PCB 'Emi proof' design

Currently I'm designing a GPS basestation which will have a radiomodem (broadcasting at 407-480MHz), a ARM7 microcontroller running at 60MHz and a FTDI USB chip. The FTDI USB chip even runs at 480MHz internally, which is in the working zone of the radio. Due to all the harmonics and these high frequencies from the PLL (which ultimately will flow out of the device's power pins), I'm extra cautious with this PCB design.

We've had some discussions among colleagues what practices are best for EMI-proof design. Especially making the microcontroller 'quiet' is important.

Currently my own approach was based on this question, which was more about decoupling. From the recommendations I've changed my PCB design to have a local ground plane under the microcontroller, which is seperated from the global ground plane. I connected this local plane to the global plane using 4 via's under the chip. Same practice goes for the FTDI USB UART bridge. All caps are routed as close as possible and orientated in a way that the VCC and GND pins have a short connection.

I feed the power in with a via from the supply layer. The GND is a local plane, so that doesn't require a via. I don't have a local supply layer, neither using ferrites to precisely separate the planes.

However, my collogue thinks that having an extra via straight to ground is better. His designs didn't involve local ground planes. All 4 layers are filled with ground, VCC is manually routed. The caps are placed closely, but sometimes the GND connection doesn't have an immediate connection to the GND pin of the controller. The ground plane under the controller isn't continuous , because it's completely broken up because of signals.

His thoughts was that the ground of the caps and pins are very secure to due the global ground plane and each via. He didn't have as much faith in my design because the ground planes are separated. His designs passed the EMCs tests, so it kind of makes me wonder whether all this trouble even makes a significant difference. I'm quite confused by that, because some appnotes tell you it's an absolute must to do local ground planes and good decoupling layouts.

My question simply put is: which design practice is better for EMI practice?

1. A GND are connected to a local plane first, which is separated from the system. This is connected to the global plane at 1 spot.


2. Each GND pin is routed manually to the global plane. Thus it means that all GND connections will get their own via. Not necessarily importance to a continuous ground plane under the controller.

Answer

I recommend Noise Reduction Techniques in Electronic Systems by Ott for such topics.

Don't have seperate grounds, but do only connect them in one place.

Connect the ground to the plane; don't route ground (or power)