Noise isolation of digital and analog power rails

[T3sl4co1l]

We must have different expectations of high resolution converters, and high precision circuits, because the first thing I do is get rid of the combined analog and digital ground plane because the digital return currents in the plane create too much noise in the converter.  They must be diverted by cutting slots into the ground plain, and the eventual outcome is if I had started with separate planes connected to a single point ground, which is where I would start anyway.

My schematics explicitly show separate grounds, single point grounds, and whatever multiple separate connections to the same node are called.

FWIW, I did a design and layout with 24b ADCs -- the slow serial S-D kind with differential inputs and programmable gain, for what it's worth.  It used +/-15V supplies (PoL converter, modestly filtered) for the analog section, and plane VDD for converter supply (though I think the VDDA side had a ferrite bead and ceramic caps for filtering?).  Layout was with analog on one corner of the board, digital on the far end.  (I had marked likely ground slot locations on the design, but the customer declined using them, which is fine.)  The acquired samples weren't exceptionally clean, however they were within datasheet expectations (I think it was around 60 counts RMS), and the slight excess was easily attributable to the op-amp in front.  So it seems to be a success, and no revisions were made.  (I wasn't involved in EMC testing, so don't know how the readings were affected by it, if any.)

If you're talking higher bandwidth (which, up to about 100kSps is achievable at 24b, beyond that you hit the Johnson-Nyquist limit, isn't it?), attention may need to be paid even just to common mode currents coming in from the external signals, even if your inputs are modestly high impedance.  That is, the ground plane itself may have enough impedance to matter.

Tim

[David Hess]

If you're talking higher bandwidth (which, up to about 100kSps is achievable at 24b, beyond that you hit the Johnson-Nyquist limit, isn't it?),

Where it really matters is with sampling converters however delta-sigma converters can still be affected.  With integrating converters, which includes delta-sigma converters, problems may manifest as higher low frequency noise or poor linearity.

attention may need to be paid even just to common mode currents coming in from the external signals, even if your inputs are modestly high impedance.  That is, the ground plane itself may have enough impedance to matter.

That is a real problem which requires careful attention to the interfacing.  These days I start out very suspicious about any high resolution application if the control signals are not isolated in some way at the high resolution converter and this gets taken to extremes in the highest resolution multimeters; it is not uncommon to find transformer isolation, optocouplers made from discrete parts, or opto-interrupters used in place of optocouplers because the later were not good enough.  And common mode rejection at the analog input is important if return currents are moving the converter's common mode reference around.

[Jan Audio]

For me its easy, i dont dont know what a ground-plane is, i have a idea and dont need to know too keep things simple.

[tszaboo]

If you're talking higher bandwidth (which, up to about 100kSps is achievable at 24b, beyond that you hit the Johnson-Nyquist limit, isn't it?),

Where it really matters is with sampling converters however delta-sigma converters can still be affected.  With integrating converters, which includes delta-sigma converters, problems may manifest as higher low frequency noise or poor linearity.

attention may need to be paid even just to common mode currents coming in from the external signals, even if your inputs are modestly high impedance.  That is, the ground plane itself may have enough impedance to matter.

That is a real problem which requires careful attention to the interfacing.  These days I start out very suspicious about any high resolution application if the control signals are not isolated in some way at the high resolution converter and this gets taken to extremes in the highest resolution multimeters; it is not uncommon to find transformer isolation, optocouplers made from discrete parts, or opto-interrupters used in place of optocouplers because the later were not good enough.  And common mode rejection at the analog input is important if return currents are moving the converter's common mode reference around.

I'm sorry, but that is not the reason for the isolation. A DMM input is floating, you could connect it to 230V common mode voltage, while the digital part is earthed. I've made boards that had 50A current sources and sinks, high speed microcontrollers, and 24 bit converters on them. For that, you reference all the analog signals to a voltage that is not your ground, and you place the reference voltage next to the ADC, and keep the digital signals and analog signals on different sides of the board. If you have a fast rising edge signal, like SPI, all the return current stays below the trace, your analog signals dont even see that. This system worked with a few PPM accuracy.
There are just simple rules, that are even highlighted in the ADCs datasheet. Dont communicate with it while it is converting. Use differential measurement if you have enough inputs. Delta-sigma is especially simple, as it typically can be driven by few KOhm sources, so you could filter out all the noise. The input is integrating, all those MHz+ noise gets integrated out.
 The real bastards are those high speed 100+KSPS 20+bit SAR ADCs, they have to be driven by a high GBW opamp, and noise could be picked up during the entire conversion. And look at this:
https://www.analog.com/media/en/technical-documentation/data-sheets/238024fa.pdf

A single solid ground plane is recommended

They even went ahead and named all the grounds GND.