Does there exist a ready-made design for implementing the measurement hardware? You can buy the analyzer package from e.g. Omicron Lab but it's a hideously expensive piece of kit (4490€) considering what you're actually buying. Never mind Ridley's version at $17.5k.

It's possible to measure a SMPS circuit feedback loop bode plot by injecting a signal into the loop and measuring the response. Theoretically you can do this with a signal generator, transformer and 2-channel oscilloscope. In practice it's not that simple - to start with, finding the right transformer to do your testing between, say, 10Hz and 1MHz. Has anyone managed to do that in a reasonable way using "homebrewed" equipment?

Like so: circuit diagram

Basically what that very very expensive equipment does for you is to generate and record the various test points which is a tedious business at best.

  • \$\begingroup\$ I think Omicron's expense if I am not mistaken on your model is the high voltage used to test large transformers 1mHz to 1MHz with constant V/F and the analysis software. I recall a cheaper Bode plotter but you can get use a step response at different current steps to gain the same insight. But yes we once did it with bench sig gen synced to sweep on a SA for 14" HDA servo response loops. and I have many times used FM with scope X sweep out to get a filter envelope response as long as blanking interval was greater than filter latency \$\endgroup\$ – Sunnyskyguy EE75 Sep 24 '18 at 18:01
  • \$\begingroup\$ not the same as I thought. This Omicron has over 7 decade sweep options with 1Hz to 50MHz and 1Hz to 5kHz variable tracking filter. That is no joke to make and is probably a UHFdown converter. with DSP \$\endgroup\$ – Sunnyskyguy EE75 Sep 24 '18 at 18:34
  • \$\begingroup\$ Impedance-converting transformers are used to inject test signals into low-impedance nets as power rails. You don't need the transformer to inject signals into high-impedance feedback network. \$\endgroup\$ – Ale..chenski Sep 24 '18 at 19:38
  • \$\begingroup\$ @TonyEErocketscientist I'm sure it's very good for the job it makes but to start with that's excessive bandwith to get the job done. Most feedback loops would go below unity in single or double digit kHz unless you did something wrong. 10Hz to 1MHz would be enough for most purposes. Another niggling issue is the interaction between current and voltage loops plus slope compensation, this is usually hand-waved away. I saw a Texas datasheet mentioning this recently, citing it's "tedious" process to do the math the hard way but actually explained how. \$\endgroup\$ – Barleyman Sep 25 '18 at 9:04
  • \$\begingroup\$ @Ale..chenski In every single diagram I've seen for gain-phase response there's a transformer involved so I'd presume it needs to be there. If for no other reason, how are you going to inject current with correct DC bias ? \$\endgroup\$ – Barleyman Sep 25 '18 at 9:53

Maxim has a good app note on winding your own isolation transformer for this application: Maxim app note

It's not very easy to make these kinds of measurements with a signal generator and scope. The network analyzers have narrow moving digital filters that allow comparison of the gain and phase of the output signal vs. the input signal in the presence of noise and harmonics.

I have access to these types of analyzers at work, but for my home lab I was able to find an old HP 3562A on Ebay for $400.00. It works fine for loops with crossover up to 100kHz.

You might also try an old gain/phase meter like the HP 3575A, which you might find even cheaper, but I'm not sure how good the results would be.

If you want to get ambitious you could get a signal acquisition board for a PC and write a bunch of code to do it yourself. Would be a fun but tough project.

  • 2
    \$\begingroup\$ In the not quite DC to ~90KHz region, this sounds like a job for a PC soundcard.... Indeed I have used such for the purpose of plotting G/B curves for sonar transducers. The two channels are nice as you can trivially sample the signal at the point you are injecting the test signal to calibrate out amplitude and phase errors while sampling the output voltage with the other channel. If faster is wanted you can build simple quadrature mixers around bus switches to allow measurement in 90KHz bands at at least into the HF bands. \$\endgroup\$ – Dan Mills Sep 24 '18 at 17:13
  • \$\begingroup\$ You can even analyze the log response on Audio with very high sampling rates using free Audacity and a Turtle Beach Audio card or maybe a good USB module?? \$\endgroup\$ – Sunnyskyguy EE75 Sep 24 '18 at 18:06
  • \$\begingroup\$ todays best price on HP3562A is less than the model number but $2600 used \$\endgroup\$ – Sunnyskyguy EE75 Sep 24 '18 at 18:08
  • \$\begingroup\$ 100kHz is a bit low, You may not see the gain margin.. But for the price I guess it's not bad. \$\endgroup\$ – Barleyman Sep 25 '18 at 9:49
  • \$\begingroup\$ HP3577A looks pretty promising with the 5Hz to 200MHz band. You can get it from Ebay for $750 delivered to UK. It can handle 25VDC input. Combine that with the Maxim appnote on wiring the Pulse Electronics transformer secondary you should have decent feedback loop analyzer on a budget. \$\endgroup\$ – Barleyman Sep 25 '18 at 10:21

You can even analyze the log response on Audio with very high sampling rates using free Audacity and a Turtle Beach Audio card or maybe a good USB module.

There are technical reasons with higher order loops**, why Bode Plots are not useful and hide unstable regions and Nyquist plots must be used in the frequency domain or Step response.

**e.g. saturation heating and many ESR*C time constants.

How to make a scope into a Network Analyzer

HOWEVER, I would rely on Step Load Voltage response to measure SMPS stability and noise using AC couple into a 50 Ohm load

You could use Fourier transform with log amplitude and phase of repetitive pulse step loads. using PC and Audacity. with channel, calibration to correct for phase shift on audio AUX input channel or get a better sound card.

But you need a tracking filter to eliminate the SMPS noise and aliasing effects..

This would require using variable repetitive step pulse, from I1 to I2 to find the worst case THD or harmonic content or Vpp result.

Often overshoot from high current to zero current with lots of ringing occurs that demands some dynamic load to maintain current feedback stability or extremely well-designed RLC reactance for high current SMPS. Loop gain often changes with load and mu with temperature.

  • \$\begingroup\$ +1 for step load voltage response advice. \$\endgroup\$ – Ale..chenski Sep 24 '18 at 20:43
  • \$\begingroup\$ Yes, step response is the old cop-out you sometimes see in datasheets since you can actually measure that with an oscilloscope. I don't really see why it'd be better than bode plot since it would be similarly affected by operating parameters. And step response won't help you much to figure out where the problem actually is if your supply is not stable. Step response is an useful parameter to measure for an SMPS of course. \$\endgroup\$ – Barleyman Sep 25 '18 at 10:27
  • \$\begingroup\$ @Barleyman if you research higher order effects, you find it is good only if highly 2nd order and no nearby higher order poles or zeros ? near RHP, so OA’s ok, SMPS, not so OK so not cop out but the acid test, BODE is still useful but not to measure actual stability margin from overshoot, ringing under worst case step loads \$\endgroup\$ – Sunnyskyguy EE75 Sep 25 '18 at 12:30
  • \$\begingroup\$ Nor for switching half bridge reactive loads but MUST be done with AC 50 Ohms \$\endgroup\$ – Sunnyskyguy EE75 Sep 25 '18 at 12:35
  • 1
    \$\begingroup\$ We used to pay much more for these transient Analyzers and programmable "Active Loads" to test power supplies in production, when I was Test Eng Mgr. \$\endgroup\$ – Sunnyskyguy EE75 Sep 25 '18 at 14:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.