Gamma Spectrometry Forum

Like I said in my introduction to this board, I'm planning to build a Farnsworth-Hirsch-Meeks fusor. In case you don't know what that is, it's an inertial electrostatic confinement nuclear fusion device. More info on that at fusor.net and its associated discussion boards.

Having recently completed and tested my neutron detection device, which will be used to confirm nuclear fusion once I achieve it, I am now concentrating on the next subsystem that I need, which is the high voltage, high power source.

Fusion doesn't require only high voltage, it also requires enough power at the high voltage. At a minimum one should supply 20kV at 20mA to obtain detectable fusion (rough minimum estimate). That's 400W of power.
I will aim higher. As a first goal, I will target 30kV (or more) at 30mA (or more). That's 900W of power.

I was unsuccessful in finding an affordable power supply with the required parameters. While I am still looking, I've decided to make my own power supply. For efficiency, it will be a switched mode power supply (SMPS) with a ferrite core transformer.
All calculations below are for a frequency of the SMPS of 20kHz.

A quick back-of-the-napkin calculation gives me this for the requirements of the ferrite transformer. For a supported power of 900W, it would need a core area of at least 35cm^2. The power supported increases with the square of the core area.
Cores that I can find at reasonable prices online are EE140 type, which have a core area of 4x4=16cm^2. One of these supports about 180W power transfer. Not enough. But if I stick 2 of them together in parallel that will give me a core area of 32cm^2, or a power of about 750W. Still not enough. How about 3 cores? That is an area of 48cm^2, and power of about 1700W. Check.

The material of these cores is a Mn-Zn ferrite, PC40. Datasheet gives for PC40:
-initial permeability 2300 at 23 degrees C
-Curie temperature above 200 degrees C
-saturation flux Bs of 0.5T at 23 degrees, or 0.38T at 100 degrees C; this is important.

I will assume that it will be pushed to its limits and calculate for max temperature of 100C.
From previous failures I have acquired the habit of over-engineering things. This not only gives me a safety factor, but also allows me to upgrade stuff in place if I need higher specs later.
So I will multiply the Bs at 100C with a factor of 0.3 and use that for calculations as maximum flux.
That gives me a Bs of 0.38 * 0.3=0.114T, or 114mT or 1140gauss.

Calculations for primary winding.
B(gauss)=V*T(on)*10^8/2*Ae*N
V=primary voltage
T(on)=duty cycle/frequency
Ae=core area (cm^2)
N=number of turns
Or expressed otherwise
N/V (turns per volt)=T(on)*10^8/2*Ae*B

T(on)=0.5(square wave with 50% duty cycle)/20000Hz=0.25*10^-4
Ae=48cm^2
B=1140gauss
thus N/V=0.25*10^4/2*48*1140=0.25/10.944=0.0228 turns per volt

For the primary I will use mains AC (120V in the US), passed through a 2KW autotransformer for power control, rectified with a full-wave bridge and filtered, then switched on/off at 20kHz/50% duty cycle with power mosfets (probably a half-bridge configuration) into the primary.
120V is the Vrms of the mains; Vpeak is 169V (Vrms/0.707); Vaverage is 108V
I will aim for an average voltage of 12000V in the secondary, thus use the 108V value in the subsequent power calculations. The Vpeak in the secondary will be 18837V, and I will use this to select components. I won't calculate a transformer that gives me more than this for practical reasons (insulation etc), and will raise the voltage to the desired values using a Cockroft-Walton voltage multipler in the secondary.

I will derive the current needed in the secondary from what the CW multiplier needs to achieve the desired output parameters.
For an input of 12000V, a 1-stage CW multiplier using 4nF capacitors at a frequency of 20kHz, in order to supply 30mA output will require at least 83mA input current. The output voltage will be 33kV with a ripple of 380V. That is acceptable.
Will round up required current to 100mA. This is what I will use in subsequent secondary calculations.

Primary turns.
N/V=0.0228 (from above) V=108V (average), 168V(peak)
N=2.46 turns for average voltage, or 3.69 for peak. Round up to 4 turns.

Secondary turns.
Vavg=12000V, Vpeak=18837V. Turn ratio=12000/108 + 10% (compensate for losses)=111.11+11.11=122.22. Round up to 123
Turns=123*4=492.
The wire needs to carry 100mA at least, so a minimum of 31AWG, or 0.226mm diameter. That will take a current of up to 113mA. Check.
If I later would require the secondary to provide moire current I could use 30AWG (up to 140mA) or AWG29 (up to 180mA). It would be pointless to increase the wire gauge above that, since at 180mA the power would be 2100W, which is more than the transformer core supports. AWG30 would support up to 1680W which is within core spec. Make a note to not exceed 140mA in the secondary.

Primary wire.
Current in primary: 100mA*123(turn ratio)=12.3A Will apply the same overengineering concept and round up to 15A minimum.
Wire size for 15A is 10AWG, or 2.58mm diameter. But hold, at 20kHz there will be a skin effect. The skin depth at 20kHz is 0.5mm, so any wire over 1mm diameter will be affected by a reduction in its current carrying capacity. For 10AWG, this leaves a surface area equivalent of 21AWG (0.72mm diameter) which can only carry 1.2A. Not good enough.

Quick area calculations taking into effect the skin effect shows that I will need primary wire of at least 6.89mm diameter. 1AWG(7.34mm diameter) is the choice. Check.

So overall I will need:
3*EE140 cores in parallel
4 turns of 1AWG wire primary wound over all 3 cores at once
492 turns of 31AWG wire wound over all 3 cores at once. (113mA max). Or 30AWG (140mA max)
Cross-sectional area of secondary winding: for 30AWG, d=0.254mm; add 50% allowance for insulation, 0.381mm diameter. Area of each wire 0.114mm^2. Times 492 gives 56mm^2. Adjusting for a fill ratio of 1/3, total secondary cross-section 168.2mm^2 or 1.68cm^2. That will fit in the core window without any problem. Can even consider using 29AWG from the start for extra tolerance and reduced DC losses (lower secondary DC resistance).
End transformer calculations.

Input: AC mains through autotransformer, full wave rectifier bridge, filtering capacitor (all specified for at least 200V, 15A); half-bridge MOSFET plus driver at 20kHz.
Output: 1-stage CW multiplier, 33kV at 30mA.
Power required for output including CW losses: 996W, approx 1kW.

Can later add a second CW stage for higher output. With the second stage, output would be 64kV/30mA with 2600V ripple, but the secondary current would rise to 161mA and total power required rise to 1937W. That is out of spec.

However if I only require a current of 18mA the output will be 65kV with 1600V ripple, and secondary current 98mA for power of 1200W. That is within spec. So this upgrade in place would be viable without needing the transformer to be rewound. More voltage and more power=better fusion efficiency.
The maximum I could extract from this design with 2-stage CW multiplier, based on the 140mA limit for the secondary current set above, would be 26mA at 65kV with 2200V ripple. That gives a power of 1688W.

Decreasing the ripple would require increasing the value of the capacitors. For instance, using the values in the max above, increasing the cap value to 8nF (2 in parallel) decreases the ripple to 1140V, and 12nF (3 each in parallel) to 758V. It would be a worthwhile upgrade, but these high-voltage caps are expensive.

The weight of the transformer, based on the weight of the cores plus that of the wire, is estimated at about 12kg. Size 14x14x12cm.

Please feel free to check my calculations and make appropriate corrections.

Silviu,

Building a fusor is challenging enough, so I admire your intentions to build your own power supply, even more so when you make the effort to wind your own transformer. This is way beyond any of my experience so I can't offer any advise.

I would like to build a high voltage supply of a different and somewhat simpler design which Chris Bradley came up with some years ago. (See pdf attachment).

This type of supply would be very quick and cheap to make.

Steven

That's an interesting design and I will have to spend some time studying it. Thanks for sharing.

In the meantime I came up with an alternative plan. It has advantages and disadvantages over the original one above. I can't decide yet between the two. I'll probably take a step back and let it rest for a few days before looking at both again with a fresh brain.

The alternate plan is to wind directly a HV transformer for the mains. I can use a big *ss laminated ferro-silllicon core (I know where to get one).

The advantages would be: no need for front-end complicated electronics. Plugs directly into the mains (or into a variac). At the other end you get HV.
The disadvantages... well for one the winding. Even with the significantly higher magnetic flux that iron cores support (approx 2T vs 0.5T for the best ferrites) you still end up winding a lot more wire with the low frequency mains. And I mean a LOT more. Primary would be 150 turns vs 4 turns for the ferrite, secondary 15000 turns vs. 492. That's a big difference.

Another advantage/disadvantage pair:
1. you get directly 12000V from the mains 120VAC/60Hz. Good, right?
2. you get 12000V, but at 60Hz. And designing a voltage multiplier with reasonable parameters for 60Hz is a lot harder. There's pesky ripple creeping in everywhere, and you have to increase the capacitance many times to tame it. And those HV capacitors are expensive and hard to find.
So increase output voltage directly to 20-30kV, correct?
Well I did mention something about many thousands of turns in the secondary. It only gets worse after that.

Soo at this point I have 2 candidates. I don't know yet which door I will choose.

Not sure what kind of current Chris Bradley's PSU can deliver, but it should depend on the CCFL inverters used. Those can be obtained from as little as $10 each. I found an old image of the prototype made by Chris, hope he doesn't mind me reposting it here. Nice clean job.

Capacitors and HV diodes are easy to find on ebay

[broken link removed - Steven]



Steven

Ebay capacitors... I bought several times HV capacitors on ebay. I'm wary of their ratings, especially at high frequencies. I haven't done in-depth tests, but I think they should be derated by a certain factor at HF. And that is assuming that you actually get what you pay for. Most Chinese sellers have no idea what they're selling, if something is round and blue it must be a capacitor (but sometimes it isn't; once I received a bag of varistors instead).

While I plan forward about a big power supply, I'm starting to build a small one, just about 2W of power, for the specific purpose of testing my designs and components. I have a (large for other purposes, but small for a power transformer) ferrite toroid of unknown material that I got a while ago from aliexpress; it's 76x40x10mm (ODxIDxh) and am waiting for the wire to wind it; the front-end will run at 12V and will have a TL494 PWN controller, a couple of gate drivers, a gate driver transformer that will provide out-of-phase gate signals to the 2 power MOSFETs that drive the transformer primary in half-bridge configuration. If that works well in small-scale (12V primary, 1000V secondary) then I can scale it up with the same drive circuit but bigger MOSFETs and power transformer.

This will allow me to test
a. my driver design
b. my skills at winding toroidal transformers, which I've done in the past but not to a great extent
c. components, especially MOSFETs, HV capacitors and diodes.
d. RF interference on my neutron detection system. A HV arc set a few cm from the detection tube should provide ample amounts of that.

OTOH, HV diodes are indeed easy to find at good prices. This particular model has been successfully tested by other people in CW multipliers with good results
[broken link removed - Steven]

Small change to the project. I have decided to use IGBTs instead of MOSFETS in the power driver. It won't change much in the other electronics part, but I expect to have smaller power losses in the driver stage, and thus less heat generated.
Why? A MOSFET is essentially a voltage-controlled resistor. Even when fully "on" it's still a resistor; a small one, but the voltage drop on it increases with the voltage it switches. An IGBT is essentially a BJT which gives it a Vcesat in the parameter list - essentially a fixed voltage drop which varies little with the switched voltage.
At low voltages, say 12V, the MOSFET is better. Say we have a MOSFET with Rds(on) of 100mOhms. That means it will drop 1.2V at 12V. IGBTs have a Vcesat on the order of 1.8-2.5V. So at 12V the MOSFET will waste less power and generate less heat. But at 120V, the MOSFET will have a voltage drop of 12V, while the IGBT will still only dissipate 1.8V.
IGBTs are a little slower than MOSFETS, but that would only matter if I were using a switching frequency of 400kHz or more. Since I use 20kHz that won't matter.
I have ordered IGBTs, gate drivers (NCP5181) and PWM controllers (SG3525A from OnSemi, a little more modern than the TL494). I expect to get them in a week or two and then I'll start experimenting. Until then I'll get on winding some toroid transformers - I expect to get the wire today. I will probably order the big ferrite cores as well as these come from China and will probably take 3-4 weeks to get here.

So I sat down and spent my Sunday by winding a transformer. I'm testing the "stacked cores" hypothesis. The medium ferrite toroids that I have, if used individually should support 2.5W, but if stacked by 4 would reach 40W or so. If this tests good I can scale up to the stacked extra-large EE cores.

Here are pictures. Testing another day, I'm too tired today. This business took me 2 hours at least. It reminded me how much I hate winding toroids. I had avoided doing it for 20+ years until today.

Anyway, specs are as follows.
4xtoroid (cheap, unknown parameters from China), 76mm OD, 38mm ID, 13mm height
secondary: 170 turns 28AWG wire (I didn't need to use 28AWG, 35AWG would have been enough but it would have been technically more difficult); hopefully will give around 1000V, 40mA.
primary: 2 turns, 2 strands 18AWG wire; operating at 12V, 20kHz, 3.25A if the power rating is correct
Need to build an appropriate driver.

(edit) After I rested a while I took a long look through my electronics stash. I have piles and piles of MOSFETs among other stuff. But that wasn't what I was looking for. Eventually I found it, a small package of IGBTs which I just knew I had somewhere. Yay. 600V, 60A of low Vce(sat) goodness. And while I was sitting I drew a schematic for the invertor and driver. I don't need any advanced PWM features, just a fixed frequency oscillator, an intermediate amplifier/gate driver and the half nridge itself. A 555, a few transistors, a small gate transformer and the IGBTs are all I need beside the large transformer itself. And if I calculate everything right I can build it once, and use both with the test transformer above and with the final transformer. The stuff I ordered earlier will get used in other projects - or in this one if the simple approach fails for any reason.

Silviu,

Nice patience job..

The high voltage transformers I have seen in commercial PSU's seem to have a lot more secondary windings (see image).

The image is from a 100 kV Gamma HV PSU that I managed to blow up twice with my STAR fusion reactor. Gamma HV were fantastic, they repaired it for free both times, even though it involved replacing the whole potted voltage multiplier.

Look forward to see how it goes..

Steven

Well this is a small test transformer, only designed to go to 1000V at a few miliamps. I've calculated that the final transformer will have a secondary that will occupy a cross-section of 3x7cm in the EE-core window.

Anyway here's a quick and dirty schematic. It's not original, I lifted a large part of it from http://www.stevehv.4hv.org/FBD/FBschematic.JPG (but even that isn't really original, it's a classic 555 oscillator with 50% duty cycle, an integrated driver which I replaced with a couple transistors, and a classic half bridge). The component values are based on what I found in my boxes and could be replaced with anything equivalent or better. The gate drive transformer should be wound with at least 24AWG wire to allow for currents up to 500mA. One of the secondaries is wound in the opposite direction compared to the primary and the other secondary.

I didn't add over-current protection, but that could go like this: small current-sensing transformer in series with the big transformer primary in the bridge; secondary from this would be rectified and fed into a comparator, then the comparator output into an optocoupler; the optocoupler output is used to reduce the drive in the driver, e.g. by switching different resistors instead of the one in the base of the second transistor. This resistor (which I marked 100-1k) is what determines the strength of the gate drive. It should be determined experimentally, and will depend on the power components in the bridge and their gate drive requirements.

Or even better, a small signal MOSFET such as BS170 connected between the base of the first transistor and the ground could do the job; perhaps even working as an AGC/voltage-controlled resistor to reduce drive signal if output exceeds a certain limit.

The IRF640 should be good enough for operating the power section at 12V; but for 120V either higher-rated MOSFETs ot the IGBTs noted will be necessary.

AGC /overcurrent protection idea.

For a modular build, everything from the 555 up to and including the gate transformer, and the optocoupler would be in one module (the signal module); everything downstream and including the power MOSFETs, and the current sensing circuitry would be in the power module. The current sensing part would be dimensioned to the power transformer and voltage in use, and would provide the same signal to the OC. I could then switch power modules depending on the transformer I use and keep the same low-signal part.