DAC- which way- oversampling or non-oversampling

[Rave]

Hello!

I've been reading for past few days a lot of info about subject... For example:

http://groups-beta.google.com/group/...0af6e619609276

and:

http://www.sakurasystems.com/articles/Kusunoki.html


So, this is theory mostly, but how is in practise? Is there so audible differences if to compare both aproaches assuming that there is no mechanical/PCB design flaws?

[ThorstenL]

Konnichiwa,

Originally posted by Rave
So, this is theory mostly, but how is in practise? Is there so audible differences if to compare both aproaches assuming that there is no mechanical/PCB design flaws?

Yes, absolutely.

Sayonara

[gmarsh]

my personal thoughts on Non-OS: oversampling is superior.

And it all comes down to this belief: A "good" DAC should produce *no* output beyond the input nyquist frequency, eg. 22.05KHz for a CD. And with a "good" DAC, the passband must be interfered with as little as possible, even if it's only a frequency response alteration.

If you somehow listened to the output of a Non-OS DAC directly (by jamming its output wires into your ears or something) then you wouldn't hear images above 22KHz and everything would be OK. However, the output of your DAC is fed through I/V converters, power amplifiers and various other things which have limited bandwidth and linearity - there's lots of ways for those high frequencies to find themselves falling back into the audible range. Some poorly designed amplifiers can even go unstable when fed with high frequency content...

Filtering the output of a Non-OS DAC using an analog filter is *painful*. I remember seeing TDA154x based CD players which had 9th order elliptical filters that included tunable coils and potentiometers, and the response of these filters still wasn't all that great.

And another thing - a Non-OS DAC doesn't put out an impulse chain, it puts out a staircase. This causes a zero order hold (sinc) frequency response, which causes a *big* reduction in treble. Due to zero-order hold, attenuation at 15KHz with a 44.1KHz sampling rate = sinc(pi*15000/22050) = 0.39486 = -8.07dB!

Oversampling with a properly designed digital filter will...

- not affect the stopband in any major way, provided a good FIR filter is used.
- require a far less aggressive output filter, with no manual tweaking or adjustment needed, and a much nicer output response.
- mitigate zero order hold response; At 4X, 15KHz attenuation is -0.417 dB. At 8X, -0.104dB... much nicer than -8dB. The interpolation FIR filter can even be designed with a shape that counteracts the output DAC's sinc response, eliminating it almost entirely.

The only big issues are...

- More hardware is required for the interpolation filter. An extra chip, sometimes extra clocks, and other things.
- The DAC must run at a higher sampling rate. Usually this requires a different, more expensive DAC.
- Power consumption goes up.

But for the gains involved, IMO it's far worth it...

Of course, this is a completely technical reasoning. Wether Non-OS or OS actually *sounds* better is a holy war I don't want to get into...

[Jocko Homo]

The higher the order of oversampling, the more likely you are to need a really good clock. Placed directly at the DAC chip.

Non-o/s is more forgiving of a mediocre clock, and perhaps that plays a part in its popularity.

Both benefit from a good I/V.......which in my world is a true transimpedance amp. Not all that hard to build.

Jocko

[carlosfm]

Originally posted by gmarsh
And another thing - a Non-OS DAC doesn't put out an impulse chain, it puts out a staircase. This causes a zero order hold (sinc) frequency response, which causes a *big* reduction in treble. Due to zero-order hold, attenuation at 15KHz with a 44.1KHz sampling rate = sinc(pi*15000/22050) = 0.39486 = -8.07dB!

You need to compensate for the treble roll-off on the analog stage, and filter.
But a simple 1st order filter does the job.

That's why I laugh when I see NON-OS dacs with passive I/V and direct outputs.
Or those guys use tone controls, or very "peaky" fullrange drivers.

That's it, I said it.

[ThorstenL]

Konnichiwa,

Originally posted by gmarsh
And another thing - a Non-OS DAC doesn't put out an impulse chain, it puts out a staircase. This causes a zero order hold (sinc) frequency response, which causes a *big* reduction in treble. Due to zero-order hold, attenuation at 15KHz with a 44.1KHz sampling rate = sinc(pi*15000/22050) = 0.39486 = -8.07dB!

Actually, it is around 1.8db, NOT 8db. Check your math please. BTW the sinc rolloff is quite easily compensated. I have documented several approaches a good few years ago.

Finally it all comes down to preference. My ears tell me that non-oversampling DAC's sound more realistic, truer to real music.

Sayonara

[Rave]

Originally posted by Kuei Yang Wang
Konnichiwa,

Actually, it is around 1.8db, NOT 8db. Check your math please. BTW the sinc rolloff is quite easily compensated. I have documented several approaches a good few years ago.

Can you target me som some materials about these aproaches, please?

Finally it all comes down to preference. My ears tell me that non-oversampling DAC's sound more realistic, truer to real music.

That is what I had seen people are reporting- non oversampling is more realistics. But seems that I'll have to build both myself.

Thank you guys for your time!

[ThorstenL]

Konnichiwa,

Originally posted by Rave
Can you target me som some materials about these aproaches, please?

Some is posted in my Technical Yahoo group:

http://groups.yahoo.com/group/Thunderstone_technical/

Pedja Rogic picked up on these and showed a variety of simulations.

Sayonara

[gmarsh]

Originally posted by Kuei Yang Wang
Actually, it is around 1.8db, NOT 8db. Check your math please. BTW the sinc rolloff is quite easily compensated. I have documented several approaches a good few years ago.

Finally it all comes down to preference. My ears tell me that non-oversampling DAC's sound more realistic, truer to real music.

Sayonara

*checks math*... *smacks head* You're right, and it's -1.72dB. ZOH causes infinite attenuation at Fs, not nyquist. Worst part is that I do signal processing for a living.

But regardless, the attenuation is still there. It can be compensated for adequately using a basic 1st-order pre-emphasis filter, but this requires increasing the order of the low pass filter. Which may or may not be a problem... personally, I detest high-order analog filters - getting a perfect response in a land of 1% resistors and 5% capacitors is painful.

Again, the 'sound' of Non-OS versus OS is something I don't want to get into - I'm one of those stereotypical spec-minded engineers that believes that an audio reproduction chain should be as transparent as possible, and I measure 'transparency' quantitatively on the Audio Precision gear. So based on that, the engineer in me is strongly Pro-OS.

Mind you, another part of me loves the sound of a record player playing a dirty/scratched vinyl, with the resulting snap-crackle-pop fed through a tube amplifier into a set of **** speakers. It sounds like all hell, but it reminds me so much of childhood and discovering my dad's record collection. You can't measure that kind of joy with a THD number.

[ThorstenL]

Konnichiwa,

Originally posted by gmarsh
*checks math*... *smacks head* You're right, and it's -1.72dB. ZOH causes infinite attenuation at Fs, not nyquist. Worst part is that I do signal processing for a living.

Oh well, almost non of us are perfect.... ;-)

Originally posted by gmarsh
But regardless, the attenuation is still there. It can be compensated for adequately using a basic 1st-order pre-emphasis filter, but this requires increasing the order of the low pass filter. Which may or may not be a problem... personally, I detest high-order analog filters - getting a perfect response in a land of 1% resistors and 5% capacitors is painful.

Well, first there is no direct requirement to have a high order LPF after a Non-OS DAC. Secondly, you can compensate the HF rolloff in a number of ways, I now prefer the one shown here:



BTW, before people ask, this is for a TDA1541, which has a -2mA output current at digital silence and swings 0 to -4mA peak for full scale. Cannot be used like this with ANY other DAC.

Sayonara