simplest possible voltage doubler?

This is "dithering" and you need to be a bit picky about the probablity
distribution of the dithering signal.

I listed a bunch of useful references in a note in Rev. Sci. Instrum.
in 1999.

Sloman A.W. "Comment on 'Noise averaging and measurement resolution"
Review of Scientific Instruments, 70 4734 (1999)

An asynchronous triangular wave would produce a uniform distribution of
off-sets, and this isn't good - you ideally want a a triangular
probability distribution, and random noise comes a lot closer to this.

----------
Bill Sloman, Nijmegen

 

[Gary Pace]

Could you change the ADC reference to 2.048V ?
Then you wouldn't need an amplifier at all
You'd need lots of local C to get the noise down, but since the signal is
from a pot I doubt the time constant would matter

"Michael Noone" <mnoone.uiuc.edu@127.0.0.1> wrote in message
news:Xns969CF1C3C8594mnooneuiucedu127001@216.196.97.136...

Hi - I need to take a signal coming from a 3K potentiometer that ranges
from 0.44-1.77V and double it (so I'd have an output of 0.88-3.54V). That
output will be then fed into an ADC. I have available on the board a very
stable 5V supply, and a less clean 6V supply. The ADC is 10b with a 4.096
reference voltage, and I'd like to keep this circuit accurate enough to a
single bit. My first thought is a non inverting amplifier - but that's one
chip and 4 resistors per circuit, and since I need three of these on the
same board, 1 chip and 12 resistors total. That's quite a bit... Is there
any better way? I've seen some dedicated chips designed for current
monitoring that would amplify a signal by a much higher gain - 50 or 100
without any external components at all, is there anything like this that
would work for me?

Thanks for your help!

-MJ Noone

 

[John Popelish]

Gary Pace wrote:

Could you change the ADC reference to 2.048V ?
Then you wouldn't need an amplifier at all
You'd need lots of local C to get the noise down, but since the signal is
from a pot I doubt the time constant would matter

I can't figure out how the O.P. is worrying about fractional percent
resolution and accuracy with the signals coming from pots. I haven't
seen many pots with fractional percent linearity, repeatability or
noise. Seems like garden variety 1% parts should keep up with the pots.

 

[Paul E. Schoen]

Look into Microchip MCP6S26
http://ww1.microchip.com/downloads/en/DeviceDoc/21117a.pdf.

It has a programmable gain of 1, 2, 4, 5, 8, 10, 16, 32 and has a front end
6 channel MUX. The gain error is only 1%, but linearity looks very good. It
comes in 14 pin SOIC and TSSOP packages.

They also come in single channel MCP6S21.

Paul E. Schoen, President
P S Technology, Inc.
www.pstech-inc.com


"Michael Noone" <mnoone.uiuc.edu@127.0.0.1> wrote in message
news:Xns969CF1C3C8594mnooneuiucedu127001@216.196.97.136...

Hi - I need to take a signal coming from a 3K potentiometer that ranges
from 0.44-1.77V and double it (so I'd have an output of 0.88-3.54V). That
output will be then fed into an ADC. I have available on the board a very
stable 5V supply, and a less clean 6V supply. The ADC is 10b with a 4.096
reference voltage, and I'd like to keep this circuit accurate enough to a
single bit. My first thought is a non inverting amplifier - but that's one
chip and 4 resistors per circuit, and since I need three of these on the
same board, 1 chip and 12 resistors total. That's quite a bit... Is there
any better way? I've seen some dedicated chips designed for current
monitoring that would amplify a signal by a much higher gain - 50 or 100
without any external components at all, is there anything like this that
would work for me?

Thanks for your help!

-MJ Noone

 

Like I said, Farnell carry these parts.

When I went to the Farnell website - www.farnell.com - and bored down
to the North American branch - Farnell in one - they list the part
under order code 3098643 as "in stock".

The great advantage of the thin film dividers is the ratio tracking -
the gain is about ten times more stable than you will get from discrete
0.1% thin film resistors.

-----------
Bill Sloman, Nijmegen

 

Like I said, Farnell carry these parts.

When I went to the Farnell website - www.farnell.com - and bored down
to the North American branch - Farnell in one - they list the part
under order code 3098643 as "in stock".

The great advantage of the thin film dividers is the ratio tracking -
the gain is about ten times more stable than you will get from discrete
0.1% thin film resistors.

-----------
Bill Sloman, Nijmegen

 

[Mochuelo]

On Sun, 24 Jul 2005 20:28:55 GMT, Fred Bloggs <nospam@nospam.com>
wrote:

Hi - I need to take a signal coming from a 3K potentiometer that ranges
from 0.44-1.77V and double it (so I'd have an output of 0.88-3.54V). That
output will be then fed into an ADC. I have available on the board a very
stable 5V supply, and a less clean 6V supply. The ADC is 10b with a 4.096
reference voltage, and I'd like to keep this circuit accurate enough to a
single bit.

4.096/1024=4mV resolution. Neglecting differential non-linearity, you
will measure voltage within the 0.44-1.77V to within 4mV accuracy.
Assuming perfectly errorless scaling into the 0.88-3.54V range, measured
with 4mV accuracy, you then get 2mV accuracy on the original signal. If
you want 2mV accuracy, then you can alternately sum 2mV and 0mV with the
original signal to obtain the same estimate. So you don't need a
doubler, you need a summer- and that is two resistors.

Can you sketch a schematic of this? I think that if you use only two
resistors, with no active part, you are going to lose even more
dynamic range. Besides, don't you need some electronic switch to
alternate between the two configurations (+0 mV and +2 mV)?

 

[Ken Smith]

In article <ud18e1pcppul4smmdar7ovakpdo1r88u0u@4ax.com>,
John Larkin <jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:
[...]

Even more fun is to sum with noise or an asynchronous triangle, smear
the samples several LSBs, and software lowpass filter. If the native
noise is enough (dicey at 10 bits, almost guaranteed at 14-16) just
filter.

If you add a high frequency boost to the anti-alias filter, you can
compensate for the digital side's low pass. This high frequency boost is
often enough to boost the noise above one LSB.

A high frequency boost also tends to give you the best sort of noise for
the job. You really want all of the noise's energy to be above the cut
off frequency of the low pass filter you are using. Any noise below the
cut off just adds noise to the measurement without helping to smear out
the lower bits.

If the ADC is part of a servo system, there normally is a natural low
passing action as part of the design. In these cases, you may not even
have to add a filter to the design. The zero you add in analog land can
help to make the phase margin of the servo loop as well as removing the
cogging effects of the LSB.


--
--
kensmith@rahul.net forging knowledge

 

[John Larkin]

On 25 Jul 2005 01:58:00 -0700, bill.sloman@ieee.org wrote:

Like I said, Farnell carry these parts.

When I went to the Farnell website - www.farnell.com - and bored down
to the North American branch - Farnell in one - they list the part
under order code 3098643 as "in stock".

The great advantage of the thin film dividers is the ratio tracking -
the gain is about ten times more stable than you will get from discrete
0.1% thin film resistors.

Vishay says they'll guarantee 5 ppm tc tracking on their 0603's if all
the parts come from the same reel. But their delivery is insane, too.
So we'll just buy a reel of somebody else's 10 ppm parts.

We were playing with the math of making various inverting and
non-inverting gains from a given number of identical resistors; lots
of combinations emerge.


John

 

[John Larkin]

On Mon, 25 Jul 2005 14:03:33 +0000 (UTC), kensmith@green.rahul.net
(Ken Smith) wrote:

In article <ud18e1pcppul4smmdar7ovakpdo1r88u0u@4ax.com>,
John Larkin <jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:
[...]
Even more fun is to sum with noise or an asynchronous triangle, smear
the samples several LSBs, and software lowpass filter. If the native
noise is enough (dicey at 10 bits, almost guaranteed at 14-16) just
filter.


If you add a high frequency boost to the anti-alias filter, you can
compensate for the digital side's low pass. This high frequency boost is
often enough to boost the noise above one LSB.

A high frequency boost also tends to give you the best sort of noise for
the job. You really want all of the noise's energy to be above the cut
off frequency of the low pass filter you are using. Any noise below the
cut off just adds noise to the measurement without helping to smear out
the lower bits.

If the ADC is part of a servo system, there normally is a natural low
passing action as part of the design. In these cases, you may not even
have to add a filter to the design. The zero you add in analog land can
help to make the phase margin of the servo loop as well as removing the
cogging effects of the LSB.

The spectroscopy boys have another cute trick: use a dac, and add a
random offset to the analog signal at each sample, and then subtract
it out digitally. This results in sub-bit resolution and essentially
perfect differential linearity, even from a fast sar adc that's not
inherently very good.

John

 

The spectroscopy boys might have got this from the audio boys.

John Watkinson's "The Art of Digital Audio" (ISBN 0-240-51270-7), which
was published in 1988, credits this to

Blesser B. "Advanced A-D conversion and filtering: data conversion" in
Digital Audo, edited by BA. Blesser, B. Locanthi and T.G. Stockman
Jnr., pages 37-5, published in New York by the Audio Engineering
Society, in 1983. Stockman has published some good stuff in the area.

-------------
Bill Sloman, Nijmegen

 

[Ken Smith]

In article <btv9e1lub2d3dkanr9ma55pdpg36kr7l24@4ax.com>,
John Larkin <jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:
[...]

The spectroscopy boys have another cute trick: use a dac, and add a
random offset to the analog signal at each sample, and then subtract
it out digitally. This results in sub-bit resolution and essentially
perfect differential linearity, even from a fast sar adc that's not
inherently very good.

Yes but there is a down side to this. The spectral resolution is degraded
as the linearity is improved. The analog and digital values do not match
to the same degree as there is a nonlinearity in the converter.


--
--
kensmith@rahul.net forging knowledge

 

[Fred Bloggs]

bill.sloman@ieee.org wrote:

This is "dithering" and you need to be a bit picky about the probablity
distribution of the dithering signal.

I listed a bunch of useful references in a note in Rev. Sci. Instrum.
in 1999.

Sloman A.W. "Comment on 'Noise averaging and measurement resolution"
Review of Scientific Instruments, 70 4734 (1999)

An asynchronous triangular wave would produce a uniform distribution of
off-sets, and this isn't good - you ideally want a a triangular
probability distribution, and random noise comes a lot closer to this.

----------
Bill Sloman, Nijmegen

Seems like the uniform distribution would in effect be a real time Monte
Carlo experiment allowing you to interpolate the sample estimate using
Mahalanobian weighting. Maybe you are averaging a different statistic.
Maybe flakey-pooh can add a single comparator and RC for a delta-sigma
modulator with 10b quantizer which ought to be good for 14b effective
resolution.

 

[John Larkin]

On Tue, 26 Jul 2005 11:19:53 GMT, Fred Bloggs <nospam@nospam.com>
wrote:

bill.sloman@ieee.org wrote:
This is "dithering" and you need to be a bit picky about the probablity
distribution of the dithering signal.

I listed a bunch of useful references in a note in Rev. Sci. Instrum.
in 1999.

Sloman A.W. "Comment on 'Noise averaging and measurement resolution"
Review of Scientific Instruments, 70 4734 (1999)

An asynchronous triangular wave would produce a uniform distribution of
off-sets, and this isn't good - you ideally want a a triangular
probability distribution, and random noise comes a lot closer to this.

----------
Bill Sloman, Nijmegen


Seems like the uniform distribution would in effect be a real time Monte
Carlo experiment allowing you to interpolate the sample estimate using
Mahalanobian weighting. Maybe you are averaging a different statistic.
Maybe flakey-pooh can add a single comparator and RC for a delta-sigma
modulator with 10b quantizer which ought to be good for 14b effective
resolution.

Never tried that; wouldn't the 10-bit quantizer have to be accurate to
1/16 LSB in order to stretch it to 14 bits? The advantage of gross
dithering is that is averages over many ADC codes, the good, the bad,
and the ugly.

I have dithered a 7-bit ADC to well over 16 bits accuracy in an
electrical metering system, digitizing AC waveforms.

You're not flakey when everything that you do works.

John

 

Fred Bloggs wrote:

bill.sloman@ieee.org wrote:
This is "dithering" and you need to be a bit picky about the probablity
distribution of the dithering signal.

I listed a bunch of useful references in a note in Rev. Sci. Instrum.
in 1999.

Sloman A.W. "Comment on 'Noise averaging and measurement resolution"
Review of Scientific Instruments, 70 4734 (1999)

An asynchronous triangular wave would produce a uniform distribution of
off-sets, and this isn't good - you ideally want a a triangular
probability distribution, and random noise comes a lot closer to this.

----------
Bill Sloman, Nijmegen


Seems like the uniform distribution would in effect be a real time Monte
Carlo experiment allowing you to interpolate the sample estimate using
Mahalanobian weighting.

"Seems like" is what fooled the first people who tried it. Read
"Dithered Quantizers" by Robert M.Gray and Thomas G.Stockham Jnr. in
the IEEE Transactions on Information Theory vol. 39 pages 805-812 May
1993, in which Stockham recapitulates the experience of listening to
the digital recording of the album "Tusk" by Fleetwoood Mac, back in
1980 where the fallacy was exposed .... Nice paper.

---------
Bill Sloman, Nijmegen

 

[Fred Bloggs]

John Larkin wrote:

On Mon, 25 Jul 2005 14:03:33 +0000 (UTC), kensmith@green.rahul.net
(Ken Smith) wrote:


In article <ud18e1pcppul4smmdar7ovakpdo1r88u0u@4ax.com>,
John Larkin <jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:
[...]

Even more fun is to sum with noise or an asynchronous triangle, smear
the samples several LSBs, and software lowpass filter. If the native
noise is enough (dicey at 10 bits, almost guaranteed at 14-16) just
filter.


If you add a high frequency boost to the anti-alias filter, you can
compensate for the digital side's low pass. This high frequency boost is
often enough to boost the noise above one LSB.

A high frequency boost also tends to give you the best sort of noise for
the job. You really want all of the noise's energy to be above the cut
off frequency of the low pass filter you are using. Any noise below the
cut off just adds noise to the measurement without helping to smear out
the lower bits.

If the ADC is part of a servo system, there normally is a natural low
passing action as part of the design. In these cases, you may not even
have to add a filter to the design. The zero you add in analog land can
help to make the phase margin of the servo loop as well as removing the
cogging effects of the LSB.






The spectroscopy boys have another cute trick: use a dac, and add a
random offset to the analog signal at each sample, and then subtract
it out digitally. This results in sub-bit resolution and essentially
perfect differential linearity, even from a fast sar adc that's not
inherently very good.

John

That sounds like sliding scale A/D conversion which uses DAC summation
to step the input sample through each and every amplitude bin for total
elimination of the differential non-linearity error, resolution remains
the same. The goal is to avoid going to a higher resolution converter in
order to achieve a given THD specification, which would be the only
other option for non-linear distortion reduction.

 

[Fred Bloggs]

Ken Smith wrote:

In article <42ECE464.7080508@nospam.com>,
Fred Bloggs <nospam@nospam.com> wrote:
[...]

That sounds like sliding scale A/D conversion which uses DAC summation
to step the input sample through each and every amplitude bin for total
elimination of the differential non-linearity error, resolution remains
the same. The goal is to avoid going to a higher resolution converter in
order to achieve a given THD specification, which would be the only
other option for non-linear distortion reduction.


There is another method that reduces THD:

Let say you have a fast ADC that has to scan a bunch of inputs and convert
each channel. By adding dithering, you can remove the non-linearity
caused by the LSB but not the differential non-linearity of the ADC.

You can add an extra channel to the scan and use it to bring in a very
good quality ramp with the same dithering. Based on this extra input, you
create a table of actual values.

When you ADC gives you, lets say, 2F7H, you look in this table for the
2F7Hth value and pass that number into the further prociessing. This can
reduce the nonlinearity to being that of the ramp.

The best ramp is going to be one produced by a DAC because you have
tight tolerances on precision of voltage versus time, so it sounds like
you're going to naturally evolve this into a sliding scale conversion
anyway.

 

[Ken Smith]

In article <42ECE464.7080508@nospam.com>,
Fred Bloggs <nospam@nospam.com> wrote:
[...]

That sounds like sliding scale A/D conversion which uses DAC summation
to step the input sample through each and every amplitude bin for total
elimination of the differential non-linearity error, resolution remains
the same. The goal is to avoid going to a higher resolution converter in
order to achieve a given THD specification, which would be the only
other option for non-linear distortion reduction.

There is another method that reduces THD:

Let say you have a fast ADC that has to scan a bunch of inputs and convert
each channel. By adding dithering, you can remove the non-linearity
caused by the LSB but not the differential non-linearity of the ADC.

You can add an extra channel to the scan and use it to bring in a very
good quality ramp with the same dithering. Based on this extra input, you
create a table of actual values.

When you ADC gives you, lets say, 2F7H, you look in this table for the
2F7Hth value and pass that number into the further prociessing. This can
reduce the nonlinearity to being that of the ramp.

--
--
kensmith@rahul.net forging knowledge

 

[John Larkin]

On Sun, 31 Jul 2005 16:58:40 +0000 (UTC), kensmith@green.rahul.net
(Ken Smith) wrote:

In article <42ECE464.7080508@nospam.com>,
Fred Bloggs <nospam@nospam.com> wrote:
[...]
That sounds like sliding scale A/D conversion which uses DAC summation
to step the input sample through each and every amplitude bin for total
elimination of the differential non-linearity error, resolution remains
the same. The goal is to avoid going to a higher resolution converter in
order to achieve a given THD specification, which would be the only
other option for non-linear distortion reduction.

There is another method that reduces THD:

Let say you have a fast ADC that has to scan a bunch of inputs and convert
each channel. By adding dithering, you can remove the non-linearity
caused by the LSB but not the differential non-linearity of the ADC.

You can improve DNL if you dither a lot, several LSBs at least, and
average samples. Without the averaging, you obviously can't fix gross
errors like missing ADC codes. The dither-average thing is often
convenient when you have an ADC that can sample a lot faster than you
really need to acquire the signal.

The other place where dithering is good is in electric meters, where
the dynamic range requirement for current measurement is extreme.

John

 

[Ken Smith]

In article <jj3qe1ponbtpv8ns18reopa2eq74lajacv@4ax.com>,
John Larkin <jjlarkin@highNOTlandTHIStechnologyPART.com> wrote:
[...]

You can improve DNL if you dither a lot, several LSBs at least, and
average samples. Without the averaging, you obviously can't fix gross
errors like missing ADC codes. The dither-average thing is often
convenient when you have an ADC that can sample a lot faster than you
really need to acquire the signal.

There is a limit on the dithering placed by the allowed noise. When you
make the dithering many LSBs tall, it and the differential non-linearity
of the ADC ends up raising the noise floor.

The other place where dithering is good is in electric meters, where
the dynamic range requirement for current measurement is extreme.

Yes and many meter like situations. Any place where the bandwidth of the
final output is low and lots of bits are wanted are good cases for
dithering.

Those folks that do the super low frequency seismic monitoring most likely
do dithering. Chances are they don't actually inject the dither. They
just allow the higher frequency noise from the environment do it for them.


--
--
kensmith@rahul.net forging knowledge