Chapter 9:
Powering Up: Lyr
Okay. So it’s time for a new amp. But before we get into this, let’s talk about product roadmaps. Yeah, more boring business stuff. But this kind of stuff is important—that is, if you’re interested in building a few amps on a hobby basis.
What’s A Product Roadmap?
In brief, it’s a plan for what products you’ll have, when you’ll introduce them, and when you’ll obsolete them. Yes. Products have lifecyles, and you need to plan for how long you expect them to be in the market. Now, this doesn’t have to be some elaborate stack of Gantt charts or backed by tens of thousands of dollars in market research. But the reality is, you need to at least have an idea of:
What products you intend to sell. If you did two headphone amps and then, say, decided to make a deep-fryer, this may not be the best strategy. If you did two headphone amps and then decided to extend the line with another amp or a DAC that works with them, this makes sense.
Where they fit in the line. Are the new products upmarket? Downmarket? Why would someone buy the new product? What need does it fill that the others don’t. Note: “I wanted to try this crazy new topology” isn’t a recipe for logical product line. And you do want to have a product line that makes sense—one where you don’t have products that overlap each other and cause confusion. Having a 1W Class A amp, and a 1.1W Class AB amp, and a 0.9W Class S amp probably doesn’t make a lot of sense.
How many years they’ll be around before you refresh them. Planning on selling something “until it doesn’t,” isn’t a recipe for success. How long do you think your products will be competitive? In mass consumer industries, you see major updates every year. In some cases, this makes sense, since the market is changing so rapidly (smartphones, tablets). In others, it makes little sense because there’s no giant change in the market (receivers, dishwashers.) In niche audio, every year is too fast. Every 8 years is a little too slow.
So, coming back from RMAF, we knew we had to have a new product—one set up from the start for the high power needs of orthodynamics. And we knew where it fit in the line: above Valhalla. We didn’t know how long it would be around, because, let’s face it, we were only about 6 months old as a company. Nothing was obsolete yet, nor would be for a long time.
The question was: what the heck should we do?
Lyr Challenges
Deciding to do a new product isn’t really worth much. You also need to have a set of target goals for it. For Lyr, our early notes were as follows:
- At least 4W RMS into 32 ohms power output
- Tube hybrid design with rollability—this was in response to early Valhallas being non-rollable, which many customers saw as a negative
- Retain as many of the key features of our other designs as possible: no overall feedback, non-Class AB output stage, etc
- A more direct signal path than either Valhalla or Asgard, which were capacitor-coupled designs
- Same chassis size as the other products
- Sexier cosmetics, without costing a fortune
The first point (4W into 32 ohms minimum power), coupled with the “same size chassis” spec was the biggest sticking point. Asgard was about at the limit of heat generation into that size of chassis, running 40-45 degrees C using the case as a heatsink. And Asgard was only 1W output. Multiplying output by 4X would result in a small hot-plate or grille—not something that could be safely used.
Of course, Asgard was Class-A, which means it runs full out all the time. When you run a real Class-A amp hard, it actually runs cooler.
An aside: “Class A” is easily the most abused term in all of audiodom. “Class A” is used variously to describe:
- Real Class A amps like Asgard and Asgard 2, which run full bias all the time and cannot come out of Class A mode, ever
- High-Bias Class AB amps that run part of the time in Class A, but transition out of Class A for higher output
- Low-Bias Class AB amps that have some kind of “sliding bias” arrangement to try to keep the bias in Class A all the time
- Other sliding bias amps with different kinds of output stages
- Preamp Class AB stages that never come out of Class A operating ranges, but are technically Class AB
- Op-amps that are internally Class B, but have the output stage biased into Class A
- Op-amps that are internally Class B, but do not have the output stage biased into Class A
- Anything else that might possibly have some aspect of Class A operation in any one part of the circuit
- Anything they think they can get away with calling Class A
Here’s a hint on how to spot real Class A amps: they are big, hot and heavy. Period. Anything else, and “Class A” is probably just a slogan.
Another note on Class A: It is in vogue with some audiophiles today to dismiss “Class A” as an inefficient affectation of doddering old designers who might be touched in the head, and therefore incapable of comprehending the amazing efficiency and performance of today’s Class D designs. That’s cool. But there’s nothing wrong with using “Class A” to accurately describe a real Class A circuit, no more than there’s anything wrong with using “Class H with switched output rails for higher power output at greater efficiency” to describe, well, a Class H amp.
Now, Where The Hell Was I?
Oh yes. Lyr. 4x the output power and not enough heatsinking for Class A operation. Plus the heat of tubes. The logical answer would be to change over to a Class AB output stage and set the quiescent current at a level where the amp wouldn’t become a George Foreman grille.
Of course, being a little (a) stubborn, (b) slow, (c) petulant, (d) affected by Not Invented Here syndrome—choose one or all of the above—we decided
not to go with the logical answer.
Instead, I wanted to create something new. Something that kept most of the characteristics of single-ended Class A operation, but seamlessly transitioned to Class-A push-pull, then finally into Class AB, as power needs increased. And I didn’t want it just to be a high-bias Class AB output stage.
Why? Several reasons:
- Class AB output stages, by nature, use complementary transistors (BJTs, MOSFETs, or, in some cases, more exotic devices)
- Complementary means two different kinds of devices: NPN and PNP, or N-channel and P-channel
- Complementary devices are never truly complementary—or, in other words, the “inverse” equivalent isn’t just an inverted version of the other
- Since they are not truly complementary, they introduce nonlinearities as you switch from one to another in a Class AB amp—and, even though these nonlinearities can be dealt with via feedback or error correction, they are an inherent part of a complementary output stage
This is why you’ll see us employing non-complementary output stages where possible. In Asgard 2, we use only N-channel MOSFETs, one as a current source. In Mjolnir, we use Circlotron-style topology to use only N-channels as well. Same with Ragnarok. Of course, this doesn’t work all the time, so Magni has a conventional Class AB output stage, with complementary devices.
So, what I wanted with Lyr was an all-N-channel output stage that would be able to “slide” out of Class A when necessary, to deliver additional power.
That’s easier said than done. I investigated various sliding-bias systems and dual-mode amplifiers, building and measuring about 20 different breadboard prototypes. Most of them worked to some extent, but all of them had some significant limitation—they couldn’t make it out of Class A, or distortion was too high, or they required 4 different trimpot tweaks per channel to make them work.
The problem was that we were working “off the roadmap.” Class AB amplifiers are well-understood. Class A amplifiers, ditto. Something in-between doesn’t have a lot of references in literature. Especially when you’re talking about such low power output. And, to make it more complicated, some of the best sliding-bias arrangements are tied up in Nelson Pass’ patents, which of course we can’t infringe on.
The Lyr project dragged on until late December with no listenable prototype. It was getting to the point where I was considering just throwing in the towel and using a Class AB output stage, because nothing was working well enough.
The Critical Stage
Sometimes the place to look for inspiration is in implementations from earlier eras in audio (sometimes not—there are genuinely better ways to do things now.) Because, in the old days, NPN and PNP components weren’t just mismatched—they were sometimes not even in the same zip code. A lot of early work avoided complementary output stages entirely. Some of these found neat ways to improve the power output of a Class A circuit. The problem was that none of them really worked the way we wanted them to.
This is where we went back to the closest match, and started tweaking. Part of it was actual S-domain control analysis, and part of it was building and testing additional prototypes, to see how close we could get to the MOSFET control characteristics we needed.
(If this was a movie, insert montage of boring engineering work with fancy camera angles, fast cuts, and a driving, heroic soundtrack.)
(Re above: ha. Engineering is a lot of heads-down work. There’s not a lot of heroics or drama. You know, like everything in real life.)
And, late one cold night in the garage, I finally had it—a stage that would do 43.2V P-P into a 32 ohm load. Sharp-eyed readers will do the P= V^2/R calcs and say, “Hey, that ain’t 4W, that’s 7.4W RMS!
Right. But then when you have two channels driven, the output falls due to power supply sag, so 6W is a nice round number.
And that’s how Lyr ended up being 6W, and not 4W—it overperformed.
On Power Ratings. Okay, please let me vent about this one thing.
Power. Ratings. Are. Done. In. RMS. Per. Channel. You do not (a) rate at peak power, (b) add power output together for both channels, (c) use an artificially low load impedance to make the output look higher.
Or, well, you shouldn’t, but some do. Maybe I should re-rate our stuff at 16 ohms—then you can brag to your friends about your 2.5W Magni or 10W Lyr. Bottom line: it’s still the same amp.
The Path to Production
Back in those early days, the path to production was pretty streamlined. As soon as I had a fully working PC board, we drew up some renderings of what Lyr would look like and announced it.
Yeah. Before the metal was in-house. Before we had boards in-house. Before we actually made a single Lyr.
And yeah, we’re idiots.
But, in this case, it actually worked out. We sent out press releases on December 27, and the renderings were sexy enough to get picked up by Wired, Engadget, and Gizmodo. We promised delivery by March 1, and actually started shipping in late February. It was the only product we ever pre-announced that met its delivery date—and the first product that sold out the first run before we started shipping.
And, it was the last product that had a run made by hand. By the time we’d started shipping, it was becoming completely clear that we couldn’t do this by ourselves anymore.