Next GH-brewed KB design - 'The Light' (open for discussion)

[rknize]

I've had plenty of embedded experience with serial busses.  Everything from UART to MIPI.  Go with SPI if you can.

[mkawa]

yes, i did look into SPI and TWI/I2C after my initial fail. since the switch matrix micro is going to do most if not all of the talking, we could do a simple 3-pin SPI setup that slaves an LED controller to the teensy's AVR. there's a teensy (standard) library that then places a reasonable send-message abstraction over the SPI bus. soarer's "straightforward, but not trivial" seems to sum up the total engineering effort needed pretty well, since we would still have to figure out how to cleanly lay out the teensy components and LED driver or micro on the board without getting in the way of anything. not to mention the need to split the firmware and find an appropriate LED driver (most of the ones i was looking at today were either overkill or underkill)

one thing i did notice is that all of the LED driver ICs that are easily slave-able would increase our current handling by about as much as we want (to 500mA or so). so, maybe a reasonable option just for that.

[Soarer]

Well I still don't think it's necessary to pump the max current, since those who want a bright backlight can simply buy the brighter LEDs. Honestly, I think would be more than bright enough even at 6% to 10% duty cycle, and I don't feel a need to prove that point. Not to mention that proving it would be tricky, what with the difficulties of scientifically photographing them or whatever!

Someone should look into what driver ICs are available though. I haven't spent much time looking, but I haven't seen any single chip solution to driving an 8x16 array with independant level control yet. There are cheap display drivers for 8x8 that do not have level control. There are RGB drivers that have level control, but need an extra shift register as well (e.g. DM163 on arduino colour shield). An 8 column RGB driver would be fine driving a 24 column mono matrix, of course.

I think the real benefit we might get from a driver chip isn't simply brightness, but brightness control... for example that DM163 only only has 6 bits per pixel, but it also has an overall brightness register of 8 bits. That has real practical value in making something that works nicely in both daylight and total darkness. But OTOH, who really wants backlighting in daylight?

Talking to a driver chip would be much less than half the work of talking to another teensy - only one end to write, no thinking about the protocol, demo code already existing, etc, etc.

[The_Ed]

It's always better to go overkill than underkill. Go overkill with the LED controller.

As to where to put everything - A separate daughter board with headers will plug into the back. It will contain all of the controlling circuitry. A slightly taller case will be required, but it would be the easiest solution.

[Soarer]

It's always better to go overkill than underkill. Go overkill with the LED controller.

As to where to put everything - A separate daughter board with headers will plug into the back. It will contain all of the controlling circuitry. A slightly taller case will be required, but it would be the easiest solution.

More than you need is a waste. Design is about compromising; you can't please 100% of people with one design, yet you have to meet 100% of the requirements for most people. Having an extra circuit board and a larger case is a big compromise, and an absolute no-no if this has to fit into an existing case (which is likely).

[Djuzuh]

PrinsValium and litster got married and are on honeymoon.
djuzah, mkawa and the_ed have huddled into an irritation of trolls.
The prototype is still absent.
There was a run on suncream at the store because of rumours of 100,000 lumen keyboards.
The network has confirmed a second series, so the first won't have any real ending...

Why does everybody keep saying Djuzah?

[Soarer]

Why does everybody keep saying Djuzah?

Heh, sorry. Probably cos I mentally pronounce it 'jizza'.

[Djuzuh]

I think somebody needs to prototype the led thing, maybe hack something quick with arduino, to know if the current you intend to send is enough.

We can't just assume, and wait for the first prototype to realize it's not enough/ it's enough .

[Soarer]

I think somebody needs to prototype the led thing, maybe hack something quick with arduino, to know if the current you intend to send is enough.

We can't just assume, and wait for the first prototype to realize it's not enough/ it's enough .

All of the asuming I'm seeing is from those who a) say it won't be bright enough without a driver, b) want someone _else_ to build prototypes, and c) don't have thought-out suggestions, just spanners to throw in.

I know from experience that getting an ultra bright LED to just glow nicely, rather than blindingly, takes a huge reduction in current and/or duty cycle. The LEDs on my teensied breadboard are using 10k resistors - which works out at only 0.25mA!! Also, doubling the current from say 20mA to 40mA won't give a subjective doubling of brightness.

[hazeluff]

All of the asuming I'm seeing is from those who a) say it won't be bright enough without a driver, b) want someone _else_ to build prototypes, and c) don't have thought-out suggestions, just spanners to throw in.

I know from experience that getting an ultra bright LED to just glow nicely, rather than blindingly, takes a huge reduction in current and/or duty cycle. The LEDs on my teensied breadboard are using 10k resistors - which works out at only 0.25mA!! Also, doubling the current from say 20mA to 40mA won't give a subjective doubling of brightness.

Prototyping that thing is one of the first things I want to do once I move back home in a few days, but I still need to gather components and even a new breadboard (silly me, I shipped it using ship freight...).

I agree, that the brightness is not the issue.

I just want to verify that the original idea is not viable AT ALL, before we start going off with the other designs. Its so much simpler and elegant (tho could just fail miserably). Occam's razor ya know.

[mkawa]

i just want to say that design brainstorming is far from throwing a spanner in. naturally it all comes down to prototyping prototyping and prototyping, but it's good to have some ideas or optimizations in mind in case we find that first prototypes are deficient in some dimension.

as for brightness, i have no idea how much current we'll need, just that we may need to increase current handling capability. remember, there's going to be a keycap on top of the LED...

[Djuzuh]

All of the asuming I'm seeing is from those who a) say it won't be bright enough without a driver, b) want someone _else_ to build prototypes, and c) don't have thought-out suggestions, just spanners to throw in.

I know from experience that getting an ultra bright LED to just glow nicely, rather than blindingly, takes a huge reduction in current and/or duty cycle. The LEDs on my teensied breadboard are using 10k resistors - which works out at only 0.25mA!! Also, doubling the current from say 20mA to 40mA won't give a subjective doubling of brightness.

Hm, you are suggesting ultra bright leds, how much more do those cost against regular leds? because we need 100 of them for each board .

[hazeluff]

You're going to need blinding LEDs to get them to go through opaque keys, no doubt. But earlier on I asked melissa about any translucent keys with legends and she said they can make a PBT that would let some light through for a glow effect.Tho I think we we should get it to a level where the keys will glow from underneath the keys possibly if we use normal opaque keys.

Otherwise the aim is probably that they light up enough for use with proper transparent legended keys.

[Djuzuh]

QWER keys now makes keycaps like on Deck boards. With custom legends too !

He can even make the opposite of deck keycaps, the whole keycap glows but not the legend.

[Soarer]

i just want to say that design brainstorming is far from throwing a spanner in. naturally it all comes down to prototyping prototyping and prototyping, but it's good to have some ideas or optimizations in mind in case we find that first prototypes are deficient in some dimension.

as for brightness, i have no idea how much current we'll need, just that we may need to increase current handling capability. remember, there's going to be a keycap on top of the LED...

Sorry, but if you have 'no idea', then you're just throwing a spanner in.

It's not hard to get an idea: old-school IBM lock LEDs are maybe 1mcd @ 25mA. 'High-brightness' types are about 50mcd. Now we have 5000+ mcd LEDs easily available. We will PWM them by a factor of between 8 and 16 - so they'll still be 300 to 600 mcd. It's just simple math, that you could easily do by yourself. Is that enough? It's certainly enough to a useful. Will it be enough for everyone? Possibly not, but if it's fine for 90% of people then I'd say it was 'good enough'.

Brainstorming is bull****. What's useful is work - research, thinking and such. If we all do a bit of it this will be fruitful and fun. Instead of just saying that we might need a driver, why not go look for some that might be suitable? You might find one that would be ideal and we'd be daft not to use it. Just google "led matrix driver" and see what you find.

And finally, I've asked more than once what keycaps might be used...

[rknize]

It's gonna be tough to make the whole cap glow from just one LED on one side.  Would be cool if it could be done, though.

[Soarer]

Hm, you are suggesting ultra bright leds, how much more do those cost against regular leds? because we need 100 of them for each board .

Why don't you go and look!!!!

[rknize]

it might be a good idea to consider which side of the cap should be lit, as it will affect layout.  For example, it makes sense to have the LED on top for the mail part of the KB, but I've always thought that the num row should have the LED on the bottom.

I wish cherry would make a switch with a completely transparent top half.

[Djuzuh]

Why don't you go and look!!!!

You seemed to have a pretty good idea about what you exactly want.

[Soarer]

You seemed to have a pretty good idea about what you exactly want.

Sure, I have an idea of what WE'd want, but I haven't shopped around. Maybe $10 on the 100, at a wild guess.

You do realize that 'regular' LEDs are orders of magnitude less bright, and the driver will only increase current by 2 or 3 times, right?

[bpiphany]

If there are locations for extra transistors and perhaps serial resistors for them, the resistor pads can be shorted, and the base/gate of the transistor can be shorted to the output. This allows for transistors for those who want them, or driving directly from the uC. The transistors need to be placed somewhere though, but there should be plenty of room.. =)

[Djuzuh]

Sure, I have an idea of what WE'd want, but I haven't shopped around. Maybe $10 on the 100, at a wild guess.

You do realize that 'regular' LEDs are orders of magnitude less bright, and the driver will only increase current by 2 or 3 times, right?

oh ok. I just remembered some sort of 1/2$ per led in my head.

[Soarer]

If there are locations for extra transistors and perhaps serial resistors for them, the resistor pads can be shorted, and the base/gate of the transistor can be shorted to the output. This allows for transistors for those who want them, or driving directly from the uC. The transistors need to be placed somewhere though, but there should be plenty of room.. =)

The only problem is that the transistors invert the signal, so the code would have to know to invert the PWM values... should have plenty of pins spare for an input from a jumper though :-)

[mkawa]

yes, i spent quite a bit of time thinking about LED efficiency, brightness claims and light emission measurement a couple of years ago when i was working on a bike light. i've also spent some time recently playing around with my backlit keyboards with both translucent and non-translucent caps. what i know is that the upper limit of this design is probably to shine through a non-translucent ABS keycap. it's a nice effect and allows the board to be compatible with a user's existing (eg, doubleshot) keycaps.

brainstorming is not "bull****" in any sense of the word, but i respect your opinion and where you're coming from and the least productive thing in this thread would be to get into a drawn out argument about that. what i will say is that i took a look at the atmel MSL series drivers yesterday and none of them seem appropriate for us. as i said initially there is a dox or phantom related thread where someone with much more design experience than i have mentioned a bunch of MCUs that looked pretty good, although I don't know what their SPI compatibility story was, and i can't for the life of me find the posts again. if i can find some time later tonight i'll see if i can page through the appropriate parts of the digikey and mouser catalogs and see if anything looks interesting.

The only problem is that the transistors invert the signal, so the code would have to know to invert the PWM values... should have plenty of pins spare for an input from a jumper though :-)

i'm personally fine with having multiple loadable versions of the firmware or some flags in the config and build scripts

[Soarer]

I always thought brainstorming was a time of voicing optimistic, out-there ideas, which don't get discounted until facts get in the way! It's when you bring in the marketing types, so they can feel they had a say in the design process* ;-)

Anyway, interesting to hear of your experiments, that's far more productive :-) Although I fear you'd need a laser to shine through DS Cherry keycaps! Any chance of finding out more about what's in the 'boards you've been playing with? Current, duty cycle, type of LED...



* actually, you do it to build confidence in junior engineers and hear their ideas, since they're now not the dumbest people in the room!

[mkawa]

in that case, consider me a junior engineer; it's pretty accurate, as i have zero experience in the embedded or practical engineering space, and thanks for suffering through my ideas

anyway, the bike light community has done some measurements and basically everything on the datasheets for "superbright leds" of the standard through-hole type is made up crap that has nothing to do with an actual sample's performance in a design. basically, if you narrow the reflector angle enough, and you do enough cherry picking, and "measurement error optimism", you can say a lot of completely useless things about LEDs. what this means, unfortunately, is that all useful bike lights use cherry-picked ultra-high-efficiency SMT LEDs with custom lenses and reflectors, large heatsinks, and heavy duty drivers, none of which is relevant to our project.

as for the backlit keyboards i've used, i've played around with both deck and vortex boards and my belief is that not only do we not need a laser to shine through SP doubleshots, but we only need a little more power than they're getting. put a cherry replica red-esc on a deck 82 and it will give a pretty nice glow. otoh, put the stock vortex keys on a pure, and it will barely shine through the translucent bits :/. so anyway, let's carry on, and we can start arguing again when we get some prototypes up and running. deal? well, i'm excited at least.

also, i'll have a lot more time after mid-july, and a work-related excuse to spend time looking at arduino/teensy code.

[hazeluff]

The only thing i want to correct is that transisters dont nessesarily invert signals. It depends on how its setup.

[Soarer]

The only thing i want to correct is that transisters dont nessesarily invert signals. It depends on how its setup.

I didn't say they always inverted ;-)

But in this application, we want a switching mode, so that imposes a few restrictions, doesn't it?

[hazeluff]

I didn't say they always inverted ;-)

But in this application, we want a switching mode, so that imposes a few restrictions, doesn't it?

Ummm The problem is we need two, one for column and one for row. And it's quite impractical to make them both work of active high signals. one of them is going to be inverted.

Morning me raging at stuff i skimmed over. = )

[Soarer]

Ummm The problem is we need two, one for column and one for row. And it's quite impractical to make them both work of active high signals. one of them is going to be inverted.

Morning me raging at stuff i skimmed over. = )

Definitely crossed wires here!

N-channel FET would be used to connect the row or column to ground when it's input is high - an inversion.
P-channel FET would be used to connect the row or column to supply when it's input is low - an inversion.

If we take out the P-channel FETs, we'd need to drive the row or column active high instead :-)

[hazeluff]

Definitely crossed wires here!

N-channel FET would be used to connect the row or column to ground when it's input is high - an inversion.
P-channel FET would be used to connect the row or column to supply when it's input is low - an inversion.

If we take out the P-channel FETs, we'd need to drive the row or column active high instead :-)

In terms of voltages you are right...I was thinking in current.

nFET conducts when gate is high. But pFET will conduct when gate is low. (Based on some diagrams a bit back).

[Soarer]

In terms of voltages you are right...I was thinking in current.

nFET conducts when gate is high. But pFET will conduct when gate is low. (Based on some diagrams a bit back).

Yes. Well... the gate has to be high or low relative to the source... and the current flow is in the opposite direction for n-channel vs p-channel, since the source voltage is opposite... :-p

[Soarer]

So... where were we?!

On IRC, __red__ mentioned some fancy-pants PWM LED driver that looked tidy...

[mkawa]

do you have the model # written down anywhere?

[Soarer]

Just found it again: AS1130.

[bpiphany]

=D Look at that. All you need in a $4 package... Does it require any form of cooling (I didn't read all the 40 pages...) Or were those 30mA the total current? Also, too bad the scroll function was designed for 5 and not 6 rows =P

[Soarer]

Total current would be 340mA, I think. See 'Operating Supply Current' on page 4. Or 360mA (Digit Drive Sink Current, Max).

[mkawa]

that is extremely slick. great find, red

[alaricljs]

Well the good news is that their site is out of date and the SSOP package is available through DigiKey.  Bad news is that the price is $9.40 in single units, although the 25 piece break is not too bad at $7.81.

Intriguing matrix layout... but I'm no EE and haven't a clue the sorts of things you can do

[alaricljs]

Okey doke, so I have 3x AS1130's on the way and I'm laying out a breakout... who's got the software skills and wants to tackle this?

[spolia optima]

I think you guys are onto something. here are my first couple ideas:




Cherry 1800 compact keyboard
-They're plentiful, easy to find. Models built in the 80's and 90's are rock solid.
-CHEAP.
-just about any G80-18* or G81-18* model will do. cherry's been making them since the 80's, and they haven't changed the basic design one iota.
-nice layout once you get the hang of it.


Cherry G86-52400 series
+plentiful and easy to find, despite having launched in the late aughts.
+simple, modern design (it's a box)
+SOLID build quality. These keyboards are pretty beefy... Unlike

-more expensive
-some might criticize the aesthetics of the thing.

[fartq]

Okey doke, so I have 3x AS1130's on the way and I'm laying out a breakout... who's got the software skills and wants to tackle this?

i can't program the thing unless i have a sample (can't simulate). soarer's the more capable though. maybe get a couple protoboards wires up and send one to him?

[alaricljs]

Okey, full LED test board schematic:



Working on a chopped down 96 LED schematic so I can get cheap-ass test boards.

96 LED test board layout:

[alaricljs]

So I paid OSH Park $4.20 to get 3 copies of this:





Okey, went ahead and got the LED boards from OSH too since the quality is so nice and the price wasn't that much less for 10 of them from Seeed ($20 vs $14)  I don't need 10, that's for sure.






edit: 2 hours later and the fab is in possession of the gerbers, wow!

[alaricljs]

SMD - no problem... except I can't see straight for an hour after I do this.  Not the prettiest I've done but it's just a breakout board.

That chunk of metal is the cut-off lead from a 3mm LED:

[absyrd]

Nice work. Why can't you see? From staring at something so small? Pick your head up once in a while and pick a spot on the horizon.

[damorgue]

Nice, now go for 3 ground paths with a potentiometer each and use them to control RGB leds...

Pwetty please?

[alaricljs]

3 ground paths?  You haven't looked at the data sheet have you?  And no pot needed, the driver has color balance settings.

[damorgue]

Which data sheet? The ones I am thinking of have one combined positive lead (anode) and three negative ones (cathodes). Did you intend to control all three of them via pwm? If so, I have missed some progress in this thread.

[alaricljs]

The as1130 LED driver datasheet (that chip up there^^) I don't intend on RGB at all, but the chip is all set to handle it.  And if you look at the LED board there is technically no ground path (at least not full time).