Separate names with a comma.
Will it be [REDACTED]...?
Thank Christ we're reaching the end of that saga.
Sent from my iPod touch using Tapatalk Pro
2016 Chapter 15:
The Vidar Chronicles, Part I
Why am I writing about Vidar now if it isn’t finished, you ask?
Well, in part because it isn’t finished. I’m sitting here, waiting for some information on heatsink availability and pricing. Which I need before I know how much space I’m working with in the chassis. Which determines the size of the board (or boards). So, while I wait for more data, I’m in limbo.
I already did this once, when we threw out the fan-forced horizontal heat tunnel approach. Now, I’m doing this one more time, as we throw out the passive vertical heat tunnel approach.
Wait, what? Why are we starting over again? Are we incompetent? Crazy? Stupid? Obsessive? Scared?
Well, maybe a bit of all of the above. To find out why, let’s pull back and discuss…
Why Power Amps Are Different
Power amps (as in, the kind that drive speakers, not headphones) are different than virtually every other audio component out there.
Let’s start with the obvious: power. Speaker power amps are way, way, way, way beefier than headphone amps. Delivering a watt or so into 32 ohms is in no way, shape or form like delivering a few hundred watts into 4 ohms—especially when that 4 ohms is really 2.1 ohms at 80 Hz and has significant reactivity and back EMF on it from two giant woofers gyrating like grandma’s jello casserole. Speaker power amps are expected to do this without breaking a sweat. This means that speaker amp output stages are way, way, way, way overbuilt compared to a headphone amp.
But there’s more: protection. Speaker amps are also expected to not self-immolate when you accidentally clank the two heavy-duty palladium-plated spade connectors together as you are hooking up the speaker while playing Megadeth at full volume. This means they need to have some kind of over-current protection in them, which is (a) unusual for a headphone amp, and (b) unobtrusive enough that you can still deliver enough current to throw serious sparks.
And more: heat. If we’re not talking Class D (and we are not), speaker amps have to dissipate quite a lot of heat. Unlike headphone amps, which probably cruise in Class A most of the time, a Class AB speaker amp will spend a lot more time on the Class B side. This means that it will have variable heat output, which will ramp way, way up when you’re running, say, Magneplanars (4 ohms and inefficient) rather than Zus (12 ohms and efficient.) This is why speaker amps usually have significant heatsinking. More on this later.
And even more: regulation. So many manufacturers engaged in so much asshattery in the 1970s that the Federal Trade Commission stepped in and actually issued rules regarding power output claims for audio power amplifiers (at least ones over a handful of watts). These rules included rating with both channels driven, continuous RMS power, and the ability to run the amp for 1 hour at 1/3 rated power. That “1/3 power” test is brutal, as it is run at the highest dissipation point for a Class AB amplifier. The 1/3 power test has been supplanted by a 1/8 power for 1 hour + 5 minutes at full power test by the FTC, but Stereophile still uses the 1/3 power test. More on this later.
And finally: it’s a make or break component. Good power amps make companies—see the glory days of the past, and look at the history of companies like GAS, Hafler, and Adcom. Those companies built their reputation on power amps. Modern companies have built their reputation on power amps as well, most notably (on the affordable, Class AB side) Emotiva. But even as good power amps make companies, bad power amps kill companies. This is what happened with GAS (Great American Sound). They made powerful amps. Unfortunately, many of them also blew themselves up. The service load from these amp failures eventually sank the company.
So yes, when it comes to speaker power amps, a degree of paranoia is warranted.
And that’s probably why you see so many amps based on standardized Class D modules these days. The hard engineering work is done. The protection system is built in. Throw a couple in a box with a power supply and you have an amp, with minimal chance of grandiose failures.
And, it’s also probably why you usually see the same Lin-type voltage-feedback topologies on the Class AB side of things. Probably 95% of the Class AB power amps on the market today are this classic topology. Again, using this topology minimizes risk. It’s well-known. It’s been tested and busted literally thousands of times. The strengths and weaknesses are well-understood. Biasing, protection, compensation, output stages, etc…they’ve been honed and refined over several decades.
At this point, you’re probably thinking, Any sane company aiming to make a low-cost power amp would probably do Class D with standard modules or Lin, right?
Right. But maybe we aren’t sane.
Balancing Sanity and Insanity
Some of you are out there, groaning into your coffees, thinking, Ah hell, Schiit’s gonna do it again. They’re gonna overcomplicate something like an inexpensive power amp, and make it a crazy, late, nontestable device I want nowhere near my stack of gear.
No. Not quite.
If we were completely crazy, we would have made Vidar a circlotron design with the Ragnarok control system and called it done. Of course, that wouldn’t have hit our cost targets, nor allowed the flexibility we wanted, nor, well, a whole bunch of things.
So if we weren’t completely insane, what exactly were we shooting for with Vidar? Again, let’s pull back and take a broader look, starting with the market.
If you look at the power amp options out there, you’re confronted with a dizzying array of choices. Slick boxes with great cosmetics from respected names. Small outfits doing interesting stuff with lots of customer accolades. Components with eye-watering price tags. And gear that’s a whole lot cheaper.
However, when you start saying, “Yeah, but I want a Class AB amp, and I want it under a grand,” the options list suddenly gets a whole lot shorter. Add the requirement for a linear power supply, and the list gets chopped again.
And if you add, “Oh yeah, and it’d be nice if it was made in the USA,” well, you get crickets.
That realization drove our first must-haves for Vidar:
Powerful and inexpensive. As in, at least a 100W stereo amp for under $1000. Way under $1000. This is a significant decision, because power amps typically have some very expensive components in them, including the power transformer and heatsinks. So cost becomes a primary driver of what you do.
Old school. As in, Class AB with linear power supply. This is also significant, because this means you have BIG heatsinks and a BIG transformer. Not cheap.
Made here. Which means it has to be simple to build. It can’t be a wiring nightmare or many-board monstrosity.
At the same time, I knew that the Ragnarok approach (circlotron and first-generation intelligent control system) wouldn’t be the best solution here. Circlotron transformers are expensive due to the number of taps and quadfilar winding. The first-gen intelligent control system might be a bit much for an inexpensive amp.
And yet, I still wanted it to be something special. So, I added the following must-haves:
No-compromise design. As in, no coupling caps, no servos, no IC gain stages, no shortcuts, no corners cut—a true, high-end design that anyone would be proud to use in any system, regardless of cost.
Versatile architecture. A 100W stereo amp is a good starting—and ending—point for many systems, especially if it delivers a solid 200W into 4 ohms as well. Add the ability to use balanced input, and it also becomes a 400W monoblock. Vidar could be two amps in one.
Interesting topology. And here’s where I went off the rails a bit. But I wasn’t completely insane—this is what would have gotten thrown out if it didn’t work. I had a complete, voltage-feedback Lin-topology schematic for Vidar in case what I had in mind didn’t work out. But I’d gotten intrigued with current-feedback topologies during the Jotunheim development, and I wanted to see how it would perform in single-ended form. More on this later.
With these must-haves in mind, I set out on the first sanity checks with Vidar: getting costs for the transformer and heatsinks.
The transformer had me sweating. If it cost as much as the transformer in Ragnarok, we were sunk already. If we had to go to China to get it, we’d be breaking one of our own internal rules. Luckily, with some negotiation of size, mounting style, and connector types, the transformer came in at a price that made the amp feasible…while still being made in the USA.
But then we came to the heatsinks.
Heatsinks can also be very, very expensive. Especially if they are cosmetically finished and proudly positioned where everyone can see them. So I thought that going with a conventional heatsink design was out of the question, and didn’t even bother exploring that route.
Note to self: never dismiss something out of hand.
Vidar The First: Fan Follies
Early in 2016, I had identified what I thought would be a good heatsink strategy for Vidar: a long, horizontal heat tunnel formed from two high-density heatsinks, backed by a variable-speed fan under microprocessor control.
On first glance, it made sense. Horizontal heat tunnels with fan-forced cooling have been done in tons of audio gear. It was a proven, well-known strategy. Furthermore, the small, lightweight, high-density heatsinks were inexpensive. Even better, you could hide the whole mess inside the chassis, so it didn’t have to be cosmetic.
And, heck, I already had a microprocessor in there, so doing real-time temperature sensing and running the fan would be no big deal. Hell, most of the time it could probably be off, or just ticking over slowly. You’d never even hear it.
All in all, it seemed like a great way to go. Even though I knew I’d hear some groans and moans about having a fan in the amp, I knew that the complaints would fall away when people found out that it would be dead-quiet. Or nearly, anyway.
Furthermore, I had a slick gain stage drawn up—a single-ended, fully-complementary interpretation of Jotunheim’s current-feedback topology. I’d done some smaller prototypes of it, so I knew it was fast, precise, and very low-distortion. I figured I’d graft it on to a linearized MOSFET output stage (so I could run very low bias, keeping the fan off even more of the time) and call it a day.
The chassis was a different matter. It took quite a bit of going back and forth to arrive at something that seemed logical: a half-width product that lined up with Saga in terms of front aspect, but could be used side-by-side as monoblocks on a typical shelf as well. I used that general idea to draw up concepts and get dimensions. However, I didn’t bother getting prototypes done, because I figured we should get the board working before we went any farther.
Hey, at least I made one good call.
The heatsinks put us behind, though, since they were done from an unusual die. We paid for finished parts…and waited…and waited…and waited.
But eventually, they came in. I did the usual building-a-prototype kind of things, stuffing the board, doing an initial power-up, doing some static measurements, etc.
And during this time, I started to get a sinking feeling about the heat tunnel approach.
Why? Lots of little things:
How do the heatsinks attach to the board? I hadn’t really thought of that.
How does the fan attach to the heatsinks? Again, oversight. Eventually I designed a complex metal girdle that solved both of the above problems, but it would add cost I didn’t expect.
How would the fan interface with the chassis in order to minimize noise and ensure the air went where it should? Again, another custom part would ne necessary.
Would the whole assembly survive shipping, or would it need additional structural support under the board?
Would the fan be as flipping loud as it seemed like it was going to be, or did we need to embalm it in sorbothane?
But it was Dave’s problem for a while, as I sent it off to have firmware done for the oversight, management, and protection of the amp. This version of the amp used something very akin to the Ragnarok control system, with full active bias oversight. That proved to be both a good and bad thing.
When Dave brought the amp back, he wasn’t smiling.
“I blew it up,” he said.
“And?” With Dave, this is to be expected.
“And the input devices are really small for their dissipation…” he said.
“I’ll fix that.”
“…and they blow up if you short the amp.”
“Huh.” That wasn’t good. That wasn’t expected. Not at all.
“And…” Dave turned on the amp. It sounded like an AM radio tuned in-between stations with the volume up full. “The fan is loud,” he shouted.
“Can’t you turn it down?”
“It needs to run pretty fast to pull air through the tunnel.”
Crap. I sat back and took a look at the amp again. Dave was right. The high-density heatsinks needed a veritable Hoover to suck air through the long tunnel.
“And it has noise from the fan PWM on the ground,” Dave said.
I shook my head. Yeah, we could go to a different extrusion. Yeah, we could get a quieter fan. Yeah, we could make all the bizarre little parts we needed to make a heat tunnel really work. All the crap about why fans suck came rushing back to me.
And I decided, right there, it wouldn’t go any further.
And that’s why Vidar the First never even got fully operational. We killed it before it played a single note.
Second note to self: you should have listened to it. Or tried to.
Vidar The Second: Heat Tunnel Hell
One of the reasons I was so eager to kill Vidar the First was simple: I’d already been thinking about an alternate approach to the heatsinking. That thought process went something like this:
Instead of using a fan, maybe there was a high-density heatsink that could be used to create a vertical heat tunnel. A vertical heat tunnel should pull air through it via convection, doing the work of a fan…silently.
Yep. Neat idea.
And, not just a neat idea—there was an affordable, compact, high-density extrusion that offered exactly the same surface area as the old Sumo 120W amp heatsinks. Coupled with an aluminum chassis, we’d be golden. Not only that, the heatsink was in stock. I had some pieces cut to size and shipped to me in a few days. Suddenly, it looked like we were on the fast track to getting an amp that worked.
I laid out a new version of the amp—this one designed to have a big hole in the middle, like a donut. This time, I designed in larger input devices, increased some resistor sizes to eliminate the short-it-and-it-blows-up problem, and made one very big decision:
This amp wouldn’t have 100% bias oversight, like Ragnarok.
This was a big decision, because it meant that the amp would need to have its initial bias set manually, and would allow for some bias variation during operation. However, the microprocessor would still allow for the elimination of the DC servo, and would provide oversight of temperature and protection. Aaanndd… the amp would be testable under the standard 1/3 power regime used at Stereophile.
Soon we were looking at another assembled prototype. This one went much more smoothly than before. Even the heatsinks fit! I ran it through the usual DC tests and sent it off to Dave for firmware.
But again, Dave came back not looking happy. “It oscillates,” he said.
“Bad. Like many amps of current bad.”
Ah. That was very bad. “When?”
“As soon as you turn up the bias.”
Hmm. I asked Dave to try a few oscillation-killing tricks (bypassing, compensation, etc), but he brought it back and plopped it on my desk, saying it was still a no-go.
Fine, I figured, I’d dig into it.
So, that weekend, I stripped the gain stage down to its bare, uncompensated, open-loop form, and measured it. (It’s important to see how the stage operates before the feedback loop is closed, in order to get the compensation right.) There were no huge surprises, so I closed everything up and slowly turned up the power.
Again, everything was just fine. So, I slowly began increasing the bias. And that’s where everything went wacky. As soon as the outputs had any bias on them, they went into violent, device-melting oscillation. I couldn’t get the amp anywhere near the target bias.
So, I spent some time re-doing what I’d told Dave to experiment with—compensation, bypassing, etc. None of that worked, so I went to more radical solutions—measuring the inductance of the output resistors, swapping them over to film versions (completely non-inductive) for temporary testing, increasing the gate stopper size on the MOSFETs, adding additional compensation poles, moving the compensation around, compensating the output stage itself, bypassing the linearization, eliminating the linearization entirely.
So I went even crazier. I pulled the gain stage back to what I knew worked fine—a basic stage that ended at the drivers. It worked fine. I pulled off all the MOSFETs (3 pairs) and went to a single pair of MOSFETs. That worked fine, too.
But when I added a second pair of MOSFET outputs, BOOM…back to the output-frying oscillation.
So, cue another montage of going back and forth with bypassing, compensation, hacking up the board to eliminate routing variables, etc, until the whole gain stage looked like I’d dumped a pile of random parts on it as the solder was cooling.
And none of that worked. Not with more than one pair of MOSFETs, anyway.
Now, I was over a week into hours-every-day work on this thing. And it was incredibly frustrating, because I’d used the same kind of MOSFET output devices in every Sumo amp I ever designed. The Sumo amps never had these kinds of problems. So why was this different? I tried a different batch of MOSFETs, I tried tighter matching, I tried even crazier ways of damping oscillation…and in all cases, the results were the same: two pairs of MOSFETs (or more) and boom.
Why did it act like this? No idea. Maybe the MOSFET design itself had changed. It was a similar part number, but it wasn’t the same…and it was from a different manufacturer…and manufacturers remix their parts from time to time. Or, maybe the combination of an ultra-fast current-feedback stage (it has a bandwidth of several megahertz before an input filter is applied) and MOSFET outputs really wasn’t fated to work.
So I decided to do something even more radical: replace the MOSFETs with bipolar transistors (BJTs.)
This is something I’d contemplated when first designing the amp, because bipolars could potentially reduce complexity. Bipolars wouldn’t need the transconductance linearization, for example. But they would also require some new, beefy drivers—which would have to be on the heatsink, which I had limited space for.
So I drilled a few more holes on the heatsink, added big drivers, threw together a bias network, and installed three pairs of Toshiba BJTs.
And the amp just fired up and worked.
Like, completely stable, no problems, like it was standing there with crossed arms, looking at me like I was an idiot and asking, why didn’t you do this before?
Sometimes life is weird. I wasn’t going to question this development—except for the heatsinking problem, going to BJT simplified the amp—and probably improved its measured performance at the same time. Fine. Win-win. We had a show coming up anyway (RMAF), and I really wanted to show the amp there.
So, I re-laid-out the board for bipolars, for the big drivers, designed up a heat spreader to attach the drivers and VAS stage to the heatsinks, and got a couple of chassis put together by our supplier.* The idea was, that if these worked well enough, they’d go to the show.
Aside: our chassis guys deserve kudos for this. The need for the heatspreader was last-minute, and they delivered. They also delivered an alternate design in a couple of days when I found out the part needed to be bigger to dissipate more heat. They did a quick-turn anodize on the heatsinks that improved their heat dissipation. They did finished-looking chassis for the show, on time. This was a wonderful showing by our metal house.
To make a long story short, the amps worked. Yeah, I had to tweak the compensation a bit, and Dave had to do some new firmware, but a week before RMAF, everything looked good. Except…
“Aren’t we going to torture test these?” Dave asked.
“Absolutely not,” I said.
“What if they blow up?”
“They won’t. We’ll torture-test them after the show.” The spectacle of blowing up the amps right before the show—and having nothing to use—wasn’t appealing. And, in reality, the amps would be loafing at the show. Typically, running amps into speaker loads is much easier than the “run it to full power and short it” test, the 1/3 power test, or anything else we do to stress an amp.
Foolhardy? Not really. If you’ve been following my bleatings here, you know the amps made it through the show just fine.
They’ve also survived an impressive array of torture-tests after the show, including:
Running to thermal shutdown to test the thermal measurement.
Running to clipping within and outside the audio bandwidth to check for nasty stuff like simultaneous conduction.
Shorting at clipping and below to test the protection system.
Running for hours into Magneplanars at high volumes.
Being handed off to employees and friends to see if they survive normal handling (accidental shorts, etc).
Validation of performance into reactive loads at high levels.
1/8 power and 1/3 power long-term testing.
And that’s where things go off the rails a bit—with the 1/8 power and 1/3 power testing. As of this writing, the amps will make it through the 1/8 power for 1 hour plus 5 minutes at full output that the FTC mandates. But they make it only about 15 minutes into the 1/3 power testing before the thermal protection shuts them down.
If you’re sitting there saying, “Well, I don’t see the problem there, you pass FTC, right?”
Well, yeah. Barely. In a 22 degree C room.
And Stereophile still tests at 1/3 power.
So, for a while we joked about adding a “panic fan” to the design that would only come on with exceptional thermal loads, such as 1/3 power testing. And, you know, that might work. But the fact is, the Vidar prototype is running much warmer than expected, especially when compared against a Polaris 2 with the same amount of heatsinking.
In retrospect, this may seem obvious. A passive heat tunnel needs to have an extreme thermal gradient to be effective. Plus, the two heatsinks are facing each other—they are radiantly heating each other. Heatsinks on the sides of chassis radiate into an effectively infinite space. Big advantage. Plus, the high-density design of the heatsink actually means there is limited thermal gradient to work with. So, the design isn’t taking advantage of heatsinking as well as it could.
So what do we do? We look at other options.
Vidar The Third: The Charm?
These “other options” mean “new heatsink design that will probably end up being structural and external.” Which means we may even end up saving some money, since the chassis becomes simpler. But that’s all yet to be seen. I’m waiting for some quotes right now, both on price and availability, that will determine our future direction.
“Oh no, so we’re gonna see Vidar in 2018, that’s what you’re saying, right?”
Not at all. In fact, I don’t see any reason to revise our timeline of shipping in Q1 2017. Even with having to go to a new heatsink design. Here’s why:
The hard parts of determining topology, etc are done. We have something, we know what it is, we know how it performs, and it’s both a logical extension of our Pivot Point topology and the Ragnarok control system.
We have an amp that’s electrically fantastic. It overperforms on power output, delivers very low distortion, is dead-stable, and has a great “Generation 2” control system that eliminates the stuff you don’t want in an amp (coupling caps and DC servos) but doesn’t de-bias on 1/3 power testing.
It’s been torture-tested. Severely. The fact that it gets VERY hot and still doesn’t fail is a big plus, because it won’t get that hot in final form.
All we need to do is get rid of the heat.
And that, I’ve done before. In going back to external heatsinks, this is something that I’m completely familiar with. It’s also an amazing boon, because it allows us more area to spread out the parts, reduce heat concentration, and eliminate heat spreaders. The amp gets simpler…again.
So now, I’ll close this part one—and look forward to seeing where the heatsink options come in…
To be continued.
I incorporated Chapter 15 in the ebook, see my signature.
Thanks Jason! Another great chapter. These insights into the birth of a Schitt product are fascinating and maybe even educational. Can't wait for part 2. Keep 'em coming.
Carver had a pro-sound amp that had something like a cooling tunnel.
I had one of these. The fan would have been on the right in this photo blowing over/through the heatsink and I guess they wanted the air to radiate up out of the top as it blew along. You can't see it in the photos but the heatsink was screwed down through the board into the bottom plate of the chassis into some stand-offs via screws located along the central valley of the fins.
The fan was LOUD. I had read accounts of rigs where a rack of these would "sing" along with the music being played as the fans ramped up and down lol.
I'll add the specs in case anyone was curious: 200/300W per ch into 8ohm and 4ohm, 700W bridged. Of course there are Crowns that put out a lot more but none of these would pass muster as "HiFi" most likely.
You're right about the Crowns. Their reliance on 741 op-amps for their voltage gain stages was anything but "HiFi". Phase Linear, OTOH, did produce a quality product with 350 W/channel. Ahh, the good old days...
See, this chapter explains why I'm going to wait patiently for Freya, and then buy it. And why I'm going to wait patiently for Vidar, then buy two. And why I'm going to wait (very) patiently for the Manhattan Project, and then buy it, even if I don't know what it is or if I need it.
Because I do know that every piece of Schiit has been conceived, designed, re-designed and sweated over by people who not only know what they're doing, but care about what they're doing. Which gives me confidence in the products.
And in the mean time, I can shop for less-efficient speakers...
I was under the impression that the Manhattan Project was the Rag/Ygg combo. This just keeps on getting better.
Thanks for the read, Jason. I really enjoy these product development chapters, and the little bit of insight they provide to the process of bringing a product from concept through "mass" production. They have, in fact, provided the final nudge in my decision to purchase a growing number of your products. Not many products out there have a diary of the design philosophy and development process published for the consumer.
Now, how about a small desktop speaker amp in a Jotunheim / Asgard footprint (taller by necessity) chassis? Something for those of us with small passive speakers on our desks that we can use in line with your existing headphone amps. Even a double-width (~18") chassis might work well if it wasn't too deep, as one could stack their Lyr 2 and Bifrost on top of it side-by-side. Solid state, ~15-30W, no remote or even volume knob necessary. Just a set of inputs and speaker terminals. The Vidar looks great, but its just too much amp for a desk.
No idea what the market demand, feasibility, or cost of a product like this would be, but you're the experts. That said, consider it my entry into the "can't always get what you want (but sometimes you do)" bucket. If there's nothing like this forthcoming from Schiit, it might be time I started looking into powered monitors. If there's a possibility there however, I certainly don't mind getting by with a Marantz 2215 awkwardly stuffed on top of the filing cabinet for a while longer.
Natural air convective heat transfer is much much slower than most people realize, especially when there isn't a clear "path" for the air to flow. It leads to some interesting conundrums like why adding vent holes on top of a metal chassis might actually decrease heat dissipation.
I wouldn't stack anything on top of an amp, even one dissipating just 25W. The heat builds up fast and can shorten the life of all the gear. But if you do, get some very thick pucks or whatnot to put a couple of free inches above the amp.
Do MOSFETs have faster current response times than Bipolar Junction Transistors? Could that be the explanation for the oscillating if current feedback is a bit slow? I am not not an EE, just an interested party trying to stretch his brain.
In another note Jason, I must compliment you on your testicular fortitude, taking a non torture tested amp to an audio show!
I've made somewhat the same request, but I wasn't willing to wait. I'm glad I didn't, because I'm now listening to a great desktop speaker system powered (6 WPC) by a set of Class A monoblocks designed by Nelson Pass that I built up out of kits from diyaudio. Modi Multibit>Magni 2 Uber>Amp Camp Monoblocks>Tang Band w5-2143 in open baffles with a subwoofer. About the same sonic signature as a K702, but a truly magnificent soundstage! Those monoblocks would be a lot of money if they were assembled and in a retail box, probably at least three times the $385 I spent for the kits. They are powered by separate switching supplies (Mr. Pass: "Get over it, they work fine.") so add another several hundred bucks for a pair of linear supplies, and we're into kilobucks for some desktop speaker amps. If Schiit could beat those prices and be as musical as my current solution I would jump in a moment.