Flies in the face of DeltaSigma oversampling by 256X or 512X etc.
My understanding is that for a second order DS converter every doubling
of oversampling buys about 9 dB in SNR.
Dave
"steve" <bungalow_steve@yahoo.com> wrote in message
news:1106449193.210985.208270@z14g2000cwz.googlegroups.com...
> Analog Devices comments on their web site that for oversampling to work
> the sample rate cannot be over 10 times the analog signal bandwidth. So
> this defines the upper bounds on how much you can reduce the noise
> floor of an analog signal. So I am thinking what magical about 10x, but
> I suppose they are implying that if you had a 100hz bandwidth limited
> signal and oversampled it 1024 times at 1Ghz rate all the samples would
> be the same value anyway since if the signal is truly bandlimited at
> 100Hz it couldn't change quick enough to be different between samples.
> So whats the question? Well I guess I actually answered it as I am
> writing this post, funny how writing the question down does that!
>
> Thinking about it some more I suppose oversampling rates greater then
> 10x the signal bandwidth could help to reduce internal a/d noise, clock
> noise, switching power supply noise and any other high frequency noise
> that sneaks in after the external analog filter or though an external
> a/d voltage reference.
>
Reply by steve●January 23, 20052005-01-23
Tim Wescott wrote:
> steve wrote:
>
> > Jerry Avins wrote:
> >
> >>steve wrote:
> >>
> >> ...
> >>
> >>
> >>>Analog Devices quote from web site
> >>>"For oversampling to work, samples need to be taken at a rate no
> >
> > faster
> >
> >>>than 10 times the analog bandwidth"
> >>
> >>Could you say where that is on the website? The context is crucial.
> >>However high the oversampling ratio, the original signal can be
> >>reconstructed. What does "working" consist of?
> >>
> >>Jerry
> >>
> >>P.S. I garbled part of my previous message. It should have been (in
> >>part) "The signal is contaminated by any noise produced by the
> >
> > sampling
> >
> >>chain within the signal's band, and any noise outside the band can
be
> >>filtered out regardless of the sample rate."
> >>--
> >>Engineering is the art of making what you want from things you can
> >
> > get.
> >
> >
> >
> > Another thing I didn't understand was why the bandwidth is reduced
by
> > oversampling, I thought the whole purpose of oversampling was to
> > maintained the signal bandwidth at reduced noise at the expense of
> > faster a/d conversions. I certainly can see if you oversampled but
kept
> > the sampling rate constant you would reduce the bandwidth, but I
> > wouldn't call that oversampling (and from the online calculator,
the
> > sample rate does increase with increase samples, yet the bandwidth
> > still drops).
> >
> It's for that specific part -- the part reads out an analog reading
as a
> duty cycle of a fixed-frequency square wave, so your actual sampling
> rate is fixed. When they say "oversampling" they should be saying
> "digital filtering". I believe that the factor of 10 is for the
> physical sampling rate vs. the analog bandwidth, not the internal
> lowpass filtering.
>
> --
>
> Tim Wescott
> Wescott Design Services
> http://www.wescottdesign.com
Ok, I think I see now, I read the datasheet, I don't completely
understand what they are talking about but it sure looks like it a
limitation of the part (PWM output is variable but limited .5 to 10ms)
thanks
Reply by Jerry Avins●January 23, 20052005-01-23
steve wrote:
> "I don't see where it says that higher sample rates reduce the
> bandwidth."
>
> Well the title for one,
>
> "Option: Reduce noise by oversampling (at expense of bandwidth)"
>
> And the first sentence too, also just type in a oversample rate and you
> will see it recompute the lower bandwidth and display it in "bandwidth
> after oversampling"
>
> you must of read the wrong web page to miss it
I remembered the title after I hit "send". Here's my take: the bandwidth
is reduced by averaging back to what it would have been before over-
sampling allowed the bandwidth to increase. If that's it, it's a crock.
I'll bet again that the auther is tied to a codec with a built-in
anti-alias filter that tracks the sample rate. If the anti-alias cutoff
stayed fixed, averageing wouldn't reduce external noise. The filter
would effectively have done the averaging by holding down the bandwidth.
I get the impression that it's not an article I'd want to learn from.
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by Tim Wescott●January 23, 20052005-01-23
steve wrote:
> Jerry Avins wrote:
>
>>steve wrote:
>>
>> ...
>>
>>
>>>Analog Devices quote from web site
>>>"For oversampling to work, samples need to be taken at a rate no
>
> faster
>
>>>than 10 times the analog bandwidth"
>>
>>Could you say where that is on the website? The context is crucial.
>>However high the oversampling ratio, the original signal can be
>>reconstructed. What does "working" consist of?
>>
>>Jerry
>>
>>P.S. I garbled part of my previous message. It should have been (in
>>part) "The signal is contaminated by any noise produced by the
>
> sampling
>
>>chain within the signal's band, and any noise outside the band can be
>>filtered out regardless of the sample rate."
>>--
>>Engineering is the art of making what you want from things you can
>
> get.
>
> �����������������������������������������������������������������������
>
> see part 8 of
>
> http://www.analog.com/Analog_Root/static/techSupport/designTools/interactiveTools/accelerometer/adxl202.html
>
> Another thing I didn't understand was why the bandwidth is reduced by
> oversampling, I thought the whole purpose of oversampling was to
> maintained the signal bandwidth at reduced noise at the expense of
> faster a/d conversions. I certainly can see if you oversampled but kept
> the sampling rate constant you would reduce the bandwidth, but I
> wouldn't call that oversampling (and from the online calculator, the
> sample rate does increase with increase samples, yet the bandwidth
> still drops).
>
It's for that specific part -- the part reads out an analog reading as a
duty cycle of a fixed-frequency square wave, so your actual sampling
rate is fixed. When they say "oversampling" they should be saying
"digital filtering". I believe that the factor of 10 is for the
physical sampling rate vs. the analog bandwidth, not the internal
lowpass filtering.
--
Tim Wescott
Wescott Design Services
http://www.wescottdesign.com
Reply by steve●January 23, 20052005-01-23
"I don't see where it says that higher sample rates reduce the
bandwidth."
Well the title for one,
"Option: Reduce noise by oversampling (at expense of bandwidth)"
And the first sentence too, also just type in a oversample rate and you
will see it recompute the lower bandwidth and display it in "bandwidth
after oversampling"
you must of read the wrong web page to miss it
Reply by Jerry Avins●January 23, 20052005-01-23
steve wrote:
> Jerry Avins wrote:
>
>>steve wrote:
>>
>> ...
>>
>>
>>>Analog Devices quote from web site
>>>"For oversampling to work, samples need to be taken at a rate no
>
> faster
>
>>>than 10 times the analog bandwidth"
>>
>>Could you say where that is on the website? The context is crucial.
>>However high the oversampling ratio, the original signal can be
>>reconstructed. What does "working" consist of?
>>
>>Jerry
>>
>>P.S. I garbled part of my previous message. It should have been (in
>>part) "The signal is contaminated by any noise produced by the
>
> sampling
>
>>chain within the signal's band, and any noise outside the band can be
>>filtered out regardless of the sample rate."
>>--
>>Engineering is the art of making what you want from things you can
>
> get.
>
> �����������������������������������������������������������������������
>
> see part 8 of
>
> http://www.analog.com/Analog_Root/static/techSupport/designTools/interactiveTools/accelerometer/adxl202.html
>
> Another thing I didn't understand was why the bandwidth is reduced by
> oversampling, I thought the whole purpose of oversampling was to
> maintained the signal bandwidth at reduced noise at the expense of
> faster a/d conversions. I certainly can see if you oversampled but kept
> the sampling rate constant you would reduce the bandwidth, but I
> wouldn't call that oversampling (and from the online calculator, the
> sample rate does increase with increase samples, yet the bandwidth
> still drops).
I don't see where it says that higher sample rates reduce the bandwidth.
As for oversampling by more than ten not working, I will lay a small bet
that that's an artifact of the way the program works and doesn't refer
to the real world. I think the author just figured that nobody would
want more than 10, just as IBM figured that nobody would want more than
640 K of RAM in a PC.
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by Fred Marshall●January 23, 20052005-01-23
"John E. Hadstate" <jh113355@hotmail.com> wrote in message
news:V8OId.6180$Gj.1929@bignews3.bellsouth.net...
>
> "Jerry Avins" <jya@ieee.org> wrote in message
> news:35gokqF4lh664U1@individual.net...
>> steve wrote:
>>
>> ...
>>
>> > Analog Devices quote from web site
>> > "For oversampling to work, samples need to be taken at a
> rate no faster
>> > than 10 times the analog bandwidth"
>>
>> Could you say where that is on the website? The context is
> crucial.
>> However high the oversampling ratio, the original signal
> can be
>> reconstructed. What does "working" consist of?
>>
>
> I think the quote from the website is clear (and correct) if
> you read it while patting your head and rubbing your tummy
> ;-) In other words, you don't have to sample faster than
> about 10 times the analog bandwidth for oversampling to be
> used effectively. It does not say that you can't sample
> faster; it just says you don't have to sample faster.
>
> Besides being confusingly worded, it's a simplistic
> statement. There may be very good reasons for oversampling
> faster than 10x the highest analog frequency. Consider, for
> example, issues involving propagation delays and latencies.
If you parse the sentence, it seems to say:
fs<=10*B
which is probably not at all what anyone would agree with because values of
fs like fs=B would be allowed.
or, if you focus on the word "need" then:
min[fs]=10*B
and, the phrase
"need to be taken at a rate no faster than"
is a very awkward way to say:
"you don't need to sample at a rate higher than"
rather than
"you need to sample at a rate not higher than"
It looks like a marketing edit with a misplaced negative...
Fred
Reply by steve●January 23, 20052005-01-23
Jerry Avins wrote:
> steve wrote:
>
> ...
>
> > Analog Devices quote from web site
> > "For oversampling to work, samples need to be taken at a rate no
faster
> > than 10 times the analog bandwidth"
>
> Could you say where that is on the website? The context is crucial.
> However high the oversampling ratio, the original signal can be
> reconstructed. What does "working" consist of?
>
> Jerry
>
> P.S. I garbled part of my previous message. It should have been (in
> part) "The signal is contaminated by any noise produced by the
sampling
> chain within the signal's band, and any noise outside the band can be
> filtered out regardless of the sample rate."
> --
> Engineering is the art of making what you want from things you can
get.
>
=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=
=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=
=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF
see part 8 of
http://www.analog.com/Analog_Root/static/techSupport/designTools/interactiv=
eTools/accelerometer/adxl202.html
Another thing I didn't understand was why the bandwidth is reduced by
oversampling, I thought the whole purpose of oversampling was to
maintained the signal bandwidth at reduced noise at the expense of
faster a/d conversions. I certainly can see if you oversampled but kept
the sampling rate constant you would reduce the bandwidth, but I
wouldn't call that oversampling (and from the online calculator, the
sample rate does increase with increase samples, yet the bandwidth
still drops).
Reply by John E. Hadstate●January 23, 20052005-01-23
"Jerry Avins" <jya@ieee.org> wrote in message
news:35gokqF4lh664U1@individual.net...
> steve wrote:
>
> ...
>
> > Analog Devices quote from web site
> > "For oversampling to work, samples need to be taken at a
rate no faster
> > than 10 times the analog bandwidth"
>
> Could you say where that is on the website? The context is
crucial.
> However high the oversampling ratio, the original signal
can be
> reconstructed. What does "working" consist of?
>
I think the quote from the website is clear (and correct) if
you read it while patting your head and rubbing your tummy
;-) In other words, you don't have to sample faster than
about 10 times the analog bandwidth for oversampling to be
used effectively. It does not say that you can't sample
faster; it just says you don't have to sample faster.
Besides being confusingly worded, it's a simplistic
statement. There may be very good reasons for oversampling
faster than 10x the highest analog frequency. Consider, for
example, issues involving propagation delays and latencies.
Reply by steve●January 23, 20052005-01-23
John E. Hadstate wrote:
> "steve" <bungalow_steve@yahoo.com> wrote in message
> news:1106449193.210985.208270@z14g2000cwz.googlegroups.com...
> > Analog Devices comments on their web site that for
> oversampling to work
> > the sample rate cannot be over 10 times the analog signal
> bandwidth. So
> > this defines the upper bounds on how much you can reduce
> the noise
> > floor of an analog signal. So I am thinking what magical
> about 10x, but
> > I suppose they are implying that if you had a 100hz
> bandwidth limited
> > signal and oversampled it 1024 times at 1Ghz rate all the
> samples would
> > be the same value anyway since if the signal is truly
> bandlimited at
> > 100Hz it couldn't change quick enough to be different
> between samples.
> > So whats the question? Well I guess I actually answered it
> as I am
> > writing this post, funny how writing the question down
> does that!
> >
> > Thinking about it some more I suppose oversampling rates
> greater then
> > 10x the signal bandwidth could help to reduce internal a/d
> noise, clock
> > noise, switching power supply noise and any other high
> frequency noise
> > that sneaks in after the external analog filter or though
> an external
> > a/d voltage reference.
> >
>
> Once upon a time, I designed a controller to sample at 1200
> Hz. even though it was for a system whose natural time
> constants were measured in tens of seconds. The reason was
> that I could easily apply a post sampling digital filter
> that would reduce 60 Hz and 120 Hz noise picked up from the
> environment to levels that were below the quantization
> level. Also, it was a lot easier to implement an
> anti-aliasing analog filter for 600 Hz than it would have
> been to implement one for 1 Hz.
Yes, very good reasons for oversampling greater then 10x signal BW,
that 1Hz filter would be huge, thanks for the info.