Transient Load Testing Buck Converter

[QQ]

Hi all,

Actually I had posted this question on another thread that I started
before but since the thread title was not reflective of the question I
think I may not be getting more opinions and advice.

The question is:

How do you do transient load testing of a buck converter to verify its
stability? I know that by varying the load current from min to max/max
to min using an external DC load and then seeing how quickly the
output voltage is able to recover is one way to verify that the
converter is stable.

But how does one determine from the step response, what the gain
crossover frequency, gain margin and phase margin is? Also I've read
that if the output voltage recovers it may still not indicate
stability. What are the telltale signs I must look for in step
response?

I am familiar with laplace transforms and know that time to frequency
domain conversion must be possible but I don't know exactly how to do
that in this case

Will appreciate any help.

Thanks
QQ

 

[legg]

On 11 May 2004 20:30:57 -0700, q_q_404@hotmail.com (QQ) wrote:

Hi all,

Actually I had posted this question on another thread that I started
before but since the thread title was not reflective of the question I
think I may not be getting more opinions and advice.

The question is:

How do you do transient load testing of a buck converter to verify its
stability? I know that by varying the load current from min to max/max
to min using an external DC load and then seeing how quickly the
output voltage is able to recover is one way to verify that the
converter is stable.

But how does one determine from the step response, what the gain
crossover frequency, gain margin and phase margin is? Also I've read
that if the output voltage recovers it may still not indicate
stability. What are the telltale signs I must look for in step
response?

I am familiar with laplace transforms and know that time to frequency
domain conversion must be possible but I don't know exactly how to do
that in this case

The whole idea behind transient load testing is that it doesn't matter
what the numbers tell you - the result of whatever you've actually
constructed, accidents and all, is demonstrated.

Your job might be to make the numbers explain real life, including the
accidents - if anything. Otherwise, you wouldn't need to do the real
physical test in the first place, would you?

Transient testing doesn't tell you where or how the stability is
achieved, or instability is engendered. You might get some useful
ideas, however, by following the control signal chain all the way
through, while the load impulse is being applied.

At last! An instance where those flakey BetaDyne app notes actually
might serve some kind of useful purpose.

"Testing Transient Response in DC-DC Converters"
BetaDyne Application Note DC-001
No author admits responsibility

http://www.beta-dyne.com/pdf/dcdc_an1_rev02.pdf

RL

 

[Genome]

"QQ" <q_q_404@hotmail.com> wrote in message
news:9ab724f3.0405111930.51c59253@posting.google.com...
| Hi all,
|
| Actually I had posted this question on another thread that I started
| before but since the thread title was not reflective of the question I
| think I may not be getting more opinions and advice.
|
| The question is:
|
| How do you do transient load testing of a buck converter to verify its
| stability?
| Thanks
| QQ

The simple answer is..... you just do. Hit it with the load transients
you expect in real life, see how it behaves and make sure you can
explain why it does what it does. Don't just look at the output, look at
what the error amplifiers do at various nodes.

Ideally you should, at least, have designed and better still modelled
the thing to the extent that you can get a good match between the two.
All the transient testing does is confirm what you thought in the first
place.

DNA

 

[Terry Given]

"Genome" <Genome@nothere.com> wrote in message
news:X9qoc.224$pR4.167@newsfe2-gui.server.ntli.net...

"QQ" <q_q_404@hotmail.com> wrote in message
news:9ab724f3.0405111930.51c59253@posting.google.com...
| Hi all,
|
| Actually I had posted this question on another thread that I started
| before but since the thread title was not reflective of the question I
| think I may not be getting more opinions and advice.
|
| The question is:
|
| How do you do transient load testing of a buck converter to verify its
| stability?
| Thanks
| QQ

The simple answer is..... you just do. Hit it with the load transients
you expect in real life, see how it behaves and make sure you can
explain why it does what it does. Don't just look at the output, look at
what the error amplifiers do at various nodes.

Ideally you should, at least, have designed and better still modelled
the thing to the extent that you can get a good match between the two.
All the transient testing does is confirm what you thought in the first
place.

DNA

Very well said. to generalise further, for any circuit (or for that matter
sw) you design, you should look carefully at a large number of nodes (all
would be good) to ensure they behave *as expected* over a wide range of
operating conditions. Check even simple stuff, like voltage dividers etc and
if you see something unexpected, explain it. If the waveform turns out to be
correct, your understanding of how it works was wrong, in which case you
learned something. Alternatively, you just found a problem.

Then do it again once you have built a real one. Actual problems come from a
wide range of sources, like diode leakage, recovery time, resonances,
capacitive/magnetic coupling, cross-conduction, IR drop, the list goes on
and on - but they are all unintentional. The real trick is figuring out what
to test, and how, then how to interpret the resulting measurements.

Conversely if you dont look, you wont find problems. It does not however
mean that there are no problems.....

To go further, when you build your first real production units, check them
thoroughly too. Anytime anything changes, problems can arise. The bigger the
change, the greater the likelihood of problems. So when you do change
things, regression test - it should still be able to pass all of the earlier
rigorous testing. In addition to 100% production testing, (by necessity only
looking at critical points), randomly select a small portion of units and
rigorously examine/test them.

cheers
terry

 

[Dave VanHorn]

To go further, when you build your first real production units, check them
thoroughly too. Anytime anything changes, problems can arise. The bigger
the
change, the greater the likelihood of problems.

How about a properly designed flyback switcher, that would occasionally
catch fire?

Turned out to be a mounting screw that was reversed.
The screw hit an electrolytic cap can.
No big deal, the can is isolated, right?




Well.. The screw is on the collector of the flyback switching transistor,
and it did NOT appreciate the extra capacitance.

 

[Terry Given]

"Dave VanHorn" <dvanhorn@cedar.net> wrote in message
news:ks-dnXU_GPnGUD_d4p2dnA@comcast.com...

To go further, when you build your first real production units, check
them
thoroughly too. Anytime anything changes, problems can arise. The bigger
the
change, the greater the likelihood of problems.

How about a properly designed flyback switcher, that would occasionally
catch fire?

Turned out to be a mounting screw that was reversed.
The screw hit an electrolytic cap can.
No big deal, the can is isolated, right?


Well.. The screw is on the collector of the flyback switching transistor,
and it did NOT appreciate the extra capacitance.

lovely. A beautiful illustration of the law of unintended consequences. BTW
Electrolytic cans are usually not very well insulated at all; they are often
at, or near, the potentil of the (-) terminal. The plastic sleeve aint much
use as insulation.

I have seen lovely "tricks" like that too. The new production guy who wasnt
taught how to properly calibrate the torque settings on air tools (building
100kW+ inverters), and did it by adjusting the air pressure valve (as
opposed to the torque setpoint grub screw on the pneumatic screwdriver).
6hrs later, explosions start happening in the test area.....

or the contract manufacturer who decided to save cost by pop-riveting 7805's
to heatsinks - which gave rise to my all-time favourite quote: "Motorola
7805's are shit, we have a 30% failure rate." Needless to say, when they
stopped pop-riveting (read as: bashing a piece of glass soldered to a piece
of nice, soft copper, with an unpredictably large hammer), the failures all
mysteriously went away. Motorola AN1040 strikes again (do you think I could
convince the manufacturer of this, even armed with AN1040? no way. actual
failure data, before and after, didnt help either. idiots.)

same manufacturer, on a flyback transistor heat sink, decided to use a nylok
nut rather than nut+schnorr washer to mount the FET. Guess what, nylon
melts, at around 90C! Little dutch time bomb, tick tock boom.

same product (its a UPS), the design engineer wrote a detailed test
procedure. lots of gate drive waveform checks, etc. But, the very first test
was to apply the national grid. Product had a near 100% failure rate, all
catastrophic failures/fires. I took the test procedure, and moved the 1st
paragraph to near the end, so the gatedrives were tested before the national
grid was applied. The failure rate was still very high, but mean time to
repair went from 4hrs (total rebuild) to about 20 minutes (find the fault),
and it turned out incorrect component insertion in the gatedrives was the
main culprit. But the catastrophic failures completely fried the gatedrives,
so no-one was able to figure out why they died.....within 2 weeks or so,
corrective action feedback to the contract manufacturer had solved most of
these problems, and production went from 2pcs per day (for 6 months!) to
100+ per day - all without changing the design, or production line, at all

cheers
terry

 

[Dave VanHorn]

lovely. A beautiful illustration of the law of unintended consequences.
BTW
Electrolytic cans are usually not very well insulated at all; they are
often
at, or near, the potentil of the (-) terminal. The plastic sleeve aint
much
use as insulation.

EVERYONE caught that one, but the can itself was insulated from the rest of
the cap, in every case that we checked. However, the large capacitance
between the can and the cap terminals (large if you're turning on and off at
300kHz) didn't show up on an ohmmeter. It did make the converter run much
slower, saturating the inductor, and eventually flaming out in a pretty
spectacular way.

or the contract manufacturer who decided to save cost by pop-riveting
7805's
to heatsinks - which gave rise to my all-time favourite quote: "Motorola
7805's are shit, we have a 30% failure rate."

VBG! I would probably agree about the 78M05s, I had an abysmal fail rate on
those as well, but we weren't shattering the dies.

I also had some fun with TI modem chips that wouldn't "shut up" when I
asserted SQT. (Squelch Transmit) The local sales weasel tried to pass it
off as ok, because there was no spec on how much the carrier level should
drop when SQT is asserted.
I was rather insistent that this should always be a very large number.

Then there was the Rockwell printer controller chip that wasn't really
tested with the mechs that they said it would work with, and the rockwell
modem chip that output a V.22bis answer tone when answering in bell 212a
mode...

same manufacturer, on a flyback transistor heat sink, decided to use a
nylok
nut rather than nut+schnorr washer to mount the FET. Guess what, nylon
melts, at around 90C! Little dutch time bomb, tick tock boom.

Too funny! Like a screen door in a submarine.
Next, non-conductive solder!

Nylon causes some other problems as well.
A fellow I'll call my "evil ex-partner", was rather insistent that we should
use Nylon and Delrin for a mechanical set, designed with very close
tolerances, because we were needing a ton of torque, and he wanted the
lowest possible friction.

This combination is kind of like twinlead, in that under ideal conditions,
it outperforms any coax cable.

However, when the prototype, which works perfectly, makes the trip from
Wisconsin in winter, to Cancun for the demo, one of the other annoying
properties of nylon shows up. It's Hygroscopic.

The tolerances went from a few mils slop, to a few mils short of an
interference fit, and the once free-spinning bearing was not turnable with
pliers.


Back in high school, I used to hang out at my dad's work where he repaired
military surveilance cameras. They gave me one to look at that they had
been puzzling over for some time. Everything seemed ok, but it just
wouldn't trip the shutter. Testing the wiring was a bear because of the
way it was built, but in a few minutes I had found the problem. It seemed
that in this camera, and in the large drawer of replacement parts, all the
solenoid plungers were made of non-magnetic metal...

 

[Terry Given]

"Dave VanHorn" <dvanhorn@cedar.net> wrote in message
news6ednXq0br1jmzndRVn-hg@comcast.com...

lovely. A beautiful illustration of the law of unintended consequences.
BTW
Electrolytic cans are usually not very well insulated at all; they are
often
at, or near, the potentil of the (-) terminal. The plastic sleeve aint
much
use as insulation.

EVERYONE caught that one, but the can itself was insulated from the rest
of
the cap, in every case that we checked. However, the large capacitance
between the can and the cap terminals (large if you're turning on and off
at
300kHz) didn't show up on an ohmmeter. It did make the converter run much
slower, saturating the inductor, and eventually flaming out in a pretty
spectacular way.

nice. I've had to redesign a fair few saturating inductors, too, including a
flyback transformer (no prizes for guessing the product) that ran at around
370mT with 3C85 material. Fine at 25C, boom at 75C......

or the contract manufacturer who decided to save cost by pop-riveting
7805's
to heatsinks - which gave rise to my all-time favourite quote: "Motorola
7805's are shit, we have a 30% failure rate."

VBG! I would probably agree about the 78M05s, I had an abysmal fail rate
on
those as well, but we weren't shattering the dies.

No experience with 78M05. I started with a simple premise: semiconductor
manufacturers products work. Surely motorola would notice 30% failure
rate.....

In general products fail because the manufacturing process abuses them in
some way. Remove the abuse, and lifetime improves. Smith & Whitehead wrote
an entire book on this subject, optimising quality in electronics assembly
IIRC.

then again, I had some 0.01% resistors from vishay that were actually
+10-25% parts! It was a dc-balance circuit in a ups, and they started
rolling off the assembly line with huge, un-correctable DC imbalances. It
took quite a bit of work to find the problem, but when I did, Vishay were
incredible! I sent them sample parts (around 100pcs were affected IIRR),
they sent me, within a week, a 30+ page report containing ESM images of the
resistive material, and a detailed analysis of what caused the problem, how
come they didnt notice (it was a material degradation, exacerbated by T, so
they started out right), how to prevent it happening again, exactly how many
and which resistors were affected, etc. And of course they replaced all the
dodgy bits for free.

I also had some fun with TI modem chips that wouldn't "shut up" when I
asserted SQT. (Squelch Transmit) The local sales weasel tried to pass it
off as ok, because there was no spec on how much the carrier level should
drop when SQT is asserted.
I was rather insistent that this should always be a very large number.

Then there was the Rockwell printer controller chip that wasn't really
tested with the mechs that they said it would work with, and the rockwell
modem chip that output a V.22bis answer tone when answering in bell 212a
mode...

or the philips micro who's RETI instruction didnt always work...but
RETI
RETI
RETI

at the end of the ISR cured the problem. 2 RETI's made it better, but didnt
fix it completely!


A company I contract to at the moment had a problem last year, with 10ns
SRAMs - our product started going mental. Turns out the SRAMs we had
purchased came from a buying house, and were 6-7 years old. The access time
was as high as 30ns in some cases, so RAM reads often returned rubbish. That
was bloody hard to find, and to prove it was the SRAM, as the designer was
beating the hell out of a CPLD, so we assumed it was a CPLD code error, and
spent weeks looking for it. And I didnt want to believe him when he said it
was the RAM, but a few simple tests proved it. We now check datecodes on all
chips, and try to only buy from the manufacturer

same manufacturer, on a flyback transistor heat sink, decided to use a
nylok
nut rather than nut+schnorr washer to mount the FET. Guess what, nylon
melts, at around 90C! Little dutch time bomb, tick tock boom.

Too funny! Like a screen door in a submarine.
Next, non-conductive solder!

I went to a company-wide health & safety meeting once. Everyone was there,
we had to sign an attendance register. The production manager then, very
seriously, told us that lead was highly poisonous, so do not lick, chew or
eat solder - apparently some of the PCB assy staff had been seen chewing on
the stuff!

Nylon causes some other problems as well.
A fellow I'll call my "evil ex-partner", was rather insistent that we
should
use Nylon and Delrin for a mechanical set, designed with very close
tolerances, because we were needing a ton of torque, and he wanted the
lowest possible friction.

This combination is kind of like twinlead, in that under ideal conditions,
it outperforms any coax cable.

However, when the prototype, which works perfectly, makes the trip from
Wisconsin in winter, to Cancun for the demo, one of the other annoying
properties of nylon shows up. It's Hygroscopic.

The tolerances went from a few mils slop, to a few mils short of an
interference fit, and the once free-spinning bearing was not turnable with
pliers.

bugger. I didnt know that about Nylon, but I'll bear it in mind, thanx.

Back in high school, I used to hang out at my dad's work where he repaired
military surveilance cameras. They gave me one to look at that they had
been puzzling over for some time. Everything seemed ok, but it just
wouldn't trip the shutter. Testing the wiring was a bear because of the
way it was built, but in a few minutes I had found the problem. It seemed
that in this camera, and in the large drawer of replacement parts, all the
solenoid plungers were made of non-magnetic metal...

even better! hey, but Aluminium is so light.....

Cheers
Terry

 

[Dave VanHorn]

No experience with 78M05. I started with a simple premise: semiconductor
manufacturers products work. Surely motorola would notice 30% failure
rate.....

I don't know why either, but we sure had a lot of failures. We switched over
to the 7805's and had no problems.

In general products fail because the manufacturing process abuses them in
some way. Remove the abuse, and lifetime improves. Smith & Whitehead wrote
an entire book on this subject, optimising quality in electronics assembly
IIRC.

Yes. It's unfortunate that the abuse dosen't always show up immediately
though.
Frequently, they live long enough to get into the user's hands before giving
up.

then again, I had some 0.01% resistors from vishay that were actually
+10-25% parts! It was a dc-balance circuit in a ups, and they started
rolling off the assembly line with huge, un-correctable DC imbalances. It
took quite a bit of work to find the problem, but when I did, Vishay were
incredible! I sent them sample parts (around 100pcs were affected IIRR),
they sent me, within a week, a 30+ page report containing ESM images of
the
resistive material, and a detailed analysis of what caused the problem,
how
come they didnt notice (it was a material degradation, exacerbated by T,
so
they started out right), how to prevent it happening again, exactly how
many
and which resistors were affected, etc. And of course they replaced all
the
dodgy bits for free.

I'll remember that!

or the philips micro who's RETI instruction didnt always work...but
RETI
RETI
RETI

at the end of the ISR cured the problem. 2 RETI's made it better, but
didnt
fix it completely!

ROTFLMAO.. That's funny, in a sick and twisted way.

A company I contract to at the moment had a problem last year, with 10ns
SRAMs - our product started going mental. Turns out the SRAMs we had
purchased came from a buying house, and were 6-7 years old. The access
time
was as high as 30ns in some cases, so RAM reads often returned rubbish.
That
was bloody hard to find, and to prove it was the SRAM, as the designer was
beating the hell out of a CPLD, so we assumed it was a CPLD code error,
and
spent weeks looking for it. And I didnt want to believe him when he said
it
was the RAM, but a few simple tests proved it. We now check datecodes on
all
chips, and try to only buy from the manufacturer

I've seen the other case, 120nS parts that reacted to 30nS glitches and
hosed the bus.
This caused a big argument over what "guaranteed access" means.
I was arguing the point that although they guarantee an access if the pulse
is >X, they make no statement about what might happen on a pulse <X.

I went to a company-wide health & safety meeting once. Everyone was there,
we had to sign an attendance register. The production manager then, very
seriously, told us that lead was highly poisonous, so do not lick, chew or
eat solder - apparently some of the PCB assy staff had been seen chewing
on
the stuff!

Evolution in action?

I got rather worried about that a few years ago, since I've handled it very
casually for 30 years.
I had a screen done for all sorts of metals and other nasties. Everything
was unreadably low.

bugger. I didnt know that about Nylon, but I'll bear it in mind, thanx.

I think that's probably pretty close to what he said, in front of the CEO
and investors, when his demo tanked.

even better! hey, but Aluminium is so light.....

"But the spec dosen't REQUIRE them to be magnetic".

I sometimes get tired of the game where I have to guess all the ways that a
vendor might screw up a part..

 

[Terry Given]

"Dave VanHorn" <dvanhorn@cedar.net> wrote in message
news:YuudnToPWcN-3zvdRVn-sw@comcast.com...

In general products fail because the manufacturing process abuses them
in
some way. Remove the abuse, and lifetime improves. Smith & Whitehead
wrote
an entire book on this subject, optimising quality in electronics
assembly
IIRC.

Yes. It's unfortunate that the abuse dosen't always show up immediately
though.
Frequently, they live long enough to get into the user's hands before
giving
up.

Static is a biggy in this regard - sometimes the damage is not catastrophic,
but instead causes degradation which shortens lifetime. I currently work
with a guy who refuses to believe in static damage. I say we should sack the
prick, but unfortunately he is our resident fpga expert, and thus plays an
important role. Our production manager wont let him in the factory though,
and we NEVER let him touch equipment that will be sold.

I once had a holiday job, and my boss (PhD EE) killed 10 87C552 micros
(glass lid ceramic package, NZ$200 EACH) in 1 week, by carrying them in his
hand from the programmer to the lab.....

Moisture absorption of large plastic smt packages is like this (hence
humidity control requirements) - the package absorbs moisture, which reflow
oven converts to high-pressure steam, potentially rupturing the hermetic
seal, allowing god-knows-what to get into the package and wreck the die.....

then again, I had some 0.01% resistors from vishay that were actually
+10-25% parts! It was a dc-balance circuit in a ups, and they started
rolling off the assembly line with huge, un-correctable DC imbalances.
It
took quite a bit of work to find the problem, but when I did, Vishay
were
incredible! I sent them sample parts (around 100pcs were affected IIRR),
they sent me, within a week, a 30+ page report containing ESM images of
the
resistive material, and a detailed analysis of what caused the problem,
how
come they didnt notice (it was a material degradation, exacerbated by T,
so
they started out right), how to prevent it happening again, exactly how
many
and which resistors were affected, etc. And of course they replaced all
the
dodgy bits for free.

I'll remember that!

or the philips micro who's RETI instruction didnt always work...but
RETI
RETI
RETI

at the end of the ISR cured the problem. 2 RETI's made it better, but
didnt
fix it completely!

ROTFLMAO.. That's funny, in a sick and twisted way.

the engineer who found it was NOT amused - months of his life wasted. The
word FUCK featured heavily in his vocabulary at this time, especially in
reference to Philips

A company I contract to at the moment had a problem last year, with 10ns
SRAMs - our product started going mental. Turns out the SRAMs we had
purchased came from a buying house, and were 6-7 years old. The access
time
was as high as 30ns in some cases, so RAM reads often returned rubbish.
That
was bloody hard to find, and to prove it was the SRAM, as the designer
was
beating the hell out of a CPLD, so we assumed it was a CPLD code error,
and
spent weeks looking for it. And I didnt want to believe him when he said
it
was the RAM, but a few simple tests proved it. We now check datecodes on
all
chips, and try to only buy from the manufacturer

I've seen the other case, 120nS parts that reacted to 30nS glitches and
hosed the bus.
This caused a big argument over what "guaranteed access" means.
I was arguing the point that although they guarantee an access if the
pulse
is >X, they make no statement about what might happen on a pulse <X.

Ooh, thats nasty - I'll bear it in mind, too!

I went to a company-wide health & safety meeting once. Everyone was
there,
we had to sign an attendance register. The production manager then, very
seriously, told us that lead was highly poisonous, so do not lick, chew
or
eat solder - apparently some of the PCB assy staff had been seen chewing
on
the stuff!

Evolution in action?

yep. A bit like drunk-driving, ultimately a self-correcting problem - I
think it is hilarious when drunk drivers wrap themselves around trees. Up
goes the national average IQ. We had a guy in NZ a few weeks ago who was
pulled over by the cops. He stopped, and when the cop came towards him he
took off at high speed, and crashed into a power pole about 1.5km away, at
around 170km/h. Im sure his parents are not happy, but the rest of us
thought of it less as a tragedy, more like a significant moment in
evolution.

I got rather worried about that a few years ago, since I've handled it
very
casually for 30 years.
I had a screen done for all sorts of metals and other nasties. Everything
was unreadably low.


bugger. I didnt know that about Nylon, but I'll bear it in mind, thanx.

I think that's probably pretty close to what he said, in front of the CEO
and investors, when his demo tanked.


even better! hey, but Aluminium is so light.....

"But the spec dosen't REQUIRE them to be magnetic".

I sometimes get tired of the game where I have to guess all the ways that
a
vendor might screw up a part..

I used to work for a US company that tried to design high-speed flywheels.
We built a number of prototypes, all of which failed miserably, none of
which got even close to desired operating speed. We paid a prof. from MIT to
come help. He went straight to the draughting dept. for dwgs, then to the
store. With a vernier, he showed that the 0.01mm tolerance parts we had
spec'd were out by 1mm or so - we never checked ANYTHING, merely assembled
them. We bought some fancy measuring eqpt, and carefully checked every
single component before assembly. It made a huge difference.

NASA have proven, quite conclusively, that:
1) If you do not look for problems, you will not find any
2) that does not mean they are not there
3) hurling expensive things at Mars is not a cost-effective way of
troubleshooting a design

Cheers
Terry