A_Routing_Scheme_for_the_IEEE-802.15.4-Enabled_Wireless_Sensor_Networks

发布时间:2021-10-14 09:52:35

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

5135

A Routing Scheme for the IEEE-802.15.4-Enabled Wireless Sensor Networks
A. Hafz Shuaib, Student Member, IEEE, and A. Hamid Aghvami, Fellow, IEEE
Abstract—A wireless sensor network (WSN) has features that t into several classes of wireless networks (e.g., mesh, ad hoc, and mobile ad hoc networks) and, at the same time, features that are unique to it. These exceptional characteristics place many demands on the WSN routing protocol. For instance, the routing protocol must assure uniform dissipation of energy across the network, quickly converge irrespective of the network node density, and be exible in terms of the routing framework and the route computation metric. All of the aforementioned conditions must be accomplished in an energy-efcient manner. Although several routing protocols have been proposed for WSNs, most approaches are usually focused on energy-efcient operations. The validity of this case is undeniable; however, one crucial element is generally assumed or ignored i.e., how one can prevent routing loops in the network. In addition to achieving the aforementioned routing objectives, in this paper, we go one step further by expressly dening and thoroughly evaluating mechanisms for loop prevention and minimization. Our proposed routing scheme leverages the services that were offered by the IEEE 802.15.4 specication to satisfy the requirements of a WSN routing protocol. Index Terms—IEEE 802.15.4, routing, wireless personal area networks (WPANs), wireless sensor networks (WSNs).

I. I NTRODUCTION IRELESS sensor network (WSN) applications can be grouped into two broad categories: 1) event based and 2) periodic monitoring [1]. Although these application categories may differ in terms of trafc characteristics and qualityof-service requirements, both categories essentially place the same communication architecture on the underlying network. This architecture is such that communication is generally between WSN nodes and their neighbors or between WSN nodes and the sink nodes. Communication between neighbor nodes is primarily one that allows for efcient data gathering. For example, data with regard to sensed phenomena can be aggregated or pruned to eliminate duplicates if nodes within the same vicinity cooperatively work with one another. These data are then transmitted to the sinks, where they are consumed. If these sink nodes reside one hop away from the data-gathering nodes, then the task of the data source to sink transmission is simple. However, these sinks may sometimes be located multiple hops away from the source of the data, thus requiring a routing protocol.
Manuscript received March 1, 2009; revised May 18, 2009. First published July 14, 2009; current version published November 11, 2009. This work was supported in part by AMYN Investments Limited, Lagos, Nigeria. The review of this paper was coordinated by Prof. Y. Xiao. The authors are with the Centre for Telecommunications Research, King's College London, WC2R 2LS London, U.K. (e-mail: abdul.shuaib@kcl.ac.uk; hamid.aghvami@kcl.ac.uk). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TVT.2009.2027440

W

Aside from the traditional qualities that have been desired of a routing protocol, e.g., the ability to quickly converge, be loop free, be correct, be stable, and be reliable, a routing protocol that has been targeted toward WSNs must also ensure that it is scalable, that the overhead is minimized, and that it supports uniform depletion of energy across the network. In terms of scalability, the sheer density of nodes in a WSN mandates that the network is administered at the level of the cluster, and as a result, routing is better served at the granularity of that administrative unit. In this case, clusters along the path from the data source to the sink must be willing to relay trafc to the sink node. To achieve this case, mechanisms must exist within the cluster that allow for bandwidth reservation, admission control, and cross-cluster communication. Coincidentally, the IEEE 802.15.4 specication [2], which is implemented on the bulk of the hardware that was targeted at WSNs (e.g., Sun Spots [3], IP-Sensor [4], small autonomous network device (SAND) [5], NanoSensor [6], MicaZ, and IRIS [7]), provides the bare mechanisms for cluster formation and maintenance, admission control, and bandwidth reservation, all of which can be harnessed to achieve the objectives of a WSN routing protocol. If we will extrapolate from previous trends, then it is likely that the IEEE 802.15.4 specication will become the de facto physical- and MAC-layer standard for WSNs, just as the IEEE 802.3 and 802.11 standards are for local area networks. With this case, it will be unwise to design higher layer protocols that are transparent to the services that were offered by the specication. To that end, in this paper, we propose a routing framework and mechanism that interacts with the IEEE 802.15.4 protocol to assure connectivity in a WSN in an energy-efcient manner. The remainder of this paper is organized as follows. In Section II, we discuss the challenges that were placed by the WSN communication architecture on routing protocols. In Section III, we discuss some of the routing protocols that have been proposed for WSNs. Section IV gives an operational overview of the IEEE 802.15.4 specication. In Section V, we present the technical details of the proposed routing mechanism. Section VI contains an evaluation of the mechanisms in Section V. We use the OPNET network simulation tool for this purpose. The Appendices follow the conclusion, which is contained in Section VII. II. O BJECTIVES OF A W IRELESS S ENSOR N ETWORK R OUTING P ROTOCOL In this section, we describe the challenges and itemize the objectives of a WSN routing protocol/mechanism/framework. The discussion in this section is illustrated in Fig. 1.

0018-9545/$26.00 2009 IEEE
Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

5136

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

below, signify the quality of a link between any two nodes on a path, i.e., → → P ath 1 = C 14 B 14 A 0→ Sink S → P ath 2 = C 7→ D 8→ E 14 Sink S. (1) (2)

Fig. 1. Cluster-to-sink communication.

A cluster is typically made up of a cluster head, cluster members, and a cluster gateway. The cluster head is the administrative node for the cluster, and the gateway is the primary relay node for the cluster. Sometimes, the cluster head acts as the cluster gateway, whereas at other times, a member of the cluster is designated as the gateway. For illustration, we will assume that the data that were generated in Cluster C needs to be transmitted to a sink, i.e., Sink S in Fig. 1. In the gure, two paths exist from Cluster C to Sink S. The rst path, i.e., Path 1, goes through Clusters B and A, and the other path, i.e., Path 2, goes through Clusters D and E. The rst challenge of a WSN routing protocol is how it can achieve efcient route propagation i.e., how sinks are announced to the network clusters, how routes to the sinks are maintained, and how nodes that are not gateway nodes or cluster heads can be shielded from taking part in the route discovery and update processes. The second challenge deals with the twin problem of admission control and bandwidth reservation by the relay clusters on behalf of other clusters. For instance, if Path 1 in the gure is the preferred path from Cluster C to Sink S, Cluster B should grant Cluster C access by reserving bandwidth for and admitting trafc from Cluster C if it can manage it. If a cluster is unwilling or cannot admit trafc from another cluster en route to a sink, then this information must be reected in the route metric, even before the rst data packet is transmitted. The idea here is to prevent other clusters from interfering with the internal activities of a relay cluster without its permission. A WSN routing protocol must infer the quality of all links on a path from a source to a destination with minimal overhead, and this condition should be reected in the route metric that qualies that path. To illustrate this challenge, let us assume that the numbers beneath the arrows in (1) and (2), shown

Now, if Cluster C picks the next hop based on its immediate link quality, it would pick Cluster B as its next hop to the sink. Notice that Cluster C cannot infer that the link quality between Cluster A and the sink is 0. This case is not ideal, because a route is only as good as its lowest quality hop. The additive-cost approach, which uses the sum of the link metric along the path, and the averaging approach, which takes the mean cost on a path, are also not ideal. Observe that, although Path 1 has a link with quality 0, it would still sum or average higher than Path 2. Another challenge deals with dissipation of energy. It has been shown in [8] and [9] that the lifetime of a WSN is extended if the rate of energy dissipation across the network is uniform. For instance, if Path 2 is the preferred path and all routed data consistently go through this path, the clusters along Path 2 will die out quicker than those on Path 1. The routing framework should recognize and appropriately adapt this case. Note that uniform energy dissipation also somewhat translates to load balancing. Another solution to the aforementioned uniform energydissipation problem is to have mobile sinks in the network. The main idea here is to have the mobile sink visit multiple locations within the network to collect data, thus saving the energy that would have been used to relay data if the sink was remotely located and multiple hops away. In [10], the authors showed that the network lifetime can theoretically be extended by up to 500% when mobile sinks are used. However, this mobile-sink scenario will constitute a major challenge for any WSN routing protocol. For example, let us assume that Cluster D announces to the network that it is one hop away from the mobile sink. Cluster C hears this announcement and accordingly updates its routing table. When the mobile sink moves away from Cluster D toward Cluster E, Cluster D purges the entries for the mobile sink soon after. At the next announcement interval, Cluster C announces that it is two hops away from the mobile sink. Cluster D hears this announcement and updates its routing table, assuming that it is three hops away from the mobile sink through Cluster C. Aside from the obvious issue of loop in the path that was introduced here, this scenario easily leads to the count-to-innity problem that plagued the Routing Information Protocol (RIP) [11]. Requirements also exist for the support of routing that is geographically oriented and secure. In addition, of course, the traditional routing challenges (e.g., scalability, exibility, loop prevention, fault tolerance, minimal overhead, adaptability to link and topology changes, and how the shortest most reliable most energy-efcient path can be chosen) must also be dealt with. III. R ELATED W ORK The routing protocol in this paper is a proactive vector routing protocol that relies on a cluster network architecture

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

SHUAIB AND AGHVAMI: ROUTING SCHEME FOR IEEE-802.15.4-ENABLED WIRELESS SENSOR NETWORKS

5137

to achieve its objectives. Thus, the discussion in this section will be restricted to a representative set of WSN protocols that share similar features with our proposed routing protocol. The interested reader should consult [12] and [13] for an exhaustive survey of routing protocols that were targeted at WSNs. The rst group of protocols are cluster routing protocols. Examples of such protocols are the low-energy adaptive clustering hierarchy (LEACH) protocol [14], the threshold-sensitive energy-efcient (TEEN) protocol [15], and the hierarchical power-aware routing (HPAR) protocol [16]. LEACH is an adaptive clustering routing protocol that elects a cluster head from a group of homogenous node to act as the relay node for the cluster for a given time interval. After this interval, the cluster head is replaced by one of its peers. As soon as a node is replaced, its elective eligibility diminishes in subsequent election rounds. The idea is to spread the energy that was utilized for relaying data for the cluster among the nodes. Each cluster head gathers data from the cluster and directly transmits these data to the sink. LEACH was designed for scenarios in which data are always available to be sent at a xed rate. TEEN, which has a similar mechanism to LEACH, was designed for reactive networks, i.e., networks that immediately transmit data upon sensing a phenomenon. The implicit assumption in both LEACH and TEEN that all cluster heads (and by extension all nodes) can directly reach the sink prevents the network from expanding, and therefore, it cannot cover large regions. The HPAR protocol groups each node into a geographical zone or cluster. A single node in each zone is periodically selected to estimate the power levels of each node in its zone. This estimate is used to compute the amount of energy that was expended to transmit data out of each corner of the zone. This information is then broadcast to other zones and is used for future routing decisions. The idea here is that nodes from other zones can infer the path that can reliably relay information using the minimum amount of energy. MintRoute [17], MultiHopLQI [18], the Collection Tree Protocol (CTP) [19], Arbutus [20], and MobiRoute [21] are a set of routing protocols that, as a group, are referred to as collection protocols. The primary difference in each of these protocols lie in how the path cost is computed. For example, in MintRoute, the computed cost of a path is a function of the ratio of the number of expected packets and the number of packets that were received on the immediate link. CTP attempts to improve upon MintRoute by summing the link costs across all hops to determine the cost of a path. The MultiHopLQI protocol uses the same principle of additive cost for path cost computation but differs from CTP in the sense that the cost is a function of the received signal strength compared to using the ratio of the expected number of received packets and the number of packets that were received. None of the aforementioned collection protocols explicitly implements a form of load balancing. Achieving load balancing is the primary motivation of the Arbutus collection protocol. It achieves its objective by using the trafc load on the immediate links of a relay node as an input to the cost computation algorithm. Observe that the collection routing protocols use either an additive or an immediate cost approach for path cost computation. As discussed in the last section, this approach is

not ideal. In addition, these protocols do not deal with the issue of mobility; however, an extension of the MintRoute protocol called MobiRoute has been proposed for this very purpose. One common feature of the collection protocols is the use of network-layer beacons to propagate route information in the network. As will be discussed in this paper, we adopt a similar approach to route propagation, except that this time, we rely on the beacon mechanism that has already been provided for by the IEEE 802.15.4 specication. The aforementioned protocols in this section can be classed as proactive distance vector routing protocols. The ad hoc ondemand distance vector protocol, which is a reactive distance vector routing protocol, was originally designed for ad hoc networks, but it has been discussed within the context of WSNs in [12] and [22]. The advantages and disadvantages of reactive routing protocols are well known. For example, nodes do not need to maintain path entries to every destination in the network, because route paths are requested on demand, thus saving memory space. Associated with this benet are some disadvantages, including the fact that route request messages are sent into the network using a broadcast mechanism, which can easily lead to a broadcast storm. The possibility of selecting a suboptimal path due to limited topological information that is available to the node [23] and the delay that was incurred when trying to acquire a route [24] are factors that should be considered when using a reactive protocol. The unique communication architecture of the WSNs makes some of the concerns that were addressed by reactive routing protocols redundant. For example, the only signicant multihop communication in a WSN is the communication between the sink and the data source; thus, the gateway/relay node need not hold route information to all nodes in a network but to an optimal number of sinks. To that end, our routing mechanism, as will be detailed in this paper, adopts a exible approach that combines some of the advantages of the proactive and reactive protocols. IV. O VERVIEW OF THE 802.15.4 N ETWORK A RCHITECTURE In this section, we discuss the network architecture of the IEEE 802.15.4 specication and its operational overview in terms of how the clusters are formed and how the shared channel is accessed in the beacon-enabled mode. We also discuss how intercluster communications can be achieved using a combination of the mechanisms that were provided by the specication. Within the specication, a cluster is referred to as a wireless personal area network (WPAN), and the cluster heads are referred to as the WPAN coordinators. In the remainder of this paper, we will use the terms WPAN and WPAN coordinator when referring to a cluster and a cluster head, respectively. A. WPAN Formation Process and the Active/Inactive Periods Based on a communications perspective, an IEEE 802.15.4 node is functional only if it is associated with a WPAN, either as the coordinator or as a member. Typically, at start up, a

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

5138

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

Fig. 2. SFD interleaving.

node scans the channel looking to receive beacons from WPAN coordinators within its surrounding areas. At the end of a scan, the node looks for the best beacon in the beacon scan list, which is usually the beacon from a WPAN that is within its personal operating space (≤ 10-m radius) and currently accepts associations. The node proceeds to associate with a WPAN if it is deemed to be suitable. As a part of this process, the WPAN coordinator assigns a 16-b short address to the associating node. All future communications between the WPAN coordinator and a member are done using this short address. On the other hand, if none of the beacons is suitable, the node does one of the following two approaches, depending on the type of the IEEE 802.15.4 node: 1) a full function device (FFD) or 2) a reduced function device (RFD). An RFD, which is relatively resource constrained, might shut down its radio and later try again to locate a suitable WPAN. An FFD, upon failing to nd a suitable WPAN, can proceed to form its own. The newly created WPAN identies itself using a 16-b WPAN ID and begins beacon transmission. Two consecutive transmitted beacons from a single WPAN will be separated by a beacon interval (BI). If the network operates on the 2.4-GHz frequency band with a bit rate of 250 Kb/s, then the duration of the BI (in seconds) is shown as follows: BI = 0.01536 × 2BO (3)

of WPANs within the communication distance of one another increases, the length of the BI also increases if each SFD will be separated in time. Although outside the scope of this paper, it is worth noting that mechanisms exist to mitigate the interference that was caused by other networks, e.g., 802.11, which operates on the industrial–scientic–medical band as the IEEE 802.15.4 network [27]. B. Channel Access Mechanisms The SFD can be divided into the contention access period (CAP) and an optional contention-free period (CFP). In the CAP of any SFD, WPAN members and the coordinator have to contend for the medium using a carrier-sense multiple access with collision avoidance (CSMA–CA) mechanism. To transmit during the CFP, WPAN members must have been granted some guaranteed time slots (GTSs) in the SFD by the WPAN coordinator. These grants are either unsolicited or a result of an explicit GTS request that was transmitted by a WPAN member. In this paper, the CFP is the most pertinent and, as such, will further be discussed in detail using Fig. 3. According to the standard, the activities of the IEEE 802.15.4 MAC layer are coordinated by an abstract entity known as the next higher layer. This entity receives and processes MAClayer commands on behalf of a node. For instance, if a node requires for GTSs, the higher layer computes the number of slots required and sends this information to the MAC layer. The MAC layer appropriately frames this GTS request command, setting all the required elds, and transmits it on the wireless medium. According to the specication, all command frames will be sent using the CSMA–CA mechanism and must be acknowledged upon receipt. When a coordinator receives a GTS request command, it acknowledges the frame and decapsulates it. The information with regard to the GTS request is then sent to the coordinator's next higher layer. The coordinator's next higher layer reserves slots in the subsequent SFD based on the request and as it deems t. The information with regard to GTS grants are included in the next outgoing beacon. Upon the receipt of a GTS grant from the coordinator, the next higher layer of a node instructs the CFP process to begin transmissions as soon as its reserved slot time begins and stops the process when it ends. If the number of allocated GTSs is not sufcient, the GTS request process is repeated as aforementioned and as depicted in the message sequence in Fig. 3. Note that only a maximum of seven nodes can be granted GTSs in any SFD. C. Inter-WPAN Communication The standard does not expressly dene how inter-WPAN communication is achieved, but it gives room for much

where BO, 0 ≤ BO ≤ 15 is referred to as the beacon order. As soon as beacon transmission commences, other unassociated nodes within the vicinity of the WPAN coordinator can elect to join the newly formed WPAN. WPAN members are only active for a certain period within the BI. This active period is referred to as the superframe duration (SFD). The length of the SFD is given as SFD = 0.01536 × 2SO (4)

where SO, SO ≤ BO is referred to as the superframe order. The SFD of a WPAN begins at the instant that the beacon is transmitted by the WPAN coordinator. The difference between the SFD and the BI is the time interval in which WPAN nodes need to be inactive and can probably go to sleep. If multiple WPANs operate on the same frequency band or frequency channel and are within the communication range of one another, it is mandatory that their SFDs are separated in time. The idea is to prevent one WPAN from interfering with the active period of another. For instance, if we assume that WPANs A–D are within the communication distance of one another, their SFDs could be arranged as shown in Fig. 2. Note that to maintain the integrity of the SFD boundaries, WPANs A–D must be synchronized with one another. Several solutions exist to ensure the aforementioned condition, including [25] and [26]. It should be apparent in Fig. 2 that as the number

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

SHUAIB AND AGHVAMI: ROUTING SCHEME FOR IEEE-802.15.4-ENABLED WIRELESS SENSOR NETWORKS

5139

Fig. 3.

Default GTS message sequence.

exibility. In [28] and [29], the authors attempt to exploit this exibility to achieve inter-WPAN communication. Two mechanisms with similar underlying principles were proposed. In the rst solution [28], which was referred to as master–slave bridging, the coordinators act as the relay nodes or bridges on behalf of their respective WPANs. Slave–slave bridging [29], on the other hand, shifts the responsibility of relaying data for a WPAN to a node other than the coordinator. To illustrate the operation of the bridge, we will assume that two WPANs A and B exist such that trafc from WPAN A needs to go through WPAN B en route to the sink. A bridge will typically accept data from WPAN A during its SFD and deliver these data to WPAN B during WPAN B's SFD. Another bridge will deliver the data from WPAN B to the sink using the same mechanism. The bridge can accept and deliver data using the CAP or CFP of the WPANs. If the bridge operates during the CAP, then it will have to compete for the channel like any other member of the WPAN. As noted in [28], depending on the internal state of a receiving WPAN, the bridge's operation during the CAP can negatively impact the activities of the WPAN members, or the other WPAN members could do the same to the bridge. Although the CFP portion is used for bridge activities, this negative impact is still possible, because the GTS request command is only transmitted during the CAP. In addition, the fact that the bridge can compete for the channel across multiple WPANs suggests that the bridge is a member of each of those WPANs. In this paper, we adopt a different approach to inter-WPAN communication by slightly extending the GTS message sequence in Fig. 3. Here, a relay node, whether it is a coordinator or another designated node, need not be a member of multiple WPANs. The communication across WPANs is strictly done using the CFP, because it aids admission control and bandwidth reservation, which, in turn, provides a mechanism for insulating a WPAN from outside activities, if need be. If a WPAN is saturated and cannot handle trafc that needs to be relayed, this information is reected in the route metric that qualies the path through that WPAN. The details of how we achieve the aforementioned approach are discussed in Section V-B, because it is an integral component of our routing mechanism.

V. P ROTOCOL D ESCRIPTION AND I MPLEMENTATION In this section, we introduce and describe the operational mechanism of our routing scheme within the context of what is expected of a WSN routing protocol. In addition, contained within this discussion is how our mechanism interacts with the IEEE 802.15.4 MAC layer to achieve the WSN routing objectives. However, before going into the discussion, a few denitions are given in order.

A. Denitions and Notations The denitions that follow will act as an aid to understanding the various components of our routing mechanism, which is described in the next section. Denition 1—Neighbor WPAN: If WPANs A and B are within the communication range of one another, then WPAN A is a neighbor of WPAN B, and vice versa. We will denote the set of neighbor WPANs of WPAN X as Nx , and its cardinality is |Nx |. Denition 2—Out-Neighbor: If WPAN A has a route to a sink through WPAN B, then WPAN B is an Out-neighbor of WPAN A for that sink. We will denote the set of Outneighbors of WPAN X for sink S as Nxsout , and Nxsout ∪ Nx = Nx . Denition 3—In-Neighbor: Correspondingly, if WPAN A has a route to a sink through WPAN B, then WPAN A is an In-neighbor of WPAN B for that sink. We will denote the set of In-neighbors of WPAN X for sink S as Nxsin , and Nxsin ∪ Nx = Nx . Denition 4: It is possible that an In-Neighbor n ∈ Nxsin for a WPAN X to sink S can also be an Out-neighbor for WPAN X to sink T, i.e., |Nxsin ∩ Nxtout | ≥ 0. Denition 5—Route Descriptor: A route descriptor contains information with regard to a single route to a sink, and this information will include the hop count (HC), the Out-neighbor, and the quality of the path to that sink. Denition 6—Route Dscrpt List: The incoming route descriptor list (Route Dscrpt List) RDinlist contains all incoming route descriptors, whereas the RDoutlist contains all the outgoing descriptors of a relay node.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

5140

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

Fig. 4. Route packet structure. (a) Beacon payload format. (b) Route descriptor format.

Fig. 5. WPANs and one sink. (a) Network structure. (b) Outgoing route descriptor.

Denition 7—Forwarding Table: The forwarding table Tf d contains the routes to the sinks and the associated computed route cost. Denition 8: The inputs to the process that produces the forwarding table are the entries of the incoming Route Dscrpt List. Denition 9—LQI: The IEEE 802.15.4 specication mandates that each incoming frame must be stamped with a link quality indicator (LQI) value. This value indicates the quality of the link at the time of frame reception. According to the standard, the LQI value must be an integer that is uniformly distributed between 0 and 255, with 255 indicating the highest signal quality. Details about how this value is computed are contained in Appendix A. Denition 10—Update Interval: The update interval, which we will denote by ∧, is the length of time that must elapse before a relay node completes the following three tasks: 1) purges its forwarding table; 2) updates the forwarding table if new information is available; and 3) announces its best routes to sinks to its surrounding WPAN neighbors. The update interval is in multiples of the BI [see (3)], i.e., ∧ = ∧coef f BI, where ∧coef f is the update coefcient. Finally, we will denote the current update interval by ∧curr . B. Mechanisms for Achieving WSN Routing Objectives The purpose of this section is to give a technical description of our proposed routing mechanism. We do this by juxtaposing each of its components with the objectives of the WSN routing protocols in Section II. 1) Route propagation: Route propagation is achieved through the insertion of route information into the payload section of an outgoing IEEE 802.15.4 beacon once in every update interval ∧. WPANs within the communication distance of one another know when a neighbor WPAN transmits its beacon, because they all have to be synchronized to prevent SFD boundary encroachment (see Section IV-A). A WPAN coordinator simply enables its receiver when any neighbor WPAN n ∈ Nx transmits a beacon that contains route information. The receiver is immediately disabled once the beacon is received. The coordinator then proceeds to extract the routing information from the beacon for processing. The frequency of

the neighborhood beacon reception also depends on the update interval ∧. The structure of the beacon payload is shown in Fig. 4(a). The protocol ID (Proto ID) eld tells the MAC layer which higher layer owns the content of the beacon payload. The route descriptor count (Route Dscrpt Count) indicates the number of route descriptors in the payload. The maximum number of route descriptors in any beacon payload is eight. The Route Dscrpt List eld contains the route descriptors to the known sinks. The formal structure of a route descriptor is shown in Fig. 4(b). The sink WPAN ID (SID) contains the WPAN ID of the sink. The Out-Neighbor ID (ONID) eld contains the WPAN ID of the Out-Neighbor. The HC contains the node's HC from the sink. Finally, the Lowest Path LQI (LPL) eld holds the value of the lowest link LQI on the path from the sink to the node. To describe the route-propagation mechanism, consider Fig. 5(a), in which nodes S and A–D are the coordinators for their respective WPANs, and the numbers by the bidirectional arcs that connect the nodes are the LQI values of the links. At time 1 [see Fig. 5(b)], node S, which is the sink for the network, inserts a payload into the beacon payload section. The beacon payload Proto ID is set to1 to indicate to the MAC layer that the payload is for the network/routing layer, and the value of the Route Dscrpt Count eld is set to1 to indicate that only one route descriptor is contained in the Route Dscrpt List. The contents of this single route descriptor are shown in the rst row in Fig. 5(b). Notice that the sink puts its address in the SID and ONID elds of the descriptor and sets HC = 0. It also sets LP L = 255. The sink then transmits this beacon during its active period or SFD. Node A enables its receiver and gets the beacon from the wireless stream. It extracts the payload and adds the route descriptor to its incoming Route Dscrpt List. At time 2, node A lls an outgoing route descriptor with the values shown in row 2 in Fig. 5(b). This time, the LPL is set to 20, because the link between Sink S and Node A had an LQI value of 20, as indicated in Fig. 5(a). HC = 1. This outgoing descriptor is inserted into the beacon payload, and the beacon is transmitted. Nodes B and D will receive the beacon that was transmitted by node A. Both nodes will process the incoming descriptor similar to process previously undertaken by node A. For their outgoing route descriptors, nodes B and D both update

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

SHUAIB AND AGHVAMI: ROUTING SCHEME FOR IEEE-802.15.4-ENABLED WIRELESS SENSOR NETWORKS

5141

the HC to 2, and the ONID is set to the address of node A. Node B updates the LPL in its outgoing descriptor to 10, whereas node D leaves the LPL as 20, although the LQI of the link that connects nodes A and D has a value of 50. Node C will receive two descriptors to Sink S: one descriptor from node B and another from node D. We designed the routing protocol such that, if a node has multiple routes to a sink, only the best route is included in the outgoing Route Dscrpt List. If other things are equal, node C will include only the path through node D in its outgoing Route Dscrpt List, primarily because the LPL on the path to the sink through D is higher than that of the path through B. In other words, the routing mechanism takes cognizance of the quality of all the links on a path to a sink, because the quality of a path is only as good as the hop on that path with the lowest link quality. At time 4, node C creates an outgoing descriptor, with the values shown in the fth row in Fig. 5(b). This descriptor is inserted into the next outgoing beacon from C. 2) Loop Avoidance: We use four mechanisms to prevent or minimize the occurrence of loops in the network. These mechanisms are itemized and discussed as follows. 1) Lists and table purge. In every update interval ∧, the forwarding table and the outgoing descriptor list are purged. If new information is available in the form of new descriptors in the incoming descriptor list, the forwarding table is populated with this information; otherwise, it is left empty. The idea here is to prevent the utilization and propagation of stale information, which minimizes loops within the network. 2) Descriptor Reject 1. In Fig. 5, at time 4, node C sends out its descriptor, as shown in the last row in Fig. 5(b). Nodes B and D will receive the route descriptor in the transmitted beacon of node C, because they are both within the communication range of node C. When node D receives node C's route descriptor, it rejects it, because node D can infer from the ONID of node C's route descriptor that it is the Out-Neighbor or the next hop for node C to Sink S. This approach effectively deals with the count-to-innity problem. During this process, node D adds node C to its In-Neighbor List NDin for Sink S. Node B will also reject node C's route descriptor, because it can infer from the HC eld of the descriptor that it is closer to the sink than node C and that the descriptor's OID = SID. A descriptor is also rejected by a node X if OID ∈ NX . The loop prevention process is captured in the algorithm in Fig. 6. 3) Descriptor Reject 2. Consider another scenario where Sink S in Fig. 5(a) is mobile and moves away from node A toward node B. At the next update interval ∧, node A will no longer have a direct path to Sink S, but node B will. Now, node A receives a route descriptor from node B, announcing that it is one hop away from the sink. Node A rejects this descriptor and removes Node B from its InNeighbor list for that sink. At the next update interval ∧, node A will accept the route descriptor from node B, because node B is no longer an In-Neighbor of node A for Sink S. The idea here is for A to err on the side of caution just in case the validity of the information

Fig. 6. Loop-prevention algorithm. X is the ID of the node that runs the algorithm, and YRDi is the ID of the source of RDi . HCxs is the current hop count of node X to Sink S.

that was propagated by node B is temporal. This loopprevention mechanism was added to the routing protocol, because we found out that, in practice, the speed and the dwelling time of the mobile sinks can sometimes negatively impact route convergence. 4) Descriptor TX limit. For the last loop avoidance mechanism, sinks are only allowed to propagate route descriptors about themselves, i.e., the Route Dscrpt Count is always 1 for a sink. For example, consider a mobile sink scenario with two sinks: one mobile sink and one static sink. Let us further assume that each sink is within the communication distance of the other one such that Sink 1 knows about Sink 2, and vice versa. Sink 1 propagates route information about Sink 2, and sink 2 does the same for Sink 1. Now, if Sink 1 becomes mobile and moves away from Sink 2, the possibility exists that Sink 1 might transmit information about Sink 2, although it is no longer within the communication range of Sink 2, thus propagating incorrect information. The descriptor TX limit mechanism prevents this instance. 3) Admission Control and Bandwidth Reservation: Bandwidth reservation and admission control across multiple WPANs is accomplished using the modied GTS message sequence in Fig. 7. Based on Fig. 5, node C had chosen node D as its OutNeighbor WPAN to Sink S. Let us now assume that node C has trafc that needs to get to Sink S through node D. The process of admission control and bandwidth reservation by node D on behalf of node C is explained as follows. Within every update interval ∧, along with the inclusion of route descriptors, node D

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

5142

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

Fig. 7. Modied GTS message sequence.

includes GTS grants in its beacon for each of its In-Neighbors (a maximum of 7 In-Neighbors). This method is coordinated by node D's routing process and its higher layer, as shown in Fig. 7. If node C requires trafc to be sent through node D en route to Sink S, it uses this rst unsolicited GTS grant to send a GTS request to node D. This GTS request would contain information about how much trafc node C seeks to route through node D. Upon receiving this GTS request, node D estimates how much resources would be required to admit node C's trafc. If node D can handle node C's trafc, node D, in its next active period/SFD, grants node C the required number of GTSs. The second grant is depicted as Grant 2 in Fig. 7. Node C begins transmission to node D at the assigned slot times. On the other hand, if node D cannot admit trafc on behalf of node C, it simply does not grant node C the second GTS grant. This method assures that the trafc from node C does not interfere with the activities of the WPAN in which node D is the coordinator without the permission of node D. In general, if the internal state of a WPAN is such that it cannot route trafc on behalf of any of its WPAN neighbors, it simply does not include GTS grants (either Grant 1 or 2 in Fig. 7) in its outgoing beacons. If node D cannot admit trafc for node C, then node B might do so. In this case, node B should be the preferred Out-Neighbor for node C to Sink S, and this approach should be reected in the route metric. How this approach is incorporated into the route metric computation is discussed in the next section. If we will contrast the modied message sequence in Fig. 7 with the default in Fig. 3, we would nd that the difference lies in how the GTS requests are sent. The default message sequence uses the CSMA–CA/CAP channel-access mechanism to transmit a request, whereas the modied version uses the CFP. If the rst GTS is not granted (i.e., Grant 1), then there is no way that a GTS request can be placed across WPANs; hence, bandwidth cannot be reserved, and trafc will not be admitted from outside the WPAN. Based on the discussions so far, we see that a coordinator can assign GTSs to nodes within and outside its WPAN. The GTS transaction between a WPAN coordinator and its WPAN

member is done using the member's assigned 16-b short address. To allow for consistency, GTS transactions outside a WPAN should be carried out using the 16-b WPAN ID. The only provision is that the coordinator of WPAN X should never assign a 16-b address that is already in use by any WPAN n ∈ Nx to any of its WPAN members. 4) Route Metric Computation: The relative preference of one path over another is based on the computed metric on those paths. Our route metric computation formula is a function of the link quality of the path and the hop distance from a sink, as shown in (5). The higher the metric Cij , the better the route. We have Cij = (HDij )γ + (LP Lij )ρ 2 (5)

where Cij ∈ [0.0, 255.0], HDij ∈ [0, 255] is the hop degree from Sink i to node j, and LP Lij ∈ [0, 255] is the value of the lowest LQI along the path from Sink i to node j. The hop degree HDij is a function of the HC HCij from a node to a sink such that HDij = 255 HCij , i.e., the closer a node is to a sink, the higher the hop degree becomes. Based the aforementioned condition, we can infer that the HC between a node and a sink can never exceed 255. In (5), ρ ∈ [0.0, 1.0] and γ ∈ [0.0, 1.0] are the LQI and hop degree exponents, respectively. These exponents are weights that dene the relative importance of the hop degree/HC and the lowest LQI on a path when computing Cij . To show how these exponents are applied, assume that two paths (Paths 1 and 2) exist from node A to Sink S. We have → → → → P ath 1 = A 50 B 60 C 50 D 70 S → → → P ath 2 = A 30 F 20 G 40 S. In addition, assume that the numbers underneath the arrows represent the quality of the hop/link. Using (5) and setting ρ = 1.0 and γ = 1.0, the route metric for Paths 1 and 2 are 150.2 and 136.0, respectively. Based on

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

SHUAIB AND AGHVAMI: ROUTING SCHEME FOR IEEE-802.15.4-ENABLED WIRELESS SENSOR NETWORKS

5143

the computed route metric, Path 1 will be the preferred route for node A to Sink S. However, Path 1 will also consume more energy for transmission than Path 2, because it consists of more hops to the sink than the other path. This case might not be ideal for all scenarios. To remedy this case, we make ρ a function of the HC HCij as follows: ρ= 1 . HCij (6)

TABLE I SIMULATION PARAMETER VALUES

If we recomputed the route metric for both paths while taking cognizance of (6) and γ = 1, we get 126.83 for Path 1 and 127.4 for Path 2, with Path 2 becoming the preferred path. Given that Path 2 is the current preferred path, the OutNeighbor for node A is node F . The possibility that node F cannot admit trafc from node A exists. In this case, node A would turn to the other path for which node B is the OutNeighbor. If node B has the resources to admit trafc from node A, then the path through node B should be the preferred path, because the path through node F is as good as useless. Basically, the inability of an Out-Neighbor to admit trafc en route to a sink should be reected in the metric that qualies that path through that Out-Neighbor. This condition is precisely what the HD exponent γ in (5) ensures. For instance, if, in the current update interval ∧curr , an Out-Neighbor to a sink grants the rst GTS (i.e., Grant 1 in Fig. 7) to an In-Neighbor, then γ is set to 1, but if it does not, then γ is set to 0, i.e., γ= 1.0, 0.0, if rst GTS allocated in ∧curr otherwise. (7)

table with the exact same values. In this paper, the WPAN coordinator also doubles as the WPAN's gateway or relay node. Finally, a WPAN, through its coordinator, can choose not to participate in the routing process by having the coordinator disable its receiver when its neighbor WPANs transmit their beacons. VI. P ERFORMANCE E VALUATION Cross-layer interactions do not lend themselves well to mathematical or analytical modeling, primarily because the multivariate parameters of the interacting layers must be considered if the performance characteristics should accurately be captured. To that end, the routing mechanism in the previous section was built atop an IEEE 802.15.4 model, which was implemented using the OPNET network simulation tool [36]. We chose this simulation tool, because it enables us to model the wireless channel in detail and with a relatively high delity. We carried out two sets of evaluations to assess how the routing mechanism in conjunction with the IEEE 802.15.4 MAC and physical layers deliver on the objectives expected of a WSN routing protocol. In the rst set of evaluations, we studied how the routing scheme achieves load balancing, uniform energy dissipation, bandwidth reservation, and admission control. We also look at the contents of the incoming descriptor lists and the forwarding tables of each of the nodes. In the second set of evaluations, we look at how the mechanism performs on scale, particularly in terms of loop control, sink mobility, energy consumption, and its adaptability to topological changes. We do this approach under three different scenarios. The plots in the next two sections are average values that were collected with a 95% condence interval over multiple simulation runs, each with a different random number seed that is uniformly distributed between 0 and 250. The entries in Table I are the simulation parameters that are common to both sets of evaluations. The table shows the nominal transmit power and receiver sensitivity, as set by the IEEE 802.15.4 specication for compliant nodes. Our energyconsumption model follows that of Texas Instrument (TI)'s CC2430 chip [37], which is currently deployed on the latest hardware platforms targeted at WSNs. This TI chip expends 24.7 mA for transmission and 27 mA for reception at 3 V. The remaining table entries are self explanatory.

Now, based on the assumption that node F does not grant GTS to node A but node B does within the current interval ∧curr , γ = 1.0 for Path 1, and γ = 0.0 for Path 2. With ρ = 1/HCij , node A's route metric CAS for Paths1 and 2 are 126.83 and 3.71, respectively, with Path1 becoming the preferred path. 5) Support for Other WSN Routing Objectives: Certain WSN applications, e.g., agricultural and environmental monitoring applications, require that the underlying network can retrieve and transmit data in a geographically oriented form. For this case to be possible, the WSN nodes must have some form of onboard location positioning component, e.g., the GPS-less solutions in [30]–[32]. If this component is present, our routing mechanism can be turned into a geographically oriented protocol simply by choosing the 16-b WPAN ID based on the geographical location of the coordinator. Note that efcient routing for WSN is done at the granularity of the cluster/WPAN. With regard to communication security within the context of the IEEE 802.15.4 specication, see [33]–[35]. Given the relative limitations of WSN nodes, particularly in terms of onboard memory, it is ideal that the size of the forwarding table is bounded and kept to a minimum. To prevent forwarding table explosion, we limit the number of entries into the forwarding table to 16, i.e., two of the best paths each for a maximum of eight distinct sinks. Note that, if a WPAN coordinator designates one of its WPAN member as the relay node for that WPAN, the designated relay node simply extracts the route information from its coordinator's transmitted beacon and updates its forwarding

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

5144

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

Fig. 8. Three WPANs and one sink.

A. Evaluation 1 Our rst network scenario consisted of four WPANs, as shown in Fig. 8. One of these WPANs was the network sink. In the scenario, we placed WPAN 3 far enough so that it is outside the communication range of the sink but within range of WPANs 1 and 2. The idea was to have WPANs 1 and 2 act as the OutNeighbors for WPAN 3 to Sink 1. Note that, with the nominal transmit power and receiver sensitivity and under a free-space path loss model, the maximum communication distance of a node is 170 m or less if we factor in the other radio attributes. In Figs. 9 and 10, we show the content of the incoming Route Dscrpt List and the resulting forwarding table for WPANs 1 and 2 for the 120th update interval. We can see in Fig. 9 that WPAN 1 has a direct route to the sink and an alternate indirect route through WPAN 2. A similar deduction about the number of available routes to WPAN 2 can be made in Fig. 10. In Fig. 9(b), the direct route is the preferred route to SINK 1 for WPAN 1 as expected, but notice that the preferred route for WPAN 2 to SINK 1 is the indirect route through WPAN 1, as shown in Fig. 10(b). The reason for this case is that the twohop path from the sink through WPAN 1 to WPAN 2 had a better link quality than the one-hop path from the sink to WPAN 2, as indicated in the low-LQI entries in Figs. 9(a) and 10(a). Such a scenario allows for data to transverse the best quality path, if longer, at the expense of consuming twice the amount of energy required for transmission and reception. If the forward error-correction option, as provided for in [38], is set for the radios of the deployed IEEE 802.15.4 WSN nodes, then these nodes can receive and improve on low-quality frames. In this case, there is no reason that the shortest path should not be the preferred option. Our nodes can receive low-quality frames up to a certain degree, given the error correction threshold in Table I, i.e., the nodes can recover frames that with 20% or fewer errors. To align the route metric with this result, we set the LQI exponent ρ = 1/HCij as discussed in Section V-B-4. The new forwarding table of WPAN 2 is shown in Fig. 11(a), with the direct path set as the best path to the sink. Fig. 12 shows the incoming Route Dscrpt List and the forwarding table for WPAN 3. The path through WPANs 1 and 2 tie in terms of the HC to the sink; thus, WPAN 3 chose WPAN 1 as its Out-Neighbor to SINK 1 based on the link quality of the path through WPAN 1. To demonstrate the load balancing, admission control, and bandwidth reservation capabilities of the protocol, WPAN 3 was set to transmit 300 frames of 100 B each to the sink. Three

scenarios were also created, and the energy consumed by the two Out-Neighbors of WPAN 3 in each of the three scenarios is shown in Fig. 13. In the rst scenario in Fig. 13, no data were transmitted by WPAN 3, and the energy that was consumed by WPANs 1 and 2 was mainly due to beacon transmission and reception. In the second scenario, WPAN 3 routed the 300 frames through WPAN 1, because WPAN 1 was the preferred Out-Neighbor, as shown in Fig. 12(b). Thus, the energy that was consumed by WPAN 1 is higher than what was consumed by WPAN 2. In the third scenario, WPAN 1 was set to reserve bandwidth and admit trafc for half of what WPAN 3 requested, after which, WPAN 1 denied all GTS grants to all its In-Neighbors. As a result of this non-GTS grant, the route through WPAN 1 was downgraded based on the hop degree exponent (γ), as shown in the forwarding table in Fig. 11(b), causing WPAN 2 to become the preferred Out-Neighbor for WPAN 3. This case caused WPAN 3 to request bandwidth reservation and trafc admittance from WPAN 2. WPAN 2 granted this request, and the remainder of the data frames was transmitted through WPAN 2 to SINK 1. Both Out-Neighbors were utilized by WPAN 3 for routing trafc; thus, the energy consumed by WPANs 1 and 2 evened out, as shown in Scenario 3 in Fig. 13. A few points to remember from the aforementioned example are given as follows: 1) A WPAN can choose to admit all, some, or none of the trafc from an In-Neighbor; 2) if trafc is not admitted by an Out-Neighbor, it is reected in the route metric of the route that goes through that Out-Neighbor; and 3) Points 1 and 2 allow for load balancing and graceful degradation of the network, which has been shown to increase the life time of the network [8], [9]. B. Evaluation 2 For the set of evaluations within this section, we created three different scenarios. Each scenario consisted of 20 randomly deployed WPAN coordinators, ve of which were set to be sinks. These nodes were deployed in a 250 m × 250 m area. In the rst scenario, which we refer to as the Static Topology scenario, all coordinators were static. In the other two scenarios, two of the ve sinks were set to be mobile. In the Dynamic Topology (IN Perimeter) scenario, the trajectories of the mobile sink nodes were conned to the perimeter of the deployment area. In the third scenario, which we refer to as the Dynamic Topology (OUT Perimeter) scenario, the trajectories of the mobile sinks were set such that they went out of the deployment area and, ultimately, out of the communication range of the other 18 coordinators. The whole idea was to study the impact of sink mobility on the routing protocol. For both scenarios with mobility enabled, the mobile sinks moved at a ground speed of 10 m/s. There were four stops along the trajectories of the mobile sinks; the dwell time for the mobile sinks at each stop was 3 min. For the dynamic scenarios, sink mobility started 3 min into the simulation. The connectivity graphs, which are based on the contents of the forwarding tables of the nodes, are shown in Figs. 14 and 15(a). The black nodes in the gures represent the nonsink nodes, whereas the white nodes are the sinks. Although the

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

SHUAIB AND AGHVAMI: ROUTING SCHEME FOR IEEE-802.15.4-ENABLED WIRELESS SENSOR NETWORKS

5145

Fig. 9. Route Dscrpt List and forwarding table for WPAN 1 (γ = 1, and ρ = 1). (a) Incoming Route Dscrpt List for Update 120. (b) Forwarding table for Update 120.

Fig. 10. Route Dscrpt List and forwarding table for WPAN 2 (γ = 1, and ρ = 1). (a) Incoming Route Dscrpt List for Update 120. (b) Forwarding table for Update 120.

Fig. 11. New forwarding tables for Update 120 for WPANs 2 and 3. (a) Forwarding table for WPAN 2 (γ = 1, and ρ = 1). (b) Forwarding table for WPAN 3 (ρ = 1).

Fig. 12. Route Dscrpt List and forwarding table for WPAN 3 (γ = 1, and ρ = 1). (a) Incoming Route Dscrpt List for Update 120. (b) Forwarding table for Update 120.

Fig. 13. Energy consumption for WPANs 1 and 2.

nodes had multiple routes to each sink (6.4 on the average), the plots in the gures only show the routes with the lowest HC to the sinks. For the static scenario, we can see that the network is fully connected. For both Dynamic Topology scenarios, Sinks 1 and 4 were the mobile sinks. The initial and nal positions of the mobile sinks for the Dynamic Topology scenarios can visually be inferred by comparing Figs. 14(b) and 15(a) with Fig. 14(a). In the IN Perimeter scenario, the links with the dotted lines were the new links formed as a result of sink mobility. For the OUT Perimeter scenario, notice that none of the nodes had a connection to Sinks 1 and 4, because they are clearly out of the

communication range (i.e., >170 m). As a result, the average number of sinks known per WPAN coordinator was lower for the OUT Perimeter scenario than for the other two scenarios, as shown in Fig. 15(b), which is a plot of the average number of known sinks versus the update coefcient. Note that the update coefcient ∧coef f denes the duration of the update interval (see Denition 10). For example, with ∧coef f = 5 and with the BO value shown in Table I, the length of the update interval ∧ is 2.5 s (i.e., the route descriptors are propagated and received, and forwarding tables are purged and updated every 2.5 s). Correspondingly, if ∧coef f = 50, then ∧ = 25 s. Note that the routing protocol is disabled if ∧coef f = 0. With the routing protocol disabled, WPAN coordinators cannot reach the sinks, because the network would not be connected. The average number of known sinks, as shown in Fig. 15(b), is consistent with what was expected for each of the three scenarios. Note that consistency is the rst of the two indicators of route convergence (the second indicator will be discussed shortly). In the OUT perimeter scenario, as the sinks steadily moved away from the deployment area, WPAN coordinators at one edge of the deployment area had to increasingly depend on the coordinators at the opposite edge to reach the mobile sinks. Thus, the mean HC from WPAN coordinators to the sinks was higher for the OUT Perimeter scenario compared with the other two scenarios, as shown in Fig. 16(a). This increased HC for the OUT perimeter scenario is more clearly shown in the HC frequency distribution of the network in Fig. 16(b), where its tail pulls outward, with at least one instance where a node had a

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

5146

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

Fig. 14. Connectivity graphs. (a) Static scenario. (b) IN Perimeter scenario.

Fig. 15. Connectivity graph and average number of known sinks. (a) OUT Perimeter scenario. (b) Average number of known sinks.

Fig. 16. Hop counts to sinks. (a) Mean hop count to sinks. (b) Hop count frequency frequency distribution (∧coef f = 5).

path with eight hops to a sink. In addition, based on the gure, the coordinators in the Static scenario were a maximum of three hops away from any sink at any time. This maximum HC for the IN perimeter scenario was 4. A system is said to be stable if it has properties that assure that errors introduced into the system at one time do not unboundedly grow at later times [39]. If the system in question is a routing protocol, then a stable routing protocol is one that

assures that a loop introduced into the network at any time does not unboundedly grow during the operation of the network. The stability of a routing protocol is the second indicator of route convergence. Fig. 16(a) and (b) implicitly convey the stability of our routing protocol. For instance, the fact that the mode for all three plots is 2 in Fig. 16(b) indicates that there are no perpetual loops in the network, irrespective of the topological changes due to sink mobility. For an unstable routing protocol,

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

SHUAIB AND AGHVAMI: ROUTING SCHEME FOR IEEE-802.15.4-ENABLED WIRELESS SENSOR NETWORKS

5147

Fig. 17. Energy expended in the network. (a) Network energy consumption. (b) Energy consumed versus the number of sinks (Static scenario; ∧coef = 10, 20 WPANs).

Fig. 18. Network life time and average time to converge. (a) Network life time (in years). (b) Average time to converge (in seconds).

the frequency of the higher HCs would have been much higher, which is generally an indication of loops in the network, similar to the case with the count-to-innity problem associated with RIP [11]. The following two cost metrics are associated with route convergence (consistency + stability): 1) the energy expended to achieve convergence and 2) the time taken to converge. In Figs. 17 and 18, we present the energy that was consumed in the network for the simulation duration, the estimated life time of the network, and the average time to converge as a function of the update coefcient. Based on the plots in Fig. 17(a), we see that when the routing protocol was disabled (i.e., ∧coef f = 0), the network consumed the least amount of energy. With the routing protocol enabled (i.e., ∧coef f > 0), we see that the higher the update coefcient is, the lower the energy consumed becomes. Fig. 17(a) expresses that the less frequent the route updates, the lower the energy consumed. As shown in the gure, the energy that was consumed in the OUT perimeter scenario was slightly lower than those of the other two scenarios. The reason for this case is that the higher the number of deployed sinks [see Fig. 15(b)], the higher the number of route descriptors in the beacon payload, which increases the size of the overall beacon itself, hence, a slight increase in

the amount of energy required for reception and transmission of beacons, as depicted in Fig. 17(b). The plot in Fig. 17(b) reaches its asymptotic level when the number of deployed sinks is greater than or equal to 8. This case is a result of the length of the Route Dscrpt Count eld in Fig. 4(a), which restricts the number of descriptors in any beacon to a maximum of 8. The lifetime of a wireless network is generally dened as the length of time for which the network functions as expected before becoming unusable. For the energy-constrained WSN, the network becomes unusable when all or a subset of the nodes, particularly those tasked with relaying data, run out of energy. To that end, based on the rate of energy consumption in Fig. 17(a), we estimated the network lifetime as a function of the update coefcient. For instance, when the update coefcient was 10 in Fig. 17(a), the energy that was consumed in the network was 10.84 J (Static scenario). This result translates to 0.54 J apiece for each of the 15 WPAN coordinators and the ve sinks for a simulation duration of 20 min. At that rate, the daily consumption is 39 J per node. Based on the total battery capacity in Table I, the nodes can last for 277 days or 0.76 year before running out of energy. As shown in Fig. 18(a), when the routing protocol is disabled, the nodes can last for about 4.3 years. We will use this result as the

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

5148

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

Fig. 19. Average time to converge and energy expended per WPAN (ve sinks; ∧coef f = 5). (a) Average time to converge (in seconds). (b) Energy expended per WPAN.

Fig. 20. Average end-to-end delay and delivery ratio. (a) Average end-to-end Delay (in seconds). (b) Average delivery ratio.

upper bound for the lifetime of the network. With the update coefcient set to 5, the nodes only last for about half a year, but as we increased the update coefcient, the lifetime of the network steadily inched toward the lifetime upper bound. It might seem like a good idea to extend the lifetime of the network by simply having a larger update interval or coefcient, but having a larger update interval also means that the network would be less reactive to topological changes, because the frequency of route propagation and route updates are dened by the update interval (see Denition 10). For instance, notice that the average time to converge is higher for larger values of the update coefcient, as shown in Fig. 18(b). It sufces to say that, if a WSN topology is one that frequently changes, then a lower update coefcient should be chosen for that network, and the converse should be the case if the topology is expected to be generally less dynamic. Another factor that indirectly affects the convergence time of the routing protocol is the number of WPANs deployed in the network, assuming that the size of the deployment area is kept constant. In Section IV-A, we mentioned that the length of the BI should be directly proportional to the number of deployed WPANs |Nx | within the communication distance of one another if we need to prevent inter-WPAN interference. Therefore, by implication, an increase in |Nx | increases the update interval, because the BI is one of the two parameters

used in its computation. In Fig. 19(a), we show that increasing the number of deployed WPANs has the same impact on the convergence time as an increase in the update interval, as shown in Fig. 18(b). One immediate benet of having a high |Nx | value is that the network is robust, fault tolerant, and connected, because a single WPAN will typically have multiple routes through its neighbor WPANs to any number of sinks. However, this benet comes at the cost of an increase in energy consumption [see Fig. 19(b)], because a WPAN, e.g., WPAN X, will have to enable its radio to receive route beacons from all Nx in every update interval to maintain high connectivity. In Appendix B, we derive an equation that optimizes the energy consumption in the network as |N x| increases without adversely affecting connectivity. To study the impact of the update coefcient on the delivery ratio and the data end-to-end delay for all three scenarios, we sent 300 data frames of 100 B each from WPAN 1 [circled black node in Figs. 14 and 15(a)] to mobile Sink 4. The OutNeighbors to Sink 4 for WPAN 1 were set to only assign enough GTSs to transmit 50 frames in every update interval. This condition meant that the more frequent the update interval becomes, the more the GTSs allocated per unit time are, and therefore, the faster the trafc travels. The impact of this result is captured in Fig. 20(a), where we see a steady increase in the average endto-end delay as we increase the update coefcient. Notice that

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

SHUAIB AND AGHVAMI: ROUTING SCHEME FOR IEEE-802.15.4-ENABLED WIRELESS SENSOR NETWORKS

5149

the end-to-end delay for the OUT Perimeter scenario was higher than that of the other two scenarios when ∧coef f > 15. This case is a direct result of the increased HC between the WPANs and Sink 4, because Sink 4 moved out of the deployment perimeter. In terms of the delivery ratio, the Static and IN perimeter scenarios performed better than the OUT perimeter scenario, as shown in Fig. 20(b). The primary reason for this result was that, for all scenarios except the OUT perimeter scenario, the mobile sink was always within the deployment area. For the OUT perimeter scenario, observe a gradual drop to 0 in the delivery ratio as the update interval increased. This case was a direct result of GTSs not being quickly allocated to data such that the data will reach the mobile sink before leaving the deployment perimeter. In addition, an equation that estimates data trafc delay and the reaction time to topological changes subject to the protocol conguration (e.g., the number of GTSs allocated to data trafc) and the network deployment conguration is derived in Appendix B. VII. C ONCLUSION In this paper, we have proposed a routing scheme that interacts with the IEEE 802.15.4 MAC- and physical-layer specication to achieve the objectives of a WSN routing protocol. So far, we have technically demonstrated the following ve cases: 1) how the route information is propagated through the use of route descriptors, which are contained within the IEEE 802.15.4 beacons; 2) how admission control and bandwidth reservation can be achieved across clusters/WPANs by simply extending the GTS message sequence of the IEEE 802.15.4 specication; 3) how loops can be prevented or minimized using route descriptor rejects and descriptor transmission limitation; 4) how the routing metric can be used to reect the highest quality path, the shortest path, or a combination of both by simply manipulating the LQI and the hop degree exponents; 5) how support for uniform energy dissipation in the network is achieved. We have shown through thorough evaluation that the proposed routing scheme is exible, adapts to topological changes, is stable, and can achieve a high degree of sink connectivity. A PPENDIX A LQI C OMPUTATION The IEEE 802.15.4 specication states that the LQI computation algorithm should use the received signal power of a frame, its signal-to-noise ratio, or a combination of both as input. The standard mandates the following two conditions: 1) The LQI value must be bounded within the interval [0, 255], with 255 being the highest quality signal and 0 being the lowest, and 2) the LQI value must be an integer value that is uniformly distributed within this interval.

It is instructive to restate here that the standard sets the nominal transmit power to 0 dBm and the receiver sensitivity to 85 dBm. Equation (8) meets the rst condition, i.e., L = 255 + 3 P rxdBm (8)

where P rxdBm is the power (in decibel–milliwatts) of a received frame. For example, if we assign P rxdBm = 0, we see that the equation gives a value of L = 255; conversely, if we set P rxdBm = 85, we get L = 0. With (8), the value of L for every other received signal power between the receiver sensitivity and the nominal transmit power is effectively bounded within the interval [0, 255]. To conform with the second mandate, we introduce (9), i.e., LQI = L , L , if L L < 0.5 otherwise. (9)

To explain this case, let us assume that P rxdBm = 65.5, which gives L = 58.5. This value of L is invalid and might have to be truncated to meet the requirement. What we did was to round the L value downward or upward, depending on whether the number after the decimal point was ≥ 5 or < 5. If the number after the decimal point is ≥ 5, we round upward, and we did the opposite when the value was <5, as shown in (9). In addition, in the case where P rxdBm = 65.5 and L = 58.5, the LQI becomes 59. A PPENDIX B O PTIMAL E NERGY C ONSUMPTION V ERSUS A CCEPTABLE D ELAY In this section, we derive two equations that can be used to estimate the amount of energy that was consumed in the network, the delay that was experienced by data trafc, and the reaction time to topological changes subject to the network conguration. There are three components that contribute to the energy E that was consumed in the network as a result of the proposed routing mechanism. The rst component is the number of route descriptors S in a beacon. As mentioned in the main body of this paper, the higher the number of route descriptors in the beacon is, the larger the beacon becomes, and the more the energy that is expended to transmit and receive these beacons [see Fig. 17(b)]. Note that S ≤ 8, as allowed by the Route Dscrpt Count eld (see Fig. 4). The second component is the number of known neighbor WPANs, i.e., |Nx |. As shown in Fig. 19(b), the energy that was consumed by a node partaking in the routing process is directly proportional to its value of |Nx |. The third component, which is the length of ∧, magnies or minimizes the impact of the other two components on E, as discussed in Section VI-A. If we dene E as a vector E = (e) and dene the vector C = (S, |Nx |), then a change in the vector C would also cause a change in the vector E. For instance, C = (S + s, |Nx | + nx ) will cause the change E = (e + ε).

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

5150

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 58, NO. 9, NOVEMBER 2009

Given the linear relationship between E and C, (10) is derived as follows: 1 (ε) = (a11 ∧ a12 ) s nx (10)

The task of the network implementer is to nd that the network conguration, among other things, meets the required objective in terms of the energy budget and delay constraints. Both equations help in this task. R EFERENCES
[1] K. Murray, A. Timm-Giel, M. Becker, C. Guo, R. Sokullu, and D. Marandin, "The European Network of Excellence CRUISE application framework and network architecture for wireless sensor networks," in Proc. IEEE Globecom Workshops, Nov. 2007, pp. 1–6. [2] LR-WPAN, Wireless medium access control (MAC) and physical layer (PHY) specications for low-rate wireless personal area networks (WPANs), 2006, IEEE Tech. Rep. [3] Sun-Spots, 2009. [Online]. Available: http://www.sunspotworld.com/ docs/general-faq.php [4] IPsensor, 2009. [Online]. Available: http://www.archrock.com/ [5] M. Ouwerkerk, F. Pasveer, and N. Engin, "SAND: A modular application development platform for miniature wireless sensors," in Proc. Int. Workshop Wearable Implantable BSN, Apr. 2006, pp. 166–170. [6] NanoSensor, 2009. [Online]. Available: http://www.sensinode.com [7] Xbow, 2009. [Online]. Available: http://www.xbow.com/ [8] V. Mhatre, C. Rosenberg, D. Kofman, R. Mazumdar, and N. Shroff, "A minimum-cost heterogeneous sensor network with a lifetime constraint," IEEE Trans. Mobile Comput., vol. 4, no. 1, pp. 4–15, Jan./Feb. 2005. [9] M. Yarvis, N. Kushalnagar, H. Singh, A. Rangarajan, Y. Liu, and S. Singh, "Exploiting heterogeneity in sensor networks," in Proc. IEEE 24th Annu. Joint Conf. INFOCOM, Mar. 2005, vol. 2, pp. 878–890. [10] J. Luo and J.-P. Hubaux, "Joint mobility and routing for lifetime elongation in wireless sensor networks," in Proc. IEEE 24th Annu. Joint Conf. INFOCOM, Mar. 2005, vol. 3, pp. 1735–1746. [11] A. S. Tanenbaum, Computer Networks. Upper Saddle River, NJ: Pearson Educ., 2003, pp. 359–360. [12] J. N. Al-Karaki and A. E. Kamal, Handbook of Sensor Networks: Compact Wireless and Wired Sensing Systems. Boca Raton, FL: CRC, 2005, pp. 116–138. [13] K. Akkaya and M. Younis, "Survey of routing protocols in wireless sensor networks," Ad Hoc Netw., vol. 3, no. 3, pp. 325–349, May 2005. [14] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, "An applicationspecic protocol architecture for wireless microsensor networks," IEEE Trans. Wireless Commun., vol. 1, no. 4, pp. 660–670, Oct. 2002. [15] A. Manjeshwar and D. Agrawal, "TEEN: A routing protocol for enhanced efciency in wireless sensor networks," in Proc. 15th Int. Parallel Distrib. Process. Symp., Apr. 2001, pp. 2009–2015. [16] Q. Li, J. Aslam, and D. Rus, "Hierarchical power-aware routing in sensor networks," in Proc. DIMACS Pervasive Netw. Workshop, May 2001. [17] A. Woo, T. Tong, and D. Culler, "Taming the underlying challenges of reliable multihop routing in sensor networks," in Proc. 1st ACM SenSys, 2003, pp. 14–27. [18] MultiHopLQI. [Online]. Available: http://www.tinyos.net/tinyos-1.x/tos/ lib/MultiHopLQI/ [19] R. Fonseca, O. Gnawali, K. Jamieson, S. Kim, P. Levis, and A. Woo, The collection tree protocol (CTP). [Online]. Available: http://www.tinyos.net/ tinyos-2.x/doc/txt/tep123.txt [20] D. Puccinelli and M. Haenggi, "Arbutus: Network-layer load balancing for wireless sensor networks," in Proc. IEEE WCNC, Apr. 2008, pp. 2063–2068. [21] J. Luo, J. Panchard, M. Piorkowski, M. Grossglauser, and J.-P. Hubaux, "Mobiroute: Routing towards a mobile sink for improving lifetime in sensor networks," in Proc. 2nd IEEE/ACM DCOSS, Jun. 2006, pp. 480–497. [22] C. Gomez, P. Salvatella, O. Alonso, and J. Paradells, "Adapting AODV for IEEE 802.15.4 mesh sensor networks: Theoretical discussion and performance evaluation in a real environment," in Proc. Int. Symp. WoWMoM, 2006, pp. 159–170. [23] H. Cheng and J. Cao, "A design framework and taxonomy for hybrid routing protocols in mobile ad hoc networks," Commun. Surveys Tuts., vol. 10, no. 3, pp. 62–73, Quarter 2008. [24] P. Samar, M. R. Pearlman, and Z. J. Haas, "Independent zone routing: an adaptive hybrid routing framework for ad hoc wireless networks," IEEE/ACM Trans. Netw., vol. 12, no. 4, pp. 595–608, Aug. 2004. [25] Zigbee-Alliance, Zigbee Specications, 2007. [Online]. Available: www. zigbee.org

where a11 = e/S, and a12 = e/|Nx |. The elements aij depend on the energy consumption model of the hardware platform and can empirically be derived. Equation (10) can be used by the network implementer to estimate how much a chosen value of |Nx | or S affects the energy that was consumed under different deployment scenarios. An increase in S assures that multiple sinks are known to a WPAN, which increases connectivity, but this case also increases E. Similarly, an increase in |Nx | increases E while also increasing the connectivity and fault tolerance of the network. Although the effects of the S component on the E value is bounded (i.e., S ≤ 8), |Nx | is not. However, the impact of |Nx | on E can be reduced without affecting connectivity and fault tolerance if the coordinator of a WPAN, e.g., WPAN X, periodically leaves its receiver disabled whenever any WPAN n ∈ Nx that / satises {n : n ∈ Nxuin n ∈ Nxtout , u = t} transmits its beacon (see Denitions 2–4). The delay that was experienced by data trafc on the path to a sink and the time delay to react to topological changes can be represented by a vector D = (d), which is affected by three components. The rst component is the HC H from a data source to a sink. The closer a data source is to a sink, the lower the value of H, and therefore, the lower the value of D. The second component G is the number of GTSs granted to trafc as it transverses the path from the source to the destination. The more the GTSs G granted by Out-Neighbors along a path, the faster the trafc travels, and hence, there is a reduction in the value of D.1 The third component ∧ multiplies the impact of D on the network, as shown in Fig. 18(b). In essence, the vector D = (d) is affected by another vector T = (H, G) such that a change in T will cause a change in D, i.e., T = (H + h, G g) will cause a change D = (d + δ). One equation similar to (10) is derived as follows: ∧ (δ) = (b11 b12 ) h g (11)

where b11 = d/H, and b12 = d/G. The elements bij in (11) can empirically be derived based on the WSN deployment conguration and the trafc characteristics. Aside from ∧, there are several subtle relationships between (10) and (11). For instance, if WPAN X has multiple paths to a sink U , i.e., with a higher number of elements in Nxuout Nx , then there is a higher probability of having a high G value and, hence, a reduction in D. However, an increase in the cardinality of Nxuout also increases the cardinality of Nx , which, in turn, increases E.

1 Note that the G component does not affect the reaction time to topological changes, because the route-propagation mechanism does not rely on GTSs.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.

SHUAIB AND AGHVAMI: ROUTING SCHEME FOR IEEE-802.15.4-ENABLED WIRELESS SENSOR NETWORKS

5151

[26] 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. 19th ECRTS, Jul. 2007, pp. 125–135. [27] R. Musaloiu-E and A. Terzis, "Minimizing the effect of Wi-Fi interference in 802.15.4 wireless sensor networks," Int. J. Sensor Netw., vol. 3, no. 1, pp. 43–54, Dec. 2007. [28] J. Misic, "Analysis of slave–slave bridging in IEEE 802.15.4 beacon-enabled networks," IEEE Trans. Veh. Technol., vol. 57, no. 3, pp. 1846–1863, May 2008. [29] J. Misic, C. Fung, and V. Misic, "On bridge residence times in master–slave connected 802.15.4 clusters," in Proc. 20th Int. Conf. AINA, Apr. 2006, vol. 2, pp. 286–290. [30] N. Bulusu, J. Heidemann, and D. Estrin, "GPS-less low-cost outdoor localization for very small devices," IEEE Pers. Commun., vol. 7, no. 5, pp. 28–34, Oct. 2000. [31] N. B. Priyantha, A. Chakraborty, and H. Balakrishnan, "The cricket location-support system," in Proc. ACM Mobicom, Boston, MA, 2000, pp. 32–43. [32] A. Savvides, C.-C. Han, and M. B. Strivastava, "Dynamic ne-grained localization in ad hoc networks of sensors," in Proc. ACM Mobicom, 2001, pp. 166–179. [33] J. Misic, F. Amini, and M. Khan, "Performance implications of periodic key exchanges and packet integrity overhead in an 802.15.4-beaconenabled cluster," Int. J. Sensor Netw., vol. 3, no. 1, pp. 38–42, Dec. 2007. [34] Y. Xiao, H.-H. Chen, B. Sun, R. Wang, and S. Sethi, "MAC security and security overhead analysis in the IEEE 802.15.4 wireless sensor networks," EURASIP J. Wireless Commun. Netw., vol. 2006, no. 2, p. 81, Apr. 2006. [35] Y. Xiao, S. Sethi, H.-H. Chen, and B. Sun, "Security services and enhancements in the IEEE 802.15.4 wireless sensor networks," in Proc. GLOBECOM, Nov./Dec. 2, 2005, vol. 3, pp. 1176–1800. [36] OPNET, 2009. [Online]. Available: www.opnet.com [37] "A true system-on-chip solution for 2.4 GHz IEEE 802.15.4/Zigbee (Rev. F)," Chipcon Products From Texas Instruments, 2008. [Online]. Available: http://focus.ti.com/lit/ds/symlink/cc2430.pdf [38] LR-WPAN, Part 15.4—Wireless medium access control (MAC) and physical layer (PHY) specications for low-rate wireless personal area networks (WPANs) amendment 1: Add alternate PHYs, 2007. [39] L. N. Trefethen, The Princeton Companion to Mathematics. Princeton, NJ: Princeton Univ. Press, 2008, pp. 604–615.

A. Hafz Shuaib (S'08) received the B.Sc. degree in computer science from the Ambrose Alli University, Ekpoma, Nigeria, in 2003 and the M.Sc. (top 5% of the class) degree in mobile and high-speed telecommunications networks from Oxford Brookes University, Oxford, U.K. in 2005. He is currently pursuing the Ph.D. degree with the Centre for Telecommunications Research (CTR), King's College London, London, U.K. Prior to joining the CTR in 2006, he was a Software Developer with the Nigerian nancial sector. In 2007, he led the European Network of Excellence Creating Ubiquitous Intelligent Sensing Environments Work Package (CRUISE NoE D113.3) on future needs, research strategy, and visionary applications for sensor networks.

A. Hamid Aghvami (M'87–SM'91–F'05) received the M.Sc. and Ph.D. degrees from the University of London, London, U.K., in 1978 and 1981, respectively. In 1984, he joined the academic staff of King's College London, where he was promoted to Reader in 1989, became a Professor of telecommunications engineering in 1993, and is currently the Director of the Centre for Telecommunications Research. He carries out consulting work on digital radio communications systems for both British and international companies. He is the author of more than 500 technical papers and has given invited talks on various aspects of personal and mobile radio communications and courses on the subject worldwide. Prof. Aghvami is a Fellow of the Royal Academy of Engineering and the Institution of Electrical Engineers. From 2001 to 2003, he was a Member of the Board of Governors of the IEEE Communications Society. He is a distinguished Lecturer of the IEEE Communications Society and has been a Member, Chairman, and Vice Chairman of the technical program and organizing committees of several international conferences. He is also the Founder of the International Conference on Personal, Indoor, and Mobile Radio Communications.

Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on July 19,2010 at 04:50:48 UTC from IEEE Xplore. Restrictions apply.


相关文档

  • A Comprehensive Analysis of low-power operation for beacon-enabled IEEE 802.15.4 wireless networks
  • Emergency access mechanism in IEEE 802.15.4 for wireless body area sensor networks
  • On the Use of IEEE Std. 802.15.4 to Enable Wireless Sensor
  • A Cross-Layer Scheme for Solving Hidden Device Problem in IEEE 802.15.4 Wireless Sensor Networks
  • Security Architecture for IEEE__ 802.15.4-based Wireless Sensor Network
  • Network-Growing Scenarios in IEEE 802.15.4 Wireless Sensor Networks
  • A pairwise key pre-distribution scheme for wireless sensor networks
  • A coverage-preserving node scheduling scheme for large wireless sensor networks
  • Precision Time Synchronization Using IEEE 1588 for Wireless Sensor Networks
  • A node scheduling scheme for energy conservation in large wireless sensor networks
  • 猜你喜欢

  • 2017单位三八妇女节活动方案
  • 小学生社会实践活动总结参考
  • 成汇建设集团有限公司上海第一分公司(企业信用报告)- 天眼查
  • 学校图书馆工作计划范文精选_工作计划.doc
  • 高中语文文学常识的详细介绍
  • 2015年南京大屠杀纪念日国旗下讲话稿
  • kindle亚马逊个人文档不显示_kindle个人文档设置
  • 糖尿病人运动应遵循的原则
  • 每天喝牛奶有什么好处
  • 特种作业持证上岗花名册
  • 四年级数学下册第9单元《数学广角_鸡兔同笼》课件新人教版
  • 高考作文指导:如何准备高考作文材料
  • “‘社会妈妈关爱团,教育帮扶青少年”项目之探讨与实践
  • ★“1”的历险作文_五年级想象作
  • 非常权威的现金流量表分析教程
  • 水泥砼面层施工专项施工
  • 开学式失恋是什么意思
  • 牛顿迭代法求根C语言
  • 非公党工委多措并举开创非公党建新局面
  • 【优质课件】冀教版语文七年级下册第14课蜘蛛1优秀课件.pptx
  • 动手术作文300字_叙事作文
  • 汽车保养的窍门有哪些
  • 育儿知识-为人母,你会发现的40个秘密
  • 在柳州职场新人学java哪里正规?
  • 山东行云物联网科技有限公司(企业信用报告)- 天眼查
  • 一天中最有效的学*时间是这4个时段!家长们知道吗?
  • 【精编范文】流产诊断证明书怎么写-精选word文档 (2页)
  • 小学征文逐梦路上作文优秀
  • vue-cli(2-4)环境模式配置
  • 小学二年级安全教育教案
  • 2016年度寻找最美孝心少年张钊事迹材料
  • 道路清扫保洁管理方案及应急方案
  • 2020年青海玉树成人高考报名时间及9月2日-8日
  • 竞争分析-KIS专业版V12.0VST3用友通标准版
  • 30《回声》ppt课件5
  • 2019年精选英语五年级上册外研版拔高训练第四篇
  • 2018-2019学年人教版政治必修一最新同步精品练习:第5课 第2框 随堂 Word版含解析
  • 衡山县农村信用合作联社瓦铺分社企业信用报告-天眼查
  • 四川成良橡胶制品有限公司(企业信用报告)- 天眼查
  • 光电阴极灵敏度和光电倍增管的总灵敏度
  • 配套K12smaAAA广西桂林市阳朔县阳朔中学2017-2018学年高二数学下学期期末质量检测试题
  • 绥中县塔山镇高氏推拿理疗馆企业信息报告-天眼查
  • 电脑版