Gekko Science
Contact Info
Weekend Builds
Nixie Tube Power Supply

Build by Kittan           kittan |at|

Gimme a break, I'll get pictures up at some point. Maybe even some schematics and photos of the device in operation powering some tubes.


We acquired for a project (stay tuned for future updates!) a case of old Russian nixie tubes and a case of old Russian nixie driver chips. These tubes run off 150-170VDC, so the first challenge was creating a power supply to drive them. The obvious test option is to rectify mains, which yields somewhere around 160V to 170V when filtered. But that's boring.

The project these tubes are for will be powered entirely by a 7.5V 1A DC adapter, so I went ahead and built a power supply taking in 7.5V and sending out 150V. There's still some tweaking to be done on it to get efficiency up, but currently it's stable and functional in prototype.

Nixie Tube Driver


Topology tests

I've built a bunch of fixed-output and adjustable linear regulators in the past, some from monolithics like the LM317, some simple emitter-followers with a fixed zener reference, and some fully-regulated using voltage feedback and error amplifiers with adjustable reference. My favorite was a 24V 20A whose pass transistor was actually regulated through a totem-pole output off a 5V op-amp and a rewired microwave oven transformer as the AC source. But all that was child's play compared to switching regulators - low efficiency, one-way conversions and simple DC mechanics.
I haven't really done much with switching regulators in the past. I have a lot of conceptual knowledge from classes, books and fixing several thousand LCD monitor power supplies in the last few years, but I had never really designed one of my own. So I decided to start small and work up from there - basically, I wanted to get a feel for "buck" and "boost" topologies before settling on the final design.
For those of you not familiar with switching regulators, "buck" generally refers to dropping the output voltage below the input voltage, and "boost" raises the output above the input. The obvious choice for a power supply taking in 7.5V and putting out 150V would be a boost, as the output is higher than the input, but that's not always the best option once you start inserting trickery.
First, a little about inductors and capacitors. Capacitors are voltage devices; that is, they store energy in an electrostatic field by pushing electrons in one side and out the other. The two leads of a capacitor are electrically isolated - nothing passes between them except pure electrical energy. In a power supply, capacitors act as either sources or loads, to try and maintain a constant voltage in the circuit. Large changes in voltage applied to a capacitor can result in large currents from the capacitor in order to try and maintain status quo. Inductors have a very similar operation on current; they store energy in an electromagnetic field by running a current through the inductor. They act as sources or loads trying to maintain a constant current in the circuit. Large changes in current applied to an inductor can result in large voltages from the inductor in order to try and maintain status quo. A typical buck converter basically works by switching input DC very quickly and then running it through a low-pass filter. The resulting voltage is equal to the average input voltage - basically, if you integrate the input voltage over one period and divide by the period length. Hey, let's explain that with a picture.

As can be inferred from this, the duty cycle (on-time for the switch, as a percentage of cycle length) is equal to the percentage of the input voltage reflected in the output.
A boost converter uses a lot of the same parts but in a different configuration. In a buck, the inductor is used to slow down current when the switch is on, and source current when the switch is off. In a boost, the inductor "charges up" when the switch is on, and then pushes that stored energy into the capacitor - pumping "up the hill" if necessary; that is, even when the capacitor's voltage exceeds the inductor's initial voltage, its voltage will increase to whatever level is necessary to push its current somewhere, as the inductor will work to keep current flowing.
Might need another diagram here.

The duty cycle of a boost transformer is more-or-less inversely proportional to the ratio of input and output voltages. Specifically, the off-time of the switch is directly proportional to said ratio, so the on-time Don = 1-Doff. What this means is, for a greater difference between input and output voltages, the on-time duty cycle becomes very large. Parasitic effects in the circuit reduce efficiency at very low and very high duty cycles. In our power supply, Vin/Vout = 7.5/150 = .05, so Don = 95% of the cycle. For a basic circuit this is far from optimal.
Another way to get high voltages out of a switching regulator without relying on a boost, is to introduce a transformer to the circuit. This sounds crazy since it's a DC circuit, but just wait. A transformer is basically two inductors with mutual magnetic flux. One winding induces current in the other winding by changing the flux linking them, in the same way a single inductor becomes a voltage source when the current through it drops. Transformers only really care about changes in flux, increasing or decreasing, and that's exactly what a switching supply provides.
A simple yet reliable topology replaces the inductor in a buck converter with a transformer and rectifier; this is referred to as a forward converter. I opted to test out the push-pull topology, which actually uses two switches driving alternately to force current back and forth through the transformer winding to trick it into thinking it was getting AC, and spitting out a higher voltage on the secondary that I could then work with.

Transformer construction

One of the benefits of switching power supplies is high-frequency operation. Mains-current transformers are huge, because they operate on very low frequencies (50-60Hz depending where you are) so the cores have to be very large in order to not saturate during the 16-20ms that current is flowing in a given direction (before reversing itself and, therefore, degrading the stored magnetic field and rebuilding it in the opposite direction). Switching power supplies can be built to run in the KHz and MHz ranges, which gives microseconds to nanoseconds of field buildup and therefore several orders of magnitude less required field management per cycle, which means a much smaller transformer core required for a given power throughput. I opted to drive my power supply at 100KHz, which is well above audible frequency range so even if it causes mechanical vibrations nobody will hear them, and gives a maximum cycle time of ten microseconds meaning a very small transformer can be used.
For a push-pull converter, you need a center-tapped primary. The center is directly powered, and the end of each leg of the winding can be switched to ground. This causes the current to flow from the center tap through that winding and out, which actually (because of autotransformer effects) induces a voltage in the other half of the winding while building a magnetic field. When the switches are reversed, the current flows through the other leg, which means reversing the direction of current flow and therefore reversing the direction of the stored magnetic field. The secondary winding sees this reversing of the field and counters it with induced currents of its own; the result is an output voltage proportional to the driving voltage roughly amplified by the turns ratio.
I tested initially with a transformer using a 30:1 turns ratio and center-tapped secondary, wherein my primary was 10 turns center-tapped, and the secondary was 300 turns center-tapped. With a center-tapped secondary and full-wave rectifier, only half the secondary is active so its turns ratio is 150:5, reduced to 30:1. I built it around a 14mmx8mm bobbin-core two-piece ferrite, and amazingly it worked exactly as expected. Unfortunately the minimal-load duty cycle was in the "sketchy operation" range and, even at a full current load, the driver was giving it about 20% duty cycle. The ratio was too high (remember, voltage out roughly equals voltage in, times turns ratio, times duty cycle), which would wreak havoc with efficiency and possibly cause audible noise. For the sake of ease of construction, I decided to use a full-bridge-rectified single secondary, and to get the duty cycle into a better range I dropped the turns ratio to 22:1 - the result was a transformer with 30 turns center-tapped and 330 turns in the secondary. Yes, those numbers don't quite add up - but remember that in push-pull, only one half of the primary is active at a time so it's actually 330:15, or 22:1. This turned out much better in every way. The reduced turns ratio put the control's duty cycle in a much more comfortable range; the single secondary was much easier to wire; and the increased number of primary turns helped increase mutual inductance coupling between the primary and secondary.

Output stage

The output from the transformer is pretty much a voltage-amplified reflection of the input. The single secondary is bridge-rectified, so the output from the secondary is a superposition of the switching pulses from both transistors - meaning that the output is a pulsetrain at 100KHz, duration equal to the on-times of the transistors. This pulsetrain is fed into an LC circult just like in a direct buck regulator; in fact, the output from the transformer is basically identical to the output of a buck regulator operating at extremely high voltage.

Driver choice

There are probably a lot of integrated or monolithic push-pull driver options out there, but I opted to play with an old fallback the TL494. I'm not sure how long ago this chip was designed, but jillions of power supplies have been built around it (and its pin-compatible counterpart the 7500B). This chip is basically the 555 of switching supplies - it is incredibly versatile, with a lot of "internal" signal lines tied to output pins to modify its operations. One of the nifty things about this chip is the completely undedicated outputs - it uses two internal NPN transistors as output drivers, with both the collector and emitter tied to output pins. You can use these for either high-side or low-side drives depending on your topology (they're built to handle 200mA each) and, by sending a control pin high or low, you can toggle whether they work in parallel or alternating. The switching frequency is set with an external resistor and capacitor, and it has two error amplifiers so you can regulate outputs based off two different factors (such as, constant voltage with a maximum current limit). There's also a "dead time" pin through which you can control the maxumim output duty cycle, effectively giving you (with some external parts) a third regulation factor, or the ability to tweak timings to suit particular needs. There's also an internal voltage regulator pinned out for creating reliable reference voltages. Terribly handy.
One thing to consider when selecting components for setting the operating frequency, is the difference between parallel and push-pull mode. In parallel mode, both transistors are timed together off one cycle of the RC oscillator; in push-pull, each transistor is alternately timed off every other cycle so each transistor maxes at 50% duty cycle at half the clock frequency.

Transistor driving

I wanted the output to be able to source at least 20mA at 150V, which equates to 3W output. This means, at 100% efficiency, a 3W input which means 7.5V, 400mA. 400mA is more current than the TL494's internal transformers can handle without roasting, so an external transistor driver is required to directly power the transformer. Something that can switch at 50KHz (as each transistor only sees half the actual clock rate), and is comfortable with at least 400mA current. The initial choice is, do I use a FET, or an NPN? FETs are nice for low power dissipation and efficient switching, but there's a trick to push-pull regulators that makes them undesirable.
In a push-pull, it is assumed that the two halves of the transformer are switched for exactly the same length of time in a given cycle, which means that after one cycle, the mangetic flux has increased one way, then increased the other way, and ends up at the same point it started. If the timings are off just a little bit, the flux endpoint can start to skew, and eventually push the transformer into saturating the core. This causes a current imbalance on each half, which can make components go up in smoke. BJTs have a positive temperature coefficient of gain, which means when they get hot they actually work better (to a point): if the core flux gets imbalanced, more current will dump through one leg than the other, increasing power dissipation in that drive transistor; if it's a BJT, its gain will increase, allowing more power to flow through it which will pull the core flux back into balance. If a FET (which has a negative temperature coefficient) is used, then as it warms up from power imbalance it will actually cause the imbalance to increase - which could be destructive.
I did my initial tests with a J44H11 NPN transistor, which I had previously fetched several of for an unrelated project. The final design ended up with a 2DC4672 surface-mount NPN, which spec'd out as having similar if not better current gain, switching speed and per-unit price. In the end it didn't make much difference, but this transistor was available in a surface-mount package, which we're trying to use as many SMD devices as possible to make assembly easier.

Feedback loop

The TL494 obviously can't handle 150VDC applied directly to its inputs; it also can't sample and regulate 150VDC directly while being powered from a 7.5V source. So I set up a voltage divider on the output in order to sample a constant portion of the output voltage, and fed it back into the error amplifier against a reference voltage derived from the internal 5V regulator.
Error amplifiers, within the context of voltage regulators, provide an output signal proportional to the difference between its two inputs; this output is fed into a PWM driver whose output duty cycle changes proportionally to the error amplifier signal. The error amplifier seeks to make its two inputs (the reference voltage, and the sampled output voltage being fed back into the system) equal. If the feedback input is higher than the reference, the output will force the PWM lower, shifting the output voltage lower; if the feedback is lower, it will force the output higher.
I used a 100K in series with a 1.5K, such that I was measuring 1.5K/101.5K of the output voltage - about 2.2V at an output of 150V. Using a trimmer potentiometer as a voltage divider between the 5V regulator output of the TL494 and ground, I could set the wiper to arbitrary voltages with the turn of a screwdriver. This changed the reference voltage on the error amplifier, which changed the driven duty cycle and therefore the output voltage.
Initially I designed the output of the regulator with a 100K load resistor in parallel with the 101.5K feedback resistor, but these two strings resulted in an effective constant resistance of about 50K, which drew 3mA even when no external load was present and represented a net loss of about 450mW. Regulators need a minimum load to function properly, but considering that the regulator will only be on for a fraction of a second before any tubes are driven, I decided to nix the load resistor and replace the feedback network with a higher resistance - 1M and 15K; the same proportion but an order of magnitude less current. This change reduced zero-state load from 3mA to 0.15mA, dropping power requirements by about 95% (from 450mW to 22mW).

Performance and Efficiency curve

I set up an adjustable load using a high-voltage FET, and measured the input current requirements for load current outputs between 0 and 25mA (0 and 3.75W). Our nixie tubes should never require more than about 17mA but I decided to test it farther than that to see how it behaved. I did reconfigure the load and tested beyond 25mA, but I didn't have anything monitoring the output voltage so it's possible it sagged which throws off calculations.
In any case, what I saw pleased me - in the operational range of 8 to 17mA, the output efficiency was between 79% and 81%, much better than initially expected and, overall, a very pleasing number. As power requirements increased, the efficiency actually increased beyond that - constant loads like the driver chip and feedback dividers became a smaller part of the equation, and as duty cycle approached 50% the transformer currents got friendlier.

If you like what you saw and you want to contribute to future awesome projects, feel free to donate some BTC: 1GekkosciLeaey8Na9siC8oD5HcMtLnWwd