SBAF DAC Talk II

That does not describe my sonic experience with my PSA DS DAC (Snowmass) through a 6SN7-based preamp and a KT88-based push/pull Class A/B ultralinear amp.....and of course my allegedly crap Tekton DI speakers.

What I hear is a good and proper amount of nice tight bass with rich (but not bloomy) midrange. The treble is present but certainly not harsh, bright, or grainy. The lows, mids, and highs are well balanced. What I get is great detail and a 3D soundstage where the instrument and voice locations are well defined yet the sound is well integrated and quite musical.

 

Why do people seem so fixated on middle-midrange frequencies (~1-2kHz) and aligning comparative FR at 1kHz when, as noted above, fundamentals for most instruments come from lower down? I could be very much mistaken, but isn't this largely overtone region? Reproduction of which is important for realism and spatial cues, yes, but they don't exactly carry the weight of music that drives everything.

Spoiler

To use headphones I'm familiar with as reference, the HP-3 and X00 are bassy as f**k-all but actually somewhat thin-sounding as far as instrumental (and vocal) fundamentals go, the Klipsch less so than the MDxFostex. The HD650 is closer to the Klipsch because it doesn't have a chunk taken out ~500Hz like the Fostex does, while the Nighthawk's... giftedness in this area makes me feel like I've got gristle stuck in my throat. Weirdly enough, I think the HE1000v2 has my favourite midrange of everything I've demoed, though the overall presentation was really thin (for my tastes); the way the midrange was balanced with the lower treble was pleasing to my ear. I kinda get the same on the Klipsch, save for more bass and sharper middle treble (which I guess I can live with).

*Above kinda relevant but also a tangent on comparative headphone tonal balance, so spoilered.

Going back to DACs, the original Hugo is an excellent example of sterile but with nice technicalities. I think of it as the HE1k of DACs, but with less resolution— there could be better analogies but I've not heard em. Had the pleasure of playing with one on an extended loaner, and I actually carried the thing around to school and back so I grew rather familiar with it. Don't shine a blacklight on it please. /s

As for my current reference, the Bifrost MB (older, famously pooed-on version) after a couple weeks, the Schiit had more body (some would say too much) and fuzzier background but somehow sounded less congested than the Chord on busier passages despite all that; I actually thought the Hugo sounded "blurrier" on bass drums, but my inexperienced ears might have just been fooled by the extra texture. Either way they traded punches well considering the disparate pricing and AFAIK shows that tonal balance in DACs should be a secondary concern since this can be tweaked with downstream components, not so technical properties.

 

. What I meant was "A midrange with a good body is to die for" is what describes my PSA DS DAC. In that, while not super on technicalities, the weight of the mid-range and the body of the music makes it delightful.

 

It would be better to build a golden schlong with the box and ears attached to it.

(or for a cheapo version, get a suction schlong with the big gonads, paint gold, cut in half, put lower half under the minidsp box, the upper half above the box, done!)

 

Motivated by @skem 's profile post, I am moving one interesting topic to this thread.

Roughly two years ago, benchmark posted about intersample overs and argued their dacs addressing this issue in their blog.

https://benchmarkmedia.com/blogs/application_notes/intersample-overs-in-cd-recordings

I don't like any dacs from benchmark (dislike dac1 and dac2/3 for different reasons though), but their reasoning regarding the issue sounds valid to me. It did not contradict to the recent AES talk which @atomicbob introduced.

http://www.aes-media.org/sections/pnw/pnwrecaps/2018/oct2018/

For this reason, I've applied a little bit of digital attenuation with foobar for years. But can't tell very strongly whether or not gains exceeded losses (lost bit resolution). Do this mostly for the peace of mind. Haha.

 

Very interesting stuff as it's common knowledge that different ADCs clip differently when fed hot signals.

WOW on the ODAC. Using an ODAC+ O2 as a powered pre was awful. Truly the worst piece of gear I've ever owned.

 

Yeah, odac or maybe sabre did just nothing, resulting in disasters. Some like Apple seem to apply compression process to prevent clipping. Not desirable but understandable for certain cases (like smartphone hp outs -- nobody expect much).

 

Sounds like they have servicable tooling for industry and are adept at pimping it out to consumers. Caveat emptor.

 

I see Siau's postings in a lot more of a charitable manner... he's essentially acknowledging that the DAC1 has audible issues, which is actually quite refreshing for a manufacturer.

The issue of inter-sample clipping is actually a little worse than Siau indicates in that article. The theoretical maximum he gives of +3.01 dB of inter-sample clipping is only true for continuous waveforms. For actual music, inter-sample clipping can be much higher. One person over at HA did an analysis of his entire music library and found one real track (Merzbow: Venereology) where inter-sample clipping peaks ranged from +7.15 to +11.3 dB depending on the filtering algorithm used.

Almost half of his 6486 track collection had some degree of inter-sample clipping with a typical filtering algorithm (which isn't all that surprising given how hot albums are mastered these days).

There's no "right" or "wrong" way, only trade-offs. Although one thing is pretty clear: mainstream ASRC ICs don't handle digital clipping well, and DACs that have an ASRC on input do need some degree of digital attenuation prior to the ASRC (which is pretty uncommon actually.) Adding digital headroom in the DAC is actually kind of rare though, and the ones that do it generally don't add more than 2dB of headroom. You'd be hard-pressed to find any recent voltage-out sigma delta DAC implementation that adds any headroom at all.

For my own listening, I'm a little more aggressive than Benchmark and add 4dB of digital headroom. One of my favorite albums has 3.8dB of inter-sample peaking. It's not a blindingly obvious difference either way, but for tracks that are mastered hot, adding some headroom does make them slightly more listenable (subjectively speaking).

Ideally mastering engineers would handle this a little better, but like the loudness wars, they're loathe to change what they do. Apple has an inter-sample peak measurement tool as part of its "Mastered for iTunes" suite and apparently rejects tracks for MfI certification if they have two much inter-sample clipping, which has helped prompt some mastering engineers to pay a little more attention to this recently.

 

I exchanged emails with John Siau two years ago on this topic. He was forward leaning in sending me test files since he did not know how, and was curious how, my then-current DAC would behave. But I then asked him:

‘Is this effect common enough in music to really matter. ​

If I understand, the intersample overs only become a problem if a passage is at the threshold of clipping, and that such a thing is probably very rare in most recordings. The article on your website mentions it occurs >500 times on a particular track, but that still seems rare to me, if these are 500 * 0.0000227 seconds each = 11ms of distorted audio.”
​

At which point he stopped responding.

 

Interesting. But what does HA stand for?

More specific reference points (hyperlink, publication citation, etc) will be appreciated. Want to see how he conducted his examination.

 

I would think HA means Hydrogen Audio https://hydrogenaud.io/

 

HydrogenAudio. I didn't link to the thread because I wasn't sure about the rules, but here it is since you asked: https://hydrogenaud.io/index.php/topic,98753.25.html

There are two people who did analyses of their collection in that thread with slightly different results. You have to read to the end of the thread because some of the early results are a little off and needed to be repeated.

If you want to try with your own collection and have a Mac, there's a command-line tool here that you can run that will identify and quantify the inter-sample peaks: https://www.apple.com/ca/itunes/mastered-for-itunes/ You can also have it create a new audio file with the inter-sample peaks marked so you can listen for yourself as to whether you can hear any audible effects.

 

Intersample overloads is a real thing. Soren Nielsen and Thomas Lund of tc electronic wrote a paper about it in 2003 with plenty of examples: http://cdn-downloads.tcelectronic.com/media/1018176/nielsen_lund_2003_overload.pdf

The problem occurs more often with very compressed music especially stuff with lots of high frequency energy. And it’s silly to say it’s only 1 millisecond or whatever of samples because how long does a drum hit last? When you hear it happen, especially in the cases where the DAC or DSP algorithm latches up (which is where I’ve heard it), it’s very obvious.

I’m glad Benchmark is trying to address it as it’s one of the real problems of digital playback.

 

Whitepapers are typically part salespaper, otherwise they wouldn't have them on their site. You need to remember that in order to put on the proper brain filter.

 

Those happened to be among the hottest mastered albums they tested. I've found much the same thing with more niche music that's mastered hot (Mars Volta, Bon Iver's latest, etc.).

You might not think we should care, but IMHO part of the reason digital sucked for so long is because everyone along the line just tended to take this "good enough"/nothing matters attitude at every stage. This isn't even a cost issue like R2R vs. sigma delta... digital headroom is easy to add.