32-bit capable DACs, Split from: "WASAPI plug-in version 3.0 beta" (94907) Options

[SoNic67]

I don't know why people obsess about 32 bit "support" when their device manufacturer clearly states that that format is not supported. The DAC chips included on Xonar soundcards are regular 24 bit chips...
IMO, 32 bit is a worthless waste of space, all the '32 bit capable' DAC chips existent today have actual resolution less than 24 bit. Their catalog data is equal or even less than the top-of-the-line 24 bit DAC.
It's just shamless marketing and amazingly, enough people bite the bait.
The new beta plug-in (ver 2) works fine on E-MU 1820 w/ Win7 x64 and foobar ver 1.1.12 beta 6.

This post has been edited by SoNic67: May 26 2012, 01:04

[stephan_g]

Typically, nonlinear distortion goes down with level, unless it is due to a discontinuity, such as crossover distortion. Crossover distortion is essentially impossible in a sigma-delta DAC (that covers just about every modern DAC for audio), and exceedingly rare or absent in properly designed amplifiers. Transducers can't have sharp discontinuities in their transfer curves unless there is something like sticking.

I wouldn't say crossover distortion is absent... people do tend to go to considerable pains to keep it in check in low-quiescent-current applications where it easily dominates high-frequency distortion at higher levels.
Borrowing a graph with horribly messed-up labeling from NwAvGuy:

(real labels should be, from bottom to top at 9 kHz: NJM2068 2.5x, NJM2068 7x, OPA2277 2.5x, OPA2277 7x)

The OPA2277 already was the best-performing out of several low-power parts he tried. And yes, of course the decay in distortion above ~10 kHz is only an artifact due to a 20 kHz LPF that filters out the harmonics.

The different characteristics and according distortion spectra of electronics vs. transducers mean that comparing distortion specs has to be carried out with care. Transducers usually have distortion dominated by low-order (2nd/3rd) products which still show considerable masking, while with electronics it is easy to generate a near-flat "zoo" of harmonics that decays very slowly, resulting in much tighter (numerical) requirements since now high-order harmonics have to be kept below the threshold of audibility (usually <=-80 dBr) at all times. Discontinuities + feedback at work there.

Incidentally, since headphone drivers also tend to be numerically optimized these days, I'd expect a rising discrepancy between objective and subjective evaluation of level handling. Same as when introducing feedback, basically - less low-order, more higher-order products. I'd rather not be trading 0.2% less 2nd harmonic for 0.03% more 7th or 8th (since it is quite possible to hear the latter, while -54 dB of 2nd should be a little below the audible minimum found in studies).

Re: why it is useful to have DR > THD+N, one only needs to consider that the required instantaneous dynamic range in any given playback or recording situation is usually smaller than the total dynamic range for all of them. For example, you may want your 80-odd dB of instantaneous DR regardless of whether you're using speakers with a sensitivity of 85 dB SPL / 2.83 V / m or others with 105 dB. (Or headphones with 95 dB / V vs. others with 130 dB / V.) It's mostly a matter of convenience and not having to worry about stuff.

Distortion that is signal-correlated (predictable?) might also be reduced through a feed-forward or feedback-loop?

The classic predistortion? That would seem worth a shot. Each batch of chips may need a different set of coefficients (internal EEPROM?), but it should be doable. Obviously that only applies if the analog part is the dominant source of distortion.

Premium DACs are already using balanced output so they're largely getting rid of even-order distortion, but what did one have to do to get rid of odd order again? Think it involved a Hilbert xform for a 90° phase shift? Sounds kinda impractical but you never know.

As far as 32-bit DACs are concerned... seems like you're losing about 4 bits with undithered output, so you're only getting 20 bits from a 24 bit converter. At 192 kHz we're already having 4X oversampling though, for an effective 22 bits in the audio band. So if anything, it would be of concern for 44/48 kHz and more or less "synthetic" material. A very, VERY small fraction of cases.

[Arnold B. Kruege...]

Typically, nonlinear distortion goes down with level, unless it is due to a discontinuity, such as crossover distortion. Crossover distortion is essentially impossible in a sigma-delta DAC (that covers just about every modern DAC for audio), and exceedingly rare or absent in properly designed amplifiers. Transducers can't have sharp discontinuities in their transfer curves unless there is something like sticking.

I wouldn't say crossover distortion is absent... people do tend to go to considerable pains to keep it in check in low-quiescent-current applications where it easily dominates high-frequency distortion at higher levels.
Borrowing a graph with horribly messed-up labeling from NwAvGuy:


(real labels should be, from bottom to top at 9 kHz: NJM2068 2.5x, NJM2068 7x, OPA2277 2.5x, OPA2277 7x)

The OPA2277 already was the best-performing out of several low-power parts he tried.

Key words: "He tried". Just two of how many 100s of similar parts.

You seem to be oblivious to the fact that the above plot is totally inappropriate for the purpose of showing the existence or absence of crossover distortion.

The OPA 2277 is not designed or recommended as a headphone amplfier by its manufacturer: OPA 2277 Manufacturer's spec sheet and application information Neither is the NJM 2068. NJM 2068 Manufacturer's spec sheet

Letsee: Irrelevant data about irrelevant parts.

<Shaking head>

This post has been edited by Arnold B. Krueger: Jun 7 2012, 12:41