The first order of business is to come up with the high voltage itself. When vacuum tubes were common, high voltage power supplies were too. These days, electronics are solid state, and run on low voltage. For a long time, 5 volts was the norm, and now voltages are getting even lower to support both devices with smaller geometry and reducing power consumption and its attendent heat generation. 3.3 volts was popular for a while, and newer devices run on 1.8 volts. In this arena, parts to produce high voltages are uncommon. As I like to share my designs, I'd prefer to use current production parts that other people can obtain fairly easily and cheaply. Happily, there is a current source for high voltage transformers. LCD screens need backlights, and one popular technology for backlights is the "cold cathode fluorescent light" (CCFL). These are long thin tubes that are light by (aha!) high voltage. Since they're so common, the associated high voltage power supplies and the parts they're built from are also common.
I found a nice design at tubetime that used a CCFL transformer in a lashup using a voltage divider to sample the high voltage, a voltage reference for comparison, an operational amplifer (op-amp) to compare them, and a power transistor to control the CCFL circuitry. I figured I'd breadboard it and see how it performed. The CCFL transformer itself, while common, is still a specialty part. However, complete CCFL modules are, inexplicably, cheaper. Eyeballing the schematic showed that the core of the design was the same lashup used in the CCFL supply. All I had to do was add a rectifier and reservoir capacitor to convert the high voltage AC output to the high voltage DC I wanted, just as was done in the tubetime design. A quick breadboard showed that it worked as desired.
However, the tubetime documentation explained that the design was derived from a Jim Williams application note, so I read it and saw that Williams' original design didn't use an linear analog feedback loop like the tubetime version, but employed a switching voltage regulator chip instead. This was appealing, as it would be both more efficient and have a smaller parts count. The switching regulator chip replaced the voltage reference, op-amp, and power transistor with a single part, and added some nice protection circuitry as a bonus. I decided to build that version. In the process, I modified the CCFL supply slightly by cutting a trace to separate the low voltage ground from the high voltage ground. I did this because the low voltage "ground" wasn't actually ground, but the switching transistor, and I didn't want to ground an 800 volt signal through the low voltage switching power supply transistor.
The first try didn't work very well, and only managed a hundred or so volts of output. My usual debugging technique for hardware or software is the same: divide and conquer. I tried operating the CCFL inverter directly from a bench supply, and it behaved nicely, even when operated from much lower input voltage than specified: 1 volt in yielded over 800 volts out, in a nice clean sine wave (The 1000X high voltage scope probe comes in handy for such measurements).
Adding the rectifier converted the AC voltage into DC nicely. Adding the feedback voltage divider produced the desired lower-voltage version of the input voltage.
Hooking it all back to the control chip went back to producing low voltage output again. Investigating with an oscilloscope revealed a bunch of of ringing at the switching node.
Perhaps all that ringing was somehow keeping the circuit from operating properly. Perusing one of the other related app notes on how to use the switching controller chips revealed a simple suggestion: add an RC "snubber" network in parallel with the (CCFL inverter) load. The suggestion gave a range of values often found to be helpful. I dug through my parts cache and came up with a capacitor with a midrange value, and a potentiometer that nicely covered the range of resistance values. I hooked them up, and watched the 'scope while twiddling the knob, and violà! The wild oscillation was tamed, and the circuit produced a nice DC output that varied smoothly with the feedback ratio. You can see that the "on" time (flat bottom of lower waveform) is longer to produce higher output voltage (upper waveform), as expected.
With that working, I grabbed a 2AP1, a common small CRT, looked up its recommended operating voltages in a manual. The usual way to produce the operating voltages is with a resistor chain operating as a voltage divider, with certain resistors actually being potentiometers that can be adjusted to fine tune the voltages to provide good brightness and focus. I worked out the ratios, then juggled the values to bring them into line with reasonable resistances and common values. I also rounded up a heater transformer to provide the necessary heater voltage (this is a much higher current draw, and therefore is impractical to supply with the resistive divider). I lashed everything together, and got a glowing spot on the face of the CRT! By adjusting the various controls, I was able to get the spot from the original big splodge to a reasonably sharp dot of light. Success!
Now that I had the CRT lit, the next order of business was to be able to move the spot around. This particular CRT uses "electrostatic" deflection to do this, which uses changing voltage on internal deflection plates to push and pull the beam around. A while back, I had designed a deflection amplifer to provide the appropriate voltages, as well as give controls for sensitivity (size) and position. I dug out the old board and hooked it up to the appropriate pins. As a quick test, I fired up the CRT again without powering the deflection board. My nice neat spot was replaced by a large fuzzy dim glow! What had gone wrong?
As it happens, the deflection plates in the CRT sit right after the accelerating anode, and the electric field between the anode and the deflection plates acts as an additional electrostatic "lense" that affects the beam focus.
I had arranged my divider with the negative end grounded, so the accelerating anode sat at about +1000 volts. The deflection plates, however, were also grounded, so they were about 1000 volts more negative than the accelerating anode. That created a strong electric field, which was what was defocussing my beam. The RCA CRT documentation actually addresses this directly in a footnote that reads:
In order to minimize spot defocusing, it is essential that ultor be returned to a point in the amplifier system which will give the lowest possible potential difference between ultor and the deflection electrodes.
There are two general ways to deal with this. One is to "float" the deflection amplifier at 1000 volts, so the voltage difference is small and doesn't create the unwanted electrostatic lense. The tubetime circuit works this way. The other approach is to ground the voltage divider near the accelerating anode end, so the accelerating anode is near ground, at a similar voltage to the deflection plates. In this diagram, you can see the ground symbol connected to the junction of R1 and R2, at the top of the voltage divider.
This puts the cathode of the CRT at around -1000V. This has the advantage that the screen of the CRT is near ground, which both reduces the shock danger from people touching the screen, and proximity effects where touching the screen warps the display, due to the electric field created by peoples' fingers. The disadvantage is that the cathode and intensity grid are now at a large negative voltage, so the heater supply needs to be "floating" at a matching voltage to not overload the heater insulation, and the intensity grid also needs to be at a large negative voltage, making it trickier to control.
I chose the latter approach, and rearranged my voltage divider to put the ground and feedback voltages near the positive end. My heater transformer is good for 4000 volts of isolation, so it's fine. If I need to control the intensity with low-voltage logic circuits, I can use a device known as an opto-isolator to bridge the voltage difference.
With that done, the CRT spot looked good again. In fact, it looked even better, as the deflection electrodes were now hooked to an appropriate voltage, so they no longer distorted the beam. When I turned on the deflection amplifier power supply, the spot jumped off the screen. The position controls could bring it partway back, but wouldn't center it. Then I remembered I had designed that amplifiers to accept a 0 to 5 volt input, and there wasn't an input connected, so they were correctly pushing the beam off to one side. Hooking the inputs to an intermediate voltage allowed the position controls to center the beam easily. This also nicely demonstrated that the deflection amplifiers were working, and could control the beam position just like I was hoping.
All this talk of electron beams and electrostatic fields strongly reminds me of my work in the Charged Particle Beam Lab at the University of Maryland. While I was there, I worked with scientists who were designing "electron guns" to produce high-quality electron beams. One of the tools they used was called the "Herrmannsfeldt code". This was written at the Stanford Linear Accelerator Center in ForTran by William B. Herrmannsfeldt, and supplied as a large deck of cards. I ended up working with this code, and working with the physicists to translate their electron gun designs into data the code could accept, and adding a plotter driver to the code to produce graphical output. The code itself dealt with the fact that the electric fields affect the electron beam, and the beam in turn affects the fields. It would iteratively solve the equations until it converged on a stable solution (if it could). It was a big slow beast on the computers of the day, but it did a good job and I enjoyed working with it. That code is still around, but now it's known as EGUN.