La Luciérnaga (Part 1, the PSU)

There is straight up a butt-load of single-ended amps out there in the design-o-sphere. The two-stage transformer-coupled amp approach sounds damn good and so its popularity shouldn’t be a surprise. With a little google fu, you can find a single-ended permutation for just about any common triode or pentode that can dissipate more than 10W.  

Here are some of my DIY favorites:

Cascade Tubes Lacewood 6V6

Mikael Abdellah SE KT88

Alex Kitic’s RH84

I thought long and hard (giggle) about publishing anything that skins this cat. The above schematics are relatively simple and sound great. They use CLC filters in the power supply for low ripple from the PSU. They use resistor loads for the input tubes to put them at tried-and-true operating points that you can easily find and tweak on a load line. They use some kind of feedback (plate to plate or ultralinear) to make the pentodes behave; alternatively they could be triode strapped if you prefer. There is nothing not to like about these circuits. They are different but successful answers to the same question.

But the rabbit hole beckons.

Single Ended Amps and Power Supplies

power and signal input

In a single-ended series-feed amplifier, we have a ground at the last cap of the power supply. Our AC output voltage appears between this ground reference and the anode of the output tube as it swings up and down against the impedance of the output transformer.  The power supply capacitor, as far as the output is concerned, is part of the AC output circuit. The same is true of other stages in the amplifier if they share the same power supply.

Ripple from our power supply also appears across this cap in parallel with the output transformer + tube (finding its ground reference at the cathode of the tube). Minimizing this ripple usually means a large capacitance part at the end of the power supply because this will shunt as much of the AC ripple as possible to ground (through the cap). Large cap values usually mean electrolytic types; these tend to be nonlinear with regards to frequency (Ref 1, Ref 2). Although ripple will not change in frequency much, the output signal, which also appears across this capacitor, does.

psu zout

Beyond the cap–further “up” the power supply–we’ll also have an output impedance from the power supply itself. This too will interact with the output signal because this output impedance is effectively in parallel with the cap. At low frequencies, where the impedance of the cap is at a maximum, the interaction with the power supply output impedance is at its worst. The larger the output impedance, the greater the amount of interaction.

What we want, in an ideal world, is minimal interaction between the output signal and the power supply. In practice, this is an assumption when we design amps: only the tube and output transformer are taken into account. Reducing the impact of our power supply brings us a little closer to the ideal. Minimal amplifier-psu interaction requires linear, low-ESR capacitors and a low output impedance supply to feed them. I’m not giving up on CLC filters, but there is only so much we can realistically accomplish with passive filters.

[TimeOut] Let’s be honest. Tube regulators look awesome and make it easy to dial in a B+ voltage. If I were to rank the factors that lead me down this road, those would be near the top and the theoretical advantages (regardless of how awesome the amp ended up sounding) would be just dressing on the window through which I’m tossing good sense and practicality. [/TimeOut]

“Regulate (ft. WTF Amps)”

For this project, I decided to have a go at a series regulator, not to be confused with a shunt regulator. The series regulator only draws the current needed for the amplifier, whereas a shunt regulator requires an additional current (proportional to the load) to maintain regulation. This amp may be impractical, but it’s not quite shunt regulator impractical.


The series/pass tube

The series/pass tube will have to be able to handle at least the current drawn by the amplifier. This being a low power speaker design, I’ll use 150mA (75mA per channel) as a target. The series tube will dissipate power equal to this current times the voltage that it is dropping from its anode to cathode. If the tube drops 100V from the raw B+, it will be dissipating:

100V * 150mA = 15W

This means that if we hold the raw B+ constant, the series tube works harder to achieve a lower voltage output (more volts are dropped across it). This will determine the minimum voltage output. The maximum voltage output is determined by the tubes perveance. Perveance is roughly a measure of how low a tube’s anode to cathode voltage can swing. Higher perveance means the tube can operate with a lower anode to cathode voltage.

Luckily for us, there is a high perveance, high current tube designed for exactly this kind of application: the 6AS7. This is a dual triode and so, assuming we’ll use both triodes, the 150mA total current can be divided between them for 75mA each. Here are the upper and lower bounds of voltage across the tube limited by 0V on the grid (25V anode to cathode) and 13W dissipation (170V anode to cathode).

6as7 LL1.png

The output voltage is the raw B+ minus the voltage dropped across the pass tube. If we have a raw B+ of about 450V, we can set the regulator between 425V and 280V (450V minus 25V or 170V). In reality, we want some headroom on either end and so 400V-300V is more realistic (with a lower current drawn by the amplifier, this range gets even a little wider). The 300V-400V range provides a nice array of potential voltages for smaller power tubes (6V6, EL84, 6AQ5, etc) and larger output tubes (6L6GC, EL34, KT88, DHTs, etc).

Telling the 6AS7 how much voltage to drop from anode to cathode is as “easy” as biasing the grid at the corresponding grid line. If we have a 75mA load (per 6AS7 triode) and want to drop 110V, we just bias the grids at -40V (point #2). If we draw less current, the needed grid bias will change, but the concept remains the same.

As you may have figured out, getting a solid output voltage requires a solid grid bias. For that we need a voltage reference.

The voltage reference

close up 0d3
VR tube

As long as we’re throwing practicalities out the window, we might as well use a good old VR tube as our reference.  They’re big, they glow, get used to them. If we wanted to save some space and cost, a zener diode would also do just fine. As you’ll find out in the next section, the amount of voltage across the reference will subtract from the available headroom in our error tube.

The 0C3 VR tube sits at a reasonably stable 105V, regardless of current. To avoid unnecessarily loading the pass tube, it can be fed from the pre-regulator portion of the supply by a ballast resistor. This resistor is sized based on the raw B+ voltage and the current we want in the 0C3 (unlike other VR applications, we don’t have to worry about an additional load). In this case we want about 10mA through the 0C3 and the raw B+ will be about 450V. So the resistor we want is:

450V – 105V / 0.01A = 34,500 ohms (or something reasonably close)

The error tube

We’ve got the pass tube and the voltage reference; that’s the easy part (sorry). The final piece of the puzzle is the error tube. The only thing the error tube knows is the difference between its cathode and its grid. The error tube’s job is to amplify the difference between the voltage reference and the output of the pass tube. It does this by sampling the output voltage via a resistor divider (grid) while simultaneously being biased by the reference (cathode). It should be an inverting amplifier so that as the grid sees the output voltage go up, the anode of the error amp goes down (dragging the pass tube’s grid negative and increasing the voltage dropped across it).

I’ve got buckets of 12AX7s. We could make an error amplifier by biasing the cathode of a 12AX7 up by the voltage reference and feeding the grid from the output. This is just like a grounded cathode amplifier and a totally valid approach, but it leaves one half of the dual triode unused (any my inner compulsion can’t abide that). We could also use a 12AX7 cascode, but this will result in a need for additional heater taps to avoid violating the maximum heater to cathode voltage. If you have a small high gain pentode around, the error amp is also a great place to use it.  But like I said, I’ve got too many 12AX7s.

dif amp error

For the error amplifier I’m using both halves of a 12AX7 in a differential amplifier. The grid of one half (configured like a cathode follower) sees the voltage reference, creating a corresponding non-inverted signal on the cathode. The other half, sharing a cathode resistor with the first half, is fed a sample of the regulator’s DC voltage output through a resistor divider. It’s this second half, fed a cathode voltage from the first half and a grid voltage from the output, that functions as the differential error amplifier. More exotic error amplifiers are certainly possible, but the differential amplifier seemed like a logical fit here for a dual triode.

Our resistor divider (right side of diagram), with a variable resistance sandwiched by two fixed resistors, sets a DC bias on the grid of the error amplifier. Because we also want the (inverting) error amplifier to help cancel ripple on the output, we bypass the upper part of the resistor divider with a capacitor so that all of the AC ripple makes it through to the grid (rather than dividing the ripple down to a lower level with the resistors). 

Just like the pass tube from earlier, the grid bias will determine the anode to cathode voltage. In this case, the voltage at the anode of the error amplifier determines the grid voltage on the pass tube, which in turn determines how many volts it drops, which sets the error tube’s grid bias through the resistor divider. Round and round we go. This is that terrible, awful, no-good, dirty feedback thing you’ve read about. But come to find out feedback is really just misunderstood. He’s a charming little audio misanthrope full of ello gobnahs and monkey shines, whose delightful hijinkx consternate the puritanicals but bring joy and fun to the world of tubes. Give feedback a chance; deep down he’s a good kid.

[TimeOut] Yes, I just anthropomorphized the shit out of feedback. [/TimeOut]

Power supply schematic

reg psu schematic.png

The pass tube’s transconductance is multiplied by the error amplifier, resulting in an output impedance in the single digits. It also rejects what little ripple is left on the B+ after a simple pre-regulator filter and so the final cap before the amp can be a modest sized film capacitor for superior linearity. What’s not to like?  Oh yeah, it’s complicated as shit.

If it’s any consolation, you’ll be able to easily vary the output voltage, and have lots of glowing tubes to gaze at as you contemplate how you reached this level of unhealthy infatuation. When you attach an amp, it sounds good too.

A power supply for the power supply

My build is fed by a 5U4GB tube rectifier followed by a cheap CLC filter (20u – 1H – 120u). The power transformer is a Hammond 274BX (750V CT). The specifics of what feeds your regulator are variable because the ultimate performance will rely more on the regulator than the raw B+ you feed it. Try to shoot for around 450Vdc unregulated input with a current capacity of at least 200mA and you should be good to go.  A simple filter before the regulator may not be necessary, but because I had some room in the chassis, I figured why not (the choke was chosen for low DCR rather than high inductance).

An important aspect in powering your regulator is the heater voltage bias.  The 6AS7 has a maximum Vhk of +/- 300V and the 12AX7’s is +/- 100V. I biased my heaters to half the output voltage with a resistor divider. This keeps both tubes in spec regardless of the output voltage. Other error amplifiers (eg cascode) may require multiple heater supplies so that they can be biased individually. This is another factor in my decision to go the differential route. I used the heater taps on the 274BX for the amplifier and a separate heater transformer for the power supply tubes’ heaters.


Bill of Materials

Luciérnaga Power Supply
1 30k, 5W+ ballast resistor for 0C3
1 22k, 1/4W+ voltage drop resistor for 12AX7
1 470k, 1W+ load resistor for 12AX7
2 1k, 1/4W+ grid stoppers
1 82k, 1W+ tail resistor for 12AX7
1 100k, 1W+ resistor divider for voltage setting
1 50k, 1/2W+ resistor divider for voltage setting
1 100k, 1/4W+ low pass for 0C3
2 10, 1/2W+ current balancing resistors for 6AS7
1 .01u, 450V+ bypass for 0C3
1 10u, 450V+ voltage drop bypass for 12AX7
1 .47u, 450V+ low pass for 0C3
1 2u+, 450V+ ripple sense capacitor on voltage divider
1 4.7u, 450V+ output bypass capacitor
1 0C3 dat voltage reference
1 12AX7 differential amplifier
1 6AS7 series tube
2 octal socket for 6AS7 and 0C3
1 9 pin for 12AX7
1 50k linear pot If not adjustible voltage, replace with resistor
Example Raw DC Supply (should supply 450V+)
1 Hammond 274BX power transformer, or similar
1 Triad F-43X filament transformer for PSU unit (6.3V @ 4A+)
1 Triad C-24X power supply choke
1 5U4GB or other rectifier, don’t forget the socket
1 20u, 500V+ filter capacitor
1 100u, 500V+ filter capacitor
2 220k, 1W+ for setting elevated heaters in PSU
switches, jacks, fuse

Practical considerations

  • For the umbilical connection between power supply and amp, I’m using a 4-way speak-on connector. It carries the B+, ground, and heater + and -. High voltage and current rated circular connectors are not cheap, but speak-ons are standard in pro audio and so they’re relatively easy to come by and affordable. Speak-ons also don’t have any exposed contact point, making them a fairly safe option. Just don’t plug a PA speaker into the power supply.
  • Once you have dialed in a voltage for an amplifier, you can replace the pot in the power supply with a fixed resistor if you don’t plan to change it. This particular pot (Alpha) is rated for 200V and there is a very small amount of current in the resistor divider.
  • I used a pair of test points on this chassis so that I can use a DMM when setting the voltage with the pot. If you make your supply adjustable, don’t forget something similar so that you don’t have to constantly flip it upside down or disassemble when playing with the voltage.

Holy crap, you still have to read about the amp