Fast transient load for DC-DC converter feedback-loop stability testing?

[David_]

Hello.

I am getting quite close to start to test a couple of DC-DC converters and for that I need a adjustable/programmable DC load circuit which there are plenty of to find online.

But I also want to test the DC-DC converters feedback-loop characteristics/stability which I gather I can do with a circuit that can supply transients to ether the input or output of the DC-DC converter, transients with very fast rise- and fall-time(in this thread I am going to be talking about this assuming it is related to applying a transient current to the output of the converter but I would love to hear whatever anyone knows to tell about circuits to apply transients to the input as well).

I'm not sure but I think that what I have written about above can be correctly called a DC-load circuit and a AC-load circuit, but the problem I have is the AC-load circuit.

All examples of such a circuit I have found are circuits which simply switch between 2 load resistors(or adding them in parallel) using N-channel MOSFETs which makes the circuit not adjustable in any way and to change the amplitudes or the vales of the current pulse to be drawn from the converter I have to de-solder the old values and then solder in new resistors with new values.

I sort of feel that there has to be a more versatile way of accomplishing this, but I haven't found one nor can I think of one.

I think the problem is the fast rise-/fall-times needed, maybe there aren't a way of adjusting the load presented to the converter output while also creating fast rise-/fall-times.

Is my best bet to get an adjustable fast transient converter tester circuit to make a circuit with a bunch of different value resistors and jumpers to select from them?

I don't like that idea since the resulting currents will depend on the output voltage of the converter under test, but if that is all I have to choose from then I will have to live with it, big high-power potentiometers are expensive aren't they? and there inductance would mess with the fast rise-/fall-times right?

What about a selection of 10 resistor values each with a relay or FET-switch and a MCU to control which of the resistors will be active and make some sort of table with all the different values possible by paralleling different combinations of these 10 resistors?
I have a hard time imagining what sort of range could be created using such a method.

As for power levels, I would be happy with anything that could reach 5A testing a 12V output converter, but 1A would be better than nothing.

I'm not pleased with this text as it is but I simply can't spend more energy writing this so I apologise if it hard to read or follow.

Regards.

[CopperCone]

To test a transient on a loaded converter I think you would load it through a big inductor then basically discharge a capacitor on the converter between where the load is? This would test immunity against transients damaging the output of the DC DC converter.

Or to test a sudden impedance variation connect a capacitor because that will be seen as a sudden low impedance on the circuit thats non permanent. This would be a momentary short circuit basically. I.e. set various DC loads and short it out with capacitors under different load conditions.

I think a short circuit is basically the worst case, and if it can handle a short it can do better with anything else.

[jbb]

Yes, using switched resistors does lead to a variable current, so some fiddling is required. Additionally the resistance will provide some damping to the power supply output. Using multiple resistors and switches would allow for automatic resistor slection (which is very nice) but still doesn’t address th3 damping issue.

In principle, an active DC load would be nice (controlled current source) provided it has enough bandwidth. But that could be hard to achieve with reasonable components, and you might get nasty interactions that cause oscillation.

Use of a switched inductor system might be effective but involves some careful thought. I guess you could use a large value inductor (to get constant current during the test) and first precharge it to the desired test current with an auxiliary circuit before switching it over to the device under test.

I suspect (but have no evidence) that the professional loop stability testers may use a mix of an electronic load (for base load) with some kind of current pulse source that is capacitor coupled to the power supply output for the load steps.

[BravoV]

Not sure if this fits in your need -> Dynamic Electronic Load Project , designed by a respected forum member Jay_Diddy_B.

[basinstreetdesign]

The last time time I tested the transient load response of a switcher was for a small booster that I built for a small tube amp.  The booster supplied about 10W at 130Vdc.  I used Jim Williams advice in his LT app note AN25 in Appendix B.  I used a HP pulse generator through a R-C to the output of the booster and tweaked the compensation values of R-C for optimum response.  The compensation values were supplied through decade boxes.  Once hooked up it took me about 60 sec to settle on values.   

[T3sl4co1l]

Load resistor.

Power transistor.

Function generator.

Stick 1 to 2 to 3 and let it rip.

No need for gate drivers, a func gen is more than fast enough.  You only need microsecond rise times.  Check output on the scope first, set it for 0/10V output, variable freq and duty cycle.

You may still want an R+C across the transistor, to dampen inductance if using wirewound resistors.  Just check that the drain peak voltage isn't crazy.

And that's it!  Put in another resistor for preload, calculate them for, say, 50 vs. 100% load, say, and that's your test.

Tim

[Ian.M]

With a fast enough OPAMP in the feedback loop, and very careful layout, a classic MOSFET electronic load circuit should be able to manage a rise or fall time of the order of 1us for a 10% step change in its control signal.  The OPAMP needs to have a high slew rate, and be unit gain stable.  The feedback loop needs to be slightly more than critically damped - tuned for flattest top after the leading edge, with minimal overshoot, and its essential to low pass filter (or otherwise slew rate limit) the control signal just enough to prevent the OPAMP slew rate limiting.   

There's no point in testing a SMPSU with a load change faster than one cycle of its operating frequency, so a classic MOSFET + OPAMP load + a function generator should be good enough for testing anything with an operating frequency up to the high hundreds of Kilohertz.  However if you want repeatable results for fast load changes, a fast sample and hold synchronised to the switching frequency, with an adjustable delay may be needed between the function generator and the load's control voltage input, so you can control the relative timing of the load step within the converter's cycle.

Edit: corrected brainfart - max SMPSU operating frequency for meaningful testing out by three orders of magnitude! 

[T3sl4co1l]

You also don't want to be too fast, because of stray inductance between the regulator output/feedback terminal and the load being measured.

There's example of this in the blackdog thread, https://www.eevblog.com/forum/beginners/how-does-blackdog_s-psu-work where the measured waveform is spikey when measured at even a slight distance, because the measurement is very sensitive.

On a recent project, I created my own ad-hoc standard for output impedance, for a number of reasons:
1. The power supply output impedance was a critical aspect of the design;
2. The output was well filtered, and filters have a characteristic impedance;
3. The load will be connected through some distance regardless, so some allowance must be made, on the load, to account for the source impedance;
4. The initial project was powered by hookup cables of modest length, so we knew there was a maximum allowable inductance of at least as much.

The way I tested to the standard, was to use the load-step circuit (idle load resistor || switched load resistor), and RC-filter the output voltage, at a given -3dB point, to obtain the measurement.  The filtering ignores high frequency properties of the internal filter structure, and allows me to approximate it as an RLC network.

Finally, I solved the RLC values, by fitting a curve to the waveform.  This gave reasonable values for inductance, when the supply compensation was reasonable.  (When compensation was not reasonable, it stood out.)

A rather lengthier process than done for most supplies, but every step of the way is perfectly reasonable, and the data are good and useful.

Tim

[David Hess]

I do it the way T3sl4co1l suggests for large signal testing.  The loop response of even a high performance regulator is not fast enough to require anything more complex.  (1) High frequencies are handled by the regulator's output capacitance and the decoupling capacitance at the load.  When I need to do this, I buy 10s of power resistors to put in parallel for the loads.

For small signal testing, coupling the output of a function generator through a resistor to limit the current (and act as a termination) and capacitor to block the DC while applying a variable DC load allows testing at different operating points.  For big regulators an amplifier will be needed at the output of the function generator to get a meaningful amount of test current.  This can be important with low dropout regulators where the transfer function varies significantly with operating point because the transconductance of the pass element varies with current and it has voltage gain.

(1) Some point of load regulators are an exception.  I think one of the Linear Technology application notes discusses how to test them.

[blackdog]

Hi,

Look at the Linear Technology App Note AN133, yes much more current than you want, but read it, so much good info!

A smaler project is Linear Technology App Note AN104 and I used the load in figure 6 many times in my LAB.
Build it like you are building a 100MHz transmittor and you wil have a perfect puls respons.

Take a look of at a topic of mine on a Dutch electronic site en use Google to translate:
https://www.circuitsonline.net/forum/view/105032

Fast puls testing is difficult, watch out fore to fast edges of your generator.
When measuring the puls response every mm wiring/trace counts, also where you are probing and probe under 90 degrees from the load wiring!

Kind regards,
Bram

[jbb]

Well, there you go. You can't ignore Jim Williams. Looks like a high bandwidth active load does the trick.

One point to note: a lot of modern switchers operate in several modes:

• Pulse skipping at very low current (huge improvement in light load efficiency)
• Discontinuous Conduction Mode (DCM) at low current (inductor current goes down to zero during each switching cycle)
• Continuous Conduction Mode (CCM) above 'low' current

The loop gain of the power supply changes a lot depending on the mode.

• Pulse Skipping uses hysteresis control, and explicitly allows a certain amount of capacitor voltage ripple. Gain is moderate. Inductor current resets to zero.
• DCM uses (typically) current mode control. Gain is moderate. Inductor current resets to zero.
• CCM uses (typically) current mode control. Gain is high. Inductor current does not reset to zero.
• Transition from DCM to CCM is governed by the physics of the circuit (Vin, Vout, inductor value, load current).

This leads to a challenge when designing the control loop: you need stability in all 3 modes, which puts limits on how much gain you can apply.  Therefore, a power supply which delivers nice step response from 50% (CCM) to 100% (CCM) load may be a lot slower (i.e. experience less voltage dip) when going from 0% to 50% (pulse skipping -> DCM -> CCM).

Compounding the situation, there may be some delay in the pulse skipping -> DCM transition due to a dead band in the control loop (depends a lot on the chip).

The upshot is that going from deep sleep to a high power state (e.g. spinning up a motor) can cause quit a big dip.  Some power converter chips feature a forced continuous mode which puts you in CCM all the time, and allows for maximum bandwidth at the cost of terrible light load efficiency.  Another approach is to apply some moderate load first to at least get the chip from pulse skipping -> DCM mode.  At work we have even deployed a little bit of ballast resistor to keep us out of pulse skipping mode (the PSU didn't run often so we could afford to waste a little power).