Separating ground planes with an impedance

Go To Last Post
7 posts / 0 new
Author
Message
#1
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Hi,

A very general question: Is there any reason ever to separate two different ground planes with a ferrite bead, inductor or resistor?

One occasionally comes across such solution in the internet, apparently in an attempt to reduce coupling of noise between the analog and digital grounds. However, I haven't yet found reliable references to why it would work better than low-impedance or completely isolated grounds.

- Simo

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

You can simply make two separate ground planes on the bottom layer, and join them at a single point with a trace, away from the signals you are trying to clean up. The idea is to keep the high currents for example from a motor away from noise sensitive circuitry, like an analog converter or some high speed IO or analog lines, etc..

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

UNiXWHoRe wrote:
You can simply make two separate ground planes on the bottom layer, and join them at a single point with a trace, away from the signals you are trying to clean up. The idea is to keep the high currents for example from a motor away from noise sensitive circuitry, like an analog converter or some high speed IO or analog lines, etc..

Yes, this seems to be what people usually suggest (or a single solid ground or star ground, depending on the application).

At least Freescale/Motorola seems to favour the use of a ferrite bead separating the AGND and DGND in their DSP application modules (for example http://www.freescale.com/files/a... -- see page A-5).

On the other hand, all other sources I've seen say you must minimize the ground impedance and possibly use split planes connected at only one point. This is confusing :?
It's itching my curiosity that the Freescale engineers may have some less known high level information that contradicts the general consensus...

- Simo

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

I am far from having an explanation, but perhaps
I can explain the point of view that I am often
using when it comes to EMC issues.

The key to better understanding is often to think
about currents and the path that different currents
probably will take.

Now assume we have to planes that are not directly
connected. But for example say on one plane
is the LED of an optocoupler, and on the other
plane is the receiving diode. And on one plane
we have a lot of HF signals and currents.

Now an optocoupler is not perfectly isolating, due
to capacitive coupling between LED and diode. So the
optocoupler provides possibility for HF noise to
couple from one plane to another.

If we now connect the two planes by a short trace, the current may find this way back and it can flow.
Now this current may induce voltages n other loops
and that increases our noise level.

If instead we use a ferrite-bead to connect the
planes, we have them connected for low frequencies
(DC ....) but keep them separated for high frequencies. So the high-fequency current-loop that
existed before is now broken and no HF-current will
flow.

In my thinking every EMC problem an every PCB needs
its own careful treatment. There is no "cures-all"
solution.

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Thanks ossi, it really starts to seem it is too application dependent to generalize.

After returning home I checked datasheets of a few random DSPs, 10MSPS 16-bit ADCs and 24-bit ADCs from different manufacturers (other than the aforementioned Freescale). All of them recommend to use a single solid ground plane with the usual 'keep digital, analog and clock stuff away from each other'.

After the survey I started to feel it'll be an overkill to use such a noiseless GND with the only 10-bit ADC of an AVR :D

- Simo

Edit: 10MSPS, not 10Mbit/s

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Update:
Browsing through the Analog Devices "Rarely Asked Questions" column archive, I came across of

http://www.analog.com/en/analog-...

where an Analog Devices engineer strongly suggests against using resistors or ferrites between AGND and DGND of a converter. So I guess the case is closed for this particular case.

Other kinds of ground connections are still open. If I recall correctly, somebody (ka7ehk?) suggested in some older thread that connecting a power ground to a star point through a ferrite reduces noise.

I however still don't have black-on-white information on why a ground separting ferrite would be beneficial in a DSP system. The usual problem of too much to read and too little time :(

- Simo

  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

The ferrite bead will block high frequency signals from crossing grounds. With your digital ground you usually have a lot of noise brought on by the clock signal and the flurry of activity from the digital systems that come as a result. This will keep the noise from interfering with your analog signals, minimizing the need for costly filtering, changes in sample frequency, etc etc.