2015 Chapter 13:
Detours in Balanced-Land, or “Improving on ‘Perfection’”
“Perfection?” some of you are already sneering. “If you’re talking about Mjolnir, that ain’t no perfect amp.”
Yep. No kidding. Nor is anything else. Hence the scare quotes on “perfection.” The fact is, though, Mjolnir was our first statement amp, our first cost-no-object design. I know this sounds crazy in these days of $15,000 headphone amps and $3,500 portables, but at the time, Mjolnir was our first shot at an amp that didn’t need to hit a price point.
“It’ll cost what it costs,” I remember telling Mike, way back in 2012.
And that’s was our mindset as we went about developing Mjolnir. Exotic transformers, sure. Premium parts, no problem. A whole new chassis design, absolutely. The only two things we didn’t pay much attention to?
- Aesthetics. Mjolnir was supposed to be a simple amp—while it wasn’t cost-constrained, it was supposed to be the simplest expression of a balanced amp we could come up with. So super-custom chassis designs weren’t really what we were focused on.
- Convenience. Also, as a simple, no-frills balanced design, we didn’t think much about convenience. Right down to making it only a balanced design—which couldn’t be unbalanced.
And balanced-only was Mjolnir’s Achilles heel. There weren’t a ton of people who wanted to commit to having only balanced headphones, no matter how high a performance level Mjolnir delivered. That really held it back. If I could go back and whisper in my ear (oh, the days a TARDIS would come in handy), I would have told myself:
“Big dummy, even if people are super hardcore into balanced headphones, what about their friends who bring their single-ended stuff by? Drop a single-ended output in there, ***!”
But even then, I might not have listened to myself. Space was at a premium in the Mjolnir design, since it was based on through-hole parts. There probably wasn’t space for summers. And even if there was, we didn’t have the trick summer design we ended up using in Ragnarok.
Between the Current and the Future
Besides the realization of how big of idiots we were for not including single-ended outputs on Mjolnir, several other things happened on the road to Mjolnir 2.
- Ragnarok happened. Once conceived as a much bigger Mjolnir that could run speakers—and could be configured as either a solid-state or tube-hybrid amp—Ragnarok morphed into a real technology statement, doing stuff that no other amp does.
- Ragnarok tube didn’t happen. And, in the process, it became very apparent that a tube hybrid version was in nevergonnahappenland.com. At least without fans. And different transformers. And maybe not even then, since I refuse to do tubes that are merely “cosmetic.” And Ragnarok’s topology doesn’t lend itself to much more than cosmetic tubeination. And Ragnarok’s topology doesn’t work without the intelligent microprocessor control we have wrapped around it (more on this later.)
- Yggy happened. Yeah. Finally.
- Yggy trickled-down happened. Even before we started shipping Yggdrasils, Mike handed me the upgrade board for what would become Gungnir Multibit. Although this happened well after Mjolnir 2 was in development (I’ve had protos running on my end table for almost a year now), it definitely provided the impetus to see how far we could take the “midrange” line.
So, in light of all of this, maybe a Mjolnir 2 was inevitable. But that didn’t mean the first prototype did much…except catch on fire.
A False Start
The first Mjolnir 2 prototype I did made a lot of sense. To me, anyway. It was nothing more than a simplified version of Ragnarok—and 100% solid state. It did away with the microprocessor management system, moved to surface-mount design, and didn’t use the crazy relay attenuators like Ragnarok. It couldn’t drive speakers, of course, so the parts were a lot smaller, as well.
And it did exactly one thing: blow itself up.
The problem with Ragnarok’s topology is that it really needs the 24/7 oversight of a microprocessor control system, constantly checking and setting bias and DC offset. Without this in place, using only a differential servo to set the bias, the first Mjolnir 2 prototype would thermally run away and self-immolate.
Yeah, I could have added a bias servo as well, but that would be even more complication. I didn’t want to end up with a design that used 200 parts to keep the 40 active ones in order. Yeah, philosophical nuttery. And perhaps I would have gotten past that eventually, but…
…but this was also about the time that it was sinking in that there would never be a tube Ragnarok. I kept trying to figure out ways to make it work within the thermal envelope of the current amp, but that kept coming out gobiteme.com.
So I had a great solid-state amp with Ragnarok (now beginning to ship), but it would never be solid state. And I had an unlistenable, unmanageable Mjolnir 2 prototype that would only work with:
- A billion more parts.
- Full microprocessor control, a la Ragnarok.
Frustrating. I put the whole mess on a shelf and tried to forget about it for a while.
Until one day, Mike asked me about it. And everything changed.
Don’t Overlook the Obvious
“Mjolnir 2?” I answered Mike, groaning inwardly. “How’s Mjolnir 2 going? Pain, fire, death.”
“That good?” Mike said, sounding amused.
“Yeah. It doesn’t work without the bias control. And the Ragnarok tube ain’t gonna happen. And we really have to think about how all this fits together.”
“Why?” Mike asked. “Does everything have to line up perfectly? Each product in its own little box?”
I sighed. No, it didn’t. But, “It has to have some kind of consistency,” I told him.
“Sounds like a way to get stuck in a rut.”
I frowned. He was right. We could internally focus-group ourselves to death—without ever doing a focus group. All we needed to do was to talk ourselves into doing the same old thing, every single time.
“There’s only one thing that makes sense,” I blurted out, completely frustrated. “A friggin Tube Mjolnir. A Tubenir. Hell with solid state, we have that solved with Ragnarok.”
“Then do it,” Mike said.
“But…” I began to protest, and trailed off. Because in the process of saying it, I realized,
This is exactly what we need to do. Not another microprocessor-controlled technological tour de force, but something simpler, something more visceral, something that would really make Mjolnir 2 stand out.
Aside: I didn’t know how much a tube hybrid balanced headphone amp would stand out at the time. As far as we can tell, Mjolnir 2 is the ONLY tube hybrid balanced amp out there, period. Of course, I could be wrong. Please enlighten me if I am.
“But…switching from tube to solid state, we’re gonna get complaints,” I told Mike, finally.
“If we do anything, we’ll get complaints,” Mike said. “Just the act of making a decision means we’ll have complaints. The only way not to have complaints is to do nothing at all.”
I nodded. I knew that.
To avoid strife, say nothing, do nothing…
But switching to tubes…while it would be undeniably cool, and while it would move Mjolnir away from every other balanced end-game headphone amp out there…what if someone wanted solid state? We always had people asking for Lyr as a solid state amp, too…
And then it hit me again. It was finally time to do the solid state tubes.
The Long History of the Schiit LISST
Solid state tubes, like I say on the product page, are not a new idea. There have been a number of companies that have tried to make them. Here’s how they go about it.
- They pick a type of tube and look at the response of the tube at different plate and grid voltages. These “tube curves” define the transfer function of the tube—a transfer function that is significantly different than most solid state devices (except a SIT, which is a story for another day.)
- Then, they try to come up with a combination of solid state devices (and passives) that mimic the curve of the tube. Sometimes this combination is very complicated—some have used literally dozens of parts. Sometimes this combination is pretty simple. For a look at one scheme, Google “trioderizer,” for a simple way to make a JFET work a lot like a tube (and also for an example of why you never let engineers name anything.)
- Finally, when they’ve gotten as close to the tube curves as possible, they announce to the world, “We have the perfect copy of the gold-grid, pinch-waisted, JimmeeJoeBob 12RU78 from 1959, come and get perfect tube sound forever!”
The reaction to #3 above is predictable, of course: tube die-hards cross their arms, squinch up their faces, and prepare to be supremely unimpressed by what the solid-state tube sounds like. Even if it sounded better, the solid-state tube company has created a perfect environment where
nobody will ever admit it.
And that usually sinks the solid-state tubes.
That’s why we’ve taken the 100% opposite approach—putting a definitively solid-state device in a can and not tweaking it to sound like a tube, and promising only, “Ya wantcher solid-state heah, we gotcher solid-state heah!”
What’s interesting is how long we’ve been playing with this idea. From literally the first days of the Lyr (early 2011), I wondered if we could replace the tubes with a solid-state device. I even did some experiments that showed, yep, you could, as long as the solid-state device could stand the tube rail voltage (about 200V.)
From there, I shelved the idea. Mainly because I wondered if something that would only work in our amps would be broad enough. I mean, Lyr, Lyr 2, and Mjolnir 2 all use current sources to bias the tubes, so plugging in a depletion mode MOSFET will bias up just fine. But there’s no guarantee it would work in every circuit.
So, one part of Mjolnir 2 was taking out the solid state tube idea, dusting it off, and making it work. Because it’s one thing to hang some parts out of a tube socket to see if it works, and a whole different thing to have something that mechanically replaces a tube.
And LISST was actually fairly complex to get working right.
First, we had the mechanical design. Stick a couple of boards in a can, sure. But how? With what connectors? In what can? How do you stick it in the can, period?
Second, we had the topology. Sure, we wanted something simple, but what if a “triodized” version sounded better? Best to do both.
Third, we had the production. Our assembly house in Simi Valley has done a bunch of complicated stuff for us, but nothing that would require assembly like this.
For mechanical design, we actually ended up doing two PC boards in a “T” configuration, with slots for alignment of pads that are soldered directly to each other (no connectors at all). The vertical part of the T holds the depletion FETs (which dissipate about a watt or so of heat), and the horizontal part of the T holds the pins that go into the tube socket. All of this is then slid into a painted steel can full of epoxy—yes, LISST are fully potted, and therefore pretty much unserviceable. If they fail, we swap them. But, if they’re used in our equipment, it’s unlikely they’ll fail. Lifespan should be in the many hundreds of thousands of hours.
For the topology, we actually built and listened to both the standard and “triodized” versions—and ended up liking the ones that didn’t have any transfer function trickery applied. Simple as that. Could we make different LISST versions? Sure. Will we? Not unless we find a way to do it better than the current model.
For production, we ended up working more closely with the assembly house than we ever have before. They were the ones who specified many of the production steps, including full potting. LISST is the first product we essential get, final and tested, from the assembly house. Kinda like an iPhone, but in a really ugly box.*
*Aside: I am sorely tempted to skewer the super-over-the-top-packaging-as-art corner we’ve gotten ourselves into by offering a future “luxury” product in both “Ugly packaging” or “Fancy packaging,” that you choose when you order. The cost of the fancy packaging would be accurately reflected in your choice. So, if this “luxury” product normally was $499 in the ugly packaging (that is, standard Schiit packaging), it might be $680 in the fancy packaging. The difference would be that we’d be giving you the choice of spending money on something you only look at once and then hide on a shelf—or not.
Aside to the aside: This future “luxury product” is entirely fictional. No points will be awarded for guessing what it might be. This has only been a thought exercise, prompted by some really over-the-top headphone packaging I’m looking at right now.
Okay, Finally, On to Mjolnir 2
Once I was past my tube phobia, Mjolnir 2 was a comparative breeze. And by “comparative,” I mean, “only three prototype board versions, after we threw away the self-immolator.”
But let’s back up a bit. Because Mjolnir 2 isn’t just “Mjolnir + Tubes.” From the start, it was intended to be much more flexible and friendly than the original Mjolnir. And, at the same time, I gave myself permission to spend some more money to improve performance even more. Most of this ended up in the board and transformers, but I’m getting ahead of myself.
Instead, let’s start with the “flexible and friendly” part. Flexibility, to us, meant at least one thing: single-ended headphone output. We actually ended up with a lot more than that, so let’s take a look at the whole picture.
- Single-ended output. In a circlotron, single-ended output means one thing: summers. Circlotrons are inherently balanced topologies that cannot be deconvolved to provide single-ended output. In fact, some circlotrons have both outputs sitting at a rail (say, 40V). Mjolnir 2 and Ragnarok aren’t like that, but this serves to show how big of a surprise you can get if you just try to use one output to ground. So, we added a version of Ragnarok’s summer to Mjolnir to provide both single-ended headphone output and single-ended preamp output.
- Gain switching. No surprise here, considering it’s in virtually all of our other amps. But good news if you’re running IEMs or high-sensitivity headphones. Although the original Mjolnir was very quiet, Mjolnir 2 is easily 20dB quieter in low gain mode. One trick thing, though: gain switching is via two relays, one placed in the exact center of each channel’s circuit.
- Input switching. Mjolnir didn’t have input switching at all—you chose one input, and grounded the single-ended input if it needed it with a rear-mounted switch. That’s it. For Mjolnir 2, we decided to add real input switching, so you can select either the balanced or single-ended input—again with a relay that switches it right at the back panel, just like Ragnarok.
- Front panel switches. And, in a controversial move, we decided to move the switches for gain and input to the front panel. A result of switchgate? Nope. The prototypes had front-panel switches while switchgate raged. We’re not categorically opposed to front-panel switches; it’s just that many of our products do better with rear switches. Take power, for example. There’s no way the Mjolnir 2 power switch would fit in the front without compromising the placement of transformers.
But I think what’s most interesting is some of the stuff that isn’t seen at all.
- 4-layer board. Look at Mjolnir 2’s PCB photo on our site, and you can’t see that it’s a 4-layer board. Yes, I know, 4-layer boards are kind of a thing right now (or even more layers), but Mjolnir 2’s board is 15.5” x 7.7”—this is a HUGE 4-layer board, especially considering the 2 oz copper. What this buys us is much more flexibility for optimal routing of power supply and output traces—which, in circlotrons, are pretty much the same thing. Going to a 4-layer board gave us the ability to create a truly optimal layout with 16x the current carrying capability of the original Mjolnir. The board also costs about 2X as much the board in the original Mjolnir.
- Crazy power supply. Note that Mjolnir 2 now has not just one, but two of the “billion pin” C-core transformers we used on Mjolnir. These complex transformers give us the capability to support the amazing range of power supplies needed in Mjolnir—2x18V, 2x6V, 4x25V, and 1x200v—without capacitive coupling, voltage doubling, or other trickery. And the billion-pin transformers, like the 4-layer board, cost over double what the standard 48VA c-core transformers cost.
- Super-trick servo. You may have noticed how proud I was to get rid of the DC servo in Ragnarok. Now, it’s not that I hate DC servos—far from it, I’d rather use a (good) DC servo than a (good) coupling cap—but the best servo is no servo at all, just like the best coupling cap. Without the microprocessor bias, I couldn’t eliminate the servo in Mjolnir 2…but I could make it act like it wasn’t there at all. Huh? Well, with a differential topology, common-mode noise disappears. So, the DC servo feeds one side of the output directly…and the other side via capacitive coupling. The result? The servo content that isn’t DC is cancelled to zero. Boom. (Almost) no servo.
The Most Boring Development in the World
Mjolnir 2 tube, or Tubenir, or whatever you want to call it, was almost comically boring in development. The first prototype started up and ran, and I’ve had versions running on my side table for almost a year.
So why so long to get this to market? Simple. Because nothing is ever simple.
Although Mjolnir 2 V0.90* started up and ran, that didn’t mean it was good to go. It had a bunch of problems. The biggest one was that the new high-voltage/SE/heater transformer was, well, rather low-voltage. We didn’t have enough volts to regulate the 200V rail. We didn’t have enough volts to regulate the 18V rail. The heaters worked fine, yeah, OK, but that doesn’t get you far.
*Aside: you can tell how confident we are in a prototype design by the numbering of
the board—V0.90 means we’re pretty confident. I’ve started stuff at 0.20. 1.0 is release. Mjolnir 2 went through 0.90, 0.95, and 0.97 before 1.0. I’ve seen us do 0.97, 0.98, 0.99, 0.995, 0.998…so sometimes we’re a little overconfident.
It was, however, enough for us to get an idea of what it would sound like. And from that day, it’s sat atop a Gungnir (first standard, then Multibit) in my main listening system.
“Wait a sec, are you saying this is better than Ragnarok?” somebody is certainly asking now. And no, I wouldn’t say that. However, for my preferences (a slightly “wetter,” tonally rich presentation), it’s ideal. Wonderful thing, personal preferences.
Aside: and before you ask, my Mjolnir 2 uses tubes—nothing super fancy, just some JAN 6ZB7s. The LISST I like just fine, and I think it's a big step up from the original Mjolnir, but the tubes are just a bit better (again, for me.) Paradoxically, I like Lyr 2 with LISST better than the real tubes (stock tubes, anyway.)
In addition to the transformer problem, my trick front-end current source was running a bit too hot…and I’d screwed up some part outlines…which meant I had a bunch of through-hole parts sticking up in the air, each leg twisted around to the right surface-mount pad.
But really, the development of Mjolnir 2 is largely a story of transformers. We went through 4 prototypes before they got it right. The next version, with higher voltages, wasn’t high enough. The version after that hit the voltages…but hummed physically. The 4
th version finally hit the right voltages and didn’t hum.
Aside: Mike wasn’t super thrilled with Mjolnir 2 until the last transformer. Getting the 200V rail in regulation really changed the character of the amp, from one that was almost a little soft, to one with great dynamics—without being hard. Now you can’t keep him away from the amp.
And, you know what? I just realized that I still need to put a chassis on the Mjolnir 2 I have—a 0.95 board on the first prototype bottom chassis.
Ah, well. Back to development. In addition to transformer stuff, Mjolnir 2 needed a completely new chassis—and it remained to be seen if our “hiding the toggle switches in a slot” would work. And yeah, the first chassis prototype showed that we had a bit of work to do. The switches didn’t line up vertically. But really not bad, for the last chassis we’ll do with 2D CAD. The tweaks were small enough to order the chassis without a second chassis prototype.
And despite everything going smoothly, we weren’t ready for either TheShow or CanJam. In this case, blame Yggdrasil. Yggdrasil and Ragnarok have changed how we make products pretty dramatically—from a single guy putting single boards in 2-piece chassis to, well, pretty much everyone working on subassemblies on a production line basis. Those complex products really were a lot to digest, but I think we’re getting a better handle on them now.
So yeah. Beyond that, how do I create drama with development that’s that easy? Short answer: you don’t. And maybe you start looking over your shoulder. I’ve said this before, but easy projects are scarce on the ground. If you have one, the next development might not be so easy.
Or maybe we really are getting good?
Nah, not possible…