Hi Eric.

If only our mutual friend Oleg Sergeev would kindly grant my wish to take part in this, just like you did.

In my new design a total of 50 wires go from the mainboard to each deriverboard:
8 lines for AD0..AD7 (to four HC573s in the driver)
4 lines for LE0..LE3 (to LE of the latches)
1 line for OFF (to /OE of HC573s whose output is poulled down through 10K resistor array before going to ULN2803. This line is used for PWMing)
5+1 lines for ADD0..ADD4 (to the 5x32 decoder made up of 4515s as column multiplexer)
2 lines for VCC (yes VCC is taken to the drivers through flat cable, hence implementing a true star topology or something)
4 lines for VDD (12v) (The LEDs are driven with 12v don't worry, already done)
Total: 25
Another 25 lines are dedicated to GND i.e. every other line on the 50 line flat is grounded.
A 50 wire flat will do. And even the stupidest techie should do it.

You are right. Not implementing this so called strobe in my first designs caused some "shadow" following the moving text. I learned to turn off the display when filling the latches to silent this shadow. The aforementioned OFF signal is used to do that. The pseudo code is something like this:

turn off the display using OFF line
put appropriate values in the latches
turn on the display
some delay or some processing
do it again

As far as the 32x32 matrixes are multiplexed in paralell (and synchrinized) the flicker problem does not increase at all.

Maybe I don't correctly get what you mean, but I don't have to multiplex the whole 256 columns, just 32 columns, hence no reason to worry about timing restrictions. I think even the "plain vanilla" could cope with it.

Best.