Well, well, well! Thinking about your problem Tom, made me go and have a look at my own bit-loadings on rsw.
And guess what
, I am also getting some 'oscilation' of the bit-loadings in the lower tones of D1, albeit not as frequently as yours. Unfortunately, for some reason I will need to investigate I can't seem to get rsw snapshots working at the moment, or I would post the graphs.
However, sometimes I am getting additional shared (blue) tones in the 30-90 tone range and these are allowing these bins to be loaded as high as 14 bits for a time; then (because of errors I guess) they get first reduced in number, then removed and the loadings fall to 4-6 bits per bin. I think this is just part of the 'normal' error correction on my line, trying to utilise unused bins to carry bits from others with errors, only to find that they are (presumably) worse! It's a very dynamic thing.
I don't remember seeing these before (although I just may not have been looking closely enough until your post) but the DSLAM did force a resync on me earlier this evening, and I don't recall having shared tones in these lower bins before that.
Perhaps somebody wiser will now be able to explain this to both of us. I don't seem to have the wild SNR oscillation you posted, it just appears like that on mine if you were to take 2 snapshots, one with the shared tones on and higher bin loadings and then later when they are removed again and the loadings fall.
/correction. The DSLAM did not force a resynch on me, it was just a data harvesting hiccup .correction/