virtins.com

Hello

Here I am again

I'm playing with CD tests.
For that, I need 16 bits test signals.

But I'm struggling with Dithering.

1. When I play an already dithered 16 bits wav file (generated from another tool), MI is adding 16 bits dithering again.
Of course results are then impacted.
I can run the generator in 24 bits to fix this, but that won't always work.

2. Comparing AP generated reference signal to MI generated, MI signal shows less noise but more distortion.
I guess that's because they apply dithering differently.

Could we get control of dithering in the generator ?
Default settings is good, but it would be helpful to be able to play with the parameters.
And/or to disable the dithering completely.

The signal generator of MI does not add dither to the generated signal implicitly. To add dither to a signal, you can use the Multitone function and add a small amount of white noise to the signal. Examples can be found in
Section 4.2 of https://www.virtins.com/doc/Measurement ... rument.pdf
and
Section 2.7.2 of https://www.virtins.com/doc/Measurement ... rument.pdf

Thanks for your answer.

Sorry, my misunderstanding.

That explains why SNR AND THD are both higher than AP, which uses 16 bits dithered signal.

I understand the Multitone generator option.
Nevertheless, it would be handy to add "Dither" options in the generator directly, IMO.
A button accessing a faceplate for setting dither on/off, function, bit depth, strength (?), would be very welcome.
Of course, I understand this would have to be saved to PSF as well.

Also, the Multitone has one major drawback as is, in my opinion:
Unless I missed something, we can't store frequencies as float values in multitone, so it's not possible to avoid spectral leakage using multitone
(fixing this would also be high added value, by the way).

Also, with the WAV file, does it add dither ?
Since it can't keep the file bit perfect, I suppose this is the case.
Correct ?
Disabling that would also be helpful.
As would be, then even more, handy to add dither in some cases, especially when we lower the WAV file playback level, since, even if the WAV file contains dither, it will decrease in level and therefore become innefficient.

I'm started measuring 16 bits digital results, and that's very important to keep dither under strict control to be able to have reproducible, meaningful and comparable measurements.

Adding dither to a signal with an absolute amplitude specified on the signal generator panel may cause inaccuracy in the specified absolute amplitude under certain conditions, e.g. [absolute amplitude] + [dither strength] > full scale and make things on MI complicated. The best place to add dither in MI is still through multitone configuration where relative amplitude ratio of the original signal and dither can be readily specified.

The Multitone function of the signal generator allows up to 200 tones to be added together in each channel. It would have been impossible to generate such a stereo signal continuously in real time on an ordinary computer if we had not limited the frequency of each tone to integer values. Also, it is impossible to avoid spectral leakage for a multitone consisting of more than one component, unless all the frequency components in the multitone are harmonics of a common fundamental frequency.

The WaveFile function of the signal generator does not add dither. It allows the alteration of the original samples through resampling, rescaling, masking, fading, modulating. It is not designed to keep the file bit perfect.

Thanks for your answer.

VirtinsTech wrote:Adding dither to a signal with an absolute amplitude specified on the signal generator panel may cause inaccuracy in the specified absolute amplitude under certain conditions, e.g. [absolute amplitude] + [dither strength] > full scale and make things on MI complicated.

I understand that.
The level can be somehow approximative then.
That's an acceptable drawback if you consider the benefits though.

VirtinsTech wrote:The best place to add dither in MI is still through multitone configuration where relative amplitude ratio of the original signal and dither can be readily specified.

I'd have no problem with this IF I could make the Multitone without spectral leakage.

VirtinsTech wrote:The Multitone function of the signal generator allows up to 200 tones to be added together in each channel. It would have been impossible to generate such a stereo signal continuously in real time on an ordinary computer if we had not limited the frequency of each tone to integer values.

OK, I didn't consider that point. I understand.
But this is a MAJOR drawback.
Would it be possible to allow float values with a more limited number of tones (say 40) ?
I don't know how to deal with that, but that would really be a major lock removed for me.

VirtinsTech wrote:Also, it is impossible to avoid spectral leakage for a multitone consisting of more than one component, unless all the frequency components in the multitone are harmonics of a common fundamental frequency.

Strange. (Or maybe I don't understand it ?)
It seems I can create composite signal without spectral leakage for 2 or even 32 tones without them to have to be harmonics of a common fundamental frequency.
That's what Audio Precision does for their 32 tones Multitone, as an example, and they use that characteristic to even retrieve the noise from the bins where they know for sure there can't be any harmonics.
(See https://www.ap.com/technical-library/using-multitones-in-audio-test/)

VirtinsTech wrote:The WaveFile function of the signal generator does not add dither. It allows the alteration of the original samples through resampling, rescaling, masking, fading, modulating. It is not designed to keep the file bit perfect.

OK, I can live with that.
I can just run the file at the next bit depth level if needed.

Thanks for the information. It is very helpful.

I'd have no problem with this IF I could make the Multitone without spectral leakage.

A "No Spectral Leakage" checkbox will be added in the MultiTone Configuration Dialog box. It will be enabled for selection if all tones are sinewaves. Other waveforms including white and pink noises cannot be added if this option is selected. No dither can be added either. The maximum number of tones in each channel will still be 200 for this option. The actually generated frequencies will be the nearest no-spectral-leakage ones to the configured frequencies. When this option is selected, the frequency specified does not have to be rounded off to an integer value, and thus it is possible to aim at any FFT bins.

It seems I can create composite signal without spectral leakage for 2 or even 32 tones without them to have to be harmonics of a common fundamental frequency.

They are actually harmonics of [Sampling Frequency]/[FFT Size]. The generated Multitone is still periodic.

That will be agreat addition !
Thanks