Every Road is a Dead End: Early Adventures with Magni
This chapter is a lesson in hubris—and in the value of chucking it all and starting over.
It happens to every company, I’m sure. There will always be a time when things are going well, reviews are great, and new products are flying off the shelves. We literally couldn’t keep Bifrost in stock. Asgard, Valhalla, and Lyr were all doing well. I’d just been contacted by the Arizona Audiophile Society, where the Bifrost had beaten all the other DACs in their blind listening test (with retail prices up to $7,500.) We had working prototypes of Mjolnir and Gungnir, and were looking forward to their launch. And we’d just started looking at space so we could move out of the garage (more on that later.)
This is the time when you start thinking, “Hey, this is going pretty well. Man, we’re really tearing it up. Wow, maybe we actually are pretty good at this!”
This isn’t good. In fact, this is the time you should be the most terrified.
Now, to be clear, we didn’t go completely over-the-top on the narcissism. We didn’t do anything truly stupid. Nobody bought a Ferrari. None of us went out and bought $1,000 bottles of Scotch. None of us created audio product derivatives to sell to Wall Street. And none of us rode to work in a sedan car borne by a dozen acolytes.
But, this run of good luck was enough to have me thinking, “Heh, a little amp? How hard can that be?”
As it turned out, it was damn difficult. Remember, 13 months from Modi proto to launch? At least 6 or 7 of those months were spent running down the wrong paths on Magni.
Philosophy Can Also Be A Prison
I spent the last part of the previous chapter nattering on about philosophy. And I truly think that every successful company should have a well-thought-through, concise philosophy that informs everything they do.
But a philosophy can also be a prison. If it’s too specific or too inflexible, you won’t be able to change when you need to. You won’t be able to adapt to new needs, new markets, new competitors. That’s also necessary.
It’s also why our philosophy is pretty broad. And if I’d done nothing more than apply that philosophy to Magni, I probably would have been fine.
Aside: our philosophy is to “make fun, affordable products that are as true to the musical source as possible,” in case you skipped the last chapter.
Instead, I larded on a bunch of additional “wants” to Magni’s initial design brief. Some of these were based on market reality. Some of them were sheer fantasy.
Let’s start with the ones based in reality:
- This amp should be versatile enough for most any headphone. We already had some very specific amps, like Valhalla for high-impedance headphones and Lyr for power-hungry orthos, but this should really be a do-all amp, since it would likely be a starter amp for many audiophiles.
- If we couldn’t do better than the inexpensive amps already out there, why bother? To me, “doing better” was a mix of more power and sonics, in a simple, attractive package.
- This amp needed to hit a very aggressive price point—a price point unimaginable when we started the company. We had to be careful about design, construction, features, reliability, etc. I had $99 as a target, to match the Modi.
And now, the fantasy:
- The topology should be as simple as possible—insanely simple, just a few transistors and a very simple power supply, almost like a solid-state version of a tube amp. That’s probably what would sound best, I thought.
- To keep costs down, we’d use a switching wall-wart to generate a single DC rail. Switching wall-warts are so cheap they show up in Cracker Jack boxes these days. I mean, unimaginably cheap. We’d have to use a switcher to keep cost down.
- It should be a neat, unique topology. I’d messed around with two-transistor gain cells. Maybe that would be cool. I’d also played around with the old JLH topology, which I remembered sounded good.
Looking back, all those extra fantasy items are really funny. Simple amps usually have to resort to Class-A output to get them linear enough to work well—and Class A was absolutely out. Magni’s tiny chassis wouldn’t be able to dissipate the heat. Switching power supplies are cheap, but absolutely scary in terms of power supply noise—not to mention the fact that a single rail would mean we’d have to use coupling capacitors at the input and output of the amp. And a neat, unique topology? Yeah, there’s a reason those are scarce. The “cool” stuff I’d played with in the past simply had too many limitations—not enough voltage swing, not linear enough, not stable into a wide range of headphone loads, etc.
But I’m An Idiot
So, of course, the first thing I had to do was to try a JLH-style amp with a switching wall-wart that I bought off of eBay. I think it was $3. Which meant, in production quantities, it could easily be a $0.50 part. Think of that—a cord, plastic chassis, PC board, switching supply doing 24V at 0.5A for half a buck.
And yeah, it was about as good as you’d expect for that price. It was so noisy that it made the JLH amp oscillate constantly at full voltage, without any input. I’m talking full-scale noise at a couple of megahertz.
To translate: instant headphone fry. Assuming the output stage lasted that long.
I tried a couple of other switching wall-warts, but they really weren’t much better. So I tried filtering them. Which doesn’t work so well when you have half a volt of noise on the ground (the engineers here are cringing).
Finally, I gave up and simply hooked the JLH topology up to our lab power supply. Now it ran fine. No oscillation. Which is what you’d expect from a clean supply.
There were only two problems:
- The JLH topology really, really doesn’t like to be transformed to a Class-AB design. It’s very nonlinear, with high distortion.
- It sounded like ass.
I mean, it sounded awful. As in, 1960s solid-state awful. I’d forgotten how bad solid state could be. Bright, nasty, confused, muddled…it simply didn’t stand up to modern designs. Not even an opamp-and-buffer design. Which I also didn’t want to do, because that’s been done to death.
After that failure, I tweaked around with the circuit for a while, and ended up with something that sounded kinda decent. But by this time, the original 16-component-per-channel design had ballooned to over twice that. It was more complicated than some of the 60- and 100-watt speaker amps I’d designed. And that was really stupid.
So what did I do? For a while I just gave up. I had a non-optimal topology and an unworkable power supply. I’d wasted a couple of months getting exactly nowhere.
The Non-Lighting Light Bulb
Sometimes when you walk away from a project, the insight will come when you least expect it. You’ll wake up one morning and have the answer. Or you’ll be driving into the office and it’ll hit you so hard you’ll say, “Hell, why didn’t I think of that before?”
In the case of Magni, walking away didn’t work. As Mjolnir and Gungnir moved towards production, and as we started our first move out of the garage, I had plenty to occupy me. I could forget about it.
But the answer didn’t come.
Not that I didn’t try. Sure, I put together a half a dozen neat circuits. JFET-MOSFET gain cell. Simple current-feedback amp. Etc.
But all of them had at least one fatal flaw. And all of them still wouldn’t work with a noisy power supply. Even if one had worked, the supply still killed it.
So I wasted more time—drawing up chassis for the Modi and the non-existent amplifier, trying still more wall-warts, tweaking circuits and hoping that something would work out. Nothing did. And I was starting to sweat. Any day now, Mike would ask me how the Magni was going, and I’d have to tell him. And he’d say, sarcastically, “I thought you said it would be easy, Sparky!”
I didn’t want to have that conversation. I didn’t want to say, “You know, an opamp and a buffer wouldn’t be so bad.”
(But, you know, even if I’d done an opamp and buffer design, it probably wouldn’t have worked because of the power supply.)
In the end, I was sitting in the garage one weekend, staring at the perfboard mess that should be a Magni. And I suddenly remembered that one thought I had: Hell, this thing has more parts than some of the speaker amps I designed.
So what if I just did it like a speaker amp? I wondered. That would eliminate the topology problem. Lin topologies could be very low-distortion—and Class AB—and direct coupled—and very, very robust.
But that was crazy! A full Lin topology for our least-expensive amp?
What the hell. I opened the schematic capture program and drew up a simple Lin amp.
It was simpler than the mess I’d designed.
But…a Lin amp really needed a bipolar power supply—that is, both positive and negative rail voltages. They didn’t like to hang halfway between a single supply and ground. That meant caps in the feedback loop, input biasing, and other ugly stuff like that.
So what if I just said, “the hell with it,” and did an AC wall-wart (basically a transformer in a box) and a half-wave bridge to create both positive and negative voltages?
Half-wave bridge, barf, I heard Mike’s voice in the back of my mind.
But I didn’t care. Maybe this was the way to go. Maybe a full Lin amp with a bipolar supply—and, what the hell, a DC servo too, might as well go crazy—maybe this would work. Maybe modern surface-mount manufacturing would make this feasible.
Aside: I really had no idea. I’d never done a surface-mount board before the Magni.
I built the Lin circuit that night and ran it on the lab supply. The damn thing worked first shot, as if to say, “Why didn’t you just do this from the start.” And it measured well. Not just well, but spectacularly. And with a few tweaks, it was running almost rail-to-rail.
Aside: “Rail to rail” is important for efficiency—a very important part of a Class AB design.
Now, I was excited. This was getting somewhere. If we could get a power supply put together to run it, we might have a product!
Except—I had no idea what a linear wall-wart would cost. They’re pretty scarce. Most people have gone over to switchers these days.
But again, like I said in the beginning, most answers are not much more than an inquiry or two away. Since I knew we were shooting for minimum cost, I wasn’t going to be able to get it from a US manufacturer. So I turned to a new source—one I’d never used before—alibaba.com.
Yes, that Alibaba. Chinese manufacturing. Now, there’s nothing wrong with that. But it was different than anything we’d done before. Luckily, Alibaba has a pretty good feedback system, so you have at least an idea of the companies you’re working with. We quickly had quotes from a half-dozen manufacturers, all at amazingly inexpensive rates. Not as inexpensive as a switcher, but still well within the envelope of a $99 product.
But what would they look like? Would they be any good? Even if they were, how well would Magni perform on a smaller power supply (smaller than the lab supply). I ordered some samples and sat back to wait.
In a week, I had my answer. They looked like standard cheap wall-warts, the kind you see on dozens of different products. But these had one big difference: they were AC wall-warts, delivering 16VAC to a half-wave rectified supply running MC-series regulators.
I did a version on perfboard and verified the performance—and sat back in shock. The Magni prototype delivered nearly 2W into 32 ohms at clipping, and distortion was less than 0.004% at 1V RMS (a much more typical headphone load.) And this was from the wall-wart. 60Hz hum from the half-wave supply was over 100dB down from 1V RMS.
It measured better than anything we made.
Still, what did it sound like? That took more waiting. Because I usually don’t listen to breadboards or perfboards—I just build a single channel and then get into the PC board, then listen to that.
Into Surface Mount
Before I did Magni, I’d never laid out a surface-mount board. It was a profoundly alien experience. I wasn’t used to the parts. I wasn’t familiar with the best way to route them. And, most of all, I still wasn’t confident it would work. The lesson from Gungnir’s pinch-off problem was too fresh in my mind. What gotchas would we find when we went to surface-mount? Would the equivalent parts even be available?
Parts turned out not to be a problem. In fact, they were a real eye-opener. When you hear someone say, “They don’t make great audio devices anymore,” and wax poetic about the glory days of Japanese transistors, they don’t work with surface mount parts. They don’t know all the cool new stuff that’s available right now—and the majority of it is in surface-mount packages.
I learned a lot throwing that first board together. But, because I wasn’t confident it would work, it wasn’t a full design. No muting relay. No servo. Hell, it didn’t have a power switch. But I wanted something we could try. Something we could listen to, and decide if it was good or bad.
In a few days, I had PC boards to play with. I threw one together and measured it. It ran pretty much the same as the prototype.
After that, it was the moment of truth. I grabbed a set of Grados and took them out to the test bench. The little Magni prototype drove them shockingly well.
But it should also be able to do better than Grados. It had tons of power. I decided I’d bring it to its knees with the LCD-2s.
Magni laughed at the LCD-2s. No problem. No big deal at all. It would easily go to ear-bleeding levels.
I sat there, laughing at the spectacle of this tiny little amp driving the LCD-2s. It looked absolutely ridiculous.
Rina came out to see if I’d lost my mind. “What are you laughing about?” she asked.
“Magni. The little amp.”
She saw that I was holding the LCD-2s. “On those?”
I laughed again. “No problem.”
“Really?” she took the headphones out of my hand and put them on. “Play my song.”
Aside: Rina has a specific song she uses to evaluate new headphones and amps. It’s not what I’d call hi-fi, but she’s heard it so many times that it’s a perfectly good reference for her. It’s Enigma’s Seven Lives (Radio Edit.) Yeah. I know. Talk to her. Hey, those of you with the earliest Asgards had them listen-tested to New Kids on the Block, thanks to her. Think about that the next time you listen.
I played her song. She listened for about a minute, poker-faced.
I frowned. What did that mean? Did she like it? Did she hate it? Was it really crap? Was I hearing things?
Eventually, she took the headphones off. She shook her head sadly and looked at me.
“So what are we going to do about Asgard?” she asked.