@TimNJ @uer166 @T3sl4co1l Thank you all for all this useful information!
Here is how the circuit actually looks like:
And a close-up of the ITH compensation network:
It's mostly based on the Figure 13 from LTC7803's typical applications (datasheet page 33). I just wanted to make a working board and maybe learn something in the process, I was too worried to change anything much
I just increased the input and output capacitance and changed a few components like the diode on the BOOST pin.
Here are some images of the output ripple:
It changes when I put some load on it and this change is audible (more load - higher audible frequency).
Are there no stitching vias? Is that what the "-" actually are?
Yes, these are the vias, Eagle is just not the best at scaling them and when they are smaller on the board view they look like that.
Try to avoid routing high currents over slots in the opposite layer. Example: the transistors themselves over the gate drive traces, which cuts that ground return path, making T4 essentially floating in air -- a high inductance path. This will be evident with an inductive loop probe held over T4 or along the respective traces. In contrast, T5 will have less. (They make a tight loop together, so their fields are overlapping; this won't be obvious just from a basic measurement. It's like a flood light beside a flashlight: both contribute, but the one just seems to modify the shape of the other's field. In reality, both are independent and superimposed.)
T4 gate drive trace I think can be routed underneath it (top side, between D and S), but there isn't enough room for both. Better placement might be with T4 and T5 end to end, so the gates come out the same side; place the controller there. Then ground can be unbroken beneath the transistors, and C13-16 can be placed at either end of the stack (or directly underneath if you don't mind bottom side placement).
Same goes for the current sense traces, which cut up the ground under R23. C18 doesn't seem to be doing anything, it must be very large indeed to do anything against a shunt resistor. Some series resistance in the Kelvin sense traces would be preferable. This is, I think, one of the oddities that LT isn't good about documenting: they prefer as simple an example as possible, even if it results in a poorer product -- see also the lack of gate resistors.
I followed your suggestions and managed to fit both gate drive traces on the top layer, and also I changed how the current sense traces are laid out:
Is this better?
I didn't fully understand what you meant with the "end to end" placement, how it would look like?
Also I will try to think a bit more of those series resistors at the sense traces.