Back to @Cloudane shortly...
Ach - I had quite a post written out, then accidentally lost it.
A quick summary of what I worked out
- Hlog graph nice and smooth, and seems to match the attenuation
- QLN graph shows the line to be noisier than comparable lines, for tones above 512, by 5dB - 10dB
- This extra noise probably accounts for 15-20Mbps lower speed
- Because the noise applies fairly evenly across the spectrum, but only above tone 512, it looks like crosstalk from other FTTC subscribers
- So most of the speed drop, over time, appears to be from crosstalk.
- Overlaid over this will be DLM interventions, and de-interventions, causing speed reduction/increase of 5-6Mbps.
- The DLM intervention that applied until this morning was at the "standard" level of intervention. No sign of anything more than that, and no banding.
- The main thing to watch will be the ES rate over each 24 hour period. 600-800 ES's per day is OK for speed-profile ISP's like Plusnet (I used to see that consistently).
I'd say the only things that can be done will be focussed on keeping DLM uninterested in your line,by either
- Reducing the slow accumulation of ES's over the course of a day, or
- Stopping the occasional "large peak" in ES's that trip you over the DLM threshold
The former could be helped by any number of small changes - the DSL cable, a new filter/faceplate, checking wiring and joints, and not running close to power lines. Whenever you make a change, restrict it to one thing at a time, and then monitor the "Errored Second" graph for a few days, to see if it has made an impact. (Tick the "Totals/Averages for ES Graphs" box in the footer, to get some extra helpful lines on the graph).
The latter is best helped by trying to figure out what event causes the large peak, and eradicating that.
On the other hand, getting BT to deploy G.INP on ECI cabinets would probably stop any more DLM intervention in one fell swoop.
Then we had ONE bad minute or so of noise on the graph and it slapped it back on again and has stayed on ever since.
Are you sure?
One bad minute, with very high CRC counts, can still only add 60 errored seconds onto the ES counter. It shouldn't have too much of an impact, by itself.