PCB practices: Grounding, and interaction b/w switching and control loops

Question

This is kind of 3rd in a series, following the posts Advice on ground plane in my first PCB and then Separating ground planes in PCBs. In terms of material, though, it can be considered a post on its own. So to summarize the PCB below (prepared using ExpressPCB) is for a flyback converter implementation using the LT3748 boundary mode flyback controller IC. The circuit is taken from the LT3748 datasheet itself (see very last page). This is for a class project.

I have attempted an improved PCB based on advice in previous posts. I have two versions, shown below.

My questions are:

1. The switching loop goes around the controller (since the tracks go under the IC - but in the same layer of the IC). Is this very bad for noise? BTW, would you consider the switching loop to be Vin-xformer-Q1-R8, or C1-xformer-Q1-R8?

2. For top version, all tracks are on top layer only, but many of them go under the IC (on the same top layer). Is that bad practice? I cannot guarantee the IC won't heat up since the datasheet quotes performance at temperatures up to 150 C.

3. I have only one ground on the primary side. I'm not sure how in my situation to keep a separate ground for the small-signal and high-current loops. I can do but this will put a few breaks in my ground plane...

Finally, as a note, another difficult constraint I was trying to meet is that the controller datasheet says the sense resistor (R8) ground should be close to controller IC, but at the same time to keep the switching loop tight R8 needs to be close to the capacitor.

Keeping all of this in mind, which design would you think would perform better - or is it neither :) Note that teh second design has a few tracks on the bottom layer (green).

Answer 1

1. You want the ESR of C1 to be much lower than the V1 so the loop is small. You also want it lower than Q1 rON to be reduce input ripple.

2. Top one is hard to analyze as the ground plane helps Q1 but then gap becomes a slotted ground radiator antenna.

I think neither is better as the external transformer is going to be excessively noisy and ferrite CM chokes will be mandatory to reduce radiation of Vout cables acting as noisy antenna.

I suggest you perform a bench test using a generator to drive Q1 C1 with Rs & transformer & output fully loaded (D,C,R). YOu don't need a PCB to test this EMI problem and see if you can measure the sense current and output accurately without noise and verify EMI egress with a shorted loop scope probe for near field noise. Designing a PCB without this awareness will be pointless.