Hi,
I just need someone too have a look at it and give feedback. 4 eyes see more than 2. Did I forget anything vital ? And yes there is currently no polyfuse.
Thnx.
Are you sure this will work. You seem to have the P channel FETs connected to V_LED which will be a voltage greater than +5V and you have the gates driven from a shift register that can only give out 0 or 5V. In that case I don't see how the FETs are ever switched off.
Ah, thanks for looking at it!
Now what about the 100k pull-up resistors... that should work if the shift register outputs were high-Z, but it seems I can't control that on a per pin basis with the 595 chips. DARN!
So I'll have to find a shift register with 'open collector' outputs instead of 'push/pull' ones. If such a thing exists. I don't have much space left on the board, so adding say transistors to deal with the gates won't be an option.
Any other traps ?
that should work if the shift register outputs were high-Z,
Not even then because even at high impedance you would be subjecting the pin to a voltage greater than the supply voltage of the chip and that is a no no.
so adding say transistors to deal with the gates won't be an option.
That's all I can think of at the moment. Surface mount transistors are small enough.
The 5V again... but "in principle" it would work IF the chip could tolerate the higher voltage. The mechanism.
Adding another set of 8 transistors, even if smd, poses a routing problem. Unless I use really small parts, which I can neither see nor handle anymore. The board is pretty packed already. And a large parts count number makes assembly unpleasant. The usual problem.
Ideally I'd need an 8-channel P-FET array in a chip with included drivers. I could use darlington NPN source driver chips, but they drop too much voltage. Or are too slow, can't get them, you name it.
This is cruel 
Edit:
Maybe I could just swap the 595 with one of the constant current sink drivers to pull the gates low. As I understand it they're not switching mode, but linear.
silly question, have you tested this design
To some degree.
This is not the very first incarnation of it. I have a version 3.04 up and running for several months now (sold a few too), but that is solely based on 4x 74HC595 shift registers. Thus the current driving capability is limited.
Here I've replaced the current sinking 595s with the constant current sink driver chips, so that shouldn't be an issue as they are shift registers too. Basically just swapping the chips and some minor adjustments to the schematic. As far as I can tell from the datasheet the logic part of the STP.... chips is compatible with 595 (+ or - slope etc.).
The current sourcing is what is "new" here. I have a source driver chip at hand and used it, but it's DIP only and turn on/off delay is too big (up to 10µs worst case, UDN2981A). There are others, but I can't get these. So I went for the P-FETs, which is the untested part here.
As I can't add more parts to the board and the 595 can't handle more than 5V to turn off the P-FETs, I thought to use the sink drivers to pull the gates low. They can handle the higher voltage. As they're not switching mode, there shouldn't be strange oscillation problems on the gates (I hope). I suppose if a current source (pull up resistor at the FET gate) with a very small current is connected to a sink-port of the chip, it will try to adjust its internal resistance to allow for the current set with an additional potentiometer. Of course the current is fixed by the pull-up, so it can't rise and the chip should just turn fully on. Just like a constant current battery charger. Without the battery the output voltage will be at the max. value, internal resistance will be minimal. I also hope that I don't run into trouble with the current regulation logic and discharging the gate capacitance.
So far with my theory.
The major disadvantage of this proposed version of the thing compared to what I have working right now is the added cost. Using the driver chips and FETs and voltage regulator would easily double the part cost. That makes it rather unattractive and I'm not sure if it is worth the effort for indoor usage at all.
So it is basically educational and I might only build one or two boards, just to see it work (or burn, blue smoke) 
How about using the 596 shift register. That has open collectors outputs. The only snag is that I can't see on the data sheet where it gives the collector emitter breakdown voltage for the open collector outputs. This needs to be greater than the LED voltage.
Sounds good, but that part seems to be obsolete. I could find a datasheet on TI's website, but no info on breakdown voltages I could find.
Farnell has a stock of >> 23 << and the unit price is £13.92. It's also in a DIP package which is too large. SOIC-16 or smaller would be OK.
That is a "no no"