Demand drives supply, I guess...
Demand drives supply, I guess...
In that case, imagination drives demand and under their conditions supply affects price above and beyond. You interested in a pair of Beats?
Now you're just trying to start something
The Pinch-Off Problem
Late in the development of the Gungnir analog stage, I was sitting in the living room, talking to Mike on the phone about a problem we’d been having on the prototype boards.
To me, this was “just another day in engineering.” Weird schiit happens. You gotta figure it out. So I didn’t think anything about our conversation…until Rina walked into the room, laughing so hard she could barely stand.
“What?” I asked her.
She just laughed harder, holding on to the kitchen counter to keep from falling over. Literally.
“What’s wrong with you?” I continued.
“You—“ she said, gasping and pointing. “You—you—“
“Me me what?”
“You have a company—called ‘Schiit’—and you’re talking about your—pinch-off problem!” Rina said, through gales of laughter.
I stopped dead. Then I started laughing, too. Schiit had a pinch-off problem.
Background: Pinch-Off is an Engineering Term. Really.
Well, more accurately, it’s an old-guy engineering term, like “plate” instead of “anode” for tubes. What it describes is the voltage it takes to turn off a JFET (or other depletion-mode device.) As described in the previous chapter, a JFET runs current through it from drain to source as soon as it’s connected to a voltage. But, if you lower the voltage at the gate of the JFET, eventually it won’t conduct at all. That’s known as Vgs(off), or, in old-dude speak, the pinch-off voltage.
So Mike and I had been sitting there, talking about our pinch-off problem, and that’s what Rina had walked in on.
Yeah, I know, stuff like this surely doesn’t happen at Sony…
Now, as far as why we were discussing our Vgs(off) problem, it was simple. I’d built some perfboard versions of the Gungnir analog stage, measured them, and been very happy with the result.
Then we got the PC boards, put the parts on them, and suddenly they were running 50x the distortion of the perfboard versions.
Yeah. Stuff like this happens all the time in engineering. The real world isn’t the same as simulations. PC board protos are different than built-in-the-air protos. Manufacturers change processes and parts don’t work the way they used to. If you want a simple life, consider a career as a fisherman in Costa Rica or something. Maybe. Who knows. I’ve never actually done that, so it might be as bad (or worse) than engineering.
So, after spending a night with Mike and Dave trying to chase down the distortion (looking at compensation, oscillation, badly routed PCB traces, bad solder, wrong parts, bad parts etc—all the obvious stuff that shows up on prototypes), we were all baffled. The circuit worked, but it didn’t work well.
So I went home and slept on it. Problems that seem huge at night sometimes become really obvious the next day.
But this one wasn’t.
I went back to it that evening, swapping parts and measuring. And a parts swap did make things better. Just not enough better. So I swapped parts again, just for the heck of it. And it got worse.
That was when the light bulb went off. The in-the-air prototypes weren’t built with surface mount parts. They were built with through-hole parts. The JFETs we were using on the PC boards were supposed to be a near-equivalent to the through-hole parts…but maybe they weren’t.
The datasheets told the story: the through-hole parts we’d been using on the perfboards had a pinch-off voltage range that was very small, and spec’d pretty tightly—about 0.2-0.5V. The ones we were using on the surface-mount PCB? 0.5V to 6V.
Yeah. 12x different. Like, duh.
And, considering where they were used in the circuit (as followers), that big pinch-off could cause all sorts of problems. I tacked some through-hole parts in their place, and suddenly the boards were acting (and measuring) just like the early prototypes.
From there, it was only a matter of finding a surface-mount JFET with similar specs, which only took a quick web search. A few days later, when the parts came in, Gungnir’s analog stage was working as it should.
But only after a painfully hilarious conversation…
But Analog Isn’t The Real Story
Okay. I front-loaded this chapter with a funny story about Gungnir’s analog stage, but in reality, that was probably the least interesting part of the DAC’s design. When Mike said he wanted to do a much more no-holds-barred design, I knew exactly what I wanted to do on the analog side. That is, a more sophisticated discrete stage, with a better topology, with higher voltage rails, and this time using a DC servo rather than coupling capacitors on the output. Other than the pinch-off problem, the development was relatively uneventful.
With Gungnir, the real story was on the digital side. Like Bifrost, we started with not much in the way of a product brief, except for Mike Moffat stating that he wanted to “do a proper hardware-balanced DAC.” Beyond that, nothing. No sizes. No feature sets. No colors. No 500-page list of specs.
But Mike is very, very specific when it comes to digital. “It needs to be big enough to keep the Hatfields and McCoys out of each other’s corrals,” he declared. “That means two transformers, one for digital and one for analog. And clock regeneration, we need to look at that a lot harder. And absolutely, positively hardware balanced, none of this single-DAC-per-channel stuff.”
Big enough to keep the Hatfields and McCoys away from each other. In Mike-speak, this means careful segregation of the analog and digital sections. Grounds. Power supplies. Clock routing. Physical space. Mike looks askance at tiny products that mix analog and digital. So, Gungnir was gonna be big.
Clock regeneration, we need to look at that. Bifrost uses a lot of the tricks Mike learned to get SPDIF jitter to acceptable levels, but Mike’s work at Theta always featured VCO clock regeneration, and he hated to give that up. Eventually, that grew into Gungnir’s unique Adapticlock system, which actually assesses the quality of the input signal (in terms of center frequency and jitter) and routes it to either a VCO or VCXO oscillator. It also meant that Gungnir needed a much bigger and more powerful microprocessor to do this analysis and routing, for all supported input resolutions and sample rates.*
*Actually, let’s talk about that a bit. In the old days, you only had to worry about 16/44.1 and 16/48. Now there’s a LOT more variations. And if you’re interested in keeping everything bitperfect, that’s a hell of a lot of management. Why do Gungnir and Bifrost click when you change sample rates? Because you have to reset the whole system to run at the new clock multiple.
Absolutely, positively hardware balanced. Hardware balancing, or using one stereo DAC per channel, pays off huge dividends. Lower distortion, lower noise floor, elimination of more of the high-frequency noise that comes out of a modern sigma-delta DAC—these are all wonderful things. It also comes at a cost of using two DACs and twice the analog components, plus discrete summers for single-ended output.
So yeah, digital is the real story. And the real story of Gungnir is probably Adapticlock, it’s unique feature. That’s a Mike Moffat original that he’s justifiably proud of. As far as we know, no other DAC tells you if your source is good or bad, and, even if bad, still provides clock regeneration. It took a ton of code to make that one work—and some very expensive VCXOs.
“If it’s bad, we’ll light up a front panel light,” Mike said. “We could call it the ‘buy better gear’ light.”
And “Buy Better Gear” is what stuck. It’s technically the “VCO Mode” light, but that’s a whole lot less interesting, right?**
**And there’s not a lot of really bad gear out there, to be honest. Pretty much any computer won’t light it. It really only comes on with really, really awful stuff, like satellite receivers and Apple Airport Express sources. And some old CD players that have gone off-frequency. That’s about it. Everything else runs in high-precision VCXO mode.
In the spirit of the last chapter, let’s talk about the parts of a digital audio system, so hopefully all of this stuff makes a little more sense:
Storage. Digital music has to be stored, whether it’s on a plastic disk, magnetic disc, or in the cloud. At this point, it is no different than any other data you have. And, like other data, it can be lost if your hard drive is made by Western Digital (er, I mean, when it breaks.) Sorry, WD has had the majority of breakage in my personal experience—this is not a statistically significant result, just a personal opinion. The important thing is to make sure it’s backed up. Or, if you’re still a dinosaur (er, I mean, using plastic disks), don’t treat them so poorly as to destroy them.
Formats. Most of you guys already know this, but let’s go ahead and be inclusive. Digital music comes in tons of different formats. Let’s cover three broad swathes:
Transmission. From the stored digital file, you need to get it to where its going. This should be relatively simple, but sometimes it isn’t.
Reception. Beyond the digital connection, there’s a receiver to process the incoming SPDIF or USB signal. In the case of SPDIF, it recovers the clocks that are embedded in the data. In the case of USB, asynchronous transfer controls the clocks locally.
Clock Management. Okay, so you have digital data. Now what? Some manufacturers choose to upsample everything to a specific datarate, no matter what’s coming into the box. This eliminates the need for clock management, but…you guessed it…asynchronous sample rate conversion, or ASRC, is not bitperfect—it replaces the original samples. So, for a bitperfect DAC, this means clock management is necessary. In short, this is the process of telling the digital filter and the DAC, “Hey, I’m sending you 16/44.1, get ready, ‘kay?” Or 16/48 or 24/96 or 32/192 or whatever. Run through the different bit depths and sample rates, and you’ll quickly see that there are many different combinations. Clock management isn’t trivial.
Digital Filter. Digital filters are where bit-perfect transfer usually dies. Digital filters upsample the incoming data to higher data rates (typically 8x) to reduce the need for analog brickwall filtering. This is handy, but again—what it outputs is a mathematical approximation. That is, unless it is a closed-form digital filter that retains the original samples. And we know of only one of those—which is what’s going in Yggdrasil.
D/A Converter. From the digital filter (which may be inside the DAC chip itself), the data gets passed off to the actual D/A converter. These typically come in two varieties:
Analog. And you thought we were done? No. Some DACs output current, which requires an I/V converter. This is a place where discrete designs, with proper low-impedance inputs, can offer huge advantages over ICs. Some DACs output voltage, but it still needs filtered, and, in some cases, summed. So there is an analog component at the very end—and it is definitely critical to the performance of the entire system.
Although I can go on and on about the engineering side, I’ll spare you the huge dissertation, as in the last chapter. Because, the more I think about it, I believe the central thing comes down to philosophy.
“Philosophy? What the heck does that have to do with DACs?” you might ask.
I’ll respond: It has everything to do with DACs. And amps. And business in general. So let’s move on to that.
Philosophy: Or, Why You Do Something
Okay. Let’s say you start a company. Why did you do it?
Let’s say that company makes products. Why did you do one, and not another?
Let’s say your products are made a certain way. Why did you choose that method?
Let’s say your products have certain features, or you choose to leave them out. Why?
If you keep asking, “Why?” you might find that there is a good reason behind all of the answers. Or you may find none at all. And, even if you find there is a good reason, you may not agree with the “why.” This is why companies have to have a philosophy—and stick to it. Because otherwise, they’re rudderless. Aimlessly wandering. Waiting for a magical hit product to pull them out of the morass. You see this in a lot of big companies—ones where products are created and approved by giant committees, endless meetings, thousands of hours of “gaining consensus.”
Car companies are a great example. How many midrange cars can you simply swap the badges on, and not even know it’s made by a different manufacturer? How many are so completely forgettable. How many seem to lose their way every decade or so—and then suddenly release a flood of models that simply copy the one hit they recently had?
There’s no philosophy. Just the endless chase of benchmarking and specs-list-stuffing and crossing your fingers and hoping that somehow, something with enough personality makes it through the bean-counters and second-guessers to make a difference for the line.
If you start a company, make a product, decide on features and specs, ask yourself, “Why?” And be specific with your answer. You’ll do a lot better for it.
So what do we mean by this? Fine. Let’s use Schiit as an example.
Our philosophy is that we want to make fun, affordable products that are as true to the musical source as possible.
Note what comes first, second, and third.
First, fun. Come on, guys, this business is far too serious at times. Let's have some fun with this.
Second, affordable. The elephant in the room in high-end audio. On a recent panel, some guys tried to make the point that "personal audio" wasn't really any younger than high-end audio, citing examples from companies making $1000-5000 products. Like, duh. I reminded them that our audience was much younger than the norm, statistically, because of one simple thing--they're affordable.
Finally, true to the source. In the digital realm, to us, this means retaining the original sampled data as much as possible. In the lower-priced realm, this also means forsaking the totality of this goal, because modern delta-sigma DACs sacrifice the original samples for a mathematical approximation.
Now, this isn’t to say these DACs can’t sound good. And, congruent with the third part of our philosophy, we choose to preserve the original bit depth and sample rate as far as possible down the chain, and to minimize errors caused by jitter. That’s the best we can do with delta-sigma.
(Now, very soon, we won’t have to compromise on this, but that’s a substantially more expensive product—though still much less than most megabuck DACs. And that takes a digital filter that runs an algorithm that retains the original samples—like the unique one we have in store. But note that this is congruent with our overall philosophy and goal.)
“Wait a minute,” you say. “You ‘want to stay as true to the source as possible,’ but you do tube amps and stuff like that. What’s up?”
What’s up is that tube amps don’t necessarily have to be high distortion, and, even then, relatively high levels of low-order THD aren’t correlated highly with audibility. And a lot of people think tube amps are fun. So we’re hitting the second part of our philosophy. If we were saying, “You must use tubes, and a great tube amp is $50,000,” that’s antithetical to our philosophy. But fun, accurate tube amps are not at all.
“But I don’t agree,” someone is saying. “You might have a philosophy, but I don’t agree with it.”
Yep. And that’s the thing. You’ll never have 100% consensus. If you don’t agree, simply move on, and find a company that meets your own goals. That’s why there isn’t one The Amp Company out there to rule them all. It’s a wonderfully wide, varied world—and that’s a very good thing.
What we can say is that we do actively ask ourselves why, and bump up the answers against our philosophy, as we develop products. Let’s see how that works:
Hey, a D/A converter manufacturer brought out a new delta-sigma device. Let’s build a new product! No, sorry. Not congruent. Just because it’s new doesn’t mean it’s better. Is it a meaningful upgrade? Inexpensive? Then maybe.
Hey, a D/A converter manufacturer brought out a new R2R device. Let’s build a new product! Hell, yes! R2R DACs are bitperfect. So let’s go for it—well, unless it’s crap or a billion dollars.
You know, some people really like the old, lush euphonic tube sound, let’s throw a really nonlinear tube in our DAC! Nope. Although it can be pleasant, and although some prefer it, it’s not staying true to the original, is it?
You know, we can make something even better and more fun, but it’s even better and less expensive than our top-of-the-line products. You bet, let’s do it! Everyone benefits!
Did you see that there’s a new Class D module that puts out 300WPC from something smaller than a cigarette pack? Let’s throw it in a DAC and make a power DAC. Not gonna happen—throwing away the signal for a nonlinear-control-system approximation of it? Not our bag.
Anyway, perhaps you see where I’m going here. Philosophy is key. Whys are important. But by sticking to them, you won’t please everyone all the time.
But then again, trying to please everyone all the time won’t do that, either.
always good reading. The Philosophy part is very much to my liking - and I guess You will be cited a lot over the next days. Especially in the brand spanking thread of 2 certain new DAPs by a korean company.
The underlying understanding of some companies philosophy is what many are missing. If those companies were able to transport their philosophy to their customers - assumed they do have such a philosophy other than "squeeze out as much as You can" - they would probably be able to sell a lot more of their products because their potential customers would understand WHY they are doing what they do.
So Kudos to you for bringing this on the table.
Another great chapter, Jason.
It was because of Schiit's philosophy that I bought my Bifrost, never having seen or heard it before. I wasn't wrong -- there isn't a day that goes by that I don't enjoy it. Well done and keep up the good work.