Dealing with DC offset in measurements

[Reet]

DC offset in measurements can be very frustrating, I thought I'd open this topic to determine best practices and ask what the recommended solutions are.

Equipment here is a Motu M4 interface, and sine sweep measurement in ARTA. Motu claims "DC coupled TRS input" so it tends to include more low frequency noise than other AC coupled equipment.

For the following measurements, I measured with amp, mic, and loopback jig connected to amp output, however no speaker connected to measurement is just background noise. Dual channel sine sweep shows increasing noise towards low frequencies, and DC offset is obvious in the impulse view:


Increasing time scale on the impulse to 2.5s to show the entire measurement, we can see that it is not strictly a DC problem, but a very low frequency wandering noise:


Single channel measurement of just the mic input, shows no low frequency / DC problem, which is expected:


Single channel measurement of the reference channel, shows the low frequency problem clearly:


Zooming out on time scale of the reference input, shows the low frequency information in the reference channel input:


VituixCAD only shows response down to 5Hz, but using ARTA we can show down to 0.5Hz, to see that there is very low frequency content in the measured signal, even though it starts to roll off at 20Hz:



Given this low frequency problem with dual channel measurements, the options I have in mind would be:

• Use "DC offset" checkbox in VituixCAD IR to FR processing, when using sine sweep meausurements
• Use PN noise measurement instead of sine sweep, to inherently high pass the measured signal.
• Use multiple measurement averaging to lower noise floor
• Place a capacitor on the reference channel input to AC couple it.(?)

Any other ideas, opinions, recommendations are welcome.

[Nil L]

The REW author suggests using the 'High Pass Filter' checkbox before measuring in the 'Sound Card' window.

[Reet]

To be honest I don't have this problem at all with REW, whether high pass is used or not. It may be something about REW's inherent window limiting when using "loopback as cal and timing reference". Additionally, the old REW method of using the loopback channel as timing only, resolves any DC offset since there is not channel normalization. In any case, the high pass feature has very little effect on overall response anyway, providing only 3dB attenuation at 2Hz.

[Reet]

DC offset is may not be the most accurate description of the problem, DC is only part of the picture, really I am looking to address the high noise floor at low frequencies with a true dual channel measurement. In the measurements above, noise floor at 20Hz is a good 10dB higher in the dual channel measurement vs single channel.

[Nil L]

DC offset is may not be the most accurate description of the problem, DC is only part of the picture

This is true. For a solution, I think you need to follow the tail of the step response. That is, seeing how far the tail differs from zero, apply an adjustment at the beginning of the impulse response. The 2 Hz high pass filter should affect exactly the end of the step response. But I can't check it now.

[Reet]

I’m still trying to comprehend what the cause is for high noise low frequency noise in dual channel measurement with ARTA especially. Individual measurement of each channel does not show any cause for concern. First thought would be that the noise at each input is not common mode, but even then I would expect a noise increase in the order of 3dB, not 10dB+ and not frequency dependent.

[Reet]

Here's a simple comparison of just background noise floor, measured in ARTA using sine sweep as the test signal, vs PN noise as the test signal, with low frequency cutoff of 20Hz.

Clearly, the low frequency noise floor is an issue inherent to the sine sweep test signal.

[Nil L]

Clearly, the low frequency noise floor is an issue inherent to the sine sweep test signal.

There are noises that the sound card picks up, from the fact that there are switching power supplies around, mains interference, or something else.

[Reet]

There are noises that the sound card picks up, from the fact that there are switching power supplies around, mains interference, or something else.

For above test, reference channel level was set about -10dBFS, SNR on reference input is very high, clean signal. Mic input measuring background noise only, in this case it is mostly my PC fans, however this is only true when using PN noise as the test signal, yellow trace in my previous post. I can easily show that the Motu M4 has a very low noise floor, with no influence from mains interference or other nearby electrical noise

Perhaps this is more of a question for the ARTA developer, as to why there is such a difference in noise floor between the different signal methods.

[Nil L]

Perhaps this is more of a question for the ARTA developer, as to why there is such a difference in noise floor between the different signal methods.

I once asked myself similar questions when I started working with ARTA. But in parallel, I did something in REW. Then the desire to do something in ARTA completely disappeared. Because REW gives all the possibilities better visually, more precisely, more conveniently. Plus constant development.

[Reet]

Only recently has REW offered a true dual channel measurement capability, and even then is still requiring some bug fixes. Maybe next version will be perfect in this regard. Personal preference, but I'm not a huge fan of REW's UI. Too much flipping back and forth between windows and tabs to get the information I need. I also don't agree with the way phase is represented in REW, it's quite annoying to have to shift the measurement to t=0 to remove time of flight from the measurement from the phase data. But, I guess every software has it's pros and cons, and John Mulcahy has been open to new ideas and changes, as long as the implementation isn't overly complex. It sounds like he has a lot of features on the long term to-do list.

For now, I will just make a point to use PN noise signal with ARTA for maximum SNR,

[Reet]

The ARTA developer unfortunately has not provided much constructive feedback on this matter. Until proven otherwise, I've decided that ARTA has a bug with what it's doing for response normalization in a dual channel sine sweep measurement.

Last comparison, remove the mic from the equation. Here is a loopback measurement, through my amp and ARTA jig, but no mic connected, measurement input is open circuit.
Cyan - Dual channel sine sweep, open input, reference -10dBFS
Yellow - Single channel sine sweep, open input for noise floor of equipment only
Green - Dual channel periodic noise, open input, reference -10dBFS
Yes, that’s a 30dB+ difference at 20Hz

[Reet]

The worst part of the dual channel sine sweep is that the noise floor result is not consistent, you can run the same measurement 10 times and get 10 different results, while any other measurement method is very consistent. I've verified my Motu is not at fault either, same behaviour can be observed with my PC onboard audio, and it's easy enough for anyone here to try for themselves.

[Nil L]

I have two questions for you.
1. When an inductor, capacitor or resistor is soldered onto the board, does it make a difference for you whether to solder the beginning or end of the passive element to the plus (input)?
2. Draw, please, how the measuring circuit of a full two-channel measurement looks like.

[Reet]

I have two questions for you.
1. When an inductor, capacitor or resistor is soldered onto the board, does it make a difference for you whether to solder the beginning or end of the passive element to the plus (input)?

Huh?

2. Draw, please, how the measuring circuit of a full two-channel measurement looks like.

https://www.htguide.com/forum/articles/do-it-yourself-diy/927384-dual-channel-measurement-jig

[Reet]

On this DC offset problem, I posted in the ARTA thread at DIYAudio, and Ivo (ARTA developer) was able to reproduce the problem with one of his audio interfaces, so it is being looked into. Solution for now is to avoid sine sweep for both high noise floor and poor repeatability, periodic noise works very well.

[Nil L]

The cirquit for a full two-channel measurement is the one that Kimmo got from REW help, which he provides in his REW measurement manual. Yes?

[Reet]

For the jig, amp output connects to jig input, jig reference output connects to audio interface input.

For analysis of audio interface only for this thread topic, "semi-dual" configuration is used, so reference loopback is just patch cable from interface output to input:


These images in VituixCAD measurement guides are all copied from the ARTA manual, btw.I use different resistor values than ARTA manual, but that's of no significance here.

https://artalabs.hr/download/ARTA-user-manual.pdf

[Nil L]

Thank you. I doubted. But it turned out I always used a fully dual-channel measurement.

[Reet]

Not that it matters much, but if REW has been your software of choice, previous versions with "loopback as timing reference" as the only option throws away most of the benefit of a full dual channel measurement, since the loopback is used for a timing reference only. In this case you might as well use simple patch cable "semi-dual" configuration in that scenario, which is what previous REW measurement guide had detailed.

[Nil L]

I took measurements with a loop, but from one output of the sound card there was a Y-cable to the amplifier and to the input of the card. Whereas many use two card outputs - one for the loop, the other for the amplifier. Several drivers belonging to one speaker were measured. Then I took measurements with acoustic synchronization of the same drivers. Both options loaded into VituixCad. I didn't notice the difference. Therefore, I still do not understand why loop synchronization is needed. Unless John changed something in the processing of measurement data in the latest versions.

[Reet]

"Semi-dual".

[Reet]

New ARTA release fixes this problem for the most part.

February, 2, 2023; ARTA software version 1.9.6 is published.

• Measurement with swept-sine signal in dual channel mode is improved by better control of sub-audio noise.
• Several small bugs are corrected
• This version no longer works in Windows XP

Note: To download ArtaSetup196.exe you will have to choose left download menu, then twice choose See more and Keep options.
Microsoft just need few days to approve download.

Ivo