Decoupling a big power burst.

[dpenev]

Hello,

I have build a system involving embedded Intel Core i5 CPU.
With Turbo mode enabled I am observing a big ~30Watt surges for about 2ms on each Linux command I send.
The power supply I have incorporated can cope with this but still strong voltage spikes appears on the input supply lines of the system.
This affects the conductive emissions of product and may be a problem.

I have experimented with a big capacitor bank put on various places along the power path but I have managed to reduce
this low frequency noise only about twice. Using bigger capacitors is impractical.

Anyone having an idea what is the best way to filter out this kind of noises?
Probably some active techniques ?

Thanks
Dimitar

[Miyuki]

Reduce CPU power limit for turbo period or need to have enough big capacitor bank to hold a charge for it
Or if overall power consumption is not a problem, take it the opposite way and disable sleep mode and burn more power at idle. It will reduce peaks relative to this state.

[dpenev]

If I disable the Turbo I get reduced CPU performance.
I am not sure if I can keep the CPU at 4GHz all the time but this is also not a solution as it will increase the total consumption dramatically.

I have made  capacitor bank with some low ESR capacitors in parallel which helps a bit
but obviously I need more and really not place for more apart
of the big inrush current I have to deal with.

During the power surges I have ~200mV bumps for about 3ms at the system input. 
I was think about some sort of active serial filtering solution.
The total system consumption is about 28V@10A. 
     

[Miyuki]

Depends on how it is powered, some local LC filter might help.
And what is regulation and resistance of power supply.
That drop can easily be an effect of the impedance of power line.

[dpenev]

Indeed the external power supply resistance influences.
I would like to decouple this noise from the external power supply and keep it inside my box.

In the measurement lab they put LISN device between the power supply and the DUT and this way
they make the conductive EMI measurements power supply independent.   

[T3sl4co1l]

What does your overall system look like (block diagram)?  Where is the overshoot?  Of what significance is it?  How is such a slow swell causing emissions problems?

I might simply wave it away as irrelevant because the front-end PFC shouldn't be responding anywhere near that fast.  But see how many assumptions I make with such a statement?  We really can't do anything without a lot of supporting information I'm afraid.

Tim

[dpenev]

Hi Tim,

I am attaching a simple sketch.
The product should be EMI measured in the lab starting from 150KHz up.

Those voltage bumps are indeed slow and isolated but due to relative fast edges may put some energy around 150Khz.
Depending how they average the spectral power in the certification lab.
I have specified 200mV peak amplitude.
In the lab our product will see 50 Ohm LISN impedance and if this bumps put some inband energy it will be greatly amplified.
I hope we don't have EMI issue but I would like to think a bit in a direction to filter this noise out.

[T3sl4co1l]

Ah, OK.

What does the power supply look like?  How fast does it respond?

In general, switching converters are low pass filters, rolling off near where the control loop rolls off.  That's in addition to whatever ripple/EMI filters are around them, which are usually going to be enough to take the edge off something like this.

Tim