2020, Chapter 11:
Not Because It’s Hard, But Because It’s Easy
Some of you no doubt recognize the inversion of JFK’s famous quote in this chapter’s title. But for those who aren’t familiar with it, here you go:
“We choose to go to the Moon in this decade and do the other things, not because they are easy, but because they are hard, because that goal will serve to organize and measure the best of our energies and skills, because that challenge is one that we are willing to accept, one we are unwilling to postpone, and one which we intend to win, and the others, too.”
Pretty inspiring stuff, huh?
And yeah, there’s something about tackling a near-impossible project and making it real. There’s something to staring at a challenge that most would say, “Nah, that ain’t gonna happen,” and taking it on, even though you know there will be setbacks that make you want to throw the whole thing in the bin, late nights that don’t go anywhere, and lots of slow, iterative slogging.
(And the NASA team ended up doing it, though it was verrrrrry near the end of the decade when it happened.)
So why did I invert the quote for this chapter?
Not because it’s hard, but because it’s easy?
Because I realized that perfectly describes Magnius.
Yes, Magnius.
Yes, the new uber-measurement balanced amp we just introduced.
Yes, the new amp that absolutely blows up the price/performance ratio in balanced amps.
“Wait a sec,” you might say. “Coming up with something like that was
easy?”
In short, yes.
Yes, another heretical statement. So let me make it even more pointed:
developing our best-measuring, highest-value amp was suuuuper easy.
And that’s at least one reason why I waited so long to do it.
Cheeky Alternate Intro
I almost started this chapter with something like this:
Magnius introduces our all-new ZZZ Technology™, with our exclusive Super Nested Opamp Recombinant Electronics (SNORE™) topology. Because composite amplifiers have been around a long time and they aren’t super exciting.
But, you know, that’s not cool. We come up with our own silly names, and our compatriots in this industry can poke fun at ours just as well.
Aside: but we don’t do the alphanumeric thing, so, well, at least we have that.
Aside to the aside: yeah, but we do come up with our own buzzword bingo, so glass houses and all that.
Aside to the aside to the aside: so consider the above stuff redacted, never to be mentioned again.
Semi-Serious Discussion About “Measurement” Amps
Okay, so why did I think getting great measurements from an amp was not particularly interesting?
It’s like I said: getting state-of-the-art measurements from a composite amplifier is super easy.
I mean, this is a path that has been well-trodden. There are dozens of implementations documented from Diyaudio.com to many manufacturers, past and present. You can fancy it up with buzzwords, but the reality is, if you throw a ton of feedback at something, it’s gonna measure well.
And, when you start playing around with measurement amp implementations, you quickly end up with designs that bottom out the APx555 analyzer. Heck, Magni Heresy, which doesn’t even use a composite topology and costs only $99, is fairly near the state of the art.
So, once you get measurements like Heresy—or, even 6dB better, like Magnius, or a couple of dB one way or the other—where do you go? I mean, the measurement problem is effectively solved. You’re operating 1000x lower than what is considered the threshold of perception. Or more.
So what do you do when you reach that level?
Sit around and sing kum-bay-yah?
See who can implement the best measurements for least cost, and pronounce each a successive winner, leading to a race to the bottom?
Or do you move on to more radical projects…crazier ideas…stuff that won’t necessarily measure at the limit of the analyzer?
Now, there’s nothing wrong with shooting for state-of-the-art measurements. It’s just that we’re getting past the limit of the analyzer, and we will eventually reach the limit of physics itself.
Hyperbole? No. Great measurements boil down to three things:
- Low distortion. Use a ton of feedback, get low distortion. This isn’t too hard, especially if you’re using a composite amplifier arrangement, where you can use each amp to its best potential—and control everything with a very high-performance device.
- Low power supply noise. This means no lumps and bumps from your 60Hz AC linear supply, or no high-frequency noise from a switcher. Lumps and bumps are pretty easy to get rid of—simple regulators and good grounding are tons for an op-amp circuit with large PSRR (power supply rejection ratio.) Physics starts rearing its head in non-zero ground impedance and in non-ideal rectifier behavior.
- Low thermal noise. This is where the physics thing really starts coming in. When you’ve squashed distortion and power supply noise, you can start getting to the limit of the equivalent amplifier impedance. Higher impedance means higher thermal noise. There’s no getting away from that. That’s why phono pre-amps designed for huge gain also use tiny input, network, and feedback resistors—to keep noise down. The lower the feedback resistor, though, the more power needed to drive it (heat) and the higher probability that driving such a heavy load will contribute to circuit nonlinearity.
Aside: fun fact: the best-measuring amp would likely be a fully discrete, current-feedback design with extreme attention paid to linearizing the open-loop operation, coupled with insanely low feedback resistors to drive the equivalent resistance down into the weeds, perhaps with an error-correction scheme on the output. All of which would mean a very complex, hot-running, fairly impractical amp. Think, 50-80W dissipation for a headphone amp, in a box the size of a Mjolnir. Maybe 500-600 parts. I’m not convinced there’s a market for such a beast, especially since you’d have to come up with new test equipment or test procedures to verify its capabilities.
Aside to the aside: or run it in a dewar of liquid nitrogen. Yeah, that’d be popular. And safe. So no.
Magni Heresy proved to me that a measurement-focused amp could sound good, and helped me understand the appeal of such products. But after Magni Heresy, I figured we were done. I mean, it’s soooo good for sooooo cheap, why bother with anything else?
Except…the nagging thought of “what can we do with a balanced implementation?” kept creeping into my consciousness…
…and Mike had Modius…
…and there were other balanced amps out there, and people seemed to like them, and back-of-the-envelope calculations made it seem like we could eviscerate them on price…
…and I began to wonder what we could do.
And that’s how we started on what would become Magnius.
Making R&D Exciting
So if this is so easy, does that mean that development was easy?
In short, kinda. In longer, well, we did our best to make it more difficult. It wasn’t just “make a prototype, say it’s good, and make them.”
Oh no.
Let’s start with the name. The original name of the project was “Jotunheim Slim.” Because, let’s face it, Magnius is very close to a Jotunheim in most objective and functional respects (and better in measurements.) And it’s thinner. Hence, “Jotunheim Slim.”
Aside: Yeah, sometimes we’re simple. Sorry about that.
But, as I got deeper into design and layout, I got less and less comfortable with the name. Because it wasn’t really a Jotunheim. Jotunheim is an all-discrete differential amp with an internal power transformer and linear supply, and it’s modular so you could use it as an all-in-one solution. The proposed amp, while differential, was not discrete, used a wall-wart, and wouldn’t take any optional modules.
At the time, we were developing Modius Maximus, so I simply pinched the concept. The first PC board prototype was labeled “Magnius.”
Aside: Which is much more appropriate. Magnius is more Magni Heresy writ large than Jotunheim diminished. Jotunheim will continue as Jotunheim—the only all-discrete, affordable balanced modular headamp and preamp out there.
I also wrestled a bit with feature set and implementation.
In terms of feature set, many balanced measurement-focused amps don’t have preamp outputs, so I thought of leaving them off. But that’s a profoundly cheap thing to do, even if you’re figuring on using top-line Neutrik connectors, like we always do. Neutriks are not that expensive.
In terms of implementation, I wondered what the best way to design a great measurement-focused differential amp. The logical way to do it was to continue down the Magni Heresy path. Add a differential op-amp (the nice-sounding LME49724 that we use in Bifrost 2 and Modius) and double the output stage full of OPA1688s from Heresy, and done. Well, except you have to add a buffer to the LME49724 to isolate it from the potentiometer, but hey, the OPA1688 is a great part for that, and it even has that magic fffffffeeedfoooooorrrward that is the buzzword of the day. We could use the standard Magni wall-wart and linear supply and see what we could get out of it.
Only one thing bothered me: I figured we’d have to use a small, 9mm 4-gang potentiometer in order to make the product hit an aggressive price point.
Not a huge deal, I told myself.
All the other measurement-focused balanced amps use 9mm potentiometers. And the custom Alps part we use in Magni was pretty darn good. All we’d have to do is get a 4-gang version of it.
Still, to get the proto going fast, I started with an off-the-shelf 9mm 4-gang audio-taper part, because that was close to the Alps. Surely it would be good enough.
Aside: why the histrionics over a potentiometer—a volume control? Sounds silly, doesn’t it? That is, until you realize that the potentiometer is arguably the most important part of a headphone amp. It’s the control you use the most. If it feels crappy, the amp feels like junk. If it doesn’t have a good audio taper, the volume will be mismatched and jumpy. A good pot is very, very important, which is why we only use custom Alps 9mm potentiometers with a deeper audio taper than normal in all of our small products, and why we really, really like to use Alps 27mm potentiometers as soon as we have the budget and room to do so. Because it feels better, and because, even if it is a part with technically the same “audio taper” as the 9mm part, the volume control is much more gradual, controllable, and matched. With pots, bigger really is better.
Aside to the aside: all this talk about “tapers” translates to how fast the volume rises when you turn the knob. A linear taper potentiometer is not suitable for audio—it will blast you out of the room as soon as you start to move the knob. Audio tapers are logarithmic tapers, and they come in different flavors, from shallow (30A and 20A) to medium (15A) to deep (10A). Perceptually, 15A and 10A tapers allow for a much more gradual, controllable volume pot. But again, you can get 15A in 9mm and 15A in 27mm…and they will feel different, in terms of how fast the volume rises as you turn the knob. There are only so many things you can do with a small pot.
So yeah, even with all that, I figured a 9mm potentiometer was a must. Like I said, it wasn’t like we’d be using anything lower-spec than everything else out there. And if we did a custom Alps part, it probably wouldn’t be too bad.
One thing that worked out very nice, though, is we had good Alps switches from the start. These are 4-gang versions of the switches we use in Sys. They feel very good, and I don’t think we’ve ever had a single failure with them in Sys.
Hum in Paradise
The first prototype worked and produced prodigious power—about 5W at 32 ohms, both channels driven. It was also stable and didn’t do anything unexpected. As I expected, it wasn’t super hard to get it running.
Aside: after the flaming protos we’ve had with discrete and tube prototypes, this aspect is very welcome.
But…
…it hummed. Not so bad through the balanced outputs, but really bad through single-ended. This isn’t a surprising result in designs with limited power-supply rejection (think discrete, tube, etc), but it was surprising in an op-amp based design with huge PSRR and a regulated supply.
…and the volume pot was crap. I mean, yeah, not total crap, but not what we are used to. It was 10dB worse than the Magni pot in terms of matching, and the taper was jumpy—it almost didn’t feel like an audio pot.
…and it had a schiitload of parts on the board. Scaling up the Magni Heresy design meant 38 op-amps total, with 32 of them being used in the output stage alone. That’s a lot of parts, especially when you start counting the balancing resistors necessary to have the 32 output buffers work together happily.
All in all, it was a bit of a letdown. I pushed it to the back of the shelf and forgot about it for a while—long enough to order a good custom 9mm quad pot prototype from Alps.
When the Alps prototypes came in, I took another look at the design. The hum turned out to be bad grounding. I’d gotten used to 4-layer boards, so the 2-layer board of Magnius meant I had to pay more attention to ground routing. With some plane cuts, I got the hum out of both the balanced and SE side.
I dropped the new potentiometer prototypes in, expecting brilliance.
The result was…not much better than the off-the-shelf part.
Sigh.
Fine, fine. I went back to Alps, told them what kind of specs we wanted to hit, and got a new prototype of a full custom part. For good measure, I also went to Alpha, another potentiometer maker who seems to have a solid product line.
Fast-forward again. Two new pot protos. Two more sets of measurements and listening.
And…still crap.
I mean, yeah, maybe I expected too much. Maybe I’ve gotten too used to 27mm potentiometers and resistor ladders. But it was still super disappointing.
I almost tabled the whole project there.
But then, a terrible, wonderful idea started brewing…
The Most Magnificent Potentiometer
I mean, I knew exactly what I needed to do to fix the pot problem…just use the custom 4-gang 27mm Alps Blue Velvet pot from Jotunheim. Done.
Except…they were shockingly expensive. As in, about 15x more expensive than the 9mm pots we use in Magni. Yes, 1500%. Yes, it’s not a cheap part.
But if we increased the number we used, maybe the price would come down to something workable, I thought. I contacted Alps again, told them the 9mm pot wouldn’t work, and asked them for a quote on a silly number of 27mm 4-gang potentiometers.
Aside: the 27mm 4-gang part we use is fully custom. It’s so custom that the option for a 4-gang 27mm Alps doesn’t exist in their parts list…at all. It’s what we’ve used in Mjolnir and Jotunheim from the beginning. Heck, Jotunheim was, at least in part, an exercise in how to use up a large order of these pots.
And that’s when fate hit.
Alex mentioned, “Well, we have like 8,000 of those pots anyway.”
I blinked. “The Alps RK27114s?”
“Yes,” Alex said, looking nervous. “I mean, I placed an order, you placed an order, our parts guys placed an order, so we have a lot extra…”
“No that’s fine,” I told him. “That’s actually good news.”
“I mean, I know you hate to carry extra inventory.”
“Yeah, I do, but why don’t we just use them for Magnius?”
Alex looked stunned. “You’re gonna put that expensive part in that amp?”
I nodded. “Thinking about it.” Plus, I liked the irony—Jotunheim helped us use a surplus of Mjolnir pots, Magnius would help us use an even bigger surplus.
But one thing nagged me:
What happens when we run out of pots? Would we still be able to make Magnius at a reasonable price?
Then, more good news: the quote from Alps came in at a nice price. We’d be able—barely—to meet the production cost on Magnius, even if we used up all of our existing stock. I grinned. Magnius was going to get an Alps RK27114…one of the most magnificent potentiometers available today.
That was completely, utterly silly…and nobody would be able to match it!
Internal Thunderdome Time
Still, one aspect of the design still nagged: the eleventy-billion op-amps. I mean, sure, yeah, it works in Magni Heresy because there are only half as many parts. But in Magnius, it was like handing 480 firefighters a garden hose, when all you need is one dood with a big-ass hose.
Why use so many op-amps in the first place? Simple, to get the output current we need to drive high power into low impedances.
Most op-amps are relatively low-current devices. Some are rated for much more, like the ones we use in Magni Heresy. But even in Heresy, we’re using 8 of them (OPA1688s, which can do 75mA each) per channel. It would be better if we had a single op-amp that could do 500mA, 750mA, or even 1000mA all by itself. It would simplify the design considerably.
And there are a few op-amps out there that can do that. Problem is, they are usually (a) noisy, (b) low voltage, (c) expensive, or (d) all of the above. But it was worth a look.
What came up was…surprising. The lead contender turned out to be a really old part (like, circa 2004) that was popular for a time, then it faded away: the TPA6120A2. This is a part that TI has billed as a “headphone amplifier.” Capable of 750mA current output, and with actually very nice distortion and noise specs, it seemed to be the ideal part.
But…I remembered some people said it didn’t sound very good.
And…it was also a current-feedback part with 1300V/uS slew rate, so it also had the reputation of being squirrely to use. (May also be why it has a reputation for not sounding good, I thought.)
So, hmm.
Maybe best to prototype that part, and do another with the old Heresy parts, and see how they compare? I thought. Because the Heresy-esque version at least sounded good. I’d taken it home and it was a pleasant, clean amplifier.
Or you can go bonkers and paste in a discrete output stage, I also thought. That seemed very interesting, maybe the best of both worlds.
So we made three different prototypes:
- One with the 32 Magni Heresy op-amps, same as it ever was.
- One with two lonely TPA6120A2s, which looked really, really simple and very silly. Look at the current board. You see what I mean.
- One with a discrete output stage pasted on to the op-amp front end. This is the one I really, really wanted to work.
The “Magni Heresy version” worked fine, as you’d expect. We’d already prototyped it. With the addition of the Alps RK27114 potentiometer, it was a great product. It sounded very good. It just used a lot of parts.
The TPA6120A2 version also worked fine, with some loop gain measurements and added Zobel network to stabilize it for any load. It measured even better than the Magni Heresy version. It was super-clean and shockingly glitch-free—much better than I expected.
The discrete output stage version was never thermally stable, even though I spent a reasonable amount of time trying to make it work. Measurement-wise, too, it wasn’t in the same league as the other two.
So the question became: which of the two prototypes would become the production Magnius? I mean, in ad-hoc listening, everyone said both were good, but I had a moderate preference for one approach.
I bet you’re not surprised with what we did next: we did an internal Thunderdome.
We sent the two amps home with several listeners, and asked them which one they liked best. This is sighted listening, yes, but we’re also talking bare boards, and we’re talking about people who wouldn’t really know anything about the circuit, other than one had more parts on the board.
The results…were surprising.
Literally everyone chose the prototype I didn’t like as much.
Yes. As in,
everyone contradicted me. Everyone liked the TPS6120A2 version over the Magni Heresy version.
A real head-scratcher? Maybe not so much. There are lots of explanations:
- I’m deaf. Sure, maybe, whatever.
- I’m biased. I’d heard that the TPS6120A2 sounded like butt for so long that it influenced my perceptions. I did, after all, know exactly what I was listening to. (And despite that, I passed several long weekends listening to the TPS6120A2 version without becoming irritated with it.)
- They’re listening for different things. I (thought) I liked a subtle harmonic richness in the Magni Heresy version. But others liked the clarity and cleanliness of the other. But again, the differences were very small—we all remarked on how small they were.
- There’s really no difference. Maybe we’re all fooling ourselves. The differences between even tube hybrid and op-amp based amps ended up being shockingly small in our last blind listening. So maybe there really aren’t any significant differences at this level (near 120dB SINAD).
Blind Listening, Private and Public
We could have just called the whole project done at that point, and gone either way. I could have called Designer’s Perogative and said we were doing the Heresy version, and that was that. Or we could have been more democratic and enlightened and said I got out-voted, so it was gonna be the TPS6120A2 version.
But I wasn’t ready to let it go. So we did two other things:
- We set up blind listening at Schiit. Under these conditions, the differences dwindled even more. It was painfully difficult to tell any difference between the two amps. It became more of a crap-shoot, though two of us could eventually call our favorites with much better than blind chance.
- We brought the prototypes to the Schiitr for more blind listening. Our last SchiitrMeet had blind listening of the two prototypes, level-matched and really, really blind. The public was evenly divided on the amps—and many ascribed the qualities I liked to the amp I didn’t like, and vice-versa.
So, really, no difference. Not really. Except there was.
Finally I made a decision: we’d be democratic. A panel of trusted listeners all liked the one I didn’t prefer, and the public didn’t seem to have a preference.
So the TPS6120A2 it was.
Coda and Second Thoughts
Should I have stuck it out for the amp that I preferred? In the end, I don’t think so. I think I was probably biased by what I heard about “the sound of the TPS6120A2.” In later listening, I’m perfectly happy where we ended up.
What I think is more important is how huge a departure this is for us:
- We started by shooting for measurements, in a high-value balanced amp.
- We verified measurements and iterated by measurements in early development.
- But…we also did listening, including sighted and blind listening.
- We took the listening outside the office and to the public.
- We used the listening results to determine which product won.
And, I think what is also more important is the insane value of this product:
- State of the art measurements.
- The only measurement-centric product with an Alps Blue Velvet RK27114 potentiometer—full boat crazy heroic potentiometer here!
- All Alps signal switches.
- All Neutrik connectors.
- Designed and built in California. And no, not by shipping finished chassis/boards/etc from a low-wage country—our chassis are made literally 5 miles away and our boards are made about 30 miles away. As usual, the wall-wart is from China (or from Taiwan, maybe, by the time you read this.)
- Covered by a 2-year warranty from a company with a decade-long track record.
- Priced 50-150% less (yes, LESS) than the cost of other high-measuring, balanced products made in low-wage countries.
I mean, think of it: for the same price as an Asgard 3 (which may still go up in price), you get a full balanced amplifier with superb measurements, super-high-power, an completely-insane-unsurpassed-by-any-competitor potentiometer, solid parts, and a class-leading feature set. If you’d told me, 10 years ago, when we started, that we’d be able to do that, I would have told you you’re insane.
Aside: yes, Asgard 3 is still holding its price. For now. We’ll continue to keep it there if we can.
Aside to the aside: and again, no, Jotunheim is not going away. It’s a truly unique product, the only fully discrete balanced differential amp that’s also affordable and modular…and I think there is space for both it and Magnius in the line.
And now, on to other things…things that, as Kennedy might have said, that we choose to do not because they’re easy,
but because they’re hard…
I'm glad that's over with (for now [...sigh...]). Re: Mark Waldrep's AES lecture. I'm not an audio engineer; however, the nuances of acoustic-science and perception keep me intrigued. I keep tabs on an interesting blog (
