Conference Paper

Real-Time Coordination and Routing in Wireless Sensor and Actor Networks

DOI: 10.1007/11759355_34 Conference: Next Generation Teletraffic and Wired/Wireless Advanced Networking, 6th International Conference, NEW2AN 2006, St. Petersburg, Russia, May 29 - June 2, 2006, Proceedings
Source: DBLP
ABSTRACT
In Wireless Sensor Actor Networks (WSAN), sensor nodes perform the sensing task and actor nodes take action based on the sensed phenomena in the field. To ensure efficient and accurate operations of WSAN, new communication protocols are imperative to provide sensor-actor coordination in order to achieve energy-efficient and reliable communication. Moreover, the protocols must honor the application-specific real-time delay bounds for the effectiveness of the actors in WSAN.
In this paper, we propose a new real-time coordination and routing (RCR) framework for WSAN. It addresses the issues of coordination among sensors and actors and honors the delay bound for routing in distributed manner. RCR configures sensors to form hierarchical clusters and provides delay-constrained energy aware routing (DEAR) mechanism. It uses only cluster-heads to coordinate with sink/actors in order to save the precious energy resources. The DEAR algorithm integrates the forwardtracking and backtracking routing approaches to establish paths from source nodes to sink/actors. In the presence of the sink in WSAN, it implements the centralized version of DEAR (C-DEAR) to coordinate with the actors through the sink. In the absence of sink or ignoring its presence, there is a distributed DEAR (D-DEAR) to provide coordination among sensors and actors. Cluster-heads then select the path among multiple alternative paths to deliver the packets to the actors within the given delay bound in an efficient way. Simulation experiments prove that RCR achieves the goal to honor the realistic application-specific delay bound.

Full-text

Available from: Ghalib Shah, Nov 30, 2015
Real-Time Coordination and Routing in
Wireless Sensor and Actor Networks
Ghalib A. Shah, Muslim Bozyi˘git
1
,
¨
Ozg¨ur B. Akan, and Buyurman Baykal
2
1
Department of Computer Engineering,
Middle East Technical University, Ankara, Turkey 06531
{e135333, bozyigit}@metu.edu.tr
2
Department of Electrical and Electronics Engineering,
Middle East Technical University, Ankara, Turkey 06531
{akan, baykal}@eee.metu.edu.tr
Abstract. In Wireless Sensor Actor Networks (WSAN), sensor nodes
perform the sensing task and actor nodes take action based on the sensed
phenomena in the field. To ensure efficient and accurate operations of
WSAN, new communication protocols are imperative to provide sensor-
actor coordination in order to achieve energy-efficient and reliable com-
munication. Moreover, the protocols must honor the application-specific
real-time delay bounds for the effectiveness of the actors in WSAN.
In this paper, we propose a new real-time coordination and routing
(RCR) framework for WSAN. It addresses the issues of coordination
among sensors and actors and honors the delay bound for routing in dis-
tributed manner. RCR configures sensors to form hierarchical clusters
and provides delay-constrained energy aware routing (DEAR) mecha-
nism. It uses only cluster-heads to coordinate with sink/actors in order to
save the precious energy resources. The DEAR algorithm integrates the
forwardtracking and backtracking routing approaches to establish paths
from source nodes to sink/actors. In the presence of the sink in WSAN,
it implements the centralized version of DEAR (C-DEAR) to coordinate
with the actors through the sink. In the absence of sink or ignoring its
presence, there is a distributed DEAR (D-DEAR) to provide coordina-
tion among sensors and actors. Cluster-heads then select the path among
multiple alternative paths to deliver the packets to the actors within the
given delay bound in an efficient way. Simulation experiments prove that
RCR achieves the goal to honor the realistic application-specific delay
bound.
1 Introduction
Recent advances in the field of sensor networks have led to the realization of
distributed wireless sensor networks (WSN). A WSN is composed of large num-
ber of sensor nodes, which are densely deployed in the sensor field in a random
fashion with a sink node. The task of sensor nodes is to detect the events in
the sensors field and route them to the sink node, which is responsible for the
monitoring of the field.
Y. Koucheryavy, J. Harju, and V.B. Iversen (Eds.): NEW2AN 2006, LNCS 4003, pp. 365–383, 2006.
c
Springer-Verlag Berlin Heidelberg 2006
Page 1
366 G.A. Shah et al.
Recently, the capabilities of the WSN are extended to include the actor nodes
responsible with taking action against the detected events [1]. Such architecture
is called a wireless sensor and actor networks (WSAN), where a small numbers
of actors, as compared to sensors, are spread in the sensor field as well. Actors
are mostly mobile and resource-rich devices and can be thought to form a mobile
ad hoc network of their own. This paradigm of WSAN is capable of observing
the physical world, processing the data, making decisions based on the sensed
observation and performing appropriate actions. Typically, the architecture of a
WSAN consists of sensors which sense the phenomena, a sink that collects the
data from the sensors to process and actors that act upon the command sent
by the sink. In the literature, such architecture is known as semi-automated ar-
chitecture. An architecture in which sensor nodes send information to the actor
nodes directly without the involvement of sink node is called an automated archi-
tecture [1]. It is apparent that the communication path in a semi-automated ar-
chitecture introduces significant delay, which is not acceptable for delay-sensitive
applications. For example, consider a military application where sensors in the
battlefield will detect the movement of red forces and send the information to
the sink which is situated in a remote command and control station. The sink
then triggers an action through an actor to counter in the threat area. In this
case unnecessary delay is introduced due to sensor-sink communication which
could be removed if the actors can take localized actions without the involve-
ment of the sink; depending on the data sent by the sensors. Hence, the most
challenging task in WSAN is the coordination between sensors and actors to
provide real-time response.
In WSAN, the effective sensor-actor coordination requires the sensors to know
the right actors and the routes to reach them. Moreover, it requires delay es-
timates for all possible routes. In addition to energy constraints as in WSN,
WSAN also imposes timing constraints in the form of end-to-end deadlines.
Clearly, there is a need for real-time communication protocols for WSAN, which
provide effective sensor-actor coordination while consuming less energy.
There have been considerable efforts to solve the routing problem in wireless
sensor networks [3], [5], [6], [10], [11], [12], [13], [14]. However, these protocols
do not consider the heterogeneity of WSAN. Moreover, none of these protocols
provide sensor-actor coordination and real-time routing. A coordination frame-
work [2] for WSAN has been proposed that is an event-based reactive model of
clustering. Cluster formation is triggered by an event so that clusters are created
on-the-fly to optimally react to the event itself and provide the required relia-
bility with minimum energy expenditure. Reactive cluster formation algorithms
have the disadvantage that they consume precious time on event occurrence for
cluster formation. Hence for real-time coordination such an architecture is not
suitable. Moreover, cluster to actor routing in [2] is done using greedy geographi-
cal approach. A packet forwarding node finds the next hop node according to the
greedy approach failing to do so results into a packet loss as the packets enters
into a void region. Since the work assumes that the network is dense therefore it
does not propose any void region prevention or recovery mode implementation.
Page 2
Real-Time Coordination and Routing in WSAN 367
Consequently, there exists no unified solution which addresses the real-time co-
ordination as well as routing problem for the heterogeneous WSANs.
In this study, we propose a real-time coordination and routing (RCR)frame-
work, which addresses the sensor-actor coordination with real-time packet deliv-
ery in the semi-automated architecture as well as automated architecture. RCR,
incorporates the two components, namely DAWC and DEAR. DAWC is our
heuristic clustering protocol used to dynamically configure the sensor nodes in
the form of clusters to achieve energy efficiency. Whereas, RCR achieves the real-
time demand τ of packet delivery through our delay-constrained energy aware
routing (DEAR) protocol that is the first and foremost aim of the protocol. The
DEAR protocol, described in Section 3, establishes a backbone network by inte-
grating the forward tracking and backtracking mechanism that provides all the
possible routes towards target nodes (sink/actors). The path selection criterion
is based on the packet delay as well as the balance consumption of energy of sen-
sor nodes. In the presence of the sink in WSAN, it implements the centralized
version of DEAR (C-DEAR) to coordinate with the actors through the sink.
On the other hand, when there is no sink or central node in WSAN, it provides
the distributed version of DEAR (D-DEAR) for coordination among sensors and
actors. Performance evaluation study reveals that RCR addresses the real-time
coordination and routing requirements of WSANs.
The remainder of the paper is organized as follows. In Section 2, we present
the cluster formation procedure. We discuss the route path computation and
sensor-actor coordination in Section 3. Performance evaluation and results are
considered in Section 4. Finally, the paper is concluded in Section 5.
2 Hierarchical Configuration of Sensor Nodes
To provide a real-time coordination among sensors and actors in WSAN, RCR
configures the sensor nodes hierarchically. The operations of the routing pro-
tocols are discussed in Section 3. The configuration of sensor nodes to achieve
real-time coordination is discussed in this section. We propose so called Dy-
namic Weighted Clustering Algorithm (DAWC). The operations of DAWC con-
sist of cluster formation of sensor nodes, delay budget estimation for forwarding
a packet from the cluster-heads and to guarantee the packet delivery within the
given delay bound τ.
There are many clustering algorithms [4], [7], [8], [9] proposed in the literature
but unlike these studies, DAWC is neither periodical clustering procedure nor the
cluster size is fixed in terms of hops. It adapts according to the dynamic topology
of the sensor and actor networks. Our work is motivated from the previous
work “A Weight Based Distributed Clustering” [8]. However, unlike [8], DAWC
provides the cluster formation procedure to cope with the dynamic number of
hops in a cluster and provides support for real-time routing. Cluster formation is
based on the weighting equation formulated in the Section 2.2, which sets weight
to different application parameters according to the applications need. DAWC
adapts to the variation in the sensors field and can be optimized accordingly.
Page 3
368 G.A. Shah et al.
Once the cluster has been formed, cluster-heads get estimates of delay budget
1
of their member nodes. The delay budgets of member nodes help to build the
delay-constrained energy efficient path.
When the sensor nodes are not uniformly deployed in the sensor field, the
density of nodes could be different in different zones of the field. Choosing an
optimal number of clusters k
opt
, which yield high throughput but incur latency
as low as possible, is an important design goal of DAWC.
2.1 Optimal Clustering
In this section, we evaluate the optimal number of clusters k
opt
and, hence, the
optimal size of clusters. We assume the uniform deployment of sensor nodes
and devise a formula to find out k
opt
. Later, we see the implication of it to
non-uniform deployment for optimal configuration in Section 2.1.
r
ClusterDangling Nodes
M
M
(a)
RSink
r
(b)
Fig. 1. Network model to formulate the optimal clusters. Fig 1(a) represents the model
to find the probability of DN nodes. Fig 1(b) illustrates the model of routing packets
from cluster-heads to the sink node.
Optimal Clusters in the Field. Each member node transmits its data packet
to the cluster-head. Let r be the transmission radius of each node regardless of
its functioning. In the clustering process, there is some probability that a number
of dangling nodes
2
(DN ) may exist due to the density of nodes or coverage of
the elected cluster-head. Let us first find out the probability of such nodes. To
do that, we map the sensor field (M × M) to non-overlapping circles of radius
r as shown in Fig 1(a) and assume that the nodes lying outside the boundary
of the circle are DN nodes and the others are member nodes. These DN nodes
require affiliating to cluster-head through the nodes insides the circle (member
nodes). The squared field M
2
can be packed by M
2
/(2r)
2
non-overlapping cir-
cles of radius r.
1
delay budget is the time to deliver the data packet from the cluster-head to the
member node.
2
Nodes which have not joined any group or cluster are referred as dangling nodes.
Page 4
Real-Time Coordination and Routing in WSAN 369
Thus, the probability γ of a multi-hop member is
γ =
M
2
(2r)
2
×
(2r)
2
πr
2
M
2
0.214
Let E
elec
be the energy consumed by the electronic circuitry in coding, mod-
ulation, filtration and spreading of the signal. Whereas,
amp
r
2
is the energy
consumed for signal amplification over a short distance r.Thus,theenergycon-
sumedbyeachmembernodeintransmittingapacketofsizel is
E
Member
= l(E
elec
+
amp
r
2
(1 + γ))
The above equation can be simplified by taking the area as circle given in Eq.
16 of [15].
E
Member
= l(E
elec
+
amp
M
2
(1 + γ)
2πk
)
Let us assume that the sensory field is covered by a circle of radius R,wherethe
sink node lies at the center of this circle as shown in Fig 1(b). This assumption
is made for routing packets from cluster-heads to the sink. The assumption is
reasonable because it is less likely that a node lying outside the boundary of circle
will be elected as cluster-head due to the low weight than the nodes inside the
circle. Cluster-heads do not extend their transmission range to transmit packets
directly to the sink node and, therefore, has the same radius r as member nodes.
We adapt the multi-hop model proposed by [17] to route packets from cluster-
head to the sink.
In the model, a circle is divided into concentric rings with the thickness r.The
energy spent to relay the packet from outside ring towards inside ring is l(2E
elec
+
amp
r
2
). The number of hops Γ require to route packet from cluster-head to sink
node can be calculated by
R
r
(1 ¯h). Where ¯h is the probability that the cluster-
head is close enough to the sink to directly transmit packets. This probability can
be calculated by using the nodes distribution in the rings given in [17].
¯h =
r
R
R/r
i=1
R
2
(ir)
2
M
2
Since packets from the cluster-heads far from the sink node are relayed through
intermediate nodes. Therefore, if Λ(i) is the number of neighbors of a node i then
Λ(i) × E
elec
is the energy consumed by the electronic circuitry of the neighbors
of forwarding node i during the propagation of packet. The number of neighbors
(Λ) of any node can be found as
Λ = n
πr
2
M
2
Hence, the energy consumed in routing data from cluster-head to sink is mea-
sured as
E
CHSink
= l(ΛE
elec
+ E
elec
+(2E
elec
+
amp
r
2
+ ΛE
elec
)Γ ).
Page 5
370 G.A. Shah et al.
The total energy dissipated by the network is
E
total
= l((n + )E
elec
+ k(2E
elec
+
amp
r
2
+ ΛE
elec
)Γ
+ n
amp
M
2
(1 + γ)
2πk
)
For r<R, the optimum value of k can be found by taking the derivative of
above equation with respect to k and equating to zero
k
opt
n(1 + γ)
(2π(1 +
2E
elec
amp
r
2
+
ΛE
elec
amp
r
2
))Γ
×
M
r
(1)
It is noteworthy that the optimal value depends on the transmission range r of
the nodes. For long range of transmission, the value of optimal clusters k
opt
is
small. This is in contrast to the optimal clustering in SEP [16] that is independent
of range parameter. For example, Let us assume that E
elec
=50nj/bit and
amp
=10pj/bit/m
2
for experiments and n = 100, M = 100 with the sink at
center of the field (x =50,y = 50). Then the value of radius R is obtained by
drawing a circle at x =50,y = 50 to cover the field. The estimated value is
R = 60 and let set the range r of individual nodes to 25. In this scenario, we
obtain the value of k
opt
10. By increasing the range of nodes to 40 meters, we
obtain k
opt
7. Whereas, the value of k
opt
in SEP [16] is 10 regardless of the
transmission coverage of individual nodes.
Optimal Cluster Size. Besides choosing the optimal value k
opt
for number of
clusters, the number of member nodes in a cluster is as important as the number
of clusters. The optimal value of member nodes M
opt
helps in load balancing of
clusters and ensures efficient MAC functioning. Head nodes use more energy
than the member nodes. Since the sensors are energy-constrained devices and
a cluster-head is selected from the homogeneous nodes, the number of member
nodes in a cluster should ensure the longevity of the cluster-head as long as
possible.
When the deployment is uniform then the M
opt
can be easily found by n/k
opt
.
However, for non-uniform deployment, the number of member nodes depends on
the density in a particular zone of the sensor field. Therefore, we put the max-
imum and minimum limits M
Min
and M
Max
respectively on the size of cluster
such that we still achieve k
opt
clusters in non-uniform deployment. Suppose M
i
is the number of neighboring nodes of any ith node and Max(M
i
)isthemaxi-
mum number of neighboring nodes that any of the ith neighbor node have. We
measure density of nodes in a particular zone by comparing the neighbor nodes
M
i
with M
opt
. We can conclude that the deployment is:
M
i
/M
opt
> 1,dense
M
i
/M
opt
1,uniform
M
i
/M
opt
< 1,sparse
Page 6
Real-Time Coordination and Routing in WSAN 371
We set the limits M
Min
and M
Max
as:
M
Max
= Max(M
opt
,Max(M
i
))
That is, the maximum of M
opt
and maximum number of neighbors of any cluster-
head at the time of cluster formation.
M
Min
= M
opt
× Min(M
opt
,Max(M
i
))/M
Max
These limits allow the configuration to manage the dense as well as sparse de-
ployment of nodes.
2.2 Cluster Formation
The first phase of DAWC is to form k
opt
number of clusters. During the forma-
tion of clusters, each cluster-head gets the delay budget of each of its member
node. The delay budget is used to identify an appropriate node to send delay-
constrained data packet. The cluster election procedure is based on calculating
weight for each sensor node in the sensor field and it chooses the head that has
the maximum weight. The weighting equation is given in cluster-head election
procedure. We define weight threshold of the cluster-head to rotate the cluster-
heads responsibility among all the potential nodes. A cluster is not strictly orga-
nized to 1-hop but it accepts the membership of a node that could not reach any
cluster in the first phase of cluster formation. Therefore, a cluster can include
n-hop members, for n 1. Although the operations of the protocol starts after
the first phase of cluster formation, there may still exist some DN nodes.
We assume that the nodes are aware of their geographical locations through
some localization devices like GPS. In the next section, we describe the details
of computing the delay budget and cluster formation procedure is presented in
Section 2.2.
Delay Measurement. When nodes are initially deployed in the field, every
node i broadcasts its ID, which is added in the neighbors list by all the nodes
that receive this broadcast. A node that receives this broadcast, computes the
delay delay
s
of the packet received from its neighbors along with the delay
budget delay
r
. delay
s
is the delay of the packet experienced and delay
r
is the
delay that the sender estimated when some packet was received from the receiver
i.e delay
r
: sender receiver, delay
s
: sender receiver.
The total delay in transmitting a packet from one node to a node in its neigh-
bor is measured by the following factors: queue, MAC, propagation and receiving
delay represented by T
q
, T
Mac
, T
Prop
and T
Rec
respectively. The wireless channel
is asymmetric that does not imply any synchronization mechanism. Therefore,
the delay is measured partially at both the sender and receiver. Sender measures
the delay L
s
until the start of transmission that includes the queue delay as well
as the MAC contention delay. Whereas, the receiver adds the factor L
r
as sum
of propagation delay and receiving delay to get the total packet delay L
h
.
L
s
= T
q
+ T
Mac
,L
r
= T
Prop
+ T
Rec
Page 7
372 G.A. Shah et al.
The hop latency L
h
can be computed as sum of these factors:
L
h
= L
s
+ L
r
Initially, the delay is measured by exchanging the hello beacons. Each node
maintains this value in its neighborhood table that contains the fields {ID,
delay
r
, delay
s
, energy, weight}. To get more close to the accurate measurement
of packet delay, the delay value is updated when the events flow from member
nodes to cluster-head. It is due to the fact that the size of data packet may differ
than the hello beacon that may experience different delay.
For the d-hop member of a cluster, packets are forwarded by the intermediate
nodes to the cluster-head. Each intermediate nodes calculates the delay L
h
of
the packet and forwards the packet to next hop by adding its L
h
in L
s
.After
following through some intermediate nodes, cluster-head gets the packet and
adds its factor L
r
as the receiver of the packet. Hence the cumulative delay
delay
s
of a member node d hops away from its cluster-head is computed as:
delay
s
=
d
i=1
L
h
i
(2)
Member nodes compute the delay estimate delay
s
of their cluster-head in this
way through cluster-head announcement beacon. When member nodes broad-
casts hello beacons, they put the delay
s
of cluster-head as delay
r
into the beacon.
Cluster-head gets the delay budget delay
r
for its members and use this value in
routing.
Cluster-Head Election Procedure. DAWC effectively combines the required
system parameters with certain weighting factors to elect cluster-heads. Values
of these factors can be chosen according to the application needs. For example
power control is very important in CDMA-based networks. Thus, weight of the
power factor can be made larger. In order to achieve the goal of energy saving,
RCR minimizes the frequency of clusters reformations. It is achieved by encour-
aging the current cluster-heads to remain cluster-heads as long as possible. That
is why we have included the time of being cluster-head in computing weight.
Similarly, if resource-rich devices are deployed to work as cluster-heads then the
weighting factors of distance can be made large and time of being head can be
kept small. The operation of the cluster-head election procedure is outlined as
follows:
Each node i maintains a list of its neighbors. Each entry of the neighbor
list contains node ID and its weight W
i
computed on the basis of the selected
parameters. Once the neighbor list is ready, the cluster-head election procedure
is initiated for the first time. Each node i does the following:
Calculates D
i
as the average distance to its neighbors, M
i
as the total number
of its neighbors, E
i
as its energy and T
i
asthetimebeingheadinthepast.
Computes weight W
i
=(c
1
D
i
+ c
2
E
i
+ c
3
M
i
)/c
4
T
i
,wherethecoecients
c
1
, c
2
, c
3
, c
4
are the weighting factors for the corresponding parameters.
Page 8
Real-Time Coordination and Routing in WSAN 373
Elects the node i as the cluster-head if it has M
i
in the range of its minimum
M
Min
and maximum M
Max
threshold of nodes and has the highest weight
among its neighbors.
Sets its threshold W
Th
=cW
i
, where c is the reduction factor to readjust
the threshold.
The cluster-head keeps computing its weight and when the weight goes down
to its threshold W
Th
, it triggers the cluster-head election procedure.
To save energy, we do not periodically reform clusters. In each round, cluster-
head recomputes its weight and compares with its threshold value. If W
i
of
cluster-head i is higher than its W
Th
value then it keeps functioning as head. if
W
i
<W
Th
then it checks whether its W
i
is also lower than any of its member
node weight. If so, it withdraws itself to function as cluster-head and cluster
election procedure is initiated.
The pseudo-code of the operations executed by a sensor node in each round
of cluster formation is reported in Algorithm 1.
Algorithm 1. Elect Cluster-head
1: Pseudo-code executed by each node N in each round
2: W
max
=0
3: for all iinΛ(n) do
4: if W
max
<W
i
then
5: W
max
= W
i
6: end if
7: end for
8: W
i
= my-weight()
9: if status = NONE then
10: if W
i
>W
max
then
11: announce-head()
12: W
th
= W
i
× c
13: where c is the threshold factor
14: else if status = HEAD then
15: if W
i
<W
th
then
16: if W
i
<W
max
then
17: withdraw-head()
18: else
19: W
th
= W
i
× c
20: end if
21: end if
22: end if
23: end if
If nodes could not join any cluster during the first phase then DAWC accom-
modates these DN nodes as follows. When a high weight node in a group of
dangling nodes have the number of neighbor nodes smaller than the lower limit
M
Min
, it decreases this value locally by one and then retires three times. Each
Page 9
374 G.A. Shah et al.
try is made during the periodic hello beacon. It continues until its M
i
becomes
equal to M
Min
or it joins any cluster. If M
i
reaches to M
Min
then it announces
itself as cluster-head and the other dangling nodes have chance to join this head.
In this way, M
Min
is reduced to manage the sparse zone of sensors. While the
M
Max
is set to disallow the nodes to make the cluster-heads overloaded in dense
zone.
2.3 Neighboring Cluster Discovery
The sink or actors can be multi-hop away from the source clusters, packets are
then forwarded through intermediate clusters. Clusters are linked with each other
to provide multi-hop cluster routing. Some member nodes within a cluster can
hear the members of neighboring clusters or heads, such nodes act as gateways.
It is also possible that there will be multiple gateways between two clusters.
Cluster-heads keep record of all of these gateways.
We build a set of forwarding gateway nodes GS, for each cluster-head, for
routing packets to neighboring clusters. Let SM
i
be the set of members of cluster-
head H
i
and SM
j
be the set of members of neighboring cluster-head H
j
. H
i
maintains a set of gateway nodes GS
i
such that
GSi(Hi)={x SM
i
/H
j
Λ(x) y SM
j
y Λ(x) i = j}
Where Λ(x) is the set of neighbors of node x. A member node x of cluster-head
H
i
belongs to the gateway set GS
i
of head H
i
if either H
j
or some member y
of H
j
exists in the neighbors set of x. The attributes of the elements of GS
i
are
{AdjacentHead, Energy, Delay, Hops}. These attributes help the cluster-heads
in selecting a particular item from the set GS. We will describe the selection
criteria in detail in Section 3.3.
Once the cluster formation is complete, each cluster gets the neighbor clusters
list along with the gateways to reach them. The route computation is discussed
in the next section.
3 Delay-Constrained Energy Aware Routing (DEAR)
The main aim of RCR framework is to provide real-time routing in WSAN with
least energy consumption. RCR achieves this by clustering sensors hierarchically
and then selecting the path on the basis of end-to-end (E2E) deadline (τ)aswell
as balanced energy consumption. We propose a delay-constrained energy aware
routing (DEAR) algorithm to deliver packets from the source clusters to the
target nodes (Sink/Actors) in WSAN. A similar idea of delay-constrained least
cost routing has been proposed in [18], [19]. Unlike these protocols, we have
combined the forward-tracking and back-tracking approach to reduce the cost
of path establishment. We establish a distributed single path, in which cluster-
head selects the outgoing link such that the packet deadline is meet with efficient
energy consumption. An energy efficient link does not merely mean the low cost
link but a link that can satisfy the delay constraint and it balances the energy
consumption on all the outgoing links.
Page 10
Real-Time Coordination and Routing in WSAN 375
3.1 Network Model
Before going into the details of the algorithm, we model the network as a con-
nected directed graph G =(V,E). The set of vertices V represents the sen-
sor nodes, where |V | = n. E is the set of directed edges such that an edge
e(u v) E if (u, v) V . Two non-negative real value functions R(e), the
available energy resource of node v V on the outgoing link e(u v) E,and
(e), the delay experienced by the data packet on the corresponding link, are as-
sociated with the edges. These real values are used to compute the weight W(u,v)
of the link e(u v) E (u, v) V . The weight of an edge e(u v) E can
be defined as follows:
W (u, v)=R(e)/∆(e),whereu,v V
Links are presumably asymmetrical because the R(e) and (e) for the link e(u
v) may not be same while going in the opposite direction of this link e(v v).
The existence of alternative paths between a pair of vertices u, v V provides the
possibility of some paths being shorter than others in terms of their associated
cost. We need to find out a minimum spanning acyclic subgraph of G having
high total weight.
Let s be a source node and d be a destination node, a set of links e
1
=
(s, v
2
),e
2
=(v
2
,v
3
), ..., e
j
=(v
j
,d) constitutes a directed path P(s,d) from s
d. The weight of this path is given as follows:
W [P (s, d)] =
eP (s,d)
W (e)
Likewise, the E2E delay experienced by following the path P(s,d) is measured
as:
[P (s, d)] =
eP (s,d)
(e)
After the formation of clusters, we can have a vertices subset H of the set
V such that the elements in H are only the cluster-heads and has an associ-
ated integral function hops[P (h target)],h H. Similarly, we obtain the set
GS
h
h H as the result of linking the clusters described in Section 2.3. Each
element h of set H maintains a set of outgoing links OUT
h
subset of GS
h
to the
single destination node either sink or actor. In the next section, we describe the
way of building the set OUT
h
h H.
3.2 Sensor-Actor Coordination
The main communication paradigm in WSANs is based on the effective sensor-
actor coordination. Right actions against the detected events cannot be per-
formed unless event information is transmitted from sensors to actors. Therefore,
the ultimate goal of any routing protocol in WSANs is to relay the event read-
ings to the actors within a certain delay limit. In the classical semi-automated
Page 11
376 G.A. Shah et al.
architecture, there is a central node that is responsible to collect the readings
and issue action commands to the actors responsible for the action. Unlike this
approach, automated architecture has also been realized due to the need of im-
mediate action on the phenomena observed in the sensory field. In the former
approach, sink is the destination of events reported by all the sources and is
responsible to coordinate with actors. In the latter case, the mobile actors in
automated architecture are the targets of the event readings observed by the
sensor nodes and, hence, the coordination is local.
In order to compute the delay-constrained paths efficiently, we decompose G
into a minimized acyclic subgraph
¯
G =(
¯
V,
¯
E) constituting a large acyclic region
within G.
¯
V is the set of nodes either in H or belong to the GS sets of cluster-
heads i.e.
¯
V = H GS
1
GS
2
... GS
k
for k number of clusters.
¯
E is the set
of directed edges such that an edge ¯e(u v)
¯
E if u, v
¯
V . The length of an
edge ¯e(u v)
¯
E may be greater than one because the members in GS may
be multi-hop far from heads. For instance, an edge ¯e(u v)
¯
E might exist
due to some member node w such that uwv),w /
¯
V,(u, v)
¯
V . Here,
Re) is the least available energy of any node visited while traversing the link
¯e(u v)ande) is the cumulative delay experienced by the data packet on
the corresponding link.
The decomposed minimized graph
¯
G is the backbone to establish the route
from the source nodes to either the sink (semi-automated architecture)orthe
actor (automated architecture). In the next section, we look into the formation
of the graph
¯
G.
Centralized DEAR (C-DEAR). In this section, we deal with the centralized
semi-automated architecture. The sink node is stationary like sensor nodes and
the path from cluster-heads to sink is built in proactive way. Sink is the destina-
tion for all the source nodes in semi-automated architecture. Source to sink path
is divided into two phases; source to cluster head and cluster-head to sink. The
first phase builds the path from source nodes to cluster-head that is done dur-
ing the cluster formation in a forward tracking manner. The next phase deals
with finding the path from cluster-heads to the sink using backtracking. It is
activated initially by the sink during the network configuration phase and is up-
dated periodically. To achieve this, the algorithm visits the graph G and marks
all the vertices h H. A mark is associated with the life of the node, which is
deleted as that vertice(node) expires. A vertex can be marked if h H has not
been already marked or the current path delay [P (sink, h)] is less than the
previously observed path delay. Once all the elements h of set H are marked, we
build a path P (sink, h) h H in proactive fashion and each element h H
set its hops[P(sink, h)] = |P (sink, h)|.
When h is marked, h adds the incoming link in(x h),x V to the set
OUT
h
in reverse-topological order out(x sink). The incoming link in may be
associated with the last marked element g H in the marking process or null if
h is the first marked item and represents link to the root (sink node). This helps
h to extend the set OUT
h
by using the pre-determined set GS
h
. The attributes
of the elements of GS set contains the AdjacentHead ID that corresponds to g.
Page 12
Real-Time Coordination and Routing in WSAN 377
For each element o(m g) GS
h
, it searches for the match of g with the
attribute AdjacentHead of o. If there exists such element o(m g)thenh adds
the link as o(m g)toOUT
h
and associate an integral value H(o)apartfrom
the other two real value functions R(o)and(o). Hence, the edges set
¯
E of
¯
G
can be obtained as OUT
1
OUT
2
, ..., OUT
k
for k number of clusters. Fig 2
illustrates the decomposed subgraph
¯
G with all the possible links to the sink
node. We use the term link for set
¯
E rather than edge because vertices of set
¯
V
may be connected by some intermediate vertices in V. The set OUT
h
provides
all the possible routes to the sink node and we exploit the multiple entries in
OUT
h
to provide delay-constrained energy aware routes and implicit congestion
control. We describe the criteria of selecting the outgoing link in Section 3.3.
The cost of marking process is O(n) and, in fact, it is the actual cost of building
route from source nodes to the sink node.
Implementation. The marking process is implemented by broadcasting sink
presence beacon in the network. That is, sink initiates the connection with the
sensor nodes by broadcasting its presence beacon periodically, where the length
of period (life of mark) is larger than the hello beacon. This periodic beacon
helps to refresh the path because the topology of the sensor nodes is dynamic.
A receiving node accepts this beacon if it meets one of the following conditions:
1. It has not already received this beacon or beacon has expired.
2. Delay of this beacon is smaller than the last received beacon.
3. The number of hops traversed by this beacon are small.
When a node receives a packet it calculates the delay and forwards the re-
quest in the direction of cluster-head. While the cluster-head forwards it to its
neighboring cluster-head. Hence, each cluster-head learns the loop free path to
thesinknodeandgetsthedelayvalueandnumberofhopssofar.
Distributed DEAR (D-DEAR). The distributed event routing approach
is imperative due to the non-existence of central controller. Events detected
by the sensor nodes are directly routed to the actor nodes without the inter-
vention of the sink node. To provide the distributed routing in the automated
architecture, RCR proposes the distributed flavor of DEAR. In D-DEAR, we
decompose the graph G into the m number of
¯
G subgraphs for each of m mo-
bile actors. The idea is similar to C-DEAR described in detail in Section 3.2
except that we have m possible destinations. The marking process is triggered
independently by all the m actors to construct m number of
¯
G representing the
paths P (h, actor
1
), (h, actor
2
), ..., (h, actor
m
) h H. The cost of D-DEAR is
O(mn).
In order to optimize the sensor-actor coordination in the distributed environ-
ment, the marking process also propagates the current load factor of the actor.
The load represents the number of sources the actor is serving at the moment.
The marking criteria in D-DEAR is modified such that h accepts the mark of an
actor on the basis of its Eculidian distance. The nearest one is the best candidate
for marking the element h of the set H. There might be the possibility that two
Page 13
378 G.A. Shah et al.
or more actors reside at the same distance to h.Insuchcase,load factor breaks
the ties among such candidates and less-loaded actor is the winner.
Actors are location aware mobile nodes. Whenever an actor moves, it triggers
the construction of graph
¯
G in addition to the periodic reconstruction of graphs.
The periodic update of graphs is required due to the highly dynamic topology
of the wireless sensor and actor networks because sensor nodes may be deployed
at any time or their energy deplete. Hence, the algorithm updates the path
proactively to reduce the chances of path failure like the path establishment.
3.3 Alternative Path Selection
Power efficiency has always been an important consideration in sensor networks.
Whereas, E2E deadline τ is another constraint for real-time applications in wire-
less sensor and actor networks. Real-time event delivery is the main aim of our
distributed routing protocol. We have described the process of building the set of
outgoing links OUT . The selection of a particular link o(m g) OUT
h
,g H
by the cluster-head h H is based on the criteria to balance the load in terms of
delay and energy of its member nodes. The operations of the alternative gateway
selection are outlined in the Algorithm 2.
Algorithm 2. Select Outgoing Link
Ensure: Delay-constrained energy aware outgoing link out OUT
h
1: Pseudo-code executed by source cluster-head h to select an outgoing link from the
set OUT
h
.
2: P =
3: for all o(m g) OUT
h
,g H do
4: if time
left
/hops[P (sink, s)] <∆(o) then
5: if R(o) <P then
6: P = R(n)
7: out = o
8: end if
9: end if
10: end for
Cluster-head adds the time
left
field to its data packet that is set to τ by the
source cluster-head. Each intermediate cluster-head looks for this time
left
field
and selects the outgoing link accordingly by executing the above procedure. If
the delay constraint can be meet through multiple links then it selects the one
according to the criteria as described below:
The link, along which the minimum power available (PA) of any node is
larger than the minimum PA of a node in any other links, is preferred”.
Every receiving node then updates the time
left
field as time
left
= time
left
delay
s
. It can be seen that the link selection criteria implicitly eliminates the
congestion by alternating the links towards destination. Whenever a link is con-
gested, the packet delay is increased and this delay is reported to the cluster-head
Page 14
Real-Time Coordination and Routing in WSAN 379
in successive hello beacon. The weight of this link is reduced and, eventually,
the cluster-head reacts to it by selecting the other available link. Hence, the
congestion is avoided in addition to the energy efficiency.
Sink
Marked Vertex (Cluster−head)
Link Edge
Element of GS set (Gateway node)
Fig. 2. Decomposition of graph G into the minimized acyclic subgraph
¯
G within the
region G
4 Performance Evaluation
The performance of RCR is evaluated by using the network simulator ns-2 [20].
The example scenario consists of a sink node, three actors and 100 sensors ran-
domly deployed in 200 × 200 meter square area. Three sensor agents developed by
NRL are also placed to trigger phenomenon at the rate of 2 events per second each.
The main aim of our proposed framework is to deliver the events triggered in
the sensors field to the actors within the given delay bound τ. Fig. 3 illustrates
the average delay against the application deadline for the four different configu-
rations; direct semi-automated, indirect semi-automated, static-actors automated
and mobile-actors automated architecture. The mobility pattern of mobile ac-
tors is random walk. The major factor in missed-deadline is mobility of actors
as clear by the mobile actors graph in Fig. 3.
Deadlines miss-ratio is an important metric in real-time systems. We mea-
sure the miss-ratio for different values of τ . Fig. 4 represents the evaluation of
deadlines miss-ratio for all the configurations. The miss-ratio reaches to 0 when
τ 50ms for direct semi-automated and static-actors automated. This value of
τ can not be set for the other configurations because there is still a significant
miss-ratio.
Besides the mobility of actors, network configuration and events congestion
are also the factors of missed-deadlines. It happens when the neighbor clus-
ter of actor itself has detected events and at the same time the other clusters
Page 15
380 G.A. Shah et al.
10 20 30 40 50 60 70 80 90 100
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Simulation time (s)
Average delay (s)
Average Delay in Semi−automated and Automated architectures
Delay bound
Semi−automated(indirect)
Semi−automated(direct)
Automated(static)
Automated(mobility)
Fig. 3. Average delay in semi-automated architecture vs automated architecture
10 20 30 40 50 60 70 80 90 100 110
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Delay bound (ms)
Packet miss−ratio
Packet miss−ratio in Semi−automated and Automated architectures
Semi−automated(indirect)
Semi−automated(direct)
Automated(static)
Automated(mobility)
Fig. 4. Deadlines Miss-ratio in semi-automated architecture and automated architecture
for τ =10110ms
are also sending their readings for the same actor through this neighbor clus-
ter. It is possible that the sensors start detecting events while the network is
in the configuration state. This scenario not only causes the loss of packets
but missed-deadlines as well. Fig. 5(a) represents the delay graph of semi-
automated architecture. The events start occurring before the network is con-
figured. There are missed-deadlines due to the non-configured network at the
beginning of the graph. The same scenario for the automated architecture is
shown in Fig. 5(b).
Although the missed-deadlines are very small in Fig. 5(b) but the peaks are
initially higher than the peaks in stable configuration and causes source of jitter
Page 16
Real-Time Coordination and Routing in WSAN 381
10 20 30 40 50 60 70 80 90 100
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Simulation time (s)
Packet delay (s)
Delay in Semi−automated architecture
Delay bound
Packet delay
Avergae delay
(a)
0 20 40 60 80 100
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Simulation time (s)
Packet delay (s)
Delay in Automated architecture
Delay bound
Packet delay
Avergae delay
(b)
Fig. 5. Event to actor routing starting before configuration of 100 sensor and 3 actor
nodes (τ =0.05sec)
50 100 150 200 250 300
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Number of Sensor Nodes
Average Delay(sec)
Average Delay in Semi−automated vs Automated Architecture
Semi−automated(Direct)
Automated
Semi−automated(Indirect)
Fig. 6. Average delay by increasing number of nodes
in the traffic. The automated architecture requires lesser configuration time
3
than the semi-automated architecture and therefore, the packet delays are not
much affected during the network configuration state. In Fig. 5(b), actors start
receiving events at about 6 sec. This value is 12 sec. in the semi-automated
architecture.
The scalability of the proposed framework is evaluated by increasing the num-
ber of sensor nodes in the field. Fig. 6 shows the results up to 300 nodes. The
average packet delay in indirect semi-automated architecture increases signifi-
cantly with the number of nodes as compared to the other configurations. The
other two configurations (direct semi-automated architecture and static-actors
automated architecture) are not much affected by deploying more sensors. The
3
Configuration time is the time to form clusters, link them and to find out the actors
for sending event readings. The network is stable after the configuration.
Page 17
382 G.A. Shah et al.
reason is apparent because the sink node is the only node that provides coordi-
nation among clusters and actors in indirect configuration. Whereas, the other
two configurations provide direct coordination among sensors and actors.
5Conclusion
There have been a number of routing protocols developed for WSN that claim
to provide real-time routing and congestion control. However, none of them has
considered the presence of actors. RCR addresses the issues in such heteroge-
neous network and provides a coordination framework for wireless sensor and
actor networks. It clusters the sensors and selects a head in each cluster to min-
imize the energy consumption, estimates the delay budget and makes routing
decisions. The communication framework works fine in semi-automated architec-
ture as well as automated architecture. Simulation results show that RCR meets
the E2E deadlines for real-time applications with a small value of miss-ratio.
References
1. I. F. Akyildiz and I. H. Kasimoglu, “Wireless Sensor and Actor Networks: Research
Challenges,” Ad Hoc Networks, Vol. 2, Issue 4, pp. 351-367, October 2004.
2. T. Melodia, D. Pompili, V. C. Gungor and I. F. Akyildiz, “A Distributed Coor-
dination Framework for Wireless Sensor and Actor Networks,” ACM MobiHoc’05,
May 2005.
3. K. Akkaya, M. Younis, “An Energy-Aware QoS Routing Protocol for Wireless
Sensor Networks,” 23rd ICDCSW, pp. 710-715, 2003.
4. S. Basagni, “Distributed Clustering for Ad Hoc Networks,” IEEE Proc. Vehicular
Technology Conference, 1999.
5. D. Tian and N.D. Georganas, “Energy Efficient Routing with Guaranteed Delivery
in Wireless Sensor Networks,” Mobile Computing and Communications Review,
2001.
6. Y. Yu, R. Govindan and D. Estrin, “Geographical and Energy Aware Rout-
ing (GEAR): a recursive dissemination protocol for wireless sensor networks,”
UCLA/CSD-TR-01-0023, Tech. Rep., 2001.
7. M. J. Handy, M. Haase and D. Timmermann, “Low Energy Adaptive Clustering
Hierarchy with Deterministic Cluster-Head Selection,” IEEE International Con-
ference on Mobile and Wireless, 2002.
8. M. Chatterjee, S.K. Das and D. Turgut, “A Weight Based Distributed Clustering
Algorithm for Mobile Ad hoc Networks,” HiPC 2000: pp. 511-521.
9. A.D. Amis, R. Prakash, T.H.P. Vuong and D.T. Huynh, “Max-Min-D-Cluster For-
mation in Wireless Ad hoc Networks,” Proc. IEEE Infocom 2000, Tel Aviv, Israel,
March 2000.
10. C. Intanagonwiwat, R. Govindan and D. Estrin, “Directed Diffusion: A Scalable
and Robust Communication Paradigm for Sensor Networks,” Proc. ACM Mobi-
Com, March 2000.
11. T. He, J.A. Stankovic, C. Lu and T. Abdelzaher, “A Real-Time Routing Protocol
forSensorNetworks,Proc. IEEE International Conference on Distributed Com-
puting Systems, May 2003.
Page 18
Real-Time Coordination and Routing in WSAN 383
12. V. Rodoplu and T.H. meng, “Minimum Energy Mobile Wireless Networks,” IEEE
JSAC, August 1999.
13. W. R. Heinzelman, A. Chandrakasan and H. Balakrishnan, “Energy Efficient Com-
munication Protocol for Wireless Microsensor Networks,” IEEE Proc. Hawaii In-
ternational Conf. Sys. Sci. January 2000.
14. W. R. Heinzelman, J. Kulik and H. Balakrishnan, “Adaptive Protocol for Infor-
mation Dissemination in Wireless Sensor Networks,” Proc. ACM MobiCom, 1999.
15. W. R. Heinzelman, A. Chandrakasan and H. Balakrishnan, “An application-specific
protocol architecture for wireless microsensor networks,” IEEE Transactions on
Wireless Communications, vol. 1, no. 4, pp. 660-670, October 2002.
16. G. Smaragdakis, I. Matta and A. Bestavros “SEP: A Stable Election Protocol for
clustered hetrogenous wireless sensor networks,” 2nd International Workshop on
Sensor and Actor Network Protocols and Applications (SANPA 2004).
17. V. Mhatre and C. Rosenberg, “Homogeneous vs. heterogeneous clustered sensor
networks: A comparative study,” In Proc. of 2004 IEEE International Conference
on Communications (ICC 2004), June 2004.
18. Q. Sun and H. Langenorfer, “A new distributed routing algorithm for supporting
delay-sensitive applications,” Journal of Computer Communications vol. 9 no. 6
May 1998.
19. R. Gawlick, A Kamath, S. Poltkin and K. G. ramkrishnan “Routing and admis-
sion control in general topology networks,” Tech. Report STAN-CS-TR-95-1548,
Stanford University, CA, 1995.
20. UC Berkeley, LBL, USC/ISI and Xerox PARC, “The Network Simulator ns-2,”
The VINT Project, http://www.isi.edu/nsnam/ns/.
Page 19
  • Source
    • "Hence, different optimization mechanisms can be used for different scenarios. As an example, authors in [42] propose an optimization protocol to obtain the optimum number of members per cluster in WSNs, that is based on minimizing energy dissipation. In mobile networks, the cluster size is continuously varying and the optimization is not straight-forward. "
    [Show abstract] [Hide abstract] ABSTRACT: Clustering became relevant in the past as a solution for the scalability problems of ad hoc networking, but, the unsuccessful application of ad hoc solutions to real scenarios, such as the projects SURAN and PRNet, decreased the interest of research community on ad hoc communications, and subsequently, on clustering algorithms. Recently, however, clustering techniques have gained renewed interest due to the emergence of cooperative communications for cellular networking. Clustering is envisaged, in this scenario, as a technique to team up nodes to support efficient data aggregation for energy saving, scalability and privacy among other benefits. Moreover, research on 5G networks also envisages a connected society, where everything and everyone will be connected under the umbrella of Internet of Everything (IoE). This novel communication paradigm has fostered new research on clustering, which has yielded novel and more advanced algorithms and applications. This article surveys the State-of-the-Art in clustering techniques and provides detailed descriptions of the basics of clustering and the latest novel ideas. Open issues, technical challenges and directions for future research are also outlined.
    Full-text · Article · May 2016
  • Source
    • "Identifying the actors that ought to handle an event is a key function of a WSAN. Such a designation can be simply based on the actor's proximity to the event [[8], by planning [9], or through inter-actor coordination [10][11] . Proximitybased actor tasking suits discrete events and can be the byproduct of sensor clustering. "
    [Show abstract] [Hide abstract] ABSTRACT: Wireless sensors and actors networks (WSANs) have the capacity for not only monitoring some phenomena through sensor nodes but also performing appropriate actions. Most of the contemporary WSAN management solutions focus on defining communication path among sensors and actors and on tasking appropriate actors to handle the detected events. In this paper we classify events based on how they evolve over time into continuous and discrete and categorize the WSAN management strategies accordingly. Unlike discrete events, a continuous event spreads quickly and becomes more serious as time passes. Such a characteristic introduces more challenges and motivates a non-conventional management strategies. This paper presents an approach for Sensor-Actuator Coordination for Handling Spreading events (SACHS). SACHS opts to enable the network to respond quickly in order to avoid the event from growing in scope, e.g., prevent a fire from spreading, while reducing the energy overhead due to the coordination messages and due to actor's relocation to the event region. SACHS limits sensor-actor and actor-actor interactions and exploits local sensor-sensor communication to determine the scope of the event, define spots for actors to position at, and schedule the actors' response. The simulation results confirm the performance advantage of SACHS compared to competing schemes.
    Full-text · Conference Paper · Aug 2014
    • "Melodia et al. [19] use an event-driven clustering paradigm to design a sensor–actor coordination model and formulate the actor coordination as a task assignment optimization problem. The real-time routing framework by Shah et al. [20] addresses the coordination of sensor and actor nodes through the delay bound for distributed routing. Another coordination algorithm among actors is introduced with the real-time communication framework by Ngai et al. [21], where an event reporting algorithm for sensor–actor communication is also given to minimize the transmission delay. "
    [Show abstract] [Hide abstract] ABSTRACT: Wireless sensor and actor networks (WSANs) have been increasingly popular for environmental monitoring applications in the last decade. While the deployment of sensor nodes enables a fine granularity of data collection, resource-rich actor nodes provide further evaluation of the information and reaction. Quality of service (QoS) and routing solutions for WSANs are challenging compared to traditional networks because of the limited node resources. WSANs also have different QoS requirements than wireless sensor networks (WSNs) since actors and sensor nodes have distinct resource constraints. In this paper, we present, LRP-QS, a lightweight routing protocol with dynamic interests and QoS support for WSANs. LRP-QS provides QoS by differentiating the rates among different types of interests with dynamic packet tagging at sensor nodes and per flow management at actor nodes. The interests, which define the types of events to observe, are distributed in the network. The weights of the interests are determined dynamically by using a nonsensitive ranking algorithm depending on the variation in the observed values of data collected in response to interests. Our simulation studies show that the proposed protocol provides a higher packet delivery ratio and a lower memory consumption than the existing state of the art protocols.
    No preview · Article · Nov 2013 · Ad Hoc Networks
Show more