mesa 7i73 4x8 keypad problems on KEY8 input

Hello,
I'm helping some friends to use a 7i73 connected to 7i96s for a CNC.
7i73 is using Firmware 14 (read from linuxcnc) with a custom adapter board to make it easier to connect a panel with switches and mpg. 7i73 is in 4x8 mode with analog in and no lcd (jumpers set accordingly & sserial=10000...).
The adapter board doesn't do much besides adapting from IDC Connectors to screw terminals.
In general the system works fine and the keypad also works with linuxcnc - buuuut all keys connected to KEY8(in) don't work. Other keypad-switches are working fine.
Cabling was done with 0,5mm² and it is quite a mess. My first guess was the big and long cables would have too much capacitance and the 10k pullups couldn't pull the line back up in time.
Bridging with a short cable made it work, so card and general set-up seems to be ok.
Replacing the 10k pullup resistors with 1k did NOT do the trick!
After some Scope measurements it actually looks like the KEY8 inputs will be measured far to soon after enabling e.g. the KEY4 output line. The outputs will be (open drain) low for ~30µs. In the scope the difference in the signals will be very little and only in the first few µs.
So the conclusion is that the KEY8 inputs will be sampled early after setting the row outputs. With my long cables and many connections it looks like this will make it unreliable to get the right switch input.
Shorter cabling is not really an option. And in case the signals are really sampled that early I'd rather like to somehow get that fixed in the firmware.
I counter-checked with another 7i73 board from another batch and did see the same result.

Attached pictures show the testing setup. Scope screenshots are from the video:
Scope cabling is not much better than the key cabling but it should still show the most important...
Yellow is KEY4 (open drain output with added pullup to see measurement moment without button pressed).
Blue is KEY8 input with pullup (not sure if original 10k or already changed to 1k).
The video is showing the difference of:
- a bad signal, "shorted" via switch that will not be seen in linuxcnc meter
- a "good" signal seen in linuxcnc meter, done by bridged signals on adapter board (very end of video)

Can anybody confirm this?
Could it be a problem in Firmware V14 that only KEY8 is sampled early?
What would be the solution for this problem?

Thanks for your help!
Alex

I'll take a look. Seems like the scan sequence needs to be changed
so the sample is done at the beginning of the next cycle so the full
period is available for settling time.

thank you for the quick response!
I am not sure what you mean by beginning of the next cycle.
If I remember right, the samling of rows is done with 3.7kHz scan frequency and each row is measured for ~30µs.
From my understanding the inputs should be sampled at the end of each cycle so here ~25µs after the output is set to low. So right before it is set back to hi-z.

7I73 version 16 firmware

Increases the time between setting the row outputs
and sampling the column inputs( for longer cables etc)

Firmware:

 

SSerial update script: