I've recently been smitten by the 15dB noise margin after stupidly resynching my modem too many times, in an attempt to evaluate pros and cons of different microfiliters.
Target SNR is a topic that's been discussed many times on this and other forums. The excellent BTW "myths and legends" pdf does explain how my margin got increased (ten resyncs in an hour), but I've never found much said from BTW as to how DLM brings the margin down again.
However, there's a BT patent application which can be viewed at the following URL...
http://www.freepatentsonline.com/EP1953959.html Now if, as seems likely to me, that patent describes BTW's current practice then a few more facts arise that I'd not seen published on the forums. Refer to the link in case I've misunderstood anything but, if I understand that patent correctly....
1) A line which is "unstable", i.e. has lots of CRC errors, may eventually pushed one step (3dB) up the noise margin ladder.
2) A line which is "very unstable", i.e. has had 10 retrains in an hour, may immediately be pushed two steps (6dB) up the ladder.
3) Your target margin won't ever be raised unless your sync speed is 800kbs better than your Fault Rate (which was established during the initial 10 day period).
4) Error rates are categorised as "very poor", "poor", "acceptable" and "very stable". If your target margin's been increased then it won't ever be reduced again just because your line subsequently exhibits an "acceptable" error rate, it has to be "very stable" for a very long time (weeks/months) to get reduced.
5) Here's one that sounds interesting... if a line completely fails to synch, it will have its target margin reduced in the hope of maintaining some service, and will be flagged for investigation.
I find (3) worrying because, sooner or later, everybody's going to see transient instability. If you were 'lucky' during the ten day period and have a decent Fault Rate you'll be fine as DLM will leave you alone. However, if you got allocated a poor Fault Rate, and your line's subsequently improved, you'll be punished and lose - maybe forever - all of the benefits that the improved line should have brought.
Tempting as it may be, I'd not recommend that anybody deliberately try emulating (5) to fool the DLM into dropping the target margin, as that could obviously cause grief for BTW, and might backfire on the cuplrit if it leads to manual intervention on a line that's wrongly believed to be dodgy, making things even worse.
Any thoughts, anybody? Anybody from BTW and willing to pass comment?