What do you make of these stats?

[#1]

Now that I'm using a Thomson router, I tried out RouterStats to see if I could find a pattern to what is happening with my line.  These two screenshots were taken last night, when my router had a sync speed of 2528.  It had been happily connected for over three days with a noise margin of 13dB before these incidents.
These incidents seem to happen randomly, approximately every couple of days, but I can't seem to find a pattern.  It doesn't seem to coincide with central heating, incoming phone calls etc.  I should be getting a much faster speed, but these disconnections followed by short periods of low margin are keeping my profile at 2000.  All day today the router has been synced at 5408 with a noise margin of 7.5dB and seems to be perfectly stable, but in a couple of days, this is going to happen again...
Any idea what could be happening?

[#2]

All one can really say is that during the periods when sync speed is held down to restore noise margin there must be considerable RF noise about. If you tune an AM radio to 612kHz during this time does it pick up any static? See this Kitz page for more on REIN.
Does the plot of  bitswaps vs time (from RouterStats) show significantly different behaviour during the periods free of noise and in the presence of noise? Usually there is considerably more bitswap activity after dark (15-30 swaps per minute on my connection) than during daytime (0-3 swaps per minute).

[#3]

Thanks spraxyt, I'll try the AM radio thing and check the bitswaps if it happens again.
However...
I tried another speed test last night, and for the first time since switching to PlusNet on the 5th of February, my speed is back to 4Mb!  It seems that at some point yesterday my profile finally increased from 2000 to 4500.  There's always the chance that another of these wierd noise events will occur and knock it back down, but I'm keeping my fingers crossed since I found this:

Quote

Planned RADIUS server maintenance - Wednesday 16th March 2.00pm-3.00pm 15/03/2011 @ 14:39 When's this work happening? Tomorrow afternoon, Wednesday 16th March. What does it affect? Broadband Connectivity. How long will it take? About fifteen minutes. What does the work involve? We're moving a number of customers from our 'IP Stream' broadband network to our 'IP Stream Connect' broadband network. Am I likely to notice the work? It's unlikely but you might. Broadband customers with connection usernames that end '@plusdsl.net' will be moved across to the alternative network the next time they disconnect/reconnect. Whilst this won't have any noticeable impact on the performance of customers' connections, some of the more observant/technically minded may notice it. Is there anything else I need to know? Further detail about these changes can be found in the blog post here - http://community.plus.net/blog/2010/12/17/new-plusnet-broadband-network/ Kind Regards, Matt Taylor Customer Support

The date doesn't quite correspond, but I am one of those '@plusdsl.net' customers.  It'll be interesting to see if PlusNet have fixed my problem by doing this.  In the meantime, I'm delighted to be able to watch iPlayer in HD again!

[#4]

I spoke too soon.  It's happened again, shortly before 6pm this evening, and dropped my speed right down again.  Complete silence on 612kHz AM.  The Bitswaps/Time chart on RouterStats seems to show the number increasing dramatically at the point of disconnection, but settling down to the same level as before immediately after, despite the lower noise margin (image attached).  Each time the connection drops, the local figures for "Loss of Signal" and "Error Seconds" on the router's stats page go up by tens of thousands.  I can't claim to understand what any of these readings mean!
I'm just going to leave my support ticket open for now (it's currently on hold until I get switched to 21CN at the end of the month) but if anyone has any insight into what might be happening I'd be interested to hear it.
I enjoyed getting my old speed back for a few hours there!

[#5]

I expect the bits per tone plot looks different between the times with a higher sync and those when sync speed is lower. May show whether the problem extends over a large band of frequencies or just part of the range.
Unfortunately these plots only identify the nature of the problem, they don't show the cause. It's a moot point whether it will follow you to ADSL2+.

[#6]

Quote from: spraxyt

I expect the bits per tone plot looks different between the times with a higher sync and those when sync speed is lower.

Looks like you're right.  For the hell of it, here are two charts. BitswapsTonesGood.jpg is from when my noise margin was 17dB and BitswapsTonesBad.jpg is from after it had suddenly dropped to 6.5dB.
I've also noticed the bitswaps/time graph shows a large spike not just when the noise margin suddingly drops but also when it suddenly increases again.  It's just shot up to 17.5dB after spending a couple of hours at 7dB and there was a big bitswaps spike at the time.  When the margin is steady tt seems to show almost no bitswaps when the noise margin is around 6-7dB and more when the margin is around 17-18dB.
Again, I don't pretend to understand what any of this means!

[#7]

It looks like the connection gives up on the higher frequencies when the noise margin drops. The plot I thought might show any effect like this more clearly was the *bits* per tone plot rather *bitswaps* per tone. My current bits per tone plot is attached.

[#8]

Sorry, I misread your post.  Here are two bits per tone graphs from this evening (this now seems to be happening daily!).  My router picked up a higher sync again when the margin increased this time, so both have similar margins, but the "good" graph with the higher sync speed covers a wider range of tones.  It doesn't look as evenly distributed as yours, though.

[#9]

Those comparisons show where your speed disappears to, I've overlaid the approximate boundary of the bad one onto the good one. Little effect on upstream, progressive problems above tone 72 (310kHz), disaster above tone 106 (457kHz) (I've ignored the small number of higher frequencies that still contribute).
From these pictures I doubt that ADSL2 will improve things at bad times, possible small effect because minimum acceptable bits per tone is 1 on ADSL2 whereas it is 2 on ADSL1.
So we can see the effect of the problem, but unfortunately this doesn't identify the cause. Since it's a night-time effect it could be lighting - does it begin when the street lights, or security lighting comes on? You don't live across the road from an electrical substation?

[#10]

Thanks for that info, I'm starting to understand a bit more about what's going on.
It happens during the day as well.  Timing of the events seems random.  That's why I'm using RouterStats, to see if I can find a pattern, but if there is one it's eluded me so far.  There are no electrical substations near me and I can't think of anything environmental that could be causing it.  To be honest, I'm still inclined to think it's a PlusNet issue as it started as soon as I switched to them after years of no similar problems, but they can't find anything wrong at their end.
I'll see what happens after I'm switched to 21CN (which should be soon), but I'm now doubting it will make a difference.  If it doesn't I now have plenty of logs and graphs I can share!

[#11]

If it is related to something at the exchange, moving from 20CN to 21CN involves your line being connected to an MSAN instead of a DSLAM, so it's always possible the problem could be eliminated. Fingers crossed.