Upstream Power Backoff (UPBO)

Discussions for BiPAC 8900 series: 8900AX-1600, 8900AX-2400, 8900X
Post Reply
gatekeeper
Posts: 167
Joined: Sat May 26, 2012 4:45 pm

Upstream Power Backoff (UPBO)

Post by gatekeeper »

Before I get to the essence of this topic, perhaps a word or two of apology is appropriate to you, billion_fan, for me contesting your assertion a few days ago that Output Power on a VDSL line was controlled by the DSLAM. I've now realised, from background research that I've now done on the Web, that a mechanism is built into ADSL and VDSL lines that enables the DSLAM to talk to the user's modem (in my case, a brand new 8900AX-2400) and limit the power level of upstream signalling from the modem. This, it turns out, has long been written into the standards for ADSL/VDSL signalling and is known as compliance G.993.2. So, certainly up to a point, you were correct, billion_fan, in suggesting that the DSLAM controls the upstream power from the user's modem (modem-router).

Apparently, UPBO-capability of modems is now a mandatory functional requirement and is designed to prevent undue levels of far-end crosstalk that would otherwise arise from short lines with high upstream output levels; excessive crosstalk would otherwise then limit the upstream speeds of all other lines in the same cable bundles. I gather that, in a modem complying to this requirement, UPBO is on by default. There is, however, a criterion for UPBO to be turned off (not by the user), and that's based on the line length; if the line is deemed to be short, UPBO is on. But if the line is a long one, then UPBO is turned off. This would make sense, since the signal from the user's modem would suffer a fair amount of attenuation (reduction in strength) on a long line.

It's left to the modem and the DSLAM to automatically decide whether, in a particular instance, the line (ie. the copper line between the user's modem and the DSLAM) is deemed long and therefore whether UPBO should be turned off or not. But therein lies a crux - thus far, I've not uncovered any written DSL standard that defines 'long' and 'short' for this context. However, my own 8900 sits on a VDSL line where the copper length to the street cabinet is some 720 metres, and the full distance to the exchange is 3.3km, so in my view I'd say that that 720 metres constitutes a LONG line. Would you not agree? But it looks like, in my connection, UPBO is ON and so is unnecessarily reducing the strength of the upstream signal. So, in practise, what's the demarkation between 'short' and 'long', one might ask?

Some experts claim that Openreach's management of UPBO is poor and indeed that there are differences between Huawei and ECI DSLAMs in how this criterion works. But it does rather look as though my 8900 is playing it far too safe and, rather than giving me the same or even an improved upstream speed (sync rate), has instead CUT IT it by some 12.5% - from the previous 8M bps on the 8800NL, to 7M bps on the new 8900. In fact, if the 8900's stats are to be believed, the amount of upstream power coming out of my 8900 is a paltry 2.9dBm at present, and this contrasts starkly with the downstream power of 12.9dBm.

I think you'll find that, in absolute terms, 2.9dBm (which is a logarithmic measurement) is a power level of just 1.95mW. The latter, therefore, is absolutely miniscule, and the upstream signal, over that 720-metre length, must be really struggling to compete with noise and crosstalk. It's remarkable, though, that the upstream error rate still appears to be reasonably good and the 8900's overall performance stable.

Since absolutely nothing else has changed in these last few days since swopping over the 8800NL for the 8900AX-2400, I highly suspect that the 8900, in conjunction with the DSLAM in the street cabinet, has got its assessment of the line length quite wrong and has consequently lowered the output power from the previous figure of 5.1dBm to 2.9dBm. Surely, nobody on this planet could, for FTTC purposes, regard 720 metres for the copper stretch as a 'short' line?!

Is there any way of finding out from the Billion firmware developers the figure that they, Broadcom, or the international standards committee have used for defining 'long' in this matter? And whether that assessment might be improved in the 8900 via a future firmware update? I'm wondering whether, rather than there being one specific cutoff point between long and short, there might be a range of length bands that can possibly be used, and the DSLAM perhaps able to instruct the user's modem to reduce its upstream power level by a variable amount that depends on which length band it sits in.

Another thought is that the ever-present zero attenuation figure on the upstream side cannot be helping in this, because if it plays any part in the 8900/DSLAM's assessment of the line length (my guess is that the way that any engineer would devise that would be for the DSLAM to look at both the downstream and upstream attenuation figures and then, from them, calculate an average figure for the line length), then it will surely give the false impression to the DSLAM that the user's modem is all but next door to the DSLAM - but here, quite the reverse is the case.
Post Reply