Original PDF Flash format wireless-sensor-networks-for-habitat-monitoring  


Wireless Sensor Networks For Habitat Monitoring

Wireless Sensor Networks for Habitat Monitoring
Alan Mainwaring1
Joseph Polastre2
Robert Szewczyk2
David Culler1,2
John Anderson3
1 Intel Research Laboratory, Berkeley
2 EECS Department
3 College of the Atlantic
Intel Corporation
University of California at Berkeley
Bar Harbor, Maine
{amm,dculler}@intel-research.net
{polastre,szewczyk,culler}@cs.berkeley.edu
jga@ecology.coa.edu
ABSTRACT
complex filtering and triggering functions, as well as to apply
application-specific or sensor-specific data compression algo-
We provide an in-depth study of applying wireless sensor
rithms. The ability to communicate not only allows informa-
networks to real-world habitat monitoring. A set of system
tion and control to be communicated across the network of
design requirements are developed that cover the hardware
nodes, but nodes to cooperate in performing more complex
design of the nodes, the design of the sensor network, and
tasks, like statistical sampling, data aggregation, and sys-
the capabilities for remote data access and management. A
tem health and status monitoring [8, 9]. Increased power
system architecture is proposed to address these require-
efficiency gives applications flexibility in resolving funda-
ments for habitat monitoring in general, and an instance of
mental design tradeoffs, e.g., between sampling rates and
the architecture for monitoring seabird nesting environment
battery lifetimes. Low-power radios with well-designed pro-
and behavior is presented. The currently deployed network
tocol stacks allow generalized communications among net-
consists of 32 nodes on a small island off the coast of Maine
work nodes, rather than point-to-point telemetry. The com-
streaming useful live data onto the web. The application-
puting and networking capabilities allow sensor networks to
driven design exercise serves to identify important areas of
be reprogrammed or retasked after deployment in the field.
further work in data sampling, communications, network re-
Nodes have the ability to adapt their operation over time
tasking, and health monitoring.
in response to changes in the environment, the condition of
the sensor network itself, or the scientific endeavor.
Categories and Subject Descriptors
We are working with members of the life science commu-
C.2.1 [Computer Communication Networks]: Network
nity to make the potential of this emerging technology a
Architecture and Design; C.3 [Computer Systems Orga-
reality. Taking an application-driven approach quickly sep-
nization]: Special-Purpose and Application-based Systems;
arates actual problems from potential ones, and relevant is-
J.3 [Computer Applications]: Life and Medical Sciences
sues from irrelevant ones. The application context helps to
differentiate problems with simple, concrete solutions from
General Terms
open research areas. However, we seek to develop an effec-
tive sensor network architecture for the domain, not just a
Design, Performance, Experimentation
particular instance, so we must look for general solutions.
Collaboration with scientists in other fields helps to define
1.
INTRODUCTION
the broader application space, as well as specific application
Habitat and environmental monitoring represent a class
requirements, allows field testing of experimental systems,
of sensor network applications with enormous potential ben-
and offers objective evaluations of the technologies.
The
efits for scientific communities and society as a whole. In-
impact of sensor networks for habitat and environmental
strumenting natural spaces with numerous networked micro-
monitoring will be measured by their ability to enable new
sensors can enable long-term data collection at scales and
applications and produce new results otherwise too difficult
resolutions that are difficult, if not impossible, to obtain oth-
to realize.
erwise. The intimate connection with its immediate physical
This paper develops a specific habitat monitoring applica-
environment allows each sensor to provide localized mea-
tion, that is largely representative of the domain. It presents
surements and detailed information that is hard to obtain
a collection of requirements, constraints and guidelines that
through traditional instrumentation. The integration of lo-
serve as a basis for a general sensor network architecture for
cal processing and storage allows sensor nodes to perform
many such applications. It describes the core components
of the sensor network for this domain – the hardware and
sensor platforms, the distinct networks involved, their inter-
connection, and the data management facilities. The design
Permission to make digital or hard copies of all or part of this work for
and implementation of the essential network services, in-
personal or classroom use is granted without fee provided that copies are
cluding power management, communications, retasking and
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
node management, can be evaluated in this context.
republish, to post on servers or to redistribute to lists, requires prior specific
The remainder of the paper is organized as follows. Sec-
permission and/or a fee.
tion 2 identifies the requirements of our habitat monitoring
WSNA’02, September 28, 2002, Atlanta, Georgia, USA.
application. Section 3 presents a tiered sensor network ar-
Copyright 2002 ACM 1-58113-589-0/02/0009 ...$5.00.
chitecture that interconnects the core system components
study sites, often limited by concerns about frequent access
ranging from very localized collections of sensor nodes to
and habitability.
the area of study to the wide-area where data is ultimately
analyzed.
Section 4 discusses the implementation of the
2.1
Great Duck Island
hardware and software modules in our instantiation of this
The College of the Atlantic (COA) is field testing in-situ
architecture and the design trade-offs present in our deploy-
sensor networks for habitat monitoring.
COA has ongo-
ment on Great Duck Island, Maine. Section 5 describes the
ing field research programs on several remote islands with
results from our initial deployment of a network of 32 nodes
well established on-site infrastructure and logistical support.
in and around the burrows of nesting Leach’s Storm Petrels.
Great Duck Island (GDI) (44.09N,68.15W) is a 237 acre is-
Section 6 draws lessons from this application-driven design
land located 15 km south of Mount Desert Island, Maine.
exercise to identify important directions for further investi-
The Nature Conservancy, the State of Maine and the College
gation. Section 7 provides concluding remarks.
of the Atlantic hold much of the island in joint tenancy.
At GDI, we are primarily interested in three major ques-
2.
HABITAT MONITORING
tions in monitoring the Leach’s Storm Petrel [2]:
Researchers in the Life Sciences are becoming increasingly
1. What is the usage pattern of nesting burrows over the
concerned about the potential impacts of human presence in
24-72 hour cycle when one or both members of a breed-
monitoring plants and animals in field conditions. At best it
ing pair may alternate incubation duties with feeding
is possible that chronic human disturbance may distort re-
at sea?
sults by changing behavioral patterns or distributions, while
at worst anthropogenic disturbance can seriously reduce or
2. What changes can be observed in the burrow and sur-
even destroy sensitive populations by increasing stress, re-
face environmental parameters during the course of
ducing breeding success, increasing predation, or causing a
the approximately 7 month breeding season (April-
shift to unsuitable habitats. While the effects of disturbance
October)?
are usually immediately obvious in animals, plant popula-
3. What are the differences in the micro-environments
tions are sensitive to trampling by even well-intended re-
with and without large numbers of nesting petrels?
searchers, introduction of exotic elements through frequent
visitation, and changes in local drainage patterns through
Each of these questions has unique data needs and suit-
path formation.
able data acquisition rates. Presence/absence data is most
Disturbance effects are of particular concern in small is-
likely acquired through occupancy detection and tempera-
land situations, where it may be physically impossible for
ture differentials between burrows with adult birds and bur-
researchers to avoid some impact on an entire population. In
rows that contain eggs, chicks, or are empty. Petrels are
addition, islands often serve as refugia for species that can-
unlikely to enter or leave during the light phase of a 24 hour
not adapt to the presence of terrestrial mammals, or may
cycle, but measurements every 5-10 minutes during the late
hold fragments of once widespread populations that have
evening and early morning are needed to capture time of
been extirpated from much of their former range.
entry or exit. More general environmental differentials be-
Seabird colonies are notorious for their sensitivity to hu-
tween burrow and surface conditions during the extended
man disturbance. Research in Maine [2] suggests that even a
breeding season can be captured by records every 2-4 hours,
15 minute visit to a cormorant colony can result in up to 20%
while differences between “popular” and “unpopular” sites
mortality among eggs and chicks in a given breeding year.
benefit from hourly sampling, especially at the beginning of
Repeated disturbance will lead to complete abandonment of
the breeding season.
the colony. On Kent Island, Nova Scotia, researchers found
It is unlikely that any one parameter recorded by wireless
that Leach’s Storm Petrels are likely to desert their nest-
sensors could determine why petrels choose a specific nest
ing burrows if they are disturbed during the first 2 weeks of
site, rather we hope that by making multiple measurements
incubation.
of many variables we will be able to develop predictive mod-
Sensor networks represent a significant advance over tra-
els. These models will correlate which conditions seabirds
ditional invasive methods of monitoring.
Sensors can be
prefer.
deployed prior to the onset of the breeding season or other
sensitive period (in the case of animals) or while plants are
2.2
Great Duck Island Requirements
dormant or the ground is frozen (in the case of botanical
studies). Sensors can be deployed on small islets where it
2.2.1
Internet access
would be unsafe or unwise to repeatedly attempt field stud-
The sensor networks at GDI must be accessible via the
ies. The results of wireless sensor-based monitoring efforts
Internet. An essential aspect of habitat monitoring appli-
can be compared with previous studies that have tradition-
cations is the ability to support remote interactions with
ally ignored or discounted disturbance effects.
in-situ networks.
Finally, sensor network deployment may represent a sub-
stantially more economical method for conducting long-term
2.2.2
Hierarchical network
studies than traditional personnel-rich methods. Presently,
The field station at GDI needs sufficient resources to host
a substantial proportion of logistics and infrastructure must
Internet connectivity and database systems. However, the
be devoted to the maintenance of field studies, often at some
habitats of scientific interest are located up to several kilo-
discomfort and occasionally at some real risk. A “deploy ’em
meters further away. A second tier of wireless networking
and leave ’em” strategy of wireless sensor usage would limit
provides connectivity to multiple patches of sensor networks
logistical needs to initial placement and occasional servicing.
deployed at each of the areas of interest.
Three to four
This could also greatly increase access to a wider array of
patches of 100 static (not mobile) nodes is sufficient to start.
2.2.3
Sensor network longevity
Sensor networks that run for 9 months from non-rechargeable
power sources would have significant audiences today. Al-
though ecological studies at GDI span multiple field seasons,
individual field seasons typically vary from 9 to 12 months.
Seasonal changes as well as the plants and animals of interest
determine their durations.
2.2.4
Operating off-the-grid
Every level of the network must operate with bounded en-
ergy supplies. Although renewable energy, for example solar
power, may be available at some locations, disconnected op-
eration remains a possibility. GDI has sufficient solar power
to run many elements of the application 24x7 with low prob-
abilities of service interruptions due to power loss.
2.2.5
Management at-a-distance
Figure 1: System architecture for habitat monitor-
The remoteness of the field sites requires the ability to
ing
monitor and manage sensor networks over the Internet. Al-
though personnel may be on site for a few months each sum-
mer, the goal is zero on-site presence for maintenance and
tion, the timely delivery of fresh sensor data is key. Lastly,
administration during the field season, except for installa-
nodal data summaries and periodic health-and-status mon-
tion and removal of nodes.
itoring requires timely delivery.
2.2.6
Inconspicuous operation
Habitat monitoring infrastructure must be inconspicuous.
3.
SYSTEM ARCHITECTURE
It should not disrupt the natural processes or behaviors un-
We now describe the system architecture, functionality
der study. Removing human presence from the study areas
of individual components and how they operate together.
both eliminates a source of error and variation in data col-
We explain how they address the requirements set forth in
lection, as well as a significant source of disturbance.
Section 2.
We developed a tiered architecture. The lowest level con-
2.2.7
System behavior
sists of the sensor nodes that perform general purpose com-
From both a systems and end-user perspective, it is criti-
puting and networking in addition to application-specific
cal that sensor networks exhibit stable, predictable, and re-
sensing. The sensor nodes may be deployed in dense patches
peatable behavior whenever possible. An unpredictable sys-
that are widely separated. The sensor nodes transmit their
tem is difficult to debug and maintain. More importantly,
data through the sensor network to the sensor network gate-
predictability is essential in developing trust in these new
way.
The gateway is responsible for transmitting sensor
technologies for life scientists.
data from the sensor patch through a local transit network
to the remote base station that provides WAN connectivity
2.2.8
In-situ interactions
and data logging. The base station connects to database
Although the majority of interactions with the sensor net-
replicas across the internet. Finally, the data is displayed
works are expected to be via the Internet, local interactions
to scientists through a user interface. Mobile devices, which
are required during initial deployment, during maintenance
we refer to as the gizmo, may interact with any of the net-
tasks, as well as during on-site visits. PDAs serve an impor-
works – whether it is used in the field or across the world
tant role in assisting with these tasks. They may directly
connected to a database replica.
The full architecture is
query a sensor, adjust operational parameters, or simply as-
depicted in Figure 1.
sist in locating devices.
The lowest level of the sensing application is provided by
autonomous sensor nodes.
These small, battery-powered
2.2.9
Sensors and sampling
devices are placed in areas of interest. Each sensor node
For our particular applications, the ability to sense light,
collects environmental data primarily about its immediate
temperature, infrared, relative humidity, and barometric pres-
surroundings. Because it is placed close to the phenomenon
sure provide an essential set of useful measurements. The
of interest, the sensors can often be built using small and in-
ability to sense additional phenomena, such as accelera-
expensive individual sensors. High spatial resolution can be
tion/vibration, weight, chemical vapors, gas concentrations,
achieved through dense deployment of sensor nodes. Com-
pH, and noise levels would augment them.
pared with traditional approaches, which use a few high
quality sensors with sophisticated signal processing, this ar-
2.2.10
Data archiving
chitecture provides higher robustness against occlusions and
Archiving sensor readings for off-line data mining and
component failures.
analysis is essential. The reliable offloading of sensor logs to
The computational module is a programmable unit that
databases in the wired, powered infrastructure is an essential
provides computation, storage, and bidirectional communi-
capability. The desire to interactively “drill-down” and ex-
cation with other nodes in the system. The computational
plore individual sensors, or a subset of sensors, in near real-
module interfaces with the analog and digital sensors on the
time complement log-based studies. In this mode of opera-
sensor module, performs basic signal processing (e.g., simple
translations based on calibration data or threshold filters),
and dispatches the data according to the application’s needs.
Compared with traditional data logging systems, networked
sensors offer two major advantages: they can be retasked in
the field and they can easily communicate with the rest of
the system. In-situ retasking allows the scientists to refocus
their observations based on the analysis of the initial results.
Suppose that initially we want to collect the absolute tem-
perature readings; however after the initial interpretation
of the data we might realize that significant temperature
changes exceeding a defined threshold are most interesting.
Individual sensor nodes communicate and coordinate with
one another. The sensors will typically form a multihop net-
work by forwarding each other’s messages, which vastly ex-
tends connectivity options. If appropriate, the network can
perform in-network aggregation (e.g., reporting the average
temperature across a region). This flexible communication
Figure 2: Mica Hardware Platform: The Mica sen-
structure allows us to produce a network that delivers the
sor node (left) with the Mica Weather Board devel-
required data while meeting the energy requirements. We
oped for environmental monitoring applications
expand on energy efficient communication protocols in Sec-
tion 6.
Ultimately, data from each sensor needs to be propagated
this kind of communication, a “custody transfer” model,
to the Internet. The propagated data may be raw, filtered,
similar to SMTP messages or bundles [10], may be applica-
or processed data. Bringing direct wide area connectivity
ble.
to each sensor path is not feasible – the equipment is too
Users interact with the sensor network data in two ways.
costly, it requires too much power and the installation of
Remote users access the replica of the base station database
all required equipment is quite intrusive to the habitat. In-
(in the degenerate case they interact with the database di-
stead, the wide area connectivity is brought to a base station,
rectly). This approach allows for easy integration with data
adequate power and housing for the equipment is provided.
analysis and mining tools, while masking the potential wide
The base station may communicate with the sensor patch
area disconnections with the base stations. Remote control
using a wireless local area network. Wireless networks are
of the network is also provided through the database inter-
particularly advantageous since often each habitat involves
face. Although this control interface is sufficient for remote
monitoring several particularly interesting areas, each with
users, on-site users may often require a more direct interac-
its own dedicated sensor patch.
tion with the network. A small, PDA-sized device, referred
Each sensor patch is equipped with a gateway which can
to as gizmo, enables such interaction. The gizmo can di-
communicate with the sensor network and provides connec-
rectly communicate with the sensor patch, provide the user
tivity to the transit network. The transit network may con-
with a fresh set of readings about the environment and mon-
sist of a single hop link or a series of networked wireless
itors the network. While the gizmo will typically not take
nodes, perhaps in a path from the gateway to base sta-
custody of any data, it allows the user to interactively con-
tion. Each transit network design has different characteris-
trol the network parameters by adjusting the sampling rates,
tics with respect to expected robustness, bandwidth, energy
power management parameters and other network parame-
efficiency, cost, and manageability.
ters. The connectivity between any sensor node and the
To provide data to remote end-users, the base station in-
gizmo does not have to rely on functioning multihop sensor
cludes WAN connectivity and persistent data storage for the
network routing, instead the user will often communicate
collection of sensor patches. Since many habitats of interest
with the mote network directly, relying on single hop prox-
are quite remote, we expect that the WAN connection will
imity. We expect that this device will be extremely useful
be wireless (e.g., two-way satellite). The components must
during the initial deployment and during retasking of the
be reliable, enclosed in environmentally protected housing,
network.
and provided with adequate power. In many environments
such conditions can be provided relatively easily at a ranger
station.
4.
IMPLEMENTATION STRATEGIES
The architecture needs to address the possibility of dis-
connection at every level. Each layer (sensor nodes, gate-
4.1
Sensor Network Node
ways, base stations) has some persistent storage which pro-
In our deployment, we are using UC Berkeley motes as the
tects against data loss in case of power outage. Each layer
sensor nodes. The latest member of the mote family, called
also provides data management services. At the sensor level,
Mica [11] (shown in Figure 2), uses a single channel, 916MHz
these will be quite primitive, taking the form of data logging.
radio from RF Monolithics to provide bidirectional commu-
The base station may offer a full-fledged relational database
nication at 40kbps, an Atmel Atmega 103 microcontroller
service.
The data management at the gateways will fall
running at 4MHz, and considerable amount of nonvolatile
somewhere in between; they may offer some database ser-
storage (512 KB). A pair of conventional AA batteries and a
vices, but perhaps over limited window of data. While many
DC boost converter provide a stable voltage source, though
types of communication can be unreliable, when it comes to
other renewable energy sources can be easily used. Small
data collection, long-latency is preferable to data loss. For
size (approximately 2.0 x 1.5 x 0.5 inches).
Sensor
Accuracy
Interchangeability
Sample Rate
Startup
Current
Photoresistor
N/A
10%
2000 Hz
10 ms
1.235 mA
I2C Temperature
1 K
0.20 K
2 Hz
500 ms
0.150 mA
Barometric Pressure
1.5 mbar
0.5%
10 Hz
500 ms
0.010 mA
Barometric Pressure Temp
0.8 K
0.24 K
10 Hz
500 ms
0.010 mA
Humidity
2%
3%
500 Hz
500-30000 ms
0.775 mA
Thermopile
3 K
5%
2000 Hz
200 ms
0.170 mA
Thermistor
5 K
10%
2000 Hz
10 ms
0.126 mA
Table 1: Mica Weather Board: Characteristics of each sensor included on the Mica Weather Board.
4.2
Sensor Board
terchangeability and accuracy, the sensors can be deployed
To provide relevant measurements to scientists, we de-
in the field quicker since little or no calibration is needed
signed and manufactured an environmental monitoring sen-
prior to deployment. Another key aspect of choosing a sen-
sor board, shown in Figure 2. The Mica Weather Board
sor is its startup time. The start up time is the time a sensor
provides sensors that monitor changing environmental con-
must be powered before its reading stabilizes. Sensors with
ditions with the same functionality as a traditional weather
long start up times require current for a longer period of
station.
The Mica Weather Board includes temperature,
time, resulting in higher power consumption. Minimizing
photoresistor, barometric pressure, humidity, and passive
start up time yields more power per day to perform other
infrared (thermopile) sensors.
tasks, such as routing and communication. Start up times
The barometric pressure module is a digital sensor man-
for each sensor are listed in Table 1.
ufactured by Intersema. The sensor is sensitive to 0.1 mbar
The unique combination of sensors can be used for a va-
of pressure and has an absolute pressure range from 300 to
riety of aggregate operations. The thermopile may be used
1100 mbar. The module is calibrated during manufacturing
in conjunction with its thermistor and the photoresistor to
and the calibration coefficients are stored in EEPROM per-
detect cloud cover [6]. The thermopile may also be used
sistent storage. The pressure module includes a calibrated
to detect occupancy, measure the temperature of a nearby
temperature sensor to compensate raw barometric pressure
object (for example, a bird or a nest), and sense changes in
readings.
the object’s temperature over time. If the initial altitude is
The humidity sensor is manufactured by General Eastern.
known, the barometer module may be used as an altimeter.
It is a polymer capacitive sensor factory calibrated to within
Strategically placed sensor boards with barometric pressure
1 picofarad (±3% relative humidity). The sensing element
sensors can detect the wind speed and direction by mod-
consists of an electrode metallization deposited over the hu-
elling the wind as a fluid flowing over a series of apertures
midity sensor polymer. The sensor is modulated by a 555
(one such method is described in [3]).
CMOS timer to sense the charge in the capacitor which is
In addition to the sensors on the Mica Weather Board,
filtered through by RC circuit. The resulting voltage is am-
we included an I2C analog to digital converter. Separat-
plified by an instrumentation amplifier for greater sensitivity
ing the ADC from the main Mica processing board provides
over the range of 0% to 100% relative humidity.
greater flexibility in developing components to reduce power
The thermopile is a passive infrared sensor manufactured
consumption. The ADC uses less power than the Atmel pro-
by Melexis.
Heat from black bodies in the sensor’s field
cessor on the Mica, may be used in parallel with processing
of view causes a temperature difference between the ther-
or radio transmission on the Mica, and can be operated in
mopile’s cold junction and the thermopile membrane. The
various low-power and sleep modes. Additionally, The sen-
temperature difference is converted to an electric potential
sor board includes an I2C 8 x 8 power switch permitting
by the thermo-electric effect in the thermopile junctions.
individual components on the board to be turned on or off.
The sensor does not require any supply voltage. The ther-
Each switch can be operated independently of each other –
mopile includes a thermistor in the silicon mass. The ther-
further reducing power consumption.
mistor may be used to measure the temperature of the cold
The Mica Weather Board was designed with interoper-
junction on the thermopile and accurately calculate the tem-
ability in mind. The Mica includes a 51 pin expansion con-
perature of the black body.
nector. The connector has the ability to stack sensor boards
The photoresistor is a variable resistor in a voltage divider
on top of each other.
Instead of allowing each board to
circuit. The divided voltage is measured by the ADC. The
compete for pins on the connector, we developed an access
final temperature sensor is a digital calibrated sensor that
protocol. The Mica will change the value of a switch on the
communicates over the I2C bus. The characteristics of each
sensor board using the I2C bus. Changing the value of the
sensor can be seen in Table 1.
switch triggers the sensor board’s hardware logic to access
The sensors were chosen with great care to ensure high
the Mica’s resources. When a board has access, it may use
interchangeability and high accuracy. Each sensor has less
the power, interrupt, ADC, and EEPROM lines that are di-
than 3% variation when interchanged with others of the
rectly connected to the microprocessor and components on
same model. The accuracy of each sensor is within 3% of
the Mica processing board.
the actual value. Through calibration, the interchangeabil-
4.3
Energy budget
ity and accuracy can be reduced to below 1% depending on
the requirements of the application. Out of the box, the
Many habitat monitoring applications need to run for nine
nodes will be accurate for most applications. Due to the in-
months – the length of a single field season. Mica runs on a
pair of AA batteries, with a typical capacity of 2.5 ampere-
hours (Ah). However we can neither use every drop of en-
ergy in the batteries nor are the batteries manufactured with
identical capacities from batch to batch or from manufac-
turer to manufacturer. We make a conservative estimate
that the batteries will be able to supply 2200 mAh at 3 volts.
Assuming the system will operate uniformly over the de-
ployment period, each node has 8.148 mAh per day available
for use. The application chooses how to allocate this energy
budget between sleep modes, sensing, local calculations and
communications. We note that since different nodes in the
network have different functions, they also may have very
different power requirements. For example, nodes near the
gateway may need to forward all messages from a patch,
whereas a node in a nest may need to merely report its own
readings. In any network, there will be some set of power
limited nodes; when these nodes exhaust their supplies, the
network is disconnected and inoperable. Consequently, we
need to budget our power with respect to the energy bottle-
Figure 3: Acrylic enclosure used for deploying the
neck of the network. To form an estimate of what is possible
Mica mote.
on a Mica mote with a pair of AA batteries, we tabulated
the costs of various basic operations in Table 2.
mounting holes for securing the sensor boards. To provide
Operation
nAh
weather-proofing, we coat the entire sensor package with a
Transmitting a packet
20.000
10 micron parylene sealant, which protects exposed electri-
Receiving a packet
8.000
cal contacts from water.
The sensors remain exposed to
Radio listening for 1 millisecond
1.250
protect their sensitivity. Each coated node is then enclosed
Operating sensor for 1 sample (analog)
1.080
in a transparent acrylic enclosure. The enclosure is venti-
Operating sensor for 1 sample (digital)
0.347
lated to not distort the sensor readings; its primary func-
Reading a sample from the ADC
0.011
tion is to provide additional protection against mechanical
Flash Read Data
1.111
failures and to raise the sensor off the ground. Acrylic pack-
Flash Write/Erase Data
83.333
aging was chosen because it is infrared and radio frequency
transparent, which won’t obstruct sensor readings or wire-
less communication.
Table 2: Power required by various Mica operations.
The acrylic enclosure shown in Figure 3 is used for de-
ploying nodes above the ground on Great Duck Island. The
The baseline life time of the node is determined by the cur-
size of the Mica mote itself was almost too large to fit in pe-
rent draw in the sleep state. Minimizing power in sleep mode
trel burrows; therefore we placed the parylene sealed motes
involves turning off the sensors, the radio, and putting the
into the burrows without enclosures. Not using the enclo-
processor into a deep sleep mode. Additionally, I/O pins on
sure is less robust; we’ve noticed expansion and contraction
the microcontroller need to be put in a pull-up state when-
of connectors over the course of four weeks leading to faulty
ever possible, as they can contribute as much as 100 µA of
electrical connections. We advocate the future use of sol-
leakage current. Mica architecture uses a DC booster to pro-
dered connections to solve this problem.
vide stable voltage from degrading alkaline batteries. With
no load, the booster draws between 200 and 300 µA, depend-
4.5
Patch Gateways
ing on the battery voltage. While this functionality is crucial
Using different gateway nodes directly affects the underly-
for predictable sensor readings and communications, it is not
ing transit network available. We implemented two designs:
needed in the sleep mode. Furthermore, the current draw
an 802.11b single hop with an embedded linux system and
of the microprocessor is proportional to the supply voltage.
a single hop mote-to-mote network.
We modified Mica motes with a Schottky diode, which al-
Initially, we chose CerfCube [1], a small, StrongArm-based
lows us to reliably bypass the DC booster while reducing
embedded system, to act as the sensor patch gateway. Each
the supply voltage in sleep modes. The modification allows
gateway is equipped with a CompactFlash 802.11b adapter.
us to achieve between 30 and 50 µA current draw (battery
Porting functionality to CerfCubes is fairly easy; they run
dependent), which reduces the energy available for tasks to
an embedded version of Linux operating system. Perma-
6.9 mAh per day.
nent storage is plentiful – the gateway can use the IBM
MicroDrive which provides up to 1 GB of storage.
Sup-
4.4
Sensor Deployment
plying adequate power for this device is a challenge, with-
We deployed a wireless sensor network using Mica motes
out power management features this device consumes about
with Mica Weather Boards in July 2002. The network con-
2.5W (two orders of magnitude more than the motes). To
tains all elements of the architecture described in Section 3.
satisfy the CerfCube power requirements, we considered a
To withstand the variable weather conditions on GDI, we
solar panel providing between 60 and 120 Watts in full sun-
designed environmental protective packaging that minimally
light connected to a rechargeable battery with capacity be-
obstruct sensing functionality. Mica motes by their design
tween 50 and 100 Watt-hours (e.g., sealed lead-acid). Re-
are fairly robust mechanically, with the battery case firmly
searchers from Intel Research and JPL have demonstrated
integrated with the main processing and sensor boards, and
delay-tolerant networking using CerfCubes and motes [10]
which will fit very well with the overall system architec-
ture. We deployed the CerfCube with a 12dbi omnidirec-
tional 2.4GHz antenna that provided a range of approxi-
mately 1000 feet.
The mote-to-mote solution consisted of a mote connected
to the base station and a mote in the sensor patch. Both
motes were connected to 14dbi directional 916MHz Yagi an-
tennae. The range of the Yagi antennae is more than 1200
feet. The differences between the mote and the CerfCube
include not only a different communication frequency and
power requirements, but also software components. Of par-
ticular interest to network connectivity is the MAC layer
– the mote’s MAC does not require a bidirectional link like
802.11b. Additionally, the mote sends raw data with a small
packet header (four bytes) directly over the radio as opposed
Figure 4: Thermopile data from a burrow mote on
to overheads imposed by 802.11b and TCP/IP connections.
GDI during a 19-day period (July 18, 2002 to Au-
For one week, we tested the packet reception and power
gust 5, 2002).
consumption of each solution. We discovered that both sys-
tems provide nearly identical packet reception rates, yet the
CerfCube consumed two orders of magnitude more power
iPaq PDA running Linux. The device interfaces with an
and required a larger, more intrusive solar panel. Before
802.11 network deployed at GDI to interact with the local
leaving GDI, we decided to only use the mote solution for
database. Equipped with a GPS unit, users of the gizmo
the gateway due to its power efficiency. Since the network
are able to take notes and tag them with location, time,
has been deployed, there have been no brownouts or power
and other attributes. We’re working on equipping the gizmo
failures observed from the gateway mote.
with a CompactFlash-based MoteNic [14] interface to com-
municate directly with the sensor network.
4.6
Base-station installation
In order to provide remote access to the habitat mon-
5.
CURRENT RESULTS
itoring networks, the collection of sensor network patches
Thirty-two motes are deployed on Great Duck Island, of
is connected to the Internet through a wide-area link. On
which nine are in underground burrows. The sensor network
GDI, we connect to the Internet through a two-way satel-
has been deployed for four weeks as of the writing of this
lite connection provided by Hughes and similar to the Di-
paper. We have calculated that the motes have sufficient
recTV system. The satellite system is connected to a laptop
power to operate for the next six months, even though biol-
which coordinates the sensor patches and provides a rela-
ogists will stop visiting the island in early September. This
tional database service. We had to solve a number of chal-
new data will provide insights into the climate and burrow
lenges to turn a consumer-grade, web-oriented service into
activity through the fall and winter, something previously
a highly reliable general-purpose network connection. The
not possible due to poor off-season weather conditions for
base station needs to function as a turnkey system, since it
island travel.
needs to run unattended. During that time we expect un-
While a complete treatment of the data is beyond the
scheduled system reboots and application failures. At this
scope of this paper, we have made the following observa-
point we have resolved many of the engineering issues sur-
tions. After deployment, the biologists set out to verify the
rounding this problem – shortly after the system boots we
accuracy of the sensor readings. On several occasions, after
can find it on the Internet and access it remotely.
noticing changes in thermopile readings, a recorded petrel
call was played above instrumented burrows. A petrel called
4.7
Database Management System
back to the recording, indicating that a bird was indeed
The base station currently uses Postgres SQL database.
present. Additional readings indicate that the humidity and
The database stores time-stamped readings from the sen-
temperature inside burrows is relatively constant, whereas
sors, health status of individual sensors, and metadata (e.g.,
outdoor motes show significant changes over time. This pat-
sensor locations). The GDI database is replicated every fif-
tern has been seen previously and bolsters confidence in the
teen minutes over the wide-area satellite link to our Postgres
system.
database in Berkeley.
Figure 4 shows occupancy data collected from July 18,
2002 to August 5, 2002. The mote was placed several feet
4.8
User Interfaces
down a burrow tunnel, approximately 1500 feet from the
We expect that many user interfaces will be implemented
lightkeeper’s house on Great Duck Island. The plotted val-
on top of the sensor network database. GIS systems pro-
ues indicate the difference between ambient temperature and
vide a widely used standard for analyzing geographical data.
the object in the thermopile’s field of view. Figure 4 illus-
Most statistics and data analysis packages, such as Matlab,
trates that a petrel left the burrow on July 21st and returned
implement powerful interfaces to relational databases. Fi-
on July 23rd. The petrel left again between July 30th and
nally, we expect a number of web based interfaces, including
August 1st. Variations in the data during times that the
a java applet we developed, to provide the ubiquitous inter-
petrel vacated can be attributed to the ±3oC accuracy of
faces to the habitat data.
the thermopile as well as changing environmental conditions
At this point, the gizmo design for local users is not well
between night and day.
developed. We experimented with a design on a Compaq
Live data from select sensors can be viewed through a Java
applet at http://www.greatduckisland.net. The database
health and status, routing tables, and forwarding network
on the island is replicated to a secondary database at the
messages. These tasks can either be tightly scheduled or
Intel Research Laboratory in Berkeley every 15 minutes. In
run on demand. On one extreme, the system is scheduled
addition to offloading web traffic from the satellite link,
at every level, from TDMA access to the channel, through
replication permits disconnected operation.
When satel-
scheduled adaptation of routes and channel quality. Over-
lite service is unavailable, which has happened several times
head costs are upfront and fixed. A TDMA system is ex-
during periods of severe thunderstorm activity around the
pected to perform well if the network is relatively static.
Hughes network operations center, data continues to be logged
On the other extreme, we use a low-power hailing channel
on the island. When connectivity is restored, the secondary
to create on-demand synchronization between a sender and
database is brought up-to-date.
a receiver. The service overhead is proportional to the use
of the service. This approach can be more robust to unex-
6.
DISCUSSION
pected changes in the network, at the expense of extra cost.
Finally, a hybrid approach is possible, where each service
All of the components in the system must operate in ac-
runs in an on-demand fashion, but the time period for when
cordance with the system’s power budget. As we pointed out
the demand can occur is scheduled on a coarse basis.
in Section 4, each node has a budget of 6.9 mAh per day.
Our deployment at GDI is sending raw data values that
Since the Mica’s processor alone draws approximately 5 mA,
are logged. As the biologists at the College of the Atlantic
we can afford to run the processor for at most 1.4 hours per
analyze the data, we intend to change the data sampling and
day, 5.8% duty cycle if no other operations are performed by
collection according to their needs via network retasking (see
the mote. In a running system, the energy budget must be
Section 6.3).
divided amongst several system services: sensor sampling,
data collection, routing and communication, health moni-
toring and network retasking. Habitat monitoring applica-
6.2
Communications
tions may need other important services in addition to those
Power efficient communication paradigms for habitat mon-
mentioned in this section. These services include localiza-
itoring must include a set of routing algorithms, media ac-
tion, time synchronization, and self configuration described
cess algorithms, and managed hardware access. The routing
by Cerpa et. al. [4].
algorithms must be tailored for efficient network commu-
nication while maintaining connectivity when required to
6.1
Data sampling and collection
source or relay packets.
In habitat monitoring the ultimate goal is data collection;
A simple routing solution for low duty cycle sensor net-
sampling rates and precision of measurements are often dic-
works is simply broadcasting data to a gateway during sched-
tated by external specifications. For every sensor we can
uled communication periods. This method is the most effi-
bound the cost of taking a single sample. By analyzing the
cient – data is only communicated in one direction and there
requirements we can place a bound on the energy spent on
is no dependency on surrounding nodes for relaying packets
data acquisition. We trade the cost of data processing and
in a multihop manner. The routing deployed on GDI is a
compression against the cost of data transmission. We can
hierarchical model. The sensor nodes in burrows are trans-
estimate the energy required by data collection by analyz-
mit only with a low duty cycle – they sample about once
ing data collected from indoor monitoring networks. Let us
per second. The gateway mote is fully powered by solar
consider an experiment where a mote collects a light sam-
power, so it is always on and relaying packets to the base
ple every minute.
The sample is represented as a 16-bit
station. We intend to equip future deployments with en-
integer, but it contains a 10-bit ADC reading. Assuming
ergy harvesting capabilities to allow nodes above ground to
that each packet can carry 25 bytes of payload, unprocessed
perform additional routing tasks with higher duty cycles.
data requires between 72 (if 10-bit samples are used) and
Many of the hard to reach research locations are beyond
116 packets (if 16-bit numbers are used). While this service
the range of a single wireless broadcast from mote to gate-
does not put a burden on the leaf nodes, the routing nodes
way. Accordingly, a multi-hop scheduled protocol must be
near the root may need to retransmit the messages from
used to collect, aggregate, and communicate data.
every leaf in the network, roughly two orders of magnitude
Methods like GAF [15] and SPAN [5] have been used to
more.
Anecdotal evidence presented in Table 3 suggests
extend the longevity of the network by selecting representa-
that this volume of data can be easily reduced by a factor
tives to participate in the network; thereby these algorithms
of 2-4 by applying a delta compression and a standard com-
reduce the average per node power consumption. Although
pression algorithm (e.g., Huffman coding or Lempel-Ziv).
these methods provide factors of 2 to 3 times longer net-
The compression performs even better when applied to a
work operation, our application requires a factor of 100 times
longer run of data. Far better results can be obtained with
longer network operation – recall that our sensor nodes are
signal-specific lossy compression techniques (much like the
on for at most 1.4 hours per day. GAF and SPAN don’t ac-
GSM voice compression schemes). Other methods include
count for infrequent sampling but rather continuous network
distributed compression involving correlating network data
connectivity and operation. Instead, we propose augment-
amongst similar nodes and using Coset codes [12]. Often
ing scheduled multihop routing or low power MAC protocols
the signal model is unknown a priori, but can be obtained
with GAF and/or SPAN to provide additional power sav-
through the analysis of the initial data. We can then use the
ings. GAF and SPAN are independent of communication
network retasking service to program the sensors to commu-
frequency, whereas our application requires increased power
nicate the data of interest.
savings that may be achieved by adjusting the communica-
Once we have allocated the energy for sampling the sen-
tion frequency.
sors and communicating the results, the remaining energy
The research challenge of the routing problem is finding
is devoted to maintaining the network – MAC protocols,
a power efficient method for scheduling the nodes such that
Compression
Huffman
Lempel-Ziv
Burrow-Wheeler
Uncompressed
algorithm
(pack)
(gzip)
(bzip2)
8-bit sample
1128
611
681
1365
10-bit sample
1827
1404
1480
1707
16-bit sample
2074
1263
1193
2730
8-bit difference
347
324
298
1365
10-bit difference
936
911
848
1707
16-bit difference
839
755
769
2730
Table 3: Compression characteristics of typical indoor light signal. We estimate the amount of information
contained within the signal by compressing various signal representations with the standard Unix compression
utilities.
long multihop paths may be used to relay the data. We pro-
interesting areas. Most of the time such updates can be en-
pose the following approaches for scheduled communication:
capsulated in network maintenance packets. More complex
• After determining an initial routing tree, set each mote’s
functionality adjustment may be implemented through vir-
level from the gateway. Schedule nodes for communi-
tual machines like Mat´
e [13]. Virtual machine-based retask-
cation on adjacent levels starting at the leaves. As
ing seems ideal when the much of the underlying function-
each level transmits to the next, it returns to a sleep
ality is implemented through underlying native functions,
state. The following level is awaken, and packets are
as is the case in making routing decisions, or processing
relayed for the scheduled time period.
The process
data through a predefined set of filters. Virtual machine
continues until all levels have completed transmission
programs can be fairly small (many fit in a single packet).
in their period. The entire network returns to a sleep
Finally, the entire code image running on a mote may be
mode. This process repeats itself at a specified point
replaced with a new one. One would use this method when
in the future.
a drastic retasking of the application is necessary; for exam-
ple if it were necessary to install a new signal-specific com-
• Instead of a horizontal approach, awaken nodes along
pression algorithm to cope with the volume of data. The
paths or subtrees in a vertical approach. Each subtree
reprogramming process is quite costly – it involves reliably
in turn completes their communication up the tree.
transmitting the binary image of the code (approximately
This method is more resilient to network contention;
10kb) to all nodes that need to be reprogrammed, and in-
however the number of subtrees in the network will
voking a reprogramming application which runs the node
likely exceed the number of levels in the network and
for 2 minutes while drawing about 10 mA. To relate this to
subtrees may be disjoint allowing them to communi-
the energy budget: we can afford to reprogram the nodes
cate in parallel.
every day during the 9 month life cycle if reprogramming
Alternatively, we have experimented with using low power
is the node’s only task. While significantly more expensive
MAC protocols. By determining our duty cycle, we can cal-
in absolute terms than virtual machine reprogramming, it
culate the frequency with which the radio samples for a start
can pay off over the period of a few days since native code
symbol. By extending the start symbol when transmitting
executes more efficiently.
packets, we can match the length of the start symbol to the
6.4
Health and Status Monitoring
sampling frequency. Other low power MAC protocols, such
as S-MAC [16] and Aloha with preamble sampling [7] em-
A major component of use to the application is one that
ploy similar techniques that turn off the radio during idle
monitors the mote’s health and the health of neighboring
periods to reduce power consumption. The difference be-
motes. Health and monitoring is essential for a variety of
tween scheduled communication and low power MACs is
purposes; the most obvious is retasking. The duty cycle of
instead of having a large power and network overhead to
a mote may be dynamically adjusted to alter its lifetime.
set up a schedule, the overhead is distributed along the life-
A simple monitoring implementation is deployed on GDI.
time of the node. Both approaches are equivalent in power
Each mote periodically includes their battery voltage level
consumption, the decision for which to use depends on the
with the sensor readings they are transmitting. The volt-
end-user interactivity required by the application. A poten-
age is represented as a one byte value in the range of 0 to
tial tradeoff of using a low power MAC is that transmitted
3.3 volts. Adding voltage measurements has greatly assisted
packets potentially wake up every node within the cell. Al-
us in remote analysis of node failures.
though early rejection can be applied, scheduling prevents
Health and status messages sent to the gateway can be
unneeded nodes from wasting power processing a packet’s
used to infer the validity of the mote’s sensor readings.
headers.
Although the health messages are not critical for correct
application execution, their use can be seen as preventive
6.3
Network Retasking
maintenance.
For this reason, we advocate a health and
As the researchers refine the experiment, it may be nec-
monitoring component that transmits status messages with
essary to adjust the functionality of individual nodes. This
lower latency in exchange for strict reliability. Health mes-
refinement can take several different forms. Scalar param-
sages may be sent rather infrequently (about once per hour
eters, like duty cycle or sampling rates, may be adjusted
or less dependent on the duty cycle) with no guarantee on
through the application manager. Even such simple adjust-
their delivery.
ment allows the researchers to focus their efforts in more
7.
CONCLUSION
for wireless communications technology. In 2001 ACM
Habitat and environmental monitoring represent an im-
SIGCOMM Workshop on Data Communications in
portant class of sensor network applications. We are col-
Latin America and the Caribbean, San Jose, Costa
laborating with biologists at the College of the Atlantic to
Rica, Apr. 2001.
define the core application requirements. Because end users
[5] B. Chen, K. Jamieson, H. Balakrishnan, and
are ultimately interested in the sensor data, the sensor net-
R. Morris. Span: An energy-efficient coordination
work system must deliver the data of interest in a confidence-
algorithm for topology maintenance in ad hoc wireless
inspiring manner. The low-level energy constraints of the
networks. In Proceedings of the 7th ACM International
sensor nodes combined with the data delivery requirements
Conference on Mobile Computing and Networking,
leave a clearly defined energy budget for all other services.
pages 85–96, Rome, Italy, July 2001.
Tight energy bounds and the need for predictable opera-
[6] R. W. Clay, N. R. Wild, D. J. Bird, B. R. Dawson,
tion guide the development of application architecture and
M. Johnston, R. Patrick, and A. Sewell. A cloud
services.
monitoring system for remote sites. Publications of the
While we believe GDI is representative of many applica-
Astronomical Society of Australia, 15(3):332–335,
tions in this domain, there may be significant differences.
Aug. 1998.
To evaluate our implementation, we have deployed an ini-
[7] A. El-Hoiydi. Aloha with preamble sampling for
tial prototype network at the James San Jacinto Mountains
sporadic traffic in ad hoc wireless sensor networks. In
Reserve (JMR) (33.48N, 116.46W) in Idyllwild, California.
Proceedings of IEEE International Conference on
JMR is a 29 acre ecological preserve, representing just one
Communications, New York, NY, USA, Apr. 2002.
of the University of California System Natural Reserve Sys-
[8] D. Estrin, L. Girod, G. Pottie, and M. Srivastava.
tem’s 34 land holdings. The deployment uses a basic sensor
Instrumenting the world with wireless sensor
package instead of the Mica Weather Board. JMR’s climate
networks. In International Conference on Acoustics,
is significantly different from GDI; it is arid and weather
Speech, and Signal Processing (ICASSP 2001), Salt
changes occur over long periods of time.
Researchers at
Lake City, UT, May 2001.
JMR are interested in microclimate readings over a large
[9] D. Estrin, R. Govindan, J. S. Heidemann, and
area as opposed to animal monitoring on GDI.
S. Kumar. Next century challenges: Scalable
Our practical experience with sensor network deployment
coordination in sensor networks. In Mobile Computing
will guide the creation of a habitat monitoring kit. This kit
and Networking, pages 263–270, 1999.
will be made available to scientists and researchers in other
[10] K. Fall. Delay-tolerant networking for extreme
fields. Users will be able to tailor the mote’s operation to a
environments. http:
variety of experimental setups, which will allow scientists to
//www.cs.berkeley.edu/~kfall/extreme-talk.pdf,
reliably collect data from locations previously unaccessible
Nov. 2001. Presentation at UCSD.
on a micro-measurement scale.
[11] J. Hill and D. Culler. A wireless embedded sensor
architecture for system-level optimization. In UC
8.
ACKNOWLEDGMENTS
Berkeley Technical Report, 2002.
This work was supported by the Intel Research Labora-
[12] J. Kusuma, L. Doherty, and K. Ramchandran.
tory at Berkeley and by DARPA grant F33615-01-C-1895
Distributed compression for wireless sensor networks.
(Network Embedded Systems Technology “NEST”). This
In Proceedings of ICIP 2001, Thessalonika, Greece,
work was done in conjunction with the University of Califor-
Oct. 2001.
nia Center for Information Technology Research in the Inter-
[13] P. Levis and D. Culler. Mat´
e: A tiny virtual machine
est of Society (CITRIS). We would like to thank Eric Paulos,
for sensor networks. In International Conference on
Ken Smith, and David Koizumi for designing and manufac-
Architectural Support for Programming Languages and
turing the acrylic enclosures, Phil Buonodonna for his net-
Operating Systems, San Jose, CA, USA, Oct. 2002. To
working expertise, Andrew Peterson and Amanda Witherell
appear.
for boat and travel logistics to Great Duck Island, and the
[14] T. Stathopuolos. MoteNIC: Overview.
College of the Atlantic for the use of Great Duck Island.
http://lecs.cs.ucla.edu/Noteworthy/quadcharts/
thanos_lecs.ppt, Feb. 2002.
9.
REFERENCES
[15] Y. Xu, J. Heidemann, and D. Estrin.
[1] Cerfcube embedded StrongARM system.
Geography-informed energy conservation for ad hoc
http://www.intrinsyc.com/products/cerfcube/.
routing. In Proceedings of the ACM/IEEE
Intrinsyc Corporation, Vancouver BC Canada.
International Conference on Mobile Computing and
[2] J. G. T. Anderson. Pilot survey of mid-coast Maine
Networking, pages 70–84, Rome, Italy, July 2001.
seabird colonies: an evaluation of techniques. Bangor,
[16] W. Ye, J. Heidemann, and D. Estrin. An
ME, 1995. Report to the State of Maine Dept. of
energy-efficient mac protocol for wireless sensor
Inland Fisheries and Wildlife.
networks. In Proceedings of the 21st International
[3] J. D. Barrick, J. A. Ritter, C. E. Watson, M. W.
Annual Joint Conference of the IEEE Computer and
Wynkoop, J. K. Quinn, and D. R. Norfolk.
Communications Societies (INFOCOM 2002), New
Calibration of NASA turbulent air motion
York, NY, USA, June 2002.
measurement system. NASA Technical Paper 3610,
Langley Research Center, Dec. 1996.
[4] A. Cerpa, J. Elson, D. Estrin, L. Girod, M. Hamilton,
and J. Zhao. Habitat monitoring: Application driver

Document Outline