04-27-2010 10:29 AM
In our application we are using several PCI-4472 cards to sample 32 channels at 1KS/s and then decimate the data.
At a sampling rate of 1KS/s a 4472 card oversamples by 128 times, effectively sampling data at 128 KS/s. The problem with this approach is that the on-board analog anti-aliasing filters do practically nothing, as their cutoff frequency is set to about 400 kHz.
Due to various system limitations it is not feasible to sample the data at higher rates. And our cards are too old in revision to support the on-board oversampling feature that the later revisions offer.
Given that we will never have to sample data above 1 KS/s I was wondering whether it would be possible to change the components on a card to set the analog filter cutoff frequency to a lower value?
Regards,
Sebulba
04-27-2010 10:46 AM
The anti-aliasing cutoff frequency is actually adjusted according to your sampling rate via combined analog and digital filtering. (page 3, http://www.ni.com/pdf/manuals/373861d.pdf).
04-27-2010 11:27 AM
Hi Ian,
Thanls for your reply. The specs that you provided are for 4472 cards of revision F and higher; these cards support enhanced low-frequency anti-aliasing. (Page 28 of http://www.ni.com/pdf/manuals/371235g.pdf) Most of the cards that I have are of an earlier revision and do not have this feature.
As for the analog anti-aliasing filter, the user’s manual for my cards states “The NI 4472 includes a two pole anti-alias lowpass filter for each input channel. This filter has a cutoff frequency of about 400 kHz.”
04-27-2010 11:49 AM
I think digital filtering is always used with analog filtering for NI DSA devices. The analog filter gets rid of frequencies above 400kHz, and then digital filter do the rest.
Maybe NI engineers can correct me or give a clearer answer.
04-27-2010 11:57 AM
That is my understanding as well. The issue is that when I specify a sampling rate of 1 KS/s the cards actually samples at 128 KS/s – which is within the pass band of the analog anti-aliasing filter (hence the problem) – then the digital filter decimates the data to produce data sampled at 1 KS/s.
04-27-2010 12:58 PM
Yes, with any of NI 24 bit delta sigma analog to digital (ADC) modules you have two low pass (anti-aliasing filters) in play. The first is the hardware filter that is around the lower oversampling rates. When you run the 4472 at the lower rates, it is actually performing 256 times oversampling. The later manuals have more details on how this works.
The second filter in line is the digital filter that is part of the 24 bit ADC. This digital filter tracks the sample rate and provides a very tight response at about 0.45 of the sampling rate. In this case at about 450 Hz.
You can verify the performance with a function generator and a sine wave of other shaped tone.
The purle line in the graph below is the location of the 4472 filter when sampling at 102.4kHz (about 48kHz). When sampling at 1024 Hz, the filter will be about 480Hz. The filters for DSA devices ar very flat in th passband with very sharp rolloff in the cut-off range.
04-27-2010 02:22 PM
The above specs appear to be for cards that support low-frequency anti-aliasing. The specs for my card are
It shows that the card oversamples 128 times for a sampling rate of 1 KS/s.
04-27-2010 02:49 PM
sas you know, you can have the 4472 sample at 1kz. Its anti-aliasing digital filter will track and set at about 480Hz.
Can you try a signal generator to verify for yourself. It works well for me.
04-28-2010 12:45 PM
Instead of a single sine wave I tested the board with a chirp (8Vpp). The result is shown below.
My concern is the almost full-scale signal that is allowed through around 256kHz. I would like to know whether the analog anti-aliasing filter may be re-programmed (i.e., have components changed) so as to shift the band around 256 kHz past the filter’s cut-off frequency.