Non Oversampling Dacs

[Bigerik]

If anyone wants to discuss how these things sound, let me know.

 

[ashok]

If anyone wants to discuss how these things sound, let me know.

My experience with NOS DACs dates back several years. I used to own the Diyparadise Monica NOS DAC. It was a nice enough DAC and saw service for about a year in my system.

However, it was a little too polite (laid back?) for my liking. I have tried oversampling DACs like the Keces DA 131.1, and Meridan Explorer (still have it), and much prefer them to the Monica. In my opinion, the dynamics are a lot better - just my preference.

I have not heard recent NOS DACs like the well known Audio Note and so cannot tell if I might like to own one again.

One of the "problems" if you will, of non-oversampling technology is the appearance of so-called image frequency components in the output of the D/A convertor very close to 20kHz. This necessitates the usage of a very steep (brickwall) filter to remove those image frequency components from the final output.

Here is what I mean. Let the signal the DAC is trying to reconstruct be a sine wave at 20kHz.

Since there is no oversampling of the digital samples, the sampling frequency stays at 44.1kHz (or whatever else it might be). This means that the "image" frequency components will be created at each of (N*44.1kHz +/- 20kHz), with N = 1, 2, 3, ...

So, we will have image frequency components at:

• 44.1 - 20 = 24.1kHz
• 44.1 + 20 = 64.1kHz
• 2*44.1 - 20 = 68.2kHz
• 2*44.1 + 20 = 108.2kHz

etc.

Notice how the lowest image frequency is at 24.1kHz. If we want to retain the 20kHz signal, and eliminate the 24.1kHz signal, the filter needs to be quite steep.

If you do not apply a low pass filter and let the image frequency components through, then you could be asking your amplifying chain to amplify those image components that do not exist in the original signal.

With oversampling, the image frequency components are pushed out well beyond 20kHz. With 4 times oversampling, the lowest image frequency component will be at 4*44.1 - 20 = 176.4 - 20 = 156.4kHz.

Now, it is a lot easier to design a filter that will preserve the 20kHz signal, and filter out 156.4kHz and up.

But that is the theory. One might just prefer the sound of the NOS DAC over that of the oversampling DAC.

 

[steerpike2]

Yes oversampling tries to eliminate the signal laddering, before it is sent through the output stage.
As far as I understand, the laddering causes high frequency garbage if nothing is done about it.
If a laddered wavefrom is put through the output stage, it will be changed in to noise, due to the limitations of the output stage.
This noise will be moved to a higher frequency range, if the ladder pattern finer, this is where oversampling comes in to play.
The laddering should be so fine, that the resulting noise is moved well beyond the audible range.

The 'laddering' is and always has been - since 1982 - beyond the audible range. And even with no oversampling, the recovery filter removes it to a sufficient degree. HOWEVER, for this recovery filter to function well at sampling frequencies of 44.1kHz (or 32kHz as was common in the 80s) it is complex and costly, and has ripples that extend into the passband (i.e., it is not entirely flat in the region you want it to be flat).
Oversampling means the demands placed on this filter are less, so it is cheaper, and its flaws become less noticable.

 

[Adda]

~steerpike2

Thanks for straightening it all out, it seems I need to do more reading and find better sources, if I am to try explaining these things again.
But it seems I don't have to