There are no stabilizing capacitors possible with the LM317 current limiter and If you add them all you will accomplish is to spread out the PWM pulses.
The data sheet seems to disagree. It recommends a 1 uF capacitor across the output to improve ripple/transient response. If you look at the transient response on page 4, you can see that it's almost 10% of output voltage without it.
However, that doesn't matter. You cannot source the input power for a LM317 for this application from a PWM pin on an AVR processor directly, because the available current output from those chips is way too low to drive the LEDs we're talking about here. (There was talk of a 74HC595 -- same problem there.)
You willl not have as much current available as you set with the sense resistor because the Fet Rdson is not in the feedback loop
That statement is wrong, as far as I can tell. I will have exactly the current set by the sense resistor. I can use the same current through as many components as I want in series. It's the same current. All I need is sufficient source voltage.
To follow your suggested logic, the current control also wouldn't include the LEDs themselves, because they are also not in the feedback loop. That's simply not how current (and current regulation) works. The voltage drop over the sense resistor is I times R. I set R so that the voltage drop at the current I want is exactly the reference value. For example, if I want 800 mA current, and reference value is 1.25V, then R == 1.25/0.8, or 1.5625 Ohms. (On another note, that needs to be at least a 1W resistor, preferably 2W.) Trying to put anything more than the sense resistor in the feedback loop (from out to sense) of the regulator will hurt the precision of the regulation.
The same argument goes to your previous suggestion that there is "no feedback as to the current drawn" -- the feedback is with the sense resistor, which is in series with the actual current drawn.
I agree that when all low-end drivers are turned off, the regulator will try to "float to top" in an attempt to increase current.
That's not necessarily a problem, especially when you just switch the load to another column, so the regulated voltage won't actually float up.
[/quote]I have used L317's for current control many times and all of my applications were stable and worked well, including some variable current designs that however aren't applicable here.[/quote]
I didn't say they weren't.
I described my suggestion for how to use double low-end drivers to solve the problem of PWM modulating the row strobes using a separate input for the "modulate" than for the "row strobe," essentially by folding the logical-and function into two switches in series.
I also expressed concern at trying to obtain a PWM power signal, and then regulating the current output of that. At a minimum, that signal will be pretty spiky (as shown on page 4 of the data sheet.) With a capacitor, you'll remove the spikes, at the cost of smearing the PWM. Maybe the spikes don't actually matter in this case, either -- you can try it and see if the LEDs blow early or not :-)