Here's an interesting piece of test equipment I came across in a dusty corner of the lab, of a kind you don't see all the time: a high-voltage low-current power supply/"DC amplifier", a Trek 609C-6 from the late 80's/early 90's. It can put out 4 kV @ 15 mA, useful for various electrostatic applications (was used originally for a controllable-viscosity fluid, in this case), biasing of detectors (avalanche photodiodes, photomultipliers), and other physics things (accelerating or deflecting beams of charged particles, etc.). For something that seems to be marketed as an externally programmable power supply, the AC performance is actually not bad, with a 13 kHz bandwidth and 150 V/µs slew rate (= 27 µs from zero to full-scale).
Let's take a look inside:
Power SupplyAt the top-left in that overview (front-right) is the power supply: there's a couple 60 Hz transformers, the larger of which feeds a switching power supply, where a TL494 drives what's basically a classic
flyback converter without the output rectifier.
The output passes through a couple wires down to a board underneath it, where a pair of voltage multipliers creates the bipolar high-voltage power supply. You can also see the back side here of the single TO-3 power MOSFET from the switching supply (a BUZ42), mounted to the chassis for heatsinking.
Everything in here was covered in a layer of dust, but notice how the dust on the output side of the high-voltage discharge resistors (the green rectangles at the far right) has built up a lot more than everywhere else - some nice electrostatic attraction effects at work.
The flyback-style drive for the voltage multiplier seems like a pretty reasonable approach for low-ish power levels, as it's an easy way to avoid the extra complexity (extra transistors, high-side gate drive, etc.) of the half-bridge or full-bridge arrangement that might first come to mind for driving a square wave. When the transistor is off, the flyback's coupled inductor discharge is inherently a current source, which is great for charging the capacitors in a
typical voltage multiplier in a controlled way. When the transistor turns on, the flyback's coupled inductor now acts as a transformer and would in theory force the transistor to sink a large short spike of current as it charges the voltage multiplier's capacitors voltage-source-style; however, intentionally adding some significant leakage inductance to the secondary winding through the winding geometry would help a lot with this, and may be the approach the designers here took (I wouldn't know without desoldering and measuring). Comparatively loose coupling would also explain the presence of the somewhat-excessive snubber action on the primary side: stacked zeners to clamp over-voltage, plus two(!) separate RC snubbers. I haven't thought through fully how the regulation would work, but since half of the voltage multiplier's switching cycle is current-controlled, I'd assume controlling the peak currents built up in the coupled inductor would still result in directly controlling how much charge gets dumped into the output capacitors, just as with a standard DC-output flyback converter.
Control SectionAt the top-right in the overview from before is the control board.
This contains the control loop which takes in the input setpoint signal, passes it to an error amplifier, and generates two separate "push" and "pull" control signals with the pairs of metal-tab-heatsink transistors: these control signals then drive the output stage.
Output StageThe output stage, in the back half of the enclosure and separated with an electrostatic shield and lots of insulation, consists of two series stacks of 9 output boards each, one as the "push" (connecting the output to the positive HV supply) and the other as the "pull" (connecting the output to the negative HV supply).
The series connections are made on the top, and each stack of output boards has their control inputs connected in series. There seems to be something a little more complicated going on than just straight series connections the whole way down, but didn't get a chance to trace the exact wiring details. The general idea though is that each output board in one stack gets the same(-ish) control voltage, and all the output boards share the voltage drop and power dissipation: it's usually easier and cheaper to use 10x series 1 kV MOSFETs than a single 10 kV MOSFET, at least up to a point.
This is what each output board looks like:
...and this is the schematic I traced:
This circuit works as a high-voltage current source, across the two output terminals (WP1 and WP2). The control signal is isolated by the optocoupler at the far left, where a variable LED current creates a variable phototransistor current. This control current is multiplied by a resistor (R6 + trimpot) to create a voltage that biases another current source (2N3906 + R7 + PTC) with over-temperature foldback which then directly sets the output. A cascoded high-voltage transistor (Q2 biased by ZD6, R1) then stands off the (potentially) high voltage across the output board, and an additional cascode transistor (Q3) stands off an additional 1/6th of the output voltage (R8/C1 vs. R1/C2).
Even though the additional comparatively-low-voltage cascode feels unnecessary, I'm guessing it was done either because Q2's power dissipation or voltage rating (with parts easily available at the time) was marginal, and with physical constraints (fitting in a standard 19" rack?) it ended up being easier to add an extra transistor to each output board than to add 2-4 extra output boards to further spread out the voltage and power dissipation.
The purpose of the additional current source connected to the optocoupler output is both (1) to not require the optocoupler's phototransistor to carry the full output current, and (2) to provide adjustability, both fixed (through the trimpot) and temperature-dependent (with the PTC thermistor thermally coupled to Q2). Optocouplers, unless intentionally sorted and matched, typically have a wide tolerance of CTR, or current transfer ratio from input to output, and so Trek would want to adjust all 9 output boards in each stack to have matching output currents with the same input current, so that they will share voltage and power dissipation relatively equally. The CTR also varies with temperature and aging, but as long as the different boards in each output stack all drift together, it works out fine. The potential need for re-calibration over time, if the optocouplers
do age differently, is a downside of this design.
Finally, the output voltage gets fed back to the control board through the orange cylinder over the control board, seen here in the overview again loosely connected to the front-middle wall:
This is almost definitely a capacitively-compensated HV voltage divider. The unusual packaging is likely for some combination of insulation, power dissipation, and fixed geometry to control the parasitic capacitance down to a low level (sub-pF). The last two points are especially important - let's suppose we're using a 1000:1 voltage divider, with a 10M high-side resistor, for example: this high-side resistor will dissipate 2.5W with a 5 kV input, which is a lot of power for a single component, great for causing overheating or thermal drift at the very least. So let's say we use a 100M high-side resistor instead - now, even 1 pF in parallel with it will put a zero at 1.6 kHz, which needs to be compensated for with a 1 pF * ~1000 = 1 nF capacitor across the low-side resistor to make the amplifier bandwidth and control loop stability manageable. To keep this capacitive compensation working correctly, it'll be important of course to keep this parasitic capacitance across the high-side resistor from varying due to external metal (positioning of the input wire, nearby metal chassis parts, nearby PCB copper, etc.). You can do this by either putting a large insulating barrier around the voltage divider to enforce a minimum spacing (large enough where any metal at this distance won't have a noticeable effect on the parasitic capacitance) or putting a conductive shield around the voltage divider, which then needs appropriate insulation on the inside, outside, or both. The shielding approach seems to be what's going on here, based on the copper visible through the sides of the cylinder.
Anyways, hope you've enjoyed this look inside. There's a few more photos in the Flickr album that I didn't include here.
Home
Help
Search
Login
Register
