line stability changes unexplained

[#1]

Here's a strange one!
On 1st Sept, my sync speed was 1248, and "Noise Margin" about 27, falling to about 21 in the evening. (6dB. Not unexpected.)
On 2nd, PPP service had been dropped ovenight but DSL session appeared to have continued unaffected. On power cycling the router (Lazy, I know) the speed was 3648 and the Noise Margin about 15. (Fine!)
That evening Noise Margin plunged to around 7, and then DSL was briefly disconnected about 21.20. On reconnection, I had sync 2179 and Noise Margin around 15. (Fine!)
On 3rd, the Noise Margin was 26ish all day, then plunging to 14 by 22.00.
On 4th is was only 19-21ish, but down to 14 again by 22.00.
On 5th I found the PPP service had been dropped overnight again but DSL session had not (according to the router status page).
Ever since I reconnected PPP from the router control page that morning, Noise Margin has been stable in the range 21-25. It has hardly changed with time of day, or changing from master to test socket, or even on a resync to 3648.
Where did the new stability come from? Any ideas more generally?

[#2]

Welcome to the forum woodsider,
Is this a new broadband account in it's training period (10 days) or a longer established one?

[#3]

This is very long established. It has been blighted from time to time with disconnections and speed isssues and I'm trying to get them sorted at last.
Two tickets open, but I get an impression of "not seeing the wood for the trees" in some responses, but others have been very helpful. I thought I'd try here for enlightenment.

[#4]

HI Woodsider,
At the moment the fault is in the correct place to be actioned. I'm afraid having checked around that I couldn't shed any more light on the matter at this time, so, for now I'm going to have to say to leave it with the faults guys.
Jojo

[#5]

... ... ...  whereas ... today's resync has dropped the Noise Margin by the expected ~10dB  to 14.5dB.

[#6]

And the explanation is ... !
Well, it appears to be mainly coincidences. Disconnecting a noisy extension phone line (by plugging the filter into the test socket instead of the main socket) increases Noise Margin. Jumping to a higher sync rate decreases Noise Margin. In my case, these must have just about cancelled out, so I thought neither was having affect. Actually, I did note a slight change (about 2dB), but by coincidence again, that value of 2dB equalled the amount of the variation with time of day afterwards.
The morals are:
1. Only change one input at a time if you want to understand what's going on. (Easier said than done when BT use my change to make one of their own to the sync rate.)
2. Don't assume, just because a telephone extension line is unused that it isn't making your line noisy and being the root cause of slower broadband.