Transient response test (alone) for Buck converter stability assurance?

Hello
I have recommended that my workplace use a AP300 gain/phase analyser for our Buck converter stability assurance, however they say that transient response testing alone will do. Do you believe this is erroneous?
…………………………………………………………………………………………………………………………………………
Firstly , My sincere apologies for the length of this email, I appreciate that you are very busy and understand entirely if you cannot respond.
This question is about the superb Dr Ridley article (“Transient response and loop gains of power supplies”) which advises against checking for power supply stability using the transient response test alone.
We are currently developing a Dual phase, Interleaved Buck converter for Vin=48V, Vout=1V5, Iout Max = 20A, Fsw = 200kHz (Interleave phased so it looks like 400kHz at output) , CCM, Current Mode, Controllers = Two LTC3892’s , Worst case load transient will be 50% to 100% load step and vice versa, Load = many processors) .
The schematic is attached in pdf should you wish to see it. (Also the LTspice simulation)
The two Buck controllers have their transconductance error amplifiers connected together, as the LTC3892 datasheet recommends.
(Incidentally, we handled the Bode plot calculation for the Dual Interleaved Buck Converter by doubling the error amplifier transconductance, also by assuming that the switching frequency was double that of the single buck case (ie 400kHz), and that the duty cycle was double what it was for the single Buck case into half-load.
I recommended that we add a loop injection resistor into the PCB layout and used the Ridley Engineering AP300 gain/phase analyser to actually measure the feedback loop..
..However, the project leader has advised against this. He says that we should simply do the stability check by transient response test alone. He says that because this is a current mode Buck Converter with a very small duty cycle, that the transient response test alone will suffice. This worries me as the worst case load transient is significant at 50% to 100% load step, and we therefore need the feedback loop to be fairly fast)
(Incidentally, the LTspice program, with an injection source as in the attached LTspice simulation, suggests a crossover frequency of 7400Hz, and phase margin of 86 degrees ………..Our own calculation gave 6kHz for crossover and 88 degrees for phase margin. We were told by authority that adding an injection source to a time domain LTspice simulation [as in the attached] was bad practice and wouldn’t give a valid result, because a frequency domain measurement is needed instead)

Anyway, Please could you possibly advise on whether transient response test alone is indeed valid for a continuous current mode Buck converter with very low max duty cycle? I doubt it is, i mean especially when we will be going for a fast response by tweaking the compensation components.

If it was a single buck then we would be confident of calculating the feedback loop, (gain and phase margin) but for various reasons, we are unsure if our feedback loop calculation is exactly correct for the dual interleaved buck case. I mean , the calculation we did corresponded almost exactly with ltspice, but we are still not 100% sure, specially when we heard that the "ltspice method" is not accurate.

Transient testing under various load conditions is a pretty quick and easy demonstrator of stability (or otherwise). Good for testing an existing completed product, especially someone else’s.
If you are evaluating several sample products for possible purchase, a quick transient test is probably all that is needed.

But if it is your own in house design, and fails the test, what then ?

A Bode plot is an excellent way to analyse the problem and more useful during design and development where it is desired to actually make running changes to improve matters.

So I suppose it really depends if you are a power supply user, or a power supply designer.

Thanks, ..the situation is that this product will go into production, and the components may as usual vary from batch to batch as they do, and if our phase margin isnt generous, then we may get caught out, since we will be making the feedback loop fast because the load transients are significant...if it was just a damped response requirement, then i appreciate that transient response tests alone may suffice..
the attached explains the pitfalls of transient response testing alone.

- - - Updated - - -

the thing is that the very low max duty cycle of this buck is said to mean that bode plots are less necessary , even though we want a fast response from the feedback loop.

A standard test used by Samsung for qualifying psu designs proffered by outside consultants, is, take full load (R) switch it with a BIG mosfet, limit the Trise (I) to 10uS via gate R, vary the freq from 0.1 Hz to 20kHz, (50% duty) SLOWLY..! and note the results at 25/30Hz, 50/60Hz, 100/120Hz, around 1kHz and any other freq's where there is odd behaviour - e.g. excessive ringing on the Vo or Io, this is a pretty solid test for any psu... if yours passes this you can have some confidence, however knowing what loads affect your psu is paramount too, we have had many instances of people measuring our psu's thru a LISN, and into low R loads, the o/p C of the psu can easily ring (in //) with the L in the LISN which is under-damped by the load R causing power supply "singing" so stability is affected by many things....

There are a couple of difficulties and complications with all of this.

The first is that the load may not be resistive, it could be hugely capacitive or even a very low impedance battery.
This load sits right across your feedback loop, so after you have very carefully optimised everything including allowance for the ESR of the output filter capacitors in the supply, connection of a really hostile load can screw up the best prepared plans.
The whole time constant and phase of the output filter can change dramatically with a few extra Farads lumped directly across the output.

One other difficulty I have encountered with doing loop gain and phase testing is high frequency noise within the loop. Its one thing to test a high end audio amplifier or a fairly benign phase locked loop with some introduced perturbing sine waves.

Quite another to test a very noisy DCM flyback by adding a few millivolts of sine wave test signal to several hundred millivolts of really vicious noise spikes.
While the power supply control loop may not mind, the phase and gain measurement readings may become very erratic or badly skewed at certain frequencies from aliasing or just plain noise.

Before buying an expensive magic phase/gain measurement test rig, I would want to hire or lease it for a while, and give it a road test on some really bad ass switching supplies.

If it works, terrific.
If all it does is tell you some unrepeatable nonsense, do not be too surprised.

thanks, the Ridley gain/phase analyser says it handles noise by adaptively adjusting the injected signal amplitude as you go through the frequency sweep, so as to always act to increase signal to noise ratio.

The Ridley engineering group make a gain phase analyser ( AP300) which is good for measuring the gain and phase margin of an SMPS…

https://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.
https://www.youtube.com/watch?v=_NMmdJYFGY0
https://www.youtube.com/watch?v=JDG70G9uGBI
https://www.youtube.com/watch?v=8TVvxB0n4qs
https://www.youtube.com/watch?v=U2ipS672M-0

If I had to pick one or the other, I'd go with a harsh load
step over the AP300 because you get to see both small
and large signal response, and I've seen more "WTF?"
involving the latter. Nothing against the AP300 or the need
to diligently stabilize the design, just that small signal
stability is not the end of it and a simple test method lets
you get to the bottom line (for the load(s) you picked)
clean and quick, everything on the table.

If I had to pick one or the other, I'd go with a harsh load
step over the AP300 because you get to see both small
and large signal response

Click to expand...

Thanks, yes, agreed that transient response testing is always needed whether or not a gain/phase analyser is used.

The article of post #3 above tells me that the gain/phase analyser is very useful too though.

I dare say that many are happy with transient response alone if compensation has been adjusted to get a very well damped response....and there are no opto's or other such widey tolerant components in the feedback loop...and the product in which the power supply sits is not very very expensive

- - - Updated - - -

I was at a big telco once who had a SMPS go unstable when they changed the opto to a "same_part_number_but_different_manufacturer" one. This wasted millions of pounds. It was the gain/phase margins not being sufficient in the first place which instigated it......a gain/phase analyser, or in that case, proper loop calculation, would have sorted it.......It was a legacy PSU that got brought in without the designer having had time to check it properly.......

They also only briefly tested it with an e-load instead of the amp with its big capacitor in front of it. They werent given access to the actual load (rf amp) FOR TESTING because it was a "secret" RF design.

I was at a big telco once who had a SMPS go unstable when they changed the opto to a "same_part_number_but_different_manufacturer" one.

Click to expand...

often a higher gain opto will take a circuit to instability, where it is a bit marginal in the first place...

thanks, the Ridley gain/phase analyser says it handles noise by adaptively adjusting the injected signal amplitude as you go through the frequency sweep, so as to always act to increase signal to noise ratio.

Click to expand...

I am sure that modern DSP technology in a recent high end instrument will be well up to the job of having a very efficient tracking filter.
But at the low cost budget instrument end, many gain phase meters use fairly simple analog techniques that work fine for most phase/gain measurement jobs, but fail miserably with high switching noise.

Buying a budget, or a vintage gain/phase meter may have quite limited usefulness for our particular application, but even a cheapie is still well worth having on the shelf.