Sie sind auf Seite 1von 12

Computer Networks 73 (2014) 142153

Contents lists available at ScienceDirect

Computer Networks
journal homepage: www.elsevier.com/locate/comnet

Event data collection in ZigBee tree-based wireless


sensor networks
Meng-Shiuan Pan , Ping-Lin Liu, Yen-Pei Lin
Department of Computer Science and Information Engineering, Tamkang University, New Taipei City, Taiwan

a r t i c l e

i n f o

Article history:
Received 18 January 2014
Received in revised form 28 May 2014
Accepted 27 June 2014
Available online 23 August 2014
Keywords:
Convergecast
Graph theory
Scheduling
Tree construction
Wireless sensor network

a b s t r a c t
Many previous studies propose to use wake-up scheduling to support energy efcient data
gathering in a wireless sensor network (WSN). However, these previous works schedule network nodes with regular patterns, and thus can only support regular and periodical data
reporting. In this work, we further consider the event data reporting scenario, where the
network may randomly have urgent events. We design an event data collection mechanism
for ZigBee tree-based WSNs. In the network, nodes are divided into routers and end devices.
Routers form the tree backbone of the network, and end devices are connected to routers.
Each router will be assigned to a regular slot and several event slots. To support regular data
reporting, routers and their child devices periodically wake up in their regular slots to
sense and to report. When an event happens, devices nearby the event can wake up more
times in their event slots, and routers can collect more event data from child devices and
report more frequently. In this paper, we design a tree formation algorithm, slot assignment schemes, and event mode operations for the event data collection scenario. Simulation and implementation results show that our designs can support both regular and event
data reporting, and the proposed schemes can effectively increase the amount of report
data when there are events.
2014 Elsevier B.V. All rights reserved.

1. Introduction
Wireless sensor networks (WSNs) are discussed to
develop various kinds of applications including monitoring
[5,13], health care [6], guiding [19], and smart home [16].
In these applications, there are many inexpensive wireless
sensors capable of collecting, storing, processing environmental information, and communicating with neighboring
nodes.
Recently, the standards ZigBee/IEEE 802.15.4 [24,7] are
widely supported in many WSN platforms, where the IEEE
802.15.4 and ZigBee specications dene the physical/link
layer protocols and the protocols above the link layer,
Corresponding author.
E-mail addresses: mspan@mail.tku.edu.tw (M.-S. Pan), 600420391@
s00.tku.edu.tw (P.-L. Liu), 602420068@s02.tku.edu.tw (Y.-P. Lin).
http://dx.doi.org/10.1016/j.comnet.2014.06.016
1389-1286/ 2014 Elsevier B.V. All rights reserved.

respectively. The ZigBee network layer supports star, tree,


and mesh networks. A ZigBee coordinator is responsible
for initializing, maintaining, and monitoring the network.
In a star network, there is a coordinator, which allows
devices associating to it. In tree and mesh networks,
devices can communicate with each other in a multihop
fashion. The network contains one ZigBee coordinator
and multiple ZigBee routers. A device can join a network
as a router or an end device by associating with the coordinator or a parent router. In a tree network, the coordinator
and routers can periodically announce beacons to start
superframes. In ZigBee, each superframe consists of an
active portion followed by an inactive portion. In this work,
we dene an active portion as a slot, and a superframe can
be logically divided into multiple slots. On receiving its
parent routers beacon, a child router or child end device

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

has to wake up for a slot time to sense the environment


and communicate with its parent. However, in a ZigBee
mesh network, regular beacons are not allowed. Since
sending beacons is an important mechanism to support
power management in ZigBee networks, the tree topology
is preferred when energy saving is a desirable feature.
Data collection is a fundamental operation in many
WSN applications. It involves a set of nodes to report their
sensory data via a data reporting tree to a sink. As discussed in [15,23], a ZigBee tree network can indeed support low power and quick data collection for regular data
patterns. But, we observe that it cannot work well when
there are urgent events. For example, in 2.4 GHz PHY, with
1.56% duty cycle, a ZigBee superframe can be up to 251.7 s
(with an active portion of 3.93 s). Based on the above settings, when there is an event, the report interval will be
251.7 s, i.e., the sink can only receive the updated event
information every 4.2 min. And, we can also see that the
amount of the reported data will be restricted by the size
of the active portion. The sink may not be able to obtain
sufcient reports of the events.
In this paper, we design schemes for ZigBee tree-based
WSNs to support both regular and event data collection.
Applications that have such demands include target tracking [11], and volcano/mine monitoring [10,20]. The network contains one sink (ZigBee coordinator), multiple
routers, and multiple end devices. Fig. 1 shows the system
ow of the network, which contains two phases. First, in
the network initialization phase, routers form a tree backbone by the proposed distributed tree formation algorithm.
Then, each router locally selects a regular slot and several
event slots by the proposed slot assignment schemes. After
selecting slots, a router can start to broadcast beacons in
its regular slots to connect end devices, and then the system can enter the next phase. Second, in the data reporting
phase, routers and end devices rst operate in regular
mode. Each router (resp., end device) periodically wakes
up in its parent routers slots, and then report collected

143

(resp., sensed) sensory data to its parent router. Sensory


data is relayed by routers to the sink along the tree backbone. When an event happens, routers and end device follow the proposed event mode operations to process events.
Routers nearby the event location will change to event
mode. An event mode router wakes up in its event slots
to react to the event, and its child end devices follow its
schedule to sense and to report. The routers that located
in the reporting path (to the sink) will be informed to
switch to event mode to facilitate data delivery. When
the event disappears, devices in event mode can switch
back to regular mode.
When assigning slots, a router should avoid using the
same regular slots with its interference neighbors. Conventionally, a router can take its two hop neighbors as its
interference neighbors. In this work, we will redene the
interference relationships between routers. We further
demand that when choosing event slots, a router cannot
select its interference neighbors regular slots as its event
slots. This is because that when a router or an end device
enters the event mode, it should not affect other nodes
regular mode operations. By the above discussions, we
can see that a router can have more choices on event slots
if it has less interference neighbors. In this work, we will
discuss how to construct a better tree to reduce interference neighbors.
The main contributions of this paper are vefold. First,
the proposed tree formation algorithm can reduce interferences by taking routers nearby links into consideration. By
our tree formation scheme, a router can obtain more slots
to support event data reporting. Second, the designed slot
assignment schemes can fully utilize the available slots
to support event data reporting without affecting other
devices regular reporting. Third, the proposed distributed
event mode operations can ensure the reliability of event
reporting. Fourth, the proposed scheme can also be applied
upon any receiver-based MAC protocols (such as RI-MAC
[17], and PW-MAC [18]). Finally, to the best of the authors

Fig. 1. The system ow of this network.

144

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

knowledge, this is the rst work that discusses regular/


event data reporting solutions in a ZigBee tree-based WSN.
The rest of this paper is organized as follows. Section 2
introduces some preliminaries. Sections 3 and 4 present
the proposed algorithms in the network initialization
phase and data reporting phase, respectively. Section 5
shows performance evaluations including packet overhead
analysis, simulations, and implementation results. Finally,
Section 6 concludes this paper.
2. Preliminaries
2.1. Related works
Energy efcient data collection schemes has been investigated in several works [3,9,15,22,23]. Ref. [3] introduces a
centralized solution to support data collection, which goal
is to increase the degree of parallel transmissions, and thus
can decrease report latency. The proposed algorithm
divides nodes into many line segments, and guarantees that
the transmissions of a node in a line segment does not cause
interference to other transmissions in the same line segment. Ref. [9] presents low latency slot assignments by considering trafc patterns (regular or one-shot). When
assigning slots, sensors choose different channels to eliminate interferences. If there are sufcient channels, sensors
can use less slots, and thus the latency can be shortened.
Although using multi-channel can effectively reduces interferences, their designs bring overheads when implementing their protocols on sensors. However, Refs. [3,9] cannot
be applied to ZigBee tree networks since a parent node in
[3,9] needs to be active multiple times per cycle to communicate with its child nodes. In a ZigBee tree network, a parent router only wakes up once per superframe.
Refs. [15,22,23] introduce scheduling schemes for ZigBee tree networks. In [15], the authors dene a minimum
delay beacon scheduling (MDBS) problem. The authors
prove the MDBS problem is a NP-Complete problem, and
propose heuristic slot assignment algorithms. In order to
reduce latency, their slot assignment algorithms assign
slots to child routers prior to parent routers under the constraint that a router cannot use the same slot with its two
hop neighbor routers. In [22], the authors investigate that
for the data collection operations, interference neighbors
can be redened based on the reporting tree structure.
The report latency can be reduced by the increased degrees
of slot reuse. Ref. [23] analyzes the types of interferences
when assigning slots, and denes the risk of slot reuse.
When the risk is low, some nodes are allowed to use the
same slot. This work can be considered as a special case
of [22]. We can see that in [3,9,15,22,23], the authors proposed data collection strategies for WSNs. These works can
indeed support regular data patterns but do not consider
the scenario that the network may have urgent events.
2.2. ZigBee superframes
In a ZigBee tree network, routers follow the denition in
the revised version of IEEE 802.15.4 (IEEE 802.15.4b) [8] to
form superframes. Before forming the network, the

coordinator decides the superframe structure by two


parameters, named beacon order (BO) and superframe order
(SO). The BO and SO decide the lengths of a superframe
and its active potion, respectively. To support low power
operations, the setting of BO and SO should satisfy the relationship 0 6 SO < BO 6 14. Changing the relationship of
BOSO allows the coordinator to adjust the on-duty time
of network nodes. According to the denitions of BO and
SO, a superframe can be divided into 2BOSO active portions.
A router selects one active portion as its outgoing superframe,
i.e., its slot. For a router, the active portion selected by its parent is called its incoming superframe. In an outgoing (resp.,
incoming) superframe, a router is expected to transmit
(resp., receive) a beacon to (resp., from) its child routers
(resp., parent router). Fig. 2 shows an example of superframe, where we set BOSO 3, and there are
2BOSO 8 slots in the superframe. Those slots are labeled
by slot 07. In Fig. 2, there is a router listening to its parents
beacons in slot 0, and sending its beacons in slot 3. The router can enter inactive mode in slots 1, 2, 4, 5, 6, and 7. To
avoid beacon collisions, in ZigBee, a router should not select
the slot that is used by any of its two hop neighbors. In this
work, we will discuss how to loosen this restriction.
2.3. Network models
Given a ZigBee network, we model it by a graph
G V; E, where V contains all routers and the sink t and
E contains all symmetric communication links between
nodes in V. End devices can only associate with routers,
but are not included in V. All nodes in this network are static. From G, we can construct a tree T V; ET , where ET
represents the tree links that span all V and jET j jVj  1.
Based on the denition of BO and SO, this system has
n 2BOSO slots. According to Brooks theorem [21], which
proves that k colors are sufcient to color any graph with a
maximum degree of k, we would assume that n P D,
where D is the maximum number of two hop neighbors
of a router in V. Considering routers having greater capabilities than end devices, we further assume that the transmission range of routers is at least twice that of end
devices. In the network, each router, say v, will be assigned
to a regular slot (sr v ) and several event slots (se v ). (Note
that se v is a set of slots.) When a router v 2 V wakes up in
its regular slot or event slot, it will broadcast a beacon.
When v sends a beacon in its regular slot (resp., in its event
slot), the beacon is used to synchronize with its child nodes
[4,12] (resp., to announce that v is active). We remark that
the decisions of parameter n and slot size should depend
on the needs of applications. For example, some applications allow nodes to aggregate their received reports. In
this case, the network manager can congure a smaller slot
size.
As mentioned above, the beacons sent in the regular
slots are used to maintain the synchronization of the network. To guarantee that the network timing can be correctly maintained, we should avoid beacon collisions in
regular slots. More specically, a routers regular slot
should not be affected by its interference neighbors regular and event slots. In this work, we adopt the interference
model in [14,22] as follow. Assuming a tree T connecting

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

145

Fig. 2. An IEEE 802.15.4b superframe.

all routers in V, for a router v 2 V, we classify its neighbors


(in V) into several sets. First, for vs one-hop neighbors
Nv 2 V, they are classied into three sets:




v p : vs parent in T.
v c : Set of vs children in T.
v m : vs not-in-tree neighbors, i.e., Nv n fv p [ v c g.

Second, for vs two hop neighbors N 2 v 2 V, where


N 2 v NNv n Nv , they are classied into eight sets:

v pp : v p s Parent in T.
v pc : Set of v p s children in T.
v pm : v p s Not-in-tree neighbors and excluding Nv , i.e.,
Nv p n fNv [ v pp [ v pc g.
 v mp : Set of v m routers parents in T.
 v mc : Set of v m routers children in T.
 v mm : Set of v m routers not-in-tree neighbors and
S
excluding Nv , i.e., 8u2v Nu n fNv [ v mp [ v mc g.
 v cc : Set of v c routers children in T.
 v cm : Set of v c routers not-in-tree neighbors and
S
excluding Nv , i.e., 8u2v Nu n fNv [ v [ v cc g.





Note that we demand that for a router v 2 V, each of its


one hop neighbors can only be classied into v p ; v c , or v m .
But, a two hop neighbor of v may be classied into multiple
sets. For example, in Fig. 3, router u is located in sets v pm

and v mp at the same time. Then, according to [22], we have


the following denition on interference neighbors.
Denition 1. The interference set of v 2 V, say Iv , with
respect to T are those routers in Nv ; v pp ; v mp ; v cm , and
v cc sets.
Note that if one of vs two hop neighbors, say u, locates
in any of v pp ; v mp ; v cm , or v cc , v will take u as its interference neighbor. The Denition 1 is so dened to guarantee
that a beacon senders child routers can receive the beacon
without collisions. For example, in Fig. 3, routers u and z
are interference neighbors of v because that u and z are
located in v mp and v cc , respectively. If v and u (resp., v
and z) send beacons at the same time, their beacons will
collide at the receiver of router x (resp., w). On the other
hand, in Fig. 3, router y is located in v mc , and v and y can
send beacons at the same time since they will not affect
the beacon receptions of each others child routers. When
assigning slots for routers, we have to ensure that the slot
assignment is interference-free.
Denition 2. A slot assignment for a network G is interference-free if for any v 2 V, its regular slot does not the
same with the regular and event slots of its interference
neighbors, i.e., 8u 2 Iv ; fsr v g \ fsr u [ se ug ;.
Based on the above denitions, we dene the goals
of tree construction based on the following two
observations:
1. By the Denition 1, we can observe that the sizes of routers interference sets are related to the structure of tree
T.
2. By the Denition 2, we can observe that a router v 2 V
can have more choices on slots if it has less interference
neighbors.

Fig. 3. An example of relationship on interference neighbors.

Denition 3. Given G V; E, our goal of tree construction is to nd a tree T V; ET in G to minimize


P
IT 8v 2V jIv j, where jIv j is the amount of vs interference neighbors based on the tree T.
Note that the above denition on interference neighbors does not include end devices. It can be proved easily
that the existence of end devices do not cause extra interferences between routers since the transmission range of
routers is at least twice that of end devices. Also note that
two interference neighbors are allowed to use the same

146

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

event slot. We will show a scheme to relieve beacon collisions in event slots in Section 4.

3. Network initialization phase


3.1. The tree formation scheme
Before presenting the details of the proposed distributed tree formation scheme, we rst give three observations on how to reduce interference neighbors based on
the view point of a router v 2 V.
 O1: by denition, those vs one hop neighbors Nv are
interference neighbors. We say those neighbors in Nv
are inevitable since the network nodes are static.
 O2: considering a router u 2 v m , we observe that us
parent, i.e., up 2 v mp , may be removed from vs interference set if (i) the up does not locate in set
v pp ; v cm , and v cc and (ii) u can reconnect to another
parent, who is an inevitable interference neighbor of
v. Since the new up is an inevitable interference neighbor of v, it will not induce one more interference
neighbor to v. The interference set of v can be reduced.
On the other hand, considering a router u 2 v mp , on us
view point, v is located in ucm . If we can remove u from
vs interference set, it is equivalent to reduce an interference node in ucm for u.
 O3: assuming a router u 2 v pp (i.e., v is ucc ), if reconnecting us child v p to an inevitable neighbor of v, we can
remove u from vs interference set. On the other hand,
to remove an interference neighbor u 2 v cc for v will
be equivalent to the above discussion since v 2 upp .
Fig. 4 shows an example on the above O2 and O3. In
Fig. 4(a), there exists a router u, which is xs parent
(x 2 Nv ). So, router u is located in v mp . At this time, router
v and u cannot use the same slot. Based on O2, Fig. 4(b)
shows that we reconnect x to p (router p is an inevitable
interference neighbor of v). Then, u will not be vs interference neighbor since u changes to be located in v mm . As a
result, jIv j can be reduced. Moreover, Fig. 4(c) shows
another example based on O3. In this example, we reconnect p (in Fig. 4(a)) to x (Router x is also an inevitable interference neighbor of v). At this time, the router w (resp., v),
which is originally located in v pp (resp., wcc ), will change to
be located in v pm (resp., wmc ). As a result, jIv j can also be
reduced.

Based on the above observations, a router vs interference neighbors can be reduced by removing its neighbors
in v pp and v mp . We have proposed a centralized tree formation scheme in [14]. In this work, we aim to provide a distributed scheme. To ease the operations of the distributed
algorithm, a router v will try to reduce its interference
neighbors by nding those v mp routers only. Before presenting the details of the distributed tree formation, we
rst dene two special sets:
 The set Pv ; v 2 V: a router u 2 Pv if it satises (i)
u 2 v m and (ii) up belongs to N 2 v and locates in set
v mp but not in sets v pp ; v cm , and v cc . Pv contains vs
neighbors, which can change their parents.
 The set Rv ; u; v 2 V and u 2 Pv : a router x 2 Rv ; u
if it satises (i) x is an inevitable interference neighbor
of v, i.e., x 2 Nv and (ii) x 2 um . Rv ; u contains potential new parents for routers in Pv .
The details of the proposed distributed algorithm are as
follows. In the proposed tree formation scheme, routers
rst perform distributed operations to form a BFS-like tree
T, and then they will try to locally reconnect some of their
nearby links to reduce IT. During the tree formation, each
router periodically broadcasts HELLO packets to its two
hop neighbors. Each HELLO contains the sender vs neighbor information (i.e., v p ; v c , and v m ) and vs depth in the
tree T (initially set to 1 before forming the tree). By HELLOs, routers can classify their neighbors into corresponding
sets dened in Section 2.3.
The procedure of forming a BFS tree is started by the
sink ooding a FORM_BFS(t; 0) packet. A router v that
receives a FORM_BFS(u; depthu) packet will connect to
u, set its depth depthv depthu 1, and then rebroadcast a FORM_BFS(v ; depthv ) packet. After broadcasting a
FORM_BFS packet, a router broadcasts HELLOs containing
its updated information. The sink waits for a tbfs time and
then broadcasts a RECONN packet to trigger its neighbors
to perform the tree reconnection procedure. When a router
v receives a RECONN packet, it checks if it receives a
RECONN packet before. If so, it will ignore the RECONN
packet. Otherwise, router v will perform the following
steps.
1. Router v identies if it has neighbors located in Pv set.
If the Pv set is empty, v rebroadcasts the RECONN
packet and ends the procedure. Otherwise, v performs
the next step.

Fig. 4. Observations on reducing interference neighbors.

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

2. Router v starts a timer T w (initially set to t wait ) and periodically broadcasts COMPETE (v ; jIv j) packets to its
two hop neighbors during the timer T w . The COMPETE
packet is used by a router to compete the right of executing tree reconnection procedure with its two hop
neighbors. During timer T w ; v may receive COMPETE,
WIN, and HELLO packets.
(a) If v receives a COMPETE(u; jIuj), v checks if one of
the following two conditions is satised: (i)
jIuj > jIv j and (ii) jIuj jIv j and IDu >
IDv . If so, v loses this competition, and then v
resets the T w timer to twait .
(b) If v receives a WIN(u), which represents that u wins
the competition, v resets its T w timer to t reconn twait ,
and will stop broadcasting its COMPETE packet for
the treconn time.
(c) If v receives a HELLO packet, v updates its two hop
neighbors information according to the HELLO. If
v realizes that its Pv set is empty according to
the updated HELLO, it will rebroadcast a RECONN
packet and end the procedure.
3. When the timer T w expires, v wins the competition.
Router v announces it is the winner by broadcasting a
WIN(v) packet to its two hop neighbors. Then, v puts
its Pv neighbors to a queue Q 1 and performs the next
step.
4. Router v sets u dequeueQ 1. There are two cases.
First, if u ;, it means that there is no router in Q 1.
At this time, v can rebroadcast the RECONN packet
and end the procedure. Second, if u ;; v computes
Rv ; u for u and then v checks all routers in Rv ; u.
For a router x 2 Rv ; u, if depthx < depthu; v puts x
into a queue Q 2. After checking all routers in Rv ; u,
if Q 2 ;, which means that v cannot reconnect u; v will
re-perform the step 4 to extract the next neighbor in
Q 1. Otherwise, v performs the step 5.
5. Router v sets x dequeueQ 2. Again, there are two
cases. First, if x ;; v goes back to step 4 to checks
the next router in Q 1. Second, if x ;; v performs the
step 6 to try to reconnect u to x.
6. In this step, router v sends a CHG(x) packet to u and
waits us reply. (Router u performs the rule R2 below,
and u will reply a NO_CHG(u) or an updated HELLO.)
If v receives a NO_CHG(u), which represents that u
refuses to reconnect, v goes back to step 5 to check
the next candidate in Q 2. If v realizes that u successfully
connects to x from us new HELLOs, v goes back to step 4
to extract the next router in Q 1.
In the reconnection procedure, there is two extra rules
for routers.
R1. For the case that a router u, which is not currently
participating any competition, receives a WIN packet,
u will inhibit itself to broadcast COMPETE for the
treconn time.
R2. For the case that a router u receives a CHG(x)
packet, u checks if jchildNu xj > jchildNu up j, where
jchildNi jj represents the number of js child routers
that are located in Ni. If so, u reconnects to x and
immediately broadcasts the updated HELLOs to its

147

two hop neighbors. Otherwise, u replies a NO_CHG(u)


packet to v.
Note that the rule R1 is to guarantee that a new competition will not be triggered around the winner. Also note
that in the beginning of the procedure, we restrict that
each router can only perform the reconnect procedure
once. This restriction is used to guarantee that the tree
reconnection procedure can be converged. In the step
2(a), we propose that the router that has more interference
neighbors can perform the reconnection procedure earlier.
In the step 2(b), when the router v receives a WIN, it stops
broadcasting COMPETE packets for a t reconn time to avoid
interfering the winners operations. After t reconn , the router
v can restart a new competition.
Theorem 1. After the tree reconnection procedure, the
reporting tree T is still loop-free.

Proof. In the step 4 of the reconnection procedure, we


only put those routers in Rv ; u, which depth values are
smaller than that of u, into the Q 2. This step ensures that
when reconnecting a router u; u will never be reconnected
to its descendants. So, the reporting tree T can be loopfree. h
Theorem 2. When a router u successfully reconnects to a
new
parent
x
in
step
6,
we
can
reduce
jchildNu xj  jchildNu up j interference neighbors for the
network.
Proof. According to the rule R2, the router u will change to
a new parent x if router x has more child routers located in
the Nu than us original parent does. Before reconnecting
u, the router w 2 up is the interference neighbor of those xs
child routers in childNu x because that for any router
i 2 childNu x, the router w is located in imp . On the other
hand, after reconnecting u to x, the router w will not be
the interference neighbor of childNu x routers since for
any router i 2 childNu x, the router w changes to be
located in imm . But, the router x becomes the interference
neighbor of those node in childNu w because that for
any router j 2 childNu w, the router x is located in jmp .
So, when the u decides whether to reconnect to a new parent x, it should consider the size of childNu x and
childNu up . When the jchildNu xj > jchildNu up j, we
can reduce jchildNu xj  jchildNu up j interference neighbors for the network. h
Fig. 5 illustrates an example of Theorem 2, where us
original parent w and the router x have one (i.e., p) and
two child routers (i.e., a and b) located in Nu, respectively. As shown in Fig. 5(a), when the router w is us parent, w is the interference neighbor of a and b because that
w is located in both amp and bmp . If we change us parent to
x (as in Fig. 5(b)), the router w will not be the interference
neighbor of a and b because that w changes to be located in
amm and bmm . However, when reconnecting u to x, the router x will become the interference neighbor of p. But, the
IT can be reduced by one in this example.

148

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

Fig. 5. An example of Theorem 2.

In this algorithm, there are three timers t bfs ; twait , and


treconn . We provide some guidelines for the network manager to set these timers. First, we can see that t bfs is used
to ensure the BFS tree can be formed. The selection of tbfs
should be related to the expected depth of the network.
Second, when setting the value of twait , the network manager should consider the network density. The twait should
be long enough for each router to broadcast several COMPETE packets to its two hop neighbors. Third, treconn is the
time for routers to wait the winner to reconnect links. So,
the value of t reconn should be proportional to the expected
size of Q 1 and Q 2 multiplying the time for unicasting
CHG/NO_CHG packets.
3.2. The slot assignment schemes
Given the T constructed in Section 3.1, in the following,
we present the distributed regular and event slot assignment schemes. Recall that there are n slots in this network.
The sink rst broadcasts an ASSIGN packet to start the regular slot assignment procedure. After selecting a regular
slot, each router v 2 V checks if all its interference neighbors have regular slots. If so, v can start to locally choose
its event slots.
In the following, we present the details of the regular
slot assignment. Initially, each router v 2 V sets sr v ;.
The sink t sets sr t n  1 and then broadcasts an
ASSIGN(t; n  1) packet. When a router v receives an
ASSIGN(u; sr u) packet, it will perform the following steps.
1. Router v checks if u v p and sr v ;. If so, v performs
the step 2. Otherwise, v ignores this packet and ends
this procedure.
2. Router v computes the smallest positive integer l such
that sr u  l mod n sr i; 8i 2 Iv . Then, v chooses
sr v sr u  l mod n as its slot.
3. Then, v broadcasts a DECIDE(v ; sr v ; jIv j) packet to its
two hop neighbors for a time period tdecide . During t decide ,
if v receives a DECIDE(i; sr i; jIij) packet from a router
i. Router v can ignore this packet if i R Iv . When
i 2 Iv ; v checks if (i) jIij > jIv j or (ii) jIij jIv j
and IDv < IDi is satised. If so, v has to stop the
timer and go back to step 2 to reselect a new sr v .
4. After tdecide ; v can nalize its regular slot selection and
broadcast an ASSIGN(v ; sr v ) packet.

Note that in step 2, in order to shorten the latency of


regular reporting, the router computes a regular slot that
is closer to its parents. In step 3, the router with more
interference neighbors can select a regular slot earlier.
When deciding the value of t decide , the network manager
should consider the network density, and the value of
tdecide should be large enough to broadcast a DECIDE packet.
In the following, we present the procedures of event
slot assignment. The basic idea is that each router can
select those slots (among n slots), which are not selected
by its interference neighbors as their regular slots. For
example, assuming n 8, a router vs sr v 2 and its
interference neighbors use slot 0; 5, and 7 as their regular
slots. In this case, v can take slot 1; 3; 4, and 6 as its event
slots. After selecting a regular slot, a router v periodically
broadcasts HELLOs including sr v to its two hop neighbors. Then, v keeps checking if it has obtained all its interference neighbors regular slots. If so, v can decide its event
slots as those n slots excluding sr u; 8u 2 Iv .
We remark that after choosing event slots, v can start to
announce beacons. An end device that receives a beacon
can perform association procedure to the beacon sender.
After associating to a router, the end device can operate
in regular mode.

4. The data reporting phase


After nishing the network initialization phases, routers
and end devices work in regular mode by a distributed
manner. In this work, when routers or end devices sensed
events, they will perform the event mode operations. An
event mode router is further labeled as a source router or
a relay router. A source router is the one that detects an
event from its child end devices, and those ancestor routers
of a source router are assigned to be relay routers. The
detailed operations are as follows. Assume that a router v
detects an event (reported from its child end devices or
detected by itself). Router v rst labels itself as a source
router and then announces an EVENT_MODE(v, on) packet
to its child end devices. When an end device receives an
EVENT_MODE(v, on) packet and realizes that v is its parent,
the device will enter event mode and follow vs event slots
to wake up and report. After broadcasting an EVENT_MODE(v, on) packet, v will inform its ancestors by sending a

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

WAKEUP(v, on) packet to its parent router u in sr u. When


u receives the WAKEUP(v, on) packet, it performs the following steps.
1. Router u records v by a list Lu . (In this operation, u uses
the list Lu is to keep track those its descendant routers
that have become source routers.)
2. If u is in regular mode, it switches to event mode and
sets itself as a relay router.
3. Router u relays the WAKEUP(v, on) packet to its parent
router.
Note that the parent router of u will perform the same
operations as u and so do the other ancestors of the source
router v. In this network, a relay router will wake up in its
event slots to facilitate event data reporting. Also note that
a relay router can also be a source router. A relay router
will not send EVENT_MODE(v, on) packets to its child end
devices unless it becomes a source router.
When the event is relieved, a source router v can broadcast an EVENT_MODE(v, off) packet to its child end devices
to inform them to leave event mode. Then v sends a
WAKEUP(v, off) packet to its parent. When a relay router
u 2 v p receives a WAKEUP(v, off) packet, it relays this
packet to its parent and removes v from list Lu . In this work,
a router u can leave event mode if Lu ; and v itself is not a
source router.
Note that when a router wakes up in its event slots, it
also needs to broadcast beacons. However, the beacon
transmissions may collide. More specically, assume that
routers v and u are neighbors and they both choose a slot,
say slot k, as one of their event slots. If v and u broadcast
beacons at the beginning of slot k, their beacons may collide at some of their child devices receivers. Since the beacons sent in event slots are not for synchronization, the
transmission times of beacons can be shifted to relieve collisions. So, in this work, event mode routers will randomly
defer their beacon transmissions in their event slots.
In the following, we discuss how to maintain the network when there is a router, say v, unexpectedly failed.
We propose network maintenance procedures for vs parent and descendant routers. First, when the parent router
v p identies that v is failed, v p sends a FAIL(v) packet to
the sink. The sink that receives a FAIL() packet will wait
instructions from the network manager or can re-perform
the network initialization phase to refresh the network.
Second, when the router v is failed, its descendant nodes
will not be able to report their data. So, for a router
u 2 v c , it will nd a temporarily parent router p 2 Nu that
can satisfy the following two conditions: (i) the router p
can induce the least regular report latency to u, and (ii)
the
router
p
does
not
locate
in
the
set
C fv c [ uc [ v cc [ ucc g. More specically, the router u
locally
nds
a
router
arg minp fsr p  sr u
nmodnjp 2 fNu n Cgg. Note that the second condition
is to guarantee that the router u will not connect to the
descendant routers of v and u, and thus loops can be
avoided. If the router u cannot nd a temporary parent, it
will dismiss its child routers by sending DISMISS(u) packets to them. When a router receives a DISMISS(.) packet
from its parent, it will pretend that its parent is failed,

149

and then it will try to nd a temporary parent by the above


procedures. After the router u dismisses its child routers, it
can connect to any router in the set C, and no loops will be
formed.
5. Performance evaluations
5.1. Analysis on computational and communication
complexities
In the following, we rst analyze the computational
complexity of the proposed tree formation and slot assignment
schemes.
Let
DI maxfjIv j; 8v 2 Vg
and
DN maxfjNv j; 8v 2 Vg. When forming the BFS tree,
each router costs O1 time to decide its parent. Then, a
router, say v, executes reconnection procedures when it
receives a RECONN packet. In the step 1, the router v costs
at most ODN 2 times to compute set Pv (because that it
has at most DN 2 two hop neighbors). Then, in the step 2,
the computational cost is dominated by the step 2(c),
where v costs ODN 2 times to update its neighbor information. From the step 3 to 6, the router v will have at most
ODN routers in Pv , which implies that the size of Q1 is
bounded by ODN . When processing a router in Q1, the
router v further puts at most ODN routers to Q2. Then,
for a router in Q2, it needs O1 time to check if it can
reconnect to another parent by the rule R2. So, the complexity from step 3 to step 6 is ODN 2 . As a result, the
total computational complexity of the proposed tree formation scheme of a router will be ODN 2 . Moreover, in
the regular slot assignment procedure, the computation
complexity is dominated by the step 2 and step 3. For a
router v, it needs ODN times to nd a slot by the step 2.
For a router v, it has at most ODN neighbors. So, it may
iteratively execute step 2 and step 3 at most ODN times.
So, the computational complexity will also be ODN 2 .
When assigning event slots, a router needs to check at
most ODI neighbors. Since DI < DN 2 , the overall computational complexity of slot assignment of a router will be
ODN 2 .
Next, we analyze the communication complexity of the
entire network when initializing the network. (Note that in
our analysis, we omit the communication complexity of
sending periodical HELLOs.) In our scheme, each router
needs to send one FORM_BFS packet and one RECONN
packet to form the BFS tree and to start the reconnection
procedures, respectively. In the step 26 of the reconnection procedure, a router v may need to compete with at
most ODI routers. So, v may broadcast COMPETE packet
at most ODI times. After winning a competition, v needs
to broadcast one WIN packet. Recall that the size of Q 1 is
bounded by ODN , and for a router in Q 1, it will generate
at most ODN in Q 2. So, v needs to send at most ODN 2
CHG packets. As a result, the communication complexity
of the router v will be O1 DI 1 DN 2 ODN 2 .
When forming the tree, the communication complexity
of the entire network will be OjVj  DN 2 . Moreover, in
the step 23 of the regular slot assignments, a router needs
to broadcast at most ODI DECIDE packets. When choosing

150

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

event slots, routers simply use periodical HELLOs determine their event slots. As a result, when assigning slots,
the communication complexity of the entire network is
OjVj  DI .

that may potentially be induced by routers two hop


neighbors.
Fig. 6(b) shows the results when using S2. We can see
that the results are similar to the one when using S1. Again,
the IT of rCEN is slight better than the one of pOUR. From
Fig. 6(a) and (b), we can see that the proposed pOUR can
efciently reduce IT even that routers in pOUR only
reconnect their neighbors by local decisions.

5.2. Simulation results


In this paper, we develop a simulator based on C language. In our simulations, we x the number of slot n to
be 32. Routers are randomly distributed in a
100  100 m2 square region, and one of them is selected
as the sink. We consider two network congurations,
named S1 and S2. First, in S1, we place 100 routers in the
network and vary the transmission ranges of routers from
17 m to 26 m. Second, in S2, we x the transmission ranges
of routers to be 22 m and vary the number of routers (from
60 to 150). The simulation results are as follows.

5.2.2. Performance on regular data reporting latency


We compare the regular data reporting latency, say
Lr G, (in unit of slots) based on pOUR, rCEN, rBFS, and
rDTA trees. After forming the tree and assigning regular
slots, the Lr G is calculated as follow. For a router v, its
one hop report latency to its parent u is
sr u  sr v modn. And, the report latency Lr v for v to
the sink t is the sum of report latencies of the links on
the tree from v to the sink. The overall latency Lr G will be

5.2.1. Performance on IT
We rst compare our tree formation algorithm
(denoted as pOUR) against the centralized tree formation
algorithm in [14] (denoted as rCEN), the distributed top
down algorithm in [22] (denoted as rDTA), and the distributed scheme in [15] (denoted as rBFS) on IT. Note that we
also apply the proposed interference model to calculate the
IT of rBFS. Fig. 6(a) shows the results when using S1. The
results indicate that when the transmission range becomes
larger, the IT increases (due to the increased one hop and
two hop neighbors of routers). Compare to the pOUR, rDTA
and rBFS, the centralized algorithm rCEN has the smallest
IT. The IT of the pOUR is only 1.5% larger than rCEN.
The rDTA has the largest IT because that, instead of
reducing interference neighbors, the authors in [22] choose
links that can minimize hop-by-hop latency for regular
data reporting. We remark that by pOUR, when setting
the transmission ranges of routers to be 17 m, 18 m, and
26 m, the average one hop neighbors (resp., interference
neighbors) of a router will be 7.66, 8.46, and 16.53 (resp.,
10.41, 11.38, and 20.01), respectively. We can see that
when using the pOUR, most of the increased interference
neighbors are induced by the increased number of one
hop neighbors. The numerical results demonstrate that
the pOUR can effectively eliminate interference neighbors

2600

8v

Fig. 7(a) shows the results when using S1, which indicate a trend that when the transmission range of routers
becomes larger, the Lr G increases because that routers
have more interference neighbors. The network formed
by the rCEN has the longest Lr G. This is because that
the rCEN may cause some routers to have longer hop count
distances to the sink, and thus increases Lr G. We can also
see that the Lr G of the pOUR is slightly higher than the
rDTA. The pOUR can shorten the Lr G by allowing a router
to have more choices when selecting its regular slot. And,
when reconnecting routers, unlike the rCEN, the pOUR will
not arbitrarily lengthen the tree height. Moreover, Fig. 7(b)
shows the results when using S2. We can see that the Lr G
of pOUR approaches the Lr G of rBFS when the network
density becomes higher. This is because that when the network density becomes higher, routers will have more
interference neighbors, and thus the pOUR has less choices
when reconnecting neighbors. On the other hand, when
the network density is lower, the Lr G of pOUR can be
slightly better than the rDTA since a router can reduce its
interference neighbors easier.

4400

pOUR
rCEN
rDTA
rBFS

4000

2000
1800
1600
1400

3200
2800
2400
2000
1600
1200

1200
1000
17

pOUR
rCEN
rDTA
rBFS

3600

2200

Averaged I (T)

Averaged I (T)

2400

Lr G maxfLr v g:

800
18

19

20

21

22

23

24

25

26

60

70

80

90

100 110 120 130 140 150

Transmission range (m)

Number of routers

(a)

(b)

Fig. 6. Simulation results on averaged IT when using (a) S1 and (b) S2.

151

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

34

39

32

36

Averaged Lr (G)

Averaged Lr (G)

30
28
26
24
22
20
pOUR
rCEN
rDTA
rBFS

18
16
17

18

19

20

21

22

23

24

33
30
27
24
21
pOUR
rCEN
rDTA
rBFS

18
25

15
60

26

70

80

Transmission range (m)

90

100 110 120 130 140 150

Number of routers

(a)

(b)

140

140

120

120

Averaged received event


data rate (kbps)

Averaged received event


data rate (kbps)

Fig. 7. Simulation results on averaged Lr G when using (a) S1 and (b) S2.

100
80
60
pOUR
rBFS
rCEN with event op
rDTA with event op
rBFS with event op

40
20
0

17

18

19

20

21

22

23

24

25

26

100
80
60
pOUR
rBFS
rCEN with event op
rDTA with event op
rBFS with event op

40
20
0

60

70

80

90

100 110 120 130 140 150

Transmission range (m)

Number of routers

(a)

(b)

Fig. 8. Simulation results on averaged received event data reports (in unit of KB) per superframe when using (a) S1 and (b) S2.

5.2.3. Performance on event reporting


In this simulation, we x the SO 4 (BO 9). We randomly generate events at any time and place, and the generated events will last for a random period of time. To
compare the performance fairly, we apply our event slot
assignment and event mode operations to the network
using rCEN, rDTA, and rBFS trees. We randomly deploy
1000 end devices in the network. When an end device
senses an event, it reports event data as fast as it can in
its parents regular slot and event slots. Each event data
is xed to 64 bytes. Fig. 8(a) and (b) shows the results of
averaged received event data rate (in kbps) of the sink
when using S1 and S2, respectively. We can observe that
when using the pOUR and rCEN, the sink can receive more
amount of event data reports (from 7% to 70%). When the
network density is higher, the pOUR and rCEN can have
better performance. The results demonstrate that the
pOUR and rCEN can effectively allow a router to have more
event slots. The rCEN can perform sightly better than pOUR
since the rCEN is a centralized approach. When using rBFS,
the sink obtains few reports because that network nodes
can only report event information in regular slots. This
simulation indicates that the proposed system can indeed
support event data reporting scenario.

5.3. Implementation results


We implement the proposed distributed scheme using a
real sensor platform (as shown in Fig. 9(a)). The platform
contains a low-power, single-chip wireless microcontroller
JN5148 [2] with a ZigBee/IEEE 802.15.4-compliant wireless
interface. There are light, temperature, and humidity sensors embedded on the platform. As shown in Fig. 9(b),
the implemented small scale network contains one sink
and 22 routers. There is no end device in our implementation since the used platforms are all router-capable. In our
implementation, we set BO 10 and SO 6. Based on the
settings of BO and SO, the network has 16 slots and the slot
length is about 1 s. We congure routers to use the smallest power, and a router v considers another router u 2 Nv
if us signal strength is higher than a threshold.1 We also
implement a JAVA graphic user interface (GUI) for the network manager to monitor the network. When there are
events, the JAVA GUI can indicate those routers that are in

1
To ease of experiment, we set the threshold to be a large value
(10 dBm). According to our tests, a router v can only put a router, which is
1025 cm away from it, into its Nv .

152

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

(a)

(b)

Fig. 9. (a) The used platform and (b) the implemented scenario in our video.

event mode. In [1], we show a video to briey introduce the


implemented JAVA GUI and our experiment procedures.
In this work, we did the experiments to compare pOUR
with rBFS on the received event data rate of the sink. A router considers that it becomes an event source router if the
sensed value of its light sensor is lower than 50 Lux. When
detecting an event, the source router will continuously
read its sensors values in its regular and event slots, and
then generate event reports. The size of event data packets
(of the application layer) is xed to be 64 bytes. Each router has a 40 KB buffer. When a relay router receives an
event data packet, it rst checks if its buffer is full. If so,
the router drops the oldest packet in its buffer. Otherwise,
it will simply forward the packet to its parent. The sink
records the amount of received event data packets from
each of the event source routers.
Fig. 10 shows the cumulative distribution function (CDF)
of the received event data rate (from 22 routers) of pOUR
and rBFS. From Fig. 10, we can see that when using the rBFS,
the received event data rates of all routers are less than
5 kbps. When using the pOUR, the received event data rates
of all routers are at least larger than 12 kbps. In such a small

CDF

0.8
0.6
0.4
0.2
pOUR
rBFS

0
0

10

20

30

40

50

60

70

Reveived event data rate (kbps)


Fig. 10. The CDF of the received event data rate of 22 routers when using
pOUR and rBFS.

scale network, the received event data rate of pOUR is at


least 2.4 times better than that of rBFS. Moreover, in ZigBee,
the ideal data rate of the low layer is 250 kbps. But, from the
Fig. 10, we can see that the maximum received event data
rate is about 61.5 kbps. We analyze the reason below. In
this experiment, the sink owns 12 event slots. When a router v, which is a child router of the sink, detects an event,
the sink will wake up in 13=16 81:3% portion of the
superframe to receive event data packets from v. Since
the event data packets are generated in vs application
layer, the ZigBees MAC and physical layers will attach some
headers and tails (about 24 bytes) to those packets. So, the
portion of the application data payload is about 72%. And,
according to our previous experiments, the saturation
throughput of the used platform is only 48% of the 250
kbps. As a result, the ideal received event data rate (in the
application layer of the sink) will be 250  81:3%
72%  48% 70:2 kbps. But, the maximum received event
data rate cannot close to the ideal value due to packet losses
during the experiments.
From Fig. 10, we can see that 4 routers can report more
packets to the sink. This is because that they are the sinks
child routers, and can fully utilize the sinks event slots. In
this experiment, 9 routers are two hops away from the sink.
Their ideal event data rate (observed by the sink) will be
less than half of the maximum event data rate since their
parent routers need to spend at least half of their times to
receive event packets. From this result, the observed smallest rate is 12.1 kbps. This is because that the corresponding
router is four hops away from the sink, and one of its relay
routers have only 8 event slots. Moreover, in this experiment, we only generate one event in one experiment. However, our implementation can indeed support multiple
events. But, when a relay router needs to relay packets from
multiple source routers, the relay router cannot fairly schedule the transmission times of those incoming data packets.
It is possible that a source routers event data packets will
be queued in the relay routers buffer for a long period of
time. So, we need to further implement an in-router packet
handling/scheduling method to facilitate relaying event
packets. (This is considered to be out of scope of this paper
and will be our future work.).

M.-S. Pan et al. / Computer Networks 73 (2014) 142153

6. Conclusions
In this paper, we have proposed a system to support
both regular and event data reporting in the low-dutycycled ZigBee tree network. We design a tree formation
algorithm, slot assignment schemes, and event mode operations to achieve our goals. The slot assignment schemes
can assign several event slots for routers by the help of
the tree formation algorithm (which aims at minimizing
interference neighbors of routers). The designed event
mode operations do not affect the regular operations of
other network nodes. Simulation results show that our tree
formation algorithm can effectively decrease total number
of interferences. Compare to other works, our slot assignment schemes can not only decrease latency for regular
data reporting, but also can effectively increase the amount
of report data when there are events. In this paper, we also
show a prototyping system, and perform some experiments on it. We are going to enhance our system, and then
deploy the system in real environment to support disaster
and wildlife monitoring.

Acknowledgement
M.-S. Pans research is sponsored by NSC Grants 1022219-E-032-001 and 101-2218-E-032-004-MY3.

References
[1] Our Prototyping System. <http://youtu.be/wOV7CtfW944>.
[2] Jennic JN5148. <http://www.jennic.com/>.
[3] H. Choi, J. Wang, E.A. Hughes, Scheduling for information gathering
on sensor network, ACM/Springer Wireless Networks 15 (1) (2009)
127140.
[4] J. Elson, L. Girod, D. Estrin, Fine-grained network time
synchronization using reference broadcasts, in: Proc. of the USENIX
Symposium
on
Operating
Systems
Design
and
Implementation(OSDI), 2002.
[5] J. Hayes, S. Beirne, K.-T. Lau, D. Diamond, Evaluation of a low cost
wireless chemical sensor network for environmental monitoring, in:
Proc. of IEEE Sensors Conference, 2008.
[6] H. Huo, Y. Xu, H. Zhang, Y.-H. Chuang, T.-C. Wu, Wireless-sensornetworks-based healthcare system: a survey on the view of
communication paradigms, Int. J. Ad Hoc Ubiquitous Comput.
(IJAHUC) 8 (3) (2011) 135154.
[7] IEEE Computer Society. IEEE 802.15.4: Wireless medium access
control and physical layer specications for low-rate WPANs,
October 2003.
[8] IEEE Computer Society. IEEE 802.15.4b: Wireless Medium Access
Control and Physical Layer Specications for Low-rate WPANs,
September 2006.
[9] O.D. Incel, A. Ghosh, B. Krishnamachari, K. Chintalapudi, Fast data
collection in tree-based wireless sensor networks, IEEE Trans. Mobile
Comput. 11 (1) (2012) 8699.
[10] M. Li, Y. Liu, Underground coal mine monitoring with wireless
sensor networks, ACM Trans. Sensor Networks 5 (2) (2009) 129.
[11] C.-Y. Lin, W.-C. Peng, Y.-C. Tseng, Efcient in-network moving object
tracking in wireless sensor networks, IEEE Trans. Mobile Comput. 5
(8) (2006) 10441056.
[12] A. Marco, R. Casas, J.L.S. Ramos, V. Coarasa, A. Asensio, M.S. Obaidat,
Synchronization of multihop wireless sensor networks at the
application layer, IEEE Wireless Commun. 18 (1) (2011) 8288.
[13] M.-S. Pan, H.-W. Fang, Y.-C. Liu, Y.-C. Tseng, Address assignment and
routing schemes for ZigBee-based long-thin wireless sensor
networks, in: Proc. of IEEE Intl Conference on Vehicular
Technology Conference (VTC), 2008.
[14] M.-S. Pan, P.-L. Liu, C.-F. Cheng. Convergecast in ZigBee tree-based
wireless sensor networks, in: Proc. of IEEE Wireless Communications
and Networking Conference (WCNC), 2013.

153

[15] M.-S. Pan, Y.-C. Tseng, Quick covergecast in ZigBee beacon-enabled


tree-based wireless sensor networks, Comput. Commun. (ComCom)
31 (5) (2008) 9991011.
[16] M.-S. Pan, L.-W. Yeh, Y.-A. Chen, Y.-H. Lin, Y.-C. Tseng, A WSN-based
intelligent light control system considering user activities and
proles, IEEE Sensors J. 8 (10) (2008) 17101721.
[17] Y. Sun, O. Gurewitz, D.B. Johnson, RI-MAC: a receiver-initiated
asynchronous duty cycle MAC protocol for dynamic trafc loads in
wireless sensor networks, in: Proc. of ACM Intl Conference on
Embedded Networked Sensor Systems (SenSys), 2008.
[18] L. Tang, Y. Sun, O. Gurewitz, D.B. Johnson, PW-MAC: an energyefcient predictive-wakeup MAC protocol for wireless sensor
networks, in: Proc. of IEEE INFOCOM, 2011.
[19] Y.-C. Tseng, M.-S. Pan, Y.-Y. Tsai, Wireless sensor networks for
emergency navigation, IEEE Comput. 39 (7) (2006) 5562.
[20] G. Werner-Allen, K. Lorincz, M. Welsh, O. Marcillo, J. Johnson, M.
Ruiz, J. Lees, Deploying a wireless sensor network on an active
volcano, IEEE Internet Comput. 10 (2) (2006) 1825.
[21] D.B. West, Introduction to Graph Theory, Prentice Hall, 2001.
[22] F.-J. Wu, Y.-C. Tseng, Distributed wake-up scheduling for data
collection in tree-based wireless sensor networks, IEEE Commun.
Lett. 13 (11) (2009) 850852.
[23] L.-H. Yen, Y.W. Law, M. Palaniswami, Risk-aware distributed beacon
scheduling for tree-based ZigBee wireless networks, IEEE Trans.
Mobile Comput. 11 (4) (2012) 692703.
[24] ZigBee specication version 2006, ZigBee document 064112, 2006.

Meng-Shiuan Pan received the BS and MS


degrees from the National Chung Cheng University and National Tsing Hua University,
Taiwan, in 2001 and 2003, respectively, and
the Ph.D. degree from the National Chiao Tung
University, Taiwan, in 2008. After obtaining
his Ph.D degree, Dr. Pan worked in HTMM
corporation, Hsin-Chu, Taiwan for three years,
and dedicated in designing and implementing
3GPP R99/R5/R6 layer 2/3 protocol stacks.
Since 2011, Dr. Pan is an assistant professor of
department of computer science and information engineering, Tamkang University, Taiwan. He also serves as an
associate editor of International Journal of Distributed Sensor Networks
(IJDSN). His research interests include wireless sensor network, mobile
computing, and LTE-A networks.

Ping-Lin Liu received the BS and MS degrees


from the Aletheia University and Tamkang
University, Taiwan, in 2011 and 2013,
respectively. His research interests include
wireless sensor network and mobile application design.

Yen-Pei Lin received the BS degree from the


Tamkang University, Taiwan, in 2013. He is
now pursuing his master degree in Tamkang
University. His research interests include
wireless sensor network and social network.

Das könnte Ihnen auch gefallen