Does anyone know where the RPi3B+ detects undervoltage?

Does anyone have a schematic that includes the 5V lines?

David

David Higton <dave@davehigton.me.uk> wrote:
> Does anyone know where the RPi3B+ detects undervoltage?
>
There's an APX803 chip which detects the voltage falling below 4.63 +/-
0.07V. That chip then asserts PWR_LOW_N, a GPIO the GPU firmware can
detect. It's also used to drive the PWR LED (ie the LED going off is the
same as PWR_LOW_N being asserted).

https://forums.raspberrypi.com/viewtopic.php?t=158777

> Does anyone have a schematic that includes the 5V lines?

https://datasheets.raspberrypi.com/rpi3/raspberry-pi-3-b-reduced-schematics.pdf

has the power supply and detection circuit.

Theo

In message <KLy*96JHz@news.chiark.greenend.org.uk>
Theo <theom+news@chiark.greenend.org.uk> wrote:

> David Higton <dave@davehigton.me.uk> wrote:
> > Does anyone know where the RPi3B+ detects undervoltage?
> >
> There's an APX803 chip which detects the voltage falling below 4.63 +/-
> 0.07V. That chip then asserts PWR_LOW_N, a GPIO the GPU firmware can
> detect. It's also used to drive the PWR LED (ie the LED going off is the
> same as PWR_LOW_N being asserted).
>
> https://forums.raspberrypi.com/viewtopic.php?t=158777
>
> > Does anyone have a schematic that includes the 5V lines?
>
> https://datasheets.raspberrypi.com/rpi3/raspberry-pi-3-b-reduced-schematics.pdf
>
> has the power supply and detection circuit.

Thanks, Theo. That's the one that everyone has, and of course it's
incomplete. But I looked up the data sheet for the RT9741C, so I have
to guess that "H5V" goes to all the USB-A sockets, which are not shown.

The upshot is that I'm going to have to reconsider power supply and
distribution for this one, which is used as NAS, so it needs to power
some spinning drives. The incoming supply is rated at 8 amps, but its
output voltage is just 5V, and it appears that its transient response
may leave something to be desired. One thing I can try is whacking
several 1000uF capacitors, via short leads, on the 5V line on the GPIO
connector. I can also see where I would need to interfere with it so
as to increase the output voltage to about 5.1V. I've cut the output
cabling as short as it can reasonably be made so that cable resistance
genuinely can be neglected.

It seems that mains-powered USB hubs are going out of fashion,
unfortunately.

David

David Higton <dave@davehigton.me.uk> wrote:
> The upshot is that I'm going to have to reconsider power supply and
> distribution for this one, which is used as NAS, so it needs to power
> some spinning drives. The incoming supply is rated at 8 amps, but its
> output voltage is just 5V, and it appears that its transient response
> may leave something to be desired. One thing I can try is whacking
> several 1000uF capacitors, via short leads, on the 5V line on the GPIO
> connector. I can also see where I would need to interfere with it so
> as to increase the output voltage to about 5.1V. I've cut the output
> cabling as short as it can reasonably be made so that cable resistance
> genuinely can be neglected.

FWIW I've just installed a 3B in my 3D printer to run Klipper. It's powered
off the main 24V PSU via a Chinese '5V 5A' DC-DC module, which outputs 5.2V
under no load. Klipper was running fine but reporting a number of power
supply dropouts, despite no USB devices connected.

I fitted a 330uF electrolytic and a 22uF ceramic (for low ESR) across the
power rails at the power input to the Pi (here via GPIO) and that resolved
the dropout problem. I'd guess the DC-DC just has insufficient capacitance
to hold up the rail when the Pi takes a gulp of current, and so needs a bit
of help.

I don't know the frequency response of the Pi's PDN (whether the gulps of
current are seconds, milliseconds or microseconds) but maybe try a range of
capacitors to handle them - not just big electolytics, which can have high
ESR and so a slow frequency response.

Theo

In message <NLy*uFPHz@news.chiark.greenend.org.uk>
Theo <theom+news@chiark.greenend.org.uk> wrote:

> David Higton <dave@davehigton.me.uk> wrote:
> > The upshot is that I'm going to have to reconsider power supply and
> > distribution for this one, which is used as NAS, so it needs to power
> > some spinning drives. The incoming supply is rated at 8 amps, but its
> > output voltage is just 5V, and it appears that its transient response may
> > leave something to be desired. One thing I can try is whacking several
> > 1000uF capacitors, via short leads, on the 5V line on the GPIO
> > connector. I can also see where I would need to interfere with it so
> > as to increase the output voltage to about 5.1V. I've cut the output
> > cabling as short as it can reasonably be made so that cable resistance
> > genuinely can be neglected.
>
> FWIW I've just installed a 3B in my 3D printer to run Klipper. It's
> powered off the main 24V PSU via a Chinese '5V 5A' DC-DC module, which
> outputs 5.2V under no load. Klipper was running fine but reporting a
> number of power supply dropouts, despite no USB devices connected.
>
> I fitted a 330uF electrolytic and a 22uF ceramic (for low ESR) across the
> power rails at the power input to the Pi (here via GPIO) and that resolved
> the dropout problem. I'd guess the DC-DC just has insufficient capacitance
> to hold up the rail when the Pi takes a gulp of current, and so needs a bit
> of help.
>
> I don't know the frequency response of the Pi's PDN (whether the gulps of
> current are seconds, milliseconds or microseconds) but maybe try a range of
> capacitors to handle them - not just big electolytics, which can have high
> ESR and so a slow frequency response.

I've just tried 9000uF (9 1000 uF capacitors in parallel) this afternoon.
I didn't see a restart in the logs when Clonezilla did a backup to OMV,
but the backup failed anyway, and now I don't know why.

I'll have to get the DSO on the job.

I've also got a bench PSU with enough current for this job, and of
course that enables me to set the voltage where I want.

David

In message <83030e515b.DaveMeUK@BeagleBoard-xM>
David Higton <dave@davehigton.me.uk> wrote:

> I've just tried 9000uF (9 1000 uF capacitors in parallel) this afternoon.
> I didn't see a restart in the logs when Clonezilla did a backup to OMV,
> but the backup failed anyway, and now I don't know why.

After a lot of experiments, I discovered that recent OpenMediaVault
versions have a bug that causes it to think the NAS becomes unavailable
suddenly, part way through a backup. I've reverted to a much earlier
version as a result of some web searching. Note that the OMV bug does
not appear to be repeatable but has been reported by several users.

I was able to get a full backup, but I'm not out of the woods yet.

I also saw OMV restart once anyway during a backup, so, although the
9000uF made things better (failed less often), the problem is still
there.

Double fault situation.

David