I'm seeing similar symptoms - but why is it so random? (I also plotted SNR over frequency and time and it looks like 66% of the D1 band is getting chopped, specifically.) I've had it happen for 15 minutes, or sometimes for 18 hours.
I also wrote some code for the HG612 that 'watches' the Attainable Rate and resyncs if there's enough gain to be achieved by doing so. An extension of that code plotted the graph. If there's interest, I could make the source available so that others can write interfaces for other modems - as well as making it more user-friendly and more fully-featured. Porting to Mono should be pretty easy - then perhaps it could be run on a Pi or something.