StimpyWan
Headphoneus Supremus
2022, Chapter 6
30 Years In Development
“You’re gonna beat those,” Mike Moffat said, pointing at the Cello monoblocks, “with that?”
That was the cheapest Sumo power amp, a Ulysses. Running $549 retail for a single stereo amplifier. Versus the $10,000-a-pair Cellos. Through Theta’s Avalon Acoustics speakers.
I shrugged. Yeah, the Sumo stuff didn’t look very exciting—it was a basic black box, with a thin aluminum front panel in front of a cheap wrinkle-finish chassis. The Cellos, for their day, were impressive—slabs of pure clear-anodized aluminum, bolted together in a clean, monolithic way.
“You already said it was closer than you thought,” I reminded Mike. We’d just compared the Ulysses to the Cellos.
Mike sniffed. “Close isn’t better. You still got those MOSFETs holding you back.”
“I already said I could change it to bipolars—”
“—but even with bipolars, I don’t know,” Mike continued, cutting me off.
“It’s all about the compensation,” I told him. “And tuning the transconductance linearization.” And, I realized, maybe tuning it to the point where the amp didn’t need any overall feedback, since the bipolars would be more linear than the MOSFETs had been.
Mike looked doubtful. He wasn’t a huge fan of the idea of transconductance linearization to start with. He drummed his fingers on the listening room couch and tilted his head back at the ceiling.
“Maybe a choke,” he said, finally.
“A choke?” I didn’t know what he was talking about.
Mike grinned. “Yeah. Maybe a choke. A big-ass choke, as big as the transformer. Maybe that would help. Of course, power output will go down, but yeah, choke-input power supply. That’s just crazy enough it might work.”
“I don’t get it,” I told him.
“Youngsters,” Mike said, rolling his eyes. “Choke input power supplies are how god intended power supplies to be made. None of this ‘capactitor bank’ stuff. What you want is a hunk of metal as big as your head, a big ‘ol inductor, right at the input. It’ll sound totally different!”
I frowned. I was still getting over my objectivist phase, and the idea that putting a big inductor in the power supply could change how something sounded was really bizarre.
“Think I’m kidding?” Mike asked. “Try it.”
So I did.
And that, arguably, is what started the development of Tyr…30 years ago.
The Turd and Turd 2
Now, before I get into the super-obvious question of “well, if it was good, why haven’t you done a choke-input amp before Tyr?” let’s dig a bit deeper into the modded-Sumo-vs-Cello deal of three decades past.
Because, upon seeing the prototype, Mike laughed out loud. “So this Turd will beat the Cellos, huh?” he asked.
To be fair, the name was accurate.
The Ulysses was never a looker. But the choke-hacked version was wayyyyyy uglier. There was no space to put the choke, so I ended up bolting it to the top of the chassis, like an old-skool tube amp. Aaaand, since the amp worked better when cranked deep into Class A, it also sported a 4” cooling fan (pinched off a computer power supply) stuck on the top as well. Between the nasty-looking EI-core choke, the whizzing fan, and the various flying wires, it was, ah, less than pleasant to behold.
“We’ll see, we’ll see,” I said.
I was still tweaking the transconductance linearization, which was needed less and less as the amp got more and more into Class A. And I was still playing with compensation, looking at the open-loop behavior of the amp far, far out on the frequency axis, wayyyy farther than any sane amp designer would look. We’re talking tens of megahertz here, stuff that shouldn’t make any difference.
“That is one ugly baby,” Mike added.
“When it beats up the Cellos and takes their lunch money, the kid’s gonna get a lot prettier.”
Mike shrugged and looked doubtful, grumbling something about “monoblock advantage.” I just kept working.
Finally, after another week of tweaking, it was time for the Sumo-vs-Cello thunderdome.
And this time, things went VERY differently.
The Turd handily beat the Cellos, pretty much by any measure. Even soundstage and dynamics, typically the purview of monoblocks, went to the Sumostein hackjob.
Mike listened for about 8 seconds before saying, “Okay, so the Turd killed ‘em.”
Then he got quiet. It took me another couple of decades to realize he was wondering what he might be able to do with a $550 retail amp, that, with another $30 or so in parts, was killing some of the most expensive amps out there. Yes, I am that thick.
“Monoblocks,” he said, finally. “What can you do with monoblocks?”
“Oh hell that’s easy,” I said. I’d been working on an extension of the differential design of the Andromeda 2—one that was not such a kludge, and also that used less parts, because (a) Sumo was really small, and (b) sometimes I had to help build the stuff, which meant that stuff with less parts meant I didn’t have to do it so often. “I have exactly the design, a differential—”
“Cool, so build them,” Mike said.
“Sumo won’t—”
“I’ll buy the boards, blah blah, let’s just get these Turd 2s going. In the meantime, I’m taking the Turd.”
That was pretty cool, I thought. Also a bit unnerving. I was working with Mike on Cobalt at the time, doing the Cobalt 307. Would this mean, like, amps? From Theta? Would it mean I was moving on from Sumo?
Whatever. When you’re young, you just go for it. And I was very, very young.
And Mike really did love the Turd. He took it home and used it with some big bizarre speakers that someone else had did for him, also horribly ugly in unfinished MDF and fiberglass. He may still have it today. Maybe it even works. I don’t know. Don’t get your hopes up.
In the meantime, I got on the Turd 2s. These, for the sake of expediency, were done in junk Polaris 2 chassis, so I had more heatsinking to play with. I also had a really, really bad layout, which meant they never really worked right. Or, well, they kinda-sorta worked, but the whole current-feedback thing was new to me, and they were very, very squirrely. Eventually I got something we could listen to, but it was always unstable.
Fun fact: Nexus is the perfected topology I was working towards at that time.
The Turd 2s, even unstable, were good enough that Mike said, “**** it, we need to go after the Krell Audio Standard.”
I gasped. Mike was talking about monoblocks that were big and stupid and car-priced. Like really good car priced, like $32,500 back in the early 1990s.
“We could kill them for half the price,” Mike said.
“Or sell a pair of monoblocks for $1 more with way better specs, and ask in the ad, “Would you pay $1 for choke input and differential current feedback and—”
Mike laughed. “And two blue LEDs rather than one!”
And so we went down that road a bit. Far enough to get some quotes on chokes and chassis. But that was when Mike and his business partner decided they wanted different directions for Theta, and Mike was getting tired of the increasingly insane prices of high-end audio, and home theater was becoming interesting…and Mike left Theta and Angstrom became a thing, while I took a 15-year hiatus from audio and started a marketing agency.
Boom. Sometimes it ain’t meant to be.
But both Mike and I always remembered the Turd. And the Turd 2. And the Reference-Beater. And the Plans That Might Have Been.
Schiit Enters the Age of the Speaker Amplifier
And yeah, I can confirm: the idea of a crazy, all-out, choke-input amp never really left. It was in the back of my mind when I started the company, and it was definitely front of mind as I worked through the prototypes that would become Vidar.
But it wasn’t time.
Just like it wasn’t time for Mike to start with Yggdrasil as our first DAC, it wasn’t time for me to start with some wacky choke-input, all-out, reference-level design.
And that’s reflected in Vidar. Vidar was intended to be a do-all amp. Like, pretty much everything you need in an amp. A 99% amp. You can use one for pretty much everything, or two for even more power. Both configurations are affordable. We don’t have any switching supplies or Class-D compromises.
In short, it’s a really good amp. It’s our Sumo Polaris 2, updated for the 21st Century.
Aegir? That was a bit of a lark, to see if we could mitigate the transconductance droop problem without active circuitry (a la Sumo) or brute force (going 100% into Class A). It’s also a great sounding amp.
Ragnarok 2? The first Nexus amp, and the first amp that I think really expresses what I want to do. Ragnarok 2 is really my favorite. At least so far.
So why haven’t we done any of those amps with a choke input?
Easy: size, cost, and power.
Vidar and Aegir would be twice the size, and at least $300 more. Ragnarok 2 doesn’t have space, unless the choke was stuck on top like the Turd. And add another $300.
Oh, and all of those amps would put out less power if they had a choke stuck in front.
Why? Chokes have DC resistance, and also can saturate. So you shouldn’t expect the same kind of power out of a choke-input amp that you’d expect from a capacitor-bank amp. (But it may be better in the real world—chokes store a ton of energy!)
So yeah. Size, cost, and power are harsh mistresses. But that’s pretty much the entire engineering game in 3 words.
And that didn’t stop me from thinking about how we might be able to do a choke-input amp that didn’t break the bank—that wasn’t reference-priced—but also deliver great sound and good power.
And, with Ragnarok 2, I realized: we have the right chassis size for a monoblock. Monoblocks would have enough heatsinking for good power output, and the Ragnarok 2 size would have plenty of internal space for a choke. Not only that, the Nexus topology used in Ragnarok 2, combined with Continuity from Aegir, would make a heckuva amp!
The real question was: how much power could we get out of it?
Fighting Chokes…and Continuity
Sounds easy, right? We have a chassis, we have the input topology, we have the output topology, stick a choke in it and we’re done!
Well, except…this is where the real engineering begins.
I mean, yeah, some things were set.
Like the chassis. We had a chassis suitable for a monoblock. It would be very efficient for production, using the same heatsinks as all of our other amps. (And this is critical to keep cost down). It would give us plenty of heat dissipation for a single channel. Dropping a choke into the chassis wasn’t a huge task.
And the topology. Nexus is logical for this application. It gives us an inherently differential, balanced design with excellent performance. And we think it sounds very good. It certainly does in Ragnarok 2, Freya S, and Jotunheim. And it’s easy to use either single-ended or balanced inputs on Nexus—you’ll notice that it’s the only one of our amps to offer both single-ended and balanced inputs for a single channel, together with a switch.
Aaand the output stage. I mean, Continuity made sense for this application. Start with Aegir, add a bunch more devices, and you’re set, right?
And, let’s face it, the control system. It builds on our Intelligent Amplifier architecture we’ve been using since Vidar. It has the Aegir “standby” trick, but other than that, the microprocessor management and protection were largely done. And that’s a good thing, because they’ve been largely bulletproof.
So. Yeah. Easy.
Except all the details. Mainly centered around the output stage. And the choke. And all the other little stuff that stands between you and a sellable product.
Here’s some of the details that meant Tyr took 18 months and 4 prototype versions to get right:
So how did it go, specifically?
- Continuity losses. Continuity is a really neat output stage. It gives you Class-A-like performance, well beyond the Class A envelope. But it’s not exactly what you’d call an efficient output stage. In fact, it’s only about 2x better than Class A. Which is why Aegir is as big as a Vidar and has only 20% the output power…and still runs hot. As it turned out, just adding more devices didn’t buy us enough power, not by a long shot. More on this later.
- Choke losses. Chokes, as I mentioned, have inherent resistance—DC resistance—and, if not properly sized, they can saturate. And the choke sits right at the beginning of the power supply (hence, choke-input power supply), which means that you’re looking at losses throughout its range of operation. Early chokes had too high DCR, so we were looking at a very weak amp.
- The other little stuff. Things like finding the right voltage gain devices for this particular version of Nexus—the stacked supply came up and above the voltage level of most candidates…and operated at a level that would fry the standard input JFETs we used. We had to tweak both voltage gain and input stage heavily in order to get the results we wanted.
- The other big stuff. To minimize losses in the output stage itself, we decided to tweak the number of transistors we were using in the output stage, from 16 to 24. Yes, that’s twenty-four 15-amp-rated Toshiba power transistors per mono channel. Yes, that’s completely insane.
Well, the first prototype was both a huge success and a huge failure. I actually put off finishing the design and starting it up for several weeks, because the board was so physically massive. I was afraid we’d end up with a horrible, unstable, oscillating mess, one that would lose its magic smoke on the least provacation.
As it was, when I finally powered it up, oscillation wasn’t a problem. The amp was actually very stable from the get-go.
The problem was power.
In fact, the first prototype of Tyr put out less than 80W into 8 ohms.
Yes, you read that right. A giant mono power amp—that put out less power than one channel of the stereo Vidar!
The problem was simple: losses. Choke losses, plus continuity losses, meant the original Tyr was pretty weak, at least when measured by continuous power output. In listening, it’s always been a dynamic, powerful-sounding amp.
But still, 80W was not a lot of power. Tyr could be more efficient.
So we did a different choke, this one with bigger core wiring and less DCR. That increased the power, but not enough. Continuity losses were still a factor, as were a couple of surprising bottlenecks:
(Or, in non-engineerese: these are huge changes, far beyond your typical prototype.)
- We were running out of stacked rail on the Nexus stage. Increasing the voltage on that stage was necessary. Sounds easy? We were already at the limit of the voltage gain devices. Which meant a new voltage gain stage. Which meant additional opportunity for oscillation and magic smoke on the next start-up.
- We were running out of current for the drivers…due to both the driver size and the regulation of the stacked rails. This meant a complete redesign of the driver and voltage regulator stage, which increased the size of the drivers, and totally revamped the voltage regulator to a unique, discrete, complementary feedback design.
And so, since I was already going to have to do a ton of changes on the board, I decided to increase the number of output devices we were using, from 16 to 24. This would allow us to use three pairs of output+Continuity corrector pairs on each side, which would eliminate any issues with the output stage being the bottleneck.
(And, by the way, this is completely bonkers for a 200W amp. Sumo’s Andromeda 3 used only 8 devices for higher rated power.)
So what happened? Smoke, flames?
No. This time, my compensation changes were correct, and the amp started up without issue. Stable and happy. Sounding good.
But power?
Yeah. Try 150W.
Now, that’s a big step forward. Absolutely.
But with the insane amount of overbuilding we were doing, I expected more. So we tweaked the choke again, and I made some adjustments to Continuity…and we got about the same power output.
What was worse, is Continuity was falling out of correction. In the pursuit of more power, I compromised Continuity performance. Which is one of the problems with Continuity—there is no free lunch, and in this case, lunch is paid with power.
So the prototype sat for a month, while I wondered if a 150W monoblock was enough. I mean, it sounded great. It would run pretty much anything. But it didn’t have the power output I expected.
In that time, I dug deeper into Continuity. Continuity is a way to make an amplifier output stage act more like Class A, even when it came out of Class A operation. And there are many ways to accomplish this feat. Some of them used different types of output stage with less loss. Some of them embedded the transconductance trick before the output stage, for no voltage loss (but trading it for gain loss.)
I decided what I needed to do was to try a couple of different ideas. One with embedded Continuity, and one new Continuity idea that used a single corrector pair to tweak the output of 4 additional outputs.
Both had their theoretical advantages and disadvantages.
I decided that what I needed to do was to make another prototype board, one that could be built with embedded Continuity or single-corrector Continuity. Then we’d run them and see how they did.
- In the case of embedded continuity, would the lower loop gain be a problem?
- In the case of the single corrector, would the sense resistor get too hot at high output?
And that’s how Tyr 0.975 came to be. That was the last prototype before the 0.99 confirming board and the 1.00 production boards.
So how’d it go?
Embedded Continuity worked! I was actually a bit surprised that it worked at all, but not only did it work, it was fairly easily tweakable to compensate for the transconductance droop of the output stage. It still would require a bit of pot-twiddling in production, but that wasn’t a showstopper.
The showstopper was, unfortunately, the lower loop gain. That increased output impedance (not a great idea in a differential amp with already kinda-low loop gain typical of CFAs) and also increased distortion. Damping factor dived and distortion climbed to unacceptable levels.
So, unfortunately, embedded Continuity was a no-go.
And single-corrector Continuity worked! In fact, it worked better than I expected. The correction resistor didn’t get too hot, and the difference in distortion performance was easily seen and tweaked. Combined with the right standing current (about 75% of Aegir), it worked like Continuity—without as much loss.
The only thing left was the power output test—and single-corrector Continuity just squeaked over 200W per channel (about 220 in our testing) into 8 ohms, and still delivered 370W per channel into 4 ohms (rated 350), which is great for a choke-input amp.
From there, all it took was ordering the 0.99 “confirmation” boards without the embedded Continuity option, and then placing production orders for the 1.00 boards.
When I placed the order for the production boards, I breathed a big sigh of relief.
And I really was thrilled. I mean, all we needed to do was to get the boards in, and drop them into chassis, and the big crazy amp was a real deal!
Because we already had the chassis!
Yes. Believe it or not, the chassis was not the pacing item on Tyr. It was worked out long ago. Which, if you looked at it, it made sense. Tyr is nothing more than a Ragnarok with different holes in it. Even the heatsinks it uses are exactly the same as what we use on every other amp. So drawing that up, and getting it done, was fairly easy. The only real trick thing we did was add a transformer “girdle” to (a) make the inside look a bit sexier—hiding the wires—and (b) to increase airflow through the chassis. You may have noticed that Tyr is covered in holes so there’s tons of airflow through it, but simply having the transformer “girdle” sit 1/8” lower than the top allowed even more convection cooling. A small thing, yes, but everything counts.
And so, since we had working boards, and chassis I did something really, really stupid:
I told everyone we released the boards to production.
About 8 months ago.
Yeah. Sigh.
The Reality of Production, AKA Never Shoot Your Mouth Off
So what can go wrong, really? I mean, we had the chassis. We had the boards. Heck, the transformers and chokes were ordered, and even at that time (early 2021), the lead times weren’t insane on those. All we needed was someone to put the parts on the board, verify they were right, and then launch the product.
Easy, right?
And now here we are, 8 months later.
This is what happens when we have long-lead parts, our partners get slammed, and there’s a holiday season in-between.
It started with a handful of mega-weird parts. Tyr, unlike our other amps, uses pricey metal-strip resistors as its output emitters. Great parts. Very low voltage and temperature coefficients. Wonderful for performance.
Unfortunately, also very hard to get. When I released Tyr to production, they were in stock. Then they went to 8 weeks. Then 16 weeks. Then 20. So we spent 5 months sitting on one part.
“So why not redesign for something else?” someone asks.
And yeah, sure, that’s a possibility. Except we’d have to scrap the boards (nothing else is this particular size), then requalify for lower-quality parts. Literally nothing has the combination of non-inductive properties, combined with huge stability and large power dissipation capability.
So we sat. We sat because the lead time kept going out.
How do I explain this?
Have you ever had a car painted? You drop it off, they tell you “2 weeks.” You come back in 2 weeks, and they tell you, “2 weeks.” You come back in two more weeks, and they tell you “2 weeks.” You come back in two more weeks, now getting frustrated, and they tell you, “2 weeks.” And you continue like that until, maybe a year later, you have a painted car.
Parts are like that. It’s rare to have a manufacturer say, “Wow, we’re boned, this might be a year or so.” Instead, a 6-8 week part becomes a 8-12 week part, then 12-16, then 16-20, and so on and so on.
Sorry. This ain’t a perfect world. If manufacturing was easy and fun, everyone would do it.
When the resistors finally came in, that wasn’t the end of the pain. Because then we were in the middle of trying to figure out how to get enough capacity to make all of our products. Our main PCB assembler was slammed. Our new PCB assembers were slammed. And the new guys we were considering bringing on board let us know they were slammed, too. The Tyr build actually got slated at one, moved to another, then moved back to the original assembler.
And, yeah, then there was the holidays. Between running stuff for existing products to try to stay in stock, and the Omicron surge, that put things back again.
Again, this is not ideal. We know that. We’re working on it. Unfortunately, things don’t change overnight.
But…
…the good news…
…is Tyr is here! Finally!
I really hope you enjoy!
Congratulations on the Try...! And one of my favorite chapters, as I love war stories.
