[GearMe]

We certainly thought so.

I mean it took us all of 2-3hrs to get it up and running, and except for the continual fussing with the battery packs, all we wanted to do was listen to this or that, or this other over there.

It was one of 2 'highlights' of our explorations into 'pushing the envelope'

JJ

Or...pushing the box in this case!

 

[33na3rd]

No reports from the preamp shootout at the Schiitr?

On the home front, I'm really enjoying Freya S.
I'll write more about that when I get more hours on, but I can say that Freya S sounded amazingly sweet right out of the box without warm up or burn in. Burning in just seems to add a bit more refinement.

 

[bluebanshee76]

Only two Rag2 owners until now? Does anybody else enjoy this new Schiit in their own chain and refrains from telling us? We need more impressions! Ok, at least I do.

 

[neogeosnk]

I want to know if it drives the HE-6Se. And how hot it is.

 

[RCBinTN]

Only two Rag2 owners until now? Does anybody else enjoy this new Schiit in their own chain and refrains from telling us? We need more impressions! Ok, at least I do.

Hopefully there will be a thread dedicated to Ragnarok2 impressions

 

[hearditontheX]

I wonder how the performance of the phono and Multibit DAC cards compare to the Mani and Modi Multibit, respectively.

After watching this video, I learned the answer is there is no difference since the Rag 2 modules are the same as the boards in the modular units.

 

[Snowpuppy77]

After watching this video, I learned the answer is there is no difference since the Rag 2 modules are the same as the boards in the modular units.

Which is a good thing as these are good products. To me even if you already have a good external DAC and Phonostage it may be worth 300 bucks to add these anyway. Could come in useful down the road.

 

[belgiangenius]

Seeing the tubes powered down when not needed, and enjoying the music popping out from a darker background whilst tubes are active due to lower noise floor... I'm very happy with the upgrade from Freya 1 to Freya+

Agreed. It is ridiculous how much better the Freya+ is than the original Freya. I can use any of my tubes now and no more mysterious tube noises.

I already sold my original Freya. In hindsight, it clearly has issues.

 

[Jason Stoddard]

2019, Chapter 9:
The Ragnarok That Is

If you haven’t read the saga about how we got to the Ragnarok 2 That Is, you might want to go back and re-read the chapter on The Ragnaroks That Weren’t. For those of you who’ve had enough of my blathering, here’s the Cliff’s Notes version:

Shortly after developing the Ragnarok (1), Jason realized he screwed up and left some critical things off of it. He then spends the next few years developing prototypes that don’t fully fix the screwups, and then decides to throw them away and come up with something much, much better.

And yeah, I know, Ragnarok 2 still ain’t perfect. But then again, nothing is.

For those of you just starting this odyssey, you may be wondering, “Is it normal to have a bunch of false starts when you re-develop a product?” The short answer is, “It depends.” But that hides the reality. Since this book is, nominally, about start-ups and business practices and management foibles, maybe that’s the best jumping-off point.

Let’s talk about perspective, make-work, fear, tweaking, and the willingness to say ‘no.’”


The Comfort of Not Screwing it Up Versus the Terror of Clean-Sheet Design

I may be crazy, but I’m pretty sure I know how the engineering teams at Jeep and Chevy felt, respectively, when they were tasked to design the new Wrangler JL and the upcoming mid-engine Corvette.

In the case of Jeep, the design direction was probably something like, “For the love of all that is holy, guys, don’t screw the pooch on this one. Fix the irritating stuff, but don’t make it too different. It’s still a Jeep, it’s still a Wrangler, don’t make it into a Range Rover or an Expedition or Tundra or anything like that. Because this thing sells great, the aftermarket is super strong, people love these things, and if you mess with it too much, you’re all gonna be looking for new jobs.”

Wow. If you’re an engineer looking to play it safe, that’s Linus’ security blanket. That’s about as warm and fuzzy as you can get. You know what you’re shooting for. Heck, they’ll probably help you aim. You aren’t going to get fired for playing it safe. And the downsides of the current product are well-documented. So if you fix some of those, add some cool new features, then you’re done. Everyone keeps buying. Everyone’s happy. You’re the golden boys.

For the Chevy guys, the brief was totally different. Something like, “Hey guys, we have this iconic car we’ve been making since the earth cooled, and it’s always pretty much had the same formula…so much so that the fans kinda went nuts when we got rid of the flippy headlights and again when we killed the round taillights. Buuuuuuttttt…you know, sales ain’t so hot, and we’re kinda maxed out, performance-wize, in the current iteration. So we’re saying, ‘Shred it all. Blow it up. Turn it around. We’re going mid-engine like Ferrari and Lamborghini.’ Make that work. Oh, and while you’re at it, remember this is gonna be a huge high-profile launch that everyone will be watching, so don’t screw it up too bad, because then we’re all mega-boned.”

And that’s the terror of clean-sheet design. Because, even though you haven’t as a company done a mid-engine car since the Fiero, this gets dropped in your lap. If you’re an engineer who loves challenges (and doesn’t mind looking for work if things don’t go well), then this is paradise. If you want something that’s easy, well, yeah, it’s clench time.

So what does this have to do with Ragnarok 2, you ask?

It’s simple: when redeveloping products, you have a choice. You can play it safe or start over. Sometimes playing it safe is the right approach. Sometimes starting over is the only way to do it.

Ragnarok 2 started with a series of play-it-safe designs, then morphed into a do-over. This is absolutely, 100% NOT the way to do it, because it’s the least efficient way to do it. We wasted a lot of time polishing turds, when we should have been flushing and starting over.

So how do you know whether to play it safe or start over? By taking the time to think about what you need, rather than engaging in make-work.

• Make-work is when you start a new design because it’s been a year or two or three, and you think you need a new design.
• Make-work is when you start a new design because you’re bored, and it might be fun to try something.
• Make-work is when someone asks you “why don’t you do it this way,” and you decide to try it, without figuring if it might be a good idea first.

Now, make-work can be fun. I like playing with what-ifs and I like laying out PC boards. That’s fun. But it’s also dangerous, because I can be wasting time on stuff that has literally no chance of ever happening, while neglecting real design that needs to be done.

What you really need is perspective. Note the examples above. There’s a hint of one perspective in each—sales. In one case, sales are good. In the other case, sales are declining. But there are other perspectives that should factor into whether or not you should be doing a clean-sheet design.

Let’s go through some of them:

1. Sales. If sales are steady or up, why are you bothering to redesign the product at all? Is there something coming out that you need to deal with, like a new Bluetooth standard or whatever? If not, leave it alone. If sales are down, though, that’s a different story. In general:

• Sales up: don’t redesign.
• Sales flat: see Past Performance, Competition, and Market. Redesign only if Performance is sub-par, or you know there are changes coming in Competition and Market.
• Sales down: redesign, but take into consideration Competition and Market to decide if it’s a tweak or clean-sheet

• Tweak: if Performance is fine, but there are no changes in Competition and Market.
• Clean Sheet: if Competition is ramping up or Market has changed.

2. Performance. If the current product is the most reliable thing in your line and has had no issues with production, then there’s no need for any huge changes. If it’s been problematic in terms of reliability, ease of production, or ergonomics, that needs to be factored in here. In general:

• Tweak: if performance has been at or above par.
• Clean sheet: if performance is a problem.

3. Competition. Did something new get introduced that’s a game-changer? Your sales may be fine this month, and the next…but then things may not be so hot. Is a new competitor promising a new product? You need to take that into consideration, too. In general:

• Tweak: If nothing much has changed, and there’s nothing on the horizon.
• Clean sheet: if there’s something new and exciting out there…or coming.

4. Market. If the market itself hasn’t changed much—in other words, if there are no new devices, no new platforms, no new software, no new hardware changes (like, say, lack of headphone jacks), then that should make you take a safer course. If things are changing rapidly, then it’s going to argue for more radical changes.

5. Time and Personality. Are you bored with nothing to do? Then you have time to play. And believe me, it’s better to have some stuff you’ve played with when the Competition and Market change in unexpected ways. However, your first job is to ensure the stellar performance of your current line, rather than always promising a future utopia. If you don’t have your own current products on your desk (or in your rack), you’re doing something wrong.

In the case of Ragnarok, sales were never rapid AND they were a monumental pain to manufacture, so a clean sheet design makes total sense, even leaving aside the competition and the market. So me embarking on a tweak design at first was stupid.

Ah well. Live and learn.

Let’s talk about…


The Real Ragnarok 2

The Ragnarok 2 we ended up with is about as clean-sheet as you can get. The chassis is entirely different, the topology is entirely different, the microprocessor management strategy is different, the look is entirely different…about the only things it shares with its predecessor are power output and (some) functionality.

The key to all of these radical changes was simple: the adoption of the heatsinks from Vidar. As soon as I decided to throw out the thermal design of the old Ragnarok, everything opened up, and everything got wayyyyyy simpler. I had already planned to move to a new topology, but when the chassis-as-heatsink idea went away, suddenly I was able to:

1. Dramatically simplify the chassis.
2. Dramatically simplify the layout.

These two “dramas” led to a huge decrease in complexity. Early Ragnarok 1s had 5 PC boards. Later ones still had 3. Both versions used irritating, sometimes-unreliable ribbon cables to connect critical circuits. Ragnarok 2 has two boards and no cables, just pin headers which have been dead-reliable in a ton of other applications.

Ragnarok 1s also were very, very hard to put together and take apart. Ragnarok 2 is dead-simple.

Ragnarok 1s were not modular in any way, shape, or form, because you wouldn’t ever want to take it apart. Ragnarok 2 is easily expandable and serviceable.

Ragnarok 1 ran very hot, because the chassis-as-a-heatsink idea had limits. Ragnarok 2 runs almost cold.

Ragnarok 1 used summers for single-ended output, resulting in differing sonic characteristics between balanced and SE output. Ragnarok 2 uses the same stage for everything.

Being so different, you’d expect there to be some pretty hilarious bumps on the road to production. And there were. Maybe less than you’d think, but that’s in part because I treated Ragnarok 2 with the paranoia befitting a clean-sheet design, and did extensive 2D and 3D mockups before moving on to metal.

Aside: our 3D printer is too small to make most of our chassis parts, so we’ve been using a service called 3Dhubs.com. They allow us to print full chassis—even chassis the size of Ragnarok or Yggdrasil—to get an idea of how they really fit. Yes, they’re plastic, and no, they are not pretty, but it’s invaluable for speeding up the process of getting good metal. They have chassis prints to us in days, not weeks (or, argh, sometimes months….)​

But that’s not to say there were no problems.

The first, and most obvious: the transformer. It took several, er, iterations to get a transformer with low enough field to keep the hum level down. Early versions were so bad that I thought we’d have to go to a toroidal design. The problem was, when we got toroid samples from two manufacturers, one was just plain bad and one wasn’t all that much better than the EI-core transformer we had. This is what happens when you compare a simple, off-the-shelf design to a custom design with all the secondaries we need for our product—the reality is sometimes much different.

When we realized a toroid wasn’t gonna work, we went to embalming the EI-core design. Literally wrapping it in mu-metal reduced the field, but it buzzed like mad. Putting the transformer in a welded steel box and potting it with epoxy finally got us to where we needed to be, but holy moly—that’s an exotic fix!

Despite that heroic effort, Ragnarok 2 is not as suited as Ragnarok 1 for sensitive headphones—it’s a bit noisier. Perhaps not surprising when it puts out 28W into 32 ohms. So why didn’t I try to make it even better? Because, as I said, using Ragnarok with IEMs is not such a hot idea. Perhaps I should design a separate “IEM and Sensitive Headphone Compatibility Box” that you can plug into the front of Ragnarok 2, if you’re really insistent on using it for that. After all, it will happily drive 8 ohms, so we could divide the output by 8 and still get an output impedance under an ohm…

Aside: file this under “I like to tweak.” It ties into the “Time and Personality” metric above.

The other problem we ran into? The chassis.

Because….the transformer, once in its box, wouldn’t fit in the chassis. What made this particularly unfortunate was the fact that itwouldfit, except for the height of the secondary coil. We spent some time with the transformer manufacturer trying to reduce the height of the secondary, but that proved to be futile—those designs didn’t meet power spec. Then it hit me: Ragnarok 2 sits on ½” tall feet. If the chassis was “belled” down just 1/8”, the secondary would clear and the transformer would fit.

Only one problem: the chassis were already made.

Yeah. Oops. Luckily, our new chassis supplier (a stamping company in Valencia) was willing to try to see if they could deboss the chassis without harming the finish. Surprisingly, it worked!

Aside: Ragnarok 2 is our first chassis without a flat bottom. It won’t be our last.​

Beyond that, it was fairly stupid stuff…early prototype main boards that didn’t fit the add-on cards very well, and some solder welling in production (solder shorting the pin headers). All in all, not too bad for a clean-sheet design.

I think what’s even more important is that Ragnarok 2 has encouraged us to step back and take a look at what we’re designing, and to see if they need a clean-sheet approach. Some do. Some don’t. Some simply need to go out to pasture forever. The main thing is we’re asking the hard questions…and taking the time to do multiple candidate designs if that’s what we need to do. The end result (I hope) will be better products.


So What’s This Nexus™ Thing?

Oh boy. Here we go.

Yeah, this ain’t gonna be easy. At least not in lay terms. Differentiating Nexus from other balanced topologies is, beyond matters of philosophy, a largely technical exercise. Yes, I’m immensely excited about it, and yes, it’s something I think is really cool, but I’m a geek and a dork and I am not you, and you may not care about Nexus at all.

But first, let’s start with a definition: Nexus is a balanced, differential gain stage with a discrete, current-feedback topology that exhibits certain unique traits…

Yeah. Now you’re all snoring.

Okay. Step back. Do you have products with balanced inputs and outputs? As in, do they use big 3- or 4-pin XLR connectors? (Or Sony’s new 4.4mm TRRRRRRRRRRRS—I exaggerate slightly) connector?)

Yes? Well, then you have a product with balanced, differential input and/or output.

Well, maybe.

Sigh. Yes. Maybe. Because some products may have XLR connectors, but they may not really be balanced or differential. They may use only one input phase on the connector, or the output may have a single phase with the other tied to ground. There’s no law against doing this. I personally think it’s poor business practice and misleading, but I am not the judge here.

Aside: I don’t believe there are a lot of products that do this. But it may be worth asking if you don’t know. It may save you a ton of money on balanced sources and cables.​

But, if you have products with XLR connectors, more than likely you have some kind of balanced gain stage inside the device.

Now, “balanced” does not have to be “differential.” And “differential” can be “very differential” or “weakly differential.” “Differential” can be “op amp” or “discrete.” And, depending on the stage, it will have its own qualities that make it better or worse for a specific use.

Let’s go back to that definition, and add to it a bit:

Nexus is a balanced, differential gain stage with a discrete, current-feedback topology that exhibits certain traits, such as inherently high input impedance and lack of N/N+1 gain structure, that makes it work well as a “do-all” circuit—SE to balanced, balanced to balanced, balanced to SE, SE to SE, all ins and outs are welcomed here.​

Okay, okay, I know that still doesn’t make much sense. Let’s break it down a bit:

• Balanced: as in, it accepts inputs with both a positive and negative phase, like from an XLR input (and it uses both phases).
• Differential: as in, it also accepts single-ended inputs, and converts them to balanced—inherently, by default, without additional stages.
• Discrete: it’s a bunch of individual transistors (actually mainly duals, I’ll get to that)
• Current-feedback: feedback is low impedance and goes to the sources of the input JFETs, rather than the gates.
• High input impedance: this is the weird one—since the feedback doesn’t go back to the gates, as with Supersymmetry, the input impedance is very high and invariant, making it easy to drive.
• Lack of N/N+1 gain structure: in amps that create a differential output by driving the low-impedance input of a second amplifier stage with the output of the first, you get an unbalanced gain structure when using a single-ended input. You can compensate for this by changing the resistors in the feedback network, but it indicates only a weakly differential design.
• “Do-all:” it’s tolerant of taking output only from one phase, making this a single-stage, balanced-to-balanced, SE-to-SE, or any combination thereof.

The catch? Nexus requires insane amounts of matching in order to work well. In fact, the entire stage typically uses paired, single-die matched transistors, plus hand-matched and then re-reeled SOT23 surface-mount JFETs, plus operational point servos, plus a number of tricks to ensure that everything is staying in balance internally. This is pretty crazy stuff.

Aside: the more astute engineers in the audience are thinking, “wow, sounds like it would work even better if done all on a single die as an integrated part.” And yeah, I hear you. Integrated parts like the LME49724 are actually very good. Unfortunately, we are not TI, and we don’t have the capability of producing ICs, however.

Aside to the aside: and others are asking, “if the LME49724 is so good, why aren’t you using that? Isn’t that a lot simpler?” Well, yes and no. Functionally, the LME49724 has conventional feedback, which means low input impedance…which means you need to add input buffers. Philosophically, we’re not here for the easiest solution, we’re here to do what sounds best. Lots of what we do is more complicated and doesn’t measure as well as a bunch of op-amps in a box, but we think it sounds better. You’re completely free to disagree.​

Now let’s compare to other balanced stages:

Circlotron: actually not a gain stage. This is an output stage. You could do a Nexus circlotron (yes, I know, your head just exploded). You could also do a Supersymmetry circlotron. I include this mainly because circlotrons have become combobulated with balanced amps, and because we use a circlotron in the Mjolnir 2, coupled with a single-stage differential amplifier. Feedback, however, goes back to the grids of the tubes, so it’s a low-impedance input, which necessitates discrete buffers before the differential stage—exactly like the LME49724. Circlotrons also have the problem of no inherent SE output—to get one, you’d have to play some capacitor-coupling tricks, and also have a very low-noise power supply. To summarize:

· Pros (of the circlotron output stage): very simple actual topology
· Cons (of the circlotron output stage): very complex power supply, no easy SE output
· Pros(of the Mjolnir single-stage+circlotron topology): very simple, no N/N+1 gain relationship
· Cons (of the Mjolnir single-stage+circlotron topology): needs input buffers, no easy SE output, requires interstage capacitors

Two Amps: some balanced amplifiers actually have two of the same exact amplifier inside, one for each phase of the balanced input. These amps are actually not differential at all. Give them a single-ended signal, and they’ll handle it fine; it’s just that one amp will sit idle. There’s nothing wrong with this, it’s just that there will be none of the benefits of differential operation, such as common-mode noise rejection.

· Pros: straightforward and simple to implement; no operation oddities from a customer perspective
· Cons: lots of parts, no benefits of differential operation, no conversion of SE to balanced or vice-versa

Open loop differential: you can get perfect differential operation by going open-loop; no pesky feedback to screw up your input impedance. We do this for the differential buffers in Freya+. There are tons of catches to this approach, though, the biggest of which is its stunning inefficiency. This is not a great idea for a power stage.

· Pros: simple, perfectly differential, can be high performance at smaller (preamp) signal levels
· Cons: extremely inefficient use of the power supply (very low swing for the rails), high distortion at high signal levels (power amp levels), high output impedance (unsuitable for driving low impedances)

Pivot Point: aha, another one of ours, and actually fairly similar to Nexus. This predecessor to Nexus has a lot of the same pluses, including high input impedance, but it is only weakly differential, leading to complex gain relationships. It is not as demanding in terms of matched parts, and still provides excellent performance, so it is still a great candidate for a simpler amp like Jotunheim.

· Pros: simpler than Nexus, does not require insane parts matching
· Cons: only weakly differential

Supersymmetry: Nelson Pass’ patented (now expired) topology is an elegant way to maximize the benefits of a differential stage, but it doesn’t get us past the low-impedance input problem. I think this may be the basis of integrated designs such as the LME49724 (TI did license his patent when it was in force), but I don’t know the internal workings of huge companies like TI.

· Pros: does not require insane parts matching (as far as I know)
· Cons: feedback is conventional, making input impedance low, so may require buffer stages for low noise or impedance-variable applications, such as after a volume pot

Whew.

Now, does this mean Nexus is the ideal balanced, differential stage? Of course not. Nothing is ideal for every application. I mean, look at the examples above. We’re currently using four different balanced, differential stages in different products. We’re using an open loop differential stage in Freya+, because Nexus is more complicated and expensive. We’re using pivot point in Jotunheim, because, again, Nexus is more complicated and expensive. We’re using a circlotron and hybrid stage with buffers in Mjolnir 2, because that’s what works best for that particular hybrid topology.

Aside: a tube hybrid Nexus, like we might use for a future Mjolnir, may never happen. Seriously. We tried it. Even the best-matched input tubes cause severe imbalances in the topology. Using a differential servo to try to pull everything in place is problematic, as it fights the operational point servos. You might end up with three servos and microprocessor oversight…and still end up with many unusable tubes. However, if we can address some cost issues, there may eventually be a Jotunheim Nexus, but that’s a bigger challenge than you might think—we get away with it in Ragnarok 2 due to extreme matching and some hand-trimming, and in Freya S because it’s a relatively simple, line-level implementation (which still requires insane matching). When you put together the requirements of power output and thermal stability that Jotunheim entails, it gets a lot more complicated in a hurry. ​

“So what about Continuity™?” someone asks. “What does that have to do with Nexus, and vice-versa?”

Good question. Frustrating answer: nothing. Nexus is a balanced topology, and Continuity is an output stage. You could have a Nexus + Continuity amplifier if you wanted (and you had a lot of heatsinking—seriously, Ragnarok 2 would be like a 20W amp if it had a Continuity output stage. Plus, it wouldn’t fit—too many output devices needed.

I’m not leaving out the possibility of a Nexus + Continuity amp in the future, but make no mistake, it’s not around the corner. Nor will it be cheap. Nor will it run cool. Thems the breaks, unless I can work up a high-sensitivity Continuity topology (many dangers there, probably won’t happen, don’t get excited, might sound like butt even if possible because it requires amplification of the sense resistor).

Okay. Take a step back. So what does this all this Nexus stuff mean?

Bottom line: It means there’s a new balanced, differential topology that we think works really neat and sounds very good. It’s applicable to a lot of things, but not everything. If you like the way it sounds, it’s probably pretty exciting. If you don’t, it probably isn’t.

We hope you like it!

 

[Snowpuppy77]

2019, Chapter 9:
The Ragnarok That Is

If you haven’t read the saga about how we got to the Ragnarok 2 That Is, you might want to go back and re-read the chapter on The Ragnaroks That Weren’t. For those of you who’ve had enough of my blathering, here’s the Cliff’s Notes version:

Shortly after developing the Ragnarok (1), Jason realized he screwed up and left some critical things off of it. He then spends the next few years developing prototypes that don’t fully fix the screwups, and then decides to throw them away and come up with something much, much better.

And yeah, I know, Ragnarok 2 still ain’t perfect. But then again, nothing is.

For those of you just starting this odyssey, you may be wondering, “Is it normal to have a bunch of false starts when you re-develop a product?” The short answer is, “It depends.” But that hides the reality. Since this book is, nominally, about start-ups and business practices and management foibles, maybe that’s the best jumping-off point.

Let’s talk about perspective, make-work, fear, tweaking, and the willingness to say ‘no.’”


The Comfort of Not Screwing it Up Versus the Terror of Clean-Sheet Design

I may be crazy, but I’m pretty sure I know how the engineering teams at Jeep and Chevy felt, respectively, when they were tasked to design the new Wrangler JL and the upcoming mid-engine Corvette.

In the case of Jeep, the design direction was probably something like, “For the love of all that is holy, guys, don’t screw the pooch on this one. Fix the irritating stuff, but don’t make it too different. It’s still a Jeep, it’s still a Wrangler, don’t make it into a Range Rover or an Expedition or Tundra or anything like that. Because this thing sells great, the aftermarket is super strong, people love these things, and if you mess with it too much, you’re all gonna be looking for new jobs.”

Wow. If you’re an engineer looking to play it safe, that’s Linus’ security blanket. That’s about as warm and fuzzy as you can get. You know what you’re shooting for. Heck, they’ll probably help you aim. You aren’t going to get fired for playing it safe. And the downsides of the current product are well-documented. So if you fix some of those, add some cool new features, then you’re done. Everyone keeps buying. Everyone’s happy. You’re the golden boys.

For the Chevy guys, the brief was totally different. Something like, “Hey guys, we have this iconic car we’ve been making since the earth cooled, and it’s always pretty much had the same formula…so much so that the fans kinda went nuts when we got rid of the flippy headlights and again when we killed the round taillights. Buuuuuuttttt…you know, sales ain’t so hot, and we’re kinda maxed out, performance-wize, in the current iteration. So we’re saying, ‘Shred it all. Blow it up. Turn it around. We’re going mid-engine like Ferrari and Lamborghini.’ Make that work. Oh, and while you’re at it, remember this is gonna be a huge high-profile launch that everyone will be watching, so don’t screw it up too bad, because then we’re all mega-boned.”

And that’s the terror of clean-sheet design. Because, even though you haven’t as a company done a mid-engine car since the Fiero, this gets dropped in your lap. If you’re an engineer who loves challenges (and doesn’t mind looking for work if things don’t go well), then this is paradise. If you want something that’s easy, well, yeah, it’s clench time.

So what does this have to do with Ragnarok 2, you ask?

It’s simple: when redeveloping products, you have a choice. You can play it safe or start over. Sometimes playing it safe is the right approach. Sometimes starting over is the only way to do it.

Ragnarok 2 started with a series of play-it-safe designs, then morphed into a do-over. This is absolutely, 100% NOT the way to do it, because it’s the least efficient way to do it. We wasted a lot of time polishing turds, when we should have been flushing and starting over.

So how do you know whether to play it safe or start over? By taking the time to think about what you need, rather than engaging in make-work.

• Make-work is when you start a new design because it’s been a year or two or three, and you think you need a new design.
• Make-work is when you start a new design because you’re bored, and it might be fun to try something.
• Make-work is when someone asks you “why don’t you do it this way,” and you decide to try it, without figuring if it might be a good idea first.

Now, make-work can be fun. I like playing with what-ifs and I like laying out PC boards. That’s fun. But it’s also dangerous, because I can be wasting time on stuff that has literally no chance of ever happening, while neglecting real design that needs to be done.

What you really need is perspective. Note the examples above. There’s a hint of one perspective in each—sales. In one case, sales are good. In the other case, sales are declining. But there are other perspectives that should factor into whether or not you should be doing a clean-sheet design.

Let’s go through some of them:

1. Sales. If sales are steady or up, why are you bothering to redesign the product at all? Is there something coming out that you need to deal with, like a new Bluetooth standard or whatever? If not, leave it alone. If sales are down, though, that’s a different story. In general:

• Sales up: don’t redesign.
• Sales flat: see Past Performance, Competition, and Market. Redesign only if Performance is sub-par, or you know there are changes coming in Competition and Market.
• Sales down: redesign, but take into consideration Competition and Market to decide if it’s a tweak or clean-sheet

• Tweak: if Performance is fine, but there are no changes in Competition and Market.
• Clean Sheet: if Competition is ramping up or Market has changed.

2. Performance. If the current product is the most reliable thing in your line and has had no issues with production, then there’s no need for any huge changes. If it’s been problematic in terms of reliability, ease of production, or ergonomics, that needs to be factored in here. In general:

• Tweak: if performance has been at or above par.
• Clean sheet: if performance is a problem.

3. Competition. Did something new get introduced that’s a game-changer? Your sales may be fine this month, and the next…but then things may not be so hot. Is a new competitor promising a new product? You need to take that into consideration, too. In general:

• Tweak: If nothing much has changed, and there’s nothing on the horizon.
• Clean sheet: if there’s something new and exciting out there…or coming.

4. Market. If the market itself hasn’t changed much—in other words, if there are no new devices, no new platforms, no new software, no new hardware changes (like, say, lack of headphone jacks), then that should make you take a safer course. If things are changing rapidly, then it’s going to argue for more radical changes.

5. Time and Personality. Are you bored with nothing to do? Then you have time to play. And believe me, it’s better to have some stuff you’ve played with when the Competition and Market change in unexpected ways. However, your first job is to ensure the stellar performance of your current line, rather than always promising a future utopia. If you don’t have your own current products on your desk (or in your rack), you’re doing something wrong.

In the case of Ragnarok, sales were never rapid AND they were a monumental pain to manufacture, so a clean sheet design makes total sense, even leaving aside the competition and the market. So me embarking on a tweak design at first was stupid.

Ah well. Live and learn.

Let’s talk about…


The Real Ragnarok 2

The Ragnarok 2 we ended up with is about as clean-sheet as you can get. The chassis is entirely different, the topology is entirely different, the microprocessor management strategy is different, the look is entirely different…about the only things it shares with its predecessor are power output and (some) functionality.

The key to all of these radical changes was simple: the adoption of the heatsinks from Vidar. As soon as I decided to throw out the thermal design of the old Ragnarok, everything opened up, and everything got wayyyyyy simpler. I had already planned to move to a new topology, but when the chassis-as-heatsink idea went away, suddenly I was able to:

1. Dramatically simplify the chassis.
2. Dramatically simplify the layout.

These two “dramas” led to a huge decrease in complexity. Early Ragnarok 1s had 5 PC boards. Later ones still had 3. Both versions used irritating, sometimes-unreliable ribbon cables to connect critical circuits. Ragnarok 2 has two boards and no cables, just pin headers which have been dead-reliable in a ton of other applications.

Ragnarok 1s also were very, very hard to put together and take apart. Ragnarok 2 is dead-simple.

Ragnarok 1s were not modular in any way, shape, or form, because you wouldn’t ever want to take it apart. Ragnarok 2 is easily expandable and serviceable.

Ragnarok 1 ran very hot, because the chassis-as-a-heatsink idea had limits. Ragnarok 2 runs almost cold.

Ragnarok 1 used summers for single-ended output, resulting in differing sonic characteristics between balanced and SE output. Ragnarok 2 uses the same stage for everything.

Being so different, you’d expect there to be some pretty hilarious bumps on the road to production. And there were. Maybe less than you’d think, but that’s in part because I treated Ragnarok 2 with the paranoia befitting a clean-sheet design, and did extensive 2D and 3D mockups before moving on to metal.

Aside: our 3D printer is too small to make most of our chassis parts, so we’ve been using a service called 3Dhubs.com. They allow us to print full chassis—even chassis the size of Ragnarok or Yggdrasil—to get an idea of how they really fit. Yes, they’re plastic, and no, they are not pretty, but it’s invaluable for speeding up the process of getting good metal. They have chassis prints to us in days, not weeks (or, argh, sometimes months….)​

But that’s not to say there were no problems.

The first, and most obvious: the transformer. It took several, er, iterations to get a transformer with low enough field to keep the hum level down. Early versions were so bad that I thought we’d have to go to a toroidal design. The problem was, when we got toroid samples from two manufacturers, one was just plain bad and one wasn’t all that much better than the EI-core transformer we had. This is what happens when you compare a simple, off-the-shelf design to a custom design with all the secondaries we need for our product—the reality is sometimes much different.

When we realized a toroid wasn’t gonna work, we went to embalming the EI-core design. Literally wrapping it in mu-metal reduced the field, but it buzzed like mad. Putting the transformer in a welded steel box and potting it with epoxy finally got us to where we needed to be, but holy moly—that’s an exotic fix!

Despite that heroic effort, Ragnarok 2 is not as suited as Ragnarok 1 for sensitive headphones—it’s a bit noisier. Perhaps not surprising when it puts out 28W into 32 ohms. So why didn’t I try to make it even better? Because, as I said, using Ragnarok with IEMs is not such a hot idea. Perhaps I should design a separate “IEM and Sensitive Headphone Compatibility Box” that you can plug into the front of Ragnarok 2, if you’re really insistent on using it for that. After all, it will happily drive 8 ohms, so we could divide the output by 8 and still get an output impedance under an ohm…

Aside: file this under “I like to tweak.” It ties into the “Time and Personality” metric above.

The other problem we ran into? The chassis.

Because….the transformer, once in its box, wouldn’t fit in the chassis. What made this particularly unfortunate was the fact that itwouldfit, except for the height of the secondary coil. We spent some time with the transformer manufacturer trying to reduce the height of the secondary, but that proved to be futile—those designs didn’t meet power spec. Then it hit me: Ragnarok 2 sits on ½” tall feet. If the chassis was “belled” down just 1/8”, the secondary would clear and the transformer would fit.

Only one problem: the chassis were already made.

Yeah. Oops. Luckily, our new chassis supplier (a stamping company in Valencia) was willing to try to see if they could deboss the chassis without harming the finish. Surprisingly, it worked!

Aside: Ragnarok 2 is our first chassis without a flat bottom. It won’t be our last.​

Beyond that, it was fairly stupid stuff…early prototype main boards that didn’t fit the add-on cards very well, and some solder welling in production (solder shorting the pin headers). All in all, not too bad for a clean-sheet design.

I think what’s even more important is that Ragnarok 2 has encouraged us to step back and take a look at what we’re designing, and to see if they need a clean-sheet approach. Some do. Some don’t. Some simply need to go out to pasture forever. The main thing is we’re asking the hard questions…and taking the time to do multiple candidate designs if that’s what we need to do. The end result (I hope) will be better products.


So What’s This Nexus™ Thing?

Oh boy. Here we go.

Yeah, this ain’t gonna be easy. At least not in lay terms. Differentiating Nexus from other balanced topologies is, beyond matters of philosophy, a largely technical exercise. Yes, I’m immensely excited about it, and yes, it’s something I think is really cool, but I’m a geek and a dork and I am not you, and you may not care about Nexus at all.

But first, let’s start with a definition: Nexus is a balanced, differential gain stage with a discrete, current-feedback topology that exhibits certain unique traits…

Yeah. Now you’re all snoring.

Okay. Step back. Do you have products with balanced inputs and outputs? As in, do they use big 3- or 4-pin XLR connectors? (Or Sony’s new 4.4mm TRRRRRRRRRRRS—I exaggerate slightly) connector?)

Yes? Well, then you have a product with balanced, differential input and/or output.

Well, maybe.

Sigh. Yes. Maybe. Because some products may have XLR connectors, but they may not really be balanced or differential. They may use only one input phase on the connector, or the output may have a single phase with the other tied to ground. There’s no law against doing this. I personally think it’s poor business practice and misleading, but I am not the judge here.

Aside: I don’t believe there are a lot of products that do this. But it may be worth asking if you don’t know. It may save you a ton of money on balanced sources and cables.​

But, if you have products with XLR connectors, more than likely you have some kind of balanced gain stage inside the device.

Now, “balanced” does not have to be “differential.” And “differential” can be “very differential” or “weakly differential.” “Differential” can be “op amp” or “discrete.” And, depending on the stage, it will have its own qualities that make it better or worse for a specific use.

Let’s go back to that definition, and add to it a bit:

Nexus is a balanced, differential gain stage with a discrete, current-feedback topology that exhibits certain traits, such as inherently high input impedance and lack of N/N+1 gain structure, that makes it work well as a “do-all” circuit—SE to balanced, balanced to balanced, balanced to SE, SE to SE, all ins and outs are welcomed here.​

Okay, okay, I know that still doesn’t make much sense. Let’s break it down a bit:

• Balanced: as in, it accepts inputs with both a positive and negative phase, like from an XLR input (and it uses both phases).
• Differential: as in, it also accepts single-ended inputs, and converts them to balanced—inherently, by default, without additional stages.
• Discrete: it’s a bunch of individual transistors (actually mainly duals, I’ll get to that)
• Current-feedback: feedback is low impedance and goes to the sources of the input JFETs, rather than the gates.
• High input impedance: this is the weird one—since the feedback doesn’t go back to the gates, as with Supersymmetry, the input impedance is very high and invariant, making it easy to drive.
• Lack of N/N+1 gain structure: in amps that create a differential output by driving the low-impedance input of a second amplifier stage with the output of the first, you get an unbalanced gain structure when using a single-ended input. You can compensate for this by changing the resistors in the feedback network, but it indicates only a weakly differential design.
• “Do-all:” it’s tolerant of taking output only from one phase, making this a single-stage, balanced-to-balanced, SE-to-SE, or any combination thereof.

The catch? Nexus requires insane amounts of matching in order to work well. In fact, the entire stage typically uses paired, single-die matched transistors, plus hand-matched and then re-reeled SOT23 surface-mount JFETs, plus operational point servos, plus a number of tricks to ensure that everything is staying in balance internally. This is pretty crazy stuff.

Aside: the more astute engineers in the audience are thinking, “wow, sounds like it would work even better if done all on a single die as an integrated part.” And yeah, I hear you. Integrated parts like the LME49724 are actually very good. Unfortunately, we are not TI, and we don’t have the capability of producing ICs, however.

Aside to the aside: and others are asking, “if the LME49724 is so good, why aren’t you using that? Isn’t that a lot simpler?” Well, yes and no. Functionally, the LME49724 has conventional feedback, which means low input impedance…which means you need to add input buffers. Philosophically, we’re not here for the easiest solution, we’re here to do what sounds best. Lots of what we do is more complicated and doesn’t measure as well as a bunch of op-amps in a box, but we think it sounds better. You’re completely free to disagree.​

Now let’s compare to other balanced stages:

Circlotron: actually not a gain stage. This is an output stage. You could do a Nexus circlotron (yes, I know, your head just exploded). You could also do a Supersymmetry circlotron. I include this mainly because circlotrons have become combobulated with balanced amps, and because we use a circlotron in the Mjolnir 2, coupled with a single-stage differential amplifier. Feedback, however, goes back to the grids of the tubes, so it’s a low-impedance input, which necessitates discrete buffers before the differential stage—exactly like the LME49724. Circlotrons also have the problem of no inherent SE output—to get one, you’d have to play some capacitor-coupling tricks, and also have a very low-noise power supply. To summarize:

· Pros (of the circlotron output stage): very simple actual topology
· Cons (of the circlotron output stage): very complex power supply, no easy SE output
· Pros(of the Mjolnir single-stage+circlotron topology): very simple, no N/N+1 gain relationship
· Cons (of the Mjolnir single-stage+circlotron topology): needs input buffers, no easy SE output, requires interstage capacitors

Two Amps: some balanced amplifiers actually have two of the same exact amplifier inside, one for each phase of the balanced input. These amps are actually not differential at all. Give them a single-ended signal, and they’ll handle it fine; it’s just that one amp will sit idle. There’s nothing wrong with this, it’s just that there will be none of the benefits of differential operation, such as common-mode noise rejection.

· Pros: straightforward and simple to implement; no operation oddities from a customer perspective
· Cons: lots of parts, no benefits of differential operation, no conversion of SE to balanced or vice-versa

Open loop differential: you can get perfect differential operation by going open-loop; no pesky feedback to screw up your input impedance. We do this for the differential buffers in Freya+. There are tons of catches to this approach, though, the biggest of which is its stunning inefficiency. This is not a great idea for a power stage.

· Pros: simple, perfectly differential, can be high performance at smaller (preamp) signal levels
· Cons: extremely inefficient use of the power supply (very low swing for the rails), high distortion at high signal levels (power amp levels), high output impedance (unsuitable for driving low impedances)

Pivot Point: aha, another one of ours, and actually fairly similar to Nexus. This predecessor to Nexus has a lot of the same pluses, including high input impedance, but it is only weakly differential, leading to complex gain relationships. It is not as demanding in terms of matched parts, and still provides excellent performance, so it is still a great candidate for a simpler amp like Jotunheim.

· Pros: simpler than Nexus, does not require insane parts matching
· Cons: only weakly differential

Supersymmetry: Nelson Pass’ patented (now expired) topology is an elegant way to maximize the benefits of a differential stage, but it doesn’t get us past the low-impedance input problem. I think this may be the basis of integrated designs such as the LME49724 (TI did license his patent when it was in force), but I don’t know the internal workings of huge companies like TI.

· Pros: does not require insane parts matching (as far as I know)
· Cons: feedback is conventional, making input impedance low, so may require buffer stages for low noise or impedance-variable applications, such as after a volume pot

Whew.

Now, does this mean Nexus is the ideal balanced, differential stage? Of course not. Nothing is ideal for every application. I mean, look at the examples above. We’re currently using four different balanced, differential stages in different products. We’re using an open loop differential stage in Freya+, because Nexus is more complicated and expensive. We’re using pivot point in Jotunheim, because, again, Nexus is more complicated and expensive. We’re using a circlotron and hybrid stage with buffers in Mjolnir 2, because that’s what works best for that particular hybrid topology.

Aside: a tube hybrid Nexus, like we might use for a future Mjolnir, may never happen. Seriously. We tried it. Even the best-matched input tubes cause severe imbalances in the topology. Using a differential servo to try to pull everything in place is problematic, as it fights the operational point servos. You might end up with three servos and microprocessor oversight…and still end up with many unusable tubes. However, if we can address some cost issues, there may eventually be a Jotunheim Nexus, but that’s a bigger challenge than you might think—we get away with it in Ragnarok 2 due to extreme matching and some hand-trimming, and in Freya S because it’s a relatively simple, line-level implementation (which still requires insane matching). When you put together the requirements of power output and thermal stability that Jotunheim entails, it gets a lot more complicated in a hurry. ​

“So what about Continuity™?” someone asks. “What does that have to do with Nexus, and vice-versa?”

Good question. Frustrating answer: nothing. Nexus is a balanced topology, and Continuity is an output stage. You could have a Nexus + Continuity amplifier if you wanted (and you had a lot of heatsinking—seriously, Ragnarok 2 would be like a 20W amp if it had a Continuity output stage. Plus, it wouldn’t fit—too many output devices needed.

I’m not leaving out the possibility of a Nexus + Continuity amp in the future, but make no mistake, it’s not around the corner. Nor will it be cheap. Nor will it run cool. Thems the breaks, unless I can work up a high-sensitivity Continuity topology (many dangers there, probably won’t happen, don’t get excited, might sound like butt even if possible because it requires amplification of the sense resistor).

Okay. Take a step back. So what does this all this Nexus stuff mean?

Bottom line: It means there’s a new balanced, differential topology that we think works really neat and sounds very good. It’s applicable to a lot of things, but not everything. If you like the way it sounds, it’s probably pretty exciting. If you don’t, it probably isn’t.

We hope you like it!

Great read. Really enjoyed that. A couple of questions:

I see that Nexus is better for single ended because it converts to balanced without summers. However how does it improve all balanced from Yggy input to the balanced headphone output? Or is the improvement really for single ended? If so still a good thing.

Your literature on the Ragnarok 1 said there was 4 watts of class A and for the Ragnarok 2 the literature says 1 watt of class A. Why couldn't you have a higher class A bias on the Ragnarok 2? Does the Ragnarok 2 have sound quality advantages when it switches to class B?

 

[audio philestine]

Put me down for November 2019 as my guess for the release date

Congratulations on beating my guess for the Nexus release date by 5 months!

 

[Ableza]

Nice chapter Jason. I believe the "insane matching" required in a Supersymmetry design is if you employ JFET buffers on the input to deal with the low input impedance. These need to be carefully matched.

I look forward to hearing your Nexus topology!

 

[artur9]

2019, Chapter 9:
The Ragnarok That Is
So What’s This Nexus™ Thing?

But first, let’s start with a definition: Nexus is a balanced, differential gain stage with a discrete, current-feedback topology that exhibits certain unique traits…

Could a Nexus device be made to compete with Jensen SE->XLR transformers like this one: https://www.cs1.net/products/jensen_transformers/PO-2RX.htm

 

[Jason Stoddard]

Great read. Really enjoyed that. A couple of questions:

I see that Nexus is better for single ended because it converts to balanced without summers. However how does it improve all balanced from Yggy input to the balanced headphone output? Or is the improvement really for single ended? If so still a good thing.

Your literature on the Ragnarok 1 said there was 4 watts of class A and for the Ragnarok 2 the literature says 1 watt of class A. Why couldn't you have a higher class A bias on the Ragnarok 2? Does the Ragnarok 2 have sound quality advantages when it switches to class B?

With respect to Nexus and SE, the benefit is that it doesn't need summers on the output, nor balanced conversion on the input. For balanced, it is highly differential--so it's ideal for both types of signal.

With respect to Nexus and bias, yep, it needs a lot less bias. Ragnarok 1 was open-loop. It needed all the bias it could stand. Nexus doesn't need it, so it doesn't get it. There's no performance benefit.

Hope that helps a bit!

Could a Nexus device be made to compete with Jensen SE->XLR transformers like this one: https://www.cs1.net/products/jensen_transformers/PO-2RX.htm

Sure. Or you can do a dozen other things. I mean, the LME49724 already slaughters even the best transformers, with respect to conventional measurements. So do the balanced line drivers and receivers (DRV and INA parts). There are plenty of ways to skin that cat...the questions are, which is best for the particular application (transformers don't require power and isolate ground, for example), and what you think sounds the best.

 

[Snowpuppy77]

With respect to Nexus and SE, the benefit is that it doesn't need summers on the output, nor balanced conversion on the input. For balanced, it is highly differential--so it's ideal for both types of signal.

With respect to Nexus and bias, yep, it needs a lot less bias. Ragnarok 1 was open-loop. It needed all the bias it could stand. Nexus doesn't need it, so it doesn't get it. There's no performance benefit.

Hope that helps a bit!



Sure. Or you can do a dozen other things. I mean, the LME49724 already slaughters even the best transformers, with respect to conventional measurements. So do the balanced line drivers and receivers (DRV and INA parts). There are plenty of ways to skin that cat...the questions are, which is best for the particular application (transformers don't require power and isolate ground, for example), and what you think sounds the best.

Thank you for the reply Jason. It does help to understand and good to know that Nexus also benefits an all balanced connection. And my turntable and phonostage is single ended so I like knowing that it would benefit even more when connecting to Nexus of Ragnarok 2 vs. Ragnarok 1.

With respect to the lower class A bias of the Ragnarok 2. My understanding is that the Aegir, with the exception of required power, has sound quality advantages over the Vidar due to being class A up to 10 watts and then eliminating transconductance droop outside of the class A bias region. I have long thought, perhaps incorrectly, that the more class A the better. Is the reason that the Ragnarok 2 does not need more class A because it does not have as much transconductance droop outside of the class A region as the Ragnarok 1?