Bandpass Filter "Ringing"

[Kallis]

Just curious, why is this 'ringing' a problem. The amplitude seems to be very low anyway (nA?). For almost all practical purposes (unless 4800 Hz sine wave that you apply is also in that range which seems unlikely), it seems you are getting 'zero'. Unlike the analog filter case where probably your sampling device is not able to resolve the analog filter's output to such low amplitudes, the digital filter will pass to the output, the in-band frequencies in the transient+noise no matter how small the amplitudes are provided everything fits within the double-precision arithmetic.

[CPWhite]

Thanks. We are operating in that range. Nominal signal is <2na. We measure the output with our 4800Hz signal off, and call that “background”. Then turn the 4800Hz on and measure again, subtracting the background signal from that to get closer to truth. Our issue is that we’re reading higher than the previous version of the tester, which used an analog filter and trying to find all the possible causes, however minor.

Thanks for your input. 

[Henrik_Volkers]

When you say you have analog 4th order filters and want to replace them with a digital filter AND need to get the same response:

Have you measured the analog filter response (or say the FRF of the old chain with and without filter)? 

(the FRF examples from the Sound and Vib module are a nice starting point, however a repeatable white noise test seems to be more appropriate ...)

It's a common problem when comparing an old test facility with a new one ... main customer wish:

I have this output signal, where I don't know (exactly) how it's  created, but I'm used to it (it works) and I want that same signal with the new one.

(Been Q-engineer and did a lot of gage-approvals ... , now (part-time) ISO 17025 auditor): Instead of hardly trying to replicate an old test-facility ... qualify the new one , establisch new  limits ..

but most important: understand and document the physics ..

[CPWhite]

Yes I have measured the frequency response of the analog and digital filters and they are the same within a few Hz and dB. We can chalk up the differences to analog component tolerances. The concern has been the ‘ringing’ and getting a meaningful RMS measurement if the ringing is too large. I was the test engineering manager and electrical engineer for this product line starting in 1992 until I tried to retire in 2013. Back as a consultant now. Unfortunately, as an FDA regulated medical device company, we can’t change test specs without a documented rationale. And changing them also requires validation which is costly.

There is more to the issue we’re trying to solve than just the filter response and it seems with the input provided by this community we understand the settings for the filter better and have reduced the no-signal ringing to acceptable levels and are limiting the signal-on measurement to valid portions of the sample array, allowing for startup time, etc. Thanks again. 

[CPWhite]

Just got back from traveling so first chance to look at this.

What is your “RMS Noise” signal? I’m seeing a ~166Hz (~6ms period) signal (solid color) superimposed over the 4.8Khz signal., which is hard to see due to the time scale 

I understand that there is a “startup delay” in the init mode and we truncate that. 

The issue remains that while our 4.8kHz band pass does a very good job of rejecting noise (random noise, various higher frequency clock and switching power supply ripple, etc ) from outside the passband, the output is not stable and therefore we can’t get an accurate measurement, especially with lower signal levels where the SN ratio is low.  Better than 0.1 but maybe 0.5. 

We ran some interesting tests today that confirms that the output of the filter is not an accurate representation of the 4.8kHz content of the input (it measures too high) and that it is not linear. What I mean by that is that increasing the signal level in the presence of constant noise does not yield a corresponding increase at the filter output. It’s like a multiple of 0.6, not 1. We’re going to do an FFT on the noisy input signal tomorrow, but visually on an O’scope, we can’t see any 4800Hz signal at small signal levels, but the output of the bandpass yields a value of 5na. On the existing fixture with the analog bandpass, we can easily see the (albeit tiny) 4800Hz signal, with high frequency (>150kHz) noise modulating it, and the system yields a value of 0.7na, which corresponds to the historical data. It’s like something in the VI which includes the lowpass filter and the bandpass filter as well as a scaling factor for volts to nA, is creating a 4800Hz signal out of “thin air”, which is why I was exploring ‘ringing’, though in this case the output is quite steady over time. We’re investigating more than the bandpass at this point and hope to find a stupid error in the VI but the SW engineer has looked quite closely at it. 🙂

[Bob_Schor]

It has been a long time since I studied this, but the act of working in the discrete digital domain, with quantization in both time (sampling) and "space" (digitization "noise") makes digital filtering have "unexpected" differences from what (continuous) theory would suggest.  I suspect this is the cause of the strange behavior that you are observing.

What the question about the origin of the RMS Noise signal addressed to me?  That's just a number used by the "Tones and Noise" signal generator to specify (wait for it ...) the RMS Noise being added to the sinusoidal signal.

[CPWhite]

So you don’t know the type of noise generated. I can look up the help file for that VI. 

I suspect that the issue may be in the acquisition section prior to our filtering module. The system has several layers to it: NI Test Stand, a proprietary layer that the developer uses on most of their projects, then custom code for our application. I guess you have to build each one independently, and I’m told that if you add a VI that is used in another ‘layer’, it sometimes will crash when building it. 

Who knows what lies in those other layers?

Thanks for your help.