Marx Three

Mission : one million volts!

May 2004 - Tower Two is now working - see below for details!

This is the first stage of a two-tower bipolar marx generator which should produce voltages in the region of one million volts in its final form. This first stage produces voltages around half a million volts. These voltages are nearly impossible to measure - the actual output voltage is the input multiplied by the number of stages, minus any (hard to quantify) losses. With a 25KV input, it will produce 25" (63.5cm) long sparks, which suggests the output is not too far from the 0.5MV target. See this page for smaller marx generators which led to the design of this one. This design uses a high voltage relay for triggering, due to problems seen in the earlier designs with unreliable sparkgap triggering.

Marx generators are basically very simple devices (see schematic, left). They work by charging a number of capacitors in parallel up to the input voltage, and then discharging them in series, producing an output of the input voltage (25KV in this case) multiplied by the number of stages (20). The switch between parallel charging and series discharge is achieved by spark gaps. In charge mode, the sparkgaps have no effect, so the generator appears as a 'ladder' array of resistors and capacitors. When one gap fires (in this case, this is forced using a relay), it causes the voltage across  the adjacent gap to double, causing it to fire, which in turn causes a ripple effect up the tower until all sparkgaps are conducting. This firing sequence happens in a matter of microseconds, so the draining effect of the charge resistor chain is negligible.

The gap in series with the trigger relay was found to be necessary to limit the peak current through the lower two resistors, which would otherwise see the full 25Kv charge from the lower 2 caps applied accross them  - a peak power of  over 2KW, which they really didn't like, erupting with a small jet of flame!
Unfortunately adding this gap remove the inherent self-discharge of the caps when the relay is in the discharge position, so if th emarx doesn't fire, all the caps remain charged (ouch!). A high value resistor should be connected across the 2 sides of the marxgen to avoid this.  

Despite its simple appearence, there are many practical issues which make marx generators quite difficult to construct once you want to get serious voltages out of them (although at lower voltages they can be built quite easily - see the Quick & Dirty Marx Generator page). Capacitors with suitable voltage ratings are difficult to obtain cheaply, except if you get lucky from surplus sources. The resistor chains must be rated for the charging voltage - resistors that will withstand several KV are hard to obtain - it is usually necessary to make up the resistors out of several ones having lower voltage ratings. Another major issue is controlling corona loss - controlling corona from the output voltage is next to impossible for a homebuilt design, but steps can be taken to reduce corona loss during charging - to ensure that all the caps charge fully, the current drain from corona loss needs to be small compoared to the charge current through the resistor chain. 

Left and right - discharges snaking around the edges of a clear acrylic sheet

Left and right - discharges tracking across the surface of a polythene sheet (Taken at UK 2003 Teslathon)

< Discharges on the surface of an acrylic sheet  (Taken at UK 2003 Teslathon)

> End-on view of a 25" spark discharge

Construction details

General views of construction. All parts are assembled on a central strip of 6mm thick acrylic (Perspex / Plexi) sheet, 90mm wide. This is then reinforced by 70mm wide strips on each side forming an 'H' section. Side plates are fixed with M3.5 nylon screws into holes tapped into the edge of the main strip.

Sparkgaps are 3/8 inch brass balls (from McMaster Carr,   Part #9617K41). These were drilled with 2mm holes, and 2mm tinned-copper wire inserted in the hole and soldered in. The wire was then sleeved with silicone sleeving (from stripped-down neon sign cable). The wire is then push-fitted into 2mm holes in the 6mm acrylic base, after first being squeezed with serrated-jaw pliers to create a spline to improve grip (shown right).

Views of base section, with 35KV rated SF6 (Sulphur Hexaflouride) filled relay for triggering. (Kilovac K61C841) Shrouded 4mm test sockets are used for connections, as these provide reasonable shrouding to reduce corona. the capped yellow socket is for future bipolar operation.

As the relay I obtained had changeover (SPDT) contacts, I could use it to disconnect the HV supply before discharge. This avoids the need for a ballast resistor to ensure the sparkgaps quench, and reduces the amount of potentially damaging impulse energy that may be kicked back into the PSU. The relay is wired so that its open to charge, and closed to discharge, ensuring the capacitors are safely discharged in the event of power loss.

<End view - H section is 9cm wide x 7cm high

>View of base, made from sheets of 4mm acrylic, heat-folded and fixed with nylon wing-bolts (wingnuts with lengths of cut-off screws glued in) .

< Before starting to chop up large bits of plastic, I found it extremely useful to build this little test-piece out of some scraps, to experiment with various dimensions, and get a feel for things like ease of adjustmemt of sparkgaps and general assembly issues.
In particular it confirmed the viability of the method of supporting the sparkgaps solely by close-fitting holes in the main plastic base. 

>Yep, that works!!!

The resistors are BC components (formerly Philips) VR68 series, 100K. These 1 watt resistors are rated for 10KV, so I used them in strings of 3 to make a 300K, 3 watt, 30KV assembly. These were joined by cutting the leads very close to the body, and soldering them end-to-end to form a butt joint. To get a good joint, they were cut with flush-cut sidecutters, to produce a square end, requiring a minimal amount of solder to fill the gap, ensuring a reasonably strong joint.  The only source I could find to buy less than a huge boxfull was from was  Allied , who do these in packs of 100 (Search for 'high voltage' in resistors and look for the 1 watt parts). Another possible part is the RV-100 series made by Micro-Ohm, which has similar specs to the VR68, but is slightly smaller. I requested quotes for the resistors from BC and Micro-ohm, the main problem seems to be buying sensible quantities - the Allied ones are US$39 per hundred, and I had a quote of UKP0.09 each from a UK distributor, but with a minimum order of 2000 (although BC say they are boxed in 500's) and 11 week leadtime! Here is another source of these in New Zealand (NZ$0.549 each at 120 off Nov 2003)

Tower Two - A Million Volts...!

Marx 3 is now complete, with a second tower for bipolar operation, giving 1 million volt discharges between the two towers (-500KV to +500KV).
Best spark length to date is about 54 inches (1.37m)!

Nearly end-on view of spark

As an experiment, I did try running both towers end-to-end in unipolar mode (+1MV above ground), however this was not very successful, probably due to massive corona and sparkgap losses - it only produced a fairly weak spark about 3 feet long.

Wear and Tear...

Whilst testing the second tower, I found that a significant number (about 16) of the resistors on the first one had gone open-circuit or increased in value. I don't know if this was a gradual process, or most of them went in one go. These film resistors are clearly not up to the job, despite their 10KV rating, and I have ordered some old-stock solid-carbon parts to replace them. The marxgen will actually still work quite well with some failed resistors, as the charging voltage will arc over the open-circuits on the resistors.
Update Oct 2005 : Since replacing the Philips VR68 resistors with old-stock Allen-Bradley solid-carbon composition types, the resistors have been extremely reliable, with no failures at all from what must now be a good few hundred shots.

Power Supply

The 25KV DC input to the Marx comes from a fairly simple PSU based on a flyback transformer from a PC monitor. This is driven by a variable duty-cycle squarewave signal at 30KHz. Most of the components are not too critical and were what was to hand at the time. The op-amp needs to be reasonably fast (GBW > approx. 2MHz) to get reasonably fast switching. C4/R4 set the frequency and should be tuned to maximise the output from the particular transformer. The transformer includes a multiplier, which has been simplified to a diode in the schematic. For lots of information on these transformers, and hints for identifying the connections, see Sam's Repair FAQ. This PSU fully charges the marxgen in about 7 seconds. For faster charging, you could use multiple identical flyback transformers with the outputs in in parallel, each driven by its own IRF830 driver, run from a common oscillator (parts R5,Q1,C1,D2 replicated for each flyback).

Inside view of PSU. The PSU includes an air operated switch (near top), to allow the marx to be triggered by an all-plastic actuator (right) - not holding anything metal while triggering a marx gen is a good idea as capacitive coupling will often couple enough energy into the operator to give a shock if touching something earthed! The black blob at the bottom is a the divider for the meter.

More Power..!

I decided that the first PSU above was a bit too "wimpy" for this marxgen, so I built a bigger, scarier one using four flyback transformers in parallel. I used a case from an one of several old cardiac defibrillators I picked up at an auction for the capacitors (destruct-o-tron upgrade coming soon...!). This was an ideal size, and had a carry handle and storage compartments for cables etc.

Design notes etc.

Below are various notes, thoughts etc. that led to the design of Marx Three. Sections in Bold are comments based on experience etc after building.   

Target output will be 1MV, for no other reason that a million colts sounds cool hopefully producing 3 foot sparks. This is for no other reason than 'a million volts' sounds really cool! The single tower produces 25" sparks from 25KV inout, suggesting that a bipolar twin-tower configuration would produce 4 foot sparks. I have enough parts to build a second tower but no idea when i might get the time and inclination to do it!

A bipolar (centre-fed) arrangement would reduce corona problems for a given output spark length, as input voltage would be halved. Jitter problems due to the large number of stages may become an issue. Physical size would also be more practical - 2 vertical 'towers' instead of one very tall one. I have made provision to try running 2 towers in unipolar mode to see what happens, but issues like ceiling height and mechanical stability will make this something of a challenge...!

A modular construction, allowing the number of stages to be changed in groups of, say, 5, as well as unipolar/bipolar modes to be investigated, would be very nice if it is practical. This would allow a 'single-tower' unipolar version for discharges to ground, or a 'dual tower' bipolar for discharges between towers. I just couldn't be bothered to do this!

Much better anti-corona measures will be needed in the construction - I want to be able to handle input voltages of at least 30KV. To reduce corona loss, I applied several coats of silicone conformal coating to the resistor chains, and used the thick silicone sleeving on the sparkgap wires, but there is still quite a lot of corona loss.   The H section construction would actually allow all the resistors and caps to be encapsulated quite easily, but this would make repair difficult or impossible, and the volume of encapsulant required would be quite expensive. Preventing corona from the output voltage will be pretty much impossible but I wanted to reduce the effect of corona loss draining the caps, preventing them reaching full charge at the top due to corona loss being signinficant compared to the charge current through the resistor chain. As the charge resistors are fairly low, this does not seem to be a major issue  - a meter across the top capacitor indicated it was getting pretty close to fully charged.

Jun 2002 : I Just received some 2n2 25KV russian caps from Tony Welsh and lashed up a 4-stage marx to test them - they seem to be pretty good so far...!

I have enough to make a twin bipolar system of 20 stages each, to give a total output voltage of 1MV from a 25KV input.

Nov 2002 :

I recently bought a Kilovac 35KV Sulphur Hexafluoride (SF6) gas-filled relay on Ebay - this will be ideal as a trigger device, as it has changeover contacts, allowing the charge supply to be disconnected when firing - this reduces the need for a ballast resistor to prevent continuous arcing across the first gap.  SF6 relays have the advantage over vacuum relays that there is no x-ray hazard - the only downside is that they are not good at disconnecting loads under power, however this is not a problem in this application. Using a relay trigger will make the spark gap spacing less critical - the only requirement will be that they don't fire at Vin, and do fire at 2 x Vin. 

As there would be a significant amount of construction involved, and resistors would need to be bought in quantity, I thought it would be a good idea to try to optimise the value properly, instead of my earlier 'use whatever is to hand or easily findable' approach.

I wanted to simulate the (non-trivial) charging behavour of charging the resistior-capacitor ladder, and found a free evaluation version of a schematic & simulation package called Micro-Cap, which can be downloaded here.

After some experimentation, it looks like 300K is about the right value of charge resistor to get the bank mostly charged in about 1 second. Here is a PDF of the simulation results for charge resistor values of 100, 300 and 500K. If you want to experiment with Marx resistor/cap values yourself, this ZIP file contains the the Micro-Cap schematics for a 20 stage marx, and a circuit to evaluate the effect of the charge resistors draining the output after firing.

The 300K value can conveniently be built using three 100K VR68 series resistors, giving a voltage rating of 30KV and power rating of 3 watts. Simulation showed that the avarage power dissipation in the first-stage resistors is about 1 watt over a 1 sec charging cycle, so there's is plenty of headroom there. Of slight concern is the peak power during the initial stage of charging, and it may prove necessary to beef up the rating of the first few resistors if they can't cope with the surge power - I'll try it and see.