1
\$\begingroup\$

My layout for an analog signal circuit uses three layers.

I have a ground plane, top layer with signal traces, and one layer with power traces --- I have +15, -15, +5, and -5; all four voltages generated from a 5V input.

Does it make sense to split the ground plane into two separate planes, one for power (i.e., for the decoupling capacitors), and one for the components that connect the various signals to GND?

And since I only have a handful of those, I wonder whether the signals should use a star ground configuration, instead of a ground plane?

The only connections to ground for the signals come from one Sallen-Keys 2nd-order 1MHz low-pass filter (two connections, since it has some bandpass gain) and one AD8138 to produce a complementary/differential output (two connections --- one resistor that connects to GND and the Vocm input that's set to GND). Plus the grounds of the two output connectors. So, it will be rather easy to set up the star ground layout for these, where the center point is the (only) point that connects to the other layer with the ground plane.

Does this make sense? Please feel free to suggest any alternatives.

\$\endgroup\$
9
  • 1
    \$\begingroup\$ What is the highest frequency that you expect to be present on your input for the filter to block? \$\endgroup\$
    – The Photon
    Dec 19, 2016 at 0:32
  • \$\begingroup\$ The input of the low-pass filter is driven by an AD8421 (instrumentation amplifier) without any signal connecting to GND. Its gain will be 10 or 20, so its bandwidth should be close to 10MHz. The input of the AD8421 will probably have much (much) higher frequency components, but there are no signals connected to GND there. Not sure whether it will still be relevant. \$\endgroup\$
    – Cal-linux
    Dec 19, 2016 at 0:45
  • \$\begingroup\$ All of these signals are presumably referenced to ground (when you say the signal is at, for example, "1 V", you mean "1 V above ground"). You'll still want to have a ground plane beneath tracks carrying these signals, if you can. \$\endgroup\$
    – The Photon
    Dec 19, 2016 at 0:49
  • 1
    \$\begingroup\$ It's about the signal frequency (both pass-band and stop-band). Old op-amps were not much good for 10 MHz and up. Even today many op-amp applications don't require handling 10 MHz signals. And even at 10 MHz you can probably get by with a star ground. But if you can afford a ground plane, why take the chance? \$\endgroup\$
    – The Photon
    Dec 19, 2016 at 0:59
  • 1
    \$\begingroup\$ Interesting this last comment "if you can afford a ground plane, why take the chance?". I always thought that a star ground is the ideal solution --- there is a (virtually) true GND and everything is referenced to it --- and that a GND plane is the second-best and we use it because star ground is almost impossible to implement when we have many GND points. But from your answer, I suspect that at high enough frequency, the effect of the parasitic inductance of the traces that connect to the center of the star are significant, and a GND plane then has better performance than a star ground? \$\endgroup\$
    – Cal-linux
    Dec 19, 2016 at 1:05

1 Answer 1

1
\$\begingroup\$

Implementing separate ground planes, one for decoupling caps, and the other for analog reference, is a good idea. It beats a star grounding scheme, and is easier to implement. The Vocm pin, especially, appears good to keep quiet.

Star grounding is sometimes used as a connect method for multiple ground planes, as noted in the references above.

\$\endgroup\$
2
  • \$\begingroup\$ Not sure how these two references suggest this. They both talk mostly about mixed-signal (analog + digital) systems. Are there any specific sections I may have missed, or is it that in general the ideas in those references lead to the conclusion that separate planes for decoupling and for analog reference is a good idea? (or did I just misinterpret your answer?) \$\endgroup\$
    – Cal-linux
    Dec 20, 2016 at 20:23
  • 1
    \$\begingroup\$ Digital sections of a design are usually considered as noise sources in a mixed-signal system. But so are the return sides of decoupling caps because, after all, they are shunting noise spikes to ground, where such spikes have to find their way to the power supply return. What you don't want is for those spikes to hit your Vocm pin on their way to the power supply return. Your low-pass filter and the AD8138 are analog, demanding the above-referenced care and handling. Your need for decoupling caps implies a switching power supply. If not, then you could get by with one ground plane. \$\endgroup\$
    – Otto Hunt
    Dec 22, 2016 at 0:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.