3–379
Motorola Sensor Device Data
For More Information On This Product,
Go to: www.freescale.com
NOISE FILTERING TECHNIQUES
AND CONSIDERATIONS
For mitigating the effects of this sensor noise, two general
approaches are effective, low pass filtering with hardware,
and low pass filtering with software. When filtering with hard-
ware, a low–pass RC filter with a cutoff frequency of 650 Hz
is recommended. A 750 ohm resistor and a 0.33
μ
F capacitor
have been determined to give the best results (see Figure 2)
since the 750 ohm series impedance is low enough for most
A/D converters.
Figure 2. Integrated Pressure Sensor with RC LP Filter to Filter Out Noise
1.0 F
IPS
0.33 F
A/D
3
5 V
0.01 F
750
2
1
This filter has been tested with an MC68HC705P9 micro-
controller which has a successive approximation A/D con-
verter. Successive approximation A/D’s are generally
compatible with the DC source impedance of the filter in
Figure 2. Results are shown in Figure 4.
Some A/D’s will not work well with the source impedance of
a single pole RC filter. Please consult your A/D converter tech-
nical data sheet if input impedance is a concern. In applica-
tions where the A/D converter is sensitive to high source
impedance, a buffer should be used. The integrated pressure
sensor has a rail–to–rail output swing, which dictates that a
rail–to–rail operational amplifier (op amp) should be used to
avoid saturating the buffer. A MC33502 rail–to–rail input and
output op amp works well for this purpose (see Figure 3).
1.0 F
IPS
0.33 F
A/D
3
5 V
0.01 F
750
2
1
+
–
MC33502
Figure 3. Use a Rail–to–Rail Buffer to Reduce Output Impedance of RC Filter
Averaging is also effective for filtering sensor noise. Averag-
ing is a form of low pass filtering in software. A rolling average
of 8 to 64 samples will clean up most of the noise. A 10 sample
average reduces the noise to about 2.5 mV peak to peak and
a 64 sample average reduces the noise to about 1 mV peak
to peak (see Figures 5 and 6).
This method is simple and requires no external compo-
nents. However, it does require RAM for data storage, extra
computation cycles and code. In applications where the
microcontroller is resource limited or pressure is changing
relatively rapidly, averaging alone may not be the best solu-
tion. In these situations, a combination of RC filtering and a
limited number of samples gives the best results. For exam-
ple, a rolling average of 4 samples combined with the RC filter
in Figure 2 results in a noise output on the order of 1 mV peak
to peak.
Another important consideration is that the incremental
effectiveness of averaging tends to fall off as the number of
samples is increased. In other words, the signal–to–noise
(S/N) ratio goes up more slowly than the number of samples.
To be more precise, the S/N ratio improves as the square root
of the number of samples is increased. For example, increas-
ing the number of samples from 10, in Figure 5, to 64, in
Figure 6, reduced noise by a factor of 2.5.
F
Freescale Semiconductor, Inc.
n
.