Electrical Engineering and Computer Science Department

Technical Report
NWU-EECS-06-09
August 3, 2006

Lucid Dreaming: Reliable Analog Event Detection

for Energy-Constrained Applications
Sasha Jevtic Mat Kotowksy†
Robert Dick Peter Dinda Charles Dowding‡

Abstract
Existing sensor network architectures are based on the assumption that data will be
polled. Therefore, they are not adequate for long-term battery-powered use in
applications that must sense or react to events that occur at unpredictable times. In
response, and motivated by a structural autonomous crack monitoring (ACM) application
from civil engineering that requires bursts of high resolution sampling in response to
aperiodic vibrations in buildings and bridges, we have designed, implemented, and
evaluated lucid dreaming, a hardware/software technique to dramatically decrease sensor
node power consumption in this and other related event-driven sensing applications. Our
hardware is an add-on board for standard Crossbow Motes that makes use of an ultra-
low-power analog comparator and an in-system programmable precision voltage
reference. The sensor, e.g., geophone, output voltage is compared to the reference. When
it exceeds the reference, an interrupt is delivered to the Mote, activating it and triggering
high-resolution sampling. In the structural integrity monitoring application, this is
achieved with 1/245 the average power consumption required by existing sensor network
architectures, thereby dramatically increasing battery lifespan. We believe that the
proposed technique will yield similar benefits in a wide range of applications.

†
Infrastructure Technology Institute, Northwestern University
‡
Civil and Environmental Engineering Department, Northwestern University
Keywords: Sensor networks, event detection, low power, analog circuits
 Lucid Dreaming: Reliable Analog Event Detection for
Energy-Constrained Applications
Sasha Jevtic† Mat Kotowsky‡ Robert P. Dick† Peter A. Dinda† Charles Dowding‡

†(sjevtic, dickrp, pdinda)@eecs.northwestern.edu ‡(kotowsky, c-dowding)@northwestern.edu

Dept. of Electrical Engg. and Computer Science Dept. of Civil and Environmental Engineering
Northwestern University Northwestern University
3 August 2006

Abstract sors, microcontrollers, and software within individual wire-

less sensor network nodes.
Existing sensor network architectures are based on the as- In this context, two universal research problems come to
sumption that data will be polled. Therefore, they are the fore: the maintenance problem and the unpredictable
not adequate for long-term battery-powered use in applica- event problem. How can we arrange for nodes to operate
tions that must sense or react to events that occur at un- without frequent intervention? Low maintenance is neces-
predictable times. In response, and motivated by a struc- sary to allow large-scale deployments in remote environ-
tural autonomous crack monitoring (ACM) application from ments. It is prevented by short battery life, hence we focus
civil engineering that requires bursts of high resolution sam- on increasing battery life. How can we arrange for nodes
pling in response to aperiodic vibrations in buildings and to react to environmental events that occur at unpredictable
bridges, we have designed, implemented, and evaluated lu- times? We cannot assume that interesting data will be pre-
cid dreaming, a hardware/software technique to dramati- sented, on a silver platter, whenever requested. Jointly ad-
cally decrease sensor node power consumption in this and dressing the maintenance and unpredictable event problems
other related event-driven sensing applications. Our hard- requires changes to the sensor network node architecture, al-
ware is an add-on board for standard Crossbow Motes that lowing it to respond to events at any time while maintain-
makes use of an ultra-low-power analog comparator and an ing ultra-low power consumption. We claim that addressing
in-system programmable precision voltage reference. The the problem requires a combined hardware and software ap-
sensor, e.g., geophone, output voltage is compared to the ref- proach. As described in Sections 2 and 5, attempts to solve
erence. When it exceeds the reference, an interrupt is deliv- these problems with software, alone, have resulted in high
ered to the Mote, activating it and triggering high-resolution power consumption or missed events.
sampling. In the structural integrity monitoring application, This work is motivated by applications that have the fol-
this is achieved with 1/245 the average power consumption lowing characteristics:
required by existing sensor network architectures, thereby
dramatically increasing battery lifespan. We believe that the 1. They are extremely power-sensitive. The nodes are
proposed technique will yield similar benefits in a wide range powered by batteries that can be replaced only after
of applications. months or years of operation.

2. Low-power sensors and computational elements can be

used for detecting, but not necessarily taking detailed
1 Introduction measurements of, events.

Wireless sensor networks have the potential to serve as plat- 3. Events are rare and the computation and/or communica-
forms for a wide range of environmental monitoring and con- tion they trigger is short relative to the event interarrival
trol applications. Applications can be considered at many time.
levels, from the individual sensors, to the individual node
hardware and software, to the local wireless network formed 4. Event interarrival times are unpredictable.
by nodes, and finally to that network’s interaction with the
broader world. Our work focuses on interaction among sen- 5. It is preferable not to miss, or ignore, events.

1
Section 3 describes the specific motivating application we paging [2], and power management algorithms [21] to in-
target. In that application, events are structural vibrations. crease battery lifespans in wireless sensor networks. Work
They cause a sensor voltage to exceed a threshold, resulting on low-power communication is largely orthogonal to the
in a burst of high-resolution data logging. idea described in this article, and can be used in combina-
Communication is not a significant power sink for our tion with it.
exemplar application, or other related applications, because Most previous research on low-power sensing focuses on
sensor data logs and events need not be aggregated in real- periodic sensing applications in which sensor network nodes
time. Thus, queuing collected data on the node and send- may safely enter low-power modes at times of their choos-
ing batch transmissions allows the radio to be powered down ing with the knowledge that data of interest will be available
most of the time. Modern ad hoc sensor network proto- whenever they choose to wake up.
cols [3, 4] can similarly keep radio transmitter and receiver Many applications, however, require the ability to reliably
off most of the time. sense and/or react to events that occur at unpredictable times,
Surprisingly, given that such applications are legion, ex- e.g., the structural integrity monitoring application described
isting and proposed sensor network node hardware and soft- in Section 3. Previous research on such event-driven applica-
ware do not adequately support this class of application. The tions [15, 17, 22] has relied on existing sensor network archi-
power consumption of the microcontroller and primary sen- tectures. However, this has proven to be a poor fit, leading to
sor are considerable for the following reasons: high power consumption that results in battery lifespans on
the order of hours or days.
1. Event detection is done in software via a sleep-read-
Some researchers have attempted to use sophisticated
test-jump polling loop. Polling requires that the primary
event prediction algorithms to improve the power consump-
sensor, analog-to-digital converter (ADC), and micro-
tion of existing sensor network architectures when used in
controller remain in active states resulting in high power
event-driven applications [21]. However, without perfect
consumption.
prediction accuracy, such techniques must necessarily miss
2. Event arrival times cannot be accurately predicted and critical events or waste battery energy. Furthermore, the pre-
should not be lost. Therefore, the amount of time spent dictability of events is largely domain-dependent and evalu-
in the sleep state, whether deterministic or random, ating it is often a goal of the application research using the
must be small. sensor network. For many applications, including the one
described in Section 3, events are too unpredictable for such
We describe the design, implementation, and evaluation methods to be feasible.
of lucid dreaming, a hardware/software technique permit- Most closely related to our work is that of Dutta et al. [12].
ting long battery lifespans in applications requiring the de- This group has carefully considered minimizing power con-
tection of unpredictable events. Specifically, lucid dreaming sumption in event-driven applications, identified the diffi-
eliminates the need for the primary sensor, ADC, and mi- culty of detecting rare, random, and ephemeral events using
crocontroller to remain continuously active. The key idea is existing sensor network architectures, and proposed a new
that event detection can be done in analog hardware much architecture that uses duty cycling and wakeup circuits to re-
more efficiently than as code running on the microproces- duce power consumption. Duty cycling sensors to reduce
sor. Hence, our analog hardware, Mote-Wake, can wake up power consumption must necessarily increase the probabil-
a standard Crossbow Mote [19, 16, 9] by raising a hardware ity of missing random events. This problem is alleviated, to
interrupt. The interrupt handler in turn causes high resolu- some degree, by allowing sensors to wake up other nearby
tion sampling to occur. sensors in response to events. Although this idea is applica-
In our exemplar application, event detection is straightfor- ble in dense sensor deployments for detecting vehicles and
ward: an event interrupt is generated when the sensor’s volt- soldiers (Dutta’s intended application), it cannot be used in
age level exceeds a sensor and application-specific thresh- cases where the events of interest are truly ephemeral, i.e.,
old. Of course, this is a quite broadly useful event genera- they last for only a moment and do not imply that other
tion function for many applications, but we believe that lu- events will, with high probability, be observed in the neigh-
cid dreaming can also be generalized to more complex event borhood of the previous event, as is the case for our moti-
generation functions. We discuss the possibilities for such vating structural integrity monitoring application. Dutta et
generalization in Section 6. al. also describe the properties of a number of wake-up cir-
cuits. Unfortunately, all the sensors and wake-up circuits de-
scribed have disturbingly high power consumption, i.e., from
2 Related work 880 µW to 19,400 µW. In the words of the authors, “We had
high hopes for the low-power wakeup circuits used with the
A number of researchers have considered designing hard- infrared and acoustic sensors. Unfortunately, these circuits
ware, communication protocols [20, 23] or multi-channel did not live up to our early expectations.” We point out the

2
difficulties Dutta et al. faced only to make clear the impor-
tance of the problem we address and highlight our contribu-
tions.
Our work makes the following contributions. First, lucid
dreaming allows the detection of unpredictable events while
maintaining ultra low-power operation. The average power
consumption of our sensor and wakeup circuit is 15 µW,
which is almost three orders of magnitude lower than the
best previously reported. This is a change in magnitude that
translates into a change in kind. The use of sensor networks
in long-term remote monitoring applications is simply in-
feasible with battery lifespans on the order of days but be-
comes practical when battery lifespans can me measured in
months or years. Second, our technique does not rely on
imperfect prediction heuristics or require that events be spa-
tially or temporally correlated. Third, it guarantees reliable
event detection without using duty cycling to trade off reli-
ability for reduced power consumption. Measurements in-
dicate that, when used in our motivating application, lucid
dreaming will increase battery lifespan by 245 without any
negative impact, bounded only by the shelf life of the batter-
ies in use.

3 Motivation
Mote-Wake was motivated by our discussions with a civil en- Figure 1: Geophone connected to Mote-Wake board mated
gineering group that is deploying sensor networks based on to Crossbow Mote
Crossbow Mote technology. It was clear that existing sensor
network architectures were inadequate for their, fairly typ-
ical, structural integrity monitoring application. Moreover, At its core, crack monitoring is a trigger-log-push appli-
we believed that a sensor network node architecture address- cation. Extremely high resolution data is needed when the
ing their specific needs would be useful in a broad class of crack is in motion. Crack motion events occur at unpre-
event-driven sensing applications. dictable times. Hence, we want to trigger when crack motion
The overall objective of the Autonomous Crack Monitor- begins, log at the limits of the sampling resolution available
ing (ACM) project [11, 10, 6] is Internet-enabled remote until motion subsides, and finally, later push the log to an
monitoring of cracks in, or deformations of, structures to analysis center.
provide timely information about the health of critical infras- This kind of application fits poorly to existing sensor net-
tructure components such as bridges and buildings. Time- work node technology, such as the Crossbow Motes the
series data collected from sensors can be analyzed to iden- ACM group is using, and to future node technologies of
tify trends and automatically alert engineers and/or regula- which we are aware. In the ACM application, logging must
tory authorities of impending problems. The ACM group’s be done at high resolution. This results in high power con-
original system [10] is being deployed to compare environ- sumption. However, we are only concerned with the logs
mental (long-term) and blast-induced (dynamic) crack width for a relatively short duration after an event, i.e., the onset
changes in residential structures, and has lead to a new of crack motion, occurs. Current node hardware provides a
approach to monitoring and controlling construction vibra- wakeup timer, but this does nothing to improve the situation
tions. It is a wired system that requires constant power and because the time of the next event is not predictable. This
significant maintenance. leaves the designer with two unsatisfactory choices: sam-
The ACM group is working to replace the existing wired ple at a high rate all the time, resulting in inadequate battery
system with a wireless sensor network [14, 18, 11]. Their lifetimes, or use the wakeup timer to implement some sam-
goal is to support a year of reliable, unattended operation pling schedule, which will result in undetected events. Nei-
powered only by the two AA batteries in each of the wireless ther choice is acceptable for large-scale critical infrastructure
nodes. The work on this application recently won third place monitoring.
honors in the 2005 Crossbow Smart Dust Challenge [14]. The ACM application uses string potentiometer and a geo-

3
  Software: The sensor network node is placed in a low-
Low-power Primary sensor
Data transmission power standby state whenever there is no sensing, data
secondary sensor (String
(Geophone) potentiometer) processing, or communication work to be done. The
Can use primary node can be activated either with a timer (for exam-
sensor if power low Data logging ple, to drive communication), or when a sensor event
ADC
occurs. In the low power state, the microcontroller is
Ultra-low-power placed in power-down mode, from which it may only
analog event Microcontroller Event filtering be awakened by a hardware interrupt or the watchdog
detection hardware
timer. ADCs are powered down and communication in-
Hardware Software terfaces are temporarily disabled. The microcontroller
is halted until an external hardware interrupt occurs. In
Figure 2: Lucid dreaming system overview response to an event interrupt, the microcontroller is ac-
tivated. The microcontroller can the, e.g., activate the
ADC and store a series of samples from the primary
phone [7, 8], which is illustrated in Figure 1. Geophones sensor.
are un-powered devices that produce output voltages. When
used to monitor a crack, crack motion induces a significant We begin by describing the criteria under which the lucid
voltage that rises above typical background noise. In the de- dreaming technique can be applied. Next, we describe our
fault ACM configuration, the string potentiometer is attached hardware implementation. Finally, we describe the software
to an ADC input on the Mote and the application detects the side of our implementation.
onset of crack motion by continually sampling the ADC and
comparing the sampled value to a threshold. It is the effect of
this polling loop that we have moved from software running 4.1 Criteria for viability
on the ATMega128 microcontroller and ADC to the custom Lucid dreaming works exceptionally well for our motivating
hardware of the Mote-Wake board. application. We also believe it will be applicable to a range
of other event-driven sensor network applications of the kind
we described in the introduction, resulting in power savings
4 Technical description that depend on a number of application-specific parameters.
However, several criteria must be met in order for the tech-
Lucid dreaming is a general hardware/software technique for nique to be applicable. We now elaborate on these criteria.
reducing power consumption in individual sensor network Sensor/sensor support circuit power requirements must be
nodes that react to events detected via, potentially straight- modest. Lucid dreaming requires that a sensor be continu-
forward, computations on values measured using sensors. ously online which, in some cases, necessitates that the sen-
The proposed technique has relatively few requirements, and sor be biased continuously. If support circuitry (such as a
thus is viable in a large number of applications. Moreover, filter or amplifier) is required, it must also be continuously
the technique may be used with more platforms other than powered. The power consumption of our technique when no
the MICA2 and MICAz, although doing this would require event is occurring is the sum of the power consumptions of
a PCB redesign. the wakeup circuitry, the sensor, and their associated elec-
Figure 2 provides a high-level overview of lucid dreaming tronics. Hence, as sensor power consumption increases, the
as used in our motivating application. The technique has two benefit of the proposed technique decreases. Fortunately,
main components: many sensors have power consumptions that are lower than
that of the fully active sensor network node.
 Hardware: Custom analog hardware observes the sen- The geophone used in the ACM application represents an
sor, detects events based on these observations, and no- ideal sensor for use with our technique as it is completely
tifies the microcontroller when more sophisticated pro- self-powered, and produces a clean, output that does not
cessing is required. In our example hardware, Mote- require amplification. Requirements for powered sensors
Wake, events are detected when the geophone output and/or active support circuits reduce the energy savings real-
voltage exceeds a threshold. Other detection methods, ized by the technique.
e.g., low-power finite state machines, may be used in To maximize the power savings possible from the pro-
other applications. Although we use separate sensors posed technique, it may be necessary to add a secondary
for event detection and data logging, the primary sensor sensor that exhibits favorable power consumption and out-
may also be used for event detection if its power con- put characteristics solely for the purpose of event detection.
sumption is sufficiently low. When an event occurs, the For example, the ACM application, the geophone is used to
hardware raises an interrupt. detect events. However, upon detecting an event, the system

4
activates a second sensor with much higher power consump-
tion to take a series of detailed measurements.
The important implication is that it is the power consump-
tion of sensor used for event detection, not data logging, that
is critical. The event detection sensor need not have linearity,
full-scale output, or other ideal characteristics. Thus, a vari-
ety of unconventional sensors, or sensors operated in uncon-
ventional manners, may be used as event detection sensors,
e.g.,

 Solar cells, for light;

 Unbiased microphones, for audio;
 Piezoelectric elements, for vibration; and
Figure 3: Top view of Mote-Wake

 Peltier elements, for temperature differences.

is impractical to implement perfectly-accurate event detec-
Event arrival times should be difficult to predict exactly. tion in low-power hardware, the proposed technique can still
If it is known when the next event is likely or sure to oc- be used in conjunction with hardware that generates occa-
cur, then lucid dreaming no more effective than conventional sional false positives to reduce Mote activation frequency
timer-based periodic or predictive wake-up is. However, the and, therefore, average power consumption. Because the
lucid dreaming technique can be beneficial when predictable Mote-Wake hardware and an attached sleeping Mote use sig-
events exhibit variation from occurrence to occurrence. nificantly less power than an active Mote, it is likely that
Events should be infrequent and quickly processed. As reducing any substantial quantity of false positives through
events become more frequent and/or more time-consuming Mote-Wake hardware enhancements will be beneficial.
to process, the Mote approaches always-on operation, result-
ing in decreasing effectiveness of lucid dreaming. Many ap- 4.2 Hardware
plications that record or react to infrequent phenomena in
the environment, e.g., the ACM application, satisfy this cri- The hardware component (Mote-Wake) is the heart of the lu-
terion. cid dreaming technique. It is a simple, ultra-low-power opti-
Communication should be infrequent and short. The ef- mized threshold detection circuit designed for direct attach-
fectiveness of the technique also depends upon the commu- ment to a Crossbow MICA2 or MICAz Mote. The Mote-
nication behavior of the application. Sensor network nodes Wake PCB layout (Gerber files) and bill of materials are
often participate in mesh network schemes that require them available for those wishing to build or have built their own
to wake up and communicate from time to time to perform Mote-Wake boards.
data aggregation. If communication is frequent and intense, The Mote-Wake PCB (Figure 3) measures
its power costs may dominate the power savings provided by 1.25 in2.25 in, and has mounting holes and a set of
lucid dreaming. The proposed technique is applicable when Hirose 51-pin Mote expansion connectors in the same
moderate to small amounts of data are transferred in response locations as the Motes and their common expansion mod-
to infrequent events. ules. The connectors, which pass through all signals,
Event detection should be simple enough to implement us- allow Mote-Wake to be placed at an arbitrary location in
ing low-power hardware. Events are detected based on sen- a MICA2/MICAz hardware stack. The mounting holes,
sor observations. For some applications, detecting events of which are connected to GND and surrounded by generous
interest may be quite complex. A key idea in lucid dreaming keep-out regions, allow Mote-Wake to be physically secured
is moving event detection from software into very low power to the hardware stack with ease, while simultaneously
analog hardware, and such hardware is limited in the com- avoiding the risk of shorts or other damage. Mote-Wake is a
plexity of measurements based upon which it detects events. two-layer board. The unused area on the top copper has been
Our hardware for the ACM application implements threshold designated as a polygon fill connected to GND, while the
detection. Hardware implementation of more complex func- unused area on the bottom copper is a polygon fill connected
tions, such as filtering or low-power finite state machines, is to VCC. This technique provides some of the benefits of
also possible, albeit with larger power requirements. Fortu- VCC/GND planes, e.g., distributed decoupling capacitance
nately, lucid dreaming event detection hardware may safely and shielding, without the expense of a four-layer board,
generate some false positive event indications, which are which would be required for full power planes. Mote-Wake
subsequently eliminated by the sensor network node micro- is powered directly from the Mote’s VCC/GND, as made
controller without impacting correctness. Thus, even if it available on the 51-pin Hirose expansion connectors.

5
 1 2 3 4

A CN1 A

GND 26
51 UART_RXD0
VSNSR 25
50 UART_TXD1 VCC
INT3 24 C1
49 PW0
INT2 23
48 PW1
INT1 22
47 PW2 CN3 J1
INT0 21 ECQ-P1H103GZ

8
46 PW3 1 1 3
CC_CCA 20 R2 0.01uF
45 PW4 2 2 4 2 U2A
LED3 19
44 PW5 CFR-25JB-100R 1
LED2 18
43 PW6 S2B-PH-K-S PRPN022PARN 100 3
LED1 17
42 ADC7 VCC MAX9020EKA-T
RD 16

3
41 ADC6 D1
WR 15

4
40 ADC5
ALE 14
39 ADC4 1 2
PW7 13
38 ADC3
USART1_CLK 12 J3
37 ADC2
PROG_MOSI 11 8 4 INT0
36 ADC1 HSMS-2702
PROG_MISO 10 7 3 INT1
9
35 ADC0
SPI_SCK 34 6 2 INT2
8 THERM_PWR
USART1_RXD 33 CN3/CN4: mate 5 1 INT3
7 THRU1
B USART1_TXD 32 with JST PHR-2 B
6 THRU2 VCC
I2C_CLK 31 housing. PRPN042PARN
5 THRU3
I2C_DATA 30
4 RSTN VCC
PWM0 29 J1-J3: use Sullins
3 PWM1B CN4 J2
PWM1A SPN02SYBN-RC

8
2 28 VCC 2 1 3
AC+ 27 R3 or equivalent.
1 GND 1 2 4 6 U2B
AC-
CFR-25JB-100R 7
S2B-PH-K-S PRPN022PARN 100 5
DF9-51P-1V(54) VCC MAX9020EKA-T

3
D2

4
1 2
CN2
VCC
AC- 1
27 GND HSMS-2702
AC+ 2
28 VCC
PWM1A 3
29 PWM1B
PWM0 4 Note: Polarity of
30 RSTN VCC
I2C_DATA 5 CN3 is opposite
31 THRU3
I2C_CLK 6 polarity of CN4.
32 THRU2 U1
USART1_TXD 7 33 R4
THRU1 1 2
C USART1_RXD 8 34 IN OUT C
THERM_PWR MFR-25FBF-1M00
SPI_SCK 9 GND
35 ADC0
PROG_MISO 10 1M
36 ADC1 MAX6018AEUR12-T
PROG_MOSI 11 3
37 ADC2
USART1_CLK 12
38 ADC3
PW7 13 39
14 ADC4
ALE 40
15 ADC5
WR 41
16 ADC6
RD 42
17 ADC7
LED1 43
18 PW6 VCC
LED2 44
19 PW5
LED3 45
CC_CCA 20
46
PW4 R1 100K Title: MOTE-WAKE (2x Large Geophone)
21 PW3 1 6
INT0 47 VDD
22 PW2 2 5 Author: Sasha Jevtic Revision: 1
INT1 48 GND
23 PW1
INT2 49
24 PW0 I2C_CLK 3 4 I2C_DATA Date: 2/27/2006 Time: 9:46:21 PM
INT3 50 SCL SDA
25 UART_TXD1
VSNSR 51
26 UART_RXD0 MAX5434/MAX-5435 File: U:\wakeup\hardware\MOTE-WAKE\MOTE-WAKE.SchDoc
GND

DF9B-51S-1V(54)
D D

1 2 3 4

Figure 4: Mote-Wake Schematic

Figure 4 is the schematic diagram for Mote-Wake, as illus- comparators contained in U2. The comparators feature 4 mV
trated in Figure 3. Sensors may be connected to CN1 and/or of hysteresis internally, providing both noise immunity and
CN3; J1 and J2 are jumpers used to enable/disable the sen- clean switching in the presence of a low slew rate, noisy in-
sors on CN1 and CN3, respectively. Disabling an unused in- put. The non-inverting inputs of the comparators are con-
put, if any, is necessary both to save power and prevent spuri- nected to a programmable voltage divider subsystem. The
ous event detection. An input protection network consisting output of the comparators are open-drain, allowing them to
of diodes and resistors protects the hardware from large tran- be directly connected to the active low/level sensitive inter-
sients which may result from vigorous shaking of the geo- rupt lines of the ATMega128L microcontroller in a wired-
phone electrostatic discharge, or other sources. D1 and D2 OR configuration merely by enabling the ATMega128L’s
are high-performance Schottky clamping diodes; they com- internal pull-up resistors. This configuration conserves re-
bine high switching speed with exceptionally low forward sources by avoiding the use of a second interrupt line or an
voltage and series resistance. R2 and R3 are current limiting OR gate. Thus, whenever the voltage of an enabled sensor
resistors that further limit the system’s exposure to damag- input exceeds that of the non-inverting input voltage level, an
ing transients. Due to exceptionally high input impedance, ATMega128L interrupt line of the user’s choice is taken low.
the R2 and R3 cause virtually no drop in the magnitude of The user may select from INT[0..3], as provided on the Hi-
the incoming sensor signal. rose connector using J3; these correspond to ATMega128L
Following the input protection network, the sensor sig- interrupts INT[5..8], respectively.
nals are passed to the inverting inputs of the low-power dual The voltage divider subsystem consists of a low-power

6
precision 1.263 V voltage reference, allowing the inverting which it entered sleep mode.
input to both comparators to remain constant over the life A library routine called the “sleep preparation routine” is
of the Mote batteries without the addition of a voltage regu- provided. This small function enables the interrupt that acti-
lator and providing immunity from power supply transients. vates the Mote-Wake board and writes to a sleep register to
The voltage reference output is connected to a series con- put the Mote into a low-power sleep mode. A second library
nection of a fixed precision 1 MΩ resistor in series with a routine is provided to configure the digital potentiometer, al-
100 KΩ, 32-tap digital potentiometer with nonvolatile wiper lowing the program to change the threshold level at which an
memory. The digital potentiometer, connected to the Mote’s event is generated by Mote-Wake.
I2 C bus provides programmatic selection of the voltage pro-
vided to the non-inverting inputs of the comparators, thereby
effectively enabling remote selection of the wakeup stimulus 5 Power and performance models and
threshold. Although the I2 C address of the digital poten- measurements
tiometer is fixed, it does not conflict with any addresses cur-
rently in use in the node hardware we support. Furthermore, We now present power and performance models for our im-
alternate addresses may be obtained with the substitution of plementation of lucid dreaming and discuss the results of
otherwise identical variants of the digital potentiometer of- bench tests with the Mote-Wake PCB. The proposed models
fered by the device’s manufacturer. The fixed resistor serves can be used by application developers to quickly determine
two roles. First, it concentrates the range of possible out- the degree to which the proposed technique will improve
put voltages of the voltage divider system around the volt- power consumption. We show the behavior of the models for
age of interest. Second, it greatly increases the resistance a range of parameter values corresponding to current hard-
of the voltage divider network, thereby avoiding overload on ware and applications. The symbols for our models can be
the voltage reference and reducing power consumption in the found in Table 1.
voltage divider itself.
The Mote-Wake hardware design is robust and versatile,
but it has notable limitations. First, the high impedance of its 5.1 Power and battery lifetime
voltage divider network, while helping to save power, pre- The average power consumption, PAVG SO, of a system us-
cludes the connection of mainstream multimeters to the non- ing software polling event detection can be approximated as
inverting comparator inputs to observe the threshold voltage. follows:
Such devices do not offer sufficient input impedance to ob-
serve the voltage divider output without notably affecting it. PAVG SO =(FDC DDC )(PAC + PS1 )+
Although this poses no problem during operation, it com- (FMC DMC )(PAC + PRT )+
plicates debugging. Second, the Mote-Wake hardware lacks
(1 FDC DDC FMC DMC )(PAC + PS1) (1)
provisions for hot installation/removal due to the design of
the Hirose 51-pin connectors used for compatibility with
Crossbow MICA2 and MICAz Motes. This connector has no The average power consumption of an equivalent system that
mechanism to guarantee that supply rails make contact prior detects events using lucid dreaming can be approximated as
to I/O lines and, furthermore, there is no general mechanism follows:
to prevent corruption during an insertion/removal event on
any of the interfaces that are made accessible through this PAVG LD =(FDC DDC )(PAC + PS1 )+
connector. (FMC DMC )(PAC + PRT )+
(1 FDC DDC FMC DMC )(PZZ )+
4.3 Software PS2 + PMW (2)

We program the node hardware in NesC [13] within the Both models assume that data collection and communica-
TinyOS [16] operating system. The software side of lucid tion are mutually exclusive events; this assumption is highly
dreaming consists of a small extension to the run-time and accurate for the types of applications where the lucid dream-
some library functions. Note that the technique can also be ing technique is most appropriate (e.g., applications with in-
used within other operating environments such as MANTIS frequent events and infrequent communication).
OS [1], or even without a third-party runtime environment. Depending on the sensor network architecture used,
An interrupt service routine for wakeup is introduced. changes in processor state or radio state may have significant
This ISR does not presently do anything. Its execution is energy costs, i.e., the power consumption of the processor or
simply a side-effect of the interrupt bringing the Mote out radio may increase before they become available for com-
of sleep. The intent is that after the ISR executes, the Mote putation or communication. This effect can be modeled by
continues executing the code immediately after the point at increasing the average duration for event processing, DDC ,

7
 Table 1: Symbols

Variable Description Example value for ACM

PAVG LD Average power consumption for lucid dreaming 1:3  10 4 W
PAVG SO Average power consumption for polling solution 3:0  10 2 W
PAVG PR Average power consumption for event prediction No example value
PRT Power consumption of Mote radio in transmitting state 3:0  10 2 W
PAC Power consumption of Mote CPU in active state 2:4  10 2 W
PZZ Power consumption of Mote CPU in sleeping state 3:0  10 5 W
PS1 Power consumption of primary sensor and data acquisition system 5:7  10 3 W
PS2 Power consumption of secondary/wakeup sensor 0W
PMW Power consumption of Mote-Wake hardware 1:6  10 5 W
FDC Average frequency of an event resulting in data collection 1:2  10 4 Hz
FMC Average frequency of a communication transmission 1:2  10 5 Hz
DDC Average duration of an event resulting in data collection 3:0 s
DMC Average duration of a communication transmission 104:0 s
FT P Average frequency of true positives No example value
FFP Average frequency of false positives No example value
pFN False negative probability (type I error) No example value
pFP False positive probability (type II error) No example value
pT P True positive probability (1 pFN ) No example value
pT N True negative probability (1 pFP ) No example value

and/or average duration of communication events, DMC , to ture [12], and a duty cycling approach. The lucid dream-
include the state transition times. ing and 2.64 mW sensor approaches are guaranteed to detect
The literature reports values for PRT , PAC , and PZZ [5]. all events. If events are not predictable, the probability, per
PS1 and PMW were determined empirically in our lab. PS2 event, that the duty cycling approach misses an event is di-
is simply the result of our geophone being a self-powered rectly related to the proportion of time the system is inactive.
sensor. FDC , FMC , DDC , and DMC are taken from the authors’ As demonstrated in the figure, lucid dreaming consistently
experience with the ACM application. outperforms the 2.64 mW sensor approach by well over an
order of magnitude. It has lower power consumption than
We now illustrate the impact of changing the parameters
the duty cycling approach except when the number of events
appearing in our models for a number of applications, sen-
per day is extremely high, i.e., over 1,000, and the accept-
sors, and sensor network node architectures. As indicated
able event miss probability is very high, i.e., over 0.9. For
in Section 2, some researchers have considered the use of
the ACM application, the expected number of events per day
reduced and/or predictive duty cycling in order to reduce
is 10. In this application, the use of lucid dreaming increases
power consumption. These approaches cannot be used in
the battery life of the application from 10.91 days to 2,669
applications for which missing short events is unacceptable
days, i.e., the battery life is bounded only by the shelf life of
and events have durations that are short compared to the pro-
the AA batteries used to power the sensor nodes.
posed duty cycle period; note that the period must not be
short because initializing a Mote carries overhead. Even if The current Crossbow port of TinyOS supports the use of
missing some events is acceptable, in most applications it is low power states for the processor and radio between the
not desirable. individual samples in a series. During bench tests, this re-
Figure 5 displays the battery life of a sensor network node sulted in lower average power consumption during sampling
used in the ACM structural integrity monitoring application than reported for a MICA2 with a continuously-active mi-
as a function of the average number of events per day and crocontroller. However, even if we assume that the power
the tolerable probability of missing each event. We used a consumption if PAC is reduced to 1/10 the reported value,
the Mote-Wake hardware still increase the battery life in the
ACM application by 92.6.
typical battery life of 2,600 mAH for each of the AA alka-
line cells. This graph compares three approaches: (1) the
proposed lucid dreaming approach, a similar approach us- Next, we model schemes in which the arrival of events is
ing the lowest-power analog wake-up hardware for event- predicted. In such schemes, the Mote predicts the interval
driven applications (2.64 mW) we have found in the litera- to the next event, and then puts itself to sleep for that in-

8
 Lucid Dream
Battery life (days) 2.64 mW
Duty cycle

10000

1000

100

10

0 10000
0.2 1000
0.4 100
Event miss 0.6 Events per day
0.8 10
probability for 1 1
duty cycle
approach

Figure 5: Battery life as a function of event miss probability and FDC

terval. Any such predictor will produce both false negatives consumption using event prediction is then a variant of that
(not predicting an event that does occur in the interval) and for lucid dreaming (Equation 2) and is
false positives (predicting an event that does not occur in the
interval). False negatives decrease power consumption (be-
PAVG PR =(FDC ( pFP + (1 pFN ))DDC )(PAC + PS1)+
cause the Mote is not awakened) and increase the miss rate
(because it should be). False positives increase the power (FMC DMC )(PAC + PRT )+
consumption (because the Mote is awakened when it should (1 FDC ( pFP + (1 pFN ))DDC
not be) and do not affect the miss rate (because we assume FMC DMC )(PZZ ) (3)
the awakened Mote can determine that the event has been
falsely predicted.
Our model assumes Poisson arrival processes for actual It is important to point out that event prediction involves
events, true positives, and false positives. The mean frequen- a tradeoff between power consumption and the probability
cies of the latter are derived from the former. Let the mean of missing an event. Furthermore, this tradeoff depends on
frequency of true positives (correctly predicted events) be the nature of the predictor bias. For an unbiased predic-
FT P = FDC pT P = FDC (1 pFN ) tor, the false positive and false negative rates will be iden-
tical (pFP = pFN ). In this situation, the power consumption
and the mean frequency of false positives be for event prediction will be virtually identical to that of lu-
cid dreaming: Equation 3 converges to Equation 2. How-
FFP = FDC pFP
ever, the probability of missing an event in the event predic-
where the pFN is the false negative probability and pFP is the tion scheme will be pFN , while the miss probability in lucid
false positive probability. Our model for the average power dreaming will always be zero.

9
5.2 Bench test observations and measure- References
ments
[1] A BRACH , H., B HATTI , S., C ARLSON , J., DAI , H., ROSE ,
We have conducted bench tests of the Mote-Wake PCB. J., S HETH , A., S HUCKER , B., AND H AN , R. MANTIS:
When used to wake the microcontroller in response to vi- System support for MultimodAl NeTworks of In-situ
bration, its power consumption is 16.5 µW. We have suc- Sensors. In ACM International Wkshp. Wireless Sensor
Networks and Applications (Sept. 2003).
cessfully used in-system programming of Mote-Wake’s non-
volatile Maxim MAX5435LEZT-T potentiometers to vary [2] AGARWAL , Y., S CHURGERS , C., AND G UPTA , R.
Dynamic power management using on demand paging for
the event interrupt triggering threshold across a wide range
networked embedded systems. In Proc. Asia & South Pacific
of voltages. Measurements of the MICA2 in different power
Design Automation Conf. (Jan. 2005), pp. 755–759.
states [5], and the impact of the Mote-Wake board upon
[3] A KKAYA , K., AND YOUNIS , M. A survey on routing
the amount of time spent in each power state, indicate that
protocols for wireless sensor networks. Ad Hoc Networks 3,
the combined long-term average power consumption of the 3 (May 2005), 325–349.
MICA2 processor/radio board, the MDA300 data acquisition
[4] A L -K ARAKI , J., AND K AMAL , A. Routing techniques in
board, the Mote-Wake board, and the sensors, in the ACM
wireless sensor networks: A survey. IEEE Wireless
application, will be reduced from 29.8 mW to 121.8 µW by Communications 11, 6 (December 2004), 6–28.
using the Mote-Wake implementation of lucid dreaming, i.e.,
[5] A NASTASI , G., C ONTI , M., FALCHI , A., G REGORI , E.,
battery life will be increased from 10.91 days to seven years. AND PASSARELLA , A. Performance measurements of more
In other words, battery life will be limited only by the shelf sensor networks. In Prof. Int. Wkshp. on Modeling, Analysis,
life of the batteries. and Simulation of Wireless and Mobile Systems (Oct. 2004).
[6] Automated crack measurement.
http://www.iti.northwestern.edu/acm.
[7] BARZILAI , A. Improving a Geophone to Produce an
6 Conclusions and Future Work Affordable Broadband Seisometer. PhD thesis, Mechanical
Engineeering, Stanford University, January 2000.
We have described the technique of lucid dreaming, an ap- [8] B RINCKER , R., L AGO , T., A NDERSEN , P., AND
V ENTURA , C. Improving the classical geophone sensor
proach to achieving ultra-low power consumption in sensor
element by digital correction. Tech. rep., Pinocchio Data
networks for applications that are driven by unpredictable
Systems, 2005.
events. Lucid dreaming consists of simple, but powerful,
[9] C ROSSBOW T ECHNOLOGY I NCORPORATED. MICAz
analog threshold detection hardware (the Mote-Wake board)
Wireless Measurement System Datasheet, 2006. Document
that awakens the and triggers detailed high-rate sampling. Part Number 6020-0060-03 Rev A.
The threshold may be set via software. We have demon-
[10] D OWDING , C. H., AND M C K ENNA , L. M. Crack response
strated the utility of lucid dreaming within the context of an
to long-term and environmental and blast vibration effects.
application that monitors cracks on critical infrastructure el- Journal of Geotechnical and Geoenvironmental Engineering
ements, reducing power consumption to 1/245 that required 131, 9 (September 2005), 1151–1161.
by existing approaches. Mote-Wake and its adjunct soft- [11] D OWDING , C. H., O ZER , H., AND KOTOWSKY, M.
ware is available from us and can be deployed with standard Wireless crack measurment for control of construction
Crossbow Motes. vibrations. In Proceedings of the Atlanta GeoCongress
Our immediate plans are to expand the capabilities of (2006), Engineering in the Information Technology Age,
Mote-Wake, specifically to support more complex event gen- Geo-Institute of the American Society of Civil Engineers.
eration functions than simple threshold detection. More [12] D UTTA , P., G RIMMER , M., A RORA , A., B IBYK , S., AND
broadly, we plan to expand Mote-Wake into a general- C ULLER , D. Design of a wireless sensor network platform
purpose analog toolbox from which power- and rate-critical for detecting rare, random, and ephemeral events. In Proc.
portions of the sensor network application can be con- Int. Conf. on Information Processing in Sensor Networks
(Apr. 2005).
structed.
[13] G AY, D., L EVIS , P., VON B EHREN , R., W ELSH , M.,
For applications similar to that described in Section 3, we B REWER , E., AND C ULLER , D. The nesc language: A
will make the electronic Gerber format printed circuit board holistic approach to networked embedded systems. In
specifications available for immediate fabrication and use. Proceedings of Programming Language Design and
For applications running on host platforms other than the Implementation (PLDI) (June 2003).
Crossbow MICA2 and MICAz, or applications with sensing [14] KOTOWSKY, M., AND O ZER , H. Wireless data acquisition.
parameters that differ greatly, we hope that the schematic de- Crossbow Smart Dust Challenge, 2004.
picted in Figure 4 and described in Section 4 provide a useful http://www.iti.northwestern.edu/research/projects/dowding/-
starting point to other researchers and designers. micro.html.

10
[15] K URATA , N., J R ., B. F. S., RUIZ -S ANDOVAL , M.,
M IYAMOTO , Y., AND S AKO , Y. A study on building risk
monitoring using wireless sensor network MICA mote. In
Proc. Int. Conf. on Structural Health Monitoring and
Intelligent Infrastructure (Nov. 2003), pp. 353–357.
[16] L EVIS , P., M ADDEN , S., G AY, D., P OLASTRE , J.,
S ZEWCZYK , R., W OO , A., B REWER , E., AND C ULLER , D.
The emergence of networking abstractions and techniques in
tinyos. In Proceedings of the First Symposium on Networked
Systems Design and Implementation (NSDI) (2004).
[17] LYNCH , J. P., L AW, K. H., K IREMIDJIAN , A. S., K ENNY,
T. W., C ARRYER , E., AND PARTRIDGE , A. The design of a
wireless sensing unit for structural health monitoring. In
Proc. Int. Wkshp. on Structural Health Monitoring (Sept.
2001).
[18] O ZER , H. Wireless crack measurement for control of
construction vibrations. Master’s thesis, Department of Civil
and Environmental Engineering, Northwestern University,
2005.
[19] P OLASTRE , J., S ZEWCZYK , R., S HARP, C., AND C ULLER ,
D. The mote revolution: Low power wireless sensor
network devices. In Proceedings of the Symposium on High
Performance Chips (HOT-CHIPS) (August 2004).
[20] R ABEY, J. M., A MMER , M. J., DA S ILVA J R ., J. L.,
PATEL , D., AND ROUNDY, S. PicoRadio supports ad hoc
ultra-low power wireless networking. IEEE Computer (July
2000), 42–48.
[21] S INHA , A., AND C HANDRAKASAN , A. Dynamic power
management in wireless sensor networks. IEEE Design and
Test of Computers (Mar. 2001), 62–74.
[22] X U , N., R ANGWALA , S., C HINTALAPUDI , K. K.,
G ANESAN , D., B ROAD , A., G OVINDAN , R., AND E STRIN ,
D. A wireless sensor network for structural monitoring. In
Proc. Conf. on Embedded and Networked Sensor Systems
(Nov. 2004).
[23] Z HENG , R., H OU , J. C., AND S HA , L. Asynchronous
wakeup for ad hoc networks. In Proc. Int. Symp. on Mobile
Ad Hoc Networking and Computing (June 2003), pp. 35–45.

11