03-17-2023 05:17 AM
I am measuring with a cDAQ-9185 controller and different modules / module types, with different sampling rates simultaneously. The corresponding tasks are started synchronously (export / import of the start triggers).
I know that corresponding group delays of the filters have to be considered by the user. In the document
it is described that there are two possibilities in my case (C series): Check the data sheet or measure it yourself.
In some data sheets (e.g. for NI9232, https://www.ni.com/docs/en-US/bundle/ni-9232-specs/page/overview.html) it is described very precisely. In other data sheets (e.g. NI9222, https://www.ni.com/docs/en-US/bundle/ni-9222-specs/page/specifications.html) the information is missing completely.
My question is: Is there somewhere a comprehensive source, from which one can get all needed information about input delays? Or is it necessary to measure by yourself in some cases?
Thanks a lot!
Solved! Go to Solution.
03-17-2023 06:40 AM
As your first linked article says, the considerations for filter delay are specific to devices with a Delta-Sigma style ADC. Devices like the 9222 that use an SAR-style ADC have no such internal filters to spec. Any delay is generally just signal propagation through circuitry, most often relatively negligible by comparison.
It's probably mostly a matter of identifying the ADC type. I'm peripherally aware of the existence of cDAQ modules that don't declare themselves as either Delta-Sigma or SAR. They only allow vaguely-defined sample rate choices like "slow" or "fast". I don't know enough about those to offer specific suggestions how to handle them.
03-24-2023 08:20 AM - edited 03-24-2023 08:21 AM
I would always try to measure the timedelays. One sine source to all channels, frequency within Nyquist but higher is better (if you know your ADCs and signal theorie about folding, subsampling is valid 😉 ) and the phase of a tone detection on all channels should give a good estinate. Validate with a second lower frequency to catch delays longer than one periode. Or apply a continious sweep and use crosscorrelations.