Electronic – Groundplanes for Wideband synthesizer

Consider this a theoretically biased answer - I've not dealt with multiple ADCs and a separate ground plane. This will (hopefully) not be your star answer but may raise some issues worth noting. Also - if any of this sounds like hogwash or ill advised (variations on the same theme :-) ) please say so (preferably gently) - leaving uncommented advice which you consider misleading reduces the worth of the material as a resource for others. .

• What you have done sounds close to ideal. A second ground plane is a luxury not always available in "lesser" systems.

• One may be tempted to partition the ground plane into N segments radially expanding from the single common ground point, but that has good and bad points.

• Considering where and how you return the grounds of the signal sources can be an interesting exercise.

  If possible you return the sources' grounds to the analog ground plane, but that then raises issues re sources which are powered but which do not themselves have separate power and analog grounds. How do you return the source power ground to the power ground plane and the source analog ground to the analog ground plane?

  In the case of eg instrumentation amplifiers this may be easy as the analog ground is conceptually separate from the power ground.

  In the case of single ended sources you may need to look closely at what happens to ground currents between power and analog. If the local power ground has a potential dc offset relative to analog ground you may wish to isolate this component from analog ground. To do this you may even go as far as providing an AC filtered DC feed to power ground for the sources analog portion and an AC ground path to the analog ground plane. This effectively creates a local analog ground for the source's circuitry - eg perhaps an inductor from power ground plane to local analog ground with a capacitor from local analog ground to analog ground plane.This sort of magic is liable to be needed only in extreme cases - it is to be hoped that in cases where DC components are large enough to matter that the device designers have accommodated it (as they have done with your dual gnd ADC's.

  An example where this may not be the case is eg a microcontroller with internal DAC being used as a signal source for an ADC. For this arrangement to make sense (DAC-ADC) there will probably be some other analog function or convolved signal as well as the DAC output. In this case, how do you treat the microcontroller ground and what differences do the choices make.

• Both ground planes will probably be interrupted by vias interconnecting other planes. In extremely demanding cases, which yours sounds like, care needs to be taken re unbalancing of go and return signal paths for critical analog signals. An analog signal track which crosses a break in it's analog ground plane creates a slot antenna which may be both a radiator and a receiver. In many cases the effect may be small enough to be neglected but you need to know that this is so by design and not by good (or bad) luck. Ground plane breaks also provide increased loop area which can be important in critical cases. (Loop area between go and return can occur in fully balanced cases when tracks are used for both paths - usually eliminated by proper groundplane use.)

Generally you don't need separate planes for power. If you have a good solid ground everywhere (you do want a plane for that in many cases), then the only issue for power is that the voltage drops due to the power current and DC resistance in the traces are acceptable. Don't worry about high frequency AC impedance in delivering power that much. Instead, bypass the power locally to the local ground at each place it is used.

In fact, especially for sensitive analog circuits, I often add a little deliberate impedance to a power feed so that a local capacitor has something to work against to filter out the high frequencies. If you put a 0805 chip inductor, for example, in series with the power connection of each IC, then follow that with a 20 uF ceramic cap to ground, you will have clean power everywhere. This is for chips and individual circuit sections that draw up to a few 10s of mA.

I just checked, and the jellybean 0805 chip inductor I commonly use for this is 950 nH and 600 mΩ. Just about any reasonable copper power trace will have less inductance and resistance than that. With this extra impedance in series with the power, a little more due to the power trace doesn't matter. With this strategy, allocating a plane to a particular power net it just a waste of routing space, and doesn't produce as good a result anyway.

As I have said many times before here, another important thing to do for low noise is to keep local high frequency currents off the main ground plane. The ground connections of a noisy subcircuit are all tied together with a local net, then that net connects to the main ground plane at a single point. The same thing works in reverse to keep noise from getting into a circuit. A sensitive analog circuit should also have its grounds connected locally, then that net tied to the main ground plane at one point.