| ??? 06/07/07 04:30 Read: times |
#140367 - Yes, yes, yes, but... Responding to: ???'s previous message |
Andy said:
A (perhaps THE) highlight of last October's Audio Engineering Society convention in San Francisco was a talk by Henry Ott. The topic was PCB layout for mixed-signal devices, obviously a subject of great interest. He started by saying, "Yes, every single converter data sheet you read tells you to split the analog and digital ground planes and connect them together at only one point, near or under the converter. So what do you do when you have more than one converter?"
He basically indicted all of the chip manufacturers as lazy. Yes, that's true. Andy said:
The key point Ott made (and it had many people scribbling in their notebooks because they never heard it before) is that return currents follow the path of least inductance, which for fast digital signals means that the return current is essentially constrained to the plane directly underneath the signal trace (which is why you need a ground plane in the first place). Yes, that's all true, if the magnetic coupling between signal trace and ground return current is high, means at places where the signal trace runs directly over the solid ground plane. But there are lots of inhomogenities on a real board. Look for instance, where the digital signal traces enter the ADC, far away from the digital ground pin. Consider the chip itself with its bonding wires, all far away from a ground plane. Where the digital signal traces hit the chip there is no strong magnetic coupling to the associated ground return currents at all, resulting in digital ground return currents leaving the digital board area. Here the splitting enormeously helps to control the digital ground return currents. Andy said:
If your digital signals have to cross a split, then you're really in trouble. Of course, this must never happen. Good filtering (and other techniques) is essential for all signals crossing the gap. Andy said:
The low-frequency signals' return currents spread out in the plane, so you do need to take care of that. Yes, the according voltage drops are negligible anyway. Andy said:
His final recommendation was: "You people need to start spending the money on multilayer boards." Yes, a good idea. But spend also money on chip packages consuming the least board space possible to keep the loop area minimal, where the magnetic coupling between digital signal traces and ground return currents decreases (see above). Andy said:
I suspect that a solid ground plane is better in terms of radiated noise than a split plain. Well, we measured the opposite. Nevertheless, what you finally see also heavily depends on the entire concept. Common mode noise is not only a question of splitting or not. I fully agree with you: Complex circuits can suffer from the splitting when the router is unexperienced. Then, dividing the board into analog and digital areas and using an overall solid ground plane might provide the best compromise. But if you need highest performance, then splitting the ground plane (by an experienced router) will offer the best results. Kai |
