Why does the (integrated) low pass filter on the analog inputs of the USB-6361/6363 DAQ boards have such slow roll off?

[Tamhleten]

 I am trying the squeeze out as much SNR from the USB-6363 DAQ board as possible for an optical measurement. Looking at the 6361/6363 Specifications, I guess that there is a PGA with a 3.4MHz RC filter (because that’s the analog trigger bandwidth (-3 dB) form the analog inputs and then another low pass filter before the ADC (this is a very neat idea allowing faster triggering, I guess). But according to the figure in the NI 6361/6363 specs sheet the 2 filters together yield a 1.7 MHz cutoff frequency with a quite slow roll-off. Given that the DAQ can sample at 2 MHz I don´t understand why is the roll off of the filter so slow, since the signal/noise at frequencies from 1.7 to 2 MHz is attenuated only by up to 4dB which should show up as aliases degrading the measurement around DC level. Is this correct or am I overlooking something? 

(If not, I would put a 3^rd order Butterworth filter (with the knee also around 1.7MHz) before the DAQ and bypass the filter for triggering via a separate APFI line)…

[JosD]

Hello Tamhleten,

To help you I would need a bit more inforation about the signal you want to measure. In which voltage range is your signal? What kind of sensor are you reading out? Are you interested in signal frequency? And if yes, in which range? And how about the source of your noise, does it fe. result from long wiring? 

Kind regards,

Jos Deurloo

[Tamhleten]

Hi, thank you for the reply.

I am measuring waveform shapes. Specifically exponential decays (about 100µs long). Once digitalized I fit by Levenberg–Marquardt algorithm the exponential decay time constants (I fit the amplitude, offset etc. of the waveform, but I am interested only in time constant measurement). The signal is coming from a photodiode in transimpedance connection, and then a voltage amplification stage (adjustable amplification and offset)  and then a THS4503 differential amplifier (also amplifying 10x) to get differential signal (so a 3 stage front end). The measured bandwidth of the circuit is about 2MHz (it is limited by the transimpedance stage) (in the 2 subsequent stages I put the RC constants to about 4MHz, so as not to accumulate attenuation on the higher frequency end). I can change the signal amplitude and offset (on the second stage) so presently I am measuring ~5V (p-p) signal shifted so it is centered on zero. (The amplitude is fluctuating +-20%). The source of the noise is the mainly the large transimpedance resistor and the photodiode. 

There is only about half a meter of (shielded differential) cable between the DAQ card and the photodiode (and circuit in the same enclosure).

I think the really important part of the bandwidth for me is 0 - 500MHz (the 100µs waveform corresponds to 10kHz) so if there would be an ideal low-pass filter I would set it to 500kHz, however with a first order filter with the -3dB knee at 1MHz I would already attenuate by 10% at 500kHz.

I higher order filter seems to me as the proper solution, I am just looking for the catch. The only thing why I am unsure about this is - why is the integrated filter on the DAQ board 1.7MHz and only first order? (Since a couple of capacitors and 2 inductors makes a nice 3^rd order filter). What would be the drawback of a higher order filter? (for any application even?)      

Kind regards,

Tomas

[JosD]

Hi Tomas,

I'd say that in your case experimenting with additional filters in front of the AI would be a good solution to reject high-frequency noise and prevent aliasing. I am not sure about the reason for applying only a first order filer in these DAQ devices. Could the linearity in phase response be a reason?

Kind regards,

Jos

[Tamhleten]

Thanks for the answer. I will indeed insert a higher order filter into the signal chain. 

Nonetheless, the linearity in phase response doesn't seem to change that radically, lets say, a 3rd order filter: 

Plot for a 1st order filter:

Plot for a 3rd order Butterworth filter:

The band pass is in both cases 1.7MHz.

There seems to be no relevant difference for signals below 1MHz (Group Delay is going in the opposite direction, but the change is of similar magnitude). The change in Phase about doubled...I guess will have to read a bit, whether this is relevant.

Kind regards,

Tomas