University: Massachusetts Institute of Technology
Team Members: Benjamin Shaya, Charles Franklin, Diony Rosa
TI Parts Used:
- TLC5947 (24 channel PWM constant current sink LED driver)
- LVC8T245 (Level Translator)
- AM26LS34C (4x RS422 Transmitter)
- BeagleBoard xM
- A 48 x 21 array of LED's, measuring 8 feet by 4 feet (that's 1008 pixels, each consisting of 3 RGB LED's, or just over 9000 dies!)
- Music synchronization via a BeagleBoard
The LED Array was fabricated as 126 long, narrow PCB's, each with a single TLC5947 driving 8 RGB LED pixels. Every board had SPI through, for easy daisy-chaining. Communication with the array is via RS-422, enabling the controller to be far from the array without sacrificing signal integrity.
Boards were assembled by hand by MIT students and incoming freshmen; for many it was an introduction to soldering and electronics!
We tested every board by hand too!
Our support structure doubled as a power distribution system; aluminum angle (literally "power rails") formed an energized frame that boards screwed in to.
Unfortunately, after getting 6 rows onto the array, the buffer chips on every board blew out... We quickly swapped them for new ones and repopulated, but another blowout occurred. Our best guess is the inductance of the signal paths causes a voltage transient on one of the buffers, which died short-circuit and shunted the fault voltage onto the 5V power supply rails, burning out the rest of the buffers.
Some quick thinking led us to wire around the buffers, cutting them out of the circuit (TI's TLC5947's proved to be much more robust than the ST Microelectronics buffers, and didn't burn out the same way).
Ultimately, we ran out of time (and willpower) to finish the array (one week of 14 hour work days was really all we could handle, and classes were starting...) We managed to get six rows working, and sort of synced to music: https://www.youtube.com/watch?v=zrmBGjx-mbU
We're not exactly thrilled with how it ended up, but we're by no means done. The next revision of board will have easily bypassed buffers (and we'll be sure to use TI buffers this time), TVS diodes on all signal lines, IDC cables for harnessing, and 12V regulators on board. These next-gen boards will be assembled by a board house (turns out untrained freshman labor just isn't as reliable as an automated SMT line). We couldn't get the BeagleBoard to work because the LVC8T245's were unhappy at 1.8V (they didn't translate reliably, but worked just fine at 2.5V and above), so we'll likely switch to a BeagleBone, using the same translation hardware.
More Photos >>