Question on testing/characterizing DCDC regulators (design process)

[E.Buer]

Hi,
I am interested to hear how you go about testing/characterizing DC/DC regulators that you design/use in a design, and which tests are necessary.

• Besides simulation, do you measure the stability of the regulator prototype? If so, how?
• Do you measure at different loads?
• Do you use small signal analysis, noise injecting etc. to obtain a loop response?
• Do you measure the efficiency?

Is it normal to just wing it, and assume that the regulator works like the simulation, given that the output voltage is correct?

If you could talk me through the design and development process, that would be helpful!
Thank you!

[David Hess]

Besides simulation, do you measure the stability of the regulator prototype? If so, how?

Yes, at a minimum I do load and line stability tests.  The load stability test may make it unnecessary to do a separate feedback loop stability test.

Do you measure at different loads?

Yes, and different input and output voltages if applicable.  The loop response will change significantly if the regulator crosses the continuous conduction mode boundary and this will need to be taken into account in one way or another.  In discontinuous conduction mode, the loop response changes with load.

Do you use small signal analysis, noise injecting etc. to obtain a loop response?

As mentioned above, the load stability test may be sufficient in place of measuring the loop response separately.  In addition, I do not have a dedicated network analyser; I just have never had enough need for one.

Do you measure the efficiency?

Only if there is a specific reason to.  I would for optimization during the design process.

Is it normal to just wing it, and assume that the regulator works like the simulation, given that the output voltage is correct?

Oh, I never do that.  And my simulation may only be paper calculations.  I usually only do simulation to verify or refine the design calculations and measurements.

[MagicSmoker]

This is really too broad a question with a different answer for each combination of parameters such as how big a budget (time and money) you have, how fast the transient response needs to be, whether the converter topology is prone to issues to begin with (e.g. - boost or flyback-derived converters in CCM), how wide the input voltage range is, how wide the output voltage and load current range is, how rapidly can/will load current change (dI/dt), is the converter transformer-isolated, how expensive the product is, how serious are the consequences of failure, what kind of MTBF is needed, etc.

Over the years I've gotten fairly good at anticipating when a very casual design and validation approach can be used and when I actually need to employ more rigorous testing schedules. I used to be a proud member of the "hates SPICE" crowd but LTSpice made a believer out of me and now I use it for at least the first-pass workup of every new design, if not full AC/transient analysis of the loop. Regardless of how much simulation you do, though, you still have to verify the circuit operates correctly in the real world. Again, whether that entails full-blown frequency response analysis with a tracking generator and signal injection transformer or just toggling a bank of resistor loads on and off and looking for ringing on the scope depends on the parameters mentioned above.

For example, non-isolated buck converters - especially if slope-compensated current-mode control is used - are so easy to get working and so forgiving of their loop compensation components that you would arguably look incompetent if you had to resort to real-world frequency analysis. In contrast, a transformer isolated CCM flyback is practically metastable by design, so unless you slug the hell out of the control loop it will almost certainly have combinations of line and load (especially load dI/dt) that are outright unstable.

[David Hess]

Often with the designs I work with, I end up doing the SPICE simulation after measuring the loop response because there are too many unknowns which I have to derive from actual measurements.

[ocset]

The Ridley engineering group make a gain phase analyser  ( AP300) which is good for measuring the gain and phase margin of an SMPS…
It can be used very quickly in most cases.

http://www.ridleyengineering.com/analyzer.html

The following 4 videos go through use of the AP300 frequency analyser, which you can use to measure gain and phase margin of the SMPS feedback loop.

[MagicSmoker]

The Ridley engineering group make a gain phase analyser  ( AP300) which is good for measuring the gain and phase margin of an SMPS…
It can be used very quickly in most cases.
...

Yeah, frequency response analyzers like this are real time savers (and, potentially, ass-savers) but not really justifiable if you are the one having to pony up the cash for it. Last time I checked the AP300 + isolation injection transformer cost something like $12000.

EDIT - just looked on the website and it is now up to $15650 for the AP310, isolation transformer and software (which used to be a glorified excel spreadsheet... look, I know how long it takes to make a nice spreadsheet with lookup tables and all that, but fercryinoutloud, include that shit for free if someone buys your $14000 piece of hardware).

[David Hess]

Picotest and Cleverscope also cater to this market.  Cleverscope does away with the need for an isolation transformer by having an AWG with isolated output:

https://www.picotest.com/products_BODE100.html
https://cleverscope.com/

SysComp Design also makes some student priced low performance USB DSOs which include AWGs and support network analysis:

https://www.syscompdesign.com/

[E.Buer]

Thanks for all the great responses!

Do you measure at different loads?

Yes, and different input and output voltages if applicable.  The loop response will change significantly if the regulator crosses the continuous conduction mode boundary and this will need to be taken into account in one way or another.  In discontinuous conduction mode, the loop response changes with load.

How would you go about testing this, using electronic loads, or simple power resistors? I guess it would not matter which is used, besides switching noise in the electronic load?

[ocset]

some e-loads, often due to their capacitance, cannot always be used in feedback loop analysis
There are links on eevblog where this has been discussed before.
.............................................................
Page 9 of the pdf..

AN038 by Richtek:
 “DC/DC converter testing with Fast Load Transient”

..shows that very often, an E load doesn’t have sufficiently high slew rate (di/dt) to be able to properly do a transient test on an SMPS.
As you can see from page 9, it says that the Step Load rise time  should be much faster than the SMPS response time. The SMPS response time in the case shown, is 0.22/fc
(where fc is the crossover frequency).
So the load step rise time should be much faster (say five times)  than the SMPS response time.
............................
..that refers to transient response alaysis, but the fact is that an eload can often effectively add capacitance to the converter  output, and thus change its frequency response.
................................
https://www.eevblog.com/forum/projects/programmable-dummy-load/msg2360178/#msg2360178

[MagicSmoker]

How would you go about testing this, using electronic loads, or simple power resistors? I guess it would not matter which is used, besides switching noise in the electronic load?

I gave you a clue in my first response: switch resistors in and out of circuit. You can use MOSFETs to do the switching, or actual mechanical switches and relays. You can flip the switches, or have an MCU do it for you (allowing all sorts of interesting test patterns). Switched-resistor loads can achieve the highest dI/dt possible, but if you need to limit the dI/dt then simply insert more inductance in series with the PSU-under-test.

Continuously-variable electronic loads tend to cause more problems than they solve, in my experience. It's nice to have one, but don't try to characterize a high speed converter with it unless you like seeing the electronic equivalent of the Tacoma Narrows bridge collapse.

[David Hess]

Do you measure at different loads?

Yes, and different input and output voltages if applicable.  The loop response will change significantly if the regulator crosses the continuous conduction mode boundary and this will need to be taken into account in one way or another.  In discontinuous conduction mode, the loop response changes with load.

How would you go about testing this, using electronic loads, or simple power resistors? I guess it would not matter which is used, besides switching noise in the electronic load?

Electronic loads are not necessarily noisy but depending on the design they may interact with the frequency compensation.  Power resistors always work and you can buy really big power rheostats if necessary.

While not ideal, the output voltage of a variable output supply can be swept into a resistive load but depending on the design this may not be a sufficient test.

[floobydust]

Simulate and test I use as a guideline, not the final result for industrial products.
Don't forget to include capacitance (and sometimes inductance) shifts due to temperature.
Noobs assume the components do not change, and ESR & value shifts can cause instability.
Go beyond tests in the lab- a temperature chamber should also be part of them.

[David Hess]

Don't forget to include capacitance (and sometimes inductance) shifts due to temperature.
Noobs assume the components do not change, and ESR & value shifts can cause instability.

Something else to be wary of is transistor gain and Ft which can vary significantly between units.  It helps during development to have some worse or extreme case parts available.

If you were designing ATE equipment or pin drivers or something which has to power a difficult or unknown load, you might get esoteric and implement frequency compensation with less than 90 degrees of phase lag and 6dB/octave of rolloff but that is beyond a discussion of performance verification.  Such a control loop will remain stable into any capacitve load.

[rbola35618]

If you don't have a FRA you could use a function generator, a Scope and a small transformer to inject a signal from the function generator and use the scope to measure the phase and gain of the feedback loop...