Chapter 39:
Unto the Second Generation
When you start updating your products, brace yourself for the questions. This is a statement that is probably the same in every industry. When Apple goes from IOS7 to 8, people start wondering where OS11 is on the Mac. No. Wait. It’s 10.10 now. More proof that someone needs a swift kick in the ass there…as they build their spaceship palace.
But, to get back on track, update one thing, and everyone will want to know when the rest of the line gets refreshed.
In our case, the inquiries started as soon as we shipped Asgard 2.
“So when do Valhalla 2 and Lyr 2 show up?” “Is there a Valhalla 2 in the pipeline?” “Hey, you should add the gain switch to Lyr and call it the Lyr 2.”
Yes, we know, I wanted to say.
But of course I couldn’t. None of us could. Of course, we knew that a Valhalla 2 and Lyr 2 would be great updates for the line. And, of course, I’d already been planning some of the things we could do for the updates. But there were no prototypes of Lyr 2 or Valhalla 2 at the time Asgard 2 was launched.
And even if we had prototypes,
there was still nothing we could say. Because once you start selling products, you can’t hint they’ll be replaced. Not soon. Not someday. Not tomorrow. Not in the distant future. Not at all. Never.
Read that again.
You can’t say anything. Period.
Because the moment you do, you kill the sales of the current line. If we’d had a Lyr 2 and Valhalla 2 in the pipeline, and we’d told people about them, then we wouldn’t be able to move Lyrs and Valhallas with boxcars of Ex-Lax.
Yes, I know. This sounds selfish. This sounds unfair.
What if I just bought a Lyr and there’s now all of a sudden a Lyr 2? you might be thinking.
And yes, there’s the chance of being caught buying a soon-to-be-discontinued product right before the new one hits. It’s unfortunate, but let’s be brutally honest:
- At least the company will most likely be around in the future to service your product, rather than going out of business by having to take write-downs of unsold products they couldn’t move after blabbing that the new version was coming.
- It’s not like the new product completely invalidates the old one. If you liked the old one enough to buy it at its current price, that indicates that it provided sufficient value to earn your money.
- Most companies will offer you a discount on the old product if you purchased it right before the new one was announced, or the option to pay shipping to get the new one.
Yes, this is from the business perspective. Remember, this is a business book. And once the business is past the first scrappy start-up years, it becomes a different business. Once your business is an ongoing concern, discussing new products becomes an even bigger problem than when you were starting up.
Let’s turn that up: discussing new products before they’re ready might
kill your business.
Read that again, too: once you’re running, the best thing you can do for yourself, your employees, and your customers is to ensure that the
business keeps running.
Talking about next-gen products that could kill current sales is not a way to keep the business running. Talking about new products that could negatively impact current sales is not a way to keep the business running.
But, on the other hand,
keeping the business running doesn’t mean
growth at all costs.
This is a mistake many businesses make. They think, “We have to keep running, we have to keep growing, we have to keep expanding.” These are not equivalent thoughts. “Keep running” could easily mean steady-state. It could even mean minor to moderate shrinkage to deal with a downturn in the economy.
“Keep growing, expanding,” is very different. Growth and expansion without any restraint is the easiest way to get yourself in big, company-shattering trouble. Mike Moffat likens a business to a plate-spinning act, making sure you keep all the plates in the air. If you work really hard, and you’re very good, you can keep quite a few plates in the air and make it seem pretty effortless.
But add one more plate…
Or add a dozen more…
Boom. Everything comes down. There’s no time to keep them spinning. All you can do is to watch everything fall.
This is one reason that, even as Schiit continues its strong growth, we have natural restraints in place—self-funding strictly through sales, no hiring forward, long product development times with long private betas (I’m sitting here listening to one right now), no artificial offers/sales to boost results in the short term, etc.
Because the most important thing is to keep it running. To me, it’s not plate-spinning. It’s a marathon.
Murphy Was A Butthead. But He Was Right.
Okay, two more doses of business reality before we get to the whys and wherefores of Valhalla 2 and Lyr 2. Because I’m sure some of you are asking, “If you’re doing so well, why couldn’t you announce all the second-gen products at once? Hire some engineering talent and get ‘er done! You’d be that much closer to the revenue from the new products.”
There are two reasons you don’t plan to launch multiple updated products at once:
- Known-good, continuously-available engineering resources are limited. Beyond that, you’re using hired guns that may or may not understand what you’re trying to do, and who may or may not be available when you need them. At Schiit, the KG-CA-ER is me, Mike, and Dave. These are the three people I can count on to:
- Understand what a great audio product is
- Give a schiit—really—about the design and performance
- Work through the initial prototype bugs without turning it into a completely different product
- Be there when the inevitable production bugs surface—see #2
- Going into production is an exercise in living Murphy’s Law. It doesn’t matter how many prototypes you’ve done, how many dozens of months the product has been running flawlessly, or how small of an incremental change it is—there will be pain. Pain like:
- A $0.008 resistor is unavailable in the value and size you need—everywhere
- A tsunami wiped out the capacitor plant, and lead times are now 18 weeks
- The distributor sent you the right part—in the wrong size
- The final rev to the PCB moved the exact wrong trace, and the first articles don’t meet spec
- The metal house built the wrong metal revision
- Chassis are late
- Boards are late
- Chassis are bad
- Transformers hum
- People are sick on the line
- A new test program needs to be written
- A new test fixture needs to be built
- For some reason, they just don’t work like the prototype
- And at least 1000 other things
So, let’s imagine a multi-launch scenario with contracted engineering. Let’s say, three new products. Each has a half a dozen line problems. Your contract engineering crew is busy working on new paying jobs, so they’re not available. You haven’t been too involved in the design, so you don’t know it inside and out. The lines are stopped. Nothing is being produced. The boardhouse is hard stopped because you can’t tell them what’s wrong yet. You’re about to lose your dedicated line there. You need to make the chassis drawing changes, but you just want to get something—anything—working.
Fun times, huh?
Updating the Classic Tube Amp: Valhalla 2
Tube amps are great. That is, if you have high-impedance headphones. 300 ohm Sennheisers and 600 ohm Beyers loved the original Valhalla.
Low-impedance headphones? Not so much. Some people liked the original Valhalla with the AKG K701s, which were 64 ohms, and some liked it with Grados (yeah, I know, weird.) But, in general, we recommended it for high-impedance headphones, where it could produce very good results.
The fact was that the original Valhalla was not well-suited for low-impedance headphones. It would quickly run out of power and distort heavily into low-impedance loads.
Why? Because it’s an OTL (output transformer-less) tube amp. OTL tube amps can’t source much current, and typically have high output impedance. High output impedance into low-impedance headphones isn’t ideal. Nor is low current capability. Valhalla, in its original form, had the following specs:
Maximum voltage output: 30V p-p (into high-impedance loads)
Maximum current output: 32mA
Output impedance: 28 ohms
Not very impressive. Especially in the face of solid-state amps that can source 250-1000mA of current without even breaking a sweat, and have 1 ohm or less output impedance.
But we still had customers who really, really wanted to use a “pure tube” amp to run low-impedance headphones. So, part of the design brief for Valhalla 2 was to make it more suitable to a wider range of headphones.
But how?
There are three ways to get better current capability and lower output impedance from a tube design:
- Use an output transformer. The transformer provides a higher impedance load that the tube likes to see, and at the same time provides a lower output impedance to the headphone. The big problem with Valhalla is that this would also make it a much larger, more expensive amp. No dice.
- Use a solid-state output stage. Make it a hybrid, in other words. This was also a no-go, since we already had Lyr, and customers were specifically asking for a pure tube amp.
- Tweak the output stage as much as possible, and add feedback. This offers smaller gains than the previous two techniques, especially in the case of current output. But this was the approach we took.
Aside: feedback? Eeeeeevillll feedback? Yes. We added feedback to Valhalla 2, both in high-gain and low-gain modes. But we never said feedback was evil. We have said that we prefer not to use it when possible. But, in this case, it worked out rather well. Read on.
First, though, let’s cover the tweaks to the output stage.
Valhalla has always used a “white cathode follower” output stage with 6N6P tubes, to maximize its possible current delivery. White cathode followers allow us to double the available current into a specified load, and the 6N6Ps are some very amazing NOS tubes that dissipate 8W on the plate and tolerate very high standing currents. There’s nothing like them in new production, except the JJ ECC99, which has lower plate dissipation and a slightly different pinout.
What we did for Valhalla 2 was simple: we increased the current through the output stage and re-tuned the White cathode follower for lower impedance loads. The end result was almost twice the current capability into low-impedance loads, without affecting performance into higher impedance loads (in fact, performance increased across the board.
And then, yes, we added the eeeevil feedback. I started with a no-feedback in high-gain mode, then did comparative listening tests to come up with what I thought sounded best. Then we added even more feedback to create the low-gain mode. In both cases, feedback is minimal (about 6dB in high gain, 16dB in low gain.)
“Wait, wait, wait!” some of you are yelling. “What about re-entrant distortion? Don’t you want to use a lot of feedback, if you’re going to use it at all?”
No. Not if the basic gain stage is very linear—which is certainly the case with Valhalla 2. In fact, we made one other change to specifically enhance linearity—adding a current source to the front end differential amplifier, which required a different power supply. It also allowed us to eliminate the input coupling capacitors. And yes, we added even more eeeevill parts, in this case transistors for the current source. But this change reduced the distortion of Valhalla 2 by a factor of 5. And, with different plate loads for more overall gain, Valhalla was now a very, very linear amplifier, running below 0.02% THD open-loop into high-impedance loads. After feedback, the THD was even lower.
The results? Valhalla 2 now measured like this:
Maximum voltage output: 60V p-p (into high-impedance loads)
Maximum current output: 60mA
Output impedance: 14 ohms in high gain, 3.5 in low gain
Not exactly Charles Atlas, no, but suddenly Valhalla 2 was able to run a much wider range of headphones. High-efficiency, low-impedance headphones could easily be run in low gain mode. And, at the same time, it sounded even better into high-impedance headphones.
A home run? No. It’s still not our first choice for planars. It’s less capable than any other amp we make (including the upcoming Fulla) into low-impedance loads. But for high-impedance headphones, it’s a great-sounding amp…and it will now comfortably run your Grados.
Valhalla 2 was also notable in that Murphy pretty much bypassed its run into production. Two prototype cycles, revised metal, new boards, and a new transformer, and it was on the line. The transformers didn’t hum, the boards fit as expected, the metal wasn’t late, and everything worked as expected.
Guys, remember this: you don’t get many of these free passes. Say thank you…and then start looking behind you for the knife coming at your back.
Taming the Beast: Lyr 2
If the original Valhalla was a flyweight, the original Lyr was a musclebound brute. Big, powerful, gutsy…but unable to keep from speaking in less than a shout, and apt to break the china in polite company.
Yes, it was powerful, and yes, it was a great match for power-hungry orthodynamics…but pair it up with efficient headphones, and you could hear the noise floor. Forget IEMs entirely. Lyr delivered the goods in the power department, and offered many, many opportunities to tune the sound via tube rolling…but it was not the last word in refinement.
So, like Valhalla 2, one of the goals for Lyr 2 was making it more suitable for a broader range of headphones.
Aside: yes, a common theme. It makes sense. You should be able to use your headphone amp for the broadest range of headphones possible, rather than having individual amps tuned for a specific headphone. So it’s not surprising we set that as a goal.
But Lyr’s problems were different than Valhalla—and, as a hybrid amp, it offered us more opportunities to change.
While the simple Valhalla 2 still uses through-hole parts, it was clear from the start that Lyr 2 would have to go to surface mount.
Aside: through-hole parts are parts that, well, go through holes on the PC board. This is an older way to make PC boards. (And, believe it or not, when companies started mass-producing color TVs, we were still in the pre-PCB era—they were hand-wired. Yes. Scary.)
Now, there’s nothing inherently wrong with through-hole PC boards, except three things:
- Through-hole parts are usually bigger, so you can’t fit as much on the board.
- It’s very hard, if not impossible, to automate through-hole PCB assembly, so you’re looking at a bunch of people stuffing the parts individually into the boards. The upshot being is that they are more expensive to make.
- Through-hole parts are on the wane. Every year, parts selection is more limited, and the interesting parts come out in surface-mount only.
Because of this, surface-mount is the way we’re going with most of our new products, and Lyr 2 was not going to be an exception. The original Lyr was already packed completely full of parts. If we wanted to stuff more refinement and capability into it, the parts would have to get smaller.
The original Lyr’s noise problem had many sources. Most were directly related to the lack of space on the board. Lack of space meant we had exactly zero regulated power supplies in the original Lyr. The 200V rail for the tube, the -25V rail for the front end, the servo supplies, even the original 6.3V tube heater was AC.
All of these supplies were smoothed as much as possible, but they weren’t as good as regulated supplies. So, noise made it through to the output. For low-efficiency headphones like orthodynamics, it was inaudible. For high-efficiency headphones, it could be problematic.
With Lyr 2, we attacked the regulation problem with full force. The 200V supply got regulated. The +/-15V supplies for the servos and front ends got precision low-noise regulation as well. Even the tube heaters got separate, dual-mono low-dropout DC regulation, rather than AC. These changes alone took the original Lyr’s noise down by a couple orders of magnitude. But these changes weren’t simple—as with Valhalla 2, they required an entirely new transformer design to implement.
But we didn’t stop there. The input stage got a working on, as well, to decrease the plate load and increase the current, for a slightly more linear operating point on the 6BZ7 tubes we’re using.
Then it was on to the output stage. That’s always been the key to the Lyr design—the Dynamically Adaptive stage. It’s really the first Schiit-specific circuit I devised, so it’s really a special thing. At the same time, though, it’s also a source of questions and uncertainty. As a unique output stage, it isn’t well-known, tested or busted to death a million times, characterized a thousand different ways in hundreds of different layouts with different analyzers. The vast majority of audio engineers are terrified to venture so far out into the weeds—much safer to do something like a Blameless design, or just throw an integrated chip amp at it.
So, I had to wonder: is there a better way to do the Lyr’s Dynamically Adaptive stage? It was easy enough to tweak the overall design and optimize its operating points, but was it really better than, say, a conventional Class AB output stage? Or even a simple diamond buffer? Or something more exotic, like a triangle buffer or error-correction stage?
I spent the vast majority of dev time on Lyr 2 trying these different output stage topologies…and, in the end, coming back to the original Dynamically Adaptive stage. Every other candidate had significant downsides—not able to swing the rails, crossover distortion, etc. I went back to a Dynamically Adaptive stage, but with one significant tweak that improved performance at higher frequencies.
Then it was time for the eeevil feedback. Yes, again. Lyr 2, like Valhalla 2, uses some feedback even at high gain, and more at low gain. The reason is also the same: because it sounds better. As with Valhalla 2, Lyr 2 is an inherently very linear circuit, so low amounts of feedback don’t result in appreciable re-entrant distortion.
Add a gain switch, and there you have it: Lyr 2. A sum of many different changes, together with a multi-month excursion into alternate output stages.
Going into production wasn’t too painful, either—other than a couple of hard-to-get parts, and a last-minute revision on one of the transformers, Lyr 2 went without a hitch.
Was this a sign we were getting better at this whole engineering thing?
No. Never think that. Run from engineers who say they’re good. Start looking up at the sky so you can dodge the lightning bolts if they say they’re infallible. Lyr 2’s painless launch was just another patch of good luck.
And…it didn’t matter anyway…
The Reality: Simultaneous Launch
For all my yammering about concentrating on one thing at a time, the grand irony was that Valhalla 2 and Lyr 2 launched at exactly the same time (together with a pre-production Wyrd, Mani, and Ragnarok, at TheShow Newport 2014.)
Now, to be fair, Lyr 2 and Valhalla 2 were developed separately, so we were following our own advice on that count. But the problem was simple: we realized that if we introduced one, the other was obvious. Introduce Valhalla 2, and everyone will know a Lyr 2 is coming. It would have the same effect as going ahead and announcing it.
So, since both products were working, finished, and on the shelf, we decided to launch them together.
Note those three caveats:
working, finished, and on the shelf. In fact, we had shelves full of Valhalla 2s for some months before we actually announced. Lyr 2s took a little longer, but they were also available in quantity well before we announced.
Which might make you wonder:
what would I see on the shelf if I were to visit right now?
Are there other simultaneous launches coming?
What’s next?
Sorry, I really can’t say.
