Oscillating DC Load

[Phaedrus]

So I have a bit of an odd project in mind...

My company sells computer power supplies. In the last few years a problem has arisen, where certain power supplies (individual units, not just series), when paired with certain graphics cards, will produce a ringing or whining noise in one or both products. It's colloquially called "coil whine", and the root cause is that the graphics card, with its large 10A+ load, does not present a plain resistive load, but instead has an oscillating current waveform of a few hundred kilohertz to a couple megahertz. This AC component can cause resonance with certain power supplies, and this resonance can become large enough to cause coils and capacitors to vibrate.

This can be ameliorated somewhat through various methods, such as winding coils more tightly, gluing capacitors in place, and other such methods. But they're ugly and not 100% reliable. I'd like to experiment to see if we can prevent this phenomenon from occurring in the first place, or at the very least testing for it before releasing a product.

So to do this I'd like to find or build a DC load that can oscillate with an adjustable frequency (between 100kHz and 2-3MHz) and amplitude. Also would be nice to be able to select between square, sin, and saw tooth waveforms. Needs to handle up to 36A @12V, max voltage 16V, max power 500W. I'll pair it with my traditional static loads to get up to 1500W.



I've only ever looked into designing static DC loads. How would you go about getting one to oscillate in magnitude as I've described?

[AndyC_772]

Should be simple enough in theory, at least in the square wave case. Connect one end of a load resistor to the drain of a suitably large N-channel MOSFET, ground the source, and drive the gate with a square wave from a function generator (or whatever else your favourite oscillator happens to be). Result: current flows through the resistor to ground when the gate is high.

There will be practical difficulties to overcome, of course. You'll probably need several circuits in parallel to achieve a large enough load and to dissipate the power without anything melting. You can parallel up the MOSFETs and/or load resistors as necessary. You may also need to buffer the gate drive signals to the MOSFETs to switch them quickly, as they'll rapidly overheat if they're kept in a partially conducting state for too long. Google "Miller capacitance" if you're not already familiar with the problem.

[Phaedrus]

Yes, I foresee cooling being a major problem. However, we also manufacture CPU heatsinks here. I figure I can fit four DPAK mosfets under one heatsink, and each heatsink can dissipate 200-300W with a powerful enough fan. It will be awkward, but easy enough and inexpensive.

[sync]

So to do this I'd like to find or build a DC load that can oscillate with an adjustable frequency (between 100kHz and 2-3MHz) and amplitude. Also would be nice to be able to select between square, sin, and saw tooth waveforms.

My DIY DC load has a modulation input. It can be used with a function generator. Very useful and more flexible than a build-in generator.

[robrenz]

I suggest taking a look at the Kepko BOP series of bipolar supplies. You can drive them with a function generator. But from memory you are not going to get above about 20kHz at 3db down.  I think those frequencies at that power level is going to be difficult/expensive even with just a sine wave input.

[plesa]

Should be simple enough in theory, at least in the square wave case. Connect one end of a load resistor to the drain of a suitably large N-channel MOSFET, ground the source, and drive the gate with a square wave from a function generator (or whatever else your favourite oscillator happens to be). Result: current flows through the resistor to ground when the gate is high.

There will be practical difficulties to overcome, of course. You'll probably need several circuits in parallel to achieve a large enough load and to dissipate the power without anything melting. You can parallel up the MOSFETs and/or load resistors as necessary. You may also need to buffer the gate drive signals to the MOSFETs to switch them quickly, as they'll rapidly overheat if they're kept in a partially conducting state for too long. Google "Miller capacitance" if you're not already familiar with the problem.

What about combinng about 5 pairs of IXYS and function generator?
EG. IXTA15N50L2 and LT1351
http://ixapps.ixys.com/DataSheet/DS100054B(IXTA-TH-TP15N50L2).pdf
http://www.ixyspower.com/images/literature/LinearL2TM.pdf
for 100USD you can build it up by yourself.

[mobiletester]

You can check this site: http://mfc-wittgenstein.net/aktionen/stromsenke/index.php

[AndyC_772]

500W is an awful lot of power to try and dissipate just in the silicon, though. I can see the transistors getting very hot indeed without a nice big, passive resistor to take care of some of the heat.

Also, without a fixed value load resistor to set the current, you'd probably want some kind of closed loop control to drive the transistors and monitor the load current somehow. Without closed loop control you'll be battling with the variation from one transistor to another, and more importantly, the change in characteristics that takes place as they warm up.

That's why I like the idea of purely digital control, it decouples the performance of the whole design from the characteristics of the individual transistors to some extent. Control becomes easy, just choose a resistor value that gives the variable part of the dissipation you're after, and drive the transistors with a square wave that's at least big enough. As soon as you go into the realm of linear operation, the circuit becomes a lot more complicated and - IMHO - not really much more useful.

[Phaedrus]

I think a resistive load switched by transistors is a good way to do it. KISS. I can live with square wave only. I'll just feed it from a function generator. The main difficulty will be switching that much power at up to 2-3MHz. Hmmmm.


I think I can use a bank of these resistors:
http://www.digikey.com/product-detail/en/KAL50FB3R00/KAL50FB3R00-ND/1646197

Mount them to a piece of scrap aluminum and get a couple of powerful fans blowing across them, should do ok. Since it's basically PWM, 48W on a 50W resistor won't be an issue.

[AndyC_772]

Those wirewound resistors can be a bit inductive. I'd be inclined to test that you can switch them fast enough before committing to a full scale circuit.

If I were building a load for this purpose, I'd probably start with a bank of thick film resistors bolted to a nice, big heat sink.

[Phaedrus]

That'll take a lot of thick film resistors. Just sayin'.

[Phaedrus]

I think I'll use four "channels" of four 5ohm, 35W resistors:
http://www.digikey.com/product-detail/en/PWR263S-35-5R00F/PWR263S-35-5R00F-ND/2620728

140W per channel, four of them gives me 560W, which is ample.

I'll make one channel first as a test and use one of those 15A IXTA15N50L2s to switch it. An ATMega to control it I think. I foresee many heatsinks, and much solder fill in my future.

[fmaimon]

Check out Jim Williams AN 133. It should fit your need nicely with a few changes.

http://cds.linear.com/docs/en/application-note/an133f.pdf

[Phaedrus]

Beauty. That's almost exactly what I had in mind from the start. Yeah, that's definitely doable, but also a lot more complex to build than the simple switched resistive load. Hm, need to think on it. Printed that AN out, I'll read it in bed tonight. 

[Phaedrus]

"Power Uncorrupted" with apologies to Lord Acton...

Jim Williams was a boss.

[Harvs]

I have done something similar-ish in my graduate project (a fast high current multi-channel arbitrary load.) 

The biggest problem when you start to go fast is inductance, and I don't think that app note really give this problem practical justice.  JW discusses placing only 20nH of parasitic inductance in the Drain, be aware that's only about 1" of PCB trace.

Every now and again a post pops up on this forum about someone's ELoad being unstable, and usually it's meet with adding some gate resistance and slowing down the loop response by providing derivative feedback.  With the idea that it's the capacitance of the FET's gate that is causing the circuit to be unstable.  In pretty much all cases, it actually isn't the root cause (go plug it all into spice with an LM324, you'll see you can never get less than about 45 degrees phase margin with the gate capacitance of a normal power FET).  The real cause is the inherent capacitance between FET drain/gate, and it's interaction with the lead inductance.  This is why something like the HP 6030B uses a snubber network to dampen the inductance.  However, that's not going to work a HF as the snubber will just absorb all the current you want to put on the PSU.

So, I recon this is going to be a really non-trivial problem to make it a closed loop system, unless you can do it without the leads on the power supply.

[Phaedrus]

Damn. With the power supply leads, breakout board, and cables to the load, I'm probably looking at >1000nH. I might minimize that to 500nH, but there's no way I'm reducing it to <20nH.

On the other hand, I only need <40A amplitude, not 100A.


I can give up on 2-3MHz. The majority of the problem is in the 100-500kHz range anyway.

[AndyC_772]

That'll take a lot of thick film resistors. Just sayin'.

http://uk.farnell.com/bourns/pwr221t-30-10r0j/resistor-thick-film-10-ohm-5-to/dp/2328272

30W continuous rating each - not that many, surely...?