Who mentioned something about negative supplies here?
It's a good caution in case of transients. Stuff come out of nowhere. Surge can be induced from nearby sources (most often, very fast pulses like ESD and EFT), from hot-plugging and accidental shorting, or from the load, or through other equipment connected to the device in question.
Never hurts to be paranoid. Not having seen it, just means you haven't probed it in the right place when it happens.
Not an argument I'm willing to accept. My linear and regulated power supply doesn't make excursions to 35V out of a sudden. 24V for a 35V absolute maximum shows that there is plenty of headroom.
Actually, 48V is the expected worst-case for hot plugging; potentially several times more if ceramic caps are involved!
See the attachments. Only a short circuit on the supply can cause spikes up to 35V, but that is rare.
On the other hand, I've used regulators rated for 16Vbeing fed by 9V DC-DC supplies, and had no problems. The headroom in that case is far smaller and the spikes can surpass 16V easily. But, again, I'm talking about extreme events like short circuits on the supply rails, upstream from the regulator. That is not what happened. I didn't have massive capacitors in my circuit the last times this happened. Anyways, even in that case, the 1uF capacitor that I've used should absorb the spike.
I see positive clipping at ~16V, and the measurement says "19?" because it knows it's clipping (evidently the clipping happens somewhere offscreen). Was the ultimate peak actually 35V?
1uF is actually a risk factor, not a helper in this case!
We can apply AC transient theory, with a few judicious guesses, and figure out what's happening.
Suppose the source is a "9V" wall wart*, say 2m cable, OCV = 12V, nominal load 500mA (so, it's internally probably something like, an impedance-protected transformer, bridge, and 1000uF filter cap with << 1 ohm ESR). The cable is probably small zip cord, which will be around 0.3uH/m, or 0.6uH total.
When this is hot-plugged to the EUT, the voltage at the end of the cord is shorted to ~zero by the input filter cap. This draws a rising current through the cable's ESL. The initial rate is dI/dt = V/L, or about (12V) / (0.6uH) = 20 A/us. If the capacitor is 1uF (nominal) and very low ESR (a ceramic would have maybe 0.1 ohm at this effective frequency), it forms a resonant circuit with the cable, so the current rises and, as the current rises, the voltage also rises. If there is no ESR in this circuit, then the resonance will continue to its full peak.
Some time later (after 1/4 wave), the voltage is nominal (say 9 to 12V) but the current is maximum, so the voltage keeps going further. By 1/2 wave, it stops at a peak of up to 24V (or, again, more if this is a nonlinear ceramic). Now the output voltage is higher than the source, and current is ramping backwards, and the same thing happens again in reverse.
What time scale is this wave? The resonant frequency is:
Fo = 1 / (2*pi*sqrt(L*C))
or 0.2MHz. The period is 1/Fo, and we can divide that period into whatever fractions of the wave we like. So, 5us total, or 1.25us to reach nominal voltage, 2.5us to reach peak, and so on.
What if we include ESR?
The source, cable, and input filter cap ESRs are all that matter so far. Any additional load there might be, should also be included, but the LDO for example acts in parallel, so we need to use a parallel-series conversion for that, which may not be representative in the conditions we're considering anyway (in which case, we might as well hand-wave it, or simulate it).
The LDO also has a constant-current input characteristic, meaning it doesn't affect the AC waveform most of the time. We can ignore it for now.
Now, ESR is generally undesirable for a power supply, right? The purpose of a supply is to deliver a stable voltage, and the definition of resistance is a change in voltage with a change in current! Well, the reality of it is this: nothing can be actually-zero ohms (except superconductors at DC, but there isn't really such a thing as true DC, either, hmm!
), so we need to look for how much impedance we can actually handle, and meet or exceed that spec.
Of course, an LDO's purpose is to stabilize a potentially messy source, so we shouldn't need to be very picky about it in this case! We don't know what impedance is required, but we do know that a mostly-capacitive impedance (i.e., the suggested 1uF) will do, so we can match that, more or less. (In particular, that impedance will only be critical around the crossover frequency of the LDO's control loop -- this isn't a tip you'll ever see in a textbook or app note, but that's what they're actually doing it for.)
So, that said, we can add ESR to the 1uF until we have a well damped resonant circuit. How do we figure that out? Well, it comes down to the solution of a quadratic equation. If the solutions are complex, it resonates; real, it dampens. This happens to cross over at what we call the characteristic impedance:
Zo = sqrt(L/C)
If ESR > Zo, it's dampened, else, resonant.
sqrt(0.6uH/1uF) ~= 0.8 ohms, so we don't need much. And really, relative to a 16mA load? That's going to be more than fine. We could get away with tens of ohms and still have a happy regulator, I would bet.
We can connect a resistor in series with the cap, or we can simply choose an electrolytic with around that much ESR. (Tantalum is handy for its built-in ESR -- and is much more stable than electrolytics' ESR -- but is not recommended for heavy supply rails and inrush, so we won't use them here.)
And we're done, right?
Well, we might take an extra moment to check worst-case conditions. Maybe we need a lower impedance at the regulator anyway (this would definitely be the case for an SMPS regulator!). Maybe we need to support longer cables, or softer supplies. In that case, we might need a bare capacitor in parallel (without much ESR). We can effectively slow it down, by connecting an R+C in parallel with it, where C >= 2.5 Co and R = ESR.
To handle longer cables (more ESL), Zo goes up, which means required ESR goes up even more. To counter this, we can grow C proportionally. This also gives us access to smaller ESRs in electrolytic type.
In short, don't worry too much about a relatively large cap -- as long as it's got enough ESR. 'Lytics are great for this; you can usually pop in, say, 100uF and not worry about it.
*I didn't notice if you mentioned what the typical source is, so I'm using an example here. This isn't a straw-man by any means; drop in the appropriate parameters for another source and see what happens. For example, a 9V battery on a 150mm clip lead, I would expect to see ESR in the several-ohms range, and ESL around 100nH. With Zo = sqrt(0.1uH/1uF) ~= 0.3 ohm, the battery ESR will dominate, and the circuit will just kind of go
thud when turned on.
Tim