Ferrite filter on switched 3V3 supply

Thread Starter

theeon

Joined Nov 5, 2010
3
Hi there

Im wondering if anyone here have some toughts on using ferrite filters on their AVCC/DVCC right after the regulator output.

My setup is powered from an xilinx card from an fmc connector.
It delivers switched supply at 3V3 and 2V5
http://www.xilinx.com/products/boards/sp601/reference_designs.htm

On my board i use this LM2831Z High Frequency 1.5A - Step-Down DC-DC
Its switching at 3Mhz and regulates from 3V3 to 1V8.
http://cache.national.com/ds/LM/LM2831.pdf
I must add that the buck regulator acts as a LC filter in itself.

This again supplies the high speed analog to digital converter (adc). It samples at 100MSPS. Its decoupled with 10nF and 1uF caps localy.


The noise that reaches the adc via the supply needs to be minimized.
Wish i had a -140dB noise floor on the samples :D

Im considering an ferrite filter between the 1V8 regulator and the adc.

BLM18PG331SN1
http://search.murata.co.jp/Ceramy/image/img/PDF/ENG/L0110S0100BLM18P.pdf
An 1Amp ferrite bead that gets to 330 Ohm when passing 100Mhz
And less than 0.2 Ohm for dc currents.
Wide band version, to prevent most of the high freq noise thru.

Im certain the regulator makes noise at 3Mhz.
When the ferrite hits 50 Ohm at that freq, will that be sufficient to dampen som of the noise?


In addition i want to place a 1nF ceramic capacitor to ground in front of the bead, viewed from the supply regulator.

Regulator <-----> 1nF to GND <-> Ferrite <----------> ADC


Anyone have some knowledge on how this will behave?

And is there some issues that may appear that i must know of?

Kind regards,
Thor
 
Last edited:

praondevou

Joined Jul 9, 2011
2,942
Yes, it will certainly do some filtering.
There are other things too that you should consider. Is your layout correct for such a high frequency application? Did you make a frequency analysis or how do you know what your noise level is? What are you sampling? Can you be sure that the high frequency noise is not coupling into your signal traces/wires? etc.
 

Thread Starter

theeon

Joined Nov 5, 2010
3
Hi and thanks
The layout is througtly planned.
The ADC input lines are terminated to 50 Ohm using the board stackup and a trace width of 0.2mm and clearance 0.3mm

Also the stackup is subject to change, its currently a four layer board.

------ High speed signal
------ GND 1
------ GND 2
------ PWR / Signal

Ive been reading alot about this subject, and found a better solution that uses a power plane.

------ High speed signal
------ GND
------ PWR
------ Signal

This should act as a high speed capacitor for the adc supply pins.
But how can i separate the analog and digital supply pins for the adc when using a power plane?

Also i use spectrum analyzer with a high frequency probe for measuring the vcc ripple. The ripple of the LM2831 that is.
Sampling the ADC without a signal gives me the current noise floor. Its about -80dB

I want to get the best wideband performance possible for sampling. For testing im feeding it 19.178456Mhz from a signal generator. Ofc i use different signals, and they are always filtered right before entering the adc pcb sma. Also there is little or no crosstalk on the channels. They follow in ground tunnels, straight into the adc.

The current layout dont have separate supply regulators for the analog and digital vcc, but ive done my best in separating them with long power traces and decoupling them before they enter the adc.

Is it worth it having separate regulators for the adc supply rails, and can i expect a significant improvement?


Filtering signal terms and wisdom are not my field, altought i got a bachlor in electronics i am still inexperient. All signal processing is black magic, hah.


Its great fun learning tho


*Thor once more leaves the physical world, entering the realm of the datasheet
 

SgtWookie

Joined Jul 17, 2007
22,230
Would -115dB do?

See the attached schematic & plots; I put it together using Elsie:
http://tonnesoftware.com/elsie.html

You're probably aware that if you're off by much anywhere, it'll throw the notch way off frequency.
[eta]
A T16-6 iron powder core with 7 turns of wire (up to AWG-23 will fit, but better use smaller to make it easier) is ~93.1nH; that will actually work fine for the left tank.
A T44-6 iron powder core with 3 turns gives ~37.8nH, which will work fine for the right tank.
 

Attachments

Last edited:

SgtWookie

Joined Jul 17, 2007
22,230
Here's another version, Chebyshev based. It's 3dB point is 200kHz, and it's ~127dB down at 3MHz, and just keeps going down from there. It will be much better for getting rid of the harmonics, too. The values aren't as critical as the previous filter. The caps need to be low ESR, of course.

You can use a couple of T30-6 iron powder toroids with 6 evenly-spaced turns of wire on them to get ~129.6nH.
 

Attachments

Top