I've been using the graphing scripts to turn my Bit-Loading numbers into graphs. I was also playing with the scripts to get good resolution to the graphs - but it made them a bit wide.
I first got some puzzling results for my old line, where the graphs seemed to show that the modem was using far too many bits in the lower quarter of the spectrum - where the comparable graph for the new line was not. Both lines were using about the same pattern of bits in the higher 3/4 of the spectrum. Both lines were at 40/10, so I'd have expected both graphs to show roughly the same number of bits (even if not in exactly the same pattern).
Then I got a stranger result...
I got vastly different results from the *same* modem, in the *same* location, taken 40 minutes apart, without a sync having happened between. There are *loads* more bits being used in the lower 1/4 of the spectrum.
Take a look at this "montage" image, which shows 3 (wide) bit-loading graphs from the same modem. The top was taken in january, and reasonably matches the one taken at 20:08 today (the second graph). The final graph (from 20:49) looks vastly different! You need to click on the image to get the full-size version.
When I look back at the raw data (which matches the graphs), I can see that the differences start at tone 33, and go on to tone 1023. Those are too nicely "power of 2" numbers to be a coincidence - suggesting some form of fault. For tone 32 and below (ie U0), and tone 1024 and above, the values match the earlier results pretty well. But between 33 and 1023, one set is just considerably bigger.
Any ideas, anyone?
Oh, in the same 30 minute period, the number of BitSwaps had changed from 3917 to 3938. Obviously not enough to cater for the changes seen in the graphs!