I hope I'm not hijacking this thread but I've been wondering about a related question. Basically when driving high currents, is it better to smooth the PWM before the current amplifying transistor or after?
Here are my thoughts:
If you do it before the transistor (i.e. to the base of the transistor) you don't need a big cap, but you are running in the "active" region (i.e. unsaturated) so it is burning energy and you are relying on a specific hopefully nearly linear Hfe transistor curve.
Also, if you do it after, some wire is carrying high current PWM... would that emit undesireable RF?
It really depends on the application - sometimes the end load can do the low-pass filtering anyway (such as an inductive load like a motor). For class D audio amplifiers the whole motivation is efficiency so you put up with large LC filter components at the output in order to get the 90 to 95% efficiency of amplification. If the load is inherently analog and does not provide low pass filtering you might want analog amplification after a low-power low-pass filter (low power filters can be cheaper and better performance).
For LEDs there is no need to filter if the frequency is high enough, the human eye does the work.
In general digital amplification is prefered because the circuitry is more efficient, cheaper and component specs are less critical. The last 20 to 30 years have seen switch-mode take over from linear-mode in many areas.
The whole question of preventing unwanted RFI is important of course, and you are right that high currents (and voltages) make the problem worse, but good design goes a long way - you have to do more to suppress RFI but then you don't have to deal with so much heat dissipation (which is usually going to cost more to cure - heatsinks, fans, larger power supply).