In response to a post on T3h GeeK ZonE, I am building a tube amplifier for once in a long time. The contest is "The Compound Tube Contest". The rules are:
To break a rule costs points. Points are awarded or removed as follows:
Clearly, the contest is aimed at using those tubes in ample supply that nobody wants anymore: TV tubes, compactron or otherwise, ranging from vertical deflection to mixers and oscillators. To name a few 9-pin miniatures, 6T8, 6BM8/ECL82, 6FD7, 6U8 (or ECFxx), 6X8, etc. There are plenty more in the 9- and 12-pin compactron sizes, including triple triodes and even some horizontal deflection types. A typical strategy would be to use a vertical deflection type such as 6BM8 (pentode) or 6FD7 or 6GF7 (triode) for SE output, and the accompanying small triode for preamplification. Two could be used for PP, using the two smaller triodes for preamp and phase splitter. Most of these will land in the 2-10W range, which is more or less sufficient.
My choice is, of course, unconventional. I picked the most unusual, cumbersome signal tube: the 6X8. For the signal path, I chose pulse width modulation, which means, potentially, an absurd amount of power for relatively little power consumption. I would've went with a 5687 for output, but it isn't dissimilar; fortunately, I was donated a 38HE7, a beam tetrode / damper diode combination, which is perfectly suitable because my PWM topology needs a diode.
This is the current revision of my design. Both 6X8s are set up as differential amplifiers, which they seem to excel at. In fact, the pentode section, in triode-strapped mode, is quite similar to the triode, making the whole tube very similar to a 6J6. On both units here, I have the triode plate and pentode screen connected to +175V, which should make relative cathode currents reasonably close (at equal grid voltages) due to the screening effect of the screen grid, which looks like a triode plate without the electron-absorbing cross section. This leaves me with a shielded plate electrode to do with as I wish.
The left side is a hysteresis RC oscillator, much like a 555 timer stripped down a lot. The pentode plate is loaded with 18k, which supplies ample current, necessary to drive plate capacitance in order to operate in the 10s of kHz range. The 1M provides positive feedback to the triode grid, while the 330k to ground and 1M to +V set the DC operating point. On the pentode side, a 100k from plate charges a 91pF capacitor (which happens to be silvered mica, but stability isn't particularly important). A 100k resistor sets operating point. Note that both grids are biased with a fixed, dissimilar ratio: the oscillator's duty cycle probably changes widely with drift (tube age, etc.) and supply voltage. It seems to work over the 150-175V range of my power supply. (Note that, as a grid's bias is changed, this style of oscillator changes duty cycle first, then frequency gets stretched, then finally the bias voltage completely passes the switch point and oscillation stops.)
The middle 6X8 is a comparator with a slight amount of hysteresis. The timing capacitor has a roughly triangular waveform on it, so the comparator output will be a square wave of duty cycle corresponding to the input voltage. Any voltage at the input jack will modulate the bottom end of the bias pot and therefore the output duty cycle.
On the right lies the output stage. When the 38HE7 is turned on, its plate voltage drops and current (through the filter capacitor) builds in the filter choke. When it turns off, the voltage flies back and the diode clamps it. This filter provides a rough averaging of the PWM signal, so the 38HE7 can operate in class D while the output transformer above operates as usual, seeing little of the high frequencies. The same action applies to the OPT, so as duty cycle changes, its voltage can vary from roughly saturation voltage (50V or so) up to about double the supply voltage, or probably 600V or so. It will need to be low impedance and high power, because the 38HE7 can sink up to about half an ampere. As such, it will also need to handle a DC bias of about 250mA (and yes, you might call this "class DA", since it is PWM class D, drawing a quiescent bias current through the OPT).
Here's a picture, as breadboarded. The power supply (on the bottom left edge of the picture) is a 175V regulated power supply I put together for my scopemod project. That it's solid state doesn't matter here.
This is the oscillator's "triangle" waveform, 5V/div vertical, 10μs/div horizontal. I'm not sure what the leading edge "nubs" are, but it may be an artifact of my scope -- this unusually short (6 div tall), unusually green shot is of the Heathkit. The falling edge has a noticable jump to it, which would be from plate-grid coupling somehow. It could be due to breadboard capacitance.
This is the PWM comparator's output, at 50% duty cycle, 20V/div vertical. Rise and fall time is about 2μs, at reduced bias (10k cathode resistor, 27k plate). Again, you can see nubs on the leading edge, which are from the oscilloscope.
This is the transfer curve of the PWM comparator, as seen through an RC filter. Input is some triangle wave. Clearly, it saturates in a rather odd way, and linearity is not very great. Still, with the power I should be getting from the 38HE7, the 3W distortion level should be good. If nothing else, I can add an error amplifier around the whole unit.
Moved off the breadboard into a chassis and wired point-to-point, the output waveform is now much faster. This is the output waveform with 5pF across the comparator's 1M feedback resistor, which seems to be a little much. This is 1μs/div, so the fall time is around 200ns and rise time around 600ns.
This is an X-Y plot of the current transfer curve. Hysteresis is clearly present, probably due to the positive feedback capacitor forcing a minimum duty cycle limit. Gain is substantially lower, which means the oscillator is "working better", another consequence of low capacitance wiring. I may need to install some shielding, as the low capacitance seems to make stray fields a big problem.
This is the newest model. I have the medium voltage transformer and regulator board on the left, switches for 120VAC, a 6.3V 5A filament transformer on the right, and bananna jacks for the high voltage transformer, for which I'm using a 240:120V control transformer in reverse. The transformer is about the size of the whole right end of the chassis, so I thought it best to keep it on cables! The Main Power switch controls the filament, MV and HV circuits. The MV switch operates the +150VDC directly, so as long as heaters are on, the power supply is energized but not connected to the circuit. (This was necessary because the 38HE7 is being heated by the 40VAC from the same transformer.) The HV has its own switch, in addition to main. The high voltage power supply, consisting of an 800V FWB and 120μF 450V capacitor, are hanging out precariously on the right end.
Underside view. 120VAC comes in to a terminal strip on the left. Wires run back and forth, and yes, those are wire ties I've used! The circuit is pretty simple: a pinch of resistors around the leftmost socket form the oscillator, the next tube is the comparator (with the PWM bias pot attached), next is the driver-- did I mention I added a driver? No? Hold on a sec then. And the driver runs the 38HE7, supplied by the power resistors on the right side. The second pot is for biasing the output tube. The negative voltage supply is also installed around the 38HE7.
Here's the update. What's changed: I found it necessary to shoehorn in not just a cathode follower, but an amplification stage as well. So, I put in a 6U8, which seems to do quite well at it. At the 38HE7 grid, rise and fall time is in the 200ns range, not bleedingly fast, but respectable, and typical of these sorts of tubes, which correspond to general purpose transistors, like the 2N4401, which switches in similar circuits at similar rates. If I went to cascode, I could probably get a lot faster, but I don't need it. The 38HE7's output waveform is similar, except for a substantial overshoot and ring due to the inductance of the wirewound resistor. I haven't tried this with the inductor yet.
Here's the complete schematic, with filtering and RC coupling to a transformer. You could call this "class DA", I suppose: the tube is class D, but the output operates with a constant bias. Moreover, the bias is supplied by a fat ugly resistor burning a good 30W non stop. But a transformer, why that must mean I can get audio into a speaker, interesting...
Underside with the output all wired up. The axial capacitor projecting off the terminal strip is the output coupling capacitor. Note the many bypass ceramics near it: these circuits may be relatively high impedance, but gain is high (tubes don't stop amplifying just because they're saturated!) and resonance is resonance, so you have to be very careful with your connections. The performance of this circuit is several times better than the breadboarded circuit, parts of which didn't even work due to the mess of stray L and C! The vitreous power resistor is the bias resistor, and it smells like melted masking tape. Ew. Just to the right is the output transformer, an hermetically sealed device, labeled suprisingly thoroughly: it is rated as 60-15,000 Hz ± 1.5dB, 6W and 600 ohms primary, 4/8 ohms secondary. With the 10μF coupling capacitor, it passes noticable voltage down to 10Hz. Output rolls off noticably by 16kHz, which could be due to the transformer or filter. The filter is one of those 0.02μF ceramic caps and the 10mH choke I wound.
This is the choke, close up. It's bank wound for minimal capacitance: measured parallel resonant frequency is 140kHz, suggesting around 80pF shunt capacitance. I wanted under a hundred, and it seems to be fine. The wire is 25AWG, scrapped from monitor yoke coils. I wound all six banks simultaneously, which was an interesting challenge and tricky to get started on. Eventually it went smoothly. The whole thing is impregnated with wax for stability. Despite the filter, there is still some RF on the output. Since the input hasn't been Nyquist filtered, there are also all sorts of interesting hetrodyne responses when the input is in the 114kHz area, or any harmonic to it. On a whim, I soldered a wire to the timing capacitor and did a little improv Theremin! ;-)
Ahh, tubes' warming glowing warming glow...
Here's the plate waveform, zero input, biased about 50% duty cycle. 50V/div vertical, 2μs horizontal. The rising and falling edges are around 150ns, which roughly corresponds to the inductor current (200mA, give or take) and parasitic capacitances (plate to etc., damper diode, and filter choke). Thanks to the diode clamping the flyback, the waveform is very clean, just a little bounce after the rising edge.
Same, with 350Hz triangle wave input (somewhat below clipping) and 1ms/div horizontal. What you are seeing is the clamp voltage (320V + diode Vf) and saturation voltage (0V, set to 1 div from the bottom), which are both quite low. This isn't much worse than MOSFET performance, quite honestly! Since forward voltage rises with current ("on resistance" is about 100 ohms here), the higher peaks correspond to nearly 400mA. Screen voltage was measured at 110V, out of a 155V supply, or 45V drop across the 1.3kohm resistor → 35mA Ig2, putting Pg2 = 3.8W, not too bad, but maybe on the hot side.
Output at a 15 ohm resistive load. I must say, the transfer curve is tantalizingly linear, especially for the RC / exponential curve oscillator waveform. Sound quality is quite acceptable, playing various things through a loose speaker. It even sounds like clipping when overdriven. There is a small amount of hysteresis at clipping, as seen on the transfer curve in the clipping region.