New power supply design

A project log for Improve the Haber process

See if ultrasonic cavitation can be used to fixate atmospheric Nitrogen less expensively than the Haber process.

Peter WalshPeter Walsh 04/29/2015 at 18:057 Comments


I'm designing a hobbyist power supply for ultrasonic transducers in the 100 watt range, with these goals:

1) Automatic resonance seeking

2) Variable power output

3) Microcontroller measurement and control

4) Generally bullet-proof and safe

Circuit and an explanation are posted below.

More Difficulties

Someone at my hackerspace took the lathe offline and didn't tell anyone, so I'm blocked from making transducer horns for the moment. They won't let me strangle him, but to be fair, they won't let anyone else strangle him either.

So I'm concentrating on a new power supply circuit for hobbyists which won't burn out and has some useful features.

NOTE: This is very much a work in progress. I'm waiting on some parts, I haven't bread-boarded and tested, and it doesn't make sense to capture a circuit before you know it will work.

Also, I'm not an electronics engineer and this puts me out of my comfort zone, so if you see something fundamentally wrong with the circuit, please let me know.


Compare with the supply from eBay.

The switcher uses the same standard totem-pole configuration, but with a proper DC supply instead of cheaply-rectified AC mains.

The user must supply their own 120 watt DC supply in the 12V - 80V range, but these are *common*. A 30 volt 10 amp supply goes for $10 on eBay: it's safe, reliable, and cheap. You can even put two of them in series for higher voltages! (I've never done this, but I'm told it will work.) Switch mode supplies have isolated outputs, so it's OK to connect the outputs in series.

I'd *like* a system that would work from 12V, thus driven from a computer PSU (those are *really* common), but this depends on the specs of the SMPS transformer. I'll know more a bit later in the project.

The transistors are a wee bit overspec for the application (35A, 600V). The difference between "50% more than needed" and "insanely overpowered" is a couple of dollars (like, $2), and I want this to be bullet proof. The transistors are typically what burns out if the user makes a mistake, and we can let the fuse do its 'goddam job for once.

On the output side, the capacitor and inductor form a low-pass filter that turns the square-wave into a sine wave.

The mathematical formula for a square wave is:

\color{White} \large \frac{4A}{\pi}[\sin(f) + \frac{1}{3}sin(3f) + \frac{1}{5}sin(5f) \ldots]Where "A" is the square wave amplitude, and "f" is the frequency.

If the low-pass filter has a cutoff somewhere between f and 3f, the higher frequencies will be suppressed, presenting only the fundamental to the transducer. As a bonus, the result is slightly higher voltage than the original square wave due to the 4/pi term.

The capacitor is TBD because the inductor is a SMPS transformer and I don't know what that inductance is yet. Needs to be high voltage with low ESR - probably polypropylene.

Synchronization and Power measurement

In addition to the transducer, the output goes through an Allegro hall-effect current sensor via PWR-1 and PWR-2, from the switcher.

The Allegro chip generates a voltage proportional to the current through the transducer. It has low resistance (a few mΩ), 2KV isolation, and an 80kHz bandwidth. This one chip solves a whole lot of issues.

The signal from the Allegro chip becomes two outputs: a voltage proportional to current, and a frequency synchronous pulse train.

The lower path routes the sin wave through an active rectifier, followed by an RC integrator. The output is a DC voltage proportional to the current through the transducer, which is delivered as feedback to the PWM controller.

The upper path requires a bit of explanation. The SG2535 controller driving the switcher has a "synchronize" input that can be used to slave the internal oscillator to an external clock. The slave signal is a pulse that switches the oscillator from its charge cycle to it's discharge cycle, so to synchronize the chip to the transducer we need to generate pulses at the *peaks* of the sine wave signal. That's what the top path does.

Referring to the signal images in the circuit, the sine wave is presented to both inputs of an LM339 comparator (typo'ed as LM324) with one input delayed by an RC constant. Since the negative input is delayed, a rising signal will result in a positive output, and a falling signal becomes a negative output. The result is a square wave which is 90 degrees off from the original signal.

The square wave is presented to both inputs of an XOR gate, and once again one side is delayed by an RC constant. This results in a series of pulses at each transition of the square wave - at twice the original frequency.

...which is what the SG2535 wants to see for synchronization.

(If I have interpreted the datasheet correctly. There's *very* little information available about this chip, and as far as I can tell no example circuits on the net use the synchronization feature. And if they did, they would be slaving one SG chip to another, without noting what signal that is.)

PWM control

The power and synchronization are presented to a UC3525 PWM management chip, which controls the switcher (first image) via PWM-A and PWM-B.

The UC3525 has an onboard 5.1 volt reference, which runs through a digital potentiometer controlled by the microcontroller. The voltage chosen by the micro is compared to the current signal from the transducer, and used to adjust the PWM width. As the transducer draws more current, the voltage goes up and the PWM controller reduces the pulse width to compensate.

Negative feedback keeps the system fixed at a power level calculated by the micro, and the sync input keeps the frequency locked to the transducer/system resonant frequency. From the point of view of the micro, it's "set and forget".

Labels "UC-xxx" represent connections to or from the microcontroller. For example, the micro can turn the output on or off using the SD ("shutdown") input to the switcher.

Possible issues

My biggest issue ATM is finding vendors and specs for SMPS power transformers, as found in computer power supplies. So far as I can tell, no standard vendor sells them (DigiKey, Farnell, Mouser, Coilcraft, et al). Lots and lots of sellers on AliBaba, but no one has specs.

I'd *like* to say "pull two transformers from old computer PSUs", because that would be an easy source for the hobbyist, but I suspect there's a lot of variation and being able to tell which ones are good to use is not a casual skill.

I'd *also* like to point to a list of component vendors, in case the hobbyist just wants to buy the parts and put it together, and maybe sell a parts kit or something. That's hard without a specific vendor and proper specs.

If anyone knows where to get SMPS transformers cheaply (meaning: not $50 each from a specialty brand) please let me know.

If anyone sees a flaw in the logic, also please let me know.


Engineer wrote 05/01/2015 at 06:20 point

You wont get a higher voltage with the LPF than a source square wave. The 4/pi you talked about is actually pi/4 when you put the equation in the right direction.

  Are you sure? yes | no

Peter Walsh wrote 05/01/2015 at 07:27 point

I'm willing to believe you, but I just checked with WolframAlpha. Also, there's this TI app note:

The app note states on the first page (bottom of 1st column) that filtering a square into a sin gives a higher amplitude than the original square.

So... are you sure? I'm willing to believe you, but can you point me to a reference that explains the math?

(And either way, thanks for looking over the circuit. I'm very interested in what people find.)

  Are you sure? yes | no

Engineer wrote 05/04/2015 at 01:25 point

I skimmed the PDF, it's talking about BPF not LPF.

I may be wrong too, but intuitively, a LPF creates a decaying exponential in response to step changes. The exponential never reaches the top or bottom of the square wave, so I don't see how it could have a larger amplitude...

Looking again I see that you have just mislabeled the circuit as a LPF, it is in fact a BPF.

  Are you sure? yes | no

Peter Walsh wrote 05/04/2015 at 02:16 point

There's definitely some dissonance here, but I'm not sure whether it's you or me. Since I'm working up the electronics this weekend, I'll do some experiments and see what the results are (and report back).

Googling "LC low pass filter" and looking at the images section shows my circuit exactly - an inductor in series and a capacitor in parallel. The "bandpass" version has both the L and C in series with each other (and the output).

I'm taking the output at the junction of the L and C. The PWR-1 and PWR-2 are effectively shorted, so the transducer is in parallel with the C, not in series.

Are you *really* sure that's a bandpass filter? Because the L and C really look like the lowpass filters on google...

  Are you sure? yes | no

Peter Walsh wrote 04/30/2015 at 18:36 point

I can purchase an inductor for the filter, but I still need a transformer. The transducer needs a couple of hundred volts to get 100 watts through its impedance (about 1K ohms), and the easy way to do that is using a transformer.

It's running at around 30KHz, so it needs to be an SMPS transformer. These are common - there's one in every switching power supply - but I can't seem to find any vendors.

Any ideas?

  Are you sure? yes | no

Jarrett wrote 04/30/2015 at 15:09 point

Why are you looking for a transformer? Don't you just want an inductor? I don't believe there are any special considerations required, other than value, current rating, and form factor.

  Are you sure? yes | no

Peter Walsh wrote 05/01/2015 at 07:28 point

I can purchase an inductor for the filter, but I still need a transformer.The transducer needs a couple of hundred volts to get 100 watts through its impedance (about 1K ohms), and the easy way to do that is using a transformer.

It's running at around 30KHz, so it needs to be an SMPS transformer. These are common - there's one in every switching power supply - but I can't seem to find any vendors.

Any ideas?

  Are you sure? yes | no