"Grounding" according to Analog Devices

[buta]

https://www.analog.com/en/analog-dialogue/articles/staying-well-grounded.html

Cant get past the stupid cookie consent.  Why do they do this....

cookie consent popup->cookie details->decline cookies at bottom right left

[m k]

What am I seeing, electricity or heat?

If electricity then how it is measured?

[Siwastaja]

The thing is that return path at high frequencies go only right under the traces if there is a solid ground plane, so they will not travel to somewhere far away into the analog part.

... but the return currents of low-frequency signals (down to DC; including power!) do not follow the track on the top, but distribute over larger area, in the ratio of path resistances.

While solid, uncut, single ground plane is known to be optimal for EMC and EMI in almost all cases, the exception where cuts can increase performance is high-resolution, high-accuracy low-frequency single-ended analog signals. It does not help with EMI; but it helps with accuracy, simply by removing or reducing the DC supply return current from the measurement path.

Careful placement of parts is of course super important. When one reaches a solution which minimizes the DC level problem even without cutouts, that placement also carries least EMC risk if/when the cuts are added. In other words, the separate area where the "ground" acts like a pseudo-differential signal, should be at some corner of the PCB, far away from any fast signals. Crossing a plane gap with a fast signal is a disaster.

[ejeffrey]

The thing is that return path at high frequencies go only right under the traces if there is a solid ground plane

The case in the article was when a large DC return current is flowing past the analog circuitry.  It's DC so it will fan out to cover the entire width of the ground plane and create a voltage gradient across the board due to the non zero resistance of the copper.  This means that each device in the analog section sees a slightly different ground voltage and if you have ground reference signals they will see that offset.

The ground slit has no traces over it but corrals the return current away from the analog circuits.

It's not an ideal situation to be in but it does happen, particularly when building very high resolution low frequency analog circuits such as DMMs.

[David Hess]

10 years ago when I was speaking to an AD Lab engineer he said they use solid ground plane for characterization of their ICs and design of evaluation boards.

They can do that because their evaluation boards are much simpler than what most of their customers are doing.  They only require the analog section and its digital interface.  Those boards, and the specifications that come from them, represent the best case performance available from the part.  Real world designs may never achieve that no matter how well they are laid out, like with multiple separate converters where a single point ground is impossible.

Definitely recommended to watch this training. This is an expert with 50 years of experience of signal integrity and EMI and he assists companies in these matters.

But he is not an expert in precision design.  He even admits that the people he advise are not dealing with precision, like telecom companies.  The difference can easily be a factor of 1000 when dealing with 16+ bit converters.

Doing what he suggests will prevent getting full performance out of high resolution converters, and for many applications that is good enough.  The RF guys get away with it because they are not combining baseband conversion and precision, and other factors like IP3 limit their ultimate performance anyway.