Capsense modeling and real life observations.
Posted: 08 Apr 2023, 01:09
This post is here should anybody try to understand how capsense actually works and how sense card looks like, capacitance-wise (and that I'll be able to later find what I did and what results I've got). You're welcome to ask questions or ask me to measure something specific (if I still have the rig set up - which should be until at least June-July 2023)
So, after measuring the encased ESSK, looks like actual key inside the keyboard looks like this (CircuitJS model), electrically:
Row and column capacitances to ground seem to be directly proportional to the PCB trace length. Mutual capacitance between row and column seems to be DOMINATED by row-to-ground and column-to-ground capacitances (connected in series). There _might_ be electrically-long-line shenanigans involved, but their effect shouldn't be _that_ bad, unless you do something stupid like overdriving the line or placing current-limiting resistors way too far from the driver circuitry, in which case ringing might kill you.
However, CircuitJS shows 50mV released/380mV pressed (62/455mV with "scope" switch off) with these parameters, and the only thing that affects output voltage is that 84pF capacitance. To bring things to observable in reality 20/150mV, I had to increase the receiving end capacitance to 260pF.
It might be my capacitance meter - I've generated a 5.00V square wave by KitProg and fed it to DUT via 999 ohm resistor, scope in parallel to the DUT triggered at 3.16V (which is 63.2% of 5.00V) and measured offset between zero crossing and trigger - 1ns = 1pF that way. However, I've noticed couple interesting things: physically larger ground electrode gives larger capacitance reading, so reversing polarity from "normal" (i.e. connecting generator ground to row line and "+" to the sense card's ground results in larger capacitance, and I'm not sure which one to consider "true".) I have a Leader 745 LCR, but it's a bit broken (I've got it already broken, but hey, it was free
). Once I'll fix it, I'll repeat some measurements and see if keyboard assembly is, indeed, a _polarized_ capacitor. BUT row-to-col capacitance is close enough to seem true (i.e. calculated capacitance of row-to-gnd + col-to-gnd in series contributes 90% of the measured capacitance), so the data is usable enough to be, well, _usable_.
So, the capacitor formed by larger pads of ESSK and a flipper over them is 10-ish pF (although it might be as low as 5 pF - R8C12 is an example. This hints at the presense of the transmission line effects, although this may be the fact that insulation under the board is somewhat flexible - CommonSense readouts float a bit depending on the strength of the keypress).
F122's pressed key only adds 2pF. That's because model F's key is essentially two capacitors in series - one is via flipper, another is between board sides. FSSK is an improvement, btw - about 4pF, because PCB is 33% thinner: 0.6mm vs F122's 0.9.
It seems like larger parasitic capacitances are actually beneficial (which is somehow contrary to what CircuitJS shows): FSSK key capacitance of bare PCB is 2pF (mutual R1C1 capacitance changes from 43pF to 45pF, 4.65%), while encased key capacitance is 4pF (68->72pF, 5.88%).
As usual, Skeletor will be back with more disturbing news - but not soon, because spring break, going to drive around The Great Plains and California with family for a week or so.
So, after measuring the encased ESSK, looks like actual key inside the keyboard looks like this (CircuitJS model), electrically:
- essk-key-real.png (8.97 KiB) Viewed 2164 times
However, CircuitJS shows 50mV released/380mV pressed (62/455mV with "scope" switch off) with these parameters, and the only thing that affects output voltage is that 84pF capacitance. To bring things to observable in reality 20/150mV, I had to increase the receiving end capacitance to 260pF.
It might be my capacitance meter - I've generated a 5.00V square wave by KitProg and fed it to DUT via 999 ohm resistor, scope in parallel to the DUT triggered at 3.16V (which is 63.2% of 5.00V) and measured offset between zero crossing and trigger - 1ns = 1pF that way. However, I've noticed couple interesting things: physically larger ground electrode gives larger capacitance reading, so reversing polarity from "normal" (i.e. connecting generator ground to row line and "+" to the sense card's ground results in larger capacitance, and I'm not sure which one to consider "true".) I have a Leader 745 LCR, but it's a bit broken (I've got it already broken, but hey, it was free
). Once I'll fix it, I'll repeat some measurements and see if keyboard assembly is, indeed, a _polarized_ capacitor. BUT row-to-col capacitance is close enough to seem true (i.e. calculated capacitance of row-to-gnd + col-to-gnd in series contributes 90% of the measured capacitance), so the data is usable enough to be, well, _usable_.So, the capacitor formed by larger pads of ESSK and a flipper over them is 10-ish pF (although it might be as low as 5 pF - R8C12 is an example. This hints at the presense of the transmission line effects, although this may be the fact that insulation under the board is somewhat flexible - CommonSense readouts float a bit depending on the strength of the keypress).
F122's pressed key only adds 2pF. That's because model F's key is essentially two capacitors in series - one is via flipper, another is between board sides. FSSK is an improvement, btw - about 4pF, because PCB is 33% thinner: 0.6mm vs F122's 0.9.
It seems like larger parasitic capacitances are actually beneficial (which is somehow contrary to what CircuitJS shows): FSSK key capacitance of bare PCB is 2pF (mutual R1C1 capacitance changes from 43pF to 45pF, 4.65%), while encased key capacitance is 4pF (68->72pF, 5.88%).
As usual, Skeletor will be back with more disturbing news - but not soon, because spring break, going to drive around The Great Plains and California with family for a week or so.