Although exhibiting better dimming performance than CFLs, commercially available LED lightbulbs seem to exhibit poor linearity compared with incandescents (especially in the bottom 20% of the dimming range).
On the workbench, I've seen that it's possible to dim LEDs to very low effective outputs using pulse width modulation. It's also reasonably straightforward to envision using comparators and zero-crossing detectors (or something more complex, like a microcontroller with an A/D converter) to analyze the AC input coming from a triac/thyristor, and generate an appropriate pulse train.
So what is the big challenge here? Why does it seem that all commercially available dimmable LED bulbs have poor linearity (especially at the low end) compared to incandescents? What am I missing here? Is it the voltage drop across the LEDs that's "offsetting" the linearity?