--===============1887408799==
Content-Type: text/plain; charset="big5"
Content-Transfer-Encoding: quoted-printable
X-MIME-Autoconverted: from 8bit to quoted-printable by pch.mit.edu id k232xegS022650

Maarten Hofman wrote:

>> Yes. That's why I said "usually". This of course means that there /are=
/
>> (IMO rare) conditions where this is not the case.
>=20
> Why rare? Wouldn't a proper design of sampling an analog signal (not fo=
r
> display on a screen by a PICmicro, but for other purposes) use the
> Nyquist=A1VShannon sampling theorem, and therefore be designed such tha=
t at
> the maximum frequency slightly more than two samples/period are taken?
> With only two samples/period the values could be anywhere within the
> range.=20

This is correct. I said rare, because I think that in most cases, you nee=
d
many more than two samples per period to fulfill your application
requirements. The theorem only says that if you take less than two sample=
s
per period of the highest frequency present in the signal, you get foldin=
g
in the frequency domain (that is, you measure frequencies that don't exis=
t
in the signal). It does /not/ say that taking two samples per period give=
s
you any useful data (for a given purpose).

So in most cases, you would measure many more than two samples per period.
The most frequent case probably is arbitrarily changing signals that do n=
ot
actually have a period as such (process signals like temperature,
acceleration, etc.). They have usually components of higher frequencies,
that also usually get filtered out, with filter cut-offs often way below
the maximum conversion frequency. Which then of course means that the
signal changes much slower than one sample at maximum and the next sample
at minimum.

I think it is rare that you get to the point where charging the holding c=
ap
becomes a problem (remember, that's all in the case of measuring one sing=
le
analog signal). Has anybody had such an application, where exactly this w=
as
a limiting factor?

At 5V Vdd, the internal resistance that charges Chold (25pF) is about 3k.
Assuming a small source resistance (op amp), that gives a time constant t=
RC
of 75ns. A full scale jump in the input takes about 7*tRC or 525ns to
dis/charge the cap to within 1/1023th (10 bit resolution) of the final
value. With a PIC running at 4MHz, that's less than a single instruction
cycle, with a PIC running at 20MHz, that's less than three instruction
cycles.=20

But you're not really measuring any useful data in this situation (in mos=
t
applications), because you're too close to the frequency limit. If you wa=
nt
to measure anything, you need more points, the jumps between points won't
be full-scale anymore, and the time to charge Chold will become smaller.

I don't say it doesn't happen... you can have a bigger source resistance,
for example. But I still think it's rare. Very rare.

Gerhard


--===============1887408799==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Disposition: inline
Content-Transfer-Encoding: 7bit

-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist

--===============1887408799==--