HAL Id: inria-00431252
https://hal.inria.fr/inria-00431252
Submitted on 11 Nov 2009
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entic research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diusion de documents
scientiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Enhancing Real-Time Delivery in Wireless Sensor
Networks With Two-Hop Information
Yanjun Li, Chung Shue Chen, Ye-Qiong Song, Zhi Wang, Youxian Sun
To cite this version:
Yanjun Li, Chung Shue Chen, Ye-Qiong Song, Zhi Wang, Youxian Sun. Enhancing Real-Time
Delivery in Wireless Sensor Networks With Two-Hop Information. IEEE Transactions on In-
dustrial Informatics, Institute of Electrical and Electronics Engineers, 2009, 5 (2), pp.113-122.
�10.1109/TII.2009.2017938�. �inria-00431252�
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 5, NO. 2, MAY 2009 113
Enhancing Real-Time Delivery in Wireless Sensor
Networks With Two-Hop Information
Yanjun Li, Chung Shue Chen, Member, IEEE, Ye-Qiong Song, Zhi Wang, Member, IEEE, and Youxian Sun
Abstract—A two-hop neighborhood information-based routing
protocol is proposed for real-time wireless sensor networks. The
approach of mapping packet deadline to a velocity is adopted
as that in SPEED; however, our routing decision is made based
on the novel two-hop velocity integrated with energy balancing
mechanism. Initiative drop control is embedded to enhance energy
utilization efficiency, while reducing packet deadline miss ratio.
Simulation and comparison show that the new protocol has led
to lower packet deadline miss ratio and higher energy efficiency
than two existing popular schemes. The result has also indicated a
promising direction in supporting real-time quality-of-service for
wireless sensor networks.
Index Terms—Deadline miss ratio, energy utilization efficiency,
quality-of-service (QoS), real-time, two-hop information, wireless
sensor networks (WSNs).
I. INTRODUCTION
W
IRELESS SENSOR NETWORKS (WSNs) have re-
cently received increasing attention in the industrial
communication community. A vivid vision of WSN can be
described by the concept of “smart dust” [1]: small and cheap
sensor nodes are embedded to sense the surroundings, com-
municate wirelessly, perform collaborative signal processing
and make the environment intelligent. With WSN, it is possible
to collect more real-time data than before, from places which
are hazardous or inaccessible by wired technology. WSN can
be used in many ways in industrial and factory automation
[2]. For example, vibration, pressure or thermal sensors can be
equipped to rotating machinery or conveyer belts to monitor
their health. This helps to detect possible system failure and to
trigger a preventive maintenance routine before a more costly
repair is needed. WSNs are also useful for tracking leakage
Manuscript received September 29, 2008; revised January 11, 2009 and
March 03, 2009. First published April 28, 2009; current version published
May 06, 2009. This work was supported by the National Natural Science
Foundation of China under Grant 60773181 and Grant 60873223, and in part
by the National High-Tech Research and Development Plan of China under
Grant 2006AA01Z218. Paper no. TII-08-09-0115.R2.
Y. Li is with the State Key Laboratory of Industrial Control Technology,
Hangzhou 310027, China and LORIA-Nancy University, TRIO, Vandoeuvre-
lès-Nancy, France (e-mail: yjli.iipc@gmail.com).
C. S. Chen was with the Department of Probability and Stochastic Networks
(PNA2), Centrum Wiskunde and Informatica, 1098 XG Amsterdam, The
Netherlands (e-mail: cschen@ieee.org). This work was carried out during the
tenure of an ERCIM “Alain Bensoussan” Fellowship Program.
Y.-Q. Song is with LORIA-Nancy University, TRIO, Vandoeuvre-lès-Nancy,
France (e-mail: song@loria.fr).
Z. Wang and Y. Sun are with the State Key Laboratory of Industrial Con-
trol Technology, Hangzhou 310027, China (e-mail: wangzhi@iipc.zju.edu.cn;
yxsun@iipc.zju.edu.cn).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TII.2009.2017938
or radiation in chemical plants. Different from some existing
best-effort services which may not have stringent packet time-
liness requirement and can tolerate a significant amount of
packet loss, these real-time industrial applications are much
more demanding [3]. Out-of-date data are usually irrelevant
and may even lead to negative effects to the system monitoring
and control.
In industrial WSNs, traffic is dominated by readings and
commands exchanged between sensors/actuators and control
units. Providing quality-of-service (QoS) in such a scenario
is to enable transmissions of periodic or sporadic messages
within predefined deadlines in a reliable fashion; timeliness
is especially important for crucial alarm messages. Since the
wireless channel is random and time-varying, conventional
deterministic QoS measures should be replaced by probabilistic
ones. An important performance measure is the deadline miss
ratio (DMR) which is defined as the ratio of messages that
cannot meet deadlines [4]. Moreover, sensor nodes usually
use battery for energy supply. Hence, energy efficiency is also
an important design goal. It is usually defined by the energy
consumed per successfully transmitted packet. Furthermore,
in order to avoid network topology holes and achieve a longer
network lifetime, node load and energy balance need to be
considered.
Generally speaking, supporting real-time QoS in WSN can
be addressed from different layers and mechanisms [5]. For ex-
ample, medium access control (MAC) can offer channel access
(one-hop) delay guarantee, while routing protocol in the net-
work layer can support multihop QoS. Transmission scheduling
can be used to provide conflict-free channel sharing based on
regular network topology (e.g., tree, hexagonal layout, etc.) with
techniques of topology control and clock synchronization [6].
Deterministic service delay bound for real-time applications is
expected. In-network data aggregation is known as a good com-
plement to routing protocols in reducing data redundancy and
alleviating network congestion. Cross-layer optimization can
provide further improvement. Among the above, without loss
of generality, routing protocol has always played a crucial role
in supporting end-to-end QoS. Here, we will focus on this do-
main and the design in this paper is oriented to more demanding
applications which emphasize packet delivery timeliness and
end-to-end QoS, e.g., alarm messages should be transmitted
from sensor nodes to control center in time so as to take prompt
actions. Energy efficiency and load balance are also among the
design goals.
It is known from the literature [3] that for system simplicity
most existing routing protocols are based on one-hop neigh-
borhood information. However, it is expected that multihop
information can lead to better performance in many issues
1551-3203/$25.00 © 2009 IEEE
Authorized licensed use limited to: Zhejiang University. Downloaded on June 10, 2009 at 10:19 from IEEE Xplore. Restrictions apply.
114 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 5, NO. 2, MAY 2009
including routing, message broadcasting, and channel access
scheduling [7]–[10]. For computing two-hop neighborhood
information in wireless ad hoc and sensor networks, some dis-
tributed algorithms and efficient information exchange schemes
are reported in [10] and [11]. In a network of
nodes, com-
puting one-hop neighbors with
messages is trivial while
computing two-hop neighbors seems to increase the complexity
and overheads. However, a complexity analysis reported in
[11] has shown that every node can obtain the knowledge of
two-hop neighborhood by a total of
messages, each of
bits, which could be enough to address the ID and
geographic position of nodes.
It is very likely that a system can perform better if more
information is available and effectively utilized. By the study
of asymptotic performance of a generic routing with multihop
routing information [7], it is observed that the number of hops
required from the source to sink decreases significantly from
one-hop to two-hop information-based routing. However, the
further gain from two-hop-based decision to three-hop-based
decision is less attractive, especially if complexity increase is
also a concern. In this paper, we propose a two-hop informa-
tion-based real-time routing protocol and show its improvement
over one-hop-based protocol SPEED [12]. The choice of two
hops is a tradeoff between performance improvement and the
complexity cost. The idea of two-hop routing is straightforward
but how to use or integrate the information effectively so as
to improve energy and real-time performance is generally non-
trivial. The resulting design has the following novel features.
1) Compared with existing protocols that utilize only one-hop
neighborhood information, it achieves lower deadline miss
ratio and also higher energy efficiency.
2) A mechanism is embedded which can release nodes
that are frequently chosen as packet forwarder. An im-
provement of energy balance throughout the network is
achieved.
3) The simulation is built on Mica2-based [13] lossy link
model, energy model and CSMA/CA MAC setting (sim-
ilar to B-MAC [14]) which are very close to real systems.
The rest of this paper is organized as follows. Section II dis-
cusses related routing protocols for real-time QoS in WSN and
explains the motivations. Section III presents our design. The
performance of proposed protocol is reported in Section IV.
Simulations and comparisons have shown its effectiveness.
In Section V, we discuss possible enhancement. Finally,
Section VI concludes this paper.
II. R
EAL-TIME ROUTING PROTOCOLS FOR
WSN
Generally speaking, there are three classes of routing poli-
cies that favor end-to-end delay performance guarantee in WSN:
(i) tree-based routing; (ii) optimal routing based on shortest-
path-first (SPF) principle by the knowledge of whole network
topology; and (iii) geographic routing by the knowledge of node
position.
Tree-based routing is popular in industrial WSN setting.
ZigBee has provided a hierarchical tree routing scheme in
which packets travel along the edges of the tree network. This
approach suits the many-to-one traffic model and does not need
routing table. End-to-end QoS (delay, energy consumption,
etc.) can be estimated by the depth of the tree. However, the
hierarchical tree routing can be very inefficient when two
nodes in different branches but mutual radio range want to
communicate with each other since packets must travel through
the ZigBee coordinators. Ad hoc on-demand distance vector
(AODV) routing is thus suggested as a supplement in this case.
As proposed in [15], another solution is to look up the neighbor
table in routing decisions so as to avoid long path and thus
shorten the worst-case delay. Another drawback of tree-based
routing is the problem of node energy consumption balancing.
Nodes near the root of the tree will consume much more energy
than the other and consequently lead to network topology holes.
A tree routing protocol is often not optimal as it does not
choose the shortest path. AODV is one of the optimal routing-
based protocols by SPF principle. However, additional over-
head (e.g., extra packet and energy consumption) will be intro-
duced in order to maintain the routing table. AODV is a reac-
tive routing which is more favorable when communication is re-
quired infrequently. The route discovery on demand adds addi-
tional latency to packet transmission. This has been investigated
in [17] and an AODV variant is proposed after introducing a
new routing metric in evaluating path efficiency which includes
end-to-end delay and energy consumption. As a result, the net-
work lifetime is prolonged and end-to-end delivery ratio is im-
proved for real-time embedded systems.
Another QoS aware routing protocol is proposed in [16] for
WSN. It finds multiple least-cost and energy-efficient paths by
extended Dijkstra’s algorithm and pick the path that can meet
end-to-end delay requirement during the connection. In addi-
tion, a class-based queueing model is employed to serve both
best-effort and real-time traffics. Their approach, however, does
not consider the impact of channel access delay. Besides, the
use of class-based priority queueing mechanism is too compli-
cated and costly for resource limited sensor nodes.
Geographic routing is popular in WSN since it does not need
to maintain routing table and consequently can reduce network
energy consumption. Resulting algorithms are highly scalable
[18]. However, geographic routing protocols are in general not
optimal since most of them are based on one-hop decision.
In addition, determining node position will introduce some
overheads and energy consumption. Several solutions exist
for finding coordinates, e.g., using global positioning system
(GPS). Note that for resource-limited WSN, using GPS can be
a problem as the required positioning chips will increase the
price and energy consumption. This problem can be alleviated
by using positioning chips only in some nodes, while other
nodes calculate positions with the assistance of their neighbors.
On the other hand, existing localization techniques such as
triangulation, multilateration and diffusion [18] can provide
GPS-free solutions. Some ranging techniques have also been
specified in the IEEE 802.15.4a standard [19], e.g., estimating
distance by measuring the difference of propagation delays.
In geographic routing, the heuristic greedy forwarding
protocol SPEED [12] is the first one addressing real-time
guarantees for WSN. Relay velocity toward a next-hop node
is identified by dividing the distance progress by its estimated
forwarding delay. Packet deadline is mapped to a velocity re-
quirement. The node with the largest relay velocity higher than
Authorized licensed use limited to: Zhejiang University. Downloaded on June 10, 2009 at 10:19 from IEEE Xplore. Restrictions apply.
LI et al.: ENHANCING REAL-TIME DELIVERY IN WIRELESS SENSOR NETWORKS WITH TWO-HOP INFORMATION 115
the velocity requirement is selected in the highest probability.
If there is no neighbor node that can meet the requirement, the
packet is dropped probabilistically to regulate network work-
load. Meanwhile, back-pressure packet rerouting in large-delay
link is conducted to divert and reduce packets injected to a
congested area. MM-SPEED [20] extends SPEED by defining
multiple delivery velocities for packets with different dead-
lines in supporting different QoS. Real-time power-aware
routing (RPAR) [21] is another variant of SPEED. A node
will adaptively change its transmission power by the progress
towards destination and packet’s due time in order to meet the
required velocity in the most energy-efficient way. Note that
all the above protocols are based on one-hop neighborhood
information.
In our proposed scheme, we also adopt the approach of
mapping packet deadline to a velocity, which is known as a
good metric to delay constrained packet delivery. However, our
routing decision will be made based on two-hop neighborhood
information and corresponding metrics. It is therefore named
as two-hop velocity-based routing (THVR). Note that gener-
ally speaking it is also possible to employ other metrics, e.g.,
by packet lifetime or hop count, to design routing protocols.
The idea of two-hop information-based routing is generic and
applicable. Here, we will focus on THVR. The routing design
and details are given in the next section.
III. D
ESIGN OF
THVR FOR RT-WSN
Although two-hop information-based routing is intuitively
helpful to improve the routing decision, an explicit mechanism
is necessary. It is worth noting that THVR primarily aims at
lowering packet DMR for demanding real-time WSN but will
also consider energy utilization efficiency that has not been ex-
plicitly addressed in SPEED and MM-SPEED.
As assumed in most geographic routing algorithms, each
node in the network is aware of the geographic location of itself
and the destination, via GPS or other localization techniques
[18], [19] as mentioned in Section II. The information can be
further exchanged among two-hop neighbors [10], [22]. Thus,
each node is aware of its immediate and two-hop neighbors,
and their locations. This is achieved by two rounds of HELLO
messages. First, each node informs its neighbors about its
existence (ID, position, remaining energy, etc.). Next, each
node sends message to all its neighbors informing about its
one-hop neighbors. If the network is static or with low mobility,
this could be done at one stroke until there is node failure.
Otherwise, in a mobile network, each node periodically emits
additional HELLO messages to maintain two-hop information.
Too old entries are removed from the neighbor table, as corre-
sponding nodes have moved out of one-hop or two-hop range.
To be detailed below, our design is mainly composed of three
components: (i) forwarding metric; (ii) delay estimation and up-
date; and (iii) initiative drop control.
A. Forwarding Metric
To begin with, some definitions are introduced. For each
node,
is used to denote the set of its one-hop neighbors.
The source and destination nodes are labeled by
and ,
respectively. The distance between a pair of nodes
and
Fig. 1. Illustration of node’s neighbor set, one-hop and two-hop forwarder set.
is denoted by . Consequently, the required end-to-end
packet delivery velocity for deadline,
, is defined as
(1)
is defined as the set of node ’s potential forwarders which
will make a progress towards the destination, i.e.,
(2)
is defined to represent the set of corresponding two-hop
potential forwarders, i.e.,
(3)
An illustration of node’s neighbor set, one-hop and two-hop for-
warder set is shown in Fig. 1.
In SPEED, the core component SNGF (stateless nondeter-
ministic geographic forwarding) works as follows. Upon re-
ceiving a packet, node
calculates the velocity provided by each
of the forwarding nodes in
, which is expressible as
(4)
where
and denotes the estimated hop delay
between
and . If there exists such that , it will
be chosen as the forwarder with probability
following the
discrete exponential distribution below [12]:
(5)
where
is the number of candidates in and is a
weighting exponent to tradeoff between load balance and
optimal delivery delay. A larger
will lead to a shorter
end-to-end delay, while a smaller one can achieve a better load
balance.
In our proposed THVR, similarly to SPEED, by two-hop in-
formation, node
will calculate the velocity provided by each
of the two-hop forwarding pairs
, i.e.,
(6)
Authorized licensed use limited to: Zhejiang University. Downloaded on June 10, 2009 at 10:19 from IEEE Xplore. Restrictions apply.
116 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 5, NO. 2, MAY 2009
where and . For node pair satisfying
, we denote it by set . Beyond comparing the
potential forwarding velocities, we also take into account node’s
remaining energy level, and thus define the following new joint
metric:
(7)
where
is the remaining energy of forwarder candidate ,
while
is its initial energy, and is the weighting
factor incorporating energy level into the joint metric. Note that
larger
tends to favor end-to-end delay performance, while
smaller one can distribute traffics to nodes in higher energy
level and result in a better energy balance. Clearly, a setting of
relies on deadline requirements. The larger the deadline, the
smaller
could be.
By (7), the node in
(e.g., node ) with the largest
will be chosen as the forwarder. The routing then proceeds and
the mechanism is repeated at the selected node iteratively. In
THVR, the sender will search the largest velocity in two-hop
neighborhood before making the forwarding decision. However,
in SPEED [12], it is only one-hop optimized. For example, if
there is a topology hole after the first forwarding node, SPEED
will get a critical problem and have to activate back-pressure
rerouting. By THVR, this kind of problems can be alleviated.
Inherently, THVR has one-hop more prediction capability as
using a “telescope” in finding the path. Generally speaking, even
if the starting choice is not the globally optimized one, it may
still have a better chance to gradually be corrected due to the
farther sight and view.
B. Delay Estimation
From (6), it is observable that packet delay estimation from
sender to its potential forwarder has played an important role in
the velocity. In general, the delay of a packet from a node
to
its immediate forwarder
is expressible as
(8)
where
and are used to represent the
MAC delay, and transmission count, respectively. The transmis-
sion time includes the queueing delay (depending on the load of
the node) and the packet transmission time (determined by the
packet/ACK size and the bandwidth). The transmission count
refers to the number of retransmissions involved since automatic
repeat-request (ARQ) is adopted when packet fails to be trans-
mitted due to collision or lossy link.
To have packet delay estimation in identifying (6), we adopt
the method of window mean with exponentially weighted
moving average (WMEWMA), which has been shown in [23]
with its best estimation performance among existing tech-
niques. The estimate of
for time instant is given
by
(9)
Fig. 2. Delay estimation performance under different values of
.
Fig. 3. Deviation from true values under different values of
. The 90% con-
fidence interval is also plotted.
where
is the time window, is the newly measured delay
(known from the most recent packet), and
is the
tunable weighting coefficient. It is clear that a large
will em-
phasize
and fits the case where delay variance is small,
while a small
is more suitable if the variance is significant. A
demonstration of the delay estimates under different
is plotted
in Fig. 2, while the sum of deviations is indicated in Fig. 3.
With a small
, the delay estimate is insensitive and
too slow to capture the system’s immediate fluctuation and thus
may result in a big deviation sum. However, when
is too large
, the update to the delay estimate appears too rigorous
while nearly ignoring the historic average and results in an even
larger deviation sum. As indicated in Fig. 3, the deviation is the
lowest when is set to 0.5 which is quite robust generally. Note
that it is also possible to design an adaptive tuning mechanism
with reference to encountered delay variance. However, we will
not go into the detail in this paper.
To identify the link delay of a packet, a sender will stamp
the time when the packet is first sent and then compare it with
the time when an acknowledgment (ACK) is received. On the
other hand, to update the link delay information to other nodes
in the routing path, after receiving the ACK with delay informa-
tion from its forwarder, the node will initiate a feedback packet,
which contains the updated delay of the forwarding link, to its
Authorized licensed use limited to: Zhejiang University. Downloaded on June 10, 2009 at 10:19 from IEEE Xplore. Restrictions apply.
