Low pass filter before NI 5112

[AFolly]

Hello,

I am currently using a NI 5112 to measure the signal from an infrared detector in a cavity ring-down experiment. Three examples of signals are shown below. My main question is how I can implement a low pass filter, preferably passive, before my NI 5112 without extremely distorting my signal due to impedance mismatching. Now some more details:

A couple of single captures for each wavelength are shown in color, while an average of 25 pulses is shown offset in black. The range and offset are chosen in each case to minimize the "digitization" noise. In the case of 3.14um, the noise you see is ~25 times the digitization noise. These were taken without the BW-limiting mode and at 100MS/s.

The detector (Vigo MIPDC-F-10) has a bandwidth of 10 MHz. I believe it is low impedance and is meant to be coupled with 50Ohms, however its documentation confuses me and I am waiting for a definitive response from the supplier. Part 2.4 of the manual says 50Ohms recommended, however the datasheet and our calibration sheet (below) seem to suggest 1 MOhm is recommended!

There are some strange oscillations with a period of around 180ns in our signal which I thought were due to the impedance mismatching that existed in the system before I changed it:

- Detector

- 1 meter 50Ohm SMC to BNC cable (RG-174)

- Inline BNC connector

- 5 meter 75Ohm BNC to BNC cable

- Digitizer, DC, no BW limit, 100MS/s, 1MOhm || 30pF

When I saw this setup I knew something wasn't right, and I even modeled it in LTSpice and it showed the same period of oscillations. However now the setup is:

- Detector

- 1 meter 50Ohm SMC to BNC cable (RG-174)

- Digitizer, DC, BW limit, 100MS/s, 50Ohm 

And we still have oscillations, although the period seems to have changed to around 320ns. These oscillations that are left are 99% likely due to our cavity ring-down experiment, however if anyone has any recommendations on possible causes or ways I can confirm it isn't due to my detection chain they would be more than welcome!!

Now the main question. Between the 1 meter 50Ohm cable and the digitizer I would like to insert a low pass filter. The BW limit helped reduce the noise, however it can definitely still be reduced further without any lose to our signal. This is because we cut off the beginning of the signal and then just measure the decay time, which is relatively long and smooth (1 to 2 us 1/e time). As well, in the future I would maybe even like to try and eliminate the 320ns oscillations, however I fear this much filtering will distort the signal too much. Therefore, for the immediate future I am really just looking to "replace" the BW filter of 20 MHz, with something like 1 or 5 MHz.

Obviously I would turn off the BW limit on the digitizer to avoid any extra confusions, however nonetheless, I am not quite sure how to approach the problem. Usually I would do lots of research and try different solutions hands-on. However, I don't have access to any electronic components at this job, so everything would have to be ordered, and I don't have much time to experiment. Ferrites seem like a possible solution, however not sure how effective they are at that low of frequency or how well they work with coaxial cables. I know the basic RC low pass filter, but the 50Ohms (or 1MOhm || 30 pF if I change it) seem to make it impossible. I guess an op-amp based one could work, however the large input impedance wouldn't match the impedance of the coaxial cable... etc...

Any recommendations of technique or must-red resources would be welcome. Thanks for your time.

[DFGray · Accepted solution]

A possible way to separate your electrical and cavity artifacts is relatively simple.  You are taking data at three wavelengths.  For each of those, do a simple exponential decay fit (e.g. Exponential Fit.vi) to your data, then subtract this from the signal.  You should have something that oscillates about a mean.  Compare the residual signals at all three wavelengths, either visually or with something like a power spectrum.  Anything in all three is probably electronics (and you could conceivably model and subtract it rather than trying to eliminate it).  This could break down if the rise times on the signals are different, since that will include different frequency elements.

I am not convinced you should filter your signal before taking the data.  As you mentioned, any filter will distort your final signal.  My preference would be to take the raw signal and apply any filtering in the analysis.  LabVIEW has a rich filter set, so you can experiment afterwards.  If you apply a filter before digitizing your data, you throw away data you will never get back.  However, if you know your data has no frequency components over your proposed filter cutoff, you should be good.  A power spectrum analysis on your current data should tell you this.  Be careful.  Your rise time shape may hold information you later want.  If you filter, you will probably slow it down.

Good luck!  Let us know if you need more information.

[AFolly]

Hello,

Thanks for your reply. I did what you recommended but haven't had a chance to analyze the data, however I think it will be a bit tricky since it seems like the sinusoidal decay envelope seems to decay differently than the exponential which will affect the exponential fit which is subsequently subtracted from the data... chicken and the egg 🙂 Either way though shouldn't effect the peaks much so I can at least see the general shape and compare.

However, I also did another experiment which is the smoking gun for me. I adjusted the cavity mirrors till we completely lost the CRDS signal. We were left with just a single peak and absolutely no ringing no matter if I changed cables, impedances, filters, etc...

Probably isn't prudent to filter however I think my professor would like the two options. I implemented a digital Butterworth filter in LabVIEW and I designed and I am going to build a hardware version as well using LC 5-pole Butterworth filter with a cutoff at 10MHz to match the detector.

Thanks for you help,!