2022, Chapter 7
There Is No Why
So, Folkvangr.
Let’s ask the obvious question: Why introduce a 10-tube headphone amp now, in this crazy we-don’t-know-if-there’s-gonna-be-tubes world?
Hell, why introduce such an insane object anytime, when it runs so hot and puts out so little power and isn’t recommended with all headphones and it ain’t even balanced and measures so, ah, ungawdly awfully?
Well, in short:
there is no why.
At least not if you look at it in those terms. But if you look at it as, first, an exploration into what an output transformerless and output capacitorless amplifier might work like, and then, a discovery that it sounded way more amazing than we expected, and finally, as an engineering challenge to work through to turn a rats-nest of wires into something that was reasonably safe to sell—compounded by some thermal mis-steps and tube drama—then maybe, just maybe, it might be interesting to you.
No worries if you don’t understand it. Shelve it with $1.5 million supercars, $450,000 watches, $15,000 espresso machines, and other such over-the-top tomfoolery. That’s cool.
Or, think, “10 tubes! Mini grille! Oh so glow! Gotta have it!”
Because there’s room for both opinions. We are adults, aren’t we?
Aren’t we?
Jason’s Nagging Thought
Like Mike, I get fixated on certain ideas. You know, like Mike wanting to do a choke-input power supply amp since the 1990s. That whole obsession eventually led to Tyr.
For me, one of my biggest what-ifs revolved around the question, “What does a tube really sound like?”
And yeah, I know, you all now think I’m nuts, because you know what tubes sound like. Kinda soft and bloomy and rolled-off and syrupy and romantic, right?
Well, except that lots of tube amps don’t sound that way, and so you know that tubes are really holographic and palpable and extended and magical, and deliver inner resolution like nothing else.
And except there are tube amps that don’t sound like either of the above, and you’ve heard them be analytical or dull or bass-light or ballsy or tons of other adjectives that, let’s face it, we only half-understand, because subjective impressions are subjective, and everyone’s vocabulary is a bit different.
Aside: Yes, this is a huge problem with subjective listening. Not denying that. If you want to talk measurements, read on, we’ll get to those.
But back to the question: what do tubes sound like?
Any attempt to answer this has to start with why different tube amps can sound wildly different. And when you start poking around into tube topologies, the answer becomes obvious:
most of the time, you’re not hearing the tube.
Or at least not
just the tube.
You see, most tube amps fall into one of two camps:
- Transformer-coupled. Tubes are inherently high-voltage, low-current devices. Well, at least the ones you’d be comfortable having in your house. Water-cooled 30kW transmitter tubes might be a different story. Since they are high-voltage and low-current, and have plate impedances that might be significantly higher than a headphone or speaker driver, many tube amps use an output transformer to couple the tubes to the load. There’s nothing wrong with this, and there are some companies with spectacular output transformer design chops. But the sound of the transformer will be convolved with the tube.
- Output transformerless (OTL). If you parallel up enough of the highest-current-output tubes, you can choose to eliminate the output transformer. Sounds great, right? Well, there’s no such thing as a free lunch. If you’re talking about a speaker amp, it’s gonna be a lot of, like 6AS7 tubes, with the associated power requirements of 2.5 amps of heater current per tube . Headphone amps can get away with something a bit less bonkers, like Valhalla’s 6N6P or JJ’s ECC99. And if you throw out the transformer, you end up with a relatively weak amp into low-impedance loads. Plus, most OTL amps still put something between your transducer and the tube: an output coupling capacitor. So the sound of the capacitor is now convolved with the tube.
And that’s the question that always bugged me:
What would an OTL and OCL (output capacitorless) amp sound like, with the transducer connected directly to the tubes?
And it was an honest question. As in, I didn’t know if it would be great or terrible. Chicken crap or chicken salad. Transcendant or trash. It was entirely possible that “tube sound” was reliant on the output transformer or capacitor for its magic. I could design a direct-coupled tube amp and it could sound, well, absolutely terrible.
And so, for a very long time, I left the idea alone. Oh yeah, I poked at it a bit, and even drew up a version of Valhalla that was direct coupled. But that never got built.
Aside: it’s a good thing it never came to be. This is a fairly bonkers idea and needs a lot of protection wrapped around it.
But time passed, and the idea wouldn’t go away. And our stocks of 6N1P and 6N6P tubes continued growing—far more than we could use in Valhalla. And with a ton of 6N6P tubes available, I finally, in a fit of pre-COVID boredom (literally my first notes are December 31, 2019), decided to take a stab at an OTL and OCL tube amp.
I didn’t know what I would call it. Or even if it would work.
But if it did, maybe I’d finally have my answer:
what does a tube sound like?
It Starts, As Many Things Do, With Total Disaster
So how did it go? About as expected: the first prototype wasn’t even listenable.
Why didn’t it make the grade? Lots of bad assumptions. The first bad assumption was that I would be able to get away with our standard flat-pack transformers. 4 of them, to be precise. Which is kinda bonkers in itself. That’s a ton of transformers.
But, early on, the 4 transformers kinda-sorta made sense. I figured I could put this crazy 10-tube amp in a Freya-sized chassis, with a thin form factor so the tubes could all hang out for better heat dissipation. 4 flat-pack transformers would allow for a Freya-thin chassis.
Aside: before you ask, this was never gonna be a balanced amp. Remember, the challenge with tubes is getting a reasonable amount of current out of them, while also minimizing output impedance. Both of these goals are diametrically opposite balanced operation. Balanced amps “see” half the load, so require twice as much current. At the same time, output impedance doubles as well. Both of these are bad. The trick with Folkvangr was always to have a bunch of output tubes in parallel, to get reasonable power output from a single-ended amp. And, since we are not wankers, we didn’t put balanced connectors on the amp to try to fool you into thinking it was really balanced.
One huge problem with flat-packs, though. They’re kinda weak. Which fubared my plan to pull full power out of the flat-packs to run the heaters. In retrospect, this was a bad assumption, for a couple of reasons:
- It’s best to derate from the core rating a bit. 48VA doesn’t really mean 48VA, especially in audio applications when you’re going for lower field.
- For a PCB-mounted transformer, you’re looking at long traces to run the heater AC voltage on. This isn’t such a huge deal at, say, 300mA or 600mA, but it becomes a very big deal when you’re looking at 7.2A of heater current, like in Folkvangr. So that nice 6.3V was barely 5V by the time it reached the tubes.
Sure. Fine. After that fall-on-your-face failure, I yanked the transformers and pasted in a lab power supply. Now I had decent heater voltage.
But I still didn’t have a working amp.
In fact, the DC level never got low enough for the output relay to engage. The protection systems I’d put into the prototype were keeping the amp from ever operating. Which is a good thing, because with a DC output of 5V, it wouldn’t be good for your headphones.
Aside: protection is an important subject with Folkvangr, because, let’s face it, connecting your headphones directly to tubes running on a +/-100V supply is more than a little bonkers. So, in addition to a DC servo, Folkvangr includes DC sensing and protection via relay, as well as AC sensing for fast shut-down. Still a bit bonkers. Still doesn’t keep me from using it with Grados. I would never in a million years use it with IEMs. Your risk tolerance may be higher or lower than mine. No judgement.
What I couldn’t understand, though, is why the DC servo wasn’t getting the output level down to a nice, comfortable 0V (or at least just a few mV). I mean, I’ve used DC servos forever, in everything.
I tried the usual tricks of increasing servo speed, reducing output impedance, etc. No dice. Everything stayed stubbornly at a couple of volts output.
Then, in a fit of inspiration, I thought: maybe the tubes are fighting each other. As in, the servo was trying to correct for one, but due to the inherent mismatch in tubes (even matched tubes aren’t perfect), it couldn’t correct for any of them.
So I pulled out 6 of the 8 output tubes.
Bam! DC output went immediately to zero and the amp clicked in.
Hmm.
That was great, and that also sucked. Because that meant that instead of one servo per channel, I’d need one per output tube. Not the end of the world, but complexity just went up.
Which, I realized, wasn’t such a huge deal. If this thing worked out, it wouldn’t be small, cheap, or practical. So a bit more complexity wouldn’t kill the idea.
Which is why I went ahead and did the first of the redesigns—to include a massively overdesigned single transformer (on the same size core as Vidar!) and 8 servos. Hopefully that would give me something I could listen to.
Second Chances and First Listens
The second prototype was a massive re-think. I lopped off over 1/3 of the board that had been used simply to mount the transformers, and redesigned all of what was left to include the multiple servos…as well as another idea I was playing with. More on that later.
When the big transformer came in, I laughed. It could easily be mistaken for something that belonged in a Vidar. But it plugged into the board and gave me the right volts. A few minutes with a drill and a sacrificial bottom chassis off a junk Gungnir, and I had a Frankenstein-esque open-sled prototype that turned on…and clicked in.
Still, the DC offset wasn’t exactly perfect…it was bouncing around about 30mV, which was OK-ish for many headphones, but I wouldn’t want it bouncing much more than that.
Probably due to the servos fighting each other, I thought. I brought the servo time constant down a bit, and the DC performance got a lot better. Good enough to have a listen.
I packed up the whole thing, took it home, and plugged it into the output of my Gungnir Multibit at home.
I turned it on, waited until the tubes started glowing, and heard the relay click in. Time for a listen.
For the heck of it, I started with the Grado Hemp, a headphone I know well.
And…
Holy schiit…
I mean, I literally sat there, mouth hanging open. Finally Rina came and found me because she was tired of waiting for dinner.
I tried to tell her how good it was, but she wasn’t having any of it. “That’s what you say about all your new products,” she told me.
“Not all of them,” I said.
“Most of them. Unless they really suck.”
And she did have a point. New toy syndrome can be really, really bad. It’s way worse when it’s new-creation syndrome. I’ve joked before that the amp I like best is the one I designed most recently, and, while not totally accurate, the allure of a new design has to be taken into account.
“So you should listen to it,” I told her.
“Nope.” She shook her head. “This one is all yours.”
“Why?”
“Because you clearly like it. And I can feel the heat coming off it from here.”
And she was right. I was sitting in a chair next to the amp, and I could clearly feel the radiant heat from the tubes. I looked sheepish.
“Yeah,” Rina said. “That’s stupid.”
“But it sounds good!”
She rolled her eyes. “Where’s dinner?”
And that was that. At least for a while. I went back to the amp—as yet unnamed—that night, and confirmed, yes, to me, it was doing some very special stuff. I tried a number of headphones, from LCD-2s to HD-650s, and it was kinda, well, universally good.
Which probably meant it was a bit too “tubey,” I thought.
But I decided I didn’t care. It sounded
spectacular.
Now I had to figure out how to make it a real product.
Fools Rush In (Where Smarter Engineers Run)
Making it into a real product was way easier said than done. I had no real idea of how the chassis would work, with a giant transformer that was wayyy thicker than a Freya profile. It was a whole new chassis design for us. Which would also mean new packaging.
Sounds like not a big deal? Well, it was. We don’t take on new packaging lightly, because it takes up a ton of room in the facility and increases complexity. I think I sold Alex on the idea mainly because:
- He knew I liked the way it sounded.
- He liked the idea of getting rid of tons of tubes.
- I promised him I’d come up with a chassis we could use for other products (though I still don’t quite know what those other products are).
Eventually, I came up with an idea for a tall chassis that had a center “scoop” that lowered the level of the chassis to nearer the tubes, so they could dissipate heat in the open air. (Or so I thought—more on that later, too.)
But even with the basic idea for the chassis nailed down, there were a couple of intractable issues, including a large transient when turning the product off, and hum with sensitive headphones.
Aside: how much hum? I don’t know. I just know it was audible with Grados in high gain. I didn’t have numbers because I had already made a decision not to measure the product until near the end, so that I wouldn’t be influenced by what it measured like.
The transient and hum meant another redesign—this one to include an AC sensor to shut everything down quickly on turn-off, and to improve the power supply filtering (going to a discrete complementary pair cap multiplier design, used on three rails in the amp).
Those changes solved the huge glitch on power-down and the audible hum, but they didn’t solve my heartburn with my other grand idea—that is, including an impedance multiplier as a switchable option.
“Impedance multiplier?” you ask. “What the heck is that?”
Yeah. I mentioned another one of my wacky ideas, so now’s the time to talk it through. From the start, this crazy 10-tube amp included a solid-state impedance multiplier. This is a circuit that makes any load connected to the amp seem like it’s 8x its actual impedance. So if you connect a Grado, at 32 ohms, the amp sees it as 256 ohms with the impedance multiplier switched in.
“Wait a sec!” some are saying. “That sounds like a transformer!”
It’s not a transformer. It’s electronics—in this case, it’s based on the TPA6120 headphone driver IC.
“So it’s a buffer, it’s cheating!” you cry.
Well, it’s not a buffer. Maybe it is cheating. But it is defeatable. Switch the impedance multiplier off, and it’s completely out of the picture. Not even its protection diodes remain in the circuit. You can have pure tube output whenever you want. Or…you can use the impedance multiplier, to increase Folkvangr’s ability to drive more difficult loads.
Aside: it was about this time in development that I started referring to the amp as “Folkvangr,” which is kinda like Valhalla, except you have to be personally selected by Freya to go there…so, yeah, I know, kinda loose, but it works, doesn’t it?
“Oh no, not so fast,” someone cries. “You can’t distract us with silly names. How is this impedance multiplier not just a buffer?”
Simple: it’s not a buffer, because even when switched in, the output of Folkvangr doesn’t change character. The distortion residual is exactly the same. It takes its cues from the tube output; it doesn’t add its own.
“Nah, no way, not buying it!” some tube die-hards are saying, crossing their arms. “You’re telling me you added a solid-state stage—”
An optional solid-state stage, I have to add.
“—optional solid-state whatever, and it doesn’t affect the way the tube output sounds?”
Ah no, I didn’t say that. I did say the distortion character is unchanged. It does change the sound, but perhaps not in the unpleasant ways you expect. I tried it, I agonized over keeping it or losing it, and, in the end, decided to keep it, because it makes Folkvangr a more versatile amp.
Aside: and, to be totally clear, “more versatile” does not mean, “will drive IEMs and HE-6s.” In fact, we can’t recommend this amp for IEMs at all. Its DC output does bounce around a bit…it shouldn’t be a problem, but I personally wouldn’t use it for highly sensitive IEMs. Nor is is gonna light the world on fire with power output, even with the impedance multiplier switched in. But for an OTL tube amp, it’s pretty versatile.
So yeah, chassis changes, electrical changes, additional prototypes, new metal, lots of angst. Are we done yet?
Not quite.
Not when the first prototypes hit 100 degrees C on the top chassis.
That’s not just hot. That’s, ah,
unsellable.
Up On The Mezzanine
Here’s what made it worse: the prototypes hitting 100 degrees C were prototypes using the already-finished, production “muffin tin,” which was our silly name for the complex, deep-drawn stamping that was designed to keep heat out of the chassis.
Yeah. So much for that design.
On reflection, the problem was clear: the tubes were still set too deep within the chassis, and were pumping heat directly into the “muffin tin,” which then resulted in sky-high operating temperatures. What we needed to do was to raise the tubes up.
Now, raising the tubes sounds easy, but options are actually very limited:
- Give everyone a set of socket savers with every amp. Yeah, no. Many of those parts don’t have the most confidence-inspiring build quality. Plus, it seems soooooo cheesy.
- Design and produce super-tall custom tube sockets. We actually looked into this. It turned out to be impractical for us in the end.
- Put the tubes on a higher board—a mezzanine board—to get them at the right height.
Option No. 3 is the one we went for in the end. It actually worked out very well, reducing chassis temperatures to below 50 degrees C, and also allowing us to have more direct connections to critical power supply traces, improving performance.
And, ah, performance…remember I said I wasn’t measuring Folkvangr until the end?
Yeah. I finally did. And, while it was not horrifying from the tube amp point of view, it wasn’t as good as, well, even Valhalla 2 into high impedances. Given that the amps are using the same tubes, this was a bit disappointing.
So I changed the operating point to be more like Valhalla.
Boom! Distortion better.
Then I listened to it…and a decent amount of the “magic” was gone. It sounded more like a Valhalla. A bit cool. A bit, well, un-tube-like.
So what did I do?
Simple:
I put it back to what it was.
And, just like that, the magic was back!
So does this mean that all tube sound is in distortion? Well, I’m sure some of it is. Some of it is going to be set by the operating points, which are different in Valhalla 2 and Folkvangr. But at the same time, setting a Valhalla 2 to run at the same operating point as Folkvangr doesn’t result in a mini Folkvangr, either. Which also makes sense because the power supplies are vastly different, and because Valhalla 2 still uses coupling capacitors.
The reality, as usual, is never simple or reducible to a single variable; basically the only thing I can say is that if you like the way Folkvangr sounds, that’s great…but if you prefer, say, Magni 3+, you’re gonna save yourself a lot of money!
What About the Future
Someone has to be sitting back there, and wondering what the future of Folkvangr is. I mean, it is a bonkers 10-tube amp, introduced at a time when tubes are heading into an uncertain horizon. What’s gonna happen in the future?
Simple: we’re making this a limited run product. We’re only gonna make a few of these. Like, 250 maximum. Hopefully there are 249 crazy people who also like it, because one’s going to me.
Beyond that? Well, nothing’s certain. Just like everything in life.
I hope you enjoy!
