Problem with PSU's power-good output momentarily tripping at load turn-off

[HwAoRrDk]

I am facing a curious problem and I can't figure out what is causing it.

I have a circuit that switches a load on a +5V rail using a TI TPS2557. The +5V supply comes from an external mains-powered switch-mode PSU. If I enable the output from the TPS2557 and load it with 5A using my DC electronic load, then turn off the load, the Power Good (PG) output signal of the PSU drops out momentarily at the point of turn-off. As in, is explicitly pulled low by the PSU, not glitching or anything like that. It stays low for about 250-ish milliseconds, then comes back again.

Schematic of the +5V load switch portion of the circuit is attached below. The PG signal (not shown below) comes in and has a 47k pull-down and a 1k series resistor before going to an MCU GPIO.

I've tried this with two different PSUs (of different brand and model), and they both exhibit the same behaviour.

At first I thought maybe there was some kind of overshoot on the +5V supply rail happening at load turn-off, and that was causing whatever kind of voltage supervision the PSU has to think it had gone out of regulation - i.e. triggering over-voltage protection (OVP). But scoping it shows nothing untoward happening. With the first PSU, the +5V rises slightly by about 150mV for around 10ms then settles to its idle level (~5.3V); the second PSU exhibits no rise. I even looked up the specifications for OVP on the kind of voltage supervisor ICs these PSUs feature, and it's minimum of 5.7V, which I am not seeing being exceeded.

I also checked the active-low enable signal to the PSU to see if it was glitching to the 'off' state, which it wasn't. I don't know how that'd be possible anyway, given it is connected to a physical slide switch pulling the enable signal hard directly to ground in the 'on' state.

One curious thing I observed is that it seemed far less likely to trip the PG if the load is turned off within about 10 seconds of turning it on.

Any suggestions as to how I can figure out why this is happening?

[HwAoRrDk]

Here's a scope capture. Pink trace is +5V rail from the PSU. Yellow trace is the PG output from the PSU.

The load turn-off is right around the first division on the left. You can see the +5V rise a little, but only to about 5.5V, then the PG goes low after about 5ms.

On a hunch, I tried adding a 470uF electrolytic capacitor across the terminals of the load, and it seems to make the problem less repeatable, but doesn't eliminate it.

I can't imagine this problem is anything other than OVP on the PSU being triggered somehow, but I'm just not seeing what would cause it. 

Unless it's some sub-microsecond spike that I'm not capturing because I simply haven't gone to that short of a timebase. But then why the hell would a PSU's OVP be that sensitive?

[selcuk]

I think there is an inductor either at the output of the PSU or on the cable. You see a small amount of overshoot but PSU sees it much larger due to the inductor.

[youngda9]

What is the controller chip that generates the Power Good (PG) output signal?  Your output voltage is increasing about 300mV or so when the load is suddenly removed.  This is a function of the transient response of the converter.  There may be a window detector that generates the PG signal.  This threshold may be tripped by the sharp increase in output voltage when the load is suddenly removed.

[HwAoRrDk]

I think there is an inductor either at the output of the PSU or on the cable. You see a small amount of overshoot but PSU sees it much larger due to the inductor.

There's nothing on the cable. I'm not sure about at the output. I mean, there are several inductors inside the PSU, but I couldn't say what their exact purpose is.

What is the controller chip that generates the Power Good (PG) output signal?

One of the PSU's has a HY510N voltage supervisor IC. I'm not sure what the other one has, I couldn't find any info from any of the markings on the chip.

Your output voltage is increasing about 300mV or so when the load is suddenly removed.  This is a function of the transient response of the converter.  There may be a window detector that generates the PG signal.  This threshold may be tripped by the sharp increase in output voltage when the load is suddenly removed.

It's not quite 300mV. The no-load idle voltage is about 5.3V, and the voltage level dips under the 5A load to about 5V, so the increase at load turn-off to about 5.4-5.5V is only about 100-200mV above no-load level.

The datasheet for that chip says its 5V over-voltage threshold is a minimum of 5.7V, typical 6.1V. This threshold appears to be an absolute level (i.e. not relative to no-load level), because the block diagram indicates that the input voltage is divided down and compared to an internal 1.2V reference. There's also a 73us de-glitch period for the whole OVP mechanism. This is what is puzzling to me, as I'm not seeing anywhere near that threshold, and if it's a really short transient at turn-off (such that I'm not seeing it on the scope), surely the de-glitching will cause it to be ignored?

[HwAoRrDk]

One thing I noticed today after more experimenting, is that it doesn't happen in a short period after first cold power-up. For a couple of minutes I can turn the load on and off without PG tripping, but after that it happens like 90% of the time. So, some kind of temperature dependent behaviour, only when warmed-up. Not sure if that would be on the part of the PSU or my circuit. (Or maybe the electronic load?)

Other things I tried:

- Setting my scope up to probe the +5V rail in AC coupling mode and a short time base (in the microseconds) to capture the rise at turn-off in more detail. Definitely no transient spikes there. I even turned off the 20MHz bandwidth limit.
- Probed the +5V standby rail (i.e. rail that is always powered) in case that was experiencing some kind of transient (because I believe the PSU's voltage supervisor IC is powered by that rail). No disturbance there at all.
- Added a 470uF cap to where the PSU connects to my board. No effect.

Still utterly mystified as what's causing this problem.

[HwAoRrDk]

I don't know why I didn't think of this before. It occurred to me that I can test the PSU in isolation by hooking the electronic load up directly to the PSU. Although a bit tricky to rig up as the PSU's connector is of a type that doesn't lend itself to being connected to anything that isn't the mating half. Grabbed a spare PCB-mount connector and hooked to the pins on that side did the trick.

It still happens when the PSU is connected directly to the load. So it's nothing at all to do with my circuit and possibly a shortcoming with the PSU.

I am surprised that two different PSUs from two different manufacturers both suffer from the same problem.

I'm not sure whether it's really worth pursing a fix for this. To start with, my 5A test load is really a worst-case scenario that is unlikely to occur in real usage, and if it does it won't really be a sharp, hard transition from zero to 5A and back again - it'll be more varied and gradual.

But I am still interested to hear any theories about why it's happening, given I can't seem to observe any external contributing factors.

[mikerj]

At first I thought maybe there was some kind of overshoot on the +5V supply rail happening at load turn-off, and that was causing whatever kind of voltage supervision the PSU has to think it had gone out of regulation - i.e. triggering over-voltage protection (OVP). But scoping it shows nothing untoward happening. With the first PSU, the +5V rises slightly by about 150mV for around 10ms then settles to its idle level (~5.3V); the second PSU exhibits no rise. I even looked up the specifications for OVP on the kind of voltage supervisor ICs these PSUs feature, and it's minimum of 5.7V, which I am not seeing being exceeded.

Over-voltage protection has quite a different intention than a power good signal.  OVP would be used to e.g. trigger a crowbar circuit to prevent/minimise damage to circuits connected to the PSU.  A PG signal lets the attached circuit know that the PSU output voltage is stable within it's rated tolerance, and is typically used to hold CPUs or other complex logic etc in a reset state until that time.  There will be a voltage tolerance and a time period associated with this, and the voltage tolerance would typically be well below a protection threshold.