[time-nuts] Re: Death of a Capacitor

[nuts at lazygranch.com]

I've only designed one LDO as a discrete chip (as opposed to a portion
of a chip where performance just has to be good enough), so I have no
guru status. That said, what spikes pass through a LDO if you do it
right is simply a capacitor divider comprised of the capacitance across
the pass device and the filter capacitor. This is a bit more
predictable with a PFET pass than a PNP. 

https://www.analog.com/en/products/lt3045.html

You can see the PSRR after a point (200kHz) rolls off and appears to
flatten. I assume the error amp is out of loop gain. It goes flat for a
while. The idea here is the drive on the pass device is constant
and just maintains the DC voltage. The AC rejection is mostly due to
capacitance ratios. This being a bipolar pass device there is some
secondary effect here where after 2MHz the rejection improves then goes
flat again.

The bipolar pass control is harder than MOS since you are trying to
keep the device out of saturation. That is besides the error amp there
is some sort of anti-saturation circuit controlling the drive on the
pass device. 

My thinking here is small signal. If you have huge spikes the
performace even in the region where you do have loop gain can be
nonlinear. For example the error amp can be slew rate limited.

This looks like fine performance given the chip only draws 2.2mA.

Just trawling the interwebs I found this on the TI website:
https://training.ti.com/ldo-architecture-review
At about the 18 minute point he goes into the regions of PSRR.  I poked
around so I can't vouch more all of the talk,

. 

On Sun, 26 Sep 2021 18:21:36 -0400
John Ackermann N8UR <jra at febo.com> wrote:

> I got some interesting and unintended data today. I was measuring low 
> phase noise oscillators using a set of power supplies I just finished 
> putting together.
> 
> The configuration is ~24 VDC into a TPS-53400 switching regulator
> that outputs 19.2 volts at up to 3 amps.  That output is fed to
> separate regulator boards for each oscillator.  Those boards each
> have an LT-1086 linear pre-regulator that drops the input to about 17
> volts, which then goes into an ultra-low-noise LT3045A outputting 15
> volt to drive the oscillator.  So there are two linear regulators and
> lots of caps, inductors, and ferrite beads to isolate the oscillators
> from the switching supply.
> 
> Due to an error by an assembly tech who will remain nameless, the
> wrong electrolytic was installed on the output side of the switching 
> regulator.  It should have been 33uF at 50 volts, but what got
> installed was 330 uF at 16 volts, so it was rated below the operating
> voltage. (I was building two boards at the same time, one for 5V and
> one for 19.2V. Apart from the voltage setting resistor, the only
> difference between the two was the output cap.  I managed to swap
> them.)
> 
> I tested the system on the bench for 24 hours and everything worked 
> fine, so I buttoned up the enclosure and started a 4 hour data
> capture. About 70 minutes in, the electrolytic became very unhappy
> and whatever it turned into caused the switcher to start spewing all
> sorts of crud. The regulator kept working (sort of) through the end
> of the run, but when I came into the lab the next morning it had shut
> down completely and troubleshooting showed that the cap had shorted
> at some point after the run completed, and the regulator chip went
> into shutdown.
> 
> Attached are a plot of frequency showing the whole run with the very 
> obvious change when the cap failed, and another zoomed view of the 
> critical moment.  The failure was very abrupt with no visible lead-in.
> 
> What I find interesting is that all that crud got through not one,
> but two linear regulators, one of which is touted for its extremely
> high PSRR (and I did my best to follow the recommended PCB layout for
> that chip).  That must have been one ugly 19V line when the cap
> went...
> 
> John