Content uploaded by Alanson P Sample
Author content
All content in this area was uploaded by Alanson P Sample on Mar 12, 2015
Content may be subject to copyright.
Revisiting Smart Dust with RFID Sensor Networks
ABSTRACT
We argue that sensing and computation platforms that
leverage RFID technology can realize “smart-dust” ap-
plications that have eluded the sensor network commu-
nity. RFID sensor networks (RSNs), which consist of
RFID readers and RFID sensor nodes (WISPs), extend
RFID to include sensing and bring the advantages of
small, inexpensive and long-lived RFID tags to wireless
sensor networks. We describe sample applications suited
to the space between existing sensor networks and RFID.
We highlight the research challenges in realizing RSNs
such as the use of intermittent power and RFID proto-
cols suited to sensor queries.
1 INTRODUCTION
In the late 1990s, the vision of “smart-dust” was articu-
lated by the research community. This vision was pred-
icated on advances in microelectronics, wireless com-
munications, and microfabricated (MEMS) sensing that
were enabling computing platforms of rapidly diminish-
ing size. The early proponents imagined devices one cu-
bic millimeter in size with capabilities sufficient to power
themselves, sense the environment, perform computa-
tion, and communicate wirelessly [9]. Large-scale de-
ployments of such devices would enable a wide range
of applications such as dense environmental monitoring,
sensor rich home automation and smart environments,
and self-identification and context awareness for every-
day objects.
The past decade has seen significant effort and
progress towards the original motivating applications. In
particular, wireless sensor networks (WSNs) based on
“mote” sensing platforms have been applied to many
real-world problems. Remote monitoring applications
have sensed animal behavior and habitat, structural in-
tegrity of bridges, volcanic activity, and forest fire dan-
ger [6], to name only a few successes. These net-
works leveraged the relatively small form-factor (ap-
proximately 1” x 2”) of motes and their multihop wire-
less communication to provide dense sensing in difficult
environments. Due to their low power design and care-
ful networking protocols these sensor networks had life-
times measured in weeks or months, which was generally
sufficient for the applications.
Despite this success, WSNs have fallen short of the
original vision of smart-dust. They have not led to an
approximation of sensing embedded in the fabric of ev-
eryday life, where walls, clothes, products, and personal
items are all equipped with networked sensors. For this
manner of deployment, truly unobtrusive sensing devices
are necessary. The size and finite lifetime of motes make
them unsuitable for these applications.
We argue in this paper that Radio Frequency Identifi-
cation (RFID) technology has a number of key attributes
that make it attractive for smart-dust applications. Pas-
sive UHF RFID already allows inexpensive tags to be re-
motely powered and interrogated for identifiers and other
information at a range of more than 30 feet. The tags can
be small as they are powered by the RF signal transmitted
from a reader rather than an onboard battery; aside from
their paper thin antennas, RFID tags are approximately
one cubic millimeter in size. Moreover, their lifetime can
be measured in decades as they are reliable and have no
power source which can be exhausted. These advantages
have resulted in the widespread deployment of RFID for
industrial supply-chain applications such as tracking pal-
lets and individual items. However, RFID technology is
limited to only identifying and inventorying items in a
given space.
The RFID Sensor Networks (RSNs) we advocate in
this paper extend RFID beyond simple identification to
in-depth sensing. This combines the advantages of RFID
technology with those of wireless sensor networks. In
our previous work, we have demonstrated the techni-
cal feasibility of building small, battery-free devices that
use the RFID PHY and MAC layer to power themselves,
sense, compute, and communicate; we refer to these de-
vices as Wireless Identification and Sensing Platforms
(WISPs)[15, 16]. While other research efforts such as
[3] have combined RFID with sensing, to the best of our
knowledge, the Intel WISP is the only RFID sensor node
with computational capabilities and that operates in the
long range UHF band.
While the feasibility of WISPs has been established
by this earlier work, how to harness many such devices
to create RSNs is an open question. An RFID sensor net-
work consists of multiple WISPs and one or more read-
ers. Consequently, realizing full-scale RSNs will require
development at both the WISP and the reader, as new
protocols and techniques must be developed unlike those
of either RFID or WSNs.
The focus of this paper is the applications that RSNs
enable and the systems challenges that must be overcome
for these to be realized. As the traditional RFID usage
1
Figure 1—Commercial UHF RFID tag, Accelerometer WISP, Telos
mote with batteries
model is very different from that of WSNs, RSNs face
substantial challenges when trying to integrate the two
technologies. For example, unlike WSNs, RSNs must
cope with intermittent power and unlike RFID must sup-
port sensor queries rather than simply identification.
2 FROM MOTES AND RFID TO RSNS
Two technologies have been widely used to realize real-
world monitoring applications: wireless sensor networks
via motes, and RFID via standard tags and readers. We
describe and contrast each technology and then present
their combination (Table 1) as RFID sensor networks
(RSNs). We use prior work on the WISP [15, 16] to
demonstrate the technical feasibility of this combina-
tion. Representative devices for the three technologies
are show in Figure 1.
2.1 Wireless Sensor Networks (Motes)
Currently, most WSN research is based on the Telos
mote [13], which is a battery powered computing plat-
form that uses an integrated 802.15.4 radio for commu-
nication. These motes are typically programmed to orga-
nize into ad-hoc networks [19] and transmit sensor data
across multiple hops to a collection point. To extend
network lifetime, motes duty cycle their CPU and radio
(e.g., with low-power listening [12]), waking up inter-
mittently to sense and communicate. With a duty cycle
of 1% motes can have a lifetime of up to three years be-
fore the batteries are exhausted.
The multihop communication paradigm of WSNs
means that they can be used to extend sensing capabil-
ities to great distances. This has made them ideal for a
wide range of sensing applications. However, the large
size of the mote and its finite lifetime makes it unsuit-
able for applications where sensing must be embedded
in small objects, or in inaccessible locations where bat-
teries cannot be replaced.
2.2 RFID
While there are a number of different RFID specifica-
tions, that of greatest interest for sensing applications
is the EPCglobal Class-1 Generation-2 (C1G2) proto-
col [4], as it is designed for long-range operation. The
C1G2 standard defines communication between RFID
readers and passive tags in the 900 MHz Ultra-High Fre-
quency (UHF) band, and has a maximum range of ap-
proximately 30 feet. A reader transmits information to a
tag by modulating an RF signal, and the tag receives both
down-link information and the entirety of its operating
energy from this RF signal. For up-link communication,
the reader transmits a continuous RF wave (CW) and the
tag modulates the reflection coefficient of its antenna. By
detecting the variation in the reflected CW the reader is
able to decode the tag response. This is referred to as
“backscattering”, and requires that a tag be within range
of a powered reader.
The MAC protocol for C1G2 systems is based on
Framed Slotted Aloha [14], where each frame has a num-
ber of slots and each tag will reply in one randomly se-
lected slot per frame. Before beginning a frame, a reader
can transmit a Select command which limits the number
of active tags by providing a bit mask, as only tags with
ID’s (or memory locations) that match this mask will re-
spond in the subsequent round. When a tag replies in
a slot, the reader can choose to singulate the tag. After
singulation the reader can read and write values to tag
memory. These mechanisms enable rapid identification
of tags and unicast read and write.
RFID tags are fixed function devices that typically use
a minimal, non-programmable state machine to report a
hard-coded ID when energized by a reader. As they are
powered by the reader the device itself can be very small,
though the antenna requires additional area. As the an-
tenna is flexible and paper thin, their small size means
they can be affixed to virtually any object and the object
can then be identified. However, RFID tags provide no
general purpose computing or sensing capabilities.
2.3 RFID sensor networks (WISPs + readers)
We define RFID sensor networks (RSNs) to consist
of small, RFID-based sensing and computing devices
(WISPs), and RFID readers that are part of the infras-
tructure and provide operating power. RSNs bring the
advantages of RFID technology to wireless sensor net-
works. While we do not expect them to replace WSNs
for all applications, they do open up new application
spaces where small form-factor, long-lived, or inacces-
sible devices are paramount. Our hope is that they will
elegantly solve many sensor network applications, e.g.,
home sensing and factory automation where installing
readers is feasible.
Prior work at Intel Research demonstrates that WISPs
can be built today. The most recent Intel WISP is a wire-
less, battery-free platform for sensing and computation
that is powered and read by a standards-compliant UHF
2
CPU Sensing Communication Range Power Lifetime Size (inches)
WSN (Mote) Yes Yes peer-to-peer Any battery <3 yrs 3.0 x 1.3 x .82 (2.16 in3)
RFID tag No No asymmetric 30 ft harvested indefinite 6.1 x 0.7 x .02 (.08 in3)
RSN (WISP) Yes Yes asymmetric 10 ft harvested indefinite 5.5 x 0.5 x .10 (.60 in3)
Table 1—Comparison of Technologies
RFID reader. The current version of the WISP has a
range of up to 10 feet. It features a wireless power sup-
ply, bidirectional UHF communication with backscatter
uplink, and a fully programmable ultra-low-power 16-
bit flash microcontroller with analog to digital converter.
This WISP includes 32K of flash program space, an
accelerometer, temperature sensor, and 8K serial flash.
Small header pins expose microcontroller ports for ex-
pansion daughter boards, external sensors and peripher-
als.
The Intel WISP has been used to implement a variety
of demonstration applications that read data from a sin-
gle sensor unit. These include the first accelerometer to
be powered and read wirelessly in the UHF band, and
also the first UHF powered-and-read strain gage [21].
Even without its sensing capabilities, the Intel WISP can
be used as an open and programmable RFID tag: the
RC5 encryption algorithm was implemented on the In-
tel WISP [2]. We believe this is the first implementation
of a strong cryptographic algorithm on a UHF tag.
3 EXAMPLE APPLICATIONS
RFID sensor networks have broad applicability wherever
sensing, small form factor, embeddability, longevity, and
low maintenance are desired and fixed or mobile readers
are feasible. This section highlights applications within
this space and some of the key design considerations.
3.1 Blood
Blood transfusions save lives, replacing blood lost dur-
ing surgery, illness, or trauma. After donation, blood is
bagged and refrigerated between 1◦and 6◦C and has a
shelf life of about 35 to 42 days. Refrigerators used to
store blood are monitored for outages and temperature
fluctuations, and collection dates are recorded on blood
bags. However, the temperature of the bag itself is rarely
monitored with any regularity. This makes it difficult to
determine if a given bag was warmed to unsafe levels,
such as if it is near the front of the refrigerator and the
door is often opened. Additionally, it is difficult or im-
possible to gauge exposure during transport from a donor
to a bank, between banks, and ultimately to a patient.
WISPs with temperature sensors could be attached di-
rectly to individual blood bags and queried for their mea-
surements. Such sensors must be small (one could imag-
ine affixing sensors with something like a price tag gun),
and inexpensive to the point of being disposable.
To understand the challenges in building such an ap-
plication, an Intel WISP was attached to a container of
milk (a suitable and widely available approximation of
a bag of blood), and its temperature was monitored over
the course of 24 hours [20]. For this study, a storage ca-
pacitor (roughly the size of a pea) was attached to the
WISP. This enabled the WISP to log sensor data for up
to a day when out of range of a reader.
3.2 Brains
Recent research in neuroscience has explored using neu-
ral sensors for controlling prosthetic limbs [18]. Sen-
sors placed outside the skull can capture neural activity
but the signals are too coarse-grained and noisy to be ef-
fective. With surgery, sensors can be placed directly on
the brain resulting in much higher resolution and finer
control of the limb. Using conventional technologies
(e.g., motes) presents difficulties with respect to power
because batteries need to be replaced via invasive surgi-
cal procedures, as is the case with pacemakers.
An RFID sensor network is well suited to this applica-
tion. A patient would have WISPs equipped with neural
probes placed inside the skull. These could then draw
power from and communicate with a device outside the
body, e.g., an RFID reader worn as a cap, bracelet, or
belt. We have completed initial studies that show the fea-
sibility of integrating neural sensors with the WISP [7].
3.3 The Elderly
Providing care for the elderly is one of the largest health-
care costs facing us today, particularly as the “baby
boomer” generation ages. Keeping people in their homes
for as long as possible significantly reduces these costs
and increases quality of life. The difficulty with this is
detecting and reacting to emergencies, such as the pa-
tient falling or forgetting to take critical medication. Cur-
rently, families have no choice but to hire costly support
personnel to regularly check-in on their loved ones.
Traditional RFID has been explored to help monitor
the behavior of the elderly. For example, by having the
patient wear a short range RFID reader bracelet and plac-
ing RFID tags on a toothbrush, toothpaste, and faucet,
software can infer that an elderly person is brushing her
teeth when these tags are read in succession [11]. Such
fine-grained sensing requires very small devices, and is
simpler and more respecting of privacy than competing
3
approaches using computer vision, where video of the
person is continuously recorded and analyzed.
Adding sensing (e.g., an accelerometer) to long range
RFID tags would have several key advantages. Rather
than requiring a person to wear a short-range reader,
which can be taken off, a single long-range reader could
be placed in the home and behavior could be determined
via direct communication with the objects that are being
interacted with. This explicit information would be more
accurate in detecting behavior than inference based only
on object identifiers.
RSNs are an appropriate solution for the above appli-
cations and those like them. Our initial studies using the
WISP show the potential of existing RFID sensing de-
vices for use in such applications. However, these stud-
ies involved only a single WISP. Combining many such
devices into a full RSN will require further research.
4 CHALLENGES
RSNs combine the technology of RFID and sensing with
the usage models of sensor networks. At the device level,
the WISP shows that it is feasible to combine sensing
with RFID. However, at the systems level, challenges
arise due to the mismatch between the RFID usage model
and that of wireless sensor networks. We detail several
challenges in this section.
4.1 Tasks and Intermittent Power
RFID tags are powered only when they are in range of
an RFID reader. For regulatory and other reasons, read-
ers do not transmit a signal continuously. Instead, they
power tags for a brief period of time before changing
channels or entirely powering down. Thus, tags have an
unpredictable and intermittent source of power. More-
over, the RFID model assumes that if a tag is powered
it will have sufficient energy to respond to a command,
e.g., to transmit its identifier.
The above is a poor fit for sensor networks that seek to
complete a task that may span many RFID commands.
For example, the WISP harvests energy only when a
reader is transmitting and at a rate largely determined
by its proximity. This energy is stored in a capacitor
and when enough energy is harvested the WISP powers
up and can begin sensing and communicating. However,
sensing and communication drain power from the WISP.
This can result in the WISP losing power in the mid-
dle of an operation depending on the task and the reader
behavior. A further complication is that receiving, trans-
mitting, performing computation, and reading/writing to
memory all consume different amounts of energy.
To run tasks to completion, or at least in a manner that
can tolerate power interruptions, WISPs are likely to re-
quire support for intermittently powered operation. They
will need to be able to estimate the energy required to
complete a task, perhaps based on task profiling or en-
ergy budgets, and compare it with estimated reserves. To
work well in this regime, it is also likely that RSN de-
vices will need to cooperate with RFID readers for power
management. This would involve signaling by either the
reader, of its intended transmission time, or by the WISP,
of its needs. Even with signaling, it may be difficult to
predict power expectations because the rate at which en-
ergy can be harvested depends on the frequency of the
reader and the proximity of the device to the reader, both
of which will change over time. Thus, to increase the
kinds of tasks that can be supported, large tasks must be
split into smaller, restartable stages and device storage
(flash or RAM) can be used to pass intermediate results
between the stages.
4.2 Unpowered Operation
In many cases, WISPs may need to gather sensor data
when they are not in the proximity of an active RFID
reader. For example, the temperature of blood plasma
should be monitored while it is out of the refrigerator;
even though these periods should be relatively short they
are crucial for the application.
To extend functionality when away from a reader, one
approach is to provide a small amount of energy storage
on the device, e.g., a capacitor, and store excess energy
when close to an active reader. This storage capacitor
can be small relative to a battery, because it is intended
only for short term usage and is wirelessly recharged
over time. The Data Logger WISP used for the milk
carton study takes this approach, using a super-capacitor
that, when fully charged, sustains low duty-cycle oper-
ation for more than a day. The type of tasks that this
enables is limited, due to energy requirements, and the
period of functionality is limited due to leakage.
This use of stored energy for unpowered operation
raises many of the same issues as completing tasks given
intermittent power. We believe that a single power API
could support both situations. Stored energy is even
likely to help by providing a known buffer period for
loss of power. However, unpowered operation is likely to
stress tradeoffs between stages. For example, writing to
flash is significantly more energy intensive than comput-
ing with RAM but preserves valuable data for later use.
Unpowered operation is also likely to benefit from task
adaptation. For example, the duty-cycle or sensor sam-
pling rate might be increased or decreased depending on
the long-term power harvesting trends.
4.3 Sensing Protocols
WSN nodes are peers in terms of the physical and link
layers of their communication, e.g., each mote has an
802.15.4 radio capable of sending and receiving trans-
missions with other nodes that are in range. In con-
4
trast, because they draw on RFID, RSN nodes will be
highly asymmetric in terms of their communication abil-
ities. With RFID, readers are able to transmit messages
to all tags and tags can transmit messages to the reader.
However, tags can do so only when the reader initiates
communication, and tags cannot communicate directly
with each other even when powered by the reader.
These differences complicate protocols designed to
gather and process sensor data. Currently, WISPs with
new sensor data must wait until they are interrogated by
a reader. This increases the likelihood of many devices
wanting to use the bandwidth limited channel at the same
time. Techniques to perform data pre-processing within
the network (on each RSN device) can help to some ex-
tent. However, the standard RFID strategy of identify-
ing and then communicating with each device is waste-
ful as most devices may not have relevant data – a more
dynamic strategy based on the value of the sensor data
would be more effective.
Consider the eldercare application. A reader might
have hundreds of accelerometer WISPs in its field of
view. Because all the WISPs share a single reader chan-
nel, the update rate per tag would be very low if every tag
were simply queried for sensor data sequentially. How-
ever, at any given moment, only a few objects will typi-
cally be in motion (and therefore producing non-trivial
accelerometer sensor values). Furthermore, the set of
objects that are moving changes dynamically, as objects
are put down and picked up. One might want a protocol
which gives priority to the most active objects, politely
“yielding” to new objects when they start to move.
Existing RFID solutions do not support anything like
this functionality. As a first step, one could have WISPs
with sensor activity below a threshold not respond to the
reader. But an appropriate threshold level may depend on
what is occurring in the room, and such a simple scheme
would not support the “polite yielding” described above.
For another example of what RSN protocols might be
asked to do, consider the blood application. When many
blood bags are read simultaneously, one might want to
interrogate the bags with the largest temperature excur-
sions first. But since the distribution of temperature ex-
cursions is not known a priori by the reader, the proto-
col would need to (implicitly) estimate this information.
It might for example ask if any WISP has a larger tem-
perature excursion than E. If no device responds, the E
response threshold would be repeatedly halved until the
appropriate scale was found. The key requirement is to
estimate an aggregate property of the data without ex-
haustively collecting that data. Finally, RSN protocols
might be power aware as well. A WISP that was about to
lose power might be given priority over those with ample
power.
New tools will be needed to explore RSN communi-
cation protocols. As a first step, we are developing an
RFID reader platform based on the Universal Software
Radio Peripheral (USRP). This platform used in conjunc-
tion with the WISP allows for the development of new
protocols at both the MAC and PHY layers. Thus far we
have used it for RFID monitoring [1].
4.4 Repurposing C1G2
There would be substantial practical benefit to realiz-
ing RSN protocols using the primitives of the C1G2
standard: commercial off-the-shelf readers could then
be used for RSN research and deployment, and WISPs
would interoperate with ordinary (non-sensing) tags.
However, the extent to which RSN protocols can be im-
plemented within the C1G2 standard is an open research
question. Additionally, there is the practical considera-
tion of commercial readers not exposing low-level func-
tionality and not implementing the complete C1G2 spec-
ification. Because of this, even RSN protocols built on
top of the C1G2 specification may not be implementable
using standard readers.
Our experience with the Intel WISP suggests that ba-
sic RSN applications can be approximated using stan-
dard C1G2 readers. To read sensor data from a C1G2
WISP, the device must first be singulated, at which point
a temporary handle is requested from the tag. A reader
can then use this handle to address the device and read
sensor data from pre-defined memory locations. How-
ever, the handle persists only until the reader singulates
another tag or the tag loses power. Thus, reading from
more than one WISP incurs substantial protocol over-
head due to singulation and handle management. Con-
sequently, simple use of the existing C1G2 protocol can
provide some level of sensing functionality, but at a sig-
nificant cost in terms of efficiency.
Along with reading sensor data, the C1G2 protocol
can support basic sensor queries using the Select com-
mand. If the reader knows that a sensor value is writ-
ten to a particular memory location, it can issue a Select
command with a mask which matches that location for
sensor values over a given threshold. Consequently, only
WISPs with sensor values over that threshold will reply
during the next frame. More generally, the Select com-
mand could be used as a general purpose broadcast chan-
nel. The bit mask in the command could be repurposed
and interpreted, in the most general case, as opcodes and
data. As multiple Selects can be sent before each frame,
complex tasking and querying could be achieved in this
manner.
The above mechanisms show that there is potential
for using the C1G2 standard to implement RSN pro-
tocols. This would have the advantage of being im-
plementable using current reader technology, given a
reader that is sufficiently programmable. However, these
5
mechanisms may prove too inefficient or may simply be
poorly matched to many applications. Further experi-
mentation is needed.
4.5 Tasking and Querying
Tasking can refer to a number of different things in the
context of sensor networks. One extreme is the transfer
of an entire code image to a device which changes its ba-
sic operation[8, 17], while the other extreme is transmit-
ting simple commands to trigger preconfigured behavior
such as modifying the sampling rate of a sensor. Several
WSN projects have chosen points in between, inventing
high-level task construction languages [5], or adopting
SQL-style declarative queries [10]. Generally, for non-
trivial applications some method must be available to ac-
tuate the behavior of devices.
The communication model of RFID results in differ-
ent trade-offs with respect to tasking than are seen in
WSNs. When tags are being powered sufficiently, the
energy cost of transmission is essentially zero and trans-
ferring large amounts of code is feasible. This allows
for the complete retasking of WISPs with costs in terms
of latency only. Conversely, because downlink commu-
nication is cheap when in range of the reader, WISPs
may not need to be as “smart” as motes. For example,
requirements for precomputation and aggregation may
be relaxed and filtering could be done through complex
queries.
In RFID terminology querying refers to identifying
tags, while in WSNs it refers to expressively indicating
the sensor data of interest. In order to realize the latter
idea using WISPs, query languages and methods of com-
municating queries need to be developed. As in WSNs,
there is a fundamental trade-off between expressiveness
and efficiency. In some regards, querying is closely cou-
pled with tasking as the complexity is split between the
reader and the WISP.
However, as discussed in the previous sections, energy
concerns play a part in determining how to divide func-
tionality between the reader and the device. For exam-
ple, interpreting complex queries on the WISP may be
efficient with respect to latency but inefficient with re-
spect to energy. Quantifying these trade-offs is an open
question.
5 CONCLUSION
By exploiting RFID technology, we believe that we can
expand the application space of wireless sensor net-
works to ubiquitous, embedded sensing tasks. We have
sketched sample sensor network applications in the space
between traditional mote networks and RFID for supply-
chain monitoring. We have described key systems and
networking challenges related to intermittent power and
RSN protocols for sensor queries. We expect RSNs to be
a fruitful new space for networking and systems research,
as there is significant work that must be done to translate
the capabilities of the WISP into full-fledged RSNs.
REFERENCES
[1] M. Buettner and D. Wetherall. An empirical study of uhf rfid
performance. In Proc. Mobicom (to appear), 2008.
[2] H. J. Chae, D. J. Yeager, J. R. Smith, and K. Fu. Maximalist
cryptography and computation on the wisp uhf rfid tag. In Proc.
Conference on RFID Security, 2007.
[3] N. Cho, S. Song, S. Kim, S. Kim, and H. Yoo. A 5.1-uw uhf
rfid tag chip integrated with sensors for wireless environmental
monitoring. In IEEE Biological Circuits and Systems (BioCAS)
2008, submitted, 2008.
[4] EPCglobal. EPC radio-frequency identity protocols class-1
generation-2 UHF RFID protocol for communications at 860
mhz-960 mhz version 1.0.9. 2005.
[5] O. Gnawali, K.-Y. Jang et al. The tenet architecture for tiered
sensor networks. In Proc. ACM SenSys, 2006.
[6] C. Hartung, R. Han, C. Seielstad, and S. Holbrook. Firewxnet:
a multi-tiered portable wireless system for monitoring weather
conditions in wildland fire environments. In Proc. MobiSys, 2006.
[7] J. Holleman, D. Yeager, R. Prasad, J. Smith, and B. Otis. Neu-
ral wisp: An energy harvesting wireless brain interface with 2m
range and 3uvrms input referred noise. In Proceedings of the 31th
European Solid-State Circuits Conference, Grenoble, 2005.
[8] J. Hui and D. Culler. The Dynamic Behavior of a Data Dissemi-
nation Algorithm at Scale. In Proc. ACM SenSys, 2004.
[9] J. Kahn, R. Katz, and K. Pister. Emerging challenges: Mobile
networking for ‘smart dust. Journal of Communication Networks,
pages 188–196, 2000.
[10] S. Madden, M. Franklin, J. Hellerstein, and W. Hong. TAG: a tiny
aggregation service for ad-hoc sensor networks. In Proc. OSDI,
2002.
[11] M. Philipose, K. Fishkin et al. Inferring activities from interac-
tions with objects. IEEE Pervasive Computing, 3(4):50–57, 2004.
[12] J. Polastre, J. Hill, and D. Culler. Versatile low power media
access for wireless sensor networks. In SenSys, 2004.
[13] J. Polastre, R. Szewczyk, and D. Culler. Telos: Enabling Ultra-
Low Power Wireless Research. In Proc. IPSN/SPOTS, 2005.
[14] L. G. Roberts. Aloha packet system with and without slots and
capture. SIGCOMM Comput. Commun. Rev., 5(2):28–42, 1975.
[15] A. P. Sample, D. J. Yeager, P. S. Powledge, and J. R. Smith.
Design of an rfid-based battery-free programmable sensing plat-
form. In IEEE Transactions on Instrumentation and Measure-
ment (accepted), 2008.
[16] J. R. Smith, A. P. Sample, P. Powledge, A. Mamishev, and S. Roy.
A wirelessly powered platform for sensing and computation. In
Proc. Ubicomp, 2006.
[17] T. Stathopoulos, J. Heidemann, and D. Estrin. A remote code up-
date mechanism for wireless sensor networks. In CENS Technical
Report No. 30.
[18] D. Taylor, S. H. Tillery, and A. Schwartz. Direct cortical control
of 3d neuroprosthetic devices. Science, 296:1829–1832, 2002.
[19] A. Woo, T. Tong, and D. Culler. Taming the underlying chal-
lenges of reliable multihop routing in sensor networks. In Proc.
SenSys, 2003.
[20] D. Yeager, R. Prasad, D. Wetherall, P. Powledge, and J. Smith.
Wirelessly-charged uhf tags for sensor data collection. In Proc.
IEEE RFID, 2008.
[21] D. J. Yeager, A. P. Sample, and J. R. Smith. Wisp: A passively
powered uhf rfid tag with sensing and computation. In M. I. Syed
A. Ahson, editor, RFID Handbook: Applications, Technology,
Security, and Privacy. CRC Press, 2008.
6
