DJ Delorie wrote: >> In my 20+ years in engineering I have yet to see one case where >> splitting a ground plane under high-speed ADCs has worked. > > What about high precision ADCs? I'm working on a design using ADE7753 > power monitor chips (16-bit ADCs) , and their own app note (AN564) > shows a ferrite isolating analog ground, and a 10R resistor isolating > AVdd. >
Ok, goes a bit OT here but maybe it helps others as well. Will take a bit longer to explain scenarios. Personally I would not split but make sure sensitive inputs are fed differentially. The datasheet shows some uncertainty about this: Page 8 says, quote "To keep ground noise around the ADE7753 to a minimum, the quiet ground plane should connected to the digital ground plane at only one point. It is acceptable to place the entire device on the analog ground plane." ... while figures 29 and 30 have both grounds tied together. Yes, in AN564 there is an inductor connecting both grounds. This concept falls apart when you have more than one converter or when stuff from the external world needs to connect to the analog side in non-diff fashion. But the real concerns are others. Examples follow, and I've seen them all. a. Since power meters are mostly mounted to an outside wall in the US this is what can happen during a thunderstorm. Split grounds act as a dipole antenna with the current maximum at the joint, or in case of AN564 an inductor. Lighting hits uncle Leroy's oak tree, a gigantic EM pulse develops, reaches the meter, finds a nice and receptive dipole antenna in the form of a split plane, fssst ... *BANG*. Meter's gone. b. Uncle Leroy has a huge antenna next to the garage where the meter is located. Inside the garage is that humongous CB amplifier that he tells everybody never to talk about because it's, ahem, not quite legit. He keys the mike, AM-modulated RF floods the woods and develops a varying voltage across that inductor that peaks high enough to upset the ADC in the meter. Upon which the meter probably goes into some kind of lock-down. c. Someone walks up to the meter and then his cell phone rings. One never knows but the chip might have some bipolar (BJT) paths between the two grounds. While it isn't sensitive to GHz frequencies per se just a wee bit of that RF can rectify at a few BJT junctions in there. Since the transmissions are burst-like the rectified signal consists of pulses of who-knows-what characteristic. This might cause more or less serious ADC conversion errors. I could go on. Bottomline is that you never really know what kind of outside disturbances a piece of electronics might be exposed to and how a chip will react to noise. Maybe not now, but what if some new 4G standard pops up in the maketplace that turns out to be a real bear in terms of susceptibility? In contrast, isolating an analog supply voltage can be a very smart thing to do. As long as either side of the isolation resistor or inductor is bypassed really well. -- Regards, Joerg http://www.analogconsultants.com/ _______________________________________________ geda-user mailing list [email protected] http://www.seul.org/cgi-bin/mailman/listinfo/geda-user
