Sampling question

Suppose we have an ADC with three inputs (multiplexed). The ADC takes
each input in turn. Suppose the frequency that the ADC scans  bewteen
inputs is fd. How should fd be related to the sampling rate per
channel  fs? 10 times bigger or 100 times bigger? and if the scan rate
is too slow what effect does this have. Obviously the inputs cannot be
matched in time because of the scan rate.

Hardy

HardySpicer wrote:
> Suppose we have an ADC with three inputs (multiplexed). The ADC takes
> each input in turn. Suppose the frequency that the ADC scans  bewteen
> inputs is fd. How should fd be related to the sampling rate per
> channel  fs? 10 times bigger or 100 times bigger? and if the scan rate
> is too slow what effect does this have. Obviously the inputs cannot be
> matched in time because of the scan rate.
> 
> Hardy
> 
That depends very much on the converter.

If its the kind of converter that produces clean independent samples 
from the input signals, and the mux switches fast enough, fd can be 
3*fs. If it something like a sigma delta converter, a change of input 
will require throwing away some samples before a clean representation of 
the new channel is available from the converter. The number of wasted 
samples used to be quite large, but many sigma delta converters are 
quite constrained these days. You can often use every other sample. You 
really have to look carefully at the converter's spec to see.

In between those extremes (I think a sigma delta converter would be the 
extreme) you have the practical messy world. Each conversion of an SAR 
converter should be completely independent of the previous one. For many 
converters, limitations in the sampler mean that is not entirely the 
case, so the channels may pollute each other. You might end up doing 
nasty things, like throwing away every other sample as the mux switches, 
to clean up that kind of pollution.

Regards,
Steve

I would guess that this depends on the type of ADC.
Does it have a build in antialias filter, and does it have a sample-hold?

Best regards,

Andre

HardySpicer wrote:
> Suppose we have an ADC with three inputs (multiplexed). The ADC takes
> each input in turn. Suppose the frequency that the ADC scans  bewteen
> inputs is fd. How should fd be related to the sampling rate per
> channel  fs? 10 times bigger or 100 times bigger? and if the scan rate
> is too slow what effect does this have. Obviously the inputs cannot be
> matched in time because of the scan rate.
> 
> Hardy
> 

On Sep 11, 5:38 am, HardySpicer <gyansor...@gmail.com> wrote:
> Suppose we have an ADC with three inputs (multiplexed). The ADC takes
> each input in turn. Suppose the frequency that the ADC scans  bewteen
> inputs is fd. How should fd be related to the sampling rate per
> channel  fs? 10 times bigger or 100 times bigger? and if the scan rate
> is too slow what effect does this have. Obviously the inputs cannot be
> matched in time because of the scan rate.
>
> Hardy

The sampling rate per channel is 1/3 fd. Or is there something hidden
here?

HardySpicer wrote:
> Suppose we have an ADC with three inputs (multiplexed). The ADC takes
> each input in turn. Suppose the frequency that the ADC scans  bewteen
> inputs is fd. How should fd be related to the sampling rate per
> channel  fs? 10 times bigger or 100 times bigger? and if the scan rate
> is too slow what effect does this have. Obviously the inputs cannot be
> matched in time because of the scan rate.
> 
> Hardy
> 

If it's a SAR ADC then fd = 3*fs.

If it's a Sigma-Delta ADC then fd = settling_time*3*fs.

So for example, you might see a sigma delta converter where the first 3 
samples are invalid.  In that case, every time you change the mux you 
need to wait for 4 samples.  That means fd needs to be 4*3*fs, i.e. 12*fs.

Brad

HardySpicer <gyansorova@gmail.com> wrote in news:1189503517.102749.244440
@g4g2000hsf.googlegroups.com:

> Suppose we have an ADC with three inputs (multiplexed). The ADC takes
> each input in turn. Suppose the frequency that the ADC scans  bewteen
> inputs is fd. How should fd be related to the sampling rate per
> channel  fs? 10 times bigger or 100 times bigger? and if the scan rate
> is too slow what effect does this have. Obviously the inputs cannot be
> matched in time because of the scan rate.
> 
> Hardy
> 

Depends on what you need to do, and the drivers for your ADC.  Often, the 
inter-channel sampling rate can be as fast as the board or chip can go, so 
you burst through your three-channel sample quickly, and then wait for the 
next trigger.  Could go all the way down to fs*3, though, and that might be 
good enough for your application.


-- 
Scott
Reverse name to reply

Steve Underwood wrote:


> In between those extremes (I think a sigma delta converter would be the 
> extreme) you have the practical messy world. Each conversion of an SAR 
> converter should be completely independent of the previous one.

If this is the case, it is good to remember that the mutual influence 
drops by a factor of ~100 per every 5*T, where T is the characteristic 
timeconst of the ADC input circuit. Thus, for the 16 bit accuracy the 
switching rate should not be faster then the 3dB bandwidth of the input.

> For many 
> converters, limitations in the sampler mean that is not entirely the 
> case, so the channels may pollute each other.

There is also charge injection effect in the multiplexer, which is 
pretty difficult to account for.

> You might end up doing 
> nasty things, like throwing away every other sample as the mux switches, 
> to clean up that kind of pollution.

Shamanism :-)

BTW, you can account for the mutual coupling so you can can cancel it out.



Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

HardySpicer wrote:
> Suppose we have an ADC with three inputs (multiplexed). The ADC takes
> each input in turn. Suppose the frequency that the ADC scans  bewteen
> inputs is fd. How should fd be related to the sampling rate per
> channel  fs? 10 times bigger or 100 times bigger? and if the scan rate
> is too slow what effect does this have. Obviously the inputs cannot be
> matched in time because of the scan rate.

There are different classes of ADC. The transient that arises from 
switching input streams to a sigma-delta converters is many sample 
times. In general, such converters are not suited for multiplexing or 
servo systems. Successive-approximation and flash converters can operate 
on a different stream for each sample provided the input bandwidth is 
great enough to accommodate the large slew that might be required. The 
actual channel rate depends on how quickly the multiplexer settles. 
Without more detail in your question,you won't get more detail in your 
answer.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Steve Underwood wrote:
> HardySpicer wrote:
>> Suppose we have an ADC with three inputs (multiplexed). The ADC takes
>> each input in turn. Suppose the frequency that the ADC scans  bewteen
>> inputs is fd. How should fd be related to the sampling rate per
>> channel  fs? 10 times bigger or 100 times bigger? and if the scan rate
>> is too slow what effect does this have. Obviously the inputs cannot be
>> matched in time because of the scan rate.
>>
>> Hardy
>>
> That depends very much on the converter.
> 
> If its the kind of converter that produces clean independent samples 
> from the input signals, and the mux switches fast enough, fd can be 
> 3*fs. If it something like a sigma delta converter, a change of input 
> will require throwing away some samples before a clean representation of 
> the new channel is available from the converter. The number of wasted 
> samples used to be quite large, but many sigma delta converters are 
> quite constrained these days. You can often use every other sample. You 
> really have to look carefully at the converter's spec to see.
> 
> In between those extremes (I think a sigma delta converter would be the 
> extreme) you have the practical messy world. Each conversion of an SAR 
> converter should be completely independent of the previous one. For many 
> converters, limitations in the sampler mean that is not entirely the 
> case, so the channels may pollute each other. You might end up doing 
> nasty things, like throwing away every other sample as the mux switches, 
> to clean up that kind of pollution.

Much of what is charged to converters is due to capacitors' dielectric 
memory. When those capacitors are on the converter chip, there's no 
difference.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

panabiker wrote:


> ... is there something hidden here?

Do you now understand some of the issues?

Jerry
-- 
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯