squarewave said:
I have not done anything near as advanced as what you're talking about but I think that all that would matter in this case would be that all grounds connect at the same point.
They have to connect in both a DC sense and an AC sense.
All of the A/D chips that I have ever looked at that I can recall have A pins and D pins grouped together so that they can straddle different ground planes.
No. That's the ancient thinking. They all get tied to the same ground.
[quote author=squarewave][quote author=me]Data sheets for recent devices all recommend one ground plane with no splits (I just checked some recent TI, AKM and Cirrus parts). That's what Henry Ott recommends, too, see here. Instead of splitting planes, you logically partition the board into analog areas and digital areas. Keep the analog signals out of the digital area, and the converse.[/quote]
I'm not sure I fully understand this. Where is this one ground plane connected to the power supply? Or does each board have it's own power supply cable and this multiple ground wired running to the same board (one for each region)?[/quote]
Let's simplify -- we are talking about one board with both digital and analog stuff on it, and one or more mixed-signal devices (converters) bridging the two domains.
The ground plane is connected to the power supply at the power-supply entrance. Perhaps that's a connector with your +/-15 V, +5 V, +3.3 V supplies (and a return for each). The returns for all of the supplies are all connected to the ground plane at that entrance.
If the later is the case, then I can understand how it could work since currents from a digital region would follow the path of least resistance which would be it's own supply cable. It's not crystal clear why it would be better though. Maybe connecting all regions just helps minimize any DC differences (but in theory if there are no connections between those regions it should not matter).
This situation seems to have two boards, an analog board and a digital board, and you are sending digital signals to a DAC on the analog board. In this case, both boards have a power-supply entrance, and by that I mean each voltage and their returns. So say analog board has +/- 15 V for op-amps, +5 V for converter analog and +3.3 V for the converter digital, and the digital board has just +3.3 V and its return. Both boards connect to a common power-supply module. At that the power supply, the grounds are all tied together.
You have a cable with some digital signals going between the two boards. That cable has the signals
and their returns. Yes, those returns are "ground." And yes, if you buzz out between the analog and digital boards, "ground" is the same in terms of DC continuity. BUT -- and this is the crux of the biscuit -- for proper signal integrity you must provide a return for the signals in your cable. That's because the physics says that the return current for those signals will be in the ground wire in that cable, and that's in an AC sense. If you do not provide that signal return in your cable, the return current has to works its way from the analog board, through the power supply cable to the power supply module, and then back to the digital board through another cable. That destroys the signal. (I've done these tests and I've seen it on an oscilloscope.)
Is that a ground loop? Strictly speaking, yes. Does it screw everything up? No, not at all.
Also, off the top of my head I cannot think of a reason why digital traces would be traversing into an analog region or visa versa which seems to be important to the single plane technique.
If your layout is correct, they won't. But in many cases, you are stuck. I used the example of putting analog mux chips to choose a signal path. You have to route your digital signals to the mux chips' control inputs, and if you look at the pin-out of most of those chips, you see you can't avoid crossing.
Also, I find it difficult to accept that this would actually be better in all cases. For example, an LED driver can generate very fast and large changes in current. I would be very weary about connecting this to any analog ground plane.
In those cases, you do have to manage the returns -- although I think that in that case, your LEDs are on a front-panel board which is already physically isolated from your analog stuff. And if you're controlling motors, you might even have to optoisolate the control signals from the driver.
It's common to use relays to control analog signal paths. Relays have a nasty inductive kick when they change state. The way to manage that is to run thick traces to the relay coil lines and not just connect it to a plane.
I use separate ground planes and separate boards with separate ground planes for digital and analog and separate ground wires in the same power cable for my external power supply and it works well enough. In fact, I use 3 separate grounds (digitial, analog and then really quiet analog). I put some digital stuff into a regular analog channel strip and I can barely measure the digital noise.
You have an "analog board" and a "digital board," and that's how you partition your domains. I don't know how the two boards are connected other than at the power supply. What signals go between them?
For something that is mostly digital but with little regions of analog for I/O then maybe that's where the signal ground plane divided up into regions would be superior.
It's superior in all cases where you have analog and digital stuff on the same board
and the two domains talk to each other.
If the analog and digital stuff on that board have no connections between them, then having separate power and ground planes is fine.