Mysterious LED problem...could use some help

[dcel]

Could you give us an idea how you are using the matrix? I mean there are 432 LEDs to address and there is only 6 pfets on your board.

Chris

[KTP]

Could you give us an idea how you are using the matrix? I mean there are 432 LEDs to address and there is only 6 pfets on your board.

Chris

Sure.  There are two banks of RGB LEDs, each bank with 6 rows and 12 columns (36 columns actually since each LED has 3 elements).  The 36 columns are driven by three Allegro 16 channel low side drivers using only 12 of the 16 channels per chip.  The 6 rows in each bank are driven on the high side by the 6 pfets, and the same pfets drive the 6 rows in the other bank (so pfet number 1 drives row 1 bank 1 and row 1 bank 2, pfet number 2 drives row 2 bank 1 and row 2 bank 2, etc.)

The 2nd bank also has three Allegro 16 channel chips driving the 36 columns.

What this all means is that at any single point in time, only 24 RGB LEDs can be on.  There is no flicker because the banks are being switched at 540 hz (for a 90 frames per second update rate).

This method reduced the amount of Allegro chips needed from 36 chips down to 6 chips....but has caused me a world of pain in debugging this problem 

On a positive note (sort of), I did find that adding 6 10K pulldowns to the pfet drains and 72 10K pullups to the Allegro nfet drains totally eliminated the problem...all LEDs stay dark.  But what a pain...78 more components...

[tnt]

Well there are resistor arrays of 16 resistors at once ...

[dcel]

On a positive note (sort of), I did find that adding 6 10K pulldowns to the pfet drains and 72 10K pullups to the Allegro nfet drains totally eliminated the problem...all LEDs stay dark.  But what a pain...78 more components...

Occam's Razor - a simpler explanation is better than a more complex one.

     I tried to prevent you from pulling most, if not all of your hair out chasing the problem. Lesson learned I hope, I had the same lesson too. In the digital realm, thou shalt use pull up/down resistors and never ever leave anything floating. One would figure that on the output of a fet that really isn't necessary, but its still digital, just at a higher voltage and/or current. And thanks to codeboy2k, I also know about the capacitances that are involved.

     I just would like to say that is a very cool project, I like the PacMan demo in the YT video. I assume that once all the kinks are ironed out, these will be available forsale?

Chris

On edit:

     On a positive note ( not sort of), you did solve the problem! Even though you have to re-spin the board and have the additional components to add, you ( along with help of the forum) trouble shot the problem and fixed it, instead of just covering it up with the plastic over it. Success is good!

[KTP]

Occam's Razor - a simpler explanation is better than a more complex one.

     I tried to prevent you from pulling most, if not all of your hair out chasing the problem. Lesson learned I hope, I had the same lesson too. In the digital realm, thou shalt use pull up/down resistors and never ever leave anything floating. One would figure that on the output of a fet that really isn't necessary, but its still digital, just at a higher voltage and/or current. And thanks to codeboy2k, I also know about the capacitances that are involved.

     I just would like to say that is a very cool project, I like the PacMan demo in the YT video. I assume that once all the kinks are ironed out, these will be available forsale?

Chris

 
Thanks Chris.  The main reason I didn't jump straight to the pullup/pulldown solution is I wanted to research the problem and see if there were any easier solution (either in hardware or software).  The resistors still feel like a "hack" and don't particularly make me happy, but I am at the point where I don't see an easier solution.

I will probably do a rev2 soon including pads for these resistors plus a few other changes.  I would like to do a few hundred boards eventually such that I can justify pick and place and reduce component costs significantly due to volume.  The board was designed for a particular application but I have tried to make it universal and modular.  It uses some fairly high quality components like the  3 amp switching regulator, 16 mbit flash, 60V common mode 15K esd protected RS485 chip, etc.  The goal would be to produce the boards for around $35 including pick and place costs and then have the ability to grab as many boards as you want, throw them together in any configuration, talk to them all via RS485, and make any type, any size RGB display at 90 frames per second.  We might try kickstarter just to get the quantity high enough to make several hundred boards near that cost.  Otherwise, it takes me about 8 hours to solder each board (more time now that I have to add all those pulldowns).