Originally Posted by Jason Stoddard
So, perhaps not a big deal in software-land, but a big deal for a hardware company. Luckily, we knew what we were getting into, and had long discussions about leaving Bifrost 24/96 only, which would eliminate the need for Windows drivers. The discussions went something like this, usually in a summer-hot garage:
“Mike, we need to offer 24/192 support. There’s 24/192 music available on HDTracks,” I told him.
“What, seven tracks of it?” Mike sneered.
“It’s limited, yes, but people are asking for 24/192 support. And it would be a good differentiator for Bifrost, now that we’ve nixed the balanced outputs.”
“Balanced should only be hardware-balanced,” Mike pontificated. “Two DACs. Summed single-ended. That’s what we did at Theta.”
“I know, but don’t change the subject. 24/192.”
Mike doesn’t change course fast, though. “If we’re going to do balanced, we’re going to do it right. Hell, Bifrost would sound better if we used two DACs per channel.”
“Yeah, and it would cost $700.”
“But it would sound good!” Mike insisted.
“Yeah, and you can do that one later. For now, Bifrost. 24/192.”
“You don’t have any balanced inputs on our amps anyway,” Mike said.
I frowned. He had a point there. Balanced inputs—done right—required a 4-gang potentiometer, which we didn’t have back then. We had grand plans, sure, but no balanced inputs yet. “Mike. 24/192. Must have it.”
“24/192 uses a half-rate master clock to the DAC,” Mike said. “24/96 probably sounds better.”
“Is this a limitation of the AKM?” I asked.
“No, it’s a limitation of most delta-sigma DACs. The master clock can only be so fast.”
“I still want 24/192. Period.”
Mike sighed. “Who’s going to do the tech support?”
“Me, for now.”
“You’re going to want to shoot yourself,” Mike predicted.
“Maybe. But we need 24/192. We’ve tested 24/192. It works. Let’s support it.”
“Ohhh…kay,” Mike said.
And that’s how Bifrost got 24/192 support. It seems funny today, with 32/384 or even higher sampling rates. Not that there’s any PCM music available there, but hey, it’s like megapixels. Meaningless numbers to use in marketing. Buzzword compliance.
Note to self: we should do a 32/384 DAC that has a switch for “easy mode,” supporting 24/96 without drivers, and “expert mode,” where you’ll need Windows drivers. Except unlike everyone else, we’ll tell everyone why 32/384 is meaningless. See a couple of chapters back.
OK, I want to do my patented gentle reply to the "32/384 is meaningless" point, but first I wanted to mention something about the Bifrost shipping:
I'd asked for my Bifrost to be sent 2-day air. When I got the tracking notice, it said it was being shipped regular ground (and I was being billed regular ground rate, so no problem there). But by the time I got the notice, here in my area of the Northeast we were in the middle of a week-long blackout caused by an early-season ice and wind storm. Everything really worked out perfectly, 'cause I wouldn't have been thrilled to pay for 2-day shipping to a blacked-out house. So I sent an email recounting this (with smileys, to let you guys know I really *was* happy at how it all turned out). And got back a nice email from Rina nevertheless apologizing. Saw the time-stamp: She sent it at 2 am!
When Jason says they "made good" re customer service, and you realize what it really means is everyone, having put in 18-hour days getting Bifrosts out the door, is now sending out apology emails (including situations like mine where an apology is totally unnecessary) at 2 in the morning - well, "made good" is quite an understatement.
Now, the 32/384 thing:
I'm just going to deal with the 384 side, because on the left side the only thing a number higher than 24 is good for is making sure any digital signal processing is silent. Producers have boards that can do signal processing with 80-bit word lengths, and the whole idea of something like that is that you *won't* hear it once it's processed and put back in the digital file. For a file that's going to be immediately converted to music, anything over 24 bits is completely meaningless. Below 20-22 bits, you're into the residual thermal noise of the electronics; in other words, the only thing going on there is entropy.
But on the 384 side - well, they say the devil's in the details, and so I suppose I'm going to let the devil loose into the discussion.
Virtually every digital recording starts out in a DSD-type format, which is then converted to some higher bit rate PCM (say 24/96 or 24/192), which is then converted to Redbook rate (16/44.1) PCM for CDs, or to MP3 for downloads. Once the CD gets to you, virtually every DAC (including all those Schiit has manufactured so far) takes that 16/44.1 PCM and does two things to it: (1) uses digital filters to double the rate three times to an "8x" rate (352.8 or or 384) PCM bitstream, then (2) uses a delta-sigma modulator to convert it to a DSD-type format. Only then is the digital bitstream converted to analog music.
You see what just happened there, right? Irrespective of any discussion of the intrinsic goodness or worthlessness of 24/96 or 24/192 or DSD or whatever, what you've done is take a DSD bitstream and put it through a bunch of conversions to wind up with - a DSD bitstream! And a little further along it was higher-res PCM, after which it was converted to CD resolution, after which your DAC did a bunch of conversions to reverse that process. This would make no difference at all if the conversions were perfect. But they're not, and with the math currently used for the filters, it's actually mathematically impossible to make them perfect. (Jason has said some *very* interesting things about the Ygg filter that have me very curious about whether it will be different from all presently available DACs in this regard.) Every filter is a compromise between various aspects of filter performance. That's (a big) part of what you're paying for in a DAC - expertise in designing these filters. (Mike Moffat pretty much invented custom-programmed DAC filters, so if you want expertise, there it is.)
Since these conversions aren't perfect, the best conversion (all else being equal!! - the best guarantee of good sound is *always* a good recording, at whatever resolution) is none at all. Playing a DSD file through a DSD-capable DAC avoids all these conversions. And a higher-res PCM file, as long as it isn't just an up-converted Redbook file, avoids some of them. (A 24/192 file avoids two rounds of in-DAC doubling; a 24/96 file avoids one round. There are a limited number of 352.8 files I'm aware of - jazz and classical genre.) Avoiding downsampling to some extent before the file gets to you, and avoiding upconversion to some extent in your DAC, should provide a better sounding result (again, all else being equal).
Note: Lest everyone think this means all the 24/192 versions of your old favorites should sound better - for some strange reason, the music companies are great fans of remastering this old stuff for the modern "loudness wars" era. Many, many of these old favorites have been compressed within an inch of their lives in the new 24/192 versions, and avoiding a round or two of sample rate conversion won't come close to fixing that. Nor will it make any new badly recorded files sound great. Every so often, though, you can find some well recorded stuff available in higher res, and that's a joy.
Edited by judmarc - 4/23/14 at 1:54pm