Help with Blinky-Light Box

[Maat Mons]

For a school project, I'm making... let me check what they called it... ah yes, it's a "Programmable and Portable Light Plate Apparatus (LPA) for Optogenetics and Photobiology."  Don't know why I have trouble remembering that.  

 

Anyway, it's an array of lights that sit below a standard 96-well plate:

 

Because, the professor assures me, exposing cells to various wavelengths of light in various patterns is useful, somehow.  

 

Anyway, each of those wells needs an RGB LED, a dedicated green LED in addition to the one that's already there as part of the RGB one, for some reason, and a UVA LED.  That's the minimum requirements, but they'd really prefer if each well has a UVC LED too.  The upshot is that I need to pack 288 or 384 LEDs into a 72 mm x 108 mm area.  And control them all individually, of course.  

 

We've designed a small test-board with just one driver chip.  

 

We're still waiting on the parts.  But if the bloody thing works, our final board will mostly just be a 2x6 array of those daisy-chained together.  

 

One regard in which the full-scale version will differ from the test board is that, on the test board, we left out one capacitor that the chip we're working with said should be there.  We put in the one that the chip said was needed for "decoupling," even though none of us have taken the coursework to understand why that's needed.  But it said to add another one to keep the "voltage ripple" caused by the LEDs turning on and off below... I think it said 5%?  

 

Can any of you guys help me figure out how to compute the size of the capacitor we'll need?  

[Maat Mons]

Thanks for the reply!  

 

The decision to use TLC5957 chips was semi-arbitrary.  We thought using drivers with a large number of outputs each would help keep the circuit board simple, since we'd need fewer chips.  The 5957s have 48 outputs each, which means we only need 12 chips.  

 

Though, looking at how tiny the contacts on the TLC5957s are, and contemplating having to solder them, I've begun to wonder if drivers with 24 outputs each might not have been a better idea.  That would increase the number of chips needed to 24.  And that seems like a lot of ICs to pack into a 7.2 cm x 10.8 cm space.  But I think it might be doable.  

 

The text from the TLC5957 datasheet that lead me to believe we should have additional capacitors was the following:

Quote

Depending on panel size, several electrolytic capacitors must be placed on board equally distributed to get a well regulated LED supply voltage (VLED). VLED voltage ripple should be less than 5% of its nominal value.

 

I'm afraid the WS2812B-2020s won't work.  If I'm reading things correctly, those are 5 mm x 5 mm.  Each well in a plate is about 6 mm in diameter.  We're aiming to have all the LEDs be roughly under the wells.  So 2020 chips would take up all the space allotted with no room for the other LEDs we need.  

 

If you look at this image, the teal circles are the areas we're looking to keep the LEDs inside of.  

 

And those "big" LEDs that hog half the space in each circle are 1206s.  

[mariushm]

For your IC, 0.1uF / 100nF ceramic as close as possible to the Vcc and ground pins ... and I'd probably add on each tiny board a 100uF electrolytic capacitor where power comes into the board.. (I'd go with 16v rating or higher, even though you're only dealing with 5v) .. can be more, let's say 150 or 220uF but would probably be pointless

 

--

 

So you actually need 4-5 leds per cell... and it looks like you have 8x12 cells or wells ... so you actually need 480  leds or something like that.

 

I would consider using something like ISSI IS31FL3742 which is a led matrix driver that can control up to 180 leds using 30 channels, so you could use 5-6 channels for each cell.

Here's a link :  https://www.digikey.com/en/products/detail/issi-integrated-silicon-solution-inc/IS31FL3742-QFLS4-TR/8563764

Example from schematic :

 

So 30 channels ... 5 leds per cell ... so each chip can do 6 x 6 cells  ... 4  chips to get your 12 x 12 cells and just not use 2 rows of cells.  or you could rearrange it as 4 x 8 for a total of 32 cells for each chip, but would make routing the traces a bit harder. 

 

edit :  or you may choose to parallel 2 channels or even 3 channels should your UV led require more current than the led driver's current per channel (in the case of this ISSI part the maximum current is 38ma)

 

edit  : or you could add that extra uv led and then use 6 channels for each cell, ending up with maximum 5x6 leds

 

4 layer boards aren't that much more expensive, jlpcb will do them for cheap, but the board you want is very doable even in 2 layers, you may just have to use a bunch of jumper links on the other side or 0 ohm resistors on the led side to jump over traces. You can have the bottom side strictly for the leds and the other side for the led driver chip and the traces from the matrix to go back to the led driver chip. The whole top surface can also act as heatsink for the led driver and you can use heat conductive pads if needed to pass heat onto a top cover or something if you want.

You could also have these driver chips on separate tiny boards and just have a 36 pin (30 channels + 6 sinks) (or as many pins as you need to use if you end up using only 4 x 8 = 32 cells)  0.1" header come out of the led board. This way you could make the whole board with leds surface mount and make a stencil to apply solder paste then put the leds on the pcb and use heat from below to heat up the solder paste and solder the leds.

Then you just manually solder the headers and then plug the driver boards into the headers.

 

 

As for the actual leds, it could be a bit more expensive but you can get 0603 or 0805 or maybe even smaller rgb leds or individual leds... you should be able to pack a lot of them with the matrix arrangement.

 

TME.eu had some leds in small packages, for example (you can change language on page from the top right corner):

 

0805 (2mm x 1.25mm x 0.8mm) : https://www.tme.eu/ro/details/ostb0805c1e-a-0.8t/diode-led-smd-colorate/optosupply/

0606 (1.6mm x 1.6mm x 0.7mm) : https://www.tme.eu/ro/details/rf-w3s198ts-a41/diode-led-smd-colorate/refond/

 

 

 

 

 

 

[Maat Mons]

Thanks for the reply!  And sorry for the late response.  

 

For that driver you mentioned, would it be rapidly switching between 6 different groups of LCDs?  Would that have any effect on the brightness?  

 

Is there a good way to keep the bigger capacitors from sticking out of the board too far?  Or will we just kind of have to plan for the spots where the capacitors stick out to be where the battery isn't, in order to keep the total package thin?  

[mariushm]

By design, the chip will multiplex, it will only show one column of leds (see picture) at a time. it turns SW1 for a short period, then SW2, then SW3, and repeats the loop. You'll have to check the datasheet to figure out the switching time. There's around 2 microseconds of dead time where all 6 are off, but that probably can't be noticed by human eyes.

 

Each of the 30 channels has a maximum configurable current limit, you set it through a resistor as can be seen in the picture (the maximum current is 38 mA per channel, but you'd probably want to stay closer to 25mA if you won't adjust brightness through i2c commands afterwards (because current limit for each SW pin is around 800mA, so 800mA / 30 = 26.66mA). You can also send commands to adjust brightness of each individual led... it has 256 pwm levels and 256 dc current levels so the brightness is quite configurable. 

It also supports configuring fewer sources (sw1...sw6) to flip faster between columns. 

 

Not sure how bright your panel has to be, some UV leds may be too low brightness at 38mA (but maybe you could have separate driver just for UV and use these for rgb and green then) ... but it may be worth grabbing one or a few such chips to experiment with and see. 

 

The chip can work with i2c up to 1 mhz (but 400kHz is probably plenty and more compatible with microcontrollers) and you can configure the address (2 bits of the address so up to 4 such chips per i2c connection), so having 4 such chips on a single i2c is super easy, and your microcontroller or whatever you end up using could talk super fast to the chips and set individual leds on and off. 

 

You would have the benefit of using just at least 4 such chips to get all your leds working, instead of 24 or however many tlc drivers you would need otherwise, and consider how much data you'd have to shift each time if all those 24 drivers are chained. 

 

If you use RGB, green, uva and uvc that means 6 leds ... so with a single chip you could split the 30 current sinks / channels into 5 groups of 6 leds  or if you get rid of the uvc you have 6 groups of 5 leds... so you can have all leds from 5 or 6 cells light at a time, then sw1 stops sourcing current and moves on to sw2 and the next group of 5 or 6 cells lights up and so on. 

 

As for the capacitors question, you could have cutouts in the pcb so that electrolytic capacitors sit horizontally / flat inside the pcb cutout, making the whole profile lower height. 

You could also go for more expensive tantalum-polymer capacitors ... for these ISL chips I'd probably consider something in the 100uF-470uF (and i'd try for 10v or more, but 6.3v would probably be fine) near the PVCC pins : https://www.digikey.com/short/7j11qmm0

They're more expensive, but again, you'd only be using one per chip and now you have 4 or 8 chips or however many you'd end up with , instead of 24 or who knows how many led drivers you'd use otherwise. 

For quick testing, these adapter boards would probably work well with the chips and they're cheap, you get 3 boards for around 6$ :: https://www.digikey.com/en/products/detail/adafruit-industries-llc/1377/5629437

 

It says it's QFN-48 7mm , the chips are QFN-48 6x6 mm , so they're probably compatible but double check), just leaving longer traces for easier soldering.

 

For actual end product, you'll want to follow the instructions in the datasheet...put Rext very close to the chip, place those 0.1uF and 1uF very close to the PVcc pins , thicker traces for the SWn pins and so on..

 

 

 

[akio123008]

On 4/27/2021 at 4:56 AM, Maat Mons said:

Is there a good way to keep the bigger capacitors from sticking out of the board too far?  Or will we just kind of have to plan for the spots where the capacitors stick out to be where the battery isn't, in order to keep the total package thin?

You could always run long wires to an external location where there's more space and have the capacitors sit there. For example along one of the sides of the contraption.

[mariushm]

8 minutes ago, akio123008 said:

You could always run long wires to an external location where there's more space and have the capacitors sit there. For example along one of the sides of the contraption.

Not a good idea. You need capacitors for bulk energy besides decoupling capacitors (the 0.1uF and 1uF ceramics) close to the ICs.

 

Long wires have their own inductance and resistance, so they can affect things when there's a lot of fluctuation in the current... which can happen if the light box switches ultraviolet on and off, or changes from a single color to full white, or does various light effects which cause fluctuation in the overall current consumption. 

Don't ignore the current consumption ... if we go with 20mA per led, we have red, green , blue in a rgb led , separate green and at least a ultraviolet led... so 5 leds x 8 x 12 cels = 480 leds  x 20mA = 9.6 amps of current if all leds are on at maximum brightness (limited to 20mA each)

You'll have voltage drop on the wires going to the panel...using multiple wires is recommended.. and the inductance in the wires can cause oscillations and fluctuations in the voltage so best to have the capacitor by the chip to "kill" / reduce those oscillations and act as a buffer for sudden changes in power consumption. The decoupling capacitors (the ceramic 0.1 and 1uF) are there for high frequency stuff they'll help a bit but not enough when it comes to imperfect "transmission lines", the power wires.

 

Look at regular atx power supplies and how a lot of them hide capacitors right by the 24pin connector or the 8 pin pci-e connector ... if the location wasn't important, they'd keep the capacitors inside the power supply where it's cheaper to solder them with automated machines, instead of manually modifying cables to hide capacitors inside them. 

 

I'd use a regular ATX power supply as it's cheap and can provide the current (up to 15-20A on 5v) and it's fairly regulated.  I'd use either a 24 pin molex connector (taking only the 5v and ground wires from it and maybe 5v standby  and a momentary button for on/off  (ps_on , short pin to ground to turn on, short for a few seconds to turn psu off) or a couple of basic hdd / molex connectors (two separate chains of molex), so that there's two pairs of 5v + ground between the psu and the panel. 

 

Alternatively, a 16-20v 40-65w (or more) laptop adapter with a standard 2.1/2.5 mm barrel jack  and have a few dc-dc converter boards convert the higher voltage down to 5v or even better, around 4v..4.5v, for less losses in the led driver chips  - they need at least voltage drop on led which should be around 3.2v for blue  plus around 0.5v but for safety I'd use at least 4v...  boards that do 2-3A of current on the output are maybe 1-2$ on eBay. 

 

[akio123008]

2 hours ago, mariushm said:

Long wires have their own inductance and resistance, so they can affect things when there's a lot of fluctuation in the current..

It's not going to matter enough. I'd just run some wires and be done with it.

 

[Maat Mons]

Since we're talking about the power supply, I should probably mention, this thing has to run on batteries.  

 

Actually, it's supposed to run for 48 hours on batteries.  But I'm pretty sure that's not going to happen.  At least, not with all the lights continuously on for that whole time.  

 

Right now, we're running all the LEDs at 5 mA, except the UV LEDs, which run at 10 mA.  Our current LEDs are:
150044M155260 (80mcd Red, 180mcd Green, 50mcd Blue @ 5mA)
IN-S63AT5G (350mcd @ 5mA)
SM1206UV-395-IL (0.65mW @ 20 mA) [but we'll be running it at 10mA, so 0.375mW?]

 

Even though those lights strike me as feeble, if we actually have to run all of them continuously for 48 hours, that 128,240 mAhs.  That's not a small battery.  

[mariushm]

No, 48 hours is not realistic. You'd need a bunch of expensive lead acid batteries or lithium battery banks... Would probably make more sense to spend 100..150$ on a harbor freight generator and recharge the batteries within a few hours to last another day.... or spend $100-200 on a 50-100w foldable solar panel that could power and top up the batteries throughout the day.

 

Maybe you could extend the battery life a bit by covering the top panel with some solar cells and hope they give you a watt or so of energy, for example : https://www.digikey.com/en/products/detail/anysolar-ltd/SM531K12L/9990471

But your product probably has to run inside a room. in the dark, not just outside in the sun.

 

As for leds, they seem ok choices ... not sure if it was a great choice as the plain leds have wide view angles at 120 degrees or 140 degrees for the rgb led ... but not sure about how that lens works, or if you want the beam of light concentrated or you'd rather have it spread wide angle.

A more narrow angle would obviously produce more mcd per mA

 

Also... wonder if you have considered choosing a green led with slightly different nm peak ... seems both your greens are in the same 520..530 nm range.

 

[mariushm]

25 minutes ago, James Evens said:

Assuming all of the LEDs consume 4 W (10-15 mA per cap) with 48 hours you need 192 Wh which means you would need roughly a "battery volume" of 500 mL. Given your footprint of 108x74mm under ideal condition the battery would be 65mm in height. You won't get close to this so it will be more like 10 cm or multiple thinner units.

Under this condition you could do the 48 hours claim.

 

He has a 8 x 12 cells, each cell has  1 RGB , 1 green, 1 uv   ... Assuming 5mA for each r/g/b and the green standalone and 10mA for the uv , we're looking at 30mA per cell. Even without the standalone green, we're at 25mA per cell.

So that's 8 x 12 x 25mA = 2400mA ... let's say we go with 3.6v (as linear led drivers are cheaper) ... that's 3.6v x 2.4a = 8.64 watts

 

So almost double...

 

[Maat Mons]

We were considering running the whole thing at 5 Volts.  This was mostly because it needs a way to program various patterns of lights.  And connecting it to a computer via USB seems a lot easier than add wifi or Bluetooth capability, or trying to put buttons/a screen on the thing and have people program it that way.  So we were figuring on having a USB port, and making that also the way we charge the battery, and we said to ourselves "USB is 5 Volts, why not make all the internals 5 Volts too?"  

 

But sure, I guess we could be real engineers and build based on efficiency instead of convenience.  Down-converting DC voltage isn't that hard anyway.  

 

I might suggest to the group that we keep the actual internal battery small, and allow it to be connected to external battery banks for extended life.  The internal battery would keep it going when you're moving things around the lab, or when you're switching power banks.  And then you could pick how chonky a battery pack you connect it to based on how long you're going to leave it.  Or maybe just plug it into the wall with a good USB power adapter.  Something that can do at least 3.25 Amps, because that's what I calculate all the LEDs and ICs together can pull in the worst case.  

 

Man, this project feels like a mess sometimes.  This is an actual thing one of the professors wants built for the lab.  I signed on for this project because I had to do something, and this sounded easy.  Only after the decision was locked in did I find out that a team had tried to build this same thing last year, and failed.  Also, I have it on good authority that none of the other groups have to work with anything nearly as tiny and fiddle as our stuff.  

 

On the subject of recovering light that doesn't get absorbed by the specimens, I'd rather try to reflect it back for another chance at absorption.  I was figuring each of the depressions for the wells could be white, or maybe even shiny.  But we're not supposed to let light bleed through from one to the next too much.  Maybe a white coating over black plastic?  I'm a little worried about making the lid bounce light back.  There's going to be a bit of a gap, and we don't want light that bounces off the lid to wind up in a different well than it started in.  Maybe the inside of the lid could be a retroreflector?  Send that light right back where it came from.  

[cachethrash]

If you didn't need UV LEDs, I wonder if an LCD screen could be used instead of an array of LEDs, and then you'd just need to program which parts of the screen are which color.