V1 Schematic Overview 

Starting on the left side of the schematic, Q7 and Q9 form a classic bistable multivibrator which oscillates at roughly 22kHz. This keeps audible noise to a minimum while prioritizing electrical efficiency. The benefit to this type of oscillator is the relatively fast turn off time of the transistors. I found a good balance of resistor and capacitor combinations that have a fall time under 75nsec, which will maximize the amount of energy extracted from the inductor every cycle. These pulses are fed into Q2 via a high pass filter which limits the current pulses through the inductor to 3uS. This is important to keep the main inductor out of saturation, which will waste energy. Also, if the oscillator gets stuck 'ON' for some reason, there is no risk of dumping current through the main inductor. When Q2 is pulled low, Vbe of Q5 drops below 0.6V, and charge from the main MOSFET gate gets pulled to ground via D3. As soon as Q2 turns off, R9 and Q5 form a voltage follower which turns the MOSFET ON. R6 prevents unwanted oscillations in the gate voltage. 

When the current is cut off from L1, the inductive EMF kickback is fed through the secondary side which adds the two voltages together. This high voltage spike goes through D1 and begins to charge the main high voltage capacitor. R5 and C5 form a snubber network that technically isn't needed, but without it, there is a lot of ringing on the FET drain which will cause a lot of unnecessary RF noise. The efficiency hit is pretty minimal, and it's the 'right' thing to do. The voltage across C1 is sensed via a resistor divider network consisting of R2, R3, R4, and R7. 100Meg resistors are needed here so the leakage is minimal. At 400V, even a 10Meg divider will be the main loss of power in this circuit. 100Meg resistors are a decent balance between leakage current and cost. 

When the voltage at the base of Q3 reaches the setpoint of 0.7 volts, it begins to conduct. The setpoint voltage can be adjusted via VR1. This lets you use a 100k adjustment pot, which is easier than trying to find a 10Meg pot. A crude voltage regulator consisting of R19, D6, and C12 provides a relatively stable 3.6V for a consistent output voltage across the entire battery discharge range.  The collector current in Q3 is amplified by Q4, and inverted by feeding the collector current from Q4 into Q8 and Q10. Even the tiniest base current into Q8 pulls the gate of Q6 low, cutting power to the oscillator circuit. This lets the high voltage capacitor slowly discharge through the resistor divider. As soon as the high voltage line drops slightly, Q6 will turn the oscillator back on, forming a negative feedback loop. Very high loop gain is preferable, as it improves load regulation and response to transient current draw from the geiger tube. R17 and C7 form delay circuit which forces the whole circuit to squeg. While this does cause some ripple on the output, overall efficiency is much better. Q10 conducts when the regulation loop is active, so D2 will light when the output is in regulation. When the battery voltage drops too low, the output falls too low and D2 turns off as a 'change battery' indicator.

The detection side starts with R1 and GM1. Normally, the current through the geiger tube is essentially zero. When a high energy particle enters the chamber, it triggers a chain reaction of electrons that slam into the anode. The result is a pulse of current through the tube which is picked up with the high pass circuit consisting of R26, C13, and R25. This current pulse gets amplified by Q14 and Q12. VR2 adjusts the sensitivity, which needs to be adjusted to discriminate against noise. Q13 drives a small buzzer and generates the characteristic clicking sound geiger counters are known for. D5, C11, and R12 form a simple pulse stretcher circuit which flashes D4 every time the geiger tube is struck. Since the pulses are only about 100uS long, the pulse stretcher lengthens them to 20-30mS so they are more visible to the user. The whole circuit is powered by two CR2032 coin cells which provide 6V. Ironically, the LEDs and buzzer are the main power draw, as the rest of the circuit only draws ~1mA. Including the LED current, I expect the circuit to run for at least 12 hours which is plenty for my needs. C3 improves the battery life considerably, smoothing out the current pulses through the main inductor. 

After waiting ~2weeks, I have my parts from DigiKey and 3 prototype boards from OSHPark. I'm a huge fan of their 'after dark' option, especially since it's the same price as the regular puple soldermask. The traces have really great contrast against the black substrate that give it a sophisticated-but-older look. Strangely, none of the silkscreen was printed on either side of the board, which I'm pretty sure is an error on OSHPark's part, but honestly it isn't that critical since most of the lettering was done on the copper layers. 

Assembly was pretty straightforward with 0805 size components. And it passes the magic smoke test on first power up! It's always a relief when theory actually matches real life for once :) Of course, there were a few passives that needed tweaking. Nothing too major - mostly the components involving the detection side. C13 I decided to change to 10nF for better immunity to voltage drifts on the high voltage, although this shouldn't really make a difference. R24 and R27 I adjusted to give a better usable sensitivity range via VR2. With the original schematic, geiger pulses were *barely* audible through the speaker, so I reduced R23 to 4.7k and replaced the flyback diode with a 10nF cap. It's still pretty quiet, so for future versions I may experiment with lower impedance speakers or switch to a piezoelectric one. The pulse stretcher circuit for the LED was way too short initially, and I eventually figured out that the C11 wasn't able to discharge completely during pulses. Since there wasn't an easy way to increase the discharge current dramatically, I opted for a 10nF cap and 100Meg charge resistor. This gives flashes that are ~300mS long for decent visibility. Also, for reason the snubber circuit across the main FET wasn't tuned right, so I reduced R5 to 100ohm. It isn't perfect, but it's good enough that I'm not going to bother fiddling any more. 

Setting the output voltage is tricky because of the very high output impedance. Even a 10Meg multimeter input impedance will drag the voltage down out of regulation. After experimenting with different setpoints, I set VR1 to ~30% of the max output voltage. I'm guessing the output voltage is 360 volts or so, but I can't say for sure until I buy or make a high impedance probe. 

At 6V, the entire circuit draws about 5mA, although the majority of that current is flowing through the indicator LEDs. With a setpoint of 360-ish volts, it stays in regulation down to a battery voltage of 3.75V. After dropping below this point, the step up converter is on constantly and the input current rises exponentially to a max of 100mA. Of course, the CR2032 cells wouldn't be able to support this much current, especially when discharged to 1.9V/cell. It's difficult to find data on CR2032 capacity vs discharge rate, so I'm going to have to determine that experimentally. I estimate ~20hrs of continuous operation until the output falls out of regulation.

Background CPM is 15-17 in Oregon where I live, which seems to be typical for the SI-29BG tube. Bringing a few Uranium  UV fluorescent beads close to the tube results in readings in the range of 50-70 CPM. I don't have any other check sources at the moment, but soon I will be exploring some abandoned Uranium mines where I will be able to do further testing.