[StimpyWan]

And no carpet in Texas.

Dirt floors huh? Easier to clean up BBQ (and beer) spills! Makes me feel right at home. Just like the Appalachian area I grew up in...!

 

[stochasticJim]

The recent discussion of acoustic transformer hum amongst the Schittizens led to an interesting journey of discovery.

When I bought my Aegir late last year, I bought it fully aware that the Aegir had some acoustic transformer hum. There's a big honkin' transformer at the front of the amp. It's not a big deal when your amp is >6 feet away across your room, probably near the floor. Since my Schtack is at arms length just to my left, I knew that the acoustic transformer hum could be an issue at my listening position.

The acoustic transformer hum from my Aegir is noticeable, but only when no music is playing. Even minor HVAC background noise at my desk is enough to mask the hum. But at times during the day, the transformer can take on a slight buzzing quality which is more objectionable. I did some research.

As mentioned on this thread previously, a DC component in the AC mains can cause excessive acoustic transformer noise, so I started shopping. My first choice was the Emotiva CMX-2, but they were out of stock. I also looked at solutions from iFi and Audiolab, but I settled on the HumDinger from Audio by Van Alstine, mostly because it had two outlets.

When I plugged in the HumDinger, it helped a bit, mostly with the more objectionable buzzing. But I could still hear the Aegir transformer. Time to go low tech.

I built a small (11" x 7") acoustic panel with 2" Ownens Corning 703 FRK fiberglass (materials I had on hand). When I have my Aegir on, the panel sits in front of it, with about 3" of clearance for air flow. Acoustic transformer noise managed.

When the thread sine wave wandered to a discussion on acoustic transformer noise, I wondered if the Emotiva CMX-2 (now in stock) would do a better job than the AVA HumDinger. On a lark, I ordered one, figuring I could check it quickly, and return it if the hum didn't improve. I plugged in the CMX-2: hum was about the same. I swapped between the CMX-2 and HumDinger to make sure they were the same. Then I tried the HumDinger and the CMX-2 in series to see if that was any better. Nope.

Even though the CMX-2 is a fine product, it just didn't make any difference for my hum. It was time send it back to Emotiva. But first, I played some music with just the CMX-2 in the AC mains path.

Wait a tick. Does my Schtack sound better?

I played a variety of music with the HumDinger, then the CMX-2, back and forth. I definitely heard an improvement with the CMX-2. I heard more ambience in the recording. Better depth in the sound field. More sparkle in the higher frequencies. Toto's Kingdom of Desire transformed from a smudgy 2-D presentation to a focused 3-D sound. In Mahler's 5th (CSO - Abbado, 2017 edition), the trombones came from the back of the orchestra as they should. One of my favorite infectious (post-punk?) pop albums, Future Me Hates Me by The Beths, has a more open sound, to my delight. I can better hear the breathiness around the edges of Edie Brickell's voice on So Familiar with Steve Martin. And Snarky Puppy's Sylva with Metropole Orkest benefitted from the improved soundstage definition.

Dammit.

I have to keep the CMX-2.

I must have some grunge on my AC mains that the CMX-2 is taking away. The difference is more subtle than stark, but it is enough for me. The CMX-2 removes grunge that really allows my Aegir to shine.

But my small acoustic panel is still necessary.

This hobby is a blast. I've got a lot more music to check out.

 

[RickB]

Definitely.

A modern DAP is mostly a software project. Fewer things in our life need software, not more. I hate when perfectly good devices get junked up by bad software. I hate software. And I'm a software developer. I can't even imagine what the hardware guys think of it.

Your Overcast app is awesome! Probably my favorite app on the iPhone.

Edit: I just realized I'm replying to an almost 8 year old post.

 

[MikeVG]

The recent discussion of acoustic transformer hum amongst the Schittizens led to an interesting journey of discovery.

When I bought my Aegir late last year, I bought it fully aware that the Aegir had some acoustic transformer hum. There's a big honkin' transformer at the front of the amp. It's not a big deal when your amp is >6 feet away across your room, probably near the floor. Since my Schtack is at arms length just to my left, I knew that the acoustic transformer hum could be an issue at my listening position.

The acoustic transformer hum from my Aegir is noticeable, but only when no music is playing. Even minor HVAC background noise at my desk is enough to mask the hum. But at times during the day, the transformer can take on a slight buzzing quality which is more objectionable. I did some research.

As mentioned on this thread previously, a DC component in the AC mains can cause excessive acoustic transformer noise, so I started shopping. My first choice was the Emotiva CMX-2, but they were out of stock. I also looked at solutions from iFi and Audiolab, but I settled on the HumDinger from Audio by Van Alstine, mostly because it had two outlets.

When I plugged in the HumDinger, it helped a bit, mostly with the more objectionable buzzing. But I could still hear the Aegir transformer. Time to go low tech.

I built a small (11" x 7") acoustic panel with 2" Ownens Corning 703 FRK fiberglass (materials I had on hand). When I have my Aegir on, the panel sits in front of it, with about 3" of clearance for air flow. Acoustic transformer noise managed.



When the thread sine wave wandered to a discussion on acoustic transformer noise, I wondered if the Emotiva CMX-2 (now in stock) would do a better job than the AVA HumDinger. On a lark, I ordered one, figuring I could check it quickly, and return it if the hum didn't improve. I plugged in the CMX-2: hum was about the same. I swapped between the CMX-2 and HumDinger to make sure they were the same. Then I tried the HumDinger and the CMX-2 in series to see if that was any better. Nope.

Even though the CMX-2 is a fine product, it just didn't make any difference for my hum. It was time send it back to Emotiva. But first, I played some music with just the CMX-2 in the AC mains path.

Wait a tick. Does my Schtack sound better?

I played a variety of music with the HumDinger, then the CMX-2, back and forth. I definitely heard an improvement with the CMX-2. I heard more ambience in the recording. Better depth in the sound field. More sparkle in the higher frequencies. Toto's Kingdom of Desire transformed from a smudgy 2-D presentation to a focused 3-D sound. In Mahler's 5th (CSO - Abbado, 2017 edition), the trombones came from the back of the orchestra as they should. One of my favorite infectious (post-punk?) pop albums, Future Me Hates Me by The Beths, has a more open sound, to my delight. I can better hear the breathiness around the edges of Edie Brickell's voice on So Familiar with Steve Martin. And Snarky Puppy's Sylva with Metropole Orkest benefitted from the improved soundstage definition.

Dammit.

I have to keep the CMX-2.

I must have some grunge on my AC mains that the CMX-2 is taking away. The difference is more subtle than stark, but it is enough for me. The CMX-2 removes grunge that really allows my Aegir to shine.

But my small acoustic panel is still necessary.

This hobby is a blast. I've got a lot more music to check out.

That is an interesting journey. I had dual Aegir's for a while and never noticed any transformer hum. Maybe I just got lucky. It's when I decided to move to dual Tyr's and a Yggy LIM that my introduction to transformer hum began. The Tyr's had a audible hum from day one that could be heard from a listening position of 6 feet away. Once Schiit was notified, they wasted no time on replacing them. I had new ones within one week and they were noticeably quieter. Fast forward a few weeks and my long awaited Yggy arrived. I plugged it in and there was the same transformer hum that was even louder the the original Tyr's. Again Schiit was very responsive and shipped out a replacement. Thankfully all is good now and I can get back to enjoying the music. As a side note, I did order and install two Emotiva CMX-2 units so that all my Schiit gear goes through them first. It was more of precaution than anything else. I have not done any critical listening yet to see if they improved the musical playback.

 

[JWMAstro]

Chapter 27:
Twilight of the Gods—Ragnarok from 2009 Until Today

Okay, so shoot me. I moved the Ragnarok chapter up a notch. But I think it makes sense, in the context of yesterday’s announcement of the Ragnarok public beta. Yes, we are almost there. Soon, we’ll be shipping Ragnaroks…for real.

Yes. Stop laughing. Rina still doesn’t believe me.

An aside: the reason Ragnarok was delayed (again) was twofold. One was my personal drama. The other was that we found another software bug in the final code—a particularly nasty one that caused production main boards to go into meltdown mode when they were first plugged into the production control boards. It took us a while to ensure that it wasn’t something inherent in the amplifier itself, and to identify the offending code. For the technically minded out there, it was related to some debugging code that remained in the software—it swapped two microprocessor output pins so we could monitor the operational parameters in real time. However, since the pins swapped included the output to the control DACs, and the updating was much slower than production, the Ragnarok microprocessor would end up trying to bump the bias up to unsustainable levels. When it finally updated, bang! Bad news. That code is now gone, and Ragnarok is running stably.

Another aside: Jude’s Ragnarok, as well as others in the very small private beta, did not have this software, and hence didn’t have the problem. All in all, we think the public beta will be short and uneventful, but in an abundance of caution, we’re proceeding with it anyway.


The Ragnarok Saga: Pre-Blab

And…it really has been a saga. Ragnarok was one of the first amps we discussed in Schiit’s pre-launch days. I have sketches dating back to late 2009.

So why was it one of the earliest concepts? Because it was a direct outgrowth of driving headphones with Sumo speaker amps. I figured, well, if a Sumo 60W amp works with headphones, why not make a reeely reeeeellly ridiculous headphone amp. We could put a warning sticker over the headphone output, saying, “If you remove this, you realize that you can make the magic smoke come out of any headphone, and agree to hold Schiit harmless for any damage.”

Yeah, silly. But an interesting concept for a publicity stunt for a then-unknown, unlaunched company.

Of course, reality soon set in—driving headphones with a speaker amp sometimes works out just fine, but most speaker amps are:

• Relatively noisy (forget your sensitive headphones and IEMs)
• Not headphone-friendly on turn-on (1V of DC is no big deal to a speaker, but a bad day for some headphones)
• Big, hot, heavy, and not desk-friendly

So, the idea got shelved, after I’d spent some time playing with the old Sumo Antares and came to the realization that, like duh, a speaker amp usually isn’t a headphone amp—at least not without a whole lot more development work.

But the kept ticking over in my head. And, after our early experiences with planar headphones and the moar power!!! phenomenon, I decided that yes, it was something that we should pursue.

And, of course, I had to mention it. This was 2011.

Yes, I am an idiot.


The Ragnarok Saga: Post-Blab

Okay, so after I’d opened my mouth and told the world that we were working on a statement amp and DAC, I realized that we’d, well, actually have to deliver something eventually. But I didn’t set any timeframe, because, well, I didn’t figure there’d be many people interested in an unnamed, un-spec’d, unreleased super-power amp.

But they were. From the first mention that Schiit was going to do a statement amp, we started getting inquiries. When will it ship? What will it be like? What’s the price?

Based on super-scientific complete-WAG BS, I started telling people it might be about a year, and it would be super-powerful, and probably about $1000.

Yes. Dig yourself deeper. This is a textbook example of what not to do. If you are silly enough to pre-announce a product, here’s what you should do:

• Say nothing more, and hope the inquiries stop.
• If pressed, say you got drunk and spoke out of turn, and the product doesn’t actually exist.
• If presented with documents you accidentally emailed someone confirming the existence of a project to develop said unreleased product, claim that you were either:
  • Blue-skying it, but it has since been abandoned
  • Developing a product for someone else.

Or, in other words, shut up until it dies. Then, when it’s ready, bring it out and let people buy it.

Yes, it really is that simple.

But no. I had to confirm its existence. I even had to opine that it would be a really, really cool product. Again: don’t do this.

But…since the cat was out of the bag, I figured that I should at least get started on the design. People who know a thing or two about large power amp design know that you really start with two things:

• Thermal design. How do you get rid of the heat? Running speakers isn’t like running headphones—there’s going to be significant thermal load.
• Transformer. How big of a transformer do you need, and does it fit in the chassis you have in mind?

Once you’ve gotten those two things set, you can then proceed on to the rest.

So, the first calculations I did were for the chassis and transformer. Would our trick of using the chassis as a heatsink work? It turned out that the answer was yes—if we were looking at a standard 2U height rack-sized product. Which gelled with other 60W amps I’d designed in the past. Transformer? Sure, you could use a 3” high product with a thick stack. Done that before, too.

But what about features? The first idea we had for Ragnarok was mighty conventional:

1. 60W/8 ohms, 100W/4 ohms
2. 3 levels of gain switching
3. Alps RK27 balanced volume pot
4. Nice Grayhill input switch
5. 5 inputs, 3 SE and 2 balanced
6. Circlotron topology
7. 16 x 12 x 4” chassis

So really, nothing too nuts, except for the gain switching and circlotron topology.

We ran with that idea for a while, long enough to get some switch samples and order the balanced pots. But we never built a prototype of it.

No. Ego stepped in, and started us down the path to insanity.

Fun fact: one of the reasons we did Mjolnir was because the Ragnarok concept had already started to grow…and I wanted to use those balanced pots somewhere. So, Mjolnir was our shot at the simplest, most basic balanced amp we could do. Ragnarok had already been elevated, in our minds, to something that was quite a bit more.

How much more?

Well, it all started with the switches. Grayhill switches are very nice, but pricey. It really wouldn’t cost any more for us to switch inputs with relays. And with relays, we could do the switching right at the back of the amp, where the inputs came in—which would save having to run a bunch of traces for each input up to the front switch. This would make things like crosstalk a whole lot better.

So hey, why not use relays?

But then you have 7 relays in the chassis (2 each for the balanced inputs, 1 each for the other inputs.) And, if you have that many relays, why not have a few more? Alps RK27 pots are nice. Sure. But a resistor-switched stepped attenuator is better. And another 12 relays would get us a 64-step attenuator at 1-1.2dB steps—much better than you’d get with a front-panel switched attenuator…and tons of control with 3 different gain levels.

So now we’re up to 19 relays. Still not a huge deal.

But if we have 19 relays, we should be switching gain with relays, too, right. So add another 4. And, of course, you have to have output muting. So that’s another 2.

Yep. 25 relays in that first Ragnarok (there are actually 29 in production, since we added a separate headphone/speaker mute, and changed the way the gain switching relays worked—but we’ll get there.)

But 25 relays wasn’t crazy enough. I also decided it might be fun to be able to run the amp with either a solid-state or tube input stage. So, that meant that the transformer would have to accommodate either one. Which meant a very large, complex transformer with like 4 Molex connectors hanging off of it, and a separate gain stage board (one for tubes, one for transistors).

And, of course, with 25 relays, we needed a microprocessor to run them—to handle switching, volume control, etc. So then we needed a shielded front control board…

…and, if we were going to all that trouble, then we might as well have a really cool volume control—that is, one that actually acted like a volume control, with stops at either end, rather than an endlessly spinning encoder. Which meant we needed a fairly beefy microprocessor, so we could A/D in a voltage and use that to set the output to the resistor ladder volume control.

Okay. So why do all this, you ask? For better sound. A better volume control and better switching is a big deal, when you’re going all-out. Plus, we’d already begun to believe our own BS, with all the inquiries about the “statement products.”

This is how you talk yourself into making something fairly insane.

And that’s where the first layout began.


The First Ragnarok

If the first Ragnarok had worked, it would have been a pretty damn good headphone amp, but a fairly crappy speaker amp. This is because I’d become used to doing headphone amps, and I wanted to do a super-simple single-gain-stage topology.

There’s only one problem with that: single gain stages have low overall gain—which means you can’t use much (if any) feedback to get the output impedance down.

Aside: Yes, Ragnarok is not a no-feedback amp. It is a no-overall-feedback amp, however…

Why is this important? It’s important for control over the speaker load. Speakers are physically large. They need a reasonable damping factor. As a single gain stage amp, Ragnarok would have had an output impedance of about 0.5 ohms in high gain—fine for a headphone amp, but equivalent to a damping factor of only 8 into 4-ohm loads. Not exactly ideal.

But, as it turned out, the first Ragnarok prototype was so problematic, we never got it fully operational. The tube stage didn’t get enough power from the transformer. The solid-state stage had so many layout errors that it wasn’t really worth fixing them all. And the board changes needed to accommodate both tube and solid-state really made it impractical to have it swappable—it would have to be built one way or the other.

(And, of course, this tube vs solid-state realization came after I’d shot my mouth off about how Ragnarok would be configurable for either. Yeah. Again: just shut up.)

We hacked around with it a bit, blew up a few output stages, and never really heard music through it. So, finally, we did something sensible.

We went back to the drawing board.


The Second Ragnarok

The next Ragnarok was stripped down and simplified. We got a new transformer, only suitable for solid state use. We designed the main board only for solid state. We went to a two-stage amp with enough gain to give us the damping factor we needed.

But, it still wasn’t the Ragnarok we’ll soon be shipping. (Stop laughing.) This Ragnarok still used DC servos and trimpots to set the bias for the amp, as we do in Mjolnir. This worked, albeit with some difficulty. Unlike Mjolnir, Ragnarok wasn’t thermally stable. You’d set the bias, and it would creep up, and up, and up…until bad things happened.

In retrospect, this isn’t surprising. Mjolnir uses 1 ohm source resistors in its output, and Ragnarok uses 0.1 ohms. Mjolnir’s operating point is stabilized by local feedback from the source resistors, whereas Ragarok didn’t have that luxury.

So what do you do in that case? Well, you either increase the source resistors (not a hot idea in a speaker amp), or you use something truly yucky, like a varistor, to try to compensate for the bias creep.

We didn’t want to use varistors (which are resistors that change value depending on temperature, which is something you usually don’t want—most resistors are designed to be as stable as possible over temperature, not the other way around—and considering that resistors set gain and other operational parameters, varistors are usually pretty bad juju.)

So, it was back to some hacks… more board hacks to fix the servos, and a whole lot of parts tacked on to try to get it thermally stable… and once again, we found ourselves at an impasse.

And that’s when I had my crazy, scary idea. The one that set back Ragnarok for over a year.


The Left Turn Into Hell

“Dave, if we have a microprocessor in Ragnarok, can’t we use it to set the bias, too?” I asked, one day we were deep in the Ragnarok Second Prototype Fiasco.

Dave is our software guy. Well, he also does digital and analog design, too, but since he’s the only one around here that does software, he’s our de facto “software guy.”

“Well, yes, but—“

“And if we use it to set the bias, we could get around the thermal stability problem.”

“Yes, but—“

I didn’t hear him, though. I was getting excited now. “And, if we used it to set the operating point on both sides of the circlotron, we could throw out the DC servo, too!”

Aside: this is very exciting. There are two ways to get rid of DC on the output of your products definitively (well, there are three, but I really can’t count that one, because, as far as I’m concerned, it’s not definitive):

• Coupling capacitor. This blocks DC. DC gone. Done. Easy. But now you have a capacitor-coupled amp. And you get to try to choose the most sonically transparent capacitor you can find.

• DC servo. This is a feedback loop that only operates at very low frequencies (say, 0.1 Hz and below.) This is usually more sonically innocuous than a capacitor, but even the best servos feed back some audio-frequency stuff…and cause phase shift in the bass. Not ideal.

• Trimpots. And, if you’re crazy, you can simply trim out the DC. And hope the operational point of the amplifier stays constant. And hope it doesn’t drift. No thanks, I like certainty.

“Well, yes, but—“ Dave tried again.

“Then that’s what we’ll do!” I cried, envisioning the truly revolutionary amplifier that Ragnarok could be with full microprocessor control. It would always run at the same operating point…and we could get rid of the DC servo…and it would be insane! Nobody was making anything like this!

“We can do it,” Dave said finally. “But the software will be pretty complicated, and I’m sure we’ll run into some issues—“

“But we’ll work them out!” I said, not hearing him.

In fact, I don’t think I heard anything past the initial “yes.”

But it didn’t matter. I knew this was the way to go. Ragnarok would be the first truly fully-managed amplifier, with a giant microprocessor brain overseeing everything it did. It would be amazing!

I got to laying out the new boards—which fundamentally changed almost everything about the Ragnarok. We had to add control lines from the DACs (yes, there are 2 DACs inside Ragnarok, but not for playing music…they take a 12-bit input from the microprocessor and use it to set the operational point of the amplifier.) We needed to add more sense lines so we could monitor what the amp was doing.

But I got it done, and we sent out for boards. This was it. I could feel it. We were going to have an incredible amp.

It was only later—a lot later—that I remembered Dave’s, “Well, but…”


Releasing the Names

So what did I do after all these changes?

Well, I didn’t do the sensible thing, which was to shut up. Instead, I believed we were far enough along that a 6-month development cycle would wrap everything up. Which meant we could make some noise. We could claim the names of our statement products (which had been the same since 2009, we simply didn’t use them). And we could show off a little.

So, I released preliminary information on Ragnarok and Yggdrasil on Head-fi, together with some product renderings and a very abbreviated spec sheet.

That was more than a year ago.

Yes. Shut up. That’s all there is to do. Else, suffer the consequences.


The Trouble With Software—and Hardware

In the early summer of 2013, Ragnarok seemed pretty buttoned down. We got the boards back, stuffed them, verified basic things like DC operating points, and gave them to Dave to do the software. All that was left was to sit back and await the software, which would ensure the amp always operated the way we expected it, each and every time.

Yeah, and software never has problems. Never has bugs. It always works entirely as expected, from day 1.

Uh-huh. Right. When you get into software, expect lots of fun—and the more fun, the more complex it is.

Actually, a primer on software should go here. Because, Ragnarok’s software isn’t really soft. Nor is it hard. So, it’s what we call “firmware.”

Yeah, I know. Sounds silly. But here you go:

1. Hardware: this is stuff that performs an analog or digital function that is set and defined. If you’re talking digital, the hardware is set up to do basic or complex logic in a set pattern, for a set purpose. You don’t go and change hardware. That’s why it’s “hard.” As in, “hard-wired.”

1. Firmware: this is software that is usually embedded in a system for a specific use, and is not intended to be updated frequently (if ever.) Ragnarok’s microprocessor uses software that is programmed into it during production, via an RJ-11 jack inside the chassis. It’s not intended to be updated, ever. That is, if we get it right. In the past, this might have been known as an “embedded system,” and burned onto a ROM. I used to do software like this. You really don’t want me doing it again. I’m a very bad software engineer. Similarly, all of our USB DACs use firmware that is programmed via the USB port; theoretically we could change this over time, but if you’ve ever used the C-Media firmware updater, you’d think twice (at least) about unleashing this capability on the world.

1. Software: This is code that’s intended to be changed from time to time. It’s also a very imprecise word, because it can refer to everything from machine-level firmware code to C# computer programs, to apps, to website Javascript and a whole host of code done in various high-level languages. So let’s leave this at that, since we have yet to do a true software product.

Okay, so where do FPGAs fit into all this? Well, let’s start with the brutally honest: “FPGA,” in audio, is simply the latest buzzword of the day. They do nothing that cannot be done with a microprocessor and code. In some cases, they may do it faster. But this is at the cost of insane development complexity. We’ll stick to microprocessors, thank you.

Smart readers are sitting back now, arms crossed, asking, So how bad was the first firmware you got for Ragnarok?

On first glance, not bad. The basic stuff—volume control, input switching, protection—all worked. The bias and DC nulling…well, not so much.

In fact, we only got it back after Dave admitted he blew up the outputs on the main board a couple of times. This isn’t surprising in development, but Dave’s efforts were hampered by his lack of experience with MOSFET-output amps. Which led to the Ragnarok main board running inconsistently, or not at all.

“But I replaced the parts that tested bad,” he told me.

“Dave, Dave, Dave,” I chided. Experience with thousands of MOSFET-output amps gives you a certain perspective. This perspective is: complete and utter paranoia.

MOSFETs are not BJTs. They are not tubes. When something goes wrong, it’s entirely possible for them to test OK…for a while. It’s also entirely possible they took out some other parts in the flames of their demise…which you might miss. When MOSFETs go bad, it’s best to simply replace everything. And then start up the amp really slow, in case you missed something. Dave was following the procedure you’d use for a BJT or tube amp…which meant more failures, later. And, for Dave, these were mystifying failures.

And we had other problems, too. The output stage oscillated, and needed to be recompensated. Again, not a big deal for someone who is used to power amp development (hell, I get suspicious if it doesn’t oscillate at some point before it’s been properly compensated in-circuit.) Problem is, an oscillating amplifier doesn’t exactly allow for accurate bias current measurement…which means even the best firmware is helpless.

And—to complicate matters—we had to have the microprocessor compensate for each of the gain stages, which required different voltages for each operational point. The problem was that at higher gains, the steps the DAC had to take were too large for good adjustment of the operating point.

This is where Dave stepped in, and came up with a new feedback arrangement which allowed:

1. The same operational point for all gain levels
2. The same amount of feedback for all gain levels

This in itself is a huge breakthrough. As far as I know, Ragnarok is the only gain-switchable amp that uses the same overall feedback for all gain levels—which means that the sonic impact of switching gain is effectively nil.

Big win, right?

Right. I was very pleased with our solutions. Ragnarok was finally running reliably…on speakers, on headphones, etc. Everything looked good. And we were still a couple of weeks away from RMAF 2013, which means we could bring an ugly prototype to the show, and ship shortly thereafter.

Except…a couple of days before the show, I turned on Ragnarok…and it smoked. No input, no load. Just smoke.

This, my friends, is not a good sign.

And that’s how Dave, Mike, and I pulled a couple of all-nighters getting Ragnarok back together, right before RMAF.

Problem solved. Right?

Wrong.


Ragnarok’s First Appearance

Ragnarok’s first appearance was at RMAF 2013. This version sported a very ugly, unanodized, unscreened, unfinished chassis that definitely didn’t show very well. We put a big sign on it saying, “This is a prototype!”

Which, of course, was ignored. People were so excited to see and hear Ragnarok that I don’t think many of them saw the sign.

The good news? Ragnarok didn’t blow up during the show.

The bad news? It didn’t work very well. In fact, with the early software, the operating points were anything but stable. It was all over the place. It ran hot. It ran cold. It ran in-between. It sounded pretty good once in a while. It sounded pretty bad most of the time.

All in all, not an auspicious debut. But let’s go back to that advice:

• Shut the hell up.
• Shutting up also means “don’t show it before it’s ready.”

Yes. Seriously.

Do you wonder why we don’t talk about products before release anymore?


Algorithmic Adventure

Once we were back from the show, Ragnarok smoked itself again.

All in all, probably a good thing. Because it was time for new boards to clear up some of the analog problems we’d encountered, and to fix some bonehead mistakes on the control board. So, we lost some time as I re-did the boards and got them stuffed again.

Then it was time for firmware.

This time…this time, it did the same thing. Weird operating points. Unreliable operation. Random blowups. In other words, it was not yet a shippable product.

Which really sucked. Here we were, at the end of the year, and Ragnarok wasn’t anywhere near ready. We were going to blow yet another deadline on it.

But why? It was a puzzle. At least to me. An enhancement-mode output stage will do exactly one thing, in the absence of bias: nothing. It will sit there all day long and not blow up. It can’t. It’s not on. And it won’t be on, unless bias is applied.

And… there was no reason it should be unstable as far as operational points go. The microprocessor was setting that. It should be, well, set.

Which pointed at the software.

“Dave, how frequently are you adjusting the bias on Ragnarok?” I asked, trying to get to the bottom of our mysteries.

“Frequently?” Dave blinked, looking mystified.

“Yes. Once a second, once every tenth of a second?”

“Once,” Dave said.

“Once? Once once?”

“Yeah, when you turn it on. Or when you switch gain modes.”

“Once?” I cried. “And then you let it go?”

“Yes,” Dave said. “But the operational point trends downward with temperature, so it’s safe.”

“Unless it gets so low the bias is turned off,” I said. Suddenly the crap-sounding Ragnarok started to make a lot more sense.

Dave nodded. “Well, there is that.”

Argh. “Dave, it has to adjust continuously, over time.”

“But then you have to parse out the difference between output current into a load for a music signal, and quiescent operating point.”

I nodded. “Exactly. But that’s what we intended to do from the start.”

Dave nodded, but fell silent. “That’s a lot harder.”

“Right.”

“I need to do some more work.”

“Right!”

A few more weeks went past, since I was tied up with other products, other problems. Eventually, Dave and I got together with new software…and a surprise from Dave.

“I also added debugging code so we can see the output current and the voltage out from the DACs at all times,” Dave said.

And, sure enough, the Ragnarok control board sprouted a new connector: an ancient DB-9 computer port. This, Dave hooked up to a computer. Soon, numbers were scrolling on the screen, updated about once a second.

“This is the output current,” Dave said. “This is the other channel. And this number is the DAC output, from 0-4096.”

“Neat!” I said. Which is true. This was an insanely helpful tool to debug Ragnarok.

“But, uh, Dave,” I added, pointing at the output current, which had risen from 250 mA to 300, then 400, then 550…”

“Oh,” Dave said, switching it off. “It hasn’t done that before. Did you change something? Maybe it’s not adjusting fast enough.”

I had changed something—the driver current. Which meant they were heating up at a different rate. Which the firmware couldn’t compensate for.

“Can it adjust faster?” I asked.

“Sure,” Dave said. He changed a few lines of code, and we restarted the Ragnarok. This time, the output current rose over 250mA, but more slowly. But it still wanted to run away.

“I can change it so that it adjusts faster, the farther it is away from the target,” Dave said. A few lines of code later, and we had an amp that didn’t overshoot the target by more than 10% before settling down to a nice, constant 250mA.

Holy moly, we really have something here, I thought.

Of course, that was before the next day, when Ragnarok smoked again.

“Dave,” I said. “Why does this keep blowing up? I thought we put in a current limit that would shut it down.”

“I was going to do that last,” Dave said, a little sheepishly.

Argh, volume 2. Do you know how much time we would have saved if this damn thing didn’t blow itself up as a failure mode? I wanted to ask. But I restrained myself.

And, after another board rebuild and some new software, we finally had a safe, operational Ragnarok. This was a day before the San Francisco Head-Fi meet in February. Satisfied, we packed off the Ragnarok to the show.

Where…it acted like a real product for the first time. No drama. No craziness. No too-hot, no too-cold.

We were done, I thought. Now, all we needed was to order boards and parts, and get to shipping.

Yeah, right.


Interfacing with the Real World

Back at the office, though, tests told a different story.

Ragnarok wasn’t good at differentiating between music and bias, as Dave had solemnly predicted. Compressed, low-dynamic-range music into 4 ohm speakers could cause itself to de-bias the output stage. Translation into English: very bad sound. The DC offset wasn’t all that hot, because Dave was only looking at bias, not DC offset, over time. At certain temperatures, it would go into a state where it wouldn’t switch into high gain mode.

Yeah. More development. We had to work out an algorithm to help differentiate between bias and music. We had to improve the bias setting over time. We had to work out the non-switching-into-high-gain problem, which was due mainly to the way we were stepping up and down the bias (and never hitting the target, in some cases.)

Which meant, in reality, more months of software time.

And, when we were finally ready to ship the first prototypes (to Jude and a couple other NDA listeners), we still got bitten. Jude’s first Ragnarok still had the “I don’t want to switch into high gain” problem, which we thought we’d worked out.

In the end, we changed the entire bias algorithm. It’s actually quite a neat bit of code…and something I’m not eager to get into. If someone else wants to do this insanity, let them have the pain, too.

Which put us at about TheShow timeframe. Ragnarok has been acting like a product for a while now. Everything seems sorted. Which made us confident enough to bring it to the show and put up a sign that said, “Shipping in June.”

Yeah. Right. Shut up.


The Final Gotchas

With all of our software problems, we were (ahem) somewhat distracted. Which meant that things we should have been paying attention to were passed by.

Things like:

1. The entire first run of transformers being the wrong voltage. Scrapped. Our fault.
2. The pots being the wrong length. Re-ordered.
3. The control board tactile switches being the wrong length. Ditto.
4. The board house stuffing a lot of the first boards wrong. Back for rework.
5. Our production line being totally unprepared for a product with 5 boards, 700 components, wiring, chassis-mount switches, transformer connectors, specific mechanical decoupling requirements, etc. Training, training.
6. Our production line literally running out of space. We’ve gotten more space.

And I’ve left out a ton of details in the saga above, details like:

1. 6 transformer changes, including splitting the circlotron output transformer and the input voltage gain transformer into to entities
2. Adventures in circlotron summing circuits, and the multiple meltdowns that occurred while figuring out the right ones
3. The insanity that is a circlotron, and the effects of nonsymmetrical loading on it
4. The crazy software just for volume control—it actually mutes the input between levels, to eliminate any popping or big glitches
5. Various blowups I’d rather forget

Suffice to say, the Ragnarok story isn’t a story at all…it’s a saga, befitting of the name. It is, by far, our most complex product to date, and it defines a number of firsts in the amplifier area—most notably as the first truly universal amplifier, using the same gain stage and output for speakers and headphones, and as the first completely managed amplifier, dispensing with DC servos and other band-aids that can affect sound.

As far as the whole saga goes—all the mis-steps, all the gotchas, all the surprises, well…this is the kind of pain it takes to make groundbreaking products.

If there’s a lesson to be learned here, it’s that blazing a new path isn’t to be taken lightly. You should take a long, hard look at your capabilities and resources, and plan for how it will impact everything you do.

And, of course…keep your damn mouth shut until it’s ready.

Welcome to Ragnarok. The end of the world.

Or the beginning?

This is the best technical saga you have described yet! I will let (i.e. force) my research team and engineers to read it. You easily justify the essential trick of multiplying any time estimate by Pi, or e ( base of natural log) depending on your relative fascination with circles or logarithms. For your circlotron, I suppose Pi would be right!

 

[Plautus001]

So, back again to the witchcraft department: I've been living with what sounded to me like a grounding problem for several months because I didn't want to take my trunk apart to figure it out. Of the eight drivers in my car, it affected only one. It didn't get louder or quieter when I turned the volume knob, but it did change volume and pitch somewhat when I stepped on the gas. It was also somewhat tolerable because road noise masked it. Yesterday I finally did the gymnastics required to lay face-up in my trunk, several times, and swap interconnects, speaker leads, etc. to confirm a bad amp channel was not the culprit. Next step was taking out a couple of trunk finishers to get access to the second of two DSP units (the first applies correction to return the signal from the head unit to a flat curve, as verified by nobody for my install ). Before I started swapping RCAs here, on a lark, I disconnected the +12V, ground and 12V trigger leads and the reconnected them. Even before I disconnected them I tugged on them and they were secure. Now they're secure-er, right?

Whatever, noise is gone and I'm happy-er...

At least that kind of schiit doesn't happen inside with my head phone gear. Wasn't there a request a few pages back for @Jason Stoddard to make some mobile audio stuff? Wouldn't have to be brushed aluminum, even @bcowen would buy C8 blue if he could put it in the trunk.

In car audio and alarms 90% of issues are related to ground. A simple way to improve car audio and sometimes car electrical performance in general is to upgrade the ground at the battery and the engine.

 

[jcflox]

The recent discussion of acoustic transformer hum amongst the Schittizens led to an interesting journey of discovery.

When I bought my Aegir late last year, I bought it fully aware that the Aegir had some acoustic transformer hum. There's a big honkin' transformer at the front of the amp. It's not a big deal when your amp is >6 feet away across your room, probably near the floor. Since my Schtack is at arms length just to my left, I knew that the acoustic transformer hum could be an issue at my listening position.

The acoustic transformer hum from my Aegir is noticeable, but only when no music is playing. Even minor HVAC background noise at my desk is enough to mask the hum. But at times during the day, the transformer can take on a slight buzzing quality which is more objectionable. I did some research.

As mentioned on this thread previously, a DC component in the AC mains can cause excessive acoustic transformer noise, so I started shopping. My first choice was the Emotiva CMX-2, but they were out of stock. I also looked at solutions from iFi and Audiolab, but I settled on the HumDinger from Audio by Van Alstine, mostly because it had two outlets.

When I plugged in the HumDinger, it helped a bit, mostly with the more objectionable buzzing. But I could still hear the Aegir transformer. Time to go low tech.

I built a small (11" x 7") acoustic panel with 2" Ownens Corning 703 FRK fiberglass (materials I had on hand). When I have my Aegir on, the panel sits in front of it, with about 3" of clearance for air flow. Acoustic transformer noise managed.



When the thread sine wave wandered to a discussion on acoustic transformer noise, I wondered if the Emotiva CMX-2 (now in stock) would do a better job than the AVA HumDinger. On a lark, I ordered one, figuring I could check it quickly, and return it if the hum didn't improve. I plugged in the CMX-2: hum was about the same. I swapped between the CMX-2 and HumDinger to make sure they were the same. Then I tried the HumDinger and the CMX-2 in series to see if that was any better. Nope.

Even though the CMX-2 is a fine product, it just didn't make any difference for my hum. It was time send it back to Emotiva. But first, I played some music with just the CMX-2 in the AC mains path.

Wait a tick. Does my Schtack sound better?

I played a variety of music with the HumDinger, then the CMX-2, back and forth. I definitely heard an improvement with the CMX-2. I heard more ambience in the recording. Better depth in the sound field. More sparkle in the higher frequencies. Toto's Kingdom of Desire transformed from a smudgy 2-D presentation to a focused 3-D sound. In Mahler's 5th (CSO - Abbado, 2017 edition), the trombones came from the back of the orchestra as they should. One of my favorite infectious (post-punk?) pop albums, Future Me Hates Me by The Beths, has a more open sound, to my delight. I can better hear the breathiness around the edges of Edie Brickell's voice on So Familiar with Steve Martin. And Snarky Puppy's Sylva with Metropole Orkest benefitted from the improved soundstage definition.

Dammit.

I have to keep the CMX-2.

I must have some grunge on my AC mains that the CMX-2 is taking away. The difference is more subtle than stark, but it is enough for me. The CMX-2 removes grunge that really allows my Aegir to shine.

But my small acoustic panel is still necessary.

This hobby is a blast. I've got a lot more music to check out.

Thanks for sharing your experience!

For my Tyr hum, as already posted, CMX-2 didn't make a noticeable difference, but I am keeping it in the system in case DC ever does appear on my AC. =-)

As additional troubleshooting, I turned off all the breakers to my house except the one feeding Tyr...no difference in hum.

As a final lark, I relocated it to a completely different room/circuit...no difference in hum.

Maybe my ears are too good, or my listening room is too quiet! Haha

I did reach out to Schiit, because I've seen the big-J himself post that hum shouldn't be audible from the listening position (6ft in my case)...unfortunately I'm not getting a very expedient response, but trying to give them the benefit of the doubt and waiting patiently.

 

[JWMAstro]

After reading all this I can't wait for the Yggdrassil chapter

Absolutely!!

 

[JWMAstro]

That sounds like an expensive amount of acrylic. Which as far as I recall is pretty rigid and transmits vibration pretty well. How about a silly putty desk?

I may hold out for the aerogel desk.

Aerogel is actually quite stiff and tougher than you might imagine. I have worked with aerogel-based Cherenkov particle detectors since the 1980s and it is remarkable stuff as it is the lightest solid. We have some of the largest aerogel panels ever made (60 cm x 60 cm) but despite what I said above, the surface is not usable for a desk. You could destroy it pretty quickly. Go for synthetic quartz or agate ( i.e. cast glass) on multi-step vibration isolators like those used in gravitational radiation detectors to isolate from ground vibrations as from traffic (easily 6 db per step and as deep as you want to construct). Of course that only works for conducted noise, not acoustics. Isolates from the floor but not from the air.

 

[Jason Stoddard]

This is the best technical saga you have described yet! I will let (i.e. force) my research team and engineers to read it. You easily justify the essential trick of multiplying any time estimate by Pi, or e ( base of natural log) depending on your relative fascination with circles or logarithms. For your circlotron, I suppose Pi would be right!

I really need to do a chapter on "Terrifying Vistas, Comforting Constraints, and Unexpected Results," documenting, well, days like today.

Here's the thing:

1. There's a lot of value in doing crazy new designs that go wayyyyyy outside your comfort zone, like the one I mentioned a few days ago. This is an as-yet-unfinished design where I allowed myself to go completely nuts, from the assumptions about the type of product it was, to going beyond the known frontiers of our discrete topologies, to doing everything absolutely right with no compromises, damn the size and heat, full speed ahead.
2. But no constraints, none at all, is also really (a) inefficient, (b) unrealistic, and (c) downright terrifying. Yes, terrifying. Engineering is not science, but you can definitely get wayyyy too far down theoretical-ish paths that may not work at all--look into log/antilog error correction for just one example. A product with 800 parts is one thing...a product with 8,000 parts is, well, maybe not producible in terms of cost and reliability.

So there have to be constraints. But where do you put them?

Sigh.

This is the Really Big Question.

I'm not going to answer it here. But maybe I can illuminate some of its dimensions. (Maybe.)

There are two reasons I do "buzzword bingo" on a lot of our proprietary approaches (Nexus, Continuity, Coherence, etc.) The first is to make it more understandable, both to me and to our customers. The second is to remind myself when I'm re-using these approaches. Because re-use is fine and great and fast and all, but it's also lazy. It's easy to fall into a rut, both in terms of performance and in terms of believing your own BS. Both are super dangerous.

Aside: and this is why op-amp based design doesn't thrill me so much...what fun is there in dropping in the "latest and greatest" chip to get 1.5dB better measurements at -120dB? None for me. Especially when the "latest and greatest" will soon be deployed by, well, everyone...and then it's a race to the (price) bottom. Same reason Mike likes his wacky multibit DACs, rather than counting zeroes on the available delta-sigma options. But then again, I may be completely and totally insane. As I've said before. YMMV. ​

So what did I decide to do with this crazy new product? I decided to split the difference between going back to first principles and making it a box of buzzword bingo. Which is to say:

1. I decided to see how far our unique discrete analog stages could go. It's one thing to do Nexus. It's another thing to do inverting Nexus (for applications where you need a virtual ground), a voltage-feedback Nexus, and a higher-function (redacted) Nexus. None of the latter I ever tried. All of the latter may not work. But it's worth trying. So I decided to build a number of "test sleds," very much in the spirit of the original Jotunheim test-sled that led to Nexus development.
2. I also decided to go back and reexamine a number of the discrete functions we use. Many of these were developed back in the dark days of Stanford Research being our top test gear. Now's the time to break them down and see how they really do, on a more detailed level. This work is especially important, given the extreme complexity of the new design.

It's too early to tell how this will all work out, but I've already had my first surprise--one of the common dual-stage blocks we use (again, sorry, redacted) works fantastically well...but I didn't know it only worked fantastically well through the cancellation of distortion from one stage to another! Now, with both stages optimized, much higher performance is possible.

This just happened today.

Where will this all lead? I'm not sure. It's still early. And I'm still wrestling with decisions like, "Well, if I do this all discrete, it'll be 158 more parts...and, where it is, it really shouldn't matter." Do I stay true to an all-out design, or do I make some logical compromises?

I don't know yet. But I do know one thing...I'll eventually write about it!

 

[A Jedi]

All out man, all out.

On your deathbed, will you smile at the compromises or the moonshots?

 

[bcowen]

All out man, all out.

On your deathbed, will you smile at the compromises or the moonshots?

That's kinda morbid, but I like the way you think.

 

[Plautus001]

I really need to do a chapter on "Terrifying Vistas, Comforting Constraints, and Unexpected Results," documenting, well, days like today.

Here's the thing:

1. There's a lot of value in doing crazy new designs that go wayyyyyy outside your comfort zone, like the one I mentioned a few days ago. This is an as-yet-unfinished design where I allowed myself to go completely nuts, from the assumptions about the type of product it was, to going beyond the known frontiers of our discrete topologies, to doing everything absolutely right with no compromises, damn the size and heat, full speed ahead.
2. But no constraints, none at all, is also really (a) inefficient, (b) unrealistic, and (c) downright terrifying. Yes, terrifying. Engineering is not science, but you can definitely get wayyyy too far down theoretical-ish paths that may not work at all--look into log/antilog error correction for just one example. A product with 800 parts is one thing...a product with 8,000 parts is, well, maybe not producible in terms of cost and reliability.

So there have to be constraints. But where do you put them?

Sigh.

This is the Really Big Question.

I'm not going to answer it here. But maybe I can illuminate some of its dimensions. (Maybe.)

There are two reasons I do "buzzword bingo" on a lot of our proprietary approaches (Nexus, Continuity, Coherence, etc.) The first is to make it more understandable, both to me and to our customers. The second is to remind myself when I'm re-using these approaches. Because re-use is fine and great and fast and all, but it's also lazy. It's easy to fall into a rut, both in terms of performance and in terms of believing your own BS. Both are super dangerous.

Aside: and this is why op-amp based design doesn't thrill me so much...what fun is there in dropping in the "latest and greatest" chip to get 1.5dB better measurements at -120dB? None for me. Especially when the "latest and greatest" will soon be deployed by, well, everyone...and then it's a race to the (price) bottom. Same reason Mike likes his wacky multibit DACs, rather than counting zeroes on the available delta-sigma options. But then again, I may be completely and totally insane. As I've said before. YMMV. ​

So what did I decide to do with this crazy new product? I decided to split the difference between going back to first principles and making it a box of buzzword bingo. Which is to say:

1. I decided to see how far our unique discrete analog stages could go. It's one thing to do Nexus. It's another thing to do inverting Nexus (for applications where you need a virtual ground), a voltage-feedback Nexus, and a higher-function (redacted) Nexus. None of the latter I ever tried. All of the latter may not work. But it's worth trying. So I decided to build a number of "test sleds," very much in the spirit of the original Jotunheim test-sled that led to Nexus development.
2. I also decided to go back and reexamine a number of the discrete functions we use. Many of these were developed back in the dark days of Stanford Research being our top test gear. Now's the time to break them down and see how they really do, on a more detailed level. This work is especially important, given the extreme complexity of the new design.

It's too early to tell how this will all work out, but I've already had my first surprise--one of the common dual-stage blocks we use (again, sorry, redacted) works fantastically well...but I didn't know it only worked fantastically well through the cancellation of distortion from one stage to another! Now, with both stages optimized, much higher performance is possible.

This just happened today.

Where will this all lead? I'm not sure. It's still early. And I'm still wrestling with decisions like, "Well, if I do this all discrete, it'll be 158 more parts...and, where it is, it really shouldn't matter." Do I stay true to an all-out design, or do I make some logical compromises?

I don't know yet. But I do know one thing...I'll eventually write about it!

Jason being excited makes me excited... I'm so happy to be a Schiit-Head

 

[dieslemat]

Schittizens

Quoting a movie: You had us at Schittizens

 

[A Jedi]

That's kinda morbid, but I like the way you think.

Just putting things in perspective