Beta Test: Multitone Loopback Analyzer software

[VintageFlanker]

The image is copied into Windows clipboard. After you press the camera button, you can go to any another application where you can paste an image, including into composing a new message on ASR, and just press Ctrl-V to paste the image.

Makes sense.

 

[Moto]

Tried to use the new version. Still having flexasio crashes or errors. The reason I think flexasio is important is in the pics. Set up is d10b into the e1da cosmos adc in mono mode. One pic is using flexasio, another is asio4all v2 in REW. Another is asio4all v2 in Multitone.
Asio4all suffers from the same noise problem in Multitone that it does in REW. Thd is very similar everywhere but noise is much worse with asio4all than flexasio.

 

[Moto]

Well I got wasapi into exclusive mode and the noise is gone. In fact it’s too good to be true.
Here are the numbers with no averaging. Noise is way better than amir measured. Why would this be the case?

 

[KSTR]

For D10B use ASIO driver and all will be fine, and in general always use ASIO when a device offers a native driver.

Noise is way better than amir measured. Why would this be the case?

I'm not seeing anything that is way better here. It all depends on the bandwidth used for the RMS noise measurement.
FFT noise floor is hard to compare as we do not know what FFT size Amir has used. Assuming it is about the same give or take 6dB (factor of two), then I would see the noise floor is the same, at around -160dBFS.

 

[pkane]

View attachment 200049Well I got wasapi into exclusive mode and the noise is gone. In fact it’s too good to be true.
Here are the numbers with no averaging. Noise is way better than amir measured. Why would this be the case?

Hi @Moto , as @KSTR also said, I'm not clear what noise do you mean? Is there a specific measurement you're looking at, or just the position of the noise floor on the chart? Some of the visual differences are for sure related to different FFT sizes.

 

[Grooved]

Great, and thanks for the new version @pkane !

For D10B use ASIO driver and all will be fine, and in general always use ASIO when a device offers a native driver.

I'm not seeing anything that is way better here. It all depends on the bandwidth used for the RMS noise measurement.
FFT noise floor is hard to compare as we do not know what FFT size Amir has used. Assuming it is about the same give or take 6dB (factor of two), then I would see the noise floor is the same, at around -160dBFS.

I believe there was a post in which Amir confirmed what he's using for the 1kHz test :
44.1kHz - 256k FFT size - 8 averages

 

[Moto]

There is this which indicates for the dashboard, 10-22.4khz, 32k fft, 44.1, thought I saw 4 averages.

Article: Understanding Digital Audio Measurements

EDIT: Kind translation to Spanish by member @RamiroColasurdo here: https://www.audiosciencereview.com/forum/index.php?threads/article-understanding-digital-audio-measurements.10523/page-8#post-1486638 A detailed tutorial on how one can characterize the true performance of DACs (whether...

audiosciencereview.com

I will try testesting.

 

[Moto]

Staticv3 pointed me to this also

Marantz SA-10 Review (SACD Player & DAC)

In this case, $7500 for ... the pleasure of "tactile use?" Not for me. If you don't get it, cool. We all hopefully know our passion for audio is not a rational thing for >99% of people out there. So IMO let's disagree in a zen way, but celebrate our irrationality. :-)

audiosciencereview.com

It’s recent and says 3 averages. Post 377

 

[Moto]

I just retested at 44.1khz, 3 averages, 10-22.4k, hann window, 20 harmonics, and various fft sizes.
FFT THD+N
64 123
128 122.9
256 122
512 121.6
1M 120.2

Taking the 64k number of -123db, that is 4.6db better than amir’s -118.4db. That’s a huge difference.
The SNR is what seems so much better. When you get the mass of the fundamental, do you just do the integral between 2 frequencies to get an rms level of the fundamental to compare to the rms level of noise? What are the cut offs? Is there a hard and fast standard there or am I just confused about the calculation?

 

[pkane]

I just retested at 44.1khz, 3 averages, 10-22.4k, hann window, 20 harmonics, and various fft sizes.
FFT THD+N
64 123
128 122.9
256 122
512 121.6
1M 120.2

Taking the 64k number of -123db, that is 4.6db better than amir’s -118.4db. That’s a huge difference.
The SNR is what seems so much better. When you get the mass of the fundamental, do you just do the integral between 2 frequencies to get an rms level of the fundamental to compare to the rms level of noise? What are the cut offs? Is there a hard and fast standard there or am I just confused about the calculation?

And yet, your first post showed -119dB. Harmonics change with temperature and, therefore, with time. Even Cosmos ADC harmonic content goes down as it warms up -- Ivan even sells a heater device to heat up the chip to optimal temperature to improve THD.

But try using a better FFT window, Hann is a low-resolution window and doesn't allow for precise measurements. Use Blackman-Harris 7, or Power Cosine 7 or 8 -- these are much better.

One thing to note is that Multitone doesn't average results in frequency domain, it averages in the time domain. The noise level is thus reduced with averaging, and that increases reported SNR.

 

[Moto]

That was no averages, 48khz, so some different parameters. This is basically a duplicate of amir.
Bkackman-Harris 7 at 64K FFT is -122.4db thd+n so still 4db better than amir.
This was at the same ambient temperature but I will try it at a few random times and warm up cycles. My hunch is it stays roughly the same.
Is there a reason not to have some smaller fft sizes available?

 

[pkane]

That was no averages, 48khz, so some different parameters. This is basically a duplicate of amir.
Bkackman-Harris 7 at 64K FFT is -122.4db thd+n so still 4db better than amir.
This was at the same ambient temperature but I will try it at a few random times and warm up cycles. My hunch is it stays roughly the same.

Then you got a better D10 sample than Amir

When you first posted REW and Multitone comparisons, THD+N was shown as very nearly the same

THD+N: Multitone=-119dB, REW=-118.9dB
THD: Multitone=-125.3dB REW=-125.7dB

 

[Moto]

That was just the number of averages. If I reduce it to 1 average I get

 

[Moto]

The snr improves by almost 8db going from 1 average to 3. Thd is basically the same between 1 and 3 averages as you’d expect. That’s a huge difference in snr. Are you doing coherent averaging or?

 

[pkane]

The snr improves by almost 8db going from 1 average to 3. Thd is basically the same between 1 and 3 averages as you’d expect. That’s a huge difference in snr. Are you doing coherent averaging or?

Averaging is done in the time domain. The two waveforms are time-aligned, and then averaged. The result should be very similar to coherent averaging.

 

[Moto]

That must be the issue then. I tried REW regular vs coherent averaging. When using regular averaging going from 1 to 4 averages( can’t do 3 in REW) results in virtually no change in noise level. When using coherent averaging, the noise goes from -120 to -126.3 and thd+n from -119 to -123db which is more like what we are seeing in Multitone(8db difference in snr and 4db in thd+n).
I don’t know what amir uses but coherent averaging, I believe, is not the norm. It is used mainly just when looking at the harmonics and wanting to ignore/suppress noise.

 

[pkane]

That must be the issue then. I tried REW regular vs coherent averaging. When using regular averaging going from 1 to 4 averages( can’t do 3 in REW) results in virtually no change in noise level. When using coherent averaging, the noise goes from -120 to -126.3 and thd+n from -119 to -123db which is more like what we are seeing in Multitone(8db difference in snr and 4db in thd+n).
I don’t know what amir uses but coherent averaging, I believe, is not the norm. It is used mainly just when looking at the harmonics and wanting to ignore/suppress noise.

I'm used to doing coherent averaging in image processing, so it was my first choice in Multitone, as it's primarily designed to show HD and IMD. I'll see if I can implement an "incoherent" average in a future version.

 

[Blumlein 88]

I'm used to doing coherent averaging in image processing, so it was my first choice in Multitone, as it's primarily designed to show HD and IMD. I'll see if I can implement an "incoherent" average in a future version.

What is the benefit of incoherent averaging?

 

[Moto]

@pkane it would be great to have an averaging method comparable to standard REW or ARTA as well as coherent.

https://link.springer.com/content/pdf/bbm%3A978-94-007-1387-1%2F1.pdf

Technical Library

www.ap.com

Time vs frequency domain averaging.
Coherent really knocks down the noise quickly to reveal harmonics but as such doesn't give the same weight to noise as power/spectral averaging.
@amirm am I correct in assuming you are doing frequency domain averaging in the Dasboard?

 

[pkane]

What is the benefit of incoherent averaging?

Incoherent averaging is usually only useful in the frequency domain. It can produce a more precise frequency/amplitude measurements because it excludes time/phase from the calculation. It is also easier to do

Coherent averaging uses time-aligned samples (i.e., the same starting phase in the frequency domain). This can result in lower noise, higher SNR more quickly (fewer averages) than when using incoherent averaging. Here's a plot stolen from an on-line book Understanding Digital Signal Processing (2nd Edition). The top line is the SNR gain per number of averages (X-axis) for coherent averaging, the bottom line is SNR gain for incoherent: