From my experience on other similar modulation schemes they do tend to fully load the tones - it probably loads the processor less in that, for example, there is only has one fully loaded tone to demod rather than two half loaded.

I think the bit loading algorithm will just try and equalise the SNR margin across the frequency range. Inevitably that means the higher frequency tones will have lower loadings. However, I wonder if there are any lines actually synced at 40mbps which fail to use the highest frequency tones? Clearly any lines which sync at less that 40mbps will not suddenly start using higher tones for the 55mbps service. I suspect that any lines actually able to hit 40mbps are already using those tones.

I think it is probably a little more subtle than that, but my reading is that this isn't going to be a real issue.

I've looked at bit loading patterns on VDSL & ADSL modems (where you can get the stats - there are also plenty online as well) and I've yet to see a single one that works as you've described. That is fully load bins to the margin limit and leave others unused. It would be a distinctly sub-optimal thing to do in that equalising the SNR margin on each bin will make for a more resilient connection. Also, the argument that it saves processing cycles is surely spurious. These chipsets are designed to work at the highest speed that the VDSL2 supports, and it's all done using dedicated DSP (digital signal processing) circuits, not using a general purpose CPU (which wouldn't stand a chance of dealing with the task). I doubt it makes the slightest difference to the efficiency of the circuits where all the heavy lifting is done whether all the bins are used or not.

This set of graphs is typical

http://vdsl2.altervista.org/

no affect, the signal is still pumped out at the same power.

Bit loading isnt the cause of crosstalk.

In reply to a post by MHC:

From my experience on other similar modulation schemes they do tend to fully load the tones - it probably loads the processor less in that, for example, there is only has one fully loaded tone to demod rather than two half loaded.

I think there is value to be gained from doing the opposite. A policy of deliberately under-using as many tones as possible, you make the whole communication channel less susceptible to noise. Every under-used tone is easier for the receiver to identify in the QAM constellation.

If the modem reported to me that actual SNRM was 12dB, I think I'd prefer that the "extra" 6dB came from every tone, rather than as 0dB (no extra) on half the tones and 12 dB on the other half (because they were unused).

I suspect that "processor load" is a strange quantity, as the part affected most will be a DSP, which will be designed for the heavy parallel load of 4000+ tones, unlike a CPU, with one (or just a few) threads.

In reply to a post by TheEulerID:

I think the bit loading algorithm will just try and equalise the SNR margin across the frequency range.

It makes sense for that to happen, yes.

In reply to a post by TheEulerID:

However, I wonder if there are any lines actually synced at 40mbps which fail to use the highest frequency tones?

Yes, I could find some lines like this. The analysis I did on some lines suggested that, even if the modem were given the extra 15Mbps of speed, it still wouldn't use the tones that weren't in use.

However, if what you suggest in the first part of the quote (about SNRM balancing) is true, then there probably *are* some lines which would have used some extra tones.

I'll have to check some more later

I think it is probably a little more subtle than that, but my reading is that this isn't going to be a real issue.

That's my thinking, currently.

I suspect the number of lines that have the potential to sync over 40mbps yet have no bit loading on the highest tones is, if it exists at all, a very small number. Given the way the SNR tails off with frequency is fairly predictable for a well behaved line it might be able to model it.

That said, given the way the ANFP changes the PSD according to the cabinet's distance (electrically speaking) from the exchange, this might not be so simple.

In any event, I'm gradually convincing myself that the 40-55mbps upgrade on BT Infinity 1 lines is not going to have any measurable impact (although I await the inevitable postings from people saying they are seeing such an impact through a loss of speed on their particular non-BT consumer service).

Shouldnt make hardly any difference - DMT uses whats called a waterfill algorithm for bitloading.
Waterfill means that it does not fill all the lower tones before moving on to the next. It means it loads all available tones with 1 bit, before starting over the whole frequency to fill up with 2 bits and so on and so on.

Therefore its highly unlikely that the EU will be utilising any new frequencies that they couldn't previously use. They will just load more bits per bin across the frequencies available up to the maximum bits that tone can load depending upon the SNR.

---
The waterfill method is much clearer to see in on an adsl line which isnt subject to as much PCB and differing PSD masks that VDSL is. A prime example is here

This line is capable of much higher speeds than the technology limitations. See how none of the tones get full bit loading (despite there being sufficient SNR to fully load a bin). There's sufficient bitloading at the higher end frequencies to be able to allow the line to sync at full speed without the need for the lower tones to need to load the full 15 bits.

If DMT didnt use the waterfill method, then that line would show fully loaded 15bits for the lower tones and no loading of the higher frequencies.

The inverted U shaping up to tone 50 is the result of PSD mask.