How to reduce effects of induced current noise caused by high in-rush?

[lpcware]

Content originally posted in LPCWare by jdupre on Wed Jun 01 17:17:06 MST 2016
I am seeing problems that appear to be related to high in-rush current on wires that run under my LPC1517 based controller card.  Although the normal operating load is less than 10 amps, I am measuring very short duration peaks of up to 160 amps.  Reducing inrush by "soft-starting" the loads has shown to be a great improvement, but there are still unexplained issues with IRQ inputs and even what appears to be memory corruption when these big in-rushes happen.

1) What is the best way to deal with unused pins?  On the 1517 the are set as inputs by default.  Is configuring them as outputs with a pull down a better way to go? (My first run of boards have the unused pins pulled out to pads for experimenting...)

2) On pins that are used as inputs, is there a method to filter out induced transients with an RC network or something like that?  [ We have an off-board SPI display driver that was going crazy with this inrush until we terminated the SPI lines at the far end with a series resistor and cap to ground at the recommendation of the manufacturer.  Can a similar method be used on inputs to the CPU? ]

3) Is there something I can do with the board layout?  Currently I have:
   1 Top - Components
   2 Ground
   3 Power (an island of 3.3v under CPU, surrounded by 5V for other parts)
   4 Bottom - Signal
Would it help to bury the traces and have ground on the bottom layer that is closest to the high current wiring?

- Joe

[lpcware]

bump

[lpcware]

Content originally posted in LPCWare by starblue on Sat Jun 04 01:15:50 MST 2016
I'm not an electrical engineer either, but here is what I'm thinking:

You are probably seeing inductive effects of the fast change of current during the inrush.  The induced voltage on other lines is proportional to the rate of change of the current, which seems to be very high (can you measure it?). The bad new is that it is hard to shield the magnetic field. You can mitigate its effect by keeping lines short and close together with the return path. Also keep the hot wires close together, hopefully their magnetic fields will cancel somewhat.

For the signal lines the ground plane ensures that the return path is very close. For the power supply I would be a bit worried about the 3.3V island. Maybe you could make that bigger to cover anything running at 3.3V (assuming the 5V parts are more robust).

I don't think the order of the layers matters.

Also, I don't think it is only an effect on the I/O lines, since you are seeing memory corruption. Maybe the power supply the microcontroller sees is dipping too low.

If you want to experiment with RC networks I would put a small capacitor (in the picofarads range) close to the controller pin, and add a resistor next to it so that you get a time constant that filters away the transient but doesn't make the signal too slow.

[lpcware]

Content originally posted in LPCWare by jdupre on Fri Jun 03 12:02:46 MST 2016
This is a dimmer for banks of up to 24 LED lamps.  Each lamp has hundreds of LEDs and it's own switching power supply that takes 90 to 265VAC input.  We are supplying dimmed ~120VAC to these lamps. The inrush is due to the capacitors in the power supplies.  This is what I have been given, so there is no changing that part of the equation.

Physically, the dimmer PCB is long and narrow.  On one end is Wi-Fi & RS-485 IO, followed by the LPC1517, then the power supply, and at the far end is the line voltage IO including a couple of triacs that do the dimming.  The line voltage side and the processor side are separated by a transformer.  The DC power and ground planes stop at the transformer.  The PCB is enclosed in a metal channel, and the two "hot" wires to the lamps pass along the length of my board.  By securing the wires, I can achieve at most 2cm separation between the load wires and the PCB.

I've got a Fluke 43B to measure the inrush.  Worst case appears to be turning on the lamps from off to a 50% duty cycle on the AC.  At first this was a little surprising to me as I would expect worse case to be 0 to full on.  But I see that at 50% the output goes from full off to the top of the sine wave, rather than ramping up with the sine wave.

I have been experimenting with "soft starting" the lamps (a quick fade rather than bumping from one level to the next), but I am still measuring substantial inrush even with small changes to the duty cycle.  Ideally I'd like to be able to "bump" the lights on, so I am looking for hardware methods to either reduce the inrush, or reduce the noise effects of the inrush.

Putting an large inductor on the line helps, but that is an expensive solution.

I think there must be methods to absorb this sort of noise with an RC network on the processor IO pins, but I am not an electrical engineer...

[lpcware]

Content originally posted in LPCWare by MarcVonWindscooting on Thu Jun 02 14:26:03 MST 2016
Interesting problem. Can you provide a bit more information?
I'm not an expert in this area, but I already had some bad experiences with inverter experiments.
What's the geometry of the power lines and through what path do you return the current (to GND)?
Sometimes it can help to spread the current in order to reduce the locally high field strength.
Similarly, don't expect a thin layer of copper to perfectly cancel transients. Copper is not an ideal conductor.
Are you sure, your disturbances arise from magnetic field and NOT at least partially from 'resistive divider' effects? Like for example raising your MCU's ground by a few volts agains some other point (connected to a pin), because the inrush return causes a considerable voltage drop across your 'GND' area and suddenly your pin 'sees' negative voltages??
160A is quite a bit, even without fast changes (statically) :(

Marc

[lpcware]

Content originally posted in LPCWare by wmues on Thu Jun 02 00:47:48 MST 2016
You can put a metal sheet under your PCB. If this improves your problem, it makes sense to add a GND plane to your PCB.