Wireless Sensor Networks (WSNs) promise an unprecedented fine-grained
interface between the virtual and physical world (Akyildiz et al.,
2007). They are one of the most rapidly developing new information technologies,
with potential applications in a wide range of fields including industrial
process control, security and surveillance, environmental sensing and
structural health monitoring, etc. Although having wide applications,
WSNs only realize the functions of data collection and transmission, which
still are open-loop from viewpoint of control system designers. The wireless
sensor/actor networks (WSANs), which is an entirely new technology, has
been sprung up for the last few years based on WSNs (Akyildiz and Kasimoglu,
2004). WSANs are capable of not only observing the physical world, processing
the data, but also making decisions based on the observations and performing
appropriate actions. In WSANs, actors take actions onto the observed object
based on the sensory data, by which way to construct the real closed-loop
feedback control and automatic systems. This network can be an integral
part of systems such as battle field surveillance and microclimate control
in buildings, nuclear, biological and chemical attack detection, home
automation and environmental monitoring (Chen et al., 2007).
As we know, efficient bandwidth utilization and quality of service are mostly
considered in traditional wireless network design. In WSAN however, sensor nodes
are deployed in the region of interest, where it is impractical or infeasible
for humans to interact with or monitor them, thus unattended. In this case,
how to maximize the network lifetime with constrained resource becomes complex
and crucial. Besides, in some scenarios such as fire suppression, the communication
traffic is typically delay-sensitive. Therefore, supporting a real-time communication
is critical in the application of WSANs. There are mainly two typical kinds
of physical system architecture of WSANs (Chaudhry et al., 2006), namely,
autonomic architecture (AA) and centralized architecture (CA), which will also
be described further. Our goal is mainly to design a routing for WSANs that
contributes to transfer sensory data to the distributed actors, solving the
problem which sensors communicate with which actors? and comparing the performance
of AA with that of CA under the same circumstances. Apparently, applying the
system architecture with better performance such as longer network lifetime
and smaller communication delay to industry will be more productive with the
WSANs characterize rigorous real-time requirement of coordination and
communication. Although, many protocols and algorithms have been proposed
for WSNs in recent years, they may not be well-suited for WSANs because
of the different requirement of WSANs. Many recently developed protocols
for WSANs are also under primary steps. For example, in Conti et al.
(2004), only actor-actor coordination is handled without any insight into
the sensor-actor coordination problem. Some recent studies (He et al.,
2003; Felemban et al., 2005) have considered the issue of real-time
communication in sensor networks. A protocol is also introduced in (Krotkov
and Blitch, 2004) where it is assumed that sensor and actor nodes are
of same type which obviously does not reflect the actual WSANs. In Vedantham
et al. (2006), the problem of hazards is considered, which consist
of the out-of-order execution of queries and commands resulting from a
lack of coordination between sensors and actors. However, coordination
problems in sensor-actor or in actor-actor communications are not considered
in the study.
Introducing distributed multi-actor to WSNs causes big challenges in
designing a new communication paradigm, much different from the centralized
architecture in WSNs. This research will deal with the principle problems
about sensor-actor communication in the WSANs and evaluate the proposed
routing by extensive simulation experiments.
SYSTEM MODEL OF WSAN
In Akyildiz and Kasimoglu (2004) sensor nodes collect data from the environment
while actors perform appropriate actions based on this data, respectively.
Sensor nodes detecting a phenomenon can transmit their readings to the
actor nodes which take appropriate actions, or route data back to the
sink which may control actors. The former case is named as AA due to the
nonexistence of central controller while the latter case is defined as
CA since the sink (central controller) collects data and coordinates the
acting process. These two architectures are given in Fig.
|| Two architectures for WSAN
The advantage of CA attributes to its topology, which is similar to the
architecture of the well-known WSNs. The community of WSNs researchers
has proposed, implemented and measured a variety of routing algorithms
for such networks which could be directly applied for CA form of WSANs.
In contrast, AA naturally decomposes a large network into separate zones
within which data processing and aggregation can be performed locally.
Because WSANs have some particular characteristics due to the coexistence
of sensors and actors, there are many challenges have to be investigated.
The principal problem, that transfers sensory data from sensors to actors,
should be solved. A suitable and effective routing should be designed
for WSANs. Further a new routing protocol is developed, which is well-suited
for AA as well as CA.
GREEDY RUMOR FORWARDING ROUTING PROTOCOL FOR WSAN
Depending on the application, the routing protocol suitable for WSANs
should be a kind of event driven and geography based routing protocol.
It is shown that geographic routing allows routers to be nearly stateless
and requires propagation of topology information for only a single hop:
each node needs only the knowledge of its neighbors` locations. The location
of a packet`s destination and locations of the next hop candidates are
information sufficient to make correct forwarding decisions without any
other topological information.
Here, GRFR protocol is proposed, an improved rumor routing protocol,
for routing sensory data from any sensor to its destination (actor) in
WSANs. Rumor routing protocol is a kind of query mechanism (Braginsky
and Estrin, 2002). The query nodes have to diffuse queries to the whole
network, which consumes too much energy. GRFR improves rumor routing protocol
to a kind of event driven and geographic routing protocol, as mentioned
above, which is well-suited for both AA and CA. GRFR mainly includes three
parts as discussed below.
Selection of the source sensor and destination actor: In WSANs,
it is assumed that distributed actors can self-localized and broadcast
their position over the network. Even if the actors are mobile, their
location will be broadcasted after they reach a new position. Each sensor
will maintain a location table for the actors.
To make it simple, we assumed the event is a vertex with circular affecting
range in the whole network topology. The network will elect source node
and destination node when an event appears. An event triggers the sensors
within its influencing scope, which then communicate mutually to elect one source node.
||Example of election of source and destination nodes
Normally, the nearest
and available sensor to the event will be elected to the source node.
However, the election of destination node in AA differs from the one in
CA, which attributes to the coexistence of sensors and actors. The nearest
actor in the actor list in the table to the event will be elected to the
destination node in AA. In contrast, the sink is the destination node
all the time in CA. So in AA, the routing is dynamic and can not be maintained
by routing table in each sensor, especially with mobile actors. But CA
can construct a relatively static routing path comparing with AA. That`s
why traditional research work on WSNs should be improved.
An example of election of the source and destination nodes in AA is shown
in Fig. 2. When an event happens, the sensor nodes within
its influencing scope all detect the event and they communicate with each
other to elect Node 0 as the source node, since Node 0 is the closest
sensor node to the event vertex. Then Node 0 will find the nearest actor
in its actor table as the destination node because the actors` locations
are known to all the sensor nodes by initially broadcasting over the network.
Finally, the nearest actor to the event vertex, actor node 4, will be
chosen as the destination node.
As introduced above, there is only one node elected as the source node
among all the nodes detected the event information. That is because if
the nodes within the influencing scope all relay the packets, it is likely
that more than one actor node being triggered and the actor nodes must
communicate and coordinate with each other and the actor-actor coordination
is beyond the research domain of this paper. Therefore, in order to simplify
the problem, without loss of generality, we assume only one node sensed
the phenomena and confine the election of source nodes to the above condition.
||Example of greedy rumor forwarding routing
Routing algorithm: In order to transmit the packet from source
sensor node to destination node, an event driven and geographic routing
algorithm is required. Therefore, GRFR algorithm is proposed, with which
packets are marked by their originators with their locations of destinations.
As a result, a forwarding node can make a greedy choice in choosing a
packet`s next hop. We assume in this work that all wireless sensor nodes
know their neighbors. We consider topologies where the wireless nodes
are roughly in a plane. Specifically, the greedy choice of next hop is
geographically closer to the packet`s destination than the current node.
Furthermore, what rumor means is that if more than one nodes accord with
the above condition, there will be a random one to forward the packet
further, i.e., they have the equal priorities. Forwarding in this regime
follows successively closer geographic hops, until the destination is
An example of next hop choice in GRFR algorithm is given in Fig.
3. Here, Node 0 receives a packet destined for D. Node 0`s radio range
is denoted by the dotted circle about Node 0 and the arc with radius equal
to the distance between Node 0 and D is shown as the dashed arc about
D. From Node 1 to Node 4 all have equal priorities to be the next hop.
This GRFR process repeats, until the packet reaches D.
TTL information in GRFR protocol: TTL, abbreviating for Time to
Live, a field in the Internet Protocol (IP) that specifies how many hops
a packet can travel before being discarded or returned. TTL contained
in every packet is another important attribute of GRFR. The TTL, a constant,
will be added in every packet when it generates and is decremented with
transmitting. The packet is forwarded if the TTL is greater than zero
and the sensor nodes will discard a packet whose TTL is zero. The packets
without TTL may cause routing loops in the network.
Simulation environment: The performance of GRFR is evaluated in
the two system architectures based on OMNeT++ (http://www.omnet.org),
objective Modular Network Test-bed in C++, which is a public-source, component-based,
modular and open-architecture simulation environment with strong GUI support
and an embeddable simulation kernel. Its primary application area is the
simulation of communication networks, but because of its generic and flexible
architecture, it has been successfully used in other areas. Recently,
many discussion groups propose electing OMNeT++ to the uniform tool for
the algorithm design and analysis in WSNs. So far, there are two simulation
frameworks, EYES Simulation Framework (http://wwwes.cs.utwente.nl/ewsnsim/)
and SensorSimulator (Iaquinto et al., 2008), built over OMNeT++.
We use the software OMNeT++ and SensorSimulator to investigate the quality
of service of GRFR under the structure of both AA and CA.
In the simulation experiments where we compare the performance of AA
with the one of CA, simulation parameters is identical to a subset of
those used in Mica2, which is a third generation mote module used for
enabling low-power, wireless, sensor networks (http://www.xbow.com).
The concrete simulation topological parameters are shown in Table
1. However, they only differ from the existence of a sink, which characterizes
the CA of WSANs.
Performance evaluations: As presented above, simulation experiments
are designed to evaluate the performance of GRFR with AA and CA. In this
study, quality of service of GRFR for WSANs is investigated including
two main parts: network lifetime and communication delay. Furthermore,
the paper also designs an experiment to analyze how the sensor nodes within
one hop from the sink affect the network lifetime.
Network lifetime of AA vs. CA: The sensor node exhausting its
own energy is always considered as node failure. Incidentally, in the
simulation, when the remainder energy of a sensor node is less than or
equal to 0.001, it is thought to be failure. The network lifetime is defined
as either the time of the first node (Chang and Tassiulas, 2004; Madan
and Lall, 2006; Seo et al., 2007; Kalpakis et al., 2003)
or a certain percentage nodes failure (Duarte-Melo et al., 2005;
Cheng et al., 2008). The former is pessimistic because the remainder
nodes can also fulfill appropriate tasks and the latter is flexible because
we can choose the percentage according as the applications.
|| Detailed simulation setting of AA/CA
||Failure time of partial nodes in AA vs CA adopting GRFR
In our simulation, the positions of the sensor nodes are stationary that
do not move anymore after once deployed and their radio radius are constant.
Thus, the network cannot work, as there are only half sensor nodes alive.
Our simulations are for networks of 100 nodes, so the failure time of
12510203040 and 50 nodes is recorded and compared in AA with CA and Fig.
4 shows the results.
Apparently, as shown in Fig. 4, the failure time points
at 50 nodes in AA are all greater than that of CA. Thus, we can conclude
that the AA have longer network lifetime than the CA with GRFR protocol
under the same given circumstances. Specially, the maximum difference
of the networks lifetime with AA and CA appears at the time when 10 nodes
fail. As a matter of fact, our simulated networks are quite dense and
there is an average of approximately 10 neighbors within range of the
average node in these networks. The relationship will be particularly
analyzed as below:
From the different physical architectures of both AA and CA, it is obvious
that the lifetime of the sensor nodes which are within one hop from the
destination nodes will affect the network lifetime. We design an experiment
to illustrate the phenomenon.
In the initialized files, we assign each sensor in the network a unique
identifier from 0 to 99. Node 99 is the sink in CA and from Node 95 to
Node 99 are 5 actors in AA. Due to the random-seed are both equal to 1,
the deployments of the two architectures are identical. In other words,
the nodes with the same identifiers have the same coordinates. Hence,
the same nodes have the same neighbors in the two architectures.
|| Neighbor lists of destination nodes
|| Former 20/10 Failure NodeID in WSANs with AA/CA
The neighbor lists of Node 95 to
Node 99 are shown in Table 2 and 3
lists the former 20/10 failure node identifiers in AA and CA, respectively.
Comparing Table 2 with Table 3, it
can be found that there are 9 neighbor nodes of the sink in the former
10 failure nodes in CA and the former 20 failure nodes are all the neighbor
nodes of the 5 actors in AA.
From the high percentage, it can be concluded that the lifetime of nodes
within one hop from the destination nodes is shorter than that of the
other nodes in the network, which will be expound as follows. In CA, wherever
the event occurs, event information always passes through the sensor nodes
within one hop from the sink. Thus, those sensor nodes have excessive
burden of relaying. When these nodes fail, the connectivity can be lost
and the network will be out of work and become useless. However, in AA,
it is much more likely that for each event different actors may be triggered.
This implies that the relay load gets evenly distributed between all nodes.
As presented in the simulation results, there is an average of approximately
10 neighbors within range of the average node in the networks and the
nodes within one hop from the sink nodes may fail faster than other nodes.
Thus, the time of 10 nodes failure may be considered as the lifetime of
CA. Where, the relaying sensor nodes in AA are also different for each
event because of the existence of 5 actors, we can consider the time of
50 nodes failure as the lifetime of network with AA. As shown in Fig.
4, the difference of the lifetime of network between AA and CA is
obvious. The above data manifests the superior of AA over CA in WSANs.
Analysis of delay in WSANs: Depending on the application there
may be a need to rapidly respond to accident event. Thus, it may request
a low communication delay in the data transmission and we also design
an experiment that manifests the superiority of AA over CA, which will
be interpreted below.
As presented in the introduction, WSANs have two rigorous requirements:
real-time and coordination.
||Failure time of partial nodes in AA vs CA adopting GRFR
We discuss the sensor-actor coordination in the protocol design. Then, the
real-time characteristic is investigated. The most important characteristic
of sensor-actor communication is to provide low communication delay due
to the proximity of sensors and actors. In some applications such as in
fire, the communication traffic is typically delay-sensitive. Therefore,
supporting a real-time traffic is more crucial in the application of WSANs.
Above all, an experiment is designed to compare the communication latency
between AA and CA in WSANs. In the simulation, we gain the hops of packet
transmission from source nodes to destination nodes in every event transmission
and we choose 20 pairs uniformly, as shown in Fig. 5.
As in Fig. 5, the hops of packet transmission in AA
are almost all smaller than the ones in CA. That is because the sensed
information is conveyed from sensors to actors in AA, since they may be
close to each other as shown in Fig. 2. As a result,
the latency is minimized in AA. Moreover, in AA since event information
is transmitted locally through sensor nodes around the event area, sensors
that are far from the event area do not function as relaying nodes, which
savings network resources in AA.
In this study, GRFR protocol for WSANs is proposed and presented in detail,
which solves the key problem of data transmission from sensors to actors
in distributed environment with multi-actors. The benefits of GRFR partially
stem from geographic routing algorithm using only immediate-neighbor information
in forwarding decisions. Then, the performances of WSANs with two architectures,
i.e., CA and AA, are evaluated using the software SensorSimulator, which
is a discrete event simulation framework for sensor networks built over
OMNeT++. The simulation results obviously manifest the superiority of
AA over CA in terms of utilizing distributed sensor nodes for information
sensing, transmission and action making. Furthermore, this paper analyzes
how the sensor nodes within one hop from the sink affect the network lifetime.
It is concluded from experiment results that in CA, the nodes near the
sink have a higher load of relaying packets, which makes the network`s
lifetime shorter. In contrast, the relay load are distributed among all
nodes in AA, though there exits the same problem with the nodes around
the actors. As sensors and actors go close to each other, the communication
latency may become much lower in AA.
In our future research, more attention will be paid on actor/actor coordination
before any specific action is to be taken. Controller design for actors
will also be our important research area.
This research is supported by Joint Funds of NSFC-Guangdong under grants
U0735003; Natural Science Foundation of China under grants No. 60604029,
60702081; Natural Science Foundation of Zhejiang Province under grants
No. Y106384; the Science and Technology Project of Zhejiang Province under
grants No. 2007C31038 and the Scientific Research Fund of Zhejiang Provincial
Education Department under grants No. 20061345.