LCD Theory
This is a graph of current draw (blue) (1mA per division) vs opacity (orange) for FSTN "shutter" / dimming LCD glasses. Time is 10ms per division. It's crazy how little current the liquid crystal draws to eventually transition to opaque vs. the capacitance of the medium. I think I have that right.
Main way to lessen the power draw is to switch as few columns/rows as possible. At some point, I could test lowering the inversion frequency, this would save a lot of power. Not sure if/when it's necessary though.
I think this display (dimming LCD glasses) is an FSTN too. Long time to decay.
On the one hand, you can make 1000s of the same pattern and shape of LCD into a mass-produced sunglasses lens that goes into clip ons or full sunglasses.
However, it's more convenient for glasses wearers to have a minimally weighted clip-on (magnetic?) that conforms to their existing glasses, and sunglasses wearers like having styles too.
Ideally, cutting a custom lens out of an LCD should be as easy as what they do for existing glasses lenses: load a blank into a machine that traces the glasses and does all the cutting for you. But...LCDs are not simple pucks of polycarbonate! They have traces and epoxy and liquid crystal that you don't want leaking out! Also, you want the shading to go all the way to the edge if possible.
So...I don't think I'll be able to get away from making the lens at the LCD factory. However, the easier I can make this process, the better and cheaper the result will be.
Is it possible to have only one ITO pattern that I re-use for all lens shapes? The portion around and including where the column ITO traces are would need to be able to be dark still for sunglasses mode and for blocking bright objects way outside the typical FOV. So place a wide "row" of ITO material underneath the column traces (it normally is non-conductive, bare glass, and covered by the device case) and the outside area and signal it like another row?
LCD module needs to ideally be the same for both sides, just flipped left to right to keep the cutout and rows the same. Columns flip, and zif connector is now on the bottom side of the LCD instead of the top.
A tradeoff of number of things:
- Number of backplanes / data lines to route and control (especially if driving manually)
- How big of a blob is needed to block a particular bright spot that is right at the intersection of, say, 4 pixels
- Smoothness of the display (shouldn't look too pixely)
- Density required in different regions of the lens. Want smaller for far distances vs. can be larger when close up (cars moving quickly) and sides and sky. However, exploiting this one requires mechanical IPD and vertical alignment so the eye is centered in the same area across users, so might not be able to take advantage of this much since IPD variation is a thing and glasses can have different shapes.
- How much light is let through (not a big issue) and refracted (see below) from ITO gaps. My 120X120 ronboe LCD is pretty ok, but could be better.
In my experimenting today (Jan 14 2024) based on the existing Ronboe LCD I have, I think even up to 2mm is ok visually.
I don't want high res TFT screen, as it introduces visual artifacts like repeated prism thing and sorta astigmatism. I want a low resolution TFT screen to address this, which starts at a million dollars easy. So...go with FSTN for now, with maybe active addressing scheme if it helps / is necessary (since low information content).
A related phenomena is the shape of the pixel. If you have a square pixel with horizontal and vertical lines for the "gaps" (I'm not entirely sure why these gaps are needed), then you will get astigmatism looking patterns. If you rotate them 45 degrees, then it looks like this:
. End result is we want to avoid the lines if possible, but if not, then we want them to be at diagonals so it obstructs as little of our vision as possible (and not horizontal lines). Hence a hexagon! (vertical and both diagonal lines, but no horizontal lines, which would come from vertical lines on the pixels). Plus it looks cool/unique and takes the least amount of "line" space, so there's technically less refraction too.
- June 2024 Nolan: I'm not sure I agree with horizontal lines being that bad. Vertical line sounds bad though. Which should I try to get rid of?
I should make a video about this...
Not much you can do about this for segmented displays apparently, other than keep it at the "two-polarizer" limit of 50% light loss and don't add additional filters if not needed.
To summarize from this excellent article: https://www.newvisiondisplay.com/types-of-lcd-technology/
Make sure you have good polarizers to start with. Should look like this when looking at a phone LED:
only_polarizers.mp4
TN will do a decent job for cheap things. However viewing angle sucks (maybe 40 degree cone) and the imperfect rotation (birefringence) of the liquid crystal material shows through and is based on light frequency. So you get a blue light shining through.
STN: Rotates the liquid crystal material more so it's twitchier / easier to turn on at middle voltages. That's about it.
DSTN (Double STN): One way to compensate for this is to add another liquid crystal layer that apparently is non-driven and flipped upside down from the first. So it sorta rotates the light backwards a little bit and somehow compensates for this imperfection without affecting the overall rotation of the light significantly. It's also more consistent across temperatures. Only problem is it's heavier (another piece of substrate) and more expensive than FSTN (below)?
- This is what welding helmets do!! I've been trying to understand why there is a 2nd LCD for a while now!
(Film STN) instead adds a film to do this birefringence correction. It's not quite perfect but cheaper! Not sure if I need a 2nd film or not and how that would affect transparency.
fstn.mp4
WVTN (Wide View Twisted Nematic) might be an option too if I want more grayscale without duty cycling the display?
(vertically aligned, vertical alignment):
- Only comes in default dark, which is:
- Not great if you don't have a robust electrical connection. But that'd be bad anyways
- Requires itself to be always on. But during the daytime, a number of people want sunglasses anyways. So might be the same situation for FSTN
- If power dies, that is bad. But you should be able to warn the user well ahead of that point. Not turn on if you don't have an hour of charge, etc.
- Looks kinda funny when off. Scary even.
- However, it is another order of magnitude of darkness compared to FSTN, which is great for sun. People would love that.
- Should be uniform darkness too, compared to weird FSTN bands that I am seeing.
- Not sure on pricing, power draw, or transparency relative to FSTN
Other things that are also important:
- Polarizer film: I have had some bad film that wasn't very dark. Gotta have good film that goes completely dark when twisted against each other with no color changes. Apparently you can flip the polarizer and get weird color changes.
- Duty cycle: Want to be as close to static mode (high duty cycle) as possible for maximum darkness. If you go less, then it starts turning off on you. This requires some special sauce if you want a high resolution display still, or you go for a TFT screen.
Great intro app note from NXP: AN3219.pdf
Great basic diagram from New Vision Display:
https://www.newvisiondisplay.com/multiplex-addressing/
Because my information density is pretty low (only a few big blocks), I can get a high resolution display with only a duty cycle for each "rectangle" I draw. So I need basically a static display driver with a large amount of outputs...
Do I want to hack in 1/2 voltage for ML1001? (alternate GND and VCC on backplane only chip to be VCC/2). Does not require level shifting data, as you can do changing of voltage after data is transferred. It adds some complexity. Maybe just separately route VCC and GND for the rows out to the PCB?
- This is apparently bad for the chip, says ChatGPT. Latch up?
Don't need to. Will need to play around with algorithms:
But it's ok to just use BAT below on larger pixel blocks to show text.
Binary_addressing_technique_with_duty_cycle_control_for_LCDs.pdf
- Somewhat readable paper! However didn't compare the contrast of his results to other methods. It should do much better than standard multiplex algorithms though.
- It basically runs through a shifted pseudo random binary sequence on the rows, and then each column decides whether it should be high or low depending on what is most beneficial for the bits it is trying to show. sum(row[j] xor col[i][j], j), and if it's greater than 50% it should be on, otherwise off.
- This is ok, but I don't need this full number of degrees of freedom, I think. So going with my own algorithm for now. https://github.com/nolanhergert/HeadlightBlocker/blob/main/passive_matrix_lcd_simulator.html
- Able to be driven by simple microcontroller, pretty simple algorithm. Rows can be the same waveforms!
- Is just on or off for each pixel, with maybe a grayscale overall. Fine with me actually.
Really just need shift registers in the end. Which the CH32V003 and similar are really cheap for that!
- Cheap microcontroller like CH32V003 or PY32F* are $.20 and work, but have 6uA off current best case. Could put onto flexible pcb connector.
- Just found a potentially better option though, the ML1001 and ML2002 are COG that are designed for running static LCDs! You just shift the data in, and if you provide your own clock <100Khz, it'll take only .1uA! And supports up to 120 pins! And is COG! And you can kill power externally when you don't want to drive stuff. Here is at least one Alibaba supplier that knows about it (Dongguan Team Source Display, and the one I've been talking with, Shenzhen Hexing Optoelectronic), and there are others in the US. https://www.alibaba.com/product-detail/Custom-Size-Lcd-Screen-7-segment_1600309167857.html
- Hopefully the cost isn't much higher. It's not! Around $1 and around the same setup cost, which is awesome!
- Other options from datasheet search: https://www.google.com/search?q=site%3Ahttps%3A%2F%2Fwww.orientdisplay.com%2Fwp-content%2Fuploads%2F+pdf+static. All others are square-ish, much bigger, and don't look to have gold bonding pad. Maybe all designed for COB and wire bonding instead of COG? KS0035, SBN1661G, HD66100F, ST7065C.
- Some low-backplane-count drivers allow you to pick static, 2,3,4 duty cycle, but you are still limited to 4 backplanes from what I can tell.
THIS ISN'T IMPORTANT RIGHT NOW!! But it's interesting... Want tiny chips that are shift registers that handle high ish voltages.
- CD4014 and CD4094 are sorta small (5x5mm) only 8 outputs though, $.30 each.
- QFN with 8 outputs: NPIC6C595BQ. Discontinued :-(
- QFN LED controllers if you're ok with multiplexing. In stock at digikey. IS31FL3748A-QFLS4-TR. https://www.lcsc.com/product-detail/LED-Drivers_RENESAS-IW7038-00-QFN4_C6577894.html