ARTA

I am trying to determine the cause of high noise floor in ARTA for dual channel sine sweep measurements. I have tried to reach out to Ivo, but the response I received was not very constructive feedback, I think perhaps the issue was not made clear enough or lost in translation.

When measuring a dual channel sine sweep, noise floor increases at low frequency substantially, compared to a dual channel periodic noise measurement, or a single channel sine sweep measurement. For a simple test, I ran a loopback cable from output to input for reference channel, set reference level at -10dBFS, and left the measurement input open with nothing connected. 96kHz, maximum measurement length, maximum FFT length. I did try as well with measurement input shorted with no difference in result. Equipment used is a Motu M4.

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
1674665146799.png


I have verified that this issue is specific to ARTA, and not some inherent characteristic of a dual channel sine sweep. SoundEasy and REW were both tested with very good noise floor results using sine sweep, behaviour is not like what I see in ARTA. Other problem is this simple noise floor meaurement with sine sweep provides different results with each sweep, I can run the measurement 10 times and get quite varying levels on the noise floor, while other measurement methods provide very consistent results with measurement repetition.

(For REW, latest early access release with "use loopback as cal and timing reference" and "merge loopback response into IR" provides a true dual channel measurement)

Solution at this point is simply to avoid sine sweep measurement and use periodic noise instead, but it would be great to have a proper solution for this in ARTA.
 
Last edited:
  • Like
Reactions: 1 user
Hi Decibel,
sorry that I did not respond quickly in email.

Here are just few words on what happens in your test.

First let see how ARTA calculate dual ch and single ch. reponse.
In single ch. mode ARTA divide spectrum of input channel with spectrum of digitally generated swept sine sequence.
In dual ch. mode ARTA divide spectrum of input channel with spectrum of reference channel captured swept sine sequence.
Difference is in timing and linearity of reference channel amplifier and input capacitor.
Even a small change or nonlinearity of these parts, makes distortion that look like DC offset, and response raise toward lower frequencies.
Farina has found that maintenance of 14 bit accuracy is very important, but in this test we deal with signal that uses just few bits.
In real life measurement this will not be the problem.

For this kind of low level measuremets the use of pink noise is better solution as that signal has random character that distribute all distortion equally over all frequencies.
That is why I recommend random noise as best signal for low level measurement. It has several dB smaller dynamic range and generally we always have to use averaging.

Best,
Ivo

P.S. I have try this test on several audio systems and results are different.
 
Hi Ivo,
Thank you for your reply. I have come to the same conclusion, is to use periodic noise whenever possible, it's more pleasant on the ears anyway. :)

I feel that keeping noise floor as low as possible at low frequency is important, as it does have an effect on the low frequency roloff "tail" end of speaker frequency response, even after windowing is applied, measuring mid-range or tweeter for example, lower noise floor of the measurement translates to better data and reduced need for elevated SPL of the measurement. For this purpose I will keep to periodic noise.

For some additional detail, not that is is of much importance, I did this test on Motu M4, which has combination of TRS and XLR input. It claims a DC coupled input, and according to the spec sheet each has different circuitry since the input impedance differs between the two. My test used TRS output and input, and since measurement input is left open, I'm not sure how the Motu decides to use TRS or XLR input in this scenario, so I tried as well with TRS pins shorted shorted, but no difference in measurement result. I tried as well with XLR loopback input using an adapter, and did confirm that the frequency response of XLR input is slightly different than TRS, but noise floor was relatively unaffected.

FYI, yellow is XLR input frequency response, green is TRS input. I will compare channel to channel matching later today, but I have no doubt that it is very good.

1674684540134.png


Finally, I did the same test with my PC onboard audio, with similar behavior of the sine sweep noise floor.

I see this as a correctable problem in software, as I mention it is only observed in ARTA. For comparison, In Soundeasy I ran both MLS and sine sweep measurement
ARTA - middle yellow trace is MLS, bottom green is single channel sine sweep, and the other dual channel sine sweep.
SoundEasy - top trace is MLS, bottom is sine sweep, here I don't see the increasing noise floor at the lower frequencies.
1674685127777.png
 
  • Like
Reactions: 1 user
I have come to the same conclusion, is to use periodic noise whenever possible, it's more pleasant on the ears anyway.
I am an ARTA user, enthusiastically ARTA user. I can't comment on the differences you are seeing... which is to say, I can't comment intelligently :)

But I did spend some time last year confirming to my own satisfaction that I get the same measured response with either swept sine or with periodic pink noise. I experimented quite a bit with sequence length, and sampling frequency, and I came to the conclusion that PN pink noise, with averaging, gave me the most repeatable results. I measured several tweeters and cone drivers.

I settled on 128k sequence length, 96k sampling rate, and 4 averages. This gives me good resolution of the impulse response, and it matches exactly the swept sine test signal. Unlike the swept sine, the PN is much more tolerant of ambient noise.

j.
 
  • Like
Reactions: 1 users
With lower quality soundcards and in noise environment it is better to use PN pink with averaging.
yeah... I am using a super-cheap $110 audio interface, a Behringer UMC202hd. My ambient acoustic is typically 35 - 40 dB SPL, and my device under test is typically 80 dB SPL, so as long as my SNR is better than 50 dB, I am good. I am fairly certain the cheap Behringer is at least -50 dB SNR and THD+n.

If I am testing for distortion, I run the device under test up to 90 or 95 dB SPL so I can measure the distortion products in the -40 to -50 dB range. I would not trust my Behringer to make a harmonic distortion measurement below 0.1%, so I ignore anything under that.

Your baby face pro looks pretty awesome in comparison to my sad-face behringer...
 
  • Like
Reactions: 1 user
JMB, Soundeasy supports both MLS and ESS (exponential sine sweep) measurement functions, compared above.

iMat and hifijim,
Thanks for your responses. Motu M4 is not a low quality device I'm sure we can all agree on that. @DaveFred also has this device and experiences the same behaviour.

I don't suspect any real timing error here, comparing measurement from a speaker, the impulse is consistently exactly the same, at the same sample, only difference is in the noise level. Here I took a cheap bookshelf speaker, applied a 1kHz high pass filter using EQ APO, with unfiltered loopback. With 4 measurements, window of 4.5ms, I get 4 different levels of noise floor on the low end:
1674725657819.png


With periodic noise, the same 4 measurements provides much better consistency on the low end. No averaging was used in either of these measurements. This is consistency is important for using spatial measurements with VituixCAD for example, since the slope of the response will be used for merging with the nearfield response. Thankfully VituixCAD does have a "force to gradient" feature to work around this, so measurements don't necessarily need to be redone if there is an issue.

1674725867586.png



Hifijim,
Behringer UMC202HD is not terrible, perfectly usable device and really the best value for a low cost audio interface, but we should all stay away from UCA202, it's junk. If you wouldn't mind, simply run a TRS patch cord from input to output, set the reference level to -10dBFS, and record with nothing connected to the measurement channel, compare PN to Sine Sweep.
 
Last edited:
  • Like
Reactions: 1 users