UCLA
Papers
Title
ASCENT: Adaptive Self-Configuring sEnsor Networks Topologies
Permalink
https://escholarship.org/uc/item/0md8w9rb
Journal
Center for Embedded Network Sensing, 3(3)
Authors
Cerpa, Alberto E
Estrin, D
Publication Date
2004-05-21
Peer reviewed
eScholarship.org Powered by the California Digital Library
University of California
ASCENT: Adaptive Self-Configuring
sEnsor Networks Topologies
Alberto Cerpa, Student Member, IEEE, and Deborah Estrin, Fellow, IEEE
Abstract—Advances in microsensor and radio technology will enable small but smart sensors to be deployed for a wide range of
environmental monitoring applications. The low per-node cost will allow these wireless networks of sensors and actuators to be
densely distributed. The nodes in these dense networks will coordinate to perform the distributed sensing and actuation tasks.
Moreover, as described in this paper, the nodes can also coordinate to exploit the redundancy provided by high density so as to extend
overall system lifetime. The large number of nodes deployed in these systems will preclude manual configuration, and the
environmental dynamics will preclude design-time preconfiguration. Therefore, nodes will have to self-configure to establish a topology
that provides communication under stringent energy constraints. ASCENT builds on the notion that, as density increases, only a
subset of the nodes are necessary to establish a routing forwarding backbone. In ASCENT, each node assesses its connectivity and
adapts its participation in the multihop network topology based on the measured operating region. This paper motivates and describes
the ASCENT algorithm and presents analysis, simulation, and experimental measurements. We show that the system achieves linear
increase in energy savings as a function of the density and the convergence time required in case of node failures while still providing
adequate connectivity.
Index Terms—Wireless sensor networks, adaptive topology, topology control, energy conservation.
æ
1INTRODUCTION
T
HE availability of microsensors and low-power wireless
communications will enable the deployment of densely
distributed sensor/actuator networks for a wide range of
environmental monitoring applications from urban to
wilderness environments; indoors and outdoors; and
encompassing a variety of data types including acoustic,
image, and various chemical and physical properties. The
sensor nodes will perform significant signal processing,
computation, and network self-configuration to achieve
scalable, robust, and long-lived networks [2], [8], [7]. More
specifically, sensor nodes will do local processing to reduce
communications and, consequently, energy costs.
In this paper, we describe and present simulation and
experimental performance studies for a form of adaptive
self-configuration designed for sensor networks. As we
argue in Section 2, these unattended systems will need to
self-configure and adapt to a wide variety of environmental
dynamics and terrain conditions. These conditions produce
regions with nonuniform communication den sity. We
suggest that one of the ways system designers can address
such challenging operating conditions is by deploying
redundant nodes and designing the system algorithms to
make use of that redundancy over time to extend the
systems life. In ASCENT, each node assesses its connectiv-
ity and adapts its participation in the multihop network
topology based on the measured operating region. For
instance, a node:
. signals when it detects high packet loss, requesting
additional nodes in the region to join the network in
order to relay messages,
. reduces its duty cycle if it detects high packet losses
due to collisions,
. probes the local communication environment and
does not join the multihop routing infrastructure
until it is “helpful” to do so.
Why can this adaptive configuration not be done from a
central node? In addition to the scaling and robustness
limitations of centralized solutions, a single node cannot
directly sense the conditions of nodes distributed elsewhere
in space. Consequently, other nodes would need to
communicate detailed information about the state of their
connectivity in order for the central node to determine who
should join the multihop network. When energy is a
constraint and the environment is dynamic, distributed
approaches are attractive and possibly are the only practical
approach [22] because they avoid transmitting dynamic
state information repeatedly across the network.
Pottie and Kaiser [22] initiated work in the general area
of wireless sensor networks by establishing that scalable
wireless sensor networks require multihop operation to
avoid sending large amounts of data over long distances.
They went on to define techniques by which wireless
nodes discover their neighbors and acquire synchronism.
Given this basic bootstrapping capability, our work
addresses the next level of automatic configuration that
will be needed to realize envisioned sensor networks,
namely, how to form the multihop topology [7]. Given the
ability to send and receive packets and the objective of
forming an energy-efficient multihop network, we apply
IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 3, NO. 3, JULY-SEPTEMBER 2004 1
. The authors are with the Center for Embedded Networked Sensing,
Computer Science Department, University of California at Los Angeles,
3531H Boelter Hall, Box 951596, Los Angeles, CA 90095-1596.
E-mail: {cerpa, destrin}@cs.ucla.edu.
Manuscript received 21 Mar. 2004; revised 21 May 2004; accepted 26 May
2004.
For information on obtaining reprints of this article, please send e-mail to:
tmc@computer.org, and reference IEEECS Log Number TMCSI-0111-0304.
1536-1233/04/$20.00 ß 2004 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS
well-known techniques from MAC layer protocols to the
problem of distributed topology formation.
In the following section, we present a sensor network
scenario, stating our assumptions and contributions. Re-
lated work is reviewed in Section 3. Section 4 describes
ASCENT in more detail. In Section 5, we present some
initial analysis, simulation, and experimental results using
ASCENT. Finally, in Section 6, we conclude.
2DISTRIBUTED SENSOR NETWORK SCENARIO
To motivate our research, consider a habitat monitoring
sensor network that is to be deployed in a remote forest.
Deployment of this network can be done, for example, by
dropping a large number of sensor nodes from a plane or
placing them by hand. In this example and in many other
anticipated applications of ad hoc wireless sensor networks
[5], the deployed systems must be designed to operate
under the following conditions and constraints:
. Ad hoc deployment: We cannot expect the sensor
field to be deployed in a regular fashion (e.g., a
linear array, 2D lattice). More importantly, uniform
deployment does not correspond to uniform con-
nectivity owing to unpredictable propagation effects
when nodes, and therefore antennae, are close to the
ground and other surfaces.
. Energy constraints: The nodes (or at least some
significant subset) will be untethered for power as
well as communications and therefore the system
must be designed to expend as little energy as is
possible in order to maximize network lifetime.
. Unattended operation under dynamics: The antici-
pated number of elements in these systems will
preclude manual configuration, and the environ-
mental dynamics will preclude desi gn-time
preconfiguration.
In many such contexts, it will be far easier to deploy
larger numbers of nodes initially than to deploy additional
nodes or additional energy reserves at a later date (similar
to the economics of stringing cable for wired networks). In
this paper, we present one way in which nodes can exploit
the resulting redundancy in order to extend system lifetime.
If we use too few of the deployed nodes, the distance
between neighboring nodes will be too great and the packet
loss rate will increase or the energy required to transmit the
data over the longer distances will be prohibitive. If we use
all deployed nodes simultaneously, the system will be
expending unnecessary energy at best and, at worst, the
nodes may interfere with one another by congesting the
channel. In the process of finding an equilibrium, we are not
trying to use a distributed localized algorithm to identify a
single optimal solution. Rather, this form of adaptive self-
configuration using localized algorithms is well suited to
problem spaces that have a large number of possible
solutions; in this context, a large solution space translates
into dense node deployment. Our simulation and experi-
mental results confirm that this is the case for our
application.
We enumerate the following assumptions that apply to
the remainder of our work:
We assume a Carrier Sense Multiple Access (CSMA)
MAC protocol with the capacity to work in promiscuous
mode. This clearly introduces the possibilities for resource
contention when too many neighboring nodes participate in
the multihop network. Our approach should be relevant to
TDMA MACs as well because distributed slot allocation
schemes will also have degrad ed per formance wit h
increased load.
Our algorithm reacts when links experience high packet
loss. The ASCENT mechanism does not detect or repair
network partitions of the underlying raw topology. Parti-
tions are more prevalent when node density is low, and our
approach is not applicable because, in general, all nodes
will be needed to form an effective network. Of course,
network partitions can occur even in dense arrays when a
swath of nodes are destroyed or obstructed. When such
network partitions do occur, complementary system me-
chanisms will be needed; for example, detecting partitions
in the multihop sensor network by exploiting information
from long range radios deployed on a subset of nodes and
used sparingly because of the power required. We leave
such complementary techniques for network partition
detection and repair to future work.
The two primary contributions of our design are:
. The use of adaptive techniques that permit applica-
tions to configure the underlying topology based on
their needs while trying to save energy to extend
network lifetime. Our work does not presume a
particular model of fairness, degree of connectivity,
or capacity required.
. The use of self-configuring techniques that react to
operating conditions measured locally. Our work is
not restricted to the radio propagation model, the
geographical distribution of nodes, or the routing
mechanisms used.
3RELATED WORK
Our work has been informed and influenced by a variety of
other research efforts. There has been a great deal of work
in the area of topology control, mostly using theoretical
analysis or simulation and involving MAC and power
control mechanisms.
There have been several important theoretical evalua-
tions of topology control. Most of this work focuses on the
analysis of algorithms for distributed construction of a
connected dominating set (CDS) of the corresponding unit-
disk graph and the routing strategies using the CDS
backbone [11], [29], [1], [12]. Gao et al. [12] present a
randomized algorithm for maintaining a CDS with low
overhead. Gao et al.’s algorithm assumes the partion of the
space in a grid and selects a small number of cluster heads.
The total number selected has an approximation factor of
Oð
ffiffiffi
n
p
Þ of the minimum theoretically possible. In later work,
Gao et al. present a distributed algorithm to construct a
restricted Delaunay graph (RDG), where only Delaunay
edges with a limited fix transmission radius are included
[11]. The work shows that the number of edges in the
restricted Delaunay graph is linear in the number of nodes,
although the maximum degree of a node may be ðnÞ in the
2 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 3, NO. 3, JULY-SEPTEMBER 2004
worstcase.Alzoubietal.[1]describeadistributed
algorithm for constructing a minimum connected dominat-
ing set (MCDS) for the unit-disk-graph with a constant
approximation ratio of the minimum possible and linear
time and message complexity. Wang and Li propose an
algorithm to build a geometric spanner that can be
implemented in a distributed manner [29]. The node degree
is bounded by a positive constant, and the resulting
backbone is a spanner for both hops and length.
The above algorithms provide the theoretical limits and
bounds of what is achievable with topology control. Our
work with ASCENT complements theirs by getting results
from experiments using real radios, rather than using only
simulation and analysis. Recent work [10], [4], [33], [30]
evaluating radio connectivity using low-power radios
suggests that these radio channels present asymmetrical
links, nonisotropic connectivity, and nonmonotonic distance
decay of power with distance. It is important to understand
the effects that these conditions impose on these topology
control algorithms since most of the r eal conditions
observed using real radios violate the assumptions in the
previous theoretical studies and may affect correctness.
There is poor correlation between the spatial distance and
reception rate, so assumptions based on geographic
proximity between nodes do not necessarily hold in
practice. Furthermore, the radio propagation is not circular,
presenting nonisotropic properties. Finally, our previous
work with SCALE [4] has shown the presence of asym-
metric links for 5-30 percent of all pairwise communication,
causing serious problems with algorithms that assume
bidirectional connectivity.
The main approach followed by MAC level protocols to
save energy has been to turn off the radios that do not have
any scheduled transmission or reception of packets in a
particular (usually small) timeframe. These protocols
usually trade network delay for energy conservation
because of the startup cost associated with turning the
radios back on. Sohrabi and Pottie [27] present a self-
configuration and synchronization TDMA scheme at the
single cluster. This work is more focused on the low-level
synchronization necessary for network self-assembly, while
we concentrate on efficient multihop topology formation.
Sparse Topology and Energy Management (STEM) [26]
accepts delays in path-setup time in exchange for energy
savings. It uses a second radio (operating at a lower duty
cycle) as a paging channel. Sensor-MAC (S-MAC) [32] treats
both per-node fairness and latency as secondary to energy
conservation. It periodically turns off the radios of idle
nodes and uses in-channel signaling to turn off radios that
are not taking part in the current communication. More
recent work [34] continues to explore MAC-level wake-up
schemes. Most of the MAC schemes mentioned above are
complementary to our work. ASCENT could establish a
particular active topology and then use any of the above
mechanisms to gain even further energy savings on the
newly created active topology.
Another approach in reducing energy consumption has
been to adaptively control the transmit power of the radio.
The lazy scheduling proposed in Prabhakar et al. [23]
transmits packets with the lowest possible transmit power
for the longest possible time such that delay constraints are
still met. Ramanathan and Rosales-Hain [24] proposed
some distributed heuristics to adaptively adjust node
transmit powers in response to topological changes caused
by mobile nodes. This work assumes that a routing
protocol is running at all times and provides basic
neighbor information that is used to dynamically adjust
transmit power. While power control can be very useful,
particularly in asymmetric networks such as cellular
telephony, their advantages are less pronounced in sensor
networks [4]. Furthermore, the power consumed by these
low-power radios in idle state is of the same order of
magnitude than the Tx or Rx state, so optimizations on
transmit power are less important. Under these conditions,
turning the radio off and putting the transceiver in sleep
state is essential to extend network lifetime.
In Xu et al. [31], GAF nodes use geographic location
information to divide the network into fixed square grids.
Nodesineachgridalternatebetweensleepingand
listening, and there is always one node active to route
packets per grid. ASCENT does not need any location aids
since it is based on connectivity. In addition, geographic
proximity may not always lead to radio connectivity; this is
why ASCENT uses local connectivity measurements. Chen
et al. [6] proposed SPAN, an energy efficient algorithm for
topology maintenance, where nodes decide whether to
sleep or join the backbone based on connectivity informa-
tion supplied by a routing protocol. ASCENT does not
depend on routing information nor need to modify the
routing state; it decides whether to join the network or sleep
based on measured local connectivity. In addition, our work
does not presume a particular model of fairness or network
capacity that the application requires.
Mobile ad hoc networks [16], [20], [21] and directed
diffusion [14] adaptively configure the routing or data
disseminatio n paths, but they do not adapt the basic
topology. Li and Rus [17] presented a scheme where mobile
nodes modify their trajectory to transmit messages in the
context of disconnected ad hoc networks. This work may
complement ours in case of mobile nodes deployment and
in the presence of network partitions.
The adaptive techniques we use were studied exten-
sively to make the MAC layer self-configuring and adaptive
more than 20 years ago during the refinement of contention
protocols [15], [18]. More recently, SRM [9] and RTCP [25]
borrowed these techniques to adaptively adjust parameters
such as session message frequency and randomization
intervals. In this work, we use those techniques to adapt the
topology of a multihop wireless network.
The following section describes the ASCENT protocol in
some detail.
4 ASCENT DESIGN
ASCENT adaptively elects “active” nodes from all nodes in
the network. Active nodes stay awake all the time and
perform multihop packet routing, while the rest of the
nodes remain “passive” and periodically check if they
should become active.
Consider a simple sensor network for data gathering
similar to the network described in Section 2. Fig. 1 shows a
simplified schematic for ASCENT during initialization in a
CERPA AND ESTRIN: ASCENT: ADAPTIVE SELF-CONFIGURING SENSOR NETWORKS TOPOLOGIES 3
high-density region. For the sake of clarity, we show only
the formation of a two-hop network. This analysis may be
extended to networks of larger sizes.
Initially, only some nodes are active. The other nodes
remain passively listening to packets but not transmitting.
This situation is depicted in Fig. 1a. The source starts
transmitting data packets toward the sink. Because the sink
is at the limit of radio range, it gets very high packet loss
from the source. We call this situation a communication hole.
The sink then starts sendin g help messages to signal
neighbors that are in listen-only mode—also called passive
neighbors—to join the network.
When a neighbor receives a help message, it may decide to
join the network. This situation is illustrated in Fig. 1b.
When a node joins the network, it starts transmitting and
receiving packets, i.e., it becomes an active neighbor. As soon
as a node decides to join the network, it signals the existence
of a new active neighbor to other passive neighbors by
sending an neighbor announcement message. This situation
continues until the number of active nodes stabilizes on a
certain value and the cycle stops (see Fig. 1c). When the
process completes, the group of newly active neighbors that
have joined the network make the delivery of data from
source to sink more reliable. The process will restart when
some future network event (e.g., node failure) or environ-
mental effect (e.g., new obstacle) causes packet loss again.
In this section, we describe the ASCENT algorithm and
their components. We elaborate on several design choices
while we describe the scheme. Our initial analysis, simula-
tions, and experiments in Section 5 focus only on a subset of
these design choices.
4.1 ASCENT State Transitions
In ASCENT, nodes are in one of four states: sleep, passive,
test, and active. Fig. 2 shows a state transition diagram.
Initially, a random timer turns on the nodes to avoid
synchronization. When a node starts, it initializes in the test
state. Nodes in the test state exchange data and routing
control messages. In addition, when a node enters the test
state, it sets up a timer T
t
and sends neighbor announcement
messages. When T
t
expires, the node enters the active state. If,
before T
t
expires, the number of active neighbors is above
the neighbor threshold (NT ) or if the average data loss rate
(DL) is higher than the average loss before entering in the
test state, then the node moves into the passive state .If
multiple nodes make a transition to the test state, then we
use the node ID in the announcement message as a tie
breaking mechanism (higher IDs win). The intuition behind
the test state is to probe the network to see if the addition of
a new node may actually improve connectivity.
When a node enters the passive state, it sets up a timer T
p
and sends new passive node announcement messages. This
information is used by active nodes to make an estimate of
the total density of nodes in the neighborhood. Active
nodes transmit this density estimate to any new passive
node in the neighborhood. When T
p
expires, the node enters
the sleep state. If, before T
p
expires, the number of
neighbors is below NT and either the DL is higher than
the loss threshold (LT )orDL is below the loss threshold but
the node received a help message from an active neighbor, it
makes a transition to the test state. While in passive state,
nodes have their radio on and are able to overhear all
packets transmitted by their active neighbors. No routing or
data packets are forwarded in this state since this is a listen-
only state. The intuition behind the passive state is to gather
information regarding the state of the network without
causing interference with the other nodes. Nodes in the
passive and test states continuously update the number of
active neighbors and data loss rate values. Energy is still
consumed in the passive state since the radio is still on when
not receiving packets. A node that enters the sleep state turns
the radio off, sets a timer T
s
, and goes to sleep. When T
s
expires, the node moves into passive state. Finally, a node in
active state continues forwarding data and routing packets
until it runs out of energy. If the data loss rate is greater than
LT, the active node sends help messages.
4 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 3, NO. 3, JULY-SEPTEMBER 2004
Fig. 1. Network self-configuration example. (a) Communication hole. (b) Transition. (c) Final state.
Fig. 2. ASCENT state transitions.
