Close

Prototype

A project log for Nixie Tube Power Supply

A flexible power supply for Nixie tubes. Input voltage can be from 3V to 12V. Output up to 200V. It uses commercially available transformers

paul-andrewsPaul Andrews 11/03/2017 at 02:192 Comments

After I had spent a month or so messing with LTSpice, I decided that I really needed to build something to try it out. From what I had read, trying to prototype this power supply using a breadboard would not be a good idea. Breadboards have too much additional resistance, capacitance and inductance that would just mess up the results. The next best thing, I decided, was to use perfboard - with this I would hopefully suffer less from all the drawbacks of breadboard.

I started by designing an adapter for the transformer and having it fabbed at OSHPark. I also ordered a SOIC8 adapter from there for the LM3478. This makes it easier (i.e. possible) to incorporate them into perfboard. Then I ordered all of the parts I thought I would need from Digikey. For the key components such as the output capacitor, the current sense resistor and the loop compensation resistor and capacitor, I ordered the values from the simulation (or close to them) and a few either side. It turns out that this was really important - the performance of the physical circuit was very much affected by some of these values.

This was the wiring diagram for the transformer:

This is what it looked like all wired up:

It looks a real mess in that picture, because that is what it looks like now, after it had been inserted, removed and re-wired several times. It still works though!

This is what the finished prototype looked like (or looks like now!):

There are some parts on the reverse side too. One important note here: It is obviously important to pay attention to the specs of the parts you order. Wattage on resistors. Voltage and ESR on capacitors. But one that is often overlooked is the voltage rating of the resistors. Typically this is around 100V, however the feedback resistor has pretty much the entire output voltage across it permanently, so that specific resistor is rated to handle it.

An interesting one is the current sense resistor. I ordered some wire wound resistors, which was a mistake because they introduce inductance to the circuit. The non-wire wound resistors worked fine.

I think I over-spec'd the diode!

What followed was a comedy of errors. Firstly, I had the output stage connected the wrong way around. Remember this diagram?

Those dots are important. I had them the wrong way around. I fixed it by swapping the inputs and presto, I was in business. However, I most definitely was not getting 186V, except with no load (actually 184.6V with no load. Near enough). BTW, this is what my test load looks like:

Three high power resistors connected in series. I had considered building a test load, but I didn't want to get diverted into yet another project!

I spent the next few weeks changing the values of various parts. One of the key ones turned out to be the frequency setting. I started with a resistor of 24k, which should have given me a frequency of about 600KHz. I gradually reduced it, and as I did so the performance kept improving. I ended up with a resistance of around 47k, which gives a frequency of about 350KHz. At that frequency, varying the other values didn't make much difference. However I noticed that the IRL640 was getting really hot. Apparently LTSpice doesn't model the thermal characteristics of the components. I could actually see the output voltage dropping as it warmed up ('warm up' is an understatement!).

My reasoning was that the Rds(on) of 0.18Ω was just too high, so I searched for a replacement with lower Rds(on) and came up with the IRLI2910. This has an Rds(on) of just 0.026Ω. When I swapped this in to the circuit, it never got warm and my output voltages improved - less power was being dissipated as heat. An added bonus, Infineon have a spice model for it, so I can easily plug it into the simulation.

The values still weren't that great though. Still not getting to 184.6V! While I was taking measurements to calculate efficiency, I just happened to notice that if I measured the input voltage at the board instead of at the power source, it was a full 1V less. My 5V source was a USB wall adapter. With a load, this voltage was dropping to around 4.6V. With the additional 1V drop, this was down to 3.6V! The culprit was one of the leads I was using. I junked it and my performance improved dramatically.

So how was it? These are the values I get from this prototype now, using the IRL640 (which still gets very hot, but that's OK for the brief time I am taking measurements):

12V inVolts outCurrent out (mA)Power  out (Watts)
20K load18091.62
15K load181122.17
10K load181.5183.267
5V in
20K load1738.71.5
15K load1419.41.33
10K load11611.61.35
3.7V in
20K load14571
15K load1318.71.4
10K load11411.41.3

Well. That isn't too great, but it can light up a bunch of IN-12s - six in this case (this is 5V in, 171V out):

Yes, I was monitoring the drain voltage on a 'scope. I'll get into that in the next log - you'll have noticed the capacitor/resistor lash-up across the inputs to the transformer, right?

This is 4 tubes at 176V from a LiPo battery. This is pretty much my minimum performance requirement. The whole point of this exercise is to get something that can run a battery powered clock. So I got to there, and with something that I didn't expect to meet the full performance of the LTSpice model, because it was lashed together on perfboard:

What you can't see is the FET's heatsink glowing red. OK I exaggerate, but it was really hot. This is the IRL640 remember - the IRLI2910 had no such problems and performs better. It was certainly encouraging and worth going on to the next step of designing an actual PCB. But first, that scope output...

Discussions

Paul Andrews wrote 11/11/2017 at 22:13 point

Their model is encrypted. I have add a diagram to the files section of this project for one of the transformers in their package. This is taken from their documentation for the LTSpice models, but it is for a different transformer than the one I used.

  Are you sure? yes | no

Johan Carlsson wrote 11/11/2017 at 21:13 point

Regarding your comment on poor performance at 600 kHz, but better at 350 kHz. Could it be the SRF (self resonance frequency) of the transformer that's the cause? Does the spice model of the transformer include its stray capacitance?

  Are you sure? yes | no