Separate names with a comma.
2018, Chapter 3:
Engineering, Part 2
If you’re just jumping into this chapter without reading Part 1, you may end up being, well, completely and utterly lost.
Because this is the second chapter in (what looks to be) a 4-chapter series. This series covers the design and production of a novelty product: the Vali Mini, which is a PC Board coaster, that can also be built into a simple hybrid amplifier.
As of the time of this writing, it looks like the chapters break down like this:
Part 1: Origins of the Project; Defining Goals, Features, and Expectations (link)
Part 2: Electronic Design, Detailed Discussion (that’s this chapter)
Part 3: Building the PCB and Design Iteration
Part 4: Measurement, Testing, and Release
If you’re looking for Part 3 and 4, they don’t exist yet. So don’t get too excited; we have a ways to run on this. And it looks like we’re going to have an interruption or two, in the form of new product chapters, along the way.
And, like everything else that ain’t done, plans might change. We might get an extra chapter, or the content of the chapters might shift.
Hell, I thought I was pretty much done with the design of a new product just a few days ago…except for a nagging question that everyone kept asking, and a desire to actually, well and truly put some of the shortcomings of the previous product to bed. Due to this, and a busy weekend checking out some new mechanical ideas, we’re gonna have a much better product…but it’s gonna take throwing the current chassis prototype away, and some significant changes on the PC board.
Don’t worry; I’m sure I’ll cover this upset in a future chapter, and you’ll understand why I decided to make some fairly major changes to something that looked mechanically done.
But I’m procrastinating. Let’s do a quick summary of the goals, features, and expectations from the last chapter, then let’s get into the nitty-gritty of design engineering.
Summary: Goals, Features, Expectations
Primary Goal: To design a functional tube hybrid headphone amplifier that fits on a coaster-sized round PC board.
PC board usable as a cool-looking coaster without adding any parts to it.
Documentation provided for moderate to advanced DIYers who want to build a functioning amp with the PCB
Bill of Materials
Low cost to build into a functional product
Uses easily available parts
Safe to use and operate
Features: This thing needs to be a functional headphone amp, so it needs some bare minimum of features in order to operate:
One line input
One headphone output
A volume control
As easy to use and safe power connector
A power switch (not necessary, but nice to have)
0.5-1% THD+N at 1VRMS out into 300 ohms
400-500mW full-scale output at clipping into 32 ohms with several percent THD
90-95dB S/N ratio, unweighted, referenced to 1V
Aaaaand, just in case anyone has forgotten: Yes, I know this is a low-performance product. Go back and review the previous chapter if you’re confused as to why.
Gain and Output Stage Design
Okay, now that we’ve decided what we want to make, how do we get there?
The answer to this big open-ended question might be very different for other amplifier types, but in the case of the constraints we have—a simple, safe tube hybrid using commonly available parts—the best place to start is figuring out how we’re going to get the voltage gain and current gain we need to run headphones. In other words, it’s best to dive right into gain stage and output stage design.
“Wait a sec,” you might be saying. “I remember Electronics 251 lab from college, where we took some transistors, biased them up, and got voltage gain out of them. Is that what you’re talking about?”
Great question, because it allows me to discuss the complexity of the real world, versus the simplicity of the lab or the engineering textbook.
If you came into electronics like I did, you probably started with a vague notion of a transistor as a kind of “valve” that allowed you to control one voltage with another. (Never mind that a transistor controls current with current input.) Then you get in the lab, and you see you actually have to use a couple of resistors and a coupling capacitor to set the operational point at the base, and then choose (carefully) the resistor in the emitter to set standing current, and then choose the collector resistor to set gain, and then decide if you need to bypass the emitter to get more gain and if you need to have a capacitor and bleed resistor at the output to get rid of DC offset, and your head starts spinning…
…and then you realize, even with all those extra resistors and capacitors, you still don’t have a functional amplifier that can drive anything like a speaker, because the output impedance at the collector is too high to really run anything, so you have to add an emitter follower, but even then you may need to start looking at negative feedback to get a usable gain range, and even then you might want to look at bipolar supplies to get rid of at least some coupling caps, but that might change the gain of the front end, and where do you put the feedback anyway, and suddenly you start realizing why commercial designs end up using symmetrical topologies with dozens of transistors instead of those simple one- and two-transistor circuits…and then you begin wondering what kind of performance improvements you might see with even more complex topologies…
…and then you realize that JFETs and MOSFETs are different types of devices than transistors, and have different biasing requirements, and might be simpler (or not), and that tubes have no actual physical connection between elements, which makes them a whole different ballgame even before you get to the heater, and why the British call them “valves” and why that makes more sense than “tubes”…
Whew. A little too much there. Let’s break it down.
To run a set of headphones at anything more than line level*, you need both voltage gain and current gain.
Pretty much any kind of device can do voltage gain (tubes, transistors, JFETs, MOSFETs)
However, tubes and modern JFETs are not good at current gain**
BJTs and MOSFETs are good at current gain
Hence, the most logical arrangement for voltage gain and current gain in a tube hybrid amp will be:
Use the tubes for voltage gain
Use BJTs or MOSFETs for current gain
*Line level is typically 2Vrms, though some phones and other sources may only output 1-1.5Vrms. To convert Vrms to peak-to-peak volts, multiply by 2.83. So 2Vrms is 5.86Vpp. This will be helpful in determining the voltage rails and gains we’ll be using in the future.
**Now, some of you are probably howling, saying, “Well, that ain’t exactly true, there are tubes that run quite a bit of output current like the 6AS7 and there were big JFETs like the old Sony stuff from the glory days.” Yes. And a pair of 6AS7s requires 5 AMPS of heater current to run—as in, 30 WATTS just from the heaters. This ain’t gonna fit on a coaster. Valhalla 2 is especially designed to provide a lot of current from tube outputs, using a White Cathode Follower arrangement (look it up), with super-robust 6N6P tubes that will dissipate 8W on the plate, but it’s still 5X weaker than our next most current-limited amp, Asgard 2. And big JFETs? Yeah, there were some. Not any more. JFETs are tiny, surface-mount parts these days. Nothing wrong with that—just don’t expect them to run speakers.
Aside: have no clue what I’m talking about? Go here to start understanding tubes: http://www.valvewizard.co.uk, then go to https://frank.pocnet.net for more tube datasheets than you can shake a stick at, then to http://www.tubecad.com to blow your mind, then to http://www.surplussales.com/homenew.html#Vacuum-Tubes for inexpensive tubes.
So how do we design a tube voltage gain stage?
Pick the tube type you want to use
Choose a topology that achieves your goals.
Set voltage gain.
Let’s run down the list.
One: Pick the Tube Type. If you’ve worked with tubes before, you know that the “safe” qualifier means we’re very limited on what we can choose. You know that most tubes are going to use 300mA or more of heater current at 6.3V, and will want plate voltages at 100V or (much) more. High voltages aren’t safe. Even the 2W of heater dissipation might be considered too warm for an open-frame design. So, you’re really down to two choices:
Use low-voltage, low-heater voltage subminiature pentodes (or Russian rod pentodes, similar idea) strapped for triode operation.
Cross your fingers and hope the 2W of heater dissipation is OK, and go with a starved-plate design and a tube that’s forgiving of low plate voltages, like a 12AU7.
So how do we choose? Let’s weigh the pluses and minuses:
Availability: Submini pentodes are still very easily purchasable on Ebay. 12AU7s are available at moderate cost for new production tubes. Wash.
Performance: Spitballing, but the 12AU7 may edge the submini even at low plate voltages. The subminis could be run much closer to their optimal operating point. Still, neither are gonna be super-linear. Slight advantage to 12AU7. Maybe
Heat: Submini pentodes use only 20-40mA heater current at 1.25V (as in, nearly nothing). 12AU7 will be about 2W from a 6.3VAC source. More if you go 6.3V DC. You’d really want a more complex power supply for something like this. Might be too complex. Big win for submini pentodes.
And that’s it. Heat—and power supply considerations—means we’re going with the subminiature pentodes. More on this when we get into power supply design. In the meantime, enjoy the 6088 pentode and triode curves:
Aside: And yep, we’re designing the power supply AFTER we nail down the gain stage. Because this is a tube hybrid, this approach makes sense. For other amps, you might want to start with the power supply you have, then think about gain stage.
Two: Choose a Tube Topology. Tubes can be run in lots of different topologies, if you have enough of them. The Valhalla 2 uses a cathode-coupled approach to get a noninverting design and to provide a convenient terminal for feedback in low gain mode. But when you need to:
minimize the number of tubes you’re using,
have voltage gain,
deal with direct-heated tubes like subminiature pentodes,
getting fancy ain’t the best approach. Nope. What you want is a simple, common-cathode gain stage. And when I say “simple,” I mean, “simple.” No active loads, no feedback, just a resistor load to set the operating point.
Three: Set Voltage Gain. Tubes don’t have anywhere near as much gain as transistors or FETs. Pentodes have more gain, but we’re actually running these submini pentodes as triodes. Why? Because they perform better as triodes than pentodes. Look at the weird kink in the pentode curves. They’re nothing to write home about as triodes, but they’re better.
With most tubes, you’ll have some leeway in setting gain by choosing plate and cathode loads, but in this case, we want to run the tubes pretty much full out.
Gain, in this case, is determined by setting the operating point at about 2/3 the available supply voltage with a 10K resistor.
This gives a gain of about 4.
You can change the gain by changing the plate resistor, but it also changes the operating point. More on this later.
Aside: sorry, no load lines here. Want to play with load lines? Go here: http://www.trioda.com/tools/triode.html
Cool. So now we have a tube amp? No. Not hardly. The tiny subminiature pentode has a huge output impedance. It can’t drive anything. Not to worry. We’ll add an output stage and then figure out how to put it all together.
So how do we design an output stage?
Pick the type of device you want to use.
Pick the topology that achieves your goals.
Set output bias.
Let’s run down the list:
One: Pick the Device Type. BJT or MOSFET? That’s the real question. Those are the two devices that do the heavy lifting, when it comes to current output. Many designers have a preference for one or another. My own preferences have been, ah, evolving. Let’s review BJTs versus MOSFETs before we get into the pluses and minuses of each.
BJTs, or Bipolar Junction Transistors, are current-input, current-output devices. They have current gain, or beta, that ranges from 50-100 (in large devices) to 250-750 (in smaller signal devices.). They also have a predictable voltage drop from base to emitter, of about 0.6V.
MOSFETs, or Metal Oxide Semiconductor Field Effect Transistors, are voltage-input, current-output devices. They have a less predictable voltage drop from gate to source (2-4V), or, if they are “depletion mode” devices, they actually flow current with no voltage at the gate, like a JFET. But you ain’t gonna find a depletion mode P-channel, so that’s out.
So how do we choose? Let’s weigh the pluses and minuses:
Availability: BJTs and MOSFETs are both available in the sizes we plan to use for this project. Wash.
Performance: Probably similar in this application (no-feedback simple Class AB gain stage). Wash.
Integration: MOSFETs don’t require much of our limited current. BJTs do. MOSFETs do have significant gate capacitance, but at the size of device we’re using, they’re still easier to drive. Advantage MOSFET.
Biasing: BJTs much more predictable, with known voltage drop. Much easier to bias without having to twiddle pots or use bias servos. Advantage BJT.
As you can see, there are advantages and disadvantages to both. I ended up picking BJTs for the output of this device, because the simpler biasing means a simpler product—and simple is one of our goals.
Getting specific, I picked two SOT-89 parts: 2SC3648 and 2SA1419. More here: http://www.onsemi.com/pub/Collateral/EN1788-D.PDF
Why this size? Well, it’s a relatively small board, so these are relatively small transistors…but transistors with a thermal pad that allows us to put some copper underneath them to spread the heat and keep overall temperature down. That’s nice.
But we could pick pretty much any SOT-89 size pair of NPN and PNP transistors. Why these, and not, say, BCX53/56? In this case, mainly personal preference and familiarity. We use the 2SC3648/SA1419 pair in a number of products, including Vidar and Magni 3, and I like their performance. Comparing to the BCX series, the f(t) and Cob are similar and hfe-ic curves are both nice and flat in their operating region, so both are fine choices. You could certainly use the BCX parts, or pick another pair entirely, and the amp will likely work just fine.
Two: Pick the Topology. This is where your options really expand towards infinity. Do we do a single BJT follower with resistor load, or go wild and do a triple darlington output stage? Or an output stage with gain? Or a (redacted)?
Well, let’s examine the pluses and minuses again of a few of them.
Single BJT Follower? Now you’re locked into a Class A design—one with limited output current. Yes, you can add a current source to get the efficiency up, but you’re still not going to run much current, and your overall output drive will be limited. Too simple.
Double Darlington Push-Pull? Now here’s a classic Class AB design, providing plenty of current gain for most applications. But it’s really more than we need. The output transistors we’re planning to use have plenty of current gain. We’re not planning on running super-hard-to-drive headphones. Too much.
Single Push-Pull? This is as simple as you can get with Class AB. We do want simple. It isn’t high-performance, but neither are our design goals. Sold.
Three: Set the Bias. Again, there are tons of ways to set output bias on a BJT design. Diodes, Vbe Multipliers (look it up), even more complex bias servos. In this case, we could go with a couple of diodes and small-ish emitter resistors to set bias.
But I wouldn’t be fully comfortable that the small emitter resistors would fully protect the small output devices we’re using. So we’ll do something a bit more fun—use an LED.
An amber LED simulates tube glow, and provides about a 1.8V drop. Subtracting the 1.2V drop across the two transistor emitters, and you get 0.6V across 20 ohms, or about 30mA output bias. Small, but sufficient for this design.
Whew! “Are we there yet?” a lot of you are asking. Do we have a functional amp yet?
Nope. All we have is a voltage gain stage and an output stage. We don’t even have the glue to bind them together. Because they don’t really work all that hot on a DC basis. In fact, you really don’t want those tiny SOT-89 transistors running across the entire 35-40V main rail for the tube, because even with copper under them, they may end up getting pretty hot.
So how do you glue this mess together? Here’s where the art really starts.
We could just plop the LED across the tube anode, and use that to run the output stage. But the tube is running very low current. The outputs need current to run, so you run the real chance of the output stage starving the voltage gain. Plus, even if you bias the tube such that it runs about midway between the rail, you’re still running the output across the entire 35-40V. Nope.
We could add an emitter follower, and run it at higher current. This doesn’t take many parts. But again, you’re running the output across the whole rail. And, when you get right down to it, it’s probably best to bias the tube so it’s not sitting midway between the rail and ground, because the tube will perform best with more voltage on it.
We could capacitor-couple the output of the tube to a resistor-biased output stage. But this is a BJT output stage, so its input impedance is nonzero, and might load down the gain stage. Plus, now you have a capacitor in the middle of things.
Or, we could get tricky, and use an inverter to level-shift the output, provide direct coupling for a lower output rail, and preserve absolute phase.
In other words, we use a transistor to invert the output of the tube (while running more current to help the output stage). At the same time, our relative DC level shifts from 27V to 9V, which is a perfect for, say, a 18V output rail.
Aside: The original Vali used a trick like this, together with an additional buffer, to further improve performance. We’re leaving the buffer out here just to keep things simple. The new Vali 2 is a whole different ballgame, using a simplified version of what we’ll soon be calling the Coherence™ hybrid topology, where we use a current mirror to combine the tube with a PNP transistor.
Still confused? Here’s what we ended up with:
Note that we’re specifying two voltage rails, but they are not complementary. Also note there’s another 1.25V supply in there. Which makes this a great place to segue into…
Power Supply Design
And here’s a whole new field to play in, one that’s nearly as big as the design of a power amplifier. Power supplies can be as simple, or as complex, as you’d like. You could decide to run everything off a 5V USB wall-wart (but not this design, which requires 40V and 20V rails, at least not without step-up switching supplies), or you could go full-boat crazy and do a linear, choke-input, discrete shunt-regulated monster.
Here, we’ll go back to that one damning word: simple.
Simple means that step-up switchers are outside the scope of this design. Simple means we’re not going to be talking about discrete regulation. Simple means we’re going to be sticking to easy, tried and true stuff. Or, in other words, boring.
The first question is how we get the power from the wall. (Because, you know, simple—no lithium-ion charge management here.) And here we’ll also pull out another word: safe.
Safe pretty much means a wall-wart, because they already have safety certifications in place, and they’re never going to output enough voltage to hurt someone
But, should we go with a DC wall-wart or AC wall-wart?
A DC wall-wart is probably not the best place to start. Even if you could find a 36V DC wall-wart, you’d need to regulate that 36V down to 18V for the output stage, and that means you’re gonna be running 60mA x 18V all the time, or 1.1W, which is a lot of power to burn simply to provide a lower-voltage rail.
An AC wall-wart is more flexible, because AC can be rectified on the board. It can then be doubled…and the doubled voltage can also be engineered to have a half-voltage point, and both can be stacked on top of ground to create the kind of power supply we need.
So, we’re going with AC.
Handily, Schiit has a number of AC wall-warts to choose from. We’re picking the 16VAC version. If you build this, it’ll work happily with a wide variety of wall-warts from 15-18VAC. Don’t go much higher or you’re gonna be burning stuff up. Don’t go too much lower, or you’ll be falling out of regulation.
Okay, so we got this 16VAC. If you know much about power supply design, you know that you should expect about 1.4 x 16V after rectification, or about 22V.
How do you get 36V from 22V? Easy. By using half-wave rectification, or, in other words, a simple voltage doubler. Now you’re looking at 44V.
And, we’ll get trickier. We’ll stack two capacitors across that 44V to get 44V and 22V rails, for the tube and output stage respectively. Then we need to regulate.
“Regulate, why?” some of you are asking. “I see plenty of amps running on unregulated rails.”
Yeah. Sure. Fully complementary amplifiers with high PSRR (power supply rejection ratio). This amp has a PSRR on the tube rail of, effectively, zero. It needs to have regulated supplies to get rid of all the ripple, especially from a half-wave rectified supply.
Aside: look up the pluses and minuses of full-wave and half-wave rectified power supplies. You’ll soon see that most people discount half-wave stuff. They’re missing out. Hell, Magni is half-wave.
So, what regulators? Let’s be boring. We can use bog-standard LM317 regulators for both rails. Just stack them. Hell, use the same feedback parts.
Not very exciting? Better parts available these days, you think. Yes. There are. Heck, following the LM317s with nothing more than a capacitance multiplier will dramatically reduce their noise floor. Yep. Gotcha.
Not only that, we can use one more LM317 with no feedback to create that 1.25V rail for the tube heaters. Yep. Because the LM317 reference voltage is 1.25V. Convenient, right?
Well, except for one thing: you probably want to filter the output of that LM317 for the heaters, because it is pretty noisy. Best to filter each one separately, unless you want your channel separation to go to heck. You’re going to be seeing audio on each tube’s cathode, after all. So, in this case, we do a simple RC filter at each tube.
And there you go. One power supply with 36V, 18V, and 1.25V regulated rails, from a 16VAC wall-wart plus 17 parts.
Putting it All Together
“Okay, so are we done now?” you might ask.
Not quite. We have an amplifier and power supply, but how do we get the signal on and off the board? How do we protect the headphone from start-up blips (which can be very large on this kind of amplifier)? How do we control volume? How about power?
Sound simple? Hmm. Maybe not so much.
Let’s start with the volume control. Should we go with a microprocessor-controlled IC volume control? A relay-switched stepped attenuator? An Alps “Blue Velvet” pot?
In short: none of the above. The first two are wayyyyy too complicated, and the last is physically quite large. What we’re looking for is a small, good potentiometer. An Alps RK09 (like we use on Magni) is just the ticket. In this case, we’ll choose an off-the-shelf part, but ideally we’d really like to see something custom, with a 10A taper, which allows even more gradual increase of volume than the standard 15A taper.
Aside: look up Alps potentiometers and the available tapers to see what I’m talking about…and to see how the simplest decisions can be, well, less than obvious.
And, just because we can, we’ll go with a vertical potentiometer, so you can run a volume knob in the middle of the board, kinda like a Fulla 2.
Next, let’s talk connectors.
RCA input? Nope. Let’s keep this really simple and go with a 1/8” stereo connector. Yeah, I can hear the groaning already. But it’s simple enough to get a 1/8” to 1/8” or dual RCA to 1/8” cable. And if it’s really that problematic, best to simply use this product as a coaster.
¼” output? Nope. Let’s keep this really really simple and do 1/8” again. Yes, I know, everything we do is ¼”, but they are physically large, and we didn’t want this to be a coaster only for Big Gulps.
Power? A DC barrel connector with 5.5mm barrel and 2.1mm pin works with our standard wall-wart. If you use something else, make sure it fits your wall-wart.
Okay, how about a power switch?
Sure, let’s do a simple vertical toggle. I’ll place it near the “back” of the board so it’ll feel familiar
And finally, headphone protection. This is another subject that can get very complicated, very fast. We can go completely bonkers on protection, like the analog-computer-style protection system used by Mjolnir 2 to calculate DC and power output and make sure everything is cool, or the microprocessor oversight of Ragnarok and Vidar, which are even more sophisticated.
What we really need to protect from is turn-on pulses. To do this, the easiest and simplest way is with an output relay and a time delay. Is it perfect? No. But it’s a great 98% solution. So that’s what we’re going with.
Aaaaaannndd….now we have an amp.
Next chapter: building the PCB. Thanks again for reading!
Aaaaaannndd….now we have an amp.
Next chapter: building the PCB. Thanks again for reading![/QUOTE]
Where is the volume control?
Where is the volume control?[/QUOTE]
Never mind, I see it now. Not a symbol I am used to and I didn't expect it by the input jack.
WOT....????? Is that it....????
Did you all notice the redacted option? Stand by fir more on that.
Oh and where does that 23volts come from for the protection circuit?
Two and a half 9 volt batteries...........
And that 1/2 battery would be the kicker, and could give those Lion battery fires a run I'll bet… hahahahahahahahaha JJ
Berlin is of course the worst candidate possible.
But only a European would understand that.
Besides, the European distributor is in Zoetermeer, The Netherlands. 5km from Delft.
On the contrary.
We doctors are trained 8 years to write unreadable.
Oh yes, Akkerman. the pen lovers heaven (or Nirvana, to stay with accepted terminology here). Man how I miss that shop living in the Middle East now....
AND readable, unlike any doctor I have ever met.