I could be wrong, but I just assumed this was something to do with line length
The basic design for UPBO seems to depend on a parameter known as a "reference length", which can be set differently per band. And I think it does include the estimate of the electrical length of the line (ie attenuation). And the outcome certainly looks like it lowers the PSD mask. All knowledge taken from my favourite white paper on UPBO:
#5 on this website. So it looks like length is involved a lot.
For my line, it looks like they've set things to really dissuade it from using U1: SNR-per-tone is 10dB lower in U1, even though QLN noise is 10dB better. I guess power must be lower in U1 by 20dB.
All good so far, so long as that estimate of electrical length doesn't change. In these circumstances, UPBO looks like it should be static.
However, I have seen times when the transmit power (per band, visible in "pbParams" output) has changed after a resync. No idea why; perhaps the length estimate went wrong on that sync, perhaps something else was detected. I just leave a marker in my head that things aren't necessarily static, and that I don't know enough...
From memory theres several (5?) different masks in use based on line length from cab.
The ANFP has a graph that shows 6 different downstream PSDs, based on 6 different CAL values. However, it is stated in text that CAL can actually vary from 0dB to 52dB in steps of 2dB, so the graph is really only a depiction of a few examples. There are really 27 different masks.
If so then neighbouring lines of similar length should have little impact because they should have a similar [upstream] PSD mask.
I suppose there could be more room for variance if your line in the bundle is getting crosstalk from a line of different length.
A line is likely to share a little of the path in a cable bundle with lines of similar length. At least the distance from, say, the street chamber to the DP at the top of a pole. But the rest of the route back to the cab is going to be shared with a variety of lengths ... I guess the likelihood of crosstalk
here from disparate lengths depends on how organised BT are about how they subdivide larger bundles at each pole/DP.
However, I suspect a fair proportion of crosstalk is sourced from the tie pairs, where your immediate neighbouring pairs could be any length.
Either that or Im not quite sure what you mean about turning up the power :/
The "turning up the power" part is an alternative way of thinking about water-filling.
In the traditional way, you start with fixed power, and fixed SNR per tone. You then iteratively ask: has this tone got an extra 3dB of available SNR? If so, give it an extra bit. Until you reach package speed, or run out of SNR.
In this alternative way, you start with minimum power, and no SNR. You then iteratively ask: can we turn the power of this tone up by 3dB? If so, increase the power and then ask if it has 3dB of available SNR? If so, give it an extra bit. Until you reach package speed, or run out of power increments.
In the latter way, you end up transmitting at lower power - in fact, just enough power to achieve the desired 6dB margin. And you wouldn't transmit on any tones where this lower power level wasn't enough to raise SNR above the margin.
This is just a thought game though - a potential way to make UPBO dynamic, and account for some of the times I have seen a change. I've seen no evidence that such a mechanism exists.