2018, Chapter 4:
Engineering, Part 3
“Okay, so we have a schematic.”
That’s how I was going to start this chapter. At least that was the plan.
But, today (March 13, 2018), I decided to plug the prototype into the Stanford (you know, just to get a head start on the measurement chapter). I didn’t expect to find anything particularly surprising. This was just a chance to do some printouts.
And. Yeah.
Nothing is ever easy.
Nothing.
Here’s what happened: I plugged the prototype into a wall-wart and turned it on. Then I realized I needed a second 1/8” to dual RCA connector to run back into the analyzer. So I went upstairs. And upstairs, I talked to Tyler, answered some emails, and generally farted around for a while. So it was fifteen minutes or so before I came back downstairs to measure the Vali Mini.
And, when I plugged it in, I was looking at power supply noise. LOTS of power supply noise. Much more than it should have. Now, Vali Mini will never win any low-noise prizes, but I figured I’d see pretty much all the power supply harmonics (above 60 Hz) below -100dB. And I figured that most of them above 180Hz would be pretty much down in the mud.
Nope. 60Hz was at like -60. 120 was -65, and 180 was -70, and down and down, but you could easily see 10x harmonics of the 60Hz in the noise. This is what I was looking at:
In comparison, here's a Jotunheim I had hanging around:
In case you’re wondering about all the engineer-speak, this is
not good.
And, this is
nothing like Vali.
Vali
never had this problem.
Sigh. On one hand, I love the fact that this simple little product turned up an example of bizarre behavior—in engineering terms, an “unexpected event,” or “unintended outcome,” or “@&%#$*^!^$$$!!!!”
On the other hand, I didn’t think this simple design would have any oddities, so I’m unhappy I have to spend time debugging it.
Aside: and yeah, at the time of this writing, I don’t know anything about why Vali Mini is exhibiting this behavior, other than it doesn’t do it when first turned on, and that it’s a noise artifact on the power supply. I’ll dig into this more tomorrow, and get into the problem in more detail in the measurements and documentation chapter.
“So what could it be?” you ask.
Good question. Given my experience, I have my suspicions:
- It’s a power supply problem. Specifically, a bad part or bad PCB trace or something like that. This is unlikely, because even though the noise manifests on the power supply side, the power supply design is bog-standard and 100% boring. Maybe I overvolted the 36V regulator—it might need a diode across it to protect it on shutdown. That’s an easy swap to see what the heck is going on.
- Something is oscillating. This is more likely, even though, as a no-overall-feedback amp, we shouldn’t have to worry about this much. Oscillation manifests in really bizarre ways. And, sometimes it’s hard to spot on today’s digital scopes. And it can be sneaky, like not showing up until the devices get warm and their beta goes up. And I’ve already had to add a 100pF cap across the plate load to stabilize it. And there are no local bypasses on the gain stages. I’ll be able to see any oscillation on a good digital scope or real analog scope when I get back to it tomorrow.
- It’s a really crappy layout. This is also possible. Maybe my design requirement for “cool looking” resulted in a layout that was, er, sub-optimal. I think it’s not bad, but I could have some parasitic capacitance or inductance that’s turning the design into an oscillator.
So. Yeah. Me, the smartass, figures “hey, I’ll show you this really simple design, because after all, what can go wrong? Vali worked fine, this is even easier…”
And I get boned.
Aside: More will be revealed as I learn what’s going on. If nothing else, it’s a great “learning experience” (AKA, “you got boned.”) or “teachable moment,” (AKA, “someone got boned.”) This is why I have great respect for any engineer who’s brought a product to market. Because this is really the easiest place for something unexpected to occur. It could be during the first production run. Yes. Eeeeeeek.
Repeat after me:
nothing is ever easy.
Nothing.
Ever.
PC Board Layout: Oh Gawd, The Choices
Okay, so we have a schematic. Are we done with the design?
Ahhahahahahhahaaahahahahahaaha! Please excuse me while I go laugh for, like, a half an hour or so. Because, yeah, even after you’ve done all the mental gyrations and design work to create a schematic, you’re nowhere near done. At least not if you’re aiming at producing more than a breadboard or hand-wired prototype or two.
The next step can be even more fun: designing the PC board.
You frown. “Seems kinda boring,” you say.
And yeah, maybe you have a point. PC board layout is kinda like the world’s least interesting video game. If you like mazes, you might like it. If you like solving mazes you made yourself, you might like it even more. If you like solving mazes you made yourself with rules involving how close the lines can get to each other, and with more unwritten rules about how you can loe if the wrong lines get too close for too long a run, and with the higher complexity of doing it in three dimensions, so lines can run over each other (and interact in new and interesting ways), you might LOVE it.
“Yeah, sounds like something I want someone else to do,” you say.
Yep. I hear you. Here’s the problem: if that “someone else” doesn’t know the circuit, its application, and the unintended gotchas as well as you do, you can end up with a really, really bad PC board. You can end up with one so bad it doesn’t work well enough to go to production.
“Well, I saw some PC board layout software once,” you say, as a savvy engineer. “It had this ‘autoroute’ button. If it can do this stuff automatically, how hard can it be?”
Aside: every single engineer who’s designed working PC boards is now hugging themselves and shivering in a corner.
Aside to the aside: I was told about one auto-routed board which had many power-hungry DSPs on it (pulling something like 3 amps) where the auto-route had run the main Vcc line all the way around the outside of the board (something like 12 inches of total trace length)…and the Vcc trace was 10 mils wide. If your hair isn’t standing on end after reading this, best to do some learning.
Yeah. No autoroute. You don’t use autoroute.
Nor do you use auto-place. Nor do you farm out your PC boards to the lowest bidder. Nor do you trust them to anyone who isn’t at least as good as you, in terms of the overall design. Or at least, Schiit doesn’t. Every PC board we do is done by me, Mike, or Dave. Period.
“That sounds inefficient,” you say. “How you gonna be a big multinational company if you can’t delegate, blah blah (insert buncha crap you read in the latest ‘pop CEO’ book here.)?”
Uh. No.
I actually
like doing PC boards, and I’m pretty damn fast at it too. Plus, I don’t need to review stuff for possible mistakes. Plus, I can do the metal changes needed as I do the PC board, so there’s more design efficiency and even less review. Dave’s at least as fast as I am. And he’s certainly better in terms of mixed analog/digital stuff.
Plus, we don’t really do that many boards.
Plus, you really don’t want anyone else doing it. Nobody else knows the design better. Nobody else is going to do a better overall job.
Aside: though yeah, we **** up. That’s what prototypes are for.
“But I’ve never designed a PC board,” you say. “I don’t know where I’d start.”
Well, here’s some blunt advice:
if you intend to make products for mass production, you’d better start learning.
Because:
- You’d better know how to route a ground for lowest noise.
- You’d better know what happens when your transformer field gets too close to your input lines.
- You’d better know about the thermal consequences of spacing products too close together.
- You’d better know what’s realistic to fit on a given size board.
- You’d better know how to run output lines, or when it’s better to go to wiring to get it off the board.
- You’d better know how to deal with mixed AC and DC lines.
- You’d better know how to deal with mixed digital and analog circuits (especially these days.)
Where do you start?
Best thing to do is probably to download Kicad, pull up a good schematic from DIYAudio.com (careful it isn’t commercial or copyright protected), and lay out a board. See if it works. Solicit feedback from the DIYaudio community.
Beyond that, there’s this amazing thing called “Google,” into which you can type search queries like “how to design PCB tutorial” and get a whole lot of results. Unfortunately, I haven’t used any of these results, so I can’t comment on how good or bad they are. Sorry. I did tell you that anyone else on the planet would be better at DIY.
“Okay, fine. I’ve downloaded KiCad. I have a schematic. What do I do now, chief?”
Oh boy. Your choices have only just begun. Because you really should know a few things before you start. Or you should at least have some good guesses. Here’s what you should know:
- What’s the chassis look like? It’s gonna suck when you deliver a beautiful 10.5” x 5” PC board…for a 9” x 6” chassis. It’s also gonna suck if your designer assumed the inputs would be on the back and the outputs would be on the front, and you ran them all along one side. It’s also gonna suck if it fits perfectly…but the board won’t fit into the chassis because the capacitors are too tall. Or they don’t fit under a rail. Or the input and output connectors assume you can push them through solid metal.
a. General rule: don’t start on the board unless you know what the chassis looks like, and where you expect the I/O to be.
b. Getting specific: Of course, we get a free pass on Vali Mini, because it’s supposed to be a coaster, no chassis at all. But even then, I figured that a 6” diameter coaster would be too big, and that people would want power on the back and I/O on the front.
- How much heat does this sucker have to dissipate? If you’re doing a simple op-amp based preamp, you may not need to worry about how hot the product is going to get, but this isn’t usual. Hell, computers are constrained by how much heat they can get out of the processor. If you’re going to be powering headphones or speakers, you’re probably looking at dissipating enough power to worry about. Do you need heatsinks? Do you need to use the chassis as a heatsink? Do you need thermal pads or gap pads? Can you use the board as a heatsink? How close can hot devices get to each other?
a. General rule: don’t start on the board unless you know how much power you have to get rid of., and what options you have to get rid of it.
b. Getting specific: Vali Mini is low-power enough to use the PC board as a heatsink.
- What’s my cost constraint? If there aren’t any cost constraints, ima doin every damn board as 10-layer, 0.093 thick, ENIG, 4/2 ox copper. Confused by this engineerese? It’s worth learning. I’ll disambiguate a bit below.
a. General rule: the more layers and the thicker the copper, the more expensive the board. 2 layer/1 oz copper boards are kinda the de facto starting point these days. These boards can be so cheap that it’s less expensive making 5-10 pieces than it costs to ship DHL from China. When you start talking more layers and thicker copper, costs can add up fast.
b. General rule: the thicker or thinner the board, relative to 0.062” or 0.047” (depending on manufacturer), the higher the cost. 0.062” boards are the de facto standard, but the thinner 0.047” boards are commonly available at similar cost (and if you are wondering why you’d want a thinner board, you haven’t done a product with a thickness constraint.)
c. General rule: the more exotic the processing, the higher the board cost. HASL (hot air solder leveling) is a cheap process that is used to tin the boards, and is pretty much the de facto (cheap) standard. ENIG (electroless nickel immersion gold) makes nice gold-colored PC boards (and can result in significantly higher reliability when you’re using exotic parts with tiny pads and buried thermal lands).
d. Getting specific: Vali Mini is going with 2-layer boards, 1 oz copper, 0.062”, ENIG. We’re using the more exotic board processing because it looks better, and it’s going to be safer around food. In practice, most Schiit boards are 4-layer, 2-oz copper, 0.062-0.093”, and HASL (most analog) or ENIG (most digital).
Okay. Got your schematic? Got your PC board layout software? Got your chassis design, heat production, and cost constraints? Now you’re ready to do your PC board.
And even then, there are choices.
- Do you put parts on only one side or both (both costs more)?
- How many through-hole parts do you use (through-hole costs more to handle, but can be more mechanically robust)?
- How critical is the layout to performance (does each side need to be exactly the same?)
- How cool do you want it to look (yes, this is a choice, and it can bite you in the butt)?
- What color is the PC board (yeah, not a huge deal, but it’s a choice)?
For Vali Mini, we’re putting parts only on one side (it’s a friggin coaster, after all). And we’re using a pretty large number of through-hole parts, including the electrolytic capacitors, I/O connectors, the power switch, and, of course, the tubes. For the audio band, the layout shouldn’t be super critical, but that doesn’t matter, we always make the two channels as close as possible to matching. “Cool looking” was a design criterion, so we’re going to take extra effort to make sure the product is as symmetrical and clean as possible.
Aside: and this is what might have caused the oscillation…if it was oscillation. Because “looking cool” is rarely “most efficient.”
And, of course, the PC board would be red, because all of our boards are red.
Vali Mini PC Board Layout and First Build
In terms of layout, I’m afraid Vali Mini was a bit of a snoozer. It’s an easy board, especially after doing (redacted). There’s plenty of space. There are no huge thermal concerns. Really, the only fun came from:
- I started with a 4” diameter board, and kept making it smaller as I made the layout more efficient. Although there’s no standard coaster size, we don’t need a combined coaster/Frisbee. So I’d finish the layout, then pull the diameter smaller, then do it again. Despite this, it really only took a couple of hours to lay out.
- I actually started with an earlier design that had a buffer, not an inverter, coupling the tube and transistor output stage. This wouldn’t have worked. That’s all right. I finished the design and sent it out for prototype before realizing this. $32 down the drain (did I mention, prototypes are insanely cheap.)
- Even after realizing my mistake and re-laying-out the board for the correct schematic, I boned the power supply. I screwed up the connection between the two regulators, so the power supply didn’t work. The prototype I’m currently running has a couple of board hacks.
And, of course, I have the weird noise problem. More on that as I learn more.
Building the prototype was also straightforward. As I mentioned, I used a lot of parts we already have lying around Schiit. Capacitors, tubes, connectors, resistors…most of that was standard.
In fact, it was actually kinda nice to build the prototype. I’d purposely chosen “0805 (hand soldering)” and similar library objects when building the board. In other words, I knew that these boards might be hand-assembled, so I used components built purposely with larger pads so they were easy to handle. If you look at an 0805 component on one of our production boards, you won’t even see the pads underneath it. A “hand soldering” version is three times as big.
So the boards were easy to assemble.
And, hell, to get right down to it, they were
assemblable. We’re using enough exotic parts (3x3mm devices with 14 irregularly shaped pads underneath, stuff with hidden thermal lands, QFNs, hell, even 100-pin 0.5mm spacing devices are no picnic) that many of our prototypes are now assembled by our PCB house, so we don’t touch them. Or, we do and there’s a lot of cussing involved.
So the board flew together. It was a simple design. Simple assembly. It should work just fine, right?
Nope.
Hell, it didn’t work at all. No LEDs. Nothing glowing. Dead. Of course, nothing was smoking, either, which you gotta count as a plus. But still, not what I expected.
Aside: powering up a design for the first time? Use a variac. Monitor the voltages. Go slow. Watch for smoke. Vali Mini is a pretty low-power design, and pretty safe, but that’s still what I do for everything. Do this 10000X more seriously with high-power designs. Turning on a speaker power amp design for the first time can be a hair-raising experience.
A bit of cautious probing with a multimeter revealed that the main power supply wasn’t working at all—not before or after the regulators. Whut? That didn’t make any sense. I mean, this was a well-known schematic, with well-known parts, what could possibly go wrong…
What was wrong? Well, how about the schematic. When you ground one half of the AC input coming into a voltage doubler, well, that ain’t no good. And when you screw up the library object for the DPAK LM317, then yep, that don’t work either.
A few hacks and the board was up and running.
Or, well, limping. The DC point at the tube anode was like 30V, or only about a 6V drop. I was looking for about 10V, so I’d be about midpoint of the 18V rail after the inverter (after accounting for the 0.6V drop).
So, I swapped out the 10K resistor for 11.3K. Closer, but no ceegar. 12.7K? Nope. 15K was the magic point, which gave me about 9.9V on one side, and 9.5V on the other.
Aside: tubes won’t be the same. Get over it. Horseshoes and hand grenades. Yep.
Aside to the aside: want to do this design with different tubes, like a 6418 or a Russian rod pentode? Expect to be changing the anode load to get the right operating point. 6418s might be a very good choice, since they like lower anode voltages. Sorry, didn’t try them, so I can’t comment definitively.
Finally, a brief listen. Sounded fine on some MrSpeakers Ether Flows. Very reminiscent of the original Vali—really vivid, 3D, maybe even a bit too enthusiastic on the high end. And, of course, with ever bump of the board….tinnggggggggggggggggggggggggg…
And that’s about all the time I got with it, because, well, this was a play project. There were other things that needed to be done. I didn’t even spend any time measuring it, other than to confirm where it clipped (about 15V p-p, with noticeable rounding on the bottom half of the sine wave, not surprising for this kind of design).
That was it. I put it on the shelf above my desk and moved on.
Until yesterday, when I made some measurements…and found some unexpected behavior.
Again: nothing is
ever easy.
Nothing.
I’ll report on what went wrong in the next chapter.
