1
\$\begingroup\$

I have a circuit with split ground planes one for analog, and one for microcontroller/DC-DC etc. The microcontroller will be using its ADC to sample from the analog circuitry.

Ideally I would like to connect the microcontroller to analog ground, so that the ADC voltage measurement is correctly referenced to the analog ground, and not digital ground. However, as the microcontroller is inherently noisy, this is likely to create return current thought the analog ground, which is not something I want. (Previous versions of the circuit have had significant problems with noise).

What is the best practise here?

\$\endgroup\$
  • 1
    \$\begingroup\$ I think I answered this in your other question, didn't I? \$\endgroup\$ – Nick Johnson Jul 10 '15 at 15:26
  • 1
    \$\begingroup\$ Why are you so set on splitting the ground plane? If you don't know what you're doing, it's going to causes you a world of problems. \$\endgroup\$ – Matt Young Jul 10 '15 at 15:27
  • \$\begingroup\$ Best practice is don't split the ground plane if at all possible. \$\endgroup\$ – Spehro Pefhany Jul 10 '15 at 15:42
  • \$\begingroup\$ Why is this best practise? Nothing i have read has suggested this (i still havent managed to read your article @NickJohnson) \$\endgroup\$ – Mauvai Jul 10 '15 at 15:46
  • 2
    \$\begingroup\$ @Mauvai In short, split planes can be done right, but they're far easier to screw up and leave you in a worse situation than you started. And, nearly any split plane board can be made into a single plane board with no loss of function. \$\endgroup\$ – Nick Johnson Jul 14 '15 at 9:01
1
\$\begingroup\$

Manufacturers recommend that you use separate planes because they don't want switching noise from the uP (which shows up as ground bounce) commuted into the sampling of the ADC (which shows up as noise in the LSB). Generally speaking, if you have a single ground plane and reasonable decoupling on the chips consuming the power, you'll be fine. and if you are getting some lsb noise you can stick an front end signal conditioner to push your input signal out of the noise floor (granted at the expense of some dynamic range).

If you can't do either of those things, and you really really want to have separate planes, then the typical suggestion is to tie the ground planes with an inductor that will block the switching noise from showing up but leave your DC bias levels the same.

I'd be interested to know what your application is that requires that level of noise suppression.

\$\endgroup\$
  • 1
    \$\begingroup\$ Cardiac monitor using new sensors. The signal is barely above the noise floor (after the 50hz notch). The last version needs serious improvement on, so a redesign form scratch \$\endgroup\$ – Mauvai Jul 11 '15 at 12:18
  • \$\begingroup\$ Have you considered having your sensor front end be battery powered and to send signal to your processing side as a differential? That might also help with the life safety issues although those are going to be a challenge either way. \$\endgroup\$ – Chuck Jul 13 '15 at 18:48
  • \$\begingroup\$ The entire thing is already battery powered, and the sensors are differential, with a diff amp shortly before the adc. \$\endgroup\$ – Mauvai Jul 13 '15 at 19:21
  • \$\begingroup\$ Ok, so you set the gain on the differential amp to give you full scale on the ADC and you're done. Any common mode noise is rejected, the bulk of your signal will be represented in the 10 - 12 upper bits of the ADC and any ground coupled noise will be down in the lsb where you can filter it out in software. Make sure you've got reasonable decoupling capacitors on your Vdd lines on the CPU and just use a single ground plane. \$\endgroup\$ – Chuck Jul 13 '15 at 19:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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