2018 Chapter 11:
Class A. Ish.
When Marv suggested that I take the Continuity™ output stage from Lyr 3 and combobulate it with a Vidar to create a low-powered, Class A amp, I thought that was a great idea.
“Hell, that’s easy,” I told him. “We could be selling that in a month!”
That was October 2017.
Today is about 11 months after the 1 month I predicted. And we still aren’t selling it.
And that, in a nutshell, is the difference between science and engineering. Theoretically, a Vidar-sized Continuity amp is dead easy. In reality, though, it turned out to be quite a fight.
Lower Power, Higher Price: Schiit Engineering at its Finest?
Now, lots of people are gonna say, “Why bother doing a lower-power amp in the first place? Isn’t that kinda dumb? Don’t most people want higher power?”
Well, yes. And when you combine that with the fact that a Continuity Vidar would cost more, that seems to make it double stupid.
But here’s the thing: low-power, high-bias amps can sound
really, really good.
Yes,
better than Vidar.
Especially if you have high-efficiency speakers.
And,
especially if you could extend the benefits of this high bias beyond the high bias region.
That’s why I was so excited about doing a lower-power amp. Because then we could deploy Continuity, which allows us to sidestep the transconductance droop outside of the Class A bias region, so we could potentially deliver a whole lot more of the Class A experience, in an amp that didn’t also double as a cat-cooker.
Aside: Bob Cordell uses the term “gm-doubling” or “transconductance doubling” to describe the same thing I call “transconductance droop.” He’s referring to transconductance doubling in the Class A bias region, I’m referring to the transconductance halving in the Class B bias region (of a Class AB amp). I prefer “transconductance droop,” because I think it is more descriptive, and, because, let’s face it, nobody wants droop.
And a lower-power, Class-A-ish amp would give us another amp option in the same chassis as Vidar, using the same heat sinks, using the same packaging...this is a huge benefit, when compared to doing an amp in a totally different size and using different heat sinks. It would be fairly efficient to implement, and it would increase the number of heatsinks we were using. Win-win.
That is, of course, if it went to plan. And, if you’ve been following the Engineering chapters this year, you know that things rarely go to plan.
And with that, let me introduce Aegir.
An amp I thought we’d be selling by the end of 2017. And now it’s probably going to be a 2019 product.
Why? Let’s have a look.
Aegir the First: The Original Idea
Here’s what I envisioned when Marv suggested doing a low-power, “Class A” amp in the Vidar chassis:
- A 20-25W into 8 ohms stereo amp that would also be able to do 80-100W when run mono
- It would use the same basic topology as Vidar
- However, the addition of a Continuity output stage would result in a more linear output
- It would also use as much the same parts as Vidar as possible
Heck, if we could just re-screen the chassis, that would be ideal, because we could get it to market even faster. We wouldn’t even need first articles.
And given that idea, we could have developed and sold something very quickly. But what we would have sold wouldn’t have been ideal, and would even have been problematic if we’d come out under a Class A banner.
Plus, I already had some ideas that would push Aegir outside of the Vidar envelope. Class A runs hot. Continuity runs cooler, but not cool—think, 50% of Class A typically for best performance. It would be ideal if we could de-bias the output stage and put it in a “standby” mode. This would be pretty easy to do if we added some optocouplers to shunt the bias, and reprogrammed the microprocessor to turn it off on command. Of course, though, this meant we needed a button.
An
accessible button.
A button
on the front of the chassis.
Yes, I know, cue the gasps from the audience.
Schiit doing something that makes ergonomic sense? Holy hell, what is the world coming to?
But if you’ve ever lived with a Class A amp, you know what I mean. They get HOT. It isn’t nice to have them pumping hundreds of watts of heat into the room all the time. It would be great to be able to put them in standby.
But a button meant a different bottom chassis. And a new daughterboard for the button and a couple of LEDs. And those two changes meant that we moved, just a little bit away from Vidar.
No matter. It made sense.
So that’s what I designed: a slightly different chassis with a front button, a daughterboard, and a couple of LEDs that would work with our light pipes.
I also explored whether or not we could do something even more exotic: a choke-input power supply. I moved parts on the board and created a deep notch to see if we could fit a choke in there. If that worked, we’d need to do additional alts to the chassis, but once you’re making changes, might as well continue to do so, right?
Aside: cutting to the chase, the choke was a no-go. There’s simply not enough space. Maybe in the future, in a bigger amp.
So the first prototype was a bit more prototype-y than usual: it had hookups for the choke, a deeply notched main board, a ribbon connector for the new daughterboard, and spaces for optocouplers that, at first, I didn’t bother installing.
But yeah, that was enough to see what it could do. I installed it in an old Vidar chassis and powered it up.
The Little Surprises
Perhaps not surprisingly, the amp started up and worked (it is, largely, a Vidar, but the output stage is vastly different.) What’s more, it was happy at some seriously Class A levels of bias (we had it initially biased at 2A of standing current). It didn’t oscillate or otherwise misbehave. Everything was looking like a quick path to production.
Except...well, the bridges were getting VERY hot. And I mean VERY. Like 125 degrees C. That is not a temperature that is conducive to long-term reliability.
We had a couple of options:
- We could create heatsinks for the bridges.
- We could switch to Schottky bridges that have much lower voltage drop—and therefore much lower heat generation.
We decided to do both. I had some aluminum heatsink protos made up, and ordered some (eek) single-source Schottky bridges.
But hot bridges weren’t the end of our problems.
Heat in general was a problem across the board. Running at 1.5A, Aegir’s heatsinks were a finger-burning 70 degrees C in places. Inside, temperatures were running only about 10 degrees cooler. And this was with a transformer that wasn’t delivering quite enough voltage to meet our power specs.
The Vidar chassis simply didn’t have enough airflow through it. We’d have to add more holes to the chassis—a lot more holes. Now, the Aegir chassis were going to be significantly different. Our path to production just got longer.
Furthermore, we were finding out more about the optimum operational point for the Continuity output stage. Turns out that it worked best at lower-than-full Class-A bias, as long as you were cool with it losing a tiny bit of voltage swing at the output. So what we really wanted to do was to change the Continuity programming resistors, then bias it lower, like about 1A or so.
At 1A, things were better. But with a hotter transformer, we still had finger-melting bridges. The heatsinks didn’t quite do the trick, but the Schottkys were amazing, running almost cool.
We were making progress! I made a bunch of changes on the board and sent out for a second prototype, thinking, “We have this in the bag.”
At this point, it’s December 2017. Yeah. In the bag.
Hmmmmmmmmmmmmmmm...
Aegir wasn’t in the bag, not at all. Not the first prototype, and not the second.
The reason
? Hum.
Now, the noise floor on big speaker power amps usually isn’t anything to write home about. They’re certainly not like headphone amplifiers. They usually have quite a bit of power supply noise...noise that is largely unavoidable without heroic measures. Luckily, this noise is usually also meaningless.
“Unavoidable? Meaningless?” you ask. “What kinda Jedi mind tricks are you trying to play on us, Stoddard?”
Well, they’re unavoidable in that the PSRR of the output stage is not, well, infinite. PSRR stands for Power Supply Rejection Ratio, or, in English, “The amount of garbage on the rails that DOESN’T get passed on to the speaker.”
So, unless you have a regulated output stage rail,
somethingis gonna get through. And doing a regulated supply that can source several amps continuously (and several tens of amps peak) is, well, kinda like making another amplifier.
So if you wanted a power supply as large as an Aegir to power an Aegir, sure, that might be possible, but said power supply will also double the price. Then you aren’t looking at such a budget-friendly solution, are you? Yeah, thems the breaks.
And they’re meaningless in that you usually won’t ever hear the hum. Well, unless you literally put your ear in the woofer cone. Because a speaker amp runs speakers, which are typically 10 feet or more from the listening position. You won’t hear it in your easy chair (especially after a couple of glasses of Springbank 15), so who cares?
Except, well, a high-bias amp has a lot more current running through it. So it’s noisier.
And a low-power amp may be used with very efficient speakers. So you may hear the noise.
So yeah, now it’s not meaningless. Now it matters.
And that was the problem with Aegir. Hmmmmmmmm....
Aside: now, let me put this in perspective. Even the worst of the Aegir prototypes were not stunningly bad. I mean, literally, you had to put your ear into the cone of a typical 90dB efficient speaker to hear it. We also lent an Aegir to Marv, who used it with compression drivers and 15” woofers with much higher efficiency, and he never noticed it as being a problem. But, we are not everyone, and nor is Marv. There are people who live in very quiet houses, with very efficient speakers. It would be nice to make sure Aegir worked for them, too.
The Quest for the Perfect Grounding
Now, when an amp is noisy, you usually first turn your attention to grounding. Grounding can make or break a product in terms of noise floor. And so the first instinct I had was to look at how we were managing the grounds. I’d come up with something that worked well for Vidar, but maybe it wasn’t optimal; maybe Aegir revealed its flaws.
“According to my grounding guru, the gospel of perfect grounding is blah blah blah,” Mike Moffat told me, when I mentioned that we had a problem with Aegir hum.
Aside: Mike also heard the Aegir hum and dismissed it as “flea fart **** in an echo chamber,” but I don’t think Mike runs 103dB efficient speakers.
I’d heard Mike’s grounding advice before, but I waved it away. “Not applicable here.”
“What do you mean?” Mike asked, irritated. “It worked for my Theta amps and preamps—“
“Your Theta amps and preamps had how many rails?” I asked Mike, referring to the number of power supply voltages they employed.
“One,” Mike said. “As God intended.”
I shook my head. “Yeah, that’s nice. Aegir has 4 main rails, plus two regulated stack rails, plus three for the digital and housekeeping ****. It’s a totally different ballgame.”
“But—“ Mike began.
“And it’s all
one PC board,” I continued, cutting him off again. “So all those old point-to-point tricks won’t work.”
“But, if you have one point of connection to the chassis—“
“Which I do.”
“And if you decouple on the AC side—“
“Which I also do—“
“And you keep everything short between the bridges—“
“Yes, I do. And?”
Mike looked nonplussed. “So what’s wrong with the grounding?”
I sighed. “Hell if I know. That’s the point.”
And, from there, I resolved to Fix The Grounds. At All Costs. I reviewed Mike’s grounding guru. I went to ESP, which has one of the greatest grounding tutorials online. And I thought I found some problems. So I did a new board. And the board came in. And I built it. And I tested it.
And it was worse than the Vidar scheme.
Argh.
So maybe, just maybe, it was something with the transformer.
I tried snubbing the bridge. No change.
I tried snubbing the transformer. No change.
I tried snubbing the bridge and transformer. No change.
In desperation, I hooked up the lab power supply to the Aegir output...
...and got great results.
****. It was the power supply.
Going further, I had a couple of small C-core transformers that had about the same output voltage. I plugged those into the board...
...and got great results.
****-a-doodle-doo. It wasn’t the Quest for the Perfect Grounding. It was...
The Quest for the Perfect Transformer
Skip at MCI is one of our longest-suffering vendors. Skip is used to our (frequently bizarre) requests:
- Can you do this C-core with double shielding for minimum field
- Oh no wait, we need 10% more volts on it
- Oh crap, it doesn’t fit, can you rewind to reduce the field with single shield
- Ah hell, this has mechanical hum, how do we get rid of it
- No, that didn’t fix the mechanical hum, what else can we do?
In all cases, Skip and MCI have delivered. They have literally done all of our transformers, with the exception of wall-warts, since we started.
But when we came too him with the complaint that some of his transformers caused our amps to hum, while others did not, I figured he would think we were crazy. I took great pains to let him know what we’d already tried: the snubbing, the different bridges, the different grounding, etc. I attached output from our APx555 to show him the problem.
And:
Skip was baffled.
Yeah. Big shock.
However, he came back to us with some ideas, the main one being the use of quadfilar winding. Or, to be more specific, the avoidance of it. It seemed that our transformers had always been wound quadfilar. Maybe this was causing too close a coupling between the different rails, something that couldn’t be snubbed out.
Skip proposed this and sent us two new prototypes: one with shielding and one without.
In short: nope. No significant change.
Argh!
In desperation, again, I wondered if it was simply a resistance problem. The smaller transformers had higher secondary resistance. Was that it? I pasted some resistors in-line with the transformer to simulate the secondary resistance of the small transformers...
...and got great results.
Oh holy hell.
The Path to Production
Let me skip the next 3 transformer prototypes, because, while necessary, they aren’t very exciting. The good news is the discovery of the secondary resistance being key is what led to Skip being able to deliver a transformer that had significantly better performance...good enough that we had a product.
Is Aegir dead-silent? No. But it’s certainly very good for a Class A amp. Or...I should be more specific, for a Continuity™ amp.
Because you won’t be seeing any more Class A from us. Like, ever.
“Oh my gawd, I love Class A!” you scream. “How can you do this? This isn’t fair!”
Okay, okay. I hear you. But let’s talk about that later. The title of this section is The Path to Production, so let’s talk more about that. Yes, I know, I’m a terrible tease. If this is unbearable, just skip forward to the next section: The End of Class A as We Know It?
Now, with a transformer quiet enough for Aegir, we were left with minor detail changes. Stuff like:
- Re-doing the grounding one more time (to bring it back more like Vidar)
- Changing the way Aegir measures DC offset to improve the microprocessor management response time
- Adding our “OBD2” diagnostic code so that the Aegir is compatible with the Vidar code reader Dave cooked up*
- Reducing the value of the emitter and sense resistors in order to keep heat generation down
- Tweaking the gain of the amplifier a bit
- Finalizing the compensation, which ended up slightly different than Vidar
- Making sure the new chassis design fit
- Doing thermal testing
- Doing overload testing
- Doing long-term diagnostic outputs, including DC, standing current, and operating temperature
*This is fun. I need to have Tony do a video on this. Dave, because he is lazy (and smart) didn’t want to help me look over some confusing Vidar “pig” boards. “Pigs” are boards that have failed in non-obvious ways, and are difficult to repair. So, instead of fixing the pigs, Dave made new firmware for the Vidar that reports the status of all the operating parameters and checks all the sensors to see if they are operational, to identify what’s wrong with it. He also created a board with an LCD screen that plugs into the programming port on a Vidar to show visually what was wrong. So now, diagnosing Vidar is as simple as plugging in the diagnostic board and reading the display. Yeah. This self-diagnostic capability will be going into all the new amps, of course.
And now, we’re pretty much ready for production. If the fourth-gen prototype boards don’t have any surprises, and the production boards don’t have a problem, and if there are no surprises with the transformers, we’re set!
Aside: we did have a failure of an Aegir at the show, which was caused by bad solder on the board. It wasn’t a flaming failure; the amp just went into protection and stayed there. This isn’t surprising, since one of the Aegirs was a first-generation prototype, and one was a second-generation prototype; the third-gen prototype with questionable grounding wasn’t worth bringing, and the fourth-gen prototypes weren’t built yet.
Yeah, I know. A lot of “ifs” in that previous paragraph, right? And now you know why introduction dates sometimes slip.
The End of Class A As We Know It?
Picking up on the thread of “No more Class A from us ever,” let’s talk about that a bit more. In short, yes, we’re serious. No more Class A. Which may mean we’re completely insane. Time will tell.
But, no matter how you slice it, Class A is profoundly limiting, and it’s fairly meaningless as a marketing buzzword.
Well, let me elaborate a bit:
- Class A in preamps and low-power circuits? Well yeah, duh, of course. No competent engineer would design a Class AB preamp or Class AB DAC output buffer. Durrrrrr.
- Class A in power amps? No. Class A has significant drawbacks that Continuity sidesteps. We’re going full in on Continuity.
“Hey, I’ve been told before that (insert magical technology here) is better than Class A, so isn’t this just one of those tricks?”
In short: no.
In long: Continuity has advantages and disadvantages. So does Class A.
Let’s summarize:
Continuity Pros:
- Continuity extends the benefits of Class A (linear transconductance) far past the Class A bias region.This is a big, big deal. Refer to John Broskie and Bob Cordell on much more detail than I have time to provide. In simpler terms, the gain of an output stage drops when it moves from Class A to Class B operation, making the transfer function look a little, well, droopy, when compared to a perfect straight line. It’s also the opposite of a tube’s square law output characteristic.
- Continuity ensures that you have both NPN and PNP devices conducting on both rails at all time, sidestepping the N and P mismatch problem.Because “matched” NPN and PNP or N-channel and P-channel devices are not actually “matched,” because physics. So every time the signal crosses zero in an amp that uses only NPNs on the positive rail and PNPs on the negative rail, it’s running into a slight mismatch.
- Continuity allows us to run more efficiently than Class A.It’s not a panacea, it doesn’t run cold, but half the standing current of a Class A design is a pretty decent reduction. Especially when you consider that Continuity allows us to extend Class A benefits BEYOND the the point where the stage drops out of Class A. In theory, an Aegir should sound pretty much the same, whether it is operating within its 10 watts of Class A, or putting 100W into a 4 ohm speaker in mono.
Continuity Cons:
- Continuity is more complex than a normal output stage.In addition to both NPN and PNP devices on both rails (important note: neither running gain, this is not a Sziklai scheme), it also needs sense resistors, which have to be carefully calculated, and capable of operating at the total power output of the amp—so, in other words, lots of BIG resistors.
- Continuity results in lower overall power output for a given rail voltage.Since we need to sense what the output stage is doing, the sense resistors drop the total available swing. Continuity will put out less power than a Class A or Class AB design using the same voltage rails.
- Continuity still runs hot.The lower the standing current, the higher the sense resistors have to be. A practical Continuity amp will never run as cool as a Class AB amp. Period. If you tried, you’d end up with an amp with very little power output, because your sense resistors would have to be too high.
To us, the pros outweigh the cons, so we’re going all-in on Continuity. I don’t think you’ll ever see another pure Class A power amp design from us, either in the speaker realm or as a headphone amp.
“Oh hell, you guys are idiots! Class A is great! Why are you throwing it away for some stupid buzzword that probably doesn’t actually work, anyway?”
Well, to answer the last part of your question first, Continuity
doeswork.
Doubt it? Refer back to John Broskie’s articles. And, also be sure to check out Bob Cordell’s new amp design book, which I understand has a full chapter on the problem of transconductance doubling (or, as I said before, transconductance
droop, because the problem isn’t the doubling in the crossover region so much—you can get around that by running your push-pull design in Class A—but the halving of the transconductance OUTSIDE the crossover region, when it’s running in Class B.)
To back up and answer the contention that Class A is great and that we should stick with it, let me be blunt:
no.
- Class A (in an output stage) is profoundly limiting.This is why Asgard 2 can’t really ever go over 1 watt, because we have to offload the heat somewhere. Again, because physics. So while Asgard has stayed stagnant at 1W, Magni’s power output has grown by 200% since its inception. Similarly, Valhalla 2 has vanishingly small power output into 32 ohms, because we can only run so much Class A bias through tubes (and this despite the fact that we’re using an output-boosting White Cathode Follower output stage to make it into a push-pull design, rather than a pure single-ended design.)
- Class A doesn’t solve the N/P mismatch problem.No matter how much current you run through a Class A design, no matter how large its operational envelope is, if you’re talking about a standard push-pull output stage on complementary rails using NPNs on one side and PNPs on the other, it will never quite match. Some N and P devices are pretty good complements, but none of them are PERFECT complements. Don’t believe me? Take a look at some data sheets for N and P devices. Compare the curves. Do they look the same? Yep, there you go. So Class A, even Class A that’s biased up to, say, 5A of standing current, will still have a discontinuity problem, albeit more slight than the transconductance doubling problem of a Class AB amp.
- Class A is expensive to implement.No matter how simple and elegant the design is, you need to throw a ton of thermal management at it to make sure it doesn’t melt down. Thermal management means big heatsinks (if you want the amp to be totally quiet) or fan-forced convection (with attendant fan noise) or exotic schemes like fluid cooling.
Aside: none of this means that we’re going to be dropping Asgard 2 or Valhalla 2 from the line anytime soon. As long as they continue to sell, we’re fine with making them. They are mature designs that work well, and I’ve said about ten billion times before that it’s really hard to make any meaningful improvements on them, without going outside of what they are.
Aside to the aside: this goes about 10x for Valhalla 2. Tubes will never be high-current devices. To get more power out of that design, we’d be looking at either an entirely different tube complement, or going to a transformer-coupled design. In either case, you’re talking about a vastly different, much more expensive amp. I’m not convinced that’s a great idea. See Gjallarhorn, an idea we abandoned.
Now, imagine that we went to Continuity for Asgard 3. With Continuity, we’d have an amp that could do 4W into 32 ohms and provide much better performance overall, no matter the metric.
Furthermore, it could take the same cards as Lyr 3 and Jotunheim.
Furthermore, with the production efficiency increases that we’re now realizing, it might be a $299 product—with the AK4490 DAC card included.
Aaaannndd furthermore, it would then be a compelling step-up from a Magni 3/Modi 3 combo, or from any Magni/Modi combo in existence.
Aaaaaaaaannnnnnd furthermore, that might be a very interesting option to the tens of thousands of Magni owners out there, who may now be viewing Asgard 2 as a sidestep or step-down, and Jotunheim as too expensive—which is a lot bigger pool than those who are insistent on Class A amplifiers.
“Crass commercialism!” Some will cry. “Sound is what matters!”
Yep, gotcha. And in that case, Continuity still delivers.
Yes, over the Class A Asgard 2. Hell, a hypothetical $199 4W Continuity Asgard 3 still is gonna run 500mW Class A bias. Not many headphones are gonna get out of that range. But for the headphones that do, now we have 4W of damn-near-Class-A performance that also sidesteps the N/P channel mismatch problem, that we could never get from Asgard 2.
“Well, yeah, I don’t believe you,” you say.
And you probably haven’t heard a Continuity amp. If you get a chance, check out a Lyr 3. Or, when they’re available, Aegir.
Aside: with luck, maybe January for Aegir? Not sure. We have metal, so that’s not a limiting factor. I’m building the final prototype boards, and it doesn’t take long to order for production. The transformer seems largely solved, and I should be able to get an order in this week.
Aside to the aside: when I said, “January” at the show, multiple people guessed February, March, May, October, and November. They’re getting to know us too well. Any of those guesses may be right.
Final aside: I’ve also heard Continuity in a whole lot of prototypes, and everywhere we use it, it sounds very, very good. When compared to Class AB, it’s more resolving and dynamic (at least to me), but this isn’t surprising when you consider that it’s eliminating the transconductance droop outside of the Class A bias region.
No, I lied, this is the real final aside: what’s even more interesting for the future is the possibility of simulating square law devices in the output stage, which can take quite a bit more complexity than the basic Continuity scheme outlined here—but suffice to say, if we can correct for droop, we can also alter the slope of the transconductance upward. Maybe even in multiple segments, to even more properly simulate a square law device (like a tube).
“Well, I’m still doubtful,” you grump.
Again, yeah, I hear you.
But I know what I hear, and I know what other expert listeners think, and this is how it’s gonna go: Continuity is our future.
