SMPS with two control loops

I got playing around with some old ATX power supplies lately. I'm trying to make'm adjustable, and so far have managed to build the external control loop circuitry. First there is a inner loop that sets the TL494 PWM from the output inductor current. The set point to this inner loop sets it up as a constant current supply. Secondly, I have another control loop that programs the current set point from the output voltage. i.e your classic current mode control.

I would like to get some idea of the frequency response of the loop(s), but found I need a frequency response analyzer. I tried injecting a signal into the control loop as explained in:

But the switching noise makes it impossible to compare two tiny sine waves on my CRO. I can get some estimation of the crossover frequency by looking when the fuzzy sine waves are approximately equal in amplitude, meaning the frequency where the open loop gain magnitude is unity. BTW, I think this transformer injection trick is just brilliant.

Prices of such frequency response analyzers (FRA) are >= $10k , so obviously not every SMPS engineer had the luxury of utilizing an FRA to help build his/her design. I'm only a hobbyist.

The current loop is stable, 0 to 25A at short circuit to open circuit, and I measured crossover point at 10kHz with about 50~80 degree margin (using transformer injection method), giving 1/10th the switching frequency as recommended by the experts. The current error amplifier has a flat gain after a zero located at 1kHz. This gives good current tracking at lower frequencies. All is well...

The voltage loop is causing problems, I cannot get it stable at light/zero loads. Am I in a situation where I'm best to plug in various caps/resistor values around the voltage error amplifier until the thing just works ?

I've given up in vain trying to mathematically predict gain and phase margins of the two control loops using Scilab/Octave software. I'm guessing this is how all the real power supply designers do it. Ok, I know I'm asking a lot out of a SMPS capable of 0% to 100% load change, adjustable constant current and constant voltage. Would this be considered as a challenging task ? I'd like to know if I'm wasting my time.

Regards,

Adam Seychell

Adam,

1) you should be able to make a low pass filter so the scope can see the injection signals without the switching frequency so strong.

2 There is a way to use op-amps (instead of a transformer) to do the injection trick so that you can see the open loop response of the loop directly without really opening the loop.

Look at the company web site of Venable.

have fun

Mark

In article , [...]

If you have a digital storage scope, triggering the scope from the signal and using the averaging mode in the scope works. You have to fiddle the frequency a bit to get the minimum noise in the sum.

You can make a simple commutating filter using a CD4052, if you can create two squarewaves at the signal frequency, with a 90 degree phase relationship:

Scope ! -------- C2A ! ! X0 !-----!!--GND C1 R1 ! ! ! C2B Vi-!!-/\\/\\/\\-+--!In X1 !-----!!--GND ! ! SQR0---!A X2 !-----!!--GND SQR90--!B ! !en X3 !-----!!--GND --------

You get a steppy waveform on the scope. The nice thing is that the phase between the sine wave and the squarewaves doesn't matter much. You calibrate by hooking Vi to the function generator. You pick off the voltages for (T0 and T180) and (T90 and T270).

The R1,C1 time constant should be low enough to let the desired signals through. R1 needs to be large enough that the currents in the 4052 are kept below about 1mA.

The R1,C2 time constant should be long.

--
--
kensmith@rahul.net   forging knowledge

In article , legg wrote: [...]

Beware that you don't want a cycle skipping pattern that looks nice on the oscilloscope. When the pattern looks nice, it has a lot of energy at some sub-harmonic of the cycle frequency. This can lead to EMI troubles. This is one case where chaos is your friend.

--
--
kensmith@rahul.net   forging knowledge

Sounds like you are using the transformer injection scheme and measuring the phase relationship one side relative to the other.

HP used to produce a very nice gain phase analyzer and they are only a couple hundred bucks used on Epay. Dean Venable used to have a bunch of decent app notes regarding making the measurements with this HP gain phase analyzer.. Since his company (ex company?) began producing the phase measurement gear his app notes were revised and only mention his gear.

Ken's suggestion of DSO avaraging is good way. Switching noise is avaraged out. You might need an expensive scope to get there tho.

Build yourself a pair of LC filters in a box having the same exact response. Send signals from both sides of transformer thru it's own filter and then when they come out the high frequency noise is gone but the phase and gain information is still valid.

regards, Bob

In article , Yzordderrex wrote: [...]

It depends on what you call expensive. The Techtronics TDS-200 can average.

--
--
kensmith@rahul.net   forging knowledge