Revision as of 23:55, 3 February 2011

Led Floor

How to turn the 11 windows of the hacklab into a big led display ?

• what have we got to do it ?
  • Windows that are H189xW95.5 centimeters
  • 1000 red leds
  • loads of RJ45 and old IDE/SCSI/floppy buses

• what we need ?
  • a system to control each diode (or group of diodes) independantly
  • that is not too expensive

Question: do we need on/off for each led, or more advanced modes (eg PWM for various intensities).

• on/off:
  • is simpler as regard to schematics
  • is simpler as regard to driving code (led bits per image, obviously)
  • isn't as cool
  • can still, to some extend, allow some variable intensities

PWM "natural" ability is prefered, but choice will be made depending on total cost of solution.

Solution #1

use a TLC5940NT that can control about 16 LED each

• datasheet: http://datasheet.octopart.com/TLC5940NT-Texas-Instruments-datasheet-153017.pdf
• schematic: http://webhome.csc.uvic.ca/~mcheng/samples/led_wheel/images/hack_schematic.jpg
• O(price) = $180 for the 64*TLC5940NT, so we'll need also several protoboard to be cut to bind them on.

Naturally leads to a PWM version

Solution #2

use the snootlab's i2c led driver solution that can control 16 LED each

• 64*i2c led driver = 9.95*64 = 636.80
• 1*i2c shield = 17.95
• 1 ardunio = 20
• O(price) = 674.75 euros

Solution #3

• use shift registers... ?

Solution #4

use two Peggy 2 board :

• 2* http://evilmadscience.com/tinykitlist/157
• O(price) = 95*2 = $180 --or far less as you won't need the giant PCB. They have a "parts kit" that includes the 328 and the IC's for $15. you'd still need LED's and transistors and other support components. --cw

looking at the peggyLE schematics, they are not using several of the outputs, it looks like a 32x30 grid is possible with the IC's included, and by adding another IC in place of the buttons 32x45 is possible. --cw => they use two 4 to 16 demux for highside driving. They therefore *have* to keep one output not connected, to allow an "off" state for each of these chips. One solution would be to use two such chips, plus two generic I/O, leading to a (quite ugly) 10 to 32 demux functionality. -- cq

Solution #5

• Check with Electrolab, where another similar project is ongoing :-)

=> actually, the (current) project there is about a 8x8x8 led cube. Which is a bit different.

Solution #6

So, the goal is to achieve a 1024 led screen, with, if possible, PWM on each led, for a total price <150€ (connectors, leds not included) ?

My proposal would be to mix #1 (eg use TLC5940 chips, for their ability to PWM their outputs) and #4 (actually, only reuse the 4 to 16 demuxes (74HC154) idea, with two additionnal GPIO from the uC to achieve a 5 to 32 demux). That is :

• two TLC5940
  • drive each row in PWM mode
  • can be daisy chained, and they take 8 I/O (or possibly less than that)
  • cost about 3€ each at www.mouser.fr
• two 74HC154
  • drive each column (one after each other)
  • cost less than 1€ each
• some power stage:
  • worst case is all 32 rows ON with 100mA each. So each column power driver should be able to sustain a 3.2 A continuous current.
  • the evilmadscientist schematic proposes 1.5A pnp transistors ;)
  • it'd be a better idea to get some (any) tougher pnp transistors (maybe few € total)
• This requires about 18 I/O for control
  • it is out of reach of an Arduino (true only when all the features of the TLC5940 are used. There are solutions to use an Arduino anyway. Note that I consider UNO and similar boards, not mega ones - which are too expensive)
  • it is quite doable with a Teensy (http://www.pjrc.com/teensy/ about $30, similar in functionalities to an arduino otherwise)
• some power input is required
  • I suggest a computer PSU (reuse a standard connector to get 5v on the circuit)
• some connectors are required
  • I suggest using standard RJ45 connectors (which stand 1.5A if I remember well. Meh...)
• some passives are required. As usual.

Total cost: O{100€ ?). Sounds good to me :)

issues/open questions

• will the TLC be well adapted to drive each column only 1/16th of the time ? That is, still work well when asked to somehow refresh data at (total refresh rate)*(number of columns)
  • according to datasheet, the max required clock speed is 4096*(refresh frequency). Even with a *16 factor, it isn't out of reach of a simple uC
  • will the leds give enough light when pulsed at max 1/16th of the time at 100mA ?
    • my guess would be: yes
• how to make the board ?
  • it can be done on a standard protoboard (quite small, I would even say)
    • Check if standard PC power connectors fit on these. If not, use a wire + standard connector.
    • Check if there are thru hole RJ45 connectors available. Looks like there are some (quite cheap, O{10€} for a 8 port version) but pins won't fit on a protoboard.
• if one want to really use 12 bit resolution for each of the 1024 leds, using a 8bit uC with 8k of RAM is... probably not the best choice
  • I guess it'd be okay to downgrade the resolution to few bits per led, eg never drive LSBs for each led and use only 4bits resolution.

=> I cannot see any big flaw in this solution. Does anyone have a better view here ? --cq

Solution #7

Similar to #6, but instead of TL+demux, use a uC which has 64 I/O that can be dedicated to the led matrix.

• I cannot think of any chip that would be as cheap and as "simple"
• software layers would be easier
• additionnal power/interfaces would be required to drive the rows

Solution #8

• 42 ?