An Introduction to
Wireless Sensor Networks On an uninhabited island off
the coast of Maine, tiny wireless sensors deep in the burrows of
mysterious sea birds monitor the environmental factors affecting the
shy creatures' comings and goings. In one of Intel’s chip
fabrication facilities, similar sensors measure the subtle
vibrations of various machines to detect malfunctions before the
equipment breaks down. At an Air Force base across the country,
dozens of small sensors scattered across a bogus battlefield
outperform tripwires, without the wires. Meanwhile, at the
University of California at Berkeley, sensors embedded in a mock
building’s walls diagnose its seismic stability after a simulated
earthquake.
Experimental sensor networks like these are
opening up new vistas for scientists and engineers to observe
physical phenomena and react to it. The building blocks of these
wireless networks are “motes,” self-contained, battery-powered
computers that measure light, temperature, humidity, and other
environmental factors. Developed by Intel Research in collaboration
with the University of California at Berkeley-based Center for
Information Technology Research in the Interest of Society (CITRIS),
the motes self-organize into ad-hoc wireless networks. The data is
then relayed from mote to neighboring mote until it reaches its
desired destination for processing.
Sensor networks are a
giant leap toward "proactive computing", a paradigm where computers
anticipate human needs and, if necessary, act on our behalf. Instead
of shuttling data between the real world and machines, the human is
at the top, reaping the benefits of ubiquitous computers. Sensor
networks and proactive computing has the potential to improve our
productivity and enhance safety, awareness, and efficiency at the
societal scale.
There are many technological hurdles that
must be overcome for ad hoc sensor networks to become practical
though. Nearly everything we take for granted in desktop computing
is at a premium in wireless sensor networks. The individual motes
are incredibly resource constrained. They have limited processing
speed, storage capacity, and communication bandwidth. Their lifetime
is determined by their ability to conserve power. All of these
constraints require new hardware designs, software applications, and
network architectures that maximize the motes’ capabilities while
keeping them inexpensive to deploy and maintain. Additionally, it
must be easy for non-computer scientists with just basic training in
the technology to extract meaningful real world data from the
networks. Finally, pervasive sensor networks raise non-trivial
security and privacy issues that call for collaboration between
engineers, social scientists, legislators, and policymakers.
Intel Research, working with the academic community and
industry, is addressing many of these significant challenges.
Already, a broad spectrum of sensor network pilot applications have
been demonstrated-- from smart home systems that improve the quality
of life for the elderly, to sensors that measure the structural
health of the Golden Gate Bridge. And this is just the beginning. As
sensor network technology emerges from research laboratories, the
ability to instrument the world is likely to transform every facet
of our lives.
In today’s model of computing, we
interact directly, one-on-one, with our desktop PCs, mobile phones,
and personal digital assistants. In the near future though, the
majority of computers will be embedded deep in the world around us,
hidden inside our homes, roads, farms, hospitals, and factories.
When we are in control of hundreds or thousands of computers each,
it will be impossible for us to interact directly with each one. The
time has come to transition from interactive to proactive computing.
These proactive computers will anticipate our needs and sometimes
act on our behalf. Sensor networks represent this paradigm shift in
computing.
On a farm, sensors buried in the soil could help
manage irrigation and fertilization. Smart smoke detectors will
guide firefighters through a building to trapped victims. Motes in
your home could monitor temperature and energy use to automatically
create comfortable microclimates while cutting your utility bill.
Of course, sensing and measuring itself is not new.
Engineers have been developing increasingly versatile and sensitive
sensors for many years. Traditionally, the cost and bulk of sensing
technology meant that only a handful of sensors could be deployed
for most applications. However, data averaged from a few sensors
does not truly represent the real world. How can computers
anticipate our needs if they can’t understand our environment? Now,
thousands of sensors can be scattered throughout a single physical
space, providing a much higher resolution picture of the real world
than ever before. This entirely new approach to instrumentation is
made possible by the intersection of several technological trends.
According to Moore's Law, the number of transistors on a
microchip doubles approximately every two years, leading to faster
and more powerful computers on our desktops with each generation. At
the same time, microprocessors with a given computing capacity are
becoming smaller and cheaper with every passing year. While silicon
scaling marches on, the same semiconductor manufacturing processes
are being utilized to build microscopic mechanical structures that
interact with the physical world. This technology, called MEMS
(microelectromechanical systems), enables the production of velocity
sensors, thermometers, and even low-power radio components that fit
on the head of a pin and cost just pennies each. These three
hardware ingredients-- microprocessors, MEMS sensors, and low-power
radios--make up a sensor node, or "mote". The "mote" nickname comes
from UC Berkeley’s Smart Dust project, an effort funded by the
Defense Advanced Research Projects Agency’s (DARPA) Network Embedded
Software Technology (NEST) program to shrink the devices down to
dust mote size through the power of Moore’s Law.
While the
motes’ low cost and small size are clearly desirable traits, they're
not sufficient on their own to open up a wide spectrum of new sensor
applications. Rather, it is the motes' ad-hoc, multi-hop networking
capabilities that make it possible to deploy larger networks of
these devices than ever before. This provides sensing closer to the
physical phenomena and with a higher granularity than previously
possible. Additionally, novel software enables the raw data
collected by the sensors to be analyzed in various ways before it
even leaves the network. After all, humans want information from
their proactive computers, not numbers.
Applying advanced networking technology
to mass-produced wireless sensors yields a new kind of "instrument".
The network literally becomes the sensor.
Researchers in
academia and industry have already deployed dozens of sensor network
pilot applications. The following is just a small sampling of those
diverse projects:
A robust sensor network on Great
Duck Island off the coast of Maine aids biologists in the
study of Leach's storm petrels, a species of seabird that have
mysteriously selected this locale as their breeding ground. (Intel
Research/UC Berkeley)
As part of the DARPA
NEST program, researchers demonstrated a sensor network at
MacDill Air Force Base that can detect, classify, and track
soldiers and vehicles in difficult-to-monitor open spaces such as
desert battlefields. (Ohio State University)
A sensor network deployed in an Oregon
vineyard guides irrigation and planting, increasing crop
yield. (Intel Research/ King Family Farms/AgCanada)
Inside an experimental
smart home at Intel’s Oregon campus, a sensor network is under
development that could someday keep tabs on an Alzheimer’s
patient’s vital signs while reminding him how to warm up his
lunch. (Intel Research) On the San
Andreas Fault, a network of motes equipped with seismometers
calculate the depth of the fault, locate accumulating stress, and
may eventually improve earthquake prediction. (UCLA Department of
Earth and Space Sciences/ Center for Embedded Networked Sensing)
Motes mounted in the treetops of UC
Botanical Garden’s Mather Redwood Grove sample environmental
data in a cross section of the canopy to help scientists
understand the massive plants' physiology. (UC Berkeley/Intel
Research)
Motes that measure vibration signatures on manufacturing
equipment are being tested for "pre-emptive
maintenance applications" to reduce downtime in
semiconductor fabrication facilities. (Intel Research/Intel
Technology and Manufacturing Group)
While the particular size, type, and
configuration of motes that form a network are mostly determined by
the intended application, all of the devices face the same
overarching design constraint. A mote is only as effective as its
ability to conserve power. Ideally, each mote should be able to
survive on its own for at least a year on a pair of AA batteries.
Yet each reading a mote takes and every bit of data it transmits
brings the device a moment closer to death. To that end, motes must
be on a strict power diet.
At its core, this diet is based
on enabling the motes to run at extremely low duty cycles. The mote
is active as little as one percent of the time. It “wakes up” only
to take scheduled readings or to transmit or receive data from
neighboring devices. Every one of the mote’s hardware and software
components is designed to support low duty cycles.
As
semiconducting circuits become smaller, they consume less power.
Simple microcontrollers like those that function as a mote’s brain
can operate with just a milliwatt of power when active, or 1-10
microwatts in standby mode. A mote’s memory must also be limited due
to the energy constraints. Each mote typically has less than 10
kilobytes of RAM, one hundred kilobytes of software, and a megabyte
of data storage. All told, that’s approximately 10,000 times less
data storage than a desktop PC.
The low power approach is
continued through a mote’s sensing system. For example,
commercially-available macroscale sensors such as thermistors and
fog detectors show a change in voltage as, respectively, they get
warmer or wetter. Analog to digital converters (ADCs) translate that
voltage into a zero or one that the microprocessor can understand.
The development of extremely efficient ADCs keep the power profile
of a mote's sensing system similar to that of the processor.
Meanwhile, MEMS provide the motes with a much broader array
of low-power sensory inputs. The simplest example of a MEMS device
resembles a diving board with a mass mounted on the end.
Gravitational forces or acceleration cause the mass to spring up and
down, forces that can easily be converted into a digital signal.
These devices, called accelerometers, are commonly used in
automobiles to trigger the release of airbags. A growing variety of
MEMS sensors are available to detect myriad factors, from the body
heat of a bird in its burrow to the presence of environmental
contaminants in the air. Intel Research is also developing biochips,
devices that can sense biological materials and organic chemistry.
While commercial sensors are already present in such
everyday products as automobiles and washing machines, motes boast
one essential capability that sets them apart from their
predecessors: wireless networking using radio. Low-power
transceivers enable the motes to transmit their sensor readings
throughout the network. Like MEMS sensors, these low power radios
can now be inexpensively produced using conventional silicon
processing techniques. This new class of RF (radio frequency)
devices is one of the key enabling technologies behind 802.11 (WiFi)
networks, Internet-enabled PDAs, ever-smaller mobile phones, and
sensor networks.
Currently, consumer AA or "coin" batteries
can keep motes alive for six months to one year. Other energy
scavenging power sources are also being developed. Ambient lighting
or sunlight could provide enough solar energy in applications where
the motes are exposed. At an earlier stage of development are MEMS
devices demonstrated at UC Berkeley that convert the ambient
vibration of structural components like air-conditioning ducts and
windows into enough electricity to keep the motes operational
indefinitely.
Several species of motes based on the
prototypes developed by Intel Research and UC Berkeley have recently
become commercially available at $50-$100 each. Through
re-engineering, Moore’s Law, and volume production, motes are
expected to drop in price to less than $5 each over the next five
years.
Crossbow
Technology Inc. was the first commercial manufacturer of motes.
Their latest generation of devices consists of a microprocessor,
memory, storage, and an internal analog-to-digital converter, all
integrated into a device roughly the size of a quarter. Various
sensor boards for measuring acceleration, magnetism, light,
temperature, and other factors can easily be snapped on to the
processor/radio.
Crossbow is also licensed to produce
Intel’s Stargate single-board gateway computer, based on the Intel
XScale® technology. These high-end nodes can improve sensor network
performance and reduce the motes’ energy consumption by offloading
some of the wireless responsibilities to devices that can be plugged
into power sources. The idea behind this kind of network is
strikingly simple. For example, a network of high bandwidth 802.11
(WiFi) gateways like Stargate could overlay a mote-based sensor
network. The structure is analogous to a highway overlaid on a
roadway system. Sensor data can then enter and exit the 802.11
highway at multiple interchanges (the Intel XScale technology
gateways) in order to bypass the side roads, the wireless motes.
This approach increases bandwidth and requires less energy on
average because the motes are not solely responsible for moving data
through the network. The resulting system is known as a
heterogeneous sensor network.
To cater to applications that
require more processing and bandwidth--vibration, audio, or image
sensing, for example--Intel Research developed the Intel® Mote,
featuring a 32-bit central processing unit and the Bluetooth
wireless standard. Bluetooth is commonly used in mobile phones and
laptop computers for short-distance communication. As a result,
users can employ these kinds of devices to easily interact with
Intel Mote-based sensor networks.
Intel Mote prototype (original size:
3x3 cm)
In the laboratory, motes are
continuing to shrink in size. As part of the Smart Dust project, UC
Berkeley researchers have shrunk the processing and wireless
functionality of the larger motes onto a single chip just a few
millimeters on a side, not including batteries. Once the incredibly
efficient radio technology behind this “Spec” mote is ready for
prime time, it will be commercialized by Dust Inc., a sensor
networking start-up spun out of the Smart Dust research effort.
Like all computers, motes require an
operating system that manages all of the hardware and software
functionality. However, commercial operating systems--Unix or
Windows, for instance--require far too much processing power and
storage space than a mote can offer. Over the last several years, UC
Berkeley researchers have designed and honed an operating system
specifically for embedded networks. Freely-available and open
source, TinyOS
has become the "industry standard" operating system for sensor
network research and applications.
At its most basic level,
TinyOS is a scheduler that manages the activities of its various
modular components. First and foremost, the operating system is the
final governor of power on the mote. Wireless communication is a
notoriously power hungry activity. And unlike mobile phones and
laptop computers, motes can't be recharged every night. In order to
keep sensor networks to alive for long periods, the motes are
programmed to pass their data bucket-brigade style from node to
node, much like packets travel through routers in the Internet. This
multi-hop approach keeps the radio’s power requirements to a
minimum. TinyOS also enables the motes to process some data locally
and only communicate the results of that processing when an
interesting event is detected.
TinyOS is also the architect
of the ad hoc wireless network. The software enables each mote to
discover its neighbors and perform an algorithm in concert with its
peers to determine how data should be routed through the network.
Finally, TinyOS is an excellent multi-tasker. It juggles the streams
of data .owing in from the sensors and the network and plays traffic
cop, directing the transmission of data to other nodes.
Once
loaded with TinyOS, the motes must maintain their flexibility in the
field. New applications may be developed over the course of a sensor
network’s life. For example, a biologist drawing insight from a
network deployed high in a tree canopy may decide six months later
that he or she would like to monitor sound as well as motion. The
sheer number of nodes makes it impossible to reprogram each one
individually. Instead, new programs are introduced into the network
similarly to the way computer viruses spread across the Internet. As
the motes communicate, they infect their kin with the new operating
instructions. The programs then run inside an easily accessible and
manageable “virtual machine,” software within TinyOS that simulates
a separate hardware computer. The beauty of TinyOS is that even with
all of these capabilities, the entire operating system is just a few
kilobytes in size.
TinyOS is written in NesC,
a programming language for motes developed at Intel Research and UC
Berkeley. An extension of the popular programming language C, NesC
(pronounced "NES-see") is a natural lingua franca for motes. Motes
are a unique species of computer, primarily because they’re asleep
most of the time. That means their processing is event driven,
occurring only when the sensors acquire data or a new message
arrives. NesC supports the motes’ reactivity to their environment
with a component model that simplifies the creation of applications
and the aggregation of data.
For example, a certain
application might require a mote’s average temperature over a period
of time. With nesC, the timer and averaging modules are two of many
reusable software components that sit between the hardware sensor
and a particular application. Depending on the task, various modules
can be “wired” together as necessary. Sensor readings can also be
aggregated as the data is routed through the network. This approach
reduces the amount of information that each mote transmits, thereby
conserving power.
Still, in a massively distributed sensor
network, numerous activities will often occur simultaneously and
must not overwhelm the motes’ limited power, processing, and memory
resources. Even with these tight constraints, NesC’s elegant
concurrency model enables the motes to be programmed to handle many
events in parallel. The NesC Compiler provides additional aid for
programmers exploiting the concurrency and component model by
detecting any potential problems in new software before it is
deployed.
Sensor network researchers around the
world have developed numerous applications using nesC and TinyOS.
For heterogeneous sensor networks to be widely deployed though, it
must be relatively simple to extract meaningful data from the
networks. Otherwise, gathering data from a sensor network would be
akin to drinking from a fire hose. TinyDB,
a database system developed at Intel Research and UC Berkeley, is
one such solution. Essentially, running TinyDB transforms diverse
kinds of sensor networks into user-friendly virtual databases rich
with useful information about the real world.
Consider this
scenario: A large corporate campus is equipped with motes in every
room keeping a constant vigil on light and sound. One use for the
sensor network would be to locate unoccupied conference rooms by
checking for noise or if the lights are on. Without an application
like TinyDB running on TinyOS, the administrator of the sensor
network would have to write several hundred lines of computer code
to collect the light and sound information from every mote,
coordinate how the data is aggregated, and forward all of the
information to a PC that determines which rooms are occupied. Once
written, the software would have to be installed in every mote
across the campus.
TinyDB greatly streamlines the process by
enabling a user to gather that same information just by posing a
simple query in SQL, a common database language. Through a graphical
user interface, the software describes what sensor readings are
available. Meanwhile, TinyDB’s declarative query language enables
the user to describe the desired data – the average noise level, for
example – without having to tell the software how to acquire that
data. The query is then sent to the TinyDB query processor
pre-installed on each mote. If a mote happens to be relaying a
message related to an unfamiliar query, it simply asks the
neighboring mote that sent the message for a copy of the query so it
too can help gather the data.
Once a query is executed,
TinyDB automatically extracts the data from the network and dumps
into a traditional database. For example, in the case of the
seabird-monitoring sensor network, the data hops across the island
to a lighthouse where it’s relayed via satellite to a Web site. The
information can then be analyzed using standard tools and
visualization techniques.
A new era of computing is on the
horizon. The vision of proactive computing calls for billions,
perhaps even trillions, of devices to be deeply embedded within our
physical environment. These tiny sensors and actuators will silently
serve us, acquiring and acting on a multitude of data to improve our
lives, help us understand our world, and make us more productive. Of
course, the ability to instrument our world poses complex questions
about security and privacy. Fortunately, these are becoming active
areas of cross-disciplinary research in academia and industry.
Addressing those concerns now is crucial. In the
not-so-distant future, motes will evolve into devices that will not
be embedded, but actually part and parcel of everyday objects.
Structures will be held together with “smart bolts” that contain
sensors and radios. Buildings will be constructed from “smart
girders” that feed seismic stability ratings to a .at screen near
the entrance. Trellis stakes will monitor the crops that they
support, while gaining energy from the sun. Instrumented watches,
teapots, and bathtubs will enable our elders to enjoy their lives
while easing the minds of their caretakers.
That is the
future of proactive computing. And sensor network research is
driving us toward it.