Analog Devices AN601 Application Notes

AN-601

a

APPLICATION NOTE

One Technology Way • P.O. Box 9106 • Norwood, MA 02062-9106 • Tel: [phone] • Fax: [phone] • www.analog.com

Minimizing Power Consumption

of iMEMS

Introduction

Portable battery-powered devices are perhaps the larg­est growth market segment today. In an effort to reduce
the size and weight of these devices, battery capacity is
often minimized. To maintain good performance,
designers are forced to carefully examine their circuits
for ways to decrease power consumption.

This application note will outline methods to reduce the
power consumption of iMEMS accelerometers using
both hardware and software techniques. A special sec­tion will cover some techniques that are specific to
certain parts.

Basic Methods

There are three basic methods to reduce power
consumption:

• Lower the supply voltage

• Turn off the accelerometer when measurements are
not taking place

• Use clever software

All of the techniques outlined in this application note are
no more than extensions of these methods.

Lower the Supply Voltage

Often the most straightforward and lowest cost way to
reduce power consumption is to simply reduce the supply
voltage. Figure 1 shows the typical power consumption ver­sus the supply voltage for several iMEMS accelerometers.

While there are great power savings to be had by simply
lowering the supply voltage, there is a price to be paid
as well. As all these accelerometers are ratiometric, low­ering their supply voltage will lower the sensitivity by
roughly the same ratio. The exception to this is the
PWM outputs of the ADXL202/ADXL210. These outputs
remain fairly constant as the supply voltage changes.

®

Accelerometers

by Harvey Weinberg

mW

Figure 1. Power Consumption vs. Supply Voltage for
Several Accelerometers

If the entire system is ratiometric (i.e., the A/D reference
voltage is proportional to V
ity is not a problem.

A potentially more serious problem is that the accel­erometer’s noise performance is generally degraded as
the supply voltage is reduced. This cannot be mitigated
by using a ratiometric system and must be kept in mind
during system design.

Turn Off the Accelerometer

Turning off the accelerometer when measurement is not
occurring can result in great power savings. This is par­ticularly true in applications where the sampling rate is
low. Figure 2 illustrates the average power consumed by
an ADXL150 versus the power cycling (sampling) fre­quency (for V
Note that the Nyquist criteria must be satisfied in any
case, and the sampling frequency must be at least twice
the input frequency.

30.0

25.0

20.0

15.0

10.0

5.0

0.0

2.5 6.53.5

ADXL250

ADXL105

= 5 V and a 25 s A/D conversion time).

DD

ADXL150

4.5 5.5

V

), this reduction of sensitiv-

DD

ADXL202/ADXL210

REV. 0

© Analog Devices, Inc., 2002

AN-601

10

1

0.1

0.01

POWER CONSUMPTION – mW

0.001
11000010

SAMPLING FREQUENCY – Hz

100 1000

Figure 2. ADXL150 Average Power Consumption vs.
Power Cycling Frequency

In iMEMS accelerometers, the turn-on time is mainly a
function of the bandwidth. While higher bandwidths will
allow lower power operation (due to faster power
cycling rates), doing so generally results in more noise.
Many systems include an antialiasing filter between the
accelerometer and the A/D converter. The time constant
of this antialiasing filter must also be considered when
power cycling.

Table I shows the approximate turn-on time (including
the internal low-pass filter) for several accelerometers.
Faster turn-on times allow the user to very quickly turn
on the accelerometer, measure the acceleration, and
then turn off the accelerometer.

Table I. Accelerometer Turn-On Time

Model Bandwidth Turn On

ADXL202/ADXL210 5000 Hz 460 s
ADXL105 12 kHz 700 s
ADXL150/ADXL250 1 kHz 360 s
ADXL190 400 Hz 750 s

Earthquakes are low frequency (below 20 Hz), low g,
multiaxis events, so we will use an ADXL202 and sample
at 40 Hz (25 ms) to avoid aliasing.

In this design, we can assume that vibrations of less
than 200 m

g

can be ignored (an earthquake is defined as

having sufficient energy to cause accelerations greater

g

than 200 m

). So our peak-to-peak noise floor must be

less than 200 mg. Using a peak-to-peak to rms ratio of

6.6, we find:

rms noise

= 200 mg/6.6 = 30.3 mg

So we will select a bandwidth that will result in an rms
noise floor that is less than 30.3 mg.

Noise Noise Density bandwidth=¥¥16.

For the ADXL202 with a typical

Noise Hz bandwidth=¥¥500 g 1.6m

noise density

of 500 g/

Hz:

÷

Rearranging the equation;

Bandwidth noise noise density=¥()(. )

22

16

In this example, the maximum bandwidth is approxi­mately 2.3 kHz. Using the closest standard value, we can
set the bandwidth to 2 kHz. Therefore

C

and CY are

X

0.0022 F and the noise floor is approximately 28 m
rms (185 mg peak to peak).

For the ADXL202, the turn-on time is approximately:

TCms

=¥=160 0 3.

ON X

where

C

is in F. The 0.3 s is the turn-on time of the

X

accelerometer itself, while the 160 

C

(or

C

X

) term is

Y

the settling time of the bandwidth limiting filter. Using a

0.0022 F capacitor, the turn-on time to steady state is
approximately 650 s. The analog outputs and an A/D
converter will be used, so a conversion time of 25 s
must be added. So the total ontime is 675 s.

g

Adding a single-pole low-pass filter (antialiasing) to the
accelerometer output will lengthen the turn-on time by
5/(2

f

) seconds, where f is the corner frequency of the filter.

For example, restricting the bandwidth of an acceler­ometer to 50 Hz by adding a single-pole low-pass filter
would add 15.9 ms to the settling time.

Low Power Design Example

Most low power designs appear in devices that measure
movements that take place infrequently. These applica­tions are ideal for power cycling. A good example is an
automatic shutoff gas valve. In the event of an earth­quake, the valve shuts down the natural gas supply to
prevent ruptured gas pipes from leaking. Minor tremors
(as can be created by large trucks passing by) and single
impulse shocks (as would be generated by bumping into
the valve) should be ignored.

Therefore the average power consumed is:

600 0 675 25 16 2



Amsms A¥=(. ) .

with a 5 V supply. Lowering the supply voltage to 3 V will
reduce the average current consumption to 10.8 A.

Obviously, 200 mg peak-to-peak noise is too high to
make a good measurement. Therefore, if a measure-

g

ment of greater than 200 m

is made, the sampling
speed can be increased for a few seconds to 800 Hz and
groups of 20 samples can be averaged. This will bring
the noise floor down to approximately 6 m
peak to peak), allowing more precise measurement.
More power will be used at this time, but as this hap­pens infrequently, the average power consumed will
still be under 20 A.

–2–

g

rms (42 m

REV. 0

g

AN-601

Software can then determine if the higher acceleration
(>200 m

g

) was actually caused by an earthquake or just
a single impulse event due to jostling and take appropri­ate action.

The ADXL202/ADXL210

With its PWM outputs, the ADXL202/ADXL210 is a special
case and merits special attention when power cycling.

Using the PWM Outputs

The duty cycle modulator of the ADXL202/ADXL210
runs asynchronously to the rest of the accelerometer.
Since we have no way of knowing what state the PWM
output will be in when the accelerometer analog output
data is valid, we must wait at least one T2 period after
the specified turn-on time (to ensure the data is valid).

This leads to two conclusions:

1. Using the analog outputs and an A/D converter

rather than the PWM outputs will allow faster power
cycling and therefore the lowest power operation.

2. If we choose to use the PWM outputs with power

cycling, we should use the shortest T2 time possible.
However, using a short T2 implies using a fast
counter, which is normally inconsistent with low
power operation.

Although not explicitly specified in its data sheet, the
maximum PWM frequency is typically 5 kHz (R

SET

=

25 kW). Therefore, the additional overhead needed to
use the PWM outputs will be at least 400 s (two T2
periods—one period wait for valid data and another
period for measurement) to calculate the counter time.
The ADXL202 has a total dynamic range of ±4

g

or

Ω8000 mgΩ. At T2 = 5 kHz, 1 mg per count = [(1/5 kHz)/
8000 m

g

] total range or 250 ns per mg). In addition, if we

want to use such a fast T2 period and maintain 10 m

g

resolution (for the ADXL202), we would need a counter
that counts in 250 ns increments.

Even if only 32 mg resolution (roughly equivalent to 2∞
of tilt) were sufficient, the counter increment would rise
to 800 ns, still too fast for most low power 8-bit
microcontrollers.

Using a counter that runs at 1 MHz (1 s per count) and
maintaining 10 m

g

resolution, we would have to run T2
at approximately 1.25 kHz. In this case, the T2 overhead
would rise to 1.6 ms compared to the approximately
25 s that an A/D converter would take to complete a
conversion. Clearly, the overall system design becomes
more involved when we are looking to minimize power
consumption and use the PWM outputs of the ADXL202/
ADXL210. Often the best choice when looking to minimize
both component cost and power consumption is to use
an A/D converter along with the ADXL202/ADXL210.

The exception to this is when the sampling rate is very
low. In a system where a measurement is only made
from time to time, the additional time required to use the

PWM output is not a great handicap. The high resolution
and low cost (i.e., no A/D converter required) may be
more important than the additional power consumption.

Reducing Turn-On Time with Charge Conservation

In the majority of applications, most of the turn-on time of
the ADXL202/ADXL210 is attributable to the time constant
of the bandwidth limiting filter (formed by the internal
32 kW resistors and C

or CY). The CX and CY values are

X

normally dictated by the resolution required in the appli­cation. Many tilt sensing applications are low speed in
nature and are therefore good candidates for power
cycling. However, these applications often require high
resolution and large C

and CY values are mandated,

X

resulting in long turn-on times. In many cases, the turn-on
time can be greatly reduced by conserving the charge on
C

and CY.

X

Figure 3 shows a typical circuit used for charge conser­vation. C
and S3) just prior to the removal of V

and CY are switched out of the circuit (via S2

X

to the ADXL202/

DD

ADXL210 (via S1). They are switched back in just after (a
few ms) V

is applied to the ADXL202/ADXL210. This

DD

removes any path for the filter capacitors to discharge
(other than leakage through C

, CY, and the switches

X

themselves). In an actual system, a CMOS analog switch
would be used for S1, S2, and S3.

By keeping the filter capacitors (CX and CY) from discharg­ing, we can speed up the settling time as shown in Figure 4.
This allows us to turn off the accelerometer quickly and
conserve power. Note that it takes the same time to arrive
at steady state with or without charge conservation.

V

R

SET

ADXL202

COM

COM

T2

X

Y

X

OUTYOUT

V

DD

DD

FILT

FILT

S1

S2

S3

C

C

Y

X

Figure 3. Charge Conservation Circuit

If the goal is to arrive at measurements that are accurate
within a certain tolerance of the steady state output
(100 m

g

as an example), it can be realized more quickly

by using charge conservation. Figure 4 shows the C

Y

(Pin 11) output of a system that is being power cycled
10 ms ON/17 ms OFF, for an effective bandwidth of
about 18 Hz. With V
tion architecture with C

= 5 V and using a charge conserva-

DD

and CY 0.1 F, the average

X

current used is:

REV. 0

–3–