28-01-2016, 11:16 AM
I decided on either 6 or 8 layers (it's along time ago) for the Quartz routers. The basic problem is to turn a 3D problem of input and output buses into a 2D one with all the IO on one edge of the PCB. So it wasn't the usual method of a couple of ground planes, a power plane and a couple of signal planes. There wa ground planing on every layer to shield critical bits from other critical bits. I specified a LOT of vias to link the ground planes. My PCB layut guy said he'd never seen so many in his life and it almost broke his CAD system.
There was a fair bit of 75R microstrip tracking. One unexpected problem was track resistance in the microstrips. The design brief had said no preset pots so all gain was determined by precision resistors. WIth a bit of "adjust on design" to give a final trim. Definitely no AOT. Because of differing track lengths we had noticeably different output channel gains. I think I compensated with different "adjust on design" resistors for some channels.
The knottiest gain problem was when Quartz and I were cosnsitently measuring the gain about 0.05dB different. This used up more or less all the design tolerance. A tricky investigation with both our network analysers showed that it was due to poor return loss on one port of their analyser. A 6dB pad sorted it. The analysers do an automated "thru" calibration so the extra loss didn't matter.
There was a fair bit of 75R microstrip tracking. One unexpected problem was track resistance in the microstrips. The design brief had said no preset pots so all gain was determined by precision resistors. WIth a bit of "adjust on design" to give a final trim. Definitely no AOT. Because of differing track lengths we had noticeably different output channel gains. I think I compensated with different "adjust on design" resistors for some channels.
The knottiest gain problem was when Quartz and I were cosnsitently measuring the gain about 0.05dB different. This used up more or less all the design tolerance. A tricky investigation with both our network analysers showed that it was due to poor return loss on one port of their analyser. A 6dB pad sorted it. The analysers do an automated "thru" calibration so the extra loss didn't matter.
www.borinsky.co.uk Jeffrey Borinsky www.becg.tv
