Bringing the IBM PC XT into the 21st Century

[DMA]

I draw them out as 2D DXF file and import them into DipTrace.

Wait, whole sense card? Does it count as one pin, or it's just "some copper fill" for diptrace?

This is all presuming that you are using a 0.8 mm thick 2 layer PCB with no vias. You improve signal detection by reducing the distance between the copper layers, so going to 0.6 or 0.4 mm PCB is helpful.

One can also electrically connect the top pad to the line, using a via. That will result in 3-4x better SNR. This also means the pad under the ex-floating pad is not needed. It will probably be beneficial to put the ground plane under both pads, but I didn't experiment with that.

4-layer PCB surely allows for awesome things like "frameless" layouts and can have an integrated ground plane (which enables whole new worlds in "rest of the keyboard" design - you can make everything from transparent acrylic, for example), but we don't yet have data on 4-layer PCB longevity in curved-plate designs.

Sem-related - my experiments with pad layouts had shown that only bottom 10mm of the pad are critical for model F - the rest can be freely used for any other purposes. You MUST keep top "dots" of the footprint - they lift the flipper's "legs" to ensure proper contact area when the flipper is down - but you can run traces thru those (ANY traces - rows, columns, ground, LEDs - whatever), because those dots are too far to interfere with sensing.

[wcass]

I draw them out as 2D DXF file and import them into DipTrace.

Wait, whole sense card? Does it count as one pin, or it's just "some copper fill" for diptrace?

It is multiple "imported object". The entire import does not add to pin count.

This is all presuming that you are using a 0.8 mm thick 2 layer PCB with no vias. You improve signal detection by reducing the distance between the copper layers, so going to 0.6 or 0.4 mm PCB is helpful.

One can also electrically connect the top pad to the line, using a via. That will result in 3-4x better SNR. This also means the pad under the ex-floating pad is not needed. It will probably be beneficial to put the ground plane under both pads, but I didn't experiment with that.

4-layer PCB surely allows for awesome things like "frameless" layouts and can have an integrated ground plane (which enables whole new worlds in "rest of the keyboard" design - you can make everything from transparent acrylic, for example), but we don't yet have data on 4-layer PCB longevity in curved-plate designs.

Curving a PCB puts the "inside" layer in compression and the "outside" layer in tension. This is made worse as the thickness increases. I think that vias might be susceptible to damage from these forces, so i try to keep the PCB thin and though hole count low.

Semi-related - my experiments with pad layouts had shown that only bottom 10mm of the pad are critical for model F - the rest can be freely used for any other purposes. You MUST keep top "dots" of the footprint - they lift the flipper's "legs" to ensure proper contact area when the flipper is down - but you can run traces thru those (ANY traces - rows, columns, ground, LEDs - whatever), because those dots are too far to interfere with sensing.

I plan to take advantage of this when i do a pad card for Unicomps upcoming Mini M. I will likely run a 6x16 matrix with the row traces going through the "dots" - giving wide clearance to the tenon holes.

[vivalarevolución]

I'm going to bump this thread in an attempt to find an answer for an XTant issue. When I took over a small run for these, I acquired a few XTs so I could make a few of these projects. Unfortunately, only one of my XTants works properly at the moment. All of the other ones have the same issue. I plug them in initially and they work fine. But after awhile, they start firing off random key presses or missing key presses. All but one of my XTants do this.

Has anybody else experienced this issue with their Model F conversions using the xwhatsit controller? Any idea what the problem is and if there is a solution?

[DMA]

define awhile. If it's like an hour - controller warms up and setpoints drift.
Use pandrew's firmware which supposedly auto-recalibrates to avoid that. Or use CS and simply forget about the issue forever (but no qmk, EVER)

[vivalarevolución]

define awhile. If it's like an hour - controller warms up and setpoints drift.
Use pandrew's firmware which supposedly auto-recalibrates to avoid that. Or use CS and simply forget about the issue forever (but no qmk, EVER)

That's probably about right. It's an hour or a couple hours after plug in. I haven't kept up on Model F firmware developments, so I am not aware of pandrew's. What do you mean by CS? The IBM CS 9000?

It also might be that my plate/PCB sandwich is too tight. They original build simply used binder clips. That doesn't really allow any sort of tightness adjustment. I just encountered a similar issue in my beam spring that was related to a back plate screwed on too tight.

[DMA]

CS is "CommonSense" - I made a replacement controller on an off the shelf prototyping kit 4-ish years ago. It has per-key thresholds and doesn't drift.

[vyquad]

How is the pcb coming along?