NOS DAC in-band distortion

[manisandher]

Here's the output of my DAC running in 24/44.1 NOS mode (1kHz sine wave input):

I understand that the out-of-band (>22.05kHz) distortion is created by imaging above the Nyquist frequency. What I don't understand is what's causing the in-band (<22.05kHz) distortion.

Here's the NOS output up to 22.05kHz:

And here it is with an 8x OS reconstruction filter:

What's causing the distortion in the NOS case?

In a recent BorderPatrol DAC review in Stereophile, JA refers to it as "aliased image energy". This makes no sense to me. My understanding is that "aliasing" occurs during analogue-to-digital conversion, where any signal above the Nyquist frequency is 'folded back' in-band. "Imaging" occurs during digital-to-analogue conversion, where any signal beyond the Nyquist frequency is 'replicated forward' out-of-band.

(I've noticed Amir using "aliasing" when talking about DACs too.)

Is JA correct to use the term "aliased image energy" for NOS DAC in-band distortion? If not, what's causing it?

Mani.

 

[gvl]

Typically this is IMD products of images.

 

[Hayabusa]

Here's the output of my DAC running in 24/44.1 NOS mode (1kHz sine wave input):

View attachment 48412

I understand that the out-of-band (>22.05kHz) distortion is created by imaging above the Nyquist frequency. What I don't understand is what's causing the in-band (<22.05kHz) distortion.

Here's the NOS output up to 22.05kHz:

View attachment 48413

And here it is with an 8x OS reconstruction filter:

View attachment 48415


What's causing the distortion in the NOS case?

In a recent BorderPatrol DAC review in Stereophile, JA refers to it as "aliased image energy". This makes no sense to me. My understanding is that "aliasing" occurs during analogue-to-digital conversion, where any signal above the Nyquist frequency is 'folded back' in-band. "Imaging" occurs during digital-to-analogue conversion, where any signal beyond the Nyquist frequency is 'replicated forward' out-of-band.

(I've noticed Amir using "aliasing" when talking about DACs too.)

Is JA correct to use the term "aliased image energy" for NOS DAC in-band distortion? If not, what's causing it?

Mani.

Most of them are harmonic distortion of the 1kHz. The peak seems to be at the 3rd harmonic (3KHz, -90db?) (there is no grid, so I have to guess)

 

[manisandher]

Typically this is IMD products of images.

JA seems to make a distinction between what he calls "aliased images in the audioband" and IMD tones.

Mani.

 

[manisandher]

Most of them are harmonic distortion of the 1kHz. The peak seems to be at the 3rd harmonic (3KHz, -90db?) (there is no grid, so I have to guess)

The 1kHz harmonics are clear from the 8x OS plot. It's the extra noise between the harmonics on the NOS plot that I'm trying to understand.

Mani.

 

[pma]

It looks similarly like non-dithered 16-bit, however it is probably not the case.

 

[Hayabusa]

The 1kHz harmonics are clear from the 8x OS plot. It's the extra noise between the harmonics on the NOS plot that I'm trying to understand.

Mani.

These are below +/- 125dB, I would not worry about them

 

[manisandher]

These are below +/- 125dB, I would not worry about them

Not worried. Just trying to understand in-band NOS behaviour .

Mani.

 

[Blumlein 88]

A DAC without a reconstruction filter is broken by design. Moral of the story is use a reconstruction filter.

 

[manisandher]

A DAC without a reconstruction filter is broken by design. Moral of the story is use a reconstruction filter.

I understand why a reconstruction filter is a good idea to stop ultrasonics. But why does it reduce in-band distortion too?

Mani.

 

[amirm]

JA seems to make a distinction between what he calls "aliased images in the audioband" and IMD tones.

Mani.

That is unrelated to this comment. You are getting intermodulation of the aliased components.

 

[Blumlein 88]

I understand why a reconstruction filter is a good idea to stop ultrasonics. But why does it reduce in-band distortion too?

Mani.

Yes, see Amir's comment. IMD of the aliased ultrasonics down into the audible band. So when you use a filter it reduces those aliased signals and the IMD results of those drop out getting rid of the hashy looking noise floor.

 

[pma]

Quite high density of the spurious products, to be IMD with mirrors, don't you think?

 

[amirm]

Quite high density of the spurious products, to be IMD with mirrors, don't you think?

There are multiple tones combining so you are going to get plenty of interactions between them.

 

[manisandher]

That is unrelated to this comment. You are getting intermodulation of the aliased components.

All the spikes are spaced exactly 100Hz apart, so it's certainly IMD. Thanks.

Edit:

But can you really use "imaging" and "aliasing" interchangeably?

Mani.

 

[Mnyb]

Yes, see Amir's comment. IMD of the aliased ultrasonics down into the audible band. So when you use a filter it reduces those aliased signals and the IMD results of those drop out getting rid of the hashy looking noise floor.

+1
And imho thats why one always should reduce ultrasonic garbage out of a DAC . This IMD circus probably repeats in the preamp and power amp to if your unlucky and maybe in a transducers to if enough of it gets trough.
Wonder how an output transformer on a tube amp reacts ?

 

[manisandher]

Thanks everyone. I've learned something today .

Mani.

 

[pma]

If the lines are equally 100 Hz spaced, I doubt it is IMD.

 

[manisandher]

If the lines are equally 100 Hz spaced, I doubt it is IMD.

The 22nd harmonic lies at 22000Hz. The first image of this beyond Nyquist is at 22100Hz. The difference is 100Hz.

It's definitely IMD .

Mani.

 

[DonH56]

Aliasing usually refers to frequencies above Fs/2 being "folded" or aliased into baseband. Imaging refers to signals that are repeated at multiples of Fs. Aliasing is usually a problem for ADCs; imaging occurs at the output of DACs.

Aliasing: https://www.audiosciencereview.com/forum/index.php?threads/digital-audio-aliasing.1920/
Imaging is shown in this thread: https://www.audiosciencereview.com/...ital-audio-converters-dacs-fundamentals.1927/

Basic sampling intro: https://www.audiosciencereview.com/forum/index.php?threads/digital-audio-sampling-101.1919/

Jitter 1 and 2: https://www.audiosciencereview.com/forum/index.php?threads/digital-audio-jitter-fundamentals.1922/ https://www.audiosciencereview.com/...igital-audio-jitter-fundamentals-part-2.1926/

Delta-sigma converters: https://www.audiosciencereview.com/...igma-delta-digital-audio-converters-dac.1928/

Distortion:

Given two tones f1 and f2 we can state where distortion terms will lie:

• Harmonic distortion (HD) terms at 2*f1, 3*f1, 4*f1, ... and 2*f2, 3*f2, 4*f2, ...
• Second-order intermodulation distortion terms at abs(f1+f2, f1-f2)
• Third-order intermodulation distortion terms at abs(f1+2*f2, f1-2*f2, 2*f1+f2, 2*f1-f2)
• where abs(x) = absolute value of x

For two tones that peak at the same level the second-order IMD distortion will be 6.021 dB higher than the second-order HD. The third-order IMD will be 9.542 dB higher than the third-order HD. The derivation is straight-forward but messy. I keep meaning to write it up but it's boring.

Distortion terms above Nyquist (Fs/2) will be aliased by an ADC and produce images from a DAC.

HTH - Don