centralized supervision of personal dose in nuclear power plant;

354 355 356 357 358 359

366

373

Used-case analysis of the described industrial applications 374 shows that in most cases, the network topology can be modeled 375 as shown in Fig. 2. OCARI End Device is a “radio-fixed” 376 network node [i.e., its position varies very little compared with 377 its initial location so that its radio link is always managed by 378 the same cell coordinator (CC)]. It is a Reduce Function Device 379 as defined in IEEE 802.15.4 specification. OCARI Cell and 380 Workshop Coordinators are Full Function Devices as specified 381 in IEEE 802.15.4. They are fixed devices in the infrastructure 382 and are equivalent to ZigBee Coordinator. The functions of the 383 CC consist of the following: 384 1) Coordinating the intracell network nodes using a star 385 topology; 386 2) Routing data packets in push mode from end-device 387 network nodes (i.e., sensors in our industrial applications) 388 to the Workshop Coordinator per Workshop domain. 389 Tree routing [18], [19] is used in a time-constrained 390 period and an energy-aware OLSR [20] is used otherwise 391 between CCs. 392 Workshop domain is a permissive volume (delimited by a 393 threshold of the BER, the SINR, or the received signal strength 394 indication [21]) to electromagnetic wave that is covered by a 395 unique Workshop-Coordinator network node. 396 Workshop Coordinator is a gateway (even in polling 397 mode used by most of supervisory control and data 398 acquisition—SCADA—applications) between the wireless sen- 399 sor networks that resided in a Workshop domain and the indus- 400 trial facility backbone. 401 As shown in Fig. 2, Sink node is a mobile-network node 402 which usually represents a patrolman/maintenance operator, 403 equipped with a personal digital assistant, collecting data from 404 sensors inside a cell. The Sink leaves the cell when it finishes 405 to acquire data (in polling mode). 406 Time server is a particular network node which is used for 407 clock synchronization in the whole Workshop domain. For an 408 accurate clock synchronization, IEEE-1588/IEC 61588-2004 409 protocol [22], [23] (Precision Time Protocol, PTP) is used. 410

IE E Pr E oo f

346

5

B. OCARI PHY and MAC Layers

411

Previous studies [24], [25] and our field tests show that 412 IEEE 802.15.4 PHY is very robust. Based on the robustness 413 of IEEE 802.15.4 PHY 2.4 GHz, we decided to adopt it for the 414 OCARI specification. However, as stated earlier, IEEE 802.15.4 415 MAC layer does not completely satisfy our requirements. Our 416 consortium is developing a derived MAC layer [18], [19] that 417 offers different access methods to the medium. 418 1) Global MaCARI Protocol: The context of industrial ap- 419 plications brings many challenges for the MAC protocol, such 420

AQ15

AQ16


IE E Pr E oo f

6

Fig. 2. OCARI network topology.

as deterministic communications or low energy consumption. In order to be suitable to the large variety of application requirements, we proposed a flexible MAC protocol, MaCARI, 424 which is introduced in [18]. 425 As with many MAC protocols for wireless sensor networks, 426 MaCARI includes some multihop routing features. 427 Justifications of main ideas of MaCARI: A deterministic 428 MAC layer has to guarantee an access to the radio channel for 429 each device of the WSN every certain period of time (which 430 is the global cycle of Fig. 3). An energy-efficient MAC layer 431 has to make all the network elements sleep as often as possible 432 during the global cycle. 433 We designed the MAC protocol for the OCARI project, 434 denoted by MaCARI, in order to achieve a tradeoff between 435 these two main features: determinism and energy efficiency. 436 The routing features of MaCARI can be explained using 437 an analogy to the return of salmons to the region where they 438 were born in. When it is too difficult for the salmon to swim 439 upstream, a fish pass can help the salmon by providing an 440 alternate path, which is easier but generally longer. Similarly, in

421 422 AQ17 423

Fig. 3.

OCARI global cycle.

MaCARI, frames can be forwarded directly (to the destination 441 node or to a chosen relay node) or indirectly, by following a 442 path by default. The indirect forwarding corresponds to the fish 443 pass, where the path by default is a spanning tree of the network 444 topology having the PAN coordinator as its root. 445 We decided to have two periodic disjoint phases, each of 446 them corresponding to a type of forwarding (direct or indirect). 447

AQ19

To do so, a synchronization of the network devices is necessary. This leads us to the design of a global cycle constituted of three periods: the synchronization, the scheduled activities, and the 451 unscheduled activities periods. These three periods are shown 452 on Fig. 3. 453 Specifications of three MaCARI periods: MaCARI uses 454 the PAN coordinator (usually the workshop coordinator) to 455 manage the global cycle. The synchronization is performed 456 using a single beacon broadcasted in the network in a hop-by457 hop manner. In order to avoid beacon collisions [29], each CC 458 repeats the beacon at a time decided by the PAN coordinator 459 so that no beacons are transmitted simultaneously. This beacon 460 carries the order in which the CCs have to transmit their beacon, 461 which is computed by the PAN coordinator. This sequence 462 of beacon transmissions is triggered by the PAN coordina463 tor at T 0. At T 1, all the network devices have a common 464 time reference. 465 The scheduled activities period, which lasts between T 1 466 and T 2, divides time into disjoint slots. The PAN coordinator 467 allocates to each cell a given number of slots (usually one). 468 During the slot of a cell, the active entities are the following: 469 the CC of the cell, the end devices of this cell, and the father 470 coordinator of the CC. During this slot, the CC is in charge 471 of the communications. The way the slot is used is described 472 in Section III-B2. Toward the end of the slot, the CC can ex473 change data with its father coordinator using polling/selecting. 474 By having attributed disjoint time slots to the cells, MaCARI 475 guarantees no intercells interferences and a dedicated channel 476 between each father and child. 477 The unscheduled-activities period takes place during T 2 and 478 T 3. During this time interval, MaCARI manages the activ479 ity of nodes according to SchEdule RoutEr Nodes Activity 480 (SERENA) requests (see Section III-D3), and tries to send 481 frames according to routes given by the energy-efficient OLSR 482 (EOLSR) routing protocol (see Section III-D2). During this 483 period, time slots are given according to a sequence of colors 484 (see the part on the SERENA protocol). All the nodes of the 485 current color can transmit simultaneously in an independent 486 way. During this period, MaCARI does not decide any spe487 cific scheduling, which is why we refer to this period as the 488 unscheduled-activities period. 489 Energy-efficiency summary: During the synchronization pe490 riod, a device of a given cell can sleep as soon as the CC of 491 its cell has transmitted the beacon, until the time when it has 492 to be active (which occurs at some point between T 1 and T 2). 493 During the scheduled-activities period, all the end devices of a 494 cell sleep, except when their cell is active. When a cell is active, 495 the father coordinator of the CC is also awaken to provide 496 father–son exchanges. During the unscheduled activities period, 497 all end devices sleep, but the activity of the CCs depends on the 498 SERENA protocol. 499 Determinism summary: Once every global cycle, each end 500 device has access to the channel during the time slot of its cell. 501 Once every global cycle, the coordinator manages the commu502 nications with its end devices according to strategies described 503 in Section III-B2. Each pair of father–son coordinators has a 504 dedicated channel (i.e., a part of the time slot) to route frames 505 on the default routing path. 448

449 450

7

The length of the global cycle determines the reactivity of the 506 protocol. The duration of [T 1;T 2] impacts on the deterministic 507 behavior of MaCARI, while the duration of [T 2;T 3] allows 508 the use of simultaneous activities, and, therefore, increases the 509 throughput of the network. 510 2) Intracellular MaCARI Protocol—An Incremental Proto- 511 col With Several Options: This part focuses on the intracellular 512 MaCARI protocol. Our proposal is an incremental protocol 513 with three different options which increase the previous one in 514 terms of bandwidth and energy saving. The MAC layer inside 515 the cell is used when the CC is authorized to communicate 516 since it has received a beacon of its father. During [T 1;T 2], the 517 cell owns a guaranteed medium access, i.e., without collision 518 or interference risks against the other nodes of the OCARI 519 network. 520 Slotted CSMA/CA in beacon-enabled mode [28]: The 521 simplest intracellular medium access method is based on 522 the well-known slotted carrier sense multiple access/collision 523 avoidance (CSMA/CA). Every slot allocated between T 1 and 524 T 2 corresponds to a superframe (as in IEEE 802.15.4 vocabu- 525 lary). Each node of the cell uses the same distributed protocol. 526 As an end device, a node can only exchange with its CC. Data 527 for end devices are transmitted by the CC during a timeslot 528 requested by the end device. Concurrent accesses are avoided 529 by the CSMA/CA method. This best effort MAC is relevant for 530 a lightly loaded cell with regard to the QoS requirements. We 531 have done a simulation and an implementation of this slotted 532 CSMA/CA protocol. Some performance analyses are presented 533 in [18], [33], and [34]. 534 CSMA/CA + GTS [28]: The random characteristic of 535 the CSMA/CA implies a collision risk between two or more 536 devices of the same cell. In fact, quasi-simultaneous clear- 537 channel assessment function calls by concurrent transceivers 538 entails a collision. The consequences of the phenomenon could 539 be insignificant for most applications but could be disastrous 540 when the submitted network load is important. Indeed, the 541 more frames to transmit, the more collisions and the more 542 frames to retransmit. A collapse phenomenon appears which 543 implies a low-bandwidth usage. Delays are also increased. This 544 can be unacceptable for time-constrained applications. This 545 problem can be avoided by using, in each cell, a specific access 546 method using a superframe with two periods: contention-access 547 period (CAP) and contention-free period (CFP). CAP uses only 548 CSMA/CA. This period permits a basic QoS for nonpriority 549 flows without temporal guarantees. This best effort algorithm 550 can be used for sporadic and unexpected flows. CFP uses some 551 dedicated timeslots allocated by the CC to its own end devices. 552 These slots are named guaranteed time slot (GTS). 553 According to the application needs of each sensor, an end 554 device requests to its coordinator a GTS allocation for the 555 next superframe. If the minimal size of the CAP has not 556 been reached, the CC can allocate a new GTS. The CAP is 557 reduced accordingly. If an end device does not use its GTS 558 during a specific duration, this GTS is automatically suppressed 559 (by timeout). The end device must request for a new GTS 560 attribution in the next superframes. The IEEE 802.15.4 standard 561 has proposed this algorithm, but it is rarely implemented on 562 available devices on the market or on evaluation kits. We have 563

IE E Pr E oo f

AQ18


AQ20

8


variation according to bandwidth and delay needs. To avoid an 622 overload of the used bandwidth, a large n is associated to this 623 first PDS. An analytical performance analysis [5] shows that 624 the bandwidth used for this PDS is small (0.78% for a PDS 625 with n = 3). At the moment, we also have implemented this 626 option which offers the best energy efficiency. Indeed, there 627 is no more energy lost by the end device to request GTS, 628 since there is no transmission of GTSrequest in the case of 629 the PDS. 630 Conclusion for intracellular MaCARI protocol: Intracel- 631 lular MaCARI proposes a core based on beaconed-slotted 632 CSMA/CA and three other optional functionalities. This core 633 should be satisfactory in most cases when only a low baud 634 rate is necessary. The second option (IEEE 802.15.4 GTS) 635 may propose a first level of determinism. The new option 636 GTS(n), which requires GTS, proposes a higher QoS flexibility 637 owing to a mechanism of service differentiation. Moreover, 638 the energy saving resulting from this option is previously 639 required by the OCARI project which aims to propose a MAC 640 layer with energy-saving functionalities. New option based 641 on PDS enables a higher level of determinism. These vari- 642 ous options are increasingly effective with regard to energy 643 optimization. 644 C) OCARI Energy-Autonomy Model: The majority of 645 OCARI nodes operate on batteries and are energy dependent. 646 The lifetime of a node depends on the lifetime of its battery. 647 Indeed, to be scalable, replacing a battery when its functioning 648 becomes critical is not permitted because of the required hu- 649 man overhead. The solution that OCARI proposes is to create 650 energy-efficient solutions that reduce the battery utilization of a 651 node and transfer traffic to nodes with more remaining energy 652 in their batteries. 653 In sensor and ad hoc networks area, many works were 654 considered a simple energy model for node operations (routing, 655 processing, . . .). The idea is to suppose an initial amount of 656 energy that is mostly equal for all nodes and then applying a 657 subtraction for every operation. Different levels of consumption 658 were calculated according to the specific operation (transmis- 659 sion, reception, waking up, . . .). Then, when the energy level 660 becomes low, the node is kept away and works only for elemen- 661 tary and basic operations to avoid its death and its elimination 662 from the network. This model seems to be linear and is, in 663 fact, far from the reality. The battery manufacturers [30], [31] 664 display the discharge curves of their products which is close 665 to a constant discharge until a critical point where the battery 666 is dead. 667 Our goal in OCARI is to integrate the energy cost of the 668 functions performed by a node in order to estimate its remaining 669 energy level. This level serves, as indicated in the next section, 670 as a metric for the routing protocol. Nodes select routes to desti- 671 nation according to the energy cost estimated by our model. The 672 idea consists in affecting an initial value of energy to a battery 673 or a node and then subtracting from this value an amount that 674 corresponds to the given operation (transmission, reception, 675 wake up, . . .) and also to the date of this operation. The date 676 takes into account how many times the battery is used. The 677 more the battery is used, the lesser its energy replenishes. We 678 define an exponential function where the slope corresponds to 679

IE E Pr E oo f

implemented this GTS option in a prototype, and performance analyses are in progress. Experimentation results from a small configuration have 567 shown that GTS option offers a better temporal stability com568 pared with the CSMA/CA option. For example, the jitter on the 569 medium-access delay is almost nonexistent with the GTS. This 570 jitter varies between 4 and 11 ms in CSMA/CA (respectively, 571 from one to five packets by superframe transmitted by four 572 end devices in a star topology). In addition to its deterministic 573 character and the low jitter, another advantage concerns the 574 energy economy. Indeed, it is possible to make end devices 575 sleep except their GTS, which is more difficult to realize using 576 CSMA/CA. 577 CSMA/CA + GTS(n) (with multiple reservation levels): 578 The previous solution is an interesting option, but its major 579 drawback is the GTS static allocation for each superframe. 580 For example, one GTS by superframe could be useless for a 581 simple temperature sensor. It would be interesting to propose 582 a service differentiation according to application communica583 tion end-device needs. This new intracellular MaCARI option 584 proposition is based on a special reservation level named n. 585 A GTS is dedicated to an end device according to its period586 icity request: when n = 0: for each superframe (as in IEEE 587 802.15.4 standard), when n = 1: every two superframes, and, 588 in the general case: every 2n superframes. The main advantage 589 is the cohabitation of different guaranteed traffics according 590 to different sensors. It is also possible to choose a reverse591 allocation principle. In a superframe, a sensor could have more 592 than one GTS. In this case, the reservation level is n = 0. 593 However, the end device must request several GTS in each 594 superframe. The bandwidth allocated to such an end device is 595 increased. 596 The other advantage of this GTS(n) option is the power 597 saving offered by this MaCARI layer. A sensor can commute 598 to doze mode, particularly when this sensor is not concerned of 599 these superframes. If a temperature sensor has a high inertia, 600 its end device can wake up only every four or eight super601 frames for a fast and short temperature transmission. After this 602 activity, the end device commutes to battery-saving mode. In 603 the classical IEEE 802.15.4 protocol, a wake up is mandatory 604 for every superframe to keep the GTS. In our proposition, 605 it is possible to save timeslots with an optimized allocation 606 only when it is necessary. This option maximizes doze mode 607 outside GTS. We are currently implementing this option in the 608 prototype. 609 CSMA/CA + GTS(n) + PDS: The principal drawback 610 of the previous option (also in the IEEE 802.15.4 standard) 611 is related to the GTSrequest which is necessarily done by 612 the end device in CSMA/CA mode. This request mode is 613 not fully deterministic because this GTS request access is 614 nonguaranteed. For critical sensor applications which need 615 guaranteed access, we propose a new intracellular MaCARI 616 option based on previously dedicated slot (PDS). A PDS is 617 allocated to every known specific sensor which wants to use 618 this high level of QoS. A PDS is in fact an a priori GTS. 619 This PDS can be used by an end device to transmit periodic 620 data with high QoS. The end device can also use this PDS to 621 request more or less GTS. This dynamic GTS allocation permits 564

565 566


Fig. 4. Battery self-discharge model (Max = 200; const = 104 ; t1 is chosen randomly between 0 and 58 ms). 680 681

9

Fig. 5. Distribution of node energy in the transmit and receive states.

the date and the amplitude to the quantity of energy consumed by the specific operation Rr (0) = Max Rr (i) = Rr (i − 1) − C(i)e−const i

IE E Pr E oo f

Δ

where Rr (i), i ∈ N remaining energy in a node at date i; Max maximum energy for a node; 685 C(i) consumed energy during the step i; 686 Δ duration in seconds of step i; 687 Const constant. 688 The value of C(i) is calculated according to the ZigBee 689 standard where

682

683 684

Etx = 30.2 × 10−6 mA · H + 8.0556 × 10−6 × t1 690

for packet transmission

Erx = 40.986 × 10−6 mA · H + 8.0556 × 10−6 × t1

for packet reception, etc. Note that a node is consuming more energy for data reception than for data transmission. Simulation results (Fig. 4) show the discharge function ac695 cording to the time when a node transmits or receives data. 696 The duration of operation were taken following a uniform 697 distribution. Those results show a better behavior of a discharge 698 than a linear model and fit better with real measurements that 699 we realized in a real environment. 700 D) OCARI Network Layer: 701 1) Energy-Efficient Strategies: The energy-constrained 702 nature of OCARI nodes requires the use of energy-efficient 703 strategies to maximize network lifetime. We can distinguish 704 four types of energy-efficient strategies [32]: 705 1) Topology control algorithms where a node adjusts its 706 power transmission instead of transmitting with the max707 imum power; 708 2) Algorithms that reduce volume of information transferred 709 by aggregating information, optimizing network flooding, 710 avoiding useless transmissions, or tuning the period of 711 control messages; 691

692 693 694

Fig. 6. Distribution of node energy in the idle, overhearing, and interference states.

3) Energy-efficient routing algorithms select routes that min- 712 imize the energy consumed by the end-to-end transmis- 713 sion of a message and avoid nodes with low residual 714 energy; 715 4) Scheduling node-activity algorithms allow nodes to sleep 716 in order to spare energy, provided that the network and 717 application functionalities are still ensured. 718 In the first version of OCARI, all nodes transmit with 719 the maximum power. Hence, all energy-efficient-strategy cat- 720 egories, except the first one, are used in OCARI. In this section, 721 we detail the two modules comprising the network layer: the 722 energy-efficient routing called EOLSR and the scheduling of 723 router nodes activity called SERENA. 724 2) EOLSR: EOLSR is an energy efficient extension of 725 the OLSR routing protocol. OLSR [33] is a proactive protocol 726 based on link state. It has been tuned to wireless ad hoc 727 networks by the concept of multipoint relays (MPRs). MPRs 728 reduce the amount of links advertised in the network and the 729 number of transmissions needed to advertise an information in 730 the network. OLSR consists of two basic functionalities. 731 1) Neighborhood Discovery: Each router node acquires the 732 knowledge of its one- and two-hop neighbors by exchang- 733 ing periodic Hello messages. It independently selects its 734 MPRs among its one-hop neighbors in order to cover all 735 its two-hop neighbors. 736

Fig. 7. OCARI stack.


IE E Pr E oo f

10

2) Topology Dissemination: A node selected as MPR sends a Topology Control message in the network; this message is forwarded by the MPRs if and only if it is the first 740 receipt. Each router node uses the information included in 741 the Topology Control messages to determine the shortest 742 route to any destination in the network. We can then 743 notice that in OLSR, the intermediate nodes of a route 744 are MPR nodes. 745 As the selection of MPRs does not take into account the 746 residual energy of nodes, and as the shortest route is not 747 always the route consuming the least energy, OLSR does not 737

738 739

maximize network lifetime. EOLSR has been introduced for 748 this purpose. It keeps the principles of OLSR. The classical 749 selection of MPRs is kept to optimize broadcasts in the net- 750 work, whereas new ones, called extended MPRs (EMPRs), 751 taking residual energy of nodes into account, are used to build 752 routes. The route selection is also modified in order to select 753 the route that minimizes the energy dissipated by the end- 754 to-end transmission of a packet, including the energy lost in 755 transmitting, receiving, overhearing, and in interferences. If 756 several routes dissipate the same energy, the shortest one is 757 chosen. 758

AQ21

Fig. 8.

IE E Pr E oo f

AQ22


OCARI application architecture.

EOLSR does not require additional messages, existing messages of OLSR are extended to include the residual energy of 761 nodes, computed as explained in Section III-C. For more details 762 on EOLSR, the reader can refer to [20]. 763 3) SERENA: SERENA deals with scheduling of router764 nodes activity; it is the second module comprising the network 765 layer. In order to spare energy, a node must sleep. However, 766 it must be awake to receive messages. Hence, a coordination 767 of nodes is required. In SERENA, this coordination is done as 768 follows: A node is awake only during its slots (i.e., the slots 769 during which it can transmit) and the slots granted to its one-hop 770 neighbors. Slots are assigned to a node according to its color 771 and its traffic. 772 As interferences are limited to two hops, a two-hop coloring 773 algorithm is used. (In OCARI specification, the interference 774 model is based on two times the estimated radio range.) The 775 coloring algorithm is decentralized: A node colors itself if and 776 only if all the nodes up to two-hop with a higher priority are 777 already colored. The node selects the smallest color available 778 up to two hop. The priority of a node is the couple (cardinality 779 of the up to two-hop neighborhood, node identifier). In the first 780 version of OCARI, a simple slot assignment is provided, using 781 cross-layering mechanisms. A more sophisticated one can be 782 found in [26]. 759 760

11

When nodes move, conflicts of colors can occur: Two nodes 783 that are either one- or two-hop neighbors have the same color. 784 Such conflicts are detected and solved: The node with the 785 highest priority keeps its color, whereas the other selects the 786 smallest color available up to two hop. 787 4) Performance Evaluation: We now quantify the ben- 788 efits brought by EOLSR and SERENA, used separately and 789 then, in combination, in a wireless ad hoc network. As the 790 MaCARI layer is under implementation, the results shown in 791 Figs. 5 and 6 have been obtained with an IEEE 802.11 network 792 at 2 Mb/s; the powers used in the different states are: 1.3 W 793 in the transmit state, 0.9 W in the receive state, 0.74 W in 794 the idle state, and 0.047 W in the sleeping state. There are 795 150 nodes with a network density (i.e., average number of 796 one-hop neighbors) of ten. The initial energy of the nodes is 797 equal to 100 J. User traffic consists of 30 flows, with randomly 798 chosen sources and destinations, and a throughput of 16 Kb/s. 799 Message size is 512 B. Messages of the routing protocol are not 800 taken into account. Each result is the average of five simulation 801 runs. 802 As expected, Figs. 5 and 6 show that with SERENA and 803 EOLSR, less energy is dissipated in the idle and interference 804 states leaving more energy for transmitting and receiving mes- 805 sages. This justifies the choice of the EOLSR and SERENA 806

12


algorithms. Both algorithms contribute to a more efficient use of node energy. Moreover, we have also shown that EOLSR alone, SERENA alone, and EOLSR+SERENA increase the network 810 lifetime by 40%, 195%, and 275%, respectively. Notice that 811 we obtain similar results for the amount of data delivered by 812 EOLSR and SERENA used separately or in combination. This 813 result highlights the true gain obtained by these protocols: the 814 application benefits from this increased network lifetime by 815 exchanging more user data. 816 In OCARI, the MAC layer provides an immediate acknowl817 edgment of any unicast message received by its destination. 818 That is why SERENA is now evolving toward a three-hop 819 coloring algorithm. We expect simulation results with three-hop 820 coloring to be similar to those shown in Figs. 5 and 6, allowing 821 us to conclude to the better energy efficiency of SERENA and 822 EOLSR. 823 E) OCARI Stack Architecture: From the work of the differ824 ent partners, an architecture is defined to organize the develop825 ments of each layer (Fig. 7). 826 This architecture reflects also the choice for prototyping the 827 system on two boards, each containing its own microcontroller. 828 The first board is for the PHY and the MAC layers, and the 829 second board is for the upper layers. It implies defining a 830 hardware interface between the MaCARI layer and the upper 831 ones allowing the configuration of radio parameters, the transfer 832 of data, as well as the interactions between SERENA and 833 MaCARI for energy management. 834 As the goal is for this stack to be implemented in low-cost 835 microcontrollers, it is kept as simple as possible. 836 F) OCARI Platform Prototype: In order to evaluate, verify, 837 and validate the OCARI specification, we define the following 838 platform in which we will implement different communication 839 layers. The PHY and MAC layers are implemented inside the 840 same communication controller module (One-RF B2400MC841 Tiny), designed by our partner, Telit. 842 OCARI routing strategy and the ZigBee NWK and APS are 843 implemented on another module, which is a PC platform for 844 ease testing and debugging. On the final architecture, the code 845 is put on a different microcontroller. 846 G) OCARI Application Architecture: To achieve a seam847 less integration of wireless sensor networks into real-world 848 applications in industrial information systems, we need to 849 develop and provide an application architecture that can in850 teroperate with existing industrial standards. The architecture 851 shown in Fig. 8 aims at responding to such requirement 852 while supporting the state of the art in industrial information 853 technology. 854 The role of a WSN-oriented middleware is to provide stan855 dard and homogenous services to user applications. It also 856 contributes to the energy saving of the WSN by implementing 857 a centralized and optimized management of network resources 858 [27]. Moreover, it serves as a gateway between different user 859 applications and different Workshop Coordinators, which man860 age the attached cells. 861 The choice of a software-bus standard such as OPC-data 862 access and the upcoming version OPC-unified architecture 863 allows us to provide the compatibility with EDDL for asset 864 management and for SCADA application. 807

IV. F UTURE W ORK

AQ23

AQ24 AQ25

865

In the next steps of our project, we will be working on the 866 implementation and the extension of different components of 867 our specification. We believe that OCARI fills some gaps, as 868 stated earlier, in ZigBee while trying to be compatible with 869 ZigBee APS and APL layers. In this way, we can preserve 870 existing efforts and investments on application development 871 already realized using ZigBee technology. 872 R EFERENCES [1] G. Breed, “Bit error rate: Fundamental concepts and measurement issues,” High Freq. Electron., vol. 2, no. 1, pp. 46–47, Jan. 2003. [2] D. R. Jeske and A. Sampath, “Signal-to-interference-plus-noise ratio estimation for wireless communication systems: Methods and analysis,” Nav. Res. Logist., vol. 51, no. 5, pp. 720–740, 2004. [3] E. Zimmermann, “Assessment of Radio-Link Technologies,” IST-2003507581 WINNER, D2.3 ver. 1.0. [4] Feb. 2008. [Online]. Available: http://www.hartcomm2.org/hart_protocol/ wireless_hart/wireless_hart_main.html [5] A. Van Den Bossche, “Proposition d’une nouvelle méthode d’accès déterministe pour un réseau personnel sans fil à fortes contraintes temporelles,” Thèse, Université Toulouse le Mirail—Toulouse II, Toulouse, France, Jul. 6, 2007. [6] OCARI Consortium Web Site. [Online]. Available: http://ocari.lri.fr [7] RFC3626 T. Clausen and P. Jacquet, Optimized Link State Routing Protocol (OLSR). [Online]. Available: http://www.ietf.org/rfc/rfc3626.txt [8] V. Gauthier, R. de Rasse, M. Marot, and M. Becker, On a Comparison of Four Ad-hoc Routing Protocols When Taking Into Account the Radio Interferences. [Online]. Available: www.comp.brad.ac.uk/het-net/HETNETs05/ReadCamera05/P42.pdf [9] C. E. Perkins, E. M. Belding-Royer, and S. Das, “Ad hoc on Demand Distance Vector (AODV) Routing,” IETF RFC 3561. [10] ZigBee Specification-Document 053474r17. [Online]. Available: http://www.zigbee.org [11] ZigBee-PRO Stack Profile, ZigBee Document 074855r05, Jan. 2008. [12] A. Wheeler, “Commercial applications of wireless sensor networks using ZigBee,” IEEE Commun. Mag., vol. 45, no. 4, pp. 70–77, Apr. 2007. [13] M. Turon and S. C. Ergen, ZigBee Low Power Router Baseline Proposal, Feb. 20, 2008, Rev. 6. [14] J. G. Castaño, “Algorithms and protocols enhancing mobility support for wireless sensor networks based on Bluetooth and ZigBee,” Ph.D. dissertations, Mälardalen Univ. Press, Västerås, Sweden, Sep., 2006. [15] ISA100.11a Release 1 Status. (2007, Oct.). [Online]. Available: www.isa.org/source/ISA100.11a_Release1_Status.ppt [16] Melanie Swiderek/ISA100 Committee, “ISA100.11a Functional Description,” ver. 2007-07-18. [17] HART Protocol Revision 7.0, Aug. 23, 2007. [18] G. Chalhoub, A. Guitton, and M. Misson, “MAC specifications for a WPAN allowing both energy saving and guaranteed delay—Part A: MaCARI: A synchronized tree-based MAC protocol,” in Proc. IFIP Conf. Wireless Sensor Actor Netw., 2008, pp. 221–232. [19] E. Livolant, A. Van den Bossche, and T. Val, “MAC specifications for a WPAN allowing both energy saving and guaranteed delay—Part B: Optimization of the intra cellular exchanges for MaCARI,” in Proc. IFIP Conf. Wireless Sensor Actor Netw., 2008, pp. 233–244. [20] S. Mahfoudh and P. Minet, “An energy efficient routing based on OLSR in wireless ad hoc and sensor networks,” in Proc. IEEE Int. Workshop PAEWN, Ginowan, Japan, Mar. 2008. [21] A. Awad, T. Frunzke, and F. Dressler, “Adaptive distance estimation and localization in WSN using RSSI measures,” in Proc. 10th Euromicro Conf. Digital Syst. Des. Architectures, Methods Tools (DSD), 2007, pp. 471–478. [22] J. C. Eidson, M. Fischer, and J. White, “IEEE-1588 Standard for a precision clock synchronization protocol for networked measurement and control systems,” in Proc. 34th Annu. PTTI Meeting, 2002. [23] Precision Clock Synchronization Protocol for Networked Measurement and Control Systems, Standard IEC 61588-2004, 2004. [24] B. Heile. (2005, Sep.–Nov.). ZigBee Alliance Tutorial. [Online]. Available: http://www.zigbee.org [25] E. Callaway. (2001, Jul.). PHY Proposal for the Low Rate 802.15.4 Standard, [Online]. Available: http://grouper.ieee.org/groups/802/15/ pub/2001/Jul01/01229r1P802-15_TG4-Motorola-PHY-Proposal.ppt

IE E Pr E oo f

808 809

873

874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 AQ26 922 923 924 925 926 927 928 AQ27 929 930 931 932 933 934 935


[26] S. Mahfoudh and P. Minet, “Performance evaluation of the SERENA algorithm to SchEdule RoutEr Nodes Activity in wireless ad hoc and sensor networks,” in Proc. IEEE 22nd Int. Conf. AINA, Ginowan, Japan, Mar. 2008, pp. 287–294. [27] S. Hadim and N. Mohamed, “Middleware challenges and approaches for wireless sensor networks,” IEEE Distrib. Syst. Online, vol. 7, no. 3, p. 1, Mar. 2006, art. no. 0603-o3001. [28] IEEE 802.15: Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs), ANSI/IEEE Standard 802.15.4 R2006, 2006. [29] A. Koubaa, A. Cunha, and M. Alves, “A time division beacon scheduling mechanism for IEEE 802.15.4/Zigbee cluster-tree wireless sensor networks,” in Proc. Euromicro Conf. Real-Time Syst., Jul. 2007, pp. 125–135. [30] [Online]. Available: http://www.mpoweruk.com/life.htm#arrhenius [31] [Online]. Available: http://www.duracell.com/oem/Pdf/others/LithBull.pdf [32] S. Mahfoudh and P. Minet, “Survey of energy efficient strategies in wireless ad hoc and sensor networks,” in Proc. IEEE ICN, Cancun, Mexico, Apr. 2008, pp. 1–7. [33] C. Adjih, T. Clausen, P. Jacquet, A. Laouiti, P. Minet, P. Muhlethaler, A. Qayyum, and L. Viennot, Optimized Link State Routing Protocol, 2003, RFC 3626, IETF. [34] G. Chalhoub, E. Livolant, A. Guitton, A. Van den Bossche, M. Misson, and T. Val, “Specifications and evaluation of a MAC protocol for a LPWPAN,” Ad Hoc Sensor Wireless Netw., vol. 7, no. 1/2, pp. 69–89, 2009. [35] M.-H. Bertin, A. Van den Bossche, G. Chalhoub, T. Dang, S. Mahfoudh, J. Rahmé, and J.-B. Viollet, “OCARI for industrial wireless sensor networks,” in Proc. IFIP Wireless Days Conf., Dubai, United Arab Emirates, Nov. 24–27, 2008, pp. 1–5.

964 965 AQ28 966 967 AQ29 968 AQ30 969 970 971 972 973 974 975 976 977 978

Khaldoun Al Agha (SM’03) received the B.S. degree in engineering from Supelec, Paris, France, in 1993, and the M.S. and Ph.D. degrees from the University of Versailles, Versailles, France, in 1995 and 1993, respectively. He received the HDR from University of Paris-Sud11, Paris, in 2002. In 1998, he was an Assistant Professor with the University of Versailles. In 1999, he was with the French National Institute for Research in Computer Science and Control (INRIA) for one year. He is currently a Full Professor with the University of Paris-Sud11. He created and conducts the networking group at the LRI laboratory. He participates in different projects (BRAIN, SAMU, Arcade, SAFARI, OCARI, SARAH, RAF, NC2, etc.). His research interests are on resource allocation, security, and Quality of Service for cellular and ad hoc networks.

979 Marc-Henry Bertin, photograph and biography not available at the time of

AQ31 980 publication.

Alexandre Guitton (M’09) received the M.Sc. and Ph.D. degrees in the field of computer networks from the University of Rennes I, Rennes, France, in 2002 and 2005, respectively. Since 2007, he has been with Clermont University, Aubière, France, where he is currently an Assistant Professor. His research interests include wireless communications, sensor networks, MAC protocols, and energy-efficiency.

996 997 998 999 1000 1001 1002 1003 1004

Pascale Minet is currently the Vice-Head of the HIPERCOM project with the French National Institute for Research in Computer Science and Control (INRIA) Le Chesnay, France. Her research interests include routing, Quality of Service and multicast in mobile ad hoc networks, energy efficiency, and deployment of wireless sensor networks, as well as real-time scheduling. Ms. Minet serves as program committee member of several international conferences.

1005 1006 1007 1008 1009 1010 1011 1012 1013 1014

IE E Pr E oo f

936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963

981 Tuan Dang (M’06) received the B.S. degree in 982 engineering from the “Ecole Centrale de Nantes,” AQ32 983 Nantes, France, and the Ph.D. degree from the “Ecole 984 Nationale Supérieure des Télécommunications de 985 Paris,” Paris, France. 986 He is the Expert Research Engineer with Simu987 lation and Information Technologies for Power gen988 eration systems (STEP) Department, Electricité De 989 France (EDF) Research and Development, Chatou, 990 France. His research fields are industrial communica991 tion networks, automation and telecontrol of power992 generation systems, home and building automation, and energy management. 993 He has published tens of papers in industrial informatics. 994 Dr. Dang is a member of IEEE Industrial Electronics Society-Building AQ33 995 Automation, Control and Management.

13

Thierry Val received the Ph.D. degree in computer science at Blaise Pascal University, ClermontFerrand, France, in 1993 and the HDR from University of Toulouse II, Toulouse, France, in 2002. He has been with the University of Toulouse where he was a Lecturer when he joined in 1994 and currently teaches networks and hardware systems. He is also the Submanager of the Laboratory of Technology and System Engineering of Toulouse (LATTIS) where he manages a research activity on engineering wireless local networks and related protocols in collaboration with IRIT-CNRS of Toulouse, LAAS-CNRS of Toulouse, LIRMM-CNRS of Montpellier, and LIMOS-CNRS of ClermontFerrand. He is also currently a Professor with the Institute of Tolouse-Blagnac, University of Toulouse.

1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 AQ35

Jean-Baptiste Viollet received the B.S. degree from Polytechnique and from Ponts et Chaussées in early 2007. Since the spring of 2007, he has been with Direction des Constructions Navales Services (DCNS) group, Lorient, France, where he is currently an Engineer in charge of the study and tests of the conduct domain and a Conduct Partner in the project FREMM and is in charge of the OCARI R&D project at DCNS.

1030 1031 1032 AQ36 1033 1034 1035 1036 1037 1038 1039