Modeling an F key mechanism

User avatar
DMA

28 May 2023, 19:45

Muirium wrote:
28 May 2023, 15:04
Model F is one of the loudest switch mechanisms ever made. Silencing that is quite a windmill to tilt, Don Quixote. ;)

IBM was explicitly targeting the Selectric feel and sound with their computer keyboards. The Beamsprings, from which these were derived, had a solenoid for a reason: to be as responsive and LOUD as an electric typewriter. Typists were used to the vibration and the racket on every keystroke. Model F dialled it down a bit, but not much. And actually, F is much more resonant.

The upside I suppose is that with all that noise, there’s a lot of scope for quiet improvement!
You do know that beamspring is actually quieter than the buckling spring? This kinda disproves the whole "and sound" part.
Vibration is a different story, but sound was definitely not a design parameter.

Anyhow, this is not about sound, this is about something manufacturable and more modern (turns out the current generation needs RGB lighting because it's not smart enough to remember the layout, as I recently read in the keyboard review)

User avatar
DMA

30 May 2023, 23:09

Turns out I modeled a flipper slightly wrong. This model is designed to lay flat on the plane. Model F bottom plate, however, has a 300mm bend radius, and the flipper is designed to lay flat against _that_.
The difference isn't that huge - 0.18mm over 21mm key pad height. But it's still there, so flat F flippers should be slightly different from the classic ones. Not to the point of not working, just producing somewhat weaker readout.
The curved flippers worked just fine on the wcass' flat F, after all.

User avatar
DMA

31 May 2023, 03:42

FreeCAD starts to show it's temper - like, it's EXTREMELY picky about point of rotation of an object.But still:
flat-vs-curved.png
flat-vs-curved.png (56.66 KiB) Viewed 10408 times
Turns out, there's ~0.25mm gap at the tip of the classic flipper when used on the flat PCB. Which is, like, SIGNIFICANT.
And yet it works - at least with CS.

Anyhow, looks like flat PCB will need special flipper molds, so good bye, flipper compatibility.

User avatar
taylorswiftttttt

31 May 2023, 07:39

...
Last edited by taylorswiftttttt on 28 Mar 2024, 07:30, edited 1 time in total.

User avatar
DMA

31 May 2023, 09:15

taylorswiftttttt wrote:
31 May 2023, 07:39
DMA wrote:
30 May 2023, 23:09
Model F bottom plate, however, has a 300mm bend radius,
I don't know what method you're using to measure the radius, but I would like to share my thoughts from when I attempted to figure out the backplate radius a couple months ago.
OH LOL. Yes, I approached it exactly like this, only because I absolutely detest math, I simply entered my measurements directly into FreeCAD as constraints to an arc. Now, FreeCAD doesn't let you specify "height" of an arc, but it has tangent constraint, so one can work around that.
o_curwa.png
o_curwa.png (18.77 KiB) Viewed 10374 times
So I have an XT (127x6.3mm = R323.17mm) and an F122 (192x17mm = R279.559mm).
And then I decided to re-read the "Remodeling the model M" post and lot_lizard was quoting R300 for bottom plate and R275 for the top one, right there.

Because R300 was conveniently located between R280 and R323, and guessing that IBM engineers weren't idiots (all critical measurements I took so far on barrels, flippers and the keys were a) metric and b) sensible (like, 0.5mm, not 0.625), R300 was chosen as "the intent". And indeed, chosing R300 led to things nicely aligned elsewhere - like continuing that R300 built from fixed points on the flipper's end and on the bottom crease nicely matching the pivot point within couple _micro_meters, and symmetric pressed/released flipper angle of 5.2 degrees instead of +5.5/-5.7.. I took those as a sign that R300 was, indeed, an original design decision.

In any case, over 21 mm, R260 is 0.2121mm dip, R280 is 0.1969, R300 is 0.1838, and R320 is 0.1723, R340 is 0.1621. So, R280 to R320 the difference per key is 0.0246mm, which is, like, comparable to the tolerances you'll get in practice. Considering that pivot point is actually 6mm from the key's boundary, actual differences are even less - 0.1081/0.1004/0.0938/0.0879/0.0827mm. So, 280vs320 is 0.0125mm difference where it matters. I won't be surprised at all if flipper actually flexes that much to conform to PCB - it would definitely explain at least some of the peak in the keypress waveform here:
Image
taylorswiftttttt wrote:
31 May 2023, 07:39
Paper is easy to print and cut at home but not very rigid.
I have some 300g/m2 paper laying around which begs to differ. That shit is so rigid my printer can't pick it up from the tray. If that doesn't satisfy you - go to home depot, buy some "ram board" (which is just 0.6mm cardboard), glue your printout to it (kudos for the calibration square btw!) and cut - you'll need either some VERY sturdy scissors or an x-acto knife to do that, but it will take quite some persuasion to flex _that_. But, as you can see above, the actual difference is minimal, not worth the effort gaining additional precision.
taylorswiftttttt wrote:
31 May 2023, 07:39
I haven't gotten around to doing this because I am lazy.
High-five! I'm only doing this all because I'm jobless since November and bored AF.

But let me tell you, 3D modeling allows for really, REALLY nice reality checks - like, you model things with sizes you've measured, and then try to put them together. All the inconsistencies, like, POP.

PS: Not sure who that Troy Fletcher guy is - but boy he's unhelpful AF. For no good reason, too :(

User avatar
Muirium
µ

31 May 2023, 20:12

How about plotting multiple radius curves on the same sheet of paper—or a harder medium—and comparing the backplate visually? The lines needn’t all be superimposed (their apex points can be separate), and could all be neatly labelled to make the result immediately apparent.

As for Mr. Fletcher: yes, it’s an objectively unhelpful response. But it’s also the friendly kind of thing you say when you don’t believe the spec, and haven’t the means to contact anyone who does. He’s likely being honest.

User avatar
DMA

31 May 2023, 20:37

Muirium wrote:
31 May 2023, 20:12
How about plotting multiple radius curves on the same sheet of paper—or a harder medium—and comparing the backplate visually? The lines needn’t all be superimposed (their apex points can be separate), and could all be neatly labelled to make the result immediately apparent.
Slight problem with that - F122 has 3 different heights. In theory they all have the same bend radius, so measuring at the ends should be OK. In practice - who knows? (yeah, one can cut out the one that should match using the edges, and verify in the middle, but where's the fun in that? Sharpening the saw is a significant part of the fun!)
Muirium wrote:
31 May 2023, 20:12
As for Mr. Fletcher: yes, it’s an objectively unhelpful response. But it’s also the friendly kind of thing you say when you don’t believe the spec, and haven’t the means to contact anyone who does. He’s likely being honest.
You are a very kind man, you know that?

User avatar
DMA

05 Jun 2023, 09:55

I accidentally acquired an XT, an AT and was loaned one of the new reproductions.
So now I have 2 XTs(older: "4870-10313"/oldest chip date 8411, newer: F3 5275, oldest chip 8538, controller has "0686" on it), AT (F4 4924/8427), F122 (F6 5218, controller desoldered and I have 4 of those, so..) and the chinese copy.
XT barrels are virtually indistinguishable - although surface polish of barrel underside is sliiiiightly different.
AT barrels are just like XT, but with barrel-side key.
F122 barrels is where it starts to get interesting: they have a fillet (rounding) on top, making the key interface ring (the part that contacts the key when bottomed out) noticeably thinner. It also has a more complex flipper retainer notch - probably to reinforce it a bit, but it's not like that part actually _needed_ any reinforcement..

So it really looks like F122 had spiffier molds (better presses maybe, capable of exquisite-r details?)
PXL_20230605_063316436.jpg
PXL_20230605_063316436.jpg (2.12 MiB) Viewed 10231 times
PXL_20230605_063658862.jpg
PXL_20230605_063658862.jpg (2.07 MiB) Viewed 10231 times
Chinese clone seems to be inspired by AT barrel design, going for
* seemingly thicker key interface ring - good!
* thicker back wall of the base (1.1 vs 1.0mm - for a moment, I thought that it's thinner side walls of the flipper chamber - but no) - good-ish? No real difference.
* certain simplification of the stem grippers access from below (likely to improve mold manufacturability) - no difference, nothing touches that part ever.
* simplifying the internal barrel chamfers a-la model M - not critical AT ALL, nothing contacts those parts either.
* tapering the outside of the barrel base a bit (likely to improve part removal - in fact, original IBM barrels seem to be a _real_ bitch to get off the mold - lots of LONG cylinders and "vertical" lines you can't taper without sacrificing stem retention stability) - no difference, may be rebound sound a bit.
* using harder plastic - good!
* but rougher surface (not sure if bad, actually - it's not _that_ rough)
Not bad overall, but somewhat scratchier/draggier vs originals. Especially if the key doesn't have an upper stabilizer (like unicomp keypad enter - but yeah, let's not even talk about unicomp :) )
* Oh, and significantly larger radius of the barrel base (yes, barrel base isn't bent upwards at some point - whole bottom of the barrel is molded as one large R300-ish arc with the axis of the barrel being the axis of that arc, and center point way above the barrel.
PXL_20230605_065139077.jpg
PXL_20230605_065139077.jpg (2.16 MiB) Viewed 10231 times
Flippers were more of a surprise: old XT had this dip in the middle (probably runner) and a larger ball under the spring support (but still not load-bearing). Newer XT and F122 flippers are indistinguishable. But AT.. AT was a total surprise: _very_ smooth plastic - almost slippery, _very_ simplistic forms - literally a model M flipper support married to a model F's flipper proper with crudely chopped off corners.

Chinese clone tries to reproduce F122 flipper, but it's thicker (1.0mm at the end, 1.4mm at the "base" vs 0.9/1.2) and has all the edges rounded (again, can be a mold manufacturability improvement) - like, the pivot on the original flippers (even on AT) is razor-sharp 135 degrees, but the clone adds R0.1-ish fillet at that corner. May be this greatly extends the life of the mold, who knows. Doesn't seem to have a huge effect on anything.

Next stop - comparing keycaps (not to unicomp, I promise. F122 will also be disqualified, as it has two-piece keys throughout for some strange reason).

User avatar
DMA

06 Jun 2023, 01:41

Keys are virtually identical. Good job!
PXL_20230605_224830427.jpg
PXL_20230605_224830427.jpg (2.44 MiB) Viewed 10210 times
The only visible difference is larger grippers (wider to simplify the mold and ~0.1-0.2mm higher). Makes keys harder to insert, but also holds them in place tighter. There's also a vertical parting line behind the gripper - hard to tell, but may be it widens just a little bit upwards - probably to facilitate easier part release from the mold.

Also, the new barrels are ~0.1mm shorter (probably because of the larder arc radius - the upper part is same size, but the barrel size along barrel centerline is 18.4(old)/18.3(new)mm.

I've recorded the sounds of middle 12 keys of XT(foam signifcantly deteriorated), AT(foam is intact-ish), F122 with no foam, F122 with taylorswifttt's foam, and the chinese clone at Jenga Studios (UMIK-1@48kSPS/24 bit). Will select AT as a "reference frame" and record swapping flippers (classic/slick/new) and barrels (AT/new). Not F122 because closing F122 subassembly with thick FSSK foam is PAIN (took me about half an hour to do - despite only 16 keys (12 in the middle and 4 corners for plate stabilization) being inserted. Oh, and also will record some model Ms, for completeness of the reference library :)

Events being recorded:
* normal keypress - pause - release
* slow pressing to actuation, slow release.
* fast tap.
Looks like fast tap is the loudest of them all with a huge lead - which might indicate that soft rubber o-rings on top of the stem might quiet down the F to the level of M, sacrificing the bottom-out feel (but seriously, the whole purpose of buckling spring is to prevent bottoming out, so that's not a big loss).

Due to extremely small event durations, spectrograms are virtually useless - so the comparison will likely be purely on a volume of a first peak plus duration of the decay to zero (if you chop off that initial snap, there are lots of interesting buzzing sounds. Will share later).
Also, only bare subassemblies are recorded to exclude case effects. The goal here is understanding the sound of a _switch_, not a whole device. The case design will be later used to enhance the result (for various values of "enhance"). I'm testing the violin _string_ here, not the whole violin - because it's the string which produces the sound, the rest of the violin "only" gives it a shape.

User avatar
pyrelink

06 Jun 2023, 02:51

As usual a very cool project DMA. Enjoyed reading through all of this.

User avatar
thefarside

06 Jun 2023, 03:12

Cool setup! Regarding the barrels, I tested some reproductions in my F107 and noticed their barrel keys are offset from the IBM originals, causing them to be shifted when installed with the IBM ones. I didn’t use them so I’m not sure if the offset is noticeable with keys installed. Below is a pic showing the repros on the bottom with the original IBM in the top rows:
IMG_5461.jpg
IMG_5461.jpg (1.95 MiB) Viewed 10191 times

Regarding the back plate radius, I discovered IBM has a corporate archives department and emailed them asking if they had info they could share on their buckling spring designs. I didn’t expect an answer but they did reply and it sounds like Lexmark got all the designs, and I’m guessing Unicomp ultimately has any remaining info. Below is the reply from IBM:
This is the IBM Corporate Archives responding to your request. Thank you for reaching out about IBM's keyboards! I've searched before in the collection for technical diagrams or blueprints for these keyboards and did not come up with anything. The current line of thinking is that when Lexmark acquired IBM's printer, typewriter, and keyboard lines in 1991 that all of that technical information went with them in the transfer. That said, I am unsure if Lexmark has an archive, but they might be able to assist further if they do.

As you are probably aware, there is some information available in the patents for these devices, such as the buckling spring patent which you can find here - https://patents.google.com/patent/US4118611 You might have luck searching for other patents such as this!

I wish that I was able to assist further with this. IBM keyboards are such cool part of IBM history!

Please let me know if you have any questions.

Sincerely,

Reference Desk
IBM Corporate Archives
2455 South Rd
Bldg 04-02 Room CSC12
Poughkeepsie, NY 12601
The patent refers to the switch which doesn’t appear to have a curve in the barrel or mention a curve. I searched for other patents by Richard Harris but there were a lot and no other keyboard related ones that I could find. I might post a question to the Dick Harris Q&A thread to see if he has any info on the curve of the back plate. Looking forward to your progress!

User avatar
DMA

06 Jun 2023, 03:22

thefarside wrote:
06 Jun 2023, 03:12
Sincerely,

Reference Desk
IBM Corporate Archives
A TALKING DESK!!!
My party and I walked into a tavern. The bar tender asked 'Why do you carry your swords inside here?' I looked him dead in the eyes and said, 'Mimics.' The bar tender laughed. I laughed. My party laughed. The table laughed. We killed the table. It was a good night.
Re: barrel key angle - this is a textbook tolerance management issue. On my AT plate, the original barrels rotate from ~-X to ~+X degrees clockwise.
Repro barrels seem to rotate from 0 to +2X, in some slots from +ε to +2X+ε. Your photo shows ~+2X angle (I'm too lazy to take and upload a photo right now). Can you try to rotate the barrel and determine the range boundaries? It might be just an unlucky turn of events and the barrels fell into the worst case position by default somehow :)

Upd: when installed into keyboard it ain't that bad - if you're bothered by the rotation, you can rotate keys into proper position and the foam will hold them. Could be worse - like, having to file off parts of the barrel key..

User avatar
DMA

06 Jun 2023, 05:30

..and now, just because I can, time to get SACRILEGIOUS!!! (yes, the sound will be recorded and analyzed. Also with nibs shaved off).
SACRILEGIOUS.jpg
SACRILEGIOUS.jpg (426.01 KiB) Viewed 10161 times

User avatar
Muirium
µ

06 Jun 2023, 12:51

Bet it sounds and feels AWSUM when sealed up and typed on. :D (Not that any keys would register but that's asking a bit much.)

User avatar
dcopellino

06 Jun 2023, 15:48

DMA wrote:
06 Jun 2023, 05:30
..and now, just because I can, time to get SACRILEGIOUS!!! (yes, the sound will be recorded and analyzed. Also with nibs shaved off).
SACRILEGIOUS.jpg
It'll be also sacrilegious, but what would happen if one shaved that central nib off and then covered the hammer surface with conductive paint? Might be called IBM model FM? or simply Model 'Marconi'.... :lol:

User avatar
DMA

06 Jun 2023, 16:58

dcopellino wrote:
06 Jun 2023, 15:48
DMA wrote:
06 Jun 2023, 05:30
..and now, just because I can, time to get SACRILEGIOUS!!! (yes, the sound will be recorded and analyzed. Also with nibs shaved off).
SACRILEGIOUS.jpg
It'll be also sacrilegious, but what would happen if one shaved that central nib off and then covered the hammer surface with conductive paint? Might be called IBM model FM? or simply Model 'Marconi'.... :lol:
I actually shaved the nib on 12 model M sliders yesterday evening, and can't get half of them to actuate in the new barrels. Today, between a court and onsite interview, I don't have much time, but I'll get there eventually and see what the hell is happening there.

User avatar
thefarside

07 Jun 2023, 03:57

I happen to be taking apart an XT and noticed the flippers are different. Mine are very shiny and there’s a little bump on the bottom end of the flipper. Here’s a pic comparing the am XT on the left to an original F107 in the middle and a reproduction on the right. The reflection gives you an idea how much smoother and shinier it is:
IMG_8211.jpeg
IMG_8211.jpeg (1.35 MiB) Viewed 9999 times
The little bump on the bottom middle of the flipper. I don’t recall this being on the other flippers:
IMG_8218.jpeg
IMG_8218.jpeg (511.97 KiB) Viewed 9999 times
Do you also have a little bump on the bottom of your XT flippers?

User avatar
DMA

07 Jun 2023, 07:12

thefarside wrote:
07 Jun 2023, 03:57
I happen to be taking apart an XT and noticed the flippers are different. Mine are very shiny and there’s a little bump on the bottom end of the flipper. Here’s a pic comparing the am XT on the left to an original F107 in the middle and a reproduction on the right. The reflection gives you an idea how much smoother and shinier it is:
So, your XT flipper is really, REALLY similar to my AT flipper in simplification of the form. The main difference is edge cutouts, but there's a smaller one: the spring holder seems to be a bit thicker, with round edge (on my AT, there's straight line near the spring holder, but doesn't look like the thickness is any different).
Can you make a photo of the side of that flipper? (so that spring is pointing up and at the left of the frame). I want to see where the mold parting line is. The best way to take that photo is on flat surface, slightly pressing on the spring so the flipper stands up. (Of course, you'll need about 3 hands to achieve that, but I'll greatly appreciate your effort. The trick there is not to press down on the spring too much - those things fly fast and they fly far!) Lightly pressing on the far flipper "leg" also works, but also not easy, because the other leg tries to lift up immediately.
thefarside wrote:
07 Jun 2023, 03:57
Do you also have a little bump on the bottom of your XT flippers?
No, I don't have any bumps on the paddle bottom - it's that large radius curve, flat on AT flippers, with a semi-circular dip where you have a bump. I'm starting to think that this dip is a direct consequence of the bump on yours. Could be that original idea was to align the flipper with it (may be the thought was it will kinda drift to the middle of the pad after several presses, because the PCB surface is ~35um lower there due to no copper?), could be something else, or I might be talking complete bullshit, who knows. 50 years passed.
But I remember a post here where somebody complained that he replaced a flipper and it didn't work - and capacitors are _really_ sensitive to the distance between plates (C = εA/d, and difference between 0.1mm and 0.01mm is factor of 10). So it might actually matter and be a defect. High-res photo of that XT's pad from an angle (to see the height variations) would be appreciated.

I've just checked a handful of those AT flippers (20-30) and the intent is definitely for the paddle bottom surface to be flat (well, technically, a section of a wall of a very large cylinder), but I found ~5 flippers _seemingly_ having a slight chamfer of the "far" edge, or being slightly bucked up at the corners. I've set those aside and will look at them at daytime with my 42x stereo microscope (42x magnification needs A LOT of light, but it definitely helps to see the flipper as 2 feet wide, with both eyes at once). The fact that those slick shiny flippers have beautiful meandering plastic flow pattern on them (which I'll try to photograph in better light) doesn't help distinguishing the surface features _at_all_. :) There are probably hi-res 3d scanners out there, but I don't have any favors with anybody who owns such a device and it's definitely beyond a budget of the unemployed man :D

User avatar
thefarside

08 Jun 2023, 01:02

Here's a side picture. Hopefully this is what you wanted but if not I’m happy to take another:
IMG_8259.jpeg
IMG_8259.jpeg (641.04 KiB) Viewed 9921 times

The pads on the PCB are raised and I think you’re right - the flipper bump on the bottom seems to fit perfectly in between the pads. I slid the flipper sideways across several pads and could feel the bump going in between the pad valley.

Here’s some pics of the PCB. I didn’t realize the controller was part of the PCB. That rules out replacing it (easily at least).
IMG_8261.jpeg
IMG_8261.jpeg (1.18 MiB) Viewed 9921 times
IMG_8262.jpeg
IMG_8262.jpeg (3.11 MiB) Viewed 9921 times
IMG_8264.jpeg
IMG_8264.jpeg (549.28 KiB) Viewed 9921 times

User avatar
DMA

08 Jun 2023, 03:31

thefarside wrote:
08 Jun 2023, 01:02
Here's a side picture. Hopefully this is what you wanted but if not I’m happy to take another:
IMG_8259.jpeg
poyfect! The angle looks the same to the rest of them - at least from what I can figure out with uninstrumented eye.
thefarside wrote:
08 Jun 2023, 01:02
The pads on the PCB are raised and I think you’re right - the flipper bump on the bottom seems to fit perfectly in between the pads. I slid the flipper sideways across several pads and could feel the bump going in between the pad valley.
There is, however, no evidence of any special processing of the PCB, and 1oz copper is 34 _micro_meters thick, that's like 0.034mm, 0.0014 inches. Are you sure you feel that bump because it's there, not because you expect it to be there? With differences that small our feelings are treacherous at best - and measurement instruments for micrometers range, while exist, are pretty rare (and quite expensive).
thefarside wrote:
08 Jun 2023, 01:02
I didn’t realize the controller was part of the PCB. That rules out replacing it (easily at least).
It's not actually that hard to replace the XT controller.
Look at the PCB where it leaves the metal enclosure. You'll see two parallel rows of vias there, connected pairwise by vertical tracks. If you cut those tracks, you can solder your new controller right into those vias (not it that makes much sense - you can get 5 new, guaranteed-not-corroded XT-size PCBs for ~$30 + ~$12 shipping nowadays). I actually did that in the early phases of my capsense experiments, when you had to fork over like $200 for 3 XT-sized PCBs.

User avatar
DMA

10 Jun 2023, 05:46

..So, I've recorded 51 tracks of 28 buckling spring inner assembly configurations over the last few days. Audacity file is 582MB, so not sharing yet - need to figure out how to chop it to manageable pieces first.

Details:
  • 12 keys in 3x4 matrix ("678", "yui", "hjk", "nm,"). Same keycaps (from AT) for all tests, same positions
  • 3 modes each:
    • normal press to the bottom, pause, fast (finger lifts faster than key) release
    • slow press till actuation, pause, slow (finger on key) release
    • fast tap
  • Configurations recorded:
    • XT - intact
    • XT - 12 keys only, all barrels
    • XT - 12 keys only, corner barrels for support
    • AT - 12 keys only, corner barrels
    • F122 - 12 keys, all flippers, no foam
    • F122 - 12 keys, corner barrels, foam from taylorswift's FSSK (boy, that was a nightmare to assemble..)
    • Same F122, after foam settled (3 days, I guess?)
    • Chinese - stock, before opening the assembly
    • M122 (F3 5605/15 MAY 87) - intact. It has backplate as thick and as yellow as F122!
    • AT&T model M Jan 90 - intact. Apparently, "typing feeling surpasses that of OG Model M", but I fail to find much difference
    • Full matrix of AT, F122 and Chinese barrels vs AT, XT, F122, M, and M(nub removed) flippers, on AT plate with 12 keys and corner barrels. That's 18 configurations.
I have '94 and '95 OG model Ms, but I can't get into them to pull the inner assembly out - because, y'know, THOSE DAMN BOLTS. They sound really, _really_ close to the AT&T version though, so I decided not to bother (unless I'll miraculously find that 7/16" screwdriver today or tomorrow).

Also, thanks Audacity for sticking recording level to "52%" by default (which is, due to log scale, is about -30dB) - which I discovered about halfway into this, so not going to re-record everything at 100% (will record interesting pieces after current data analysis). It turned out not that quiet - in fact, at 100% fast taps would probably clip anyway, so..

What's obvious so far: The flipper hitting the PCB isn't the biggest source of noise. Fast taps are 5x+ louder, and the loudest part is bottoming out. Slow press/release, to minimize all but the flipper sound, seems to result in key-up being ~3x louder than key-down. But those are very approximate observation - need to think how to batch-analyze this data better. It's, like, 1.5k events to process - a bit too much for manual analysis..

User avatar
DMA

12 Jun 2023, 07:19

..looks like I screwed up recording somehow - maybe didn't control for the recording levels, or something.
Everything sounds about the same, volume is about the same (except model M - it's different, way quieter. I think that's because both key travel is limited by flipper protrusions and flipper deformation on tap dissipates substantial amount of keypress energy. Anyway, it was too quiet to reliably automate event detection, so they aren't used).

One thing that astonishes me is how it's virtually impossible to understand if two adjacent clicks are same key or different keys, once you separate press sound from release sound. They are literally all the same (It may, however, be my shitty desktop speakers - however, it's hard to do so using the spectrogram, too - the heaviest hint is the quiet buzzing of spring visible on the spectrogram). They aren't, like, _completely_ the same - Chinese barrel seem to have more higher-frequency components but less lower-frequency, so is quieter overall (probably harder plastic) - but the difference seems superficial. When press+release are combined - it's kinda easier to tell them apart, but the variability between two recordings of the same key is comparable to variability between different keys, so it's only "kinda" :)

Attached are audacity projects for normal tap press and release, slow tap press and release, and fast tap (couldn't quickly write a python script to reliably split upstroke sound from downstroke, decided not to do that at all).
Clicks in those files are standardized - as in, every click is given 500ms window, and the peak is placed at exactly 100 ms from the start of that window.
"key 5" is "U" position, surrounded by barrels from all sides, key 6 is "I" position, and doesn't have anything to the right of it. (It seems to have more high frequency components - so may be I wasn't completely wrong with recordings).

b_<something> barrel summary - so, b_AT is AT, Chinese, F122 and XT flipper in the AT barrel.
f_<something> is a flipper summary - so, f_AT is AT flipper in all barrels.
Because everything was recorded twice - there's 2x more samples than barrels. They are grouped together - so, first and second samples are first flipper, next two are second, etc.

PS: and yeah, flipper's contribution to the loudness is negligible - main disturbance of the peace comes from the bottoming-out of the key, and this is likely the sound of a stem hitting the PCB. I'll try to remove some PCB material from under the stem legs - that AT's PCB is really badly corroded, so it can't really be used for anything anymore and will be disposed of anyway - and see how it affects the sound. It won't affect the key-up sound though - that one seems to be caused exclusively by flipper retaining legs hitting the barrel, and so it can't really be silenced (except probably by wrapping the barrel into mass-loaded vinyl).
Attachments
F-sounds.7z
(2.56 MiB) Downloaded 83 times

User avatar
DMA

12 Jun 2023, 23:40

So, I decided to try o-rings to silence the fast taps - but I didn't have suitable size o-rings laying around, so I used 2mm craft foam instead:
PXL_20230612_204300637.jpg
PXL_20230612_204300637.jpg (1.98 MiB) Viewed 9761 times
The result is interesting: without affecting key feel much (the only thing that changed is that bottoming out is now softer), the bottoming out is ~12dB quieter (4x dampening factor).

So when you're typing as you normally would, the bottoming-out is noticeably quieter than the key resetting, when you're MASHING the keys it becomes marginally louder than, and if you REALLY MASHING the key - my finger still hurts from that - it's comparable to the undampened mechanism (likely because the foam fully compresses and the stem hits the PCB).

Incidentally, it looks like default recording level was adequate, because I had to reduce recording level to -30dB in "MASHING" scenario, and had to even go to -36dB in the "REALLY MASHING" one to avoid clipping (on the undampened key - dampened key still had ~6dB headroom at -30dB)

I also modified some barrels with rebound silencers made from silicone sealant - they are now curing. Will see tomorrow how it pans out.
Attachments
craft_foam_damper.wav.7z
(165.34 KiB) Downloaded 81 times

User avatar
DMA

14 Jun 2023, 00:40

Only one out of 4 modified barrels materially quieted down the rebound (around 11 seconds in the attached wav) - but again, 12db. Which is not to sneeze at, especially on the already-quiet rebound. Note that this recording isn't loud taps - it's "press as slowly as possible to the bottom, then release quickly".

The loudest sound now is the ping of the spring, actually. So, it is now definitely quieter than the MX blue. I don't have any linear switches - they will likely be even quieter, but we're definitely in the office space acceptance range now.
Attachments
rebound_dampeners.wav.7z
(1.32 MiB) Downloaded 77 times

User avatar
DMA

15 Jun 2023, 05:43

So, "unbending" the flipper results in the following:
Thickness: 0.92 -> 0.85mm
Swing: 2.73 -> 1.72mm (12 degrees -> 7)

This should reduce the flipper landing noice materially.

To silence the flipper upstroke, a small silicone plug/bumper added:
rebound-damper.png
rebound-damper.png (84.16 KiB) Viewed 9640 times
Rebound flipper inside the barrel should be added, will take a swing on that one tomorrow.
The downstroke dampeners can be simply an o-ring (topre S-like), or silicone dampers can be added looking thru the PCB (will try to model tomorrow)

User avatar
DMA

16 Jun 2023, 06:01

bottom-out-damper.png
bottom-out-damper.png (14.15 KiB) Viewed 9607 times
Bottoming-out damper. Comes in contact with the stem 0.7mm above the barrel contact point.
stem-on-damper.png
stem-on-damper.png (28.48 KiB) Viewed 9607 times
If made of 30 Shore A silicone, this 0.7mm should be good for couple kg of force (which is kinda A LOT for a keypress). If more force is applied, the key will contact the barrel and noisiness will suddenly JUMP (although probably not as high as stem hitting the PCB - the key is naturally dampened by the finger pressing on it, while PCB readily conducts the sound to outside, where it can be heard). 50 Shore A will probably need, like, a pound to sag that 0.7mm.

Tests on a school eraser (supposedly 55 Shore A) show that bottom-out feel is not compromised at all - it's just a lot quieter.
The stem can be improved by widening the bottom part to increase the contact area - so even 30 Shore A will be good for inhumanly heavy hitters.

Still can't figure out the way to silence the stem upstroke - at the plus side there's not much energy to dissipate (the stem is pushed back by a spring, and nothing you can do to make it go faster), but on the minus side the stem gripper is ~0.6mm wide, and the cavity it's in is only ~1.1mm wide by ~4mm deep - plus I can only bore straight down, no fancy shapes (or the barrel won't come off the mold).
I can design some H-shaped hole, but how does one squeeze 0.5mm-thick silicone part in without tearing it apart?

PS: the round part on top of the stem can be removed with the bottom dampers - no risk of damaging the PCB or the stem - which will make it a bit nicer.
thicc-feet.png
thicc-feet.png (69.39 KiB) Viewed 9594 times
Also made feet thicc without affecting the key insertion/removal. The contact area is now ~2.5x0.8mm - so, ~2 mm2 per side, ~4 mm2 total. This should allow us to safely use of 30 Shore A silicone.

User avatar
DMA

16 Jun 2023, 10:27

..Also, now that we don't have anything huge on top of the stem, we can cut channels in the barrel for the top key stabilizers (corners) all the way through, and that will allow us to insert the stems from the bottom (so we can make grippers as large as we want).
This will need additional guides for the flipper to find it's final resting place on the bottom of the barrel - 1.5mm high, 1.2mm in diameter. The flipper will protrude higher than that - about 2mm - but the guides should restrict it enough to be nudged into position without breaking anything. Auto-alignment!
flipper-guides.png
flipper-guides.png (83.59 KiB) Viewed 9589 times
The empty space between bottom-out damper and the barrel is PCB.

PS: if we make holes in the damper a bit smaller, and guides less tapered, we can probably assemble things on the bottom plate: individually, springs only pull with 65 grams of force, and I believe we can create enough friction between the damper and the barrel to hold the barrel tight. In fact, first millimeter of spring compression probably contributes, like, 10-20 grams of force - easy to hold.

User avatar
DMA

20 Jun 2023, 07:18

Here's rough idea of how keys will be placed inside the keyboard.
perspective-view.png
perspective-view.png (96.58 KiB) Viewed 9501 times
The bolt holes are countersunk M2.5, screwed into threaded spacer from both sides. There are 29 of them, to prevent drop damage (just the plates will weigh about a kilogram) and add some rigidity (not that 1.5mm steel plates need any help there, but..).
The outline between plates will be made of some transclusent, but not transparent plastic (may be even polyethylene), with ambience LEDs placed between bolts plus in corners (that's 9x4 outline - K-Type has 13x5, but I don't really think it will make a material difference)

I don't think any extra bolt holes need to be added in the middle - but it's easy to add one between backslash and Delete, plus another one beneath the right shift.
Adding things inside the alpha block will be hard - not impossible, but a) they'll need to be no larger than M2, b) there must be a nut on the bottom side (no space for the spacer - only for the bolt) and с) cutouts in the barrels are pretty tricky to get right (every shifted row is shifted differently, and for some reason I don't want to be shoehorned into "OK, those holes are only for 'between R2 and R3'" - although now that I'm thinking of it, 'betwen R2 and R3' covers fullsize, TKL and 60% - so, like, 99.999% of configurations encountered in the field).

Question: what's the best slant angle? I don't think I have a preference (although I always end up with legs in raised position for some reason) - but there's a technical reason: bluetooth antenna needs to be at least 8mm from the nearest metal plane (and the plates are definitely metal planes), plus the module will have to be mounted perpendicular to the plate - so, needs at least 18mm of space under the plates. Plus one needs to hide the batteries somewhere. Now, 18mm is about 6.7 degrees slant - so, 7.5 degrees?

User avatar
DMA

22 Jun 2023, 06:48

I've modelled the outline of the keyboard and internal layer structure.
cake.png
cake.png (81 KiB) Viewed 9356 times
Plates will be 1.5mm 304 steel (or may be 904 - "swiss watch steel" sounds better (it also holds polish better, but not sure if that's even relevant), held together by 10 mm spacers. Inside, top to bottom - 0.5mm silicone insulator, 1.6mm LED PCB, 2mm foam (compressed from 2.5mm), barrel, 0.8mm sense card PCB, 1mm silicone damper.
This leaves 5mm between top plate and MX keycap skirt when pressed, 9.1mm at rest (keys shown pressed in this render, btw).
Total height with "cylindrical" profile MX keycaps - ~32mm, with modern "standard" profile (max. key height 12.5mm) - ~35mm, which is _really_ close to an average MX keyboard (MX switch is 8.3mm from plate to the tip of contacts, and 6mm from the plate to the bottom of keycaps at rest).

It should weigh about 1.5 kg (a bit less) in this configuration.
However, because we have extra 4mm of unused space between plate and the bottom of the keycaps, we can put either 4mm of mass-loaded vynil between LEDs PCB and the foam, or 4.5mm MLV (that's very close to thickness of standard 1.5lb/sf MLV) on top of the LED PCB - but I think it will be better to put it under the LED PCB, because LED cutouts will allow a lot of sound to escape. MLV-under-LED-PCB approach will allow for a very snug fit of the barrels (cutouts can be made even a bit smaller then barrels), allowing for superior sound deadening, especially higher-frequency components.
Now, standard 1.0lb/sf MLV is 0.125" or ~3.2mm thick. Remaining 0.8mm can be spent on thickening of the bottom silicone damper to 1.8mm - which should improve damping further.
MLV area will be ~0.5sf, holes will take 30% of that, so, about 160 grams of vinyl. So, may be 4 pounds total weight.

Weight is easy to add - additional 1mm of the bottom plate thickness is ~500 grams - but how much is heavy enough, really?

PS: wired keyboard can be made exactly as shown - there's enough internal space for the MCU, but not enough for the non-custom BLE antenna or battery. I'm trying to think of something modular - along the lines of "pluggable keyboard stand enabling BLE connectivity", but it's kinda hard (all the options that came to my mind so far are both expensive _and_ fragile).

User avatar
DMA

01 Jul 2023, 10:08

I _might_ have been overdoing the fasteners a bit.
fasteners.png
fasteners.png (93.9 KiB) Viewed 9144 times
39 pairs of M2.5 bolts so far. Probably don't need that much - but really don't want those bolt heads to be ripped out if keyboad is dropped onto the floor.

In other news - inventoried the keys on hand and looks like barrels have evolved over time (both XT and AT variety), while flippers haven't really changed that much.
I have "original", 2-, 3-, 4-, 5-, A and P XT barrel series, and B, G and P AT barrels today. Only the P series stands out - top of the barrel is chamfered (same chamfer size on both XT and AT barrels) - the rest don't have any _obvious_ differences, but might have smaller ones.
Flippers are actually used interchangeably across boards - A, B, C, D and G series (series are suffix, so full mark is "26A" or "5C") and don't seem to have any material differences. Notable exceptions - older XT flippers (with round pit on the bottom) are "prefix-A" series (full mark is "A8"), and simplified flippers from the AT are "PGC" series, separated by space (so, "PGC 27").

Post Reply

Return to “Workshop”