# Sampling Question

Started by August 1, 2008
```Suppose I take samples at once per minute for a data logger. Do I
still need an anti-aliasing filter? The sampling freq would be 1/T
where T=60 secs...so how is this possible? The values of  the
capacitors etc would be hugh. Of course I could over-sample1000s of
times...is this the norm?

K.
```
```On 1 Aug, 22:12, kronec...@yahoo.co.uk wrote:
> Suppose I take samples at once per minute for a data logger. Do I
> still need an anti-aliasing filter? The sampling freq would be 1/T
> where T=60 secs...so how is this possible? The values of &#4294967295;the
> capacitors etc would be hugh. Of course I could over-sample1000s of
> times...is this the norm?

What kind of process is this? What will you use the data for?

Rune
```
```kronecker@yahoo.co.uk wrote:
> Suppose I take samples at once per minute for a data logger. Do I
> still need an anti-aliasing filter? The sampling freq would be 1/T
> where T=60 secs...so how is this possible? The values of  the
> capacitors etc would be hugh. Of course I could over-sample1000s of
> times...is this the norm?

An anti-alias filter is rarely used for such measurements. Similarly,
neither for tide measurements, daily temperatures, or closing stock
prices. For many applications this is fine, but it imposes limits on how
that data can be used.

Jerry
--
Engineering is the art of making what you want from things you can get.
&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;
```
```On 2008-08-01, kronecker@yahoo.co.uk <kronecker@yahoo.co.uk> wrote:
> Suppose I take samples at once per minute for a data logger. Do I
> still need an anti-aliasing filter?

Imagine you took samples of a GPS position once a minute.  If you put it
in a car going cross-country you might get perfectly usable data.  If you
put it in an Indy car on an oval track you might have trouble telling the
difference between the car going around the track fast vs forward slow
vs backwards (!).  So you have to know something about what you're
measuring before you know what you stand to lose at your sample rate.

Keep in mind that you don't have to sample at your logging rate.  You
could sample at 100Hz and filter that in software and decimate by 60000.
Choose your sample rate to be highly over-sampled for your process, and
then you can log at any rate greater than the Nyquist rate.

--
<ben@ben.com>
http://www.ben.com/
```
```<kronecker@yahoo.co.uk> wrote in message
> Suppose I take samples at once per minute for a data logger. Do I
> still need an anti-aliasing filter? The sampling freq would be 1/T
> where T=60 secs...so how is this possible? The values of  the
> capacitors etc would be hugh. Of course I could over-sample1000s of
> times...is this the norm?
>
>
> K.

Only you can really answer the question "do I still need an anti-aliasing
filter?".  Your idea of over-sampling and then digital filtering might well
be a good one.  No capacitors.....

You might want to examine what aliasing does and what that means to you.
- If you are going to do frequency analysis then you need to know that
higher frequencies can be aliased into the analysis range.  If you aren't
going to do that then does it matter?
- If you are going to do reconstruction then you aren't going to be able to
do it perfectly and perhaps it will even be done terribly.  Reconstruction
is closely related to interpolation - so if you're planning to interpolate
to a high degree then that might be a problem too.
- If all you're going to do is compute statistics then maybe you will be OK
as the samples you take *are* valid samples.  We often are forced to live
with this situation.

Sometimes the best thing to do is just try it and see.
For example, compare oversampled data that's "properly" decimated with
undersampled data from the same time span in your process - whatever that
may be.  Are the results comparable or not?

Fred

```
```kronecker@yahoo.co.uk wrote:
> Suppose I take samples at once per minute for a data logger. Do I
> still need an anti-aliasing filter? The sampling freq would be 1/T
> where T=60 secs...so how is this possible? The values of  the
> capacitors etc would be hugh. Of course I could over-sample1000s of
> times...is this the norm?
>
>
> K.

Nyquist never said you _had_ to have anti-aliasing filters.  This may
not answer your question, but it may shed some light:
http://www.wescottdesign.com/articles/Sampling/sampling.html.

If the thing you're sampling just doesn't have any content above 1/120th
of a Hz, then you don't need to anti-alias at all.  If it has content at
10Hz and you desperately need to know what that content is, then
sampling at 1/60th Hz will never, ever get you what you want.

Consider that you may do better sampling once/three minutes, and saving
away the minimum, maximum, and average signal seen in that period.

And read Ben's post - he makes some good points, from a time-domain
point of view.  Sometimes we get so wound up in the frequency-domain
analysis that we forget that we live in the time domain, and a little
bit of reasoning in the here-and-now will save us a long trip through
mathemagic land.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" gives you just what it says.
See details at http://www.wescottdesign.com/actfes/actfes.html
```
```On Aug 2, 8:31 am, Rune Allnor <all...@tele.ntnu.no> wrote:
> On 1 Aug, 22:12, kronec...@yahoo.co.uk wrote:
>
> > Suppose I take samples at once per minute for a data logger. Do I
> > still need an anti-aliasing filter? The sampling freq would be 1/T
> > where T=60 secs...so how is this possible? The values of  the
> > capacitors etc would be hugh. Of course I could over-sample1000s of
> > times...is this the norm?
>
> What kind of process is this? What will you use the data for?
>
> Rune

It's strain measurements.

K.
```
```kronecker@yahoo.co.uk wrote:
> Suppose I take samples at once per minute for a data logger. Do I
> still need an anti-aliasing filter? The sampling freq would be 1/T
> where T=60 secs...so how is this possible? The values of  the
> capacitors etc would be hugh. Of course I could over-sample1000s of
> times...is this the norm?

Remember it is RC that makes a filter, either large C or
large R will work.  With an appropriate amplifier, you can make
good filters without huge capacitors.

But you really need to understand the problem.  The human
auditory system is pretty sensitive to non-harmonic signals,
so the filters are pretty important for audio signals.

The visual system is somewhat less sensitive to aliasing.
Also, video hardware naturally has some filtering due to
the width of the sensor elements (or beam size in scanning
electron tube devices).   Even more, the color TV standards
put the color information where high spatial frequency video
information should go, but most people aren't especially
bothered by it.

At low frequencies over sampling and digital filtering
are fairly easy to do, so that may be a fine solution.

Audio and video are fairly easy, as there are upper limits
to the audio frequency and spatial frequency that we can
hear/see.

-- glen

-- glen

```
```On 2 Aug, 02:13, kronec...@yahoo.co.uk wrote:
> On Aug 2, 8:31 am, Rune Allnor <all...@tele.ntnu.no> wrote:
>
> > On 1 Aug, 22:12, kronec...@yahoo.co.uk wrote:
>
> > > Suppose I take samples at once per minute for a data logger. Do I
> > > still need an anti-aliasing filter? The sampling freq would be 1/T
> > > where T=60 secs...so how is this possible? The values of &#4294967295;the
> > > capacitors etc would be hugh. Of course I could over-sample1000s of
> > > times...is this the norm?
>
> > What kind of process is this? What will you use the data for?
>
> > Rune
>
> It's strain measurements.

If these are static strains (which the sampling rate might indicate)
I wouldn't bother with the antialias filters.

Rune
```
```Rune Allnor wrote:
> On 2 Aug, 02:13, kronec...@yahoo.co.uk wrote:
>> On Aug 2, 8:31 am, Rune Allnor <all...@tele.ntnu.no> wrote:
>>
>>> On 1 Aug, 22:12, kronec...@yahoo.co.uk wrote:
>>>> Suppose I take samples at once per minute for a data logger. Do I
>>>> still need an anti-aliasing filter? The sampling freq would be 1/T
>>>> where T=60 secs...so how is this possible? The values of  the
>>>> capacitors etc would be hugh. Of course I could over-sample1000s of
>>>> times...is this the norm?
>>> What kind of process is this? What will you use the data for?
>>> Rune
>> It's strain measurements.
>
> If these are static strains (which the sampling rate might indicate)
> I wouldn't bother with the antialias filters.

It depends on the sensitivity, the nature of the phenomena, and the
cut-off frequency of the filter. Water-hight gauges used to measure
rivers and tides usually incorporate a stilling basin to exclude the
effects of waves and ripples. The averaging time is usually much shorter
than the sampling interval. The strain gauges in scales for infants and
animals are filtered to give steady readings while the loads move.

Jerry
--
Engineering is the art of making what you want from things you can get.
&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;&#4294967295;
```