How to sample multiple signals at the same time with ADC?

[Mike Noone]

Hi - I'm working on a project where I'm going to have 12 different
channels of data that I will be sampling at about 2KHz. I need I think
about 16 bits of accuracy, maybe a little less. I would really like to
take all these samples at *exactly* the same time. So - I could have 12
ADCs on the board... But that'd be just a tad spendy, to say the least.
What I don't care too much about is how long it takes to digitize each
signal, as long as all are digitized in under half a milisecond (1/2K
seconds). So ideally I'd like a chip that could sample and hold all the
signals at the same time, and then go back and individually digitize
them. Are there any ADCs that can do this for me? If not - are there
any good solutions to this problem? I should mention that I might end
up needing only 4 channels, not 12. (I'll have 3 boards with 4 channels
each - but I'm not sure if I want an ADC on each board or just one ADC
for all 3).

Other details if they matter - the devices being samples will be an
Analog Devices accelerometer and an Analog Devices gyro as well as
probably a Honeywell magnetometer. Fourth channel is the temperature
sensor on the gyro.

Thanks!

-Mike

 

[PeteS]

Hi - I'm working on a project where I'm going to have 12 different
channels of data that I will be sampling at about 2KHz. I need I think
about 16 bits of accuracy, maybe a little less. I would really like to
take all these samples at *exactly* the same time. So - I could have 12
ADCs on the board... But that'd be just a tad spendy, to say the least.
What I don't care too much about is how long it takes to digitize each
signal, as long as all are digitized in under half a milisecond (1/2K
seconds). So ideally I'd like a chip that could sample and hold all the
signals at the same time, and then go back and individually digitize
them. Are there any ADCs that can do this for me? If not - are there
any good solutions to this problem? I should mention that I might end
up needing only 4 channels, not 12. (I'll have 3 boards with 4 channels
each - but I'm not sure if I want an ADC on each board or just one ADC
for all 3).

Other details if they matter - the devices being samples will be an
Analog Devices accelerometer and an Analog Devices gyro as well as
probably a Honeywell magnetometer. Fourth channel is the temperature
sensor on the gyro.

Thanks!

-Mike

You could use a multiplexed A-D and front end sample hold ICs - it's
possible to make your own, of course, but unless you are trying to
squeeze the last cent from the design or some other special reason, I'd
go with a premade one.

16 bits of resolution implies 0.001% accuracy for all errors. Even the
best ICs I could find (Analog devices) spec 14 bit accuracy (and are
expensive - $9 - $34 each depending on which one you choose).
http://www.analog.com/en/subCat/0,2879,773%5F925%5F0%5F%5F0%5F,00.html

Much cheaper is the LF398 (298/198) [A] range from National
http://www.national.com/pf/LF/LF398.html

Has 0.002% gain accuracy, [15 bit accuracy without including other
errors], so droop will be the major limiting factor.

You could probably get 12 bits [enob] out of this without much trouble.

Cheers

PeteS

 

[Mike Noone]

You could use a multiplexed A-D and front end sample hold ICs - it's
possible to make your own, of course, but unless you are trying to
squeeze the last cent from the design or some other special reason, I'd
go with a premade one.

16 bits of resolution implies 0.001% accuracy for all errors. Even the
best ICs I could find (Analog devices) spec 14 bit accuracy (and are
expensive - $9 - $34 each depending on which one you choose).
http://www.analog.com/en/subCat/0,2879,773%5F925%5F0%5F%5F0%5F,00.html

Much cheaper is the LF398 (298/198) [A] range from National
http://www.national.com/pf/LF/LF398.html

Has 0.002% gain accuracy, [15 bit accuracy without including other
errors], so droop will be the major limiting factor.

You could probably get 12 bits [enob] out of this without much trouble.

Cheers

PeteS

Hi PeteS - thanks for the response. I didn't realize there were chips
designed just for this purpose. The National part looks like exactly
what I want. The droop shouldn't be too serious of a problem as long as
I have a fast ADC running. (currently looking at the TI ADS1258). Do
you think that my ADC readings will be inaccurate if I have the ADC on
a board separate from the sensors? Specifically - I need the sets of
sensors at right angles to each other - so there will be three boards
all orthogonal to each other - and I'm thinking just one of those
boards would have the ADC on it. They would then be connected to each
other with right angle pin headers or something like that. The trace
length from the ADC to the sensor should be under 4cm.

Thanks!

-Mike

 

[joseph2k]

You could use a multiplexed A-D and front end sample hold ICs - it's
possible to make your own, of course, but unless you are trying to
squeeze the last cent from the design or some other special reason, I'd
go with a premade one.

16 bits of resolution implies 0.001% accuracy for all errors. Even the
best ICs I could find (Analog devices) spec 14 bit accuracy (and are
expensive - $9 - $34 each depending on which one you choose).
http://www.analog.com/en/subCat/0,2879,773%5F925%5F0%5F%5F0%5F,00.html

Much cheaper is the LF398 (298/198) [A] range from National
http://www.national.com/pf/LF/LF398.html

Has 0.002% gain accuracy, [15 bit accuracy without including other
errors], so droop will be the major limiting factor.

You could probably get 12 bits [enob] out of this without much trouble.

Cheers

PeteS

Hi PeteS - thanks for the response. I didn't realize there were chips
designed just for this purpose. The National part looks like exactly
what I want. The droop shouldn't be too serious of a problem as long as
I have a fast ADC running. (currently looking at the TI ADS1258). Do
you think that my ADC readings will be inaccurate if I have the ADC on
a board separate from the sensors? Specifically - I need the sets of
sensors at right angles to each other - so there will be three boards
all orthogonal to each other - and I'm thinking just one of those
boards would have the ADC on it. They would then be connected to each
other with right angle pin headers or something like that. The trace
length from the ADC to the sensor should be under 4cm.

Thanks!

-Mike

While it is doable, maintaining 12 or more bit accuracy across a connector
is increasingly problematic as bits increase. I would put ADCs on each
board. Designers choice though.

 

[Mike Noone]

While it is doable, maintaining 12 or more bit accuracy across a connector
is increasingly problematic as bits increase. I would put ADCs on each
board. Designers choice though.

Hmm in that case... the Analog SF7656:
http://www.analog.com/en/prod/0,,AD7656,00.html doesn't look too bad at
all. Expensive part ($17 each for 1K parts) but the added simplicity of
the part might make a whole lot of sense. I could put one on each board
and be done with it - no muxes, no worrying about signal degradation
between parts and boards, no sample and hold chips. Just a shared SPI
interface between the three boards.

-Mike

 

[PeteS]

Hmm in that case... the Analog SF7656:
http://www.analog.com/en/prod/0,,AD7656,00.html doesn't look too bad at
all. Expensive part ($17 each for 1K parts) but the added simplicity of
the part might make a whole lot of sense. I could put one on each board
and be done with it - no muxes, no worrying about signal degradation
between parts and boards, no sample and hold chips. Just a shared SPI
interface between the three boards.

-Mike

I agree with Joseph, and the choices look nice. My general rule is it's
best to keep the signal in the digital domain as much as possible. It's
much better to run digital signals across connectors than analog ones,
and will make the connector much cheaper than you would otherwise need.

Your droop issue will depend on the timing of the hold signal to the
track/hold to the time the last sample is fully converted (or latched
if the A-D has an internal hold cap). That will be dominated by the
leakage of the hold amplifer, the hold cap (use a nice one, *not* a
ceramic - there was a thread on such things recently) and the internal
droop, leakage and input resistance of the A-D.

Cheers

PeteS

 

[vasile]

Hi all,

The problem here is not the sample&hold but the time required for
reading the 12 holded values with a reasonable accuracy (as fastest as
possible after sampling is finished) using a multiplexed ADC.

regards,
Vasile

You could use a multiplexed A-D and front end sample hold ICs - it's
possible to make your own, of course, but unless you are trying to
squeeze the last cent from the design or some other special reason, I'd
go with a premade one.

16 bits of resolution implies 0.001% accuracy for all errors. Even the
best ICs I could find (Analog devices) spec 14 bit accuracy (and are
expensive - $9 - $34 each depending on which one you choose).
http://www.analog.com/en/subCat/0,2879,773%5F925%5F0%5F%5F0%5F,00.html

Much cheaper is the LF398 (298/198) [A] range from National
http://www.national.com/pf/LF/LF398.html

Has 0.002% gain accuracy, [15 bit accuracy without including other
errors], so droop will be the major limiting factor.

You could probably get 12 bits [enob] out of this without much trouble.

Cheers

PeteS

Hi PeteS - thanks for the response. I didn't realize there were chips
designed just for this purpose. The National part looks like exactly
what I want. The droop shouldn't be too serious of a problem as long as
I have a fast ADC running. (currently looking at the TI ADS1258). Do
you think that my ADC readings will be inaccurate if I have the ADC on
a board separate from the sensors? Specifically - I need the sets of
sensors at right angles to each other - so there will be three boards
all orthogonal to each other - and I'm thinking just one of those
boards would have the ADC on it. They would then be connected to each
other with right angle pin headers or something like that. The trace
length from the ADC to the sensor should be under 4cm.

Thanks!

-Mike

 

[Adrian Jansen]

Hmm in that case... the Analog SF7656:
http://www.analog.com/en/prod/0,,AD7656,00.html doesn't look too bad at
all. Expensive part ($17 each for 1K parts) but the added simplicity of
the part might make a whole lot of sense. I could put one on each board
and be done with it - no muxes, no worrying about signal degradation
between parts and boards, no sample and hold chips. Just a shared SPI
interface between the three boards.

-Mike

You say 'at the same time' and then 'within 0.5 msec'. You can do a lot
of processing in 0.5 msec. For instance the AD7651 will do 16 bit
conversions on single channel at 100 KSPS rate, so you could multiplex
12 inputs and still get around 8 KSPS on all channels. Thats far faster
than your 0.5 msec cycle time, and a much cheaper part.

--
Regards,

Adrian Jansen adrianjansen at internode dot on dot net
Design Engineer J & K Micro Systems
Microcomputer solutions for industrial control
Note reply address is invalid, convert address above to machine form.

 

[joseph2k]

I agree with Joseph, and the choices look nice. My general rule is it's
best to keep the signal in the digital domain as much as possible. It's
much better to run digital signals across connectors than analog ones,
and will make the connector much cheaper than you would otherwise need.

Your droop issue will depend on the timing of the hold signal to the
track/hold to the time the last sample is fully converted (or latched
if the A-D has an internal hold cap). That will be dominated by the
leakage of the hold amplifer, the hold cap (use a nice one, *not* a
ceramic - there was a thread on such things recently) and the internal
droop, leakage and input resistance of the A-D.

Cheers

PeteS

Yes, watch the hold capacitors closely, i prefer polystyrene, polypropylene,
or mylar(PET). And not "Stacked".