2017, Chapter 8:
The Vidar Chronicles, Part 2
For those of you who’ve been following the Vidar updates breathlessly, waiting for its release, let me cut to the chase: nope, Vidar won’t be shipping this month.
Yes, I know. Huge shock. Especially considering there are only a few days left in this month.
Next month is another story. All the gears are turning, and we are simply waiting for a few critical parts to be able to ship. Most critical are heatsinks, which are currently scheduled for June 13. They’re late because of a debacle that is becoming all-too-familiar with some USA-based companies, which I’ll get into later. Also critical are the custom Sorbothane washers that isolate the transformer from the chassis and keep Vidar from humming like a refrigerator. Those will come before the heatsinks. We’re also still waiting for the output binding posts, but those are relatively minor—we can steal from Ragnarok to get the run started if necessary.
So now it’s a waiting game. Will we be able to ship in June? Almost certainly. Will it be June 13, the day the heatsinks arrive? Certainly not. Will we ship 50, 500, or 5000 amps by the end of the month? Maybe, probably not, and certainly not are the answers on that one.
But we’ll see. Maybe this will slop into July. Maybe aliens will invade Santa Clarita, and we’ll be pressed into service making Glorbsplort Hyperblasters. Maybe Google will offer a billion dollars to buy us out. Maybes are listed in increasing levels of improbability.
In any case, you’ll soon be seeing a shipping Vidar, so I thought it would be good to catch you up on what’s been happening with it. If you haven’t already read it, you may want to go back and have a look at
https://www.head-fi.org/f/threads/s...obable-start-up.701900/page-926#post-12997153 to see what a saga it’s been to date. In short, we’ve thrown away several complete, working prototype approaches already, in a quest for truly over-the-top performance.
Where I left it, in the last chapter, we’d decided to go to external heatsinks. I’ve posted a couple of informal updates about our progress on that, but let’s start with a recap of where we are, thermally…
Heatsink Follies and Clip Madness
It really wasn’t that hard to find a heatsink extrusion that worked externally and didn’t cost a fortune—and, ironically, it turned out to be the same heatsink extrusion I used internally in the Ulysses and Ulysses 2 power amplifiers for Sumo! I didn’t realize it was the same until the samples arrived. The heatsinks looked awfully familiar…and, compared to the Antares (a Sumo integrated that never made production) heatsink, it was soon clear why it was familiar—it was the same extrusion. A clear case of
great minds think alike (or
a foolish consistency is the hobgoblin of little minds, take your pick depending on how you feel about me.)
The new heatsinks were more expensive than the high-density versions I’d planned to use internally, but much of the cost ended up being offset by the simpler chassis. With huge, heavy heatsinks running from front to back along both sides of the chassis (and when I mean heavy, I mean heavy, the base thickness of this heatsink is almost 3/8”), the bottom chassis got a whole lot simpler.
The amp also got a lot more sturdy overall. With the top and bottom both bolting directly to the heatsink, it really feels like a brick (or, “hewn from a single ingot,” in audiophile terms more appropriate to car-priced components.
So, with new heatsinks chosen, I embarked on redesigning the PC board. The old PC board was a large affair, essentially the size of the chassis. It also had two holes in it—one for the heatsinks, and one for the power transformer.
The damn thing was more air than board, I recall thinking. Not an efficient way to do things.
I fixed that with the new board. Since we had much larger heatsinks on either side of the chassis, I was able to spread out all the power devices, from VAS to drivers to outputs, without using heatspreaders (and offsetting more of the cost of the bigger heatsinks.) I also didn’t need to have a donut hole in the middle of the thing for heatsinks, which made for much more optimal layout of the power supplies and grounding.
Also, since the board shrank significantly, I was able to go from a 2-layer board to a 4-layer board, which also helped optimize the routing. This is particularly important on a power amplifier, since the high-current traces can interact with more sensitive parts of the circuit if they aren’t carefully routed.
So, with some blank heatsinks in hand, an unpainted chassis, and a hand-assembled prototype board, we set about making (what we thought was) the final prototype of Vidar.
Yeah, right.
Here’s the problem: there was no way to assemble the product in any reasonable length of time. I’d planned on screwing all the outputs, drivers, and VAS to the heatsink from the inside of the chassis, but the layout was so tight that I would have had to use non-ratcheting right-angled screwdrivers to make it happen. Plus, it would require 22 separate screws—any one of which, if not tightened down correctly, could cause problems on testing, or worse, on burn-in. Torque-settable right-angled screwdrivers had exactly zero chance of fitting into the chassis.
So, what to do? There were too many fins to simply space the outputs between them and drill through (and then you’d be looking at 22 screws and 22 nuts…ouch.)
I really had two choices:
- Design an output board that bolted to the heatsink directly and connected to an internal motherboard with some kind of board-to-board connector. This would be a huge departure, though. Also, board-to-board connectors come in two types: inexpensive ones that suck, and expensive ones that also suck.
- Design a spring steel clip to hold the outputs in their current locations, and eliminate the threaded holes for each output device. This would really simplify assembly, but we’d never done a clip like that before. Never-done-before and critical-product usually aren’t a recipe for a good night’s sleep.
I decided to go with the second option—mainly because we had a local stamping company that worked with spring steel, and I could work with them directly.
And it was the right decision. In just a couple of weeks, I had a prototype of the clip and was able to test it on Vidar. It worked great. I asked for some changes the stamping company should have charged me for, because they were my mistake, and they did them gratis. A few more weeks and we had final clips, and we were off to production.
Or so I thought.
We were still waiting for the revised first article heatsinks—the ones that fit the clips. And waiting. And waiting. After some bitching, we finally got our first articles, bolted the clips onto them, and verified that they worked.
Except I screwed up the location of one of the clip mounting holes.
No big deal, that’s what first articles are for, I thought. I contacted the heatsink company, said everything looked good except for the one dimension, and sent them a new drawing.
The next day, I had a response from the heatsink company…but it wasn’t what I expected.
Well, they said,
we can do that, but it’ll take longer and cost more, because we’ve already started the production of the parts.
What? Then what were the first articles for, I wondered. I also wondered about the clause in the PO that said, “Production to proceed upon approval of first articles.”
They insisted that this was how they did business, and cited the revision level of the part, etc. We dug in our heels. To make a long story short, we worked it out…but that “working out,” coupled with the waiting for first articles, lost us a couple of months.
Aside: And yes, I would go elsewhere for this heatsink, except that bearing the tooling cost for a nearly 13” wide extrusion doesn’t make me go wild with delight. It’s not out of the question, but first we’ll see how it goes with this supplier.
But, as I’d see, this wouldn’t be the first case of stupidity from a US manufacturer. Sadly, it seems that some really don’t care. More on this later.
A Review: Vidar’s Total Insanity Power Supply
In the process of laying out the Vidar PCB, I shared some of the design challenges with a friend, a person who’s done amp design in the past. I did this because he couldn’t figure out why I was having so much trouble getting it right.
But when he understood what I was doing, his eyes went big, and he said, “That’s total insanity!”
“On an amp of this price, yep,” I agreed.
“No. On any amp. You’re nuts!”
Now, he has his opinion, and I have mine. I think that Vidar’s power supply would be perfectly fine on a $5,000 amp (as sold typically at retail.) But he does have a point. It’s way above and beyond what most people will expect…and far, far beyond even the pricey amps we did at Sumo.
You see, a speaker power amp is usually a fairly simple beast, at least when it comes to the power supply. Give it a couple of full-wave rectified rails with a couple of big capacitors shared between the channels, at a voltage that allows you to reach your power output (after the losses through the predriver, driver and VAS stages), and you’re good. Not a regulator in sight, no worries about channel modulation of the power supply, not a big deal if it runs a bit hot because you have to throw more volts at it. A great example of this type of power supply would be on Sumo’s original Polaris amplifier. Pop the top and you’ll see the big rectifier, the two shared capacitors, and a nest of wiring leading to the output boards (single-sided, natch!)
Most amps get away with this kind of simple design, because most of the noise on the rails won’t make it to the speakers, and speakers are usually far away from the listener, and they usually aren’t that efficient anyway, so a bit of buzz isn’t a big deal.
The more sophisticated amps in the Sumo arsenal (we’re talking Andromeda II and later stuff, $1499 amps in 1990, or nearly $3K today) used a “boost” supply to increase the amount of voltage to the voltage gain and drivers. This “boost” supply was literally a boost…two extra windings on the transformer that were stacked on top of the main power supply rails. This helped the amp in a number of ways:
- It allowed the amplifier to swing all the way to the output rails, which reduced the voltage needed on the output rails for its rated power output.
- Lower volts meant cooler running, and kept the devices down deeper in their safe operating area.
- Clipping the output stage first is usually the most benign way to do it, so it sounded better when driven to its limit.
But a boost supply is still pretty damn primitive. It wasn’t regulated, so all the ripple of the main supply made its way to all parts of the circuit. So there was no better hum rejection. Worse, the boost supply was modulated by the main supply, since it literally rode on top of it. If the main supply took a 6V hit due to high power output (hitting a bass note at high volume, for example), the boost supply took that same hit. A boost supply, while an improvement, was still a sloppy mess.
For Vidar, as with all of our amps, I wanted a completely separate supply for the higher voltage to the front end voltage gain and driver stages—not one that rode on the output supply. This meant a more complex transformer with a larger core, since I was shooting for a regulated +/-62V supply. Regulation also meant heatsinking and more parts. But the payoff was a dead-quiet power supply that isn’t affected by what the output stage is doing.
And that’s only part of the insanity.
In addition to its separate supply for the voltage gain and drivers, Vidar also employs a “dual mono back to the transformer” design for the main high-current (+/-52V) rails. Four separate transformer secondaries are tied to 4 separate bridges, with 4 separate capacitors providing final filtering. Two each are used per channel to create the rails.
Why such a bizarre arrangement? For precision current measurement. More on this later. But the big deal is that each channel of Vidar has its own high-current power supply that goes all the way back to the transformer. Each channel has its own separate transformer windings. And so, where it’s most critical—where the power supply is bouncing around due to high current draw into difficult loads—neither channel can influence the other. Except for using a single transformer, it’s very close to a dual mono design.
In addition, Vidar needs housekeeping voltages, unlike a typical amp. It has a microprocessor to run. And relays. And op-amps for nulling DC. That means it also needs a completely separate +/- 15V and 5V supply. That’s another set of windings on the transformer. Which helps isolate the analog and digital sections of this amp.
So, to put it in context, my amp designer friend was expecting to see this for the Vidar power supply:
- 1 center-tapped transformer winding
- 1 bridge rectifier
- 2 capacitors
Instead, he saw this:
- 1 center-tapped transformer winding (for the HV regulated supply)
- 4 transformer windings (for the high current supply)
- 1 transformer winding (for the housekeeping stuff)
- 1 bridge for the HV supply
- 2 discrete-regulated HV supplies
- 8 capacitors in the HV supplies
- 4 bridges for the high-current supplies
- 4 capacitors for the high-current supplies
- 2 power diodes (for housekeeping rectification)
- 2 capacitors for the housekeeping supplies
- 3 separate regulated supplies for housekeeping functions
So, whew. Perhaps “total insanity” isn’t such a stretch.
More Weird Choice Gotchas
If the total insanity power supply isn’t enough, we also made a whole lot of rather unusual choices on the Vidar—many of which I didn’t want to talk about until I knew we’d be retaining them for production. Now that we have the final boards at the assembly house, let’s talk weird.
Weird #1: Surface-Mount Output Resistors. Sure, surface mount is fine for low-power use, but it isn’t something you see when you look at amplifier output resistors. If you’re familiar with amp design, you’re used to seeing large, “cement” style ceramic-cased resistors used in the output stage. And that’s exactly what Vidar started with. But during the course of the design, I started despising the “typical” output resistor. Here’s why:
- Even the “non-inductive” ones are inductive. Inductive enough to cause all sorts of weird aberrations in square-wave output testing—even without a load attached. Inductive output resistors are no bueno.
- Their specs are straight outta the 1970s. 5% tolerance? 10%? Really? Holy hell, did we get in a time machine and go back to Marantz or something? Wow, where are the polyester leisure suits?
- They are big and inconvenient—as thru-hole parts, they have to be hand-inserted, driving up the cost of production.
And that’s why I experimented with surface-mount output resistors. They’re flat, so they’re completely noninductive. 1% tolerance is easy to get. And they’re SMD, so the robots put them in, simplifying production.
Sounds great, right? Yeah, except for the fact that we have to parallel up a whole lot of parts to get to the dissipation levels we need. In early Vidar prototypes, we stacked three 1 ohm, 2W resistors to make a 0.33 ohm resistor. In the final version, we’re stacking four, for a 0.25 ohm resistor equivalent.
Why the change? Because, even though we have tons of dissipation at a 6W combined rating, we neglected the fact that these resistors get HOT. Hot enough to radiate enough heat to cause the thermal sensor to turn off prematurely (yes, you read that right—radiant heating from the output resistors drove the thermal sensor above 85 degrees C.) So, we added a fourth resistor to reduce the total heat, and added big lands to draw the heat away from the parts. (And moved the sensor a bit…)
And, in exchange, we get better matching in the output stage and resistors that are truly non-inductive…and that also keep costs down. Win, win, win.
Weird #2: Current Sensing Resistors. When we started the Vidar project, I had a complex way of measuring output current…which had to measure all outputs…on both sides…which led to a maze of traces, and the need for super-precision parts to reject common mode noise. That is, until Dave came up with a super-simple way to measure current using just two resistors per channel.
The catch? We would need to go to 4 separate transformer secondaries. Yeah, that’s where that came from.
But Dave’s solution offered simplicity in the amp’s most critical measurement, so that’s the way we went. And it worked.
Except…it made for a very bizarre failure mode. Shorting the output supply to ground (before it was drawn down by the bleeds, think “impatient engineer with prototype thermal clips”) could cause a resistor failure. Which left the supply floating with no reference to ground. Which would lead to other problems when the same engineer thought the supply was bled to nothing.
What we had to do was reduce the value of the measurement resistor (down to a vanishingly low 0.005 ohms) and double up the resistors to get the bulletproof dissipation we needed.
Aside: and, in case I didn’t mention this, the clips work spectacularly, when you’re not impatient and let the supplies bleed down. They allow repairs in minutes, since the whole amp can be stripped down so quickly.
So yeah, weird choices…weird solutions.
Avoiding The Labor Solution
You’ll notice that a whole lot of the Vidar chronicles are about reducing labor. Making it fast to assemble with clips. Keeping the chassis simple so it goes together easy. Using as many surface-mount parts as possible to increase the amount of automated assembly. This isn’t just a fetish or phobia—this is something we have to do, if we’re going to make something in the USA.
You see, if we were manufacturing in a low-labor-rate country, it’d be easy to simply throw more labor at it. This is
The Labor Solution, and this is what we want to avoid here, because labor is costly.
The old Sumo Polaris I mentioned before is a great example of a product that we wouldn’t want to do today. It didn’t even use a PC board for its power supply—each was hand-soldered using bus bars. It had two single-sided PC boards that were hand-stuffed and wave soldered. It was a ratsnest of wiring inside, with half a dozen connections to each circuit board. Its “protection system” was rail fuses (located inside the chassis, so if you dropped a channel, you’d have to disassemble the amp.) And, as stated before, it had the most simple power supply possible.
Aside: this was a $699 amp in 1990. Yes, same as Vidar is today in depreciated dollars. Let that sink in.
Vidar, in contrast, uses a single, 4-layer PCB with a super-sophisticated power supply, offers higher power output, has sophisticated intelligent oversight and protection, and absolutely minimizes any kind of rats nest of wires, as well as manual assembly. Vidar transformers are delivered with high-current pin connectors in place, ready to plug into the board. Vidar boards are delivered with the wiring to the outputs and to the single power LED in place. For assembly, you just drop the PCB in the chassis, solder the outputs, plug in the transformer, and drop on the heatsink/clip assembly. Done. It takes less than 1/20 of the time it took to put a Polaris 2 together to build a Vidar.
(And yeah, it’s been a long road, but we’ve learned a lot from this one…expect to see us apply a lot of this learning in other places.)
The Sad State of US Manufacturing, Part 2
I already told you the story of the Stupid Heatsink Supplier who Didn’t Know What First Articles Were For. Now, prepare yourself for the story of the Lazy Inductor Manufacturers.
One of the weird parts we needed for Vidar was an output inductor. This is a simple part: literally 8 turns of 14 ga enameled copper wire on a 0.5” center, with tinned leads. I’d been making the ones for the prototype by winding wire on an AAA battery. Close enough.
So, I never thought there’d be a problem getting an inductor like that made. We used them back in the Sumo days, so they had to be around, right?
Wrong. After a long search, I realized that I was looking at a custom part. But it was just a bit of wire. I still wasn’t worried. Or at least not until I started contacting manufacturers.
I did this like I normally do when we’re looking for a new product. I chose a half a dozen US-based manufacturers, and sent them off a simple message, describing the inductor in much the same way I did in the second paragraph of this section. I told them I wanted a quote for 10,000 and 50,000 pcs, figuring that we’d be using a lot of these, and that it couldn’t be that expensive (I figured about $0.40-0.50 in the smaller quantity.)
I only waited a day until I got the first response. Unfortunately, it wasn’t what I thought it would be. It was a guy asking about operating currents, frequency ranges, and self-resonant modes.
I responded that I’d described exactly what we needed, in terms of materials and everything, and all we needed was a quote.
He asked for a drawing.
Yes. A drawing. Of a coil of wire.
By this time, I was two days in, and I hadn’t heard from anyone else. So I went ahead and did the drawing. I also sent it to the other guys for good measure.
On Day 3, I got my first quote: $2.15 for 1000 pcs. Delivery in 8-12 weeks.
I emailed them back and said I was looking for 10,000 pcs, could they re-quote me for that?
They responded, saying the quote was still good for 10,000 pcs.
I sat and fumed, thinking,
No, that quote isn’t good at any more than about 100 pcs. So I tried contacting the other guys again. More silence. I was now 5 days in, and I still didn’t have a supplier. The clock was ticking on Vidar.
Silence.
Okay, fine. I decided to do something dumb: to see if I could get this from a Chinese manufacturer. After all, we use some Chinese parts in our products, most notably the wall-warts, and we’ve had reasonable luck with them. It was worth a shot.
So, I did exactly the same thing: sent a note to a half-dozen Chinese manufacturers, describing the inductor and the quantities. The only thing different was that they had the drawing from the start.
In 12 hours, I had 6 quotes, ranging from $0.08 to $0.22 each in 10K pc quantity. And they were quoting 2-3 week lead times, too.
Sigh.
So, to make a long story short, this is why the output inductor of Vidar is being made in China. While I did finally get some quotes from US companies at acceptable prices ($0.40-0.50), they came after 10 days, and they all had crazy lead times (like 6-12 weeks.) I’ll revisit them in the future, but for the first runs, it was way too easy to get the inductors from China.
And no, this is not an apologia or a justification for us to start moving all manufacturing to China. As usual, our chassis, transformers, boards, and PCB assembly are all done here. And they will remain that way.
But it does make you wonder…
