Kitz Forum

Broadband Related => Broadband Technology => Topic started by: WWWombat on September 24, 2016, 01:18:02 AM

Title: Vectoring - Dutch results from 180,000 lines
Post by: WWWombat on September 24, 2016, 01:18:02 AM: Here's an interesting graph from the Netherlands on the behaviour of VDSL2 Vectoring.

Averaging 100Mbps at 400m, 80Mbps at 500m, 65Mbps at 600m, 45Mbps at 800m.

Dutch cables seem to be slightly different from UK. 0.5mm copper, but they are formed as twisted quads (ie two pairs twisted together). The graphs suggest they allow 20dB per km for insertion loss, which I understand to be about right for measurements performed at 1MHz (which is different from the method and values used by BT).
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: S.Stephenson on September 24, 2016, 11:06:02 AM: Great results just a shame that I'll never see VDSL2 vectoring. :'(
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: Weaver on September 24, 2016, 11:42:20 AM: :-(
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: Chrysalis on September 24, 2016, 02:48:08 PM: that non vectored is highly variable, if I owned a telco selling unvectored services with such a wide lottery I would be ashamed.
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: WWWombat on September 25, 2016, 11:07:22 AM: Interesting you should mention the variability. If I look at the vectored results, I'd say that was rather more variable than was to be expected from previous trials: where the "vectored" results all came up close to a theoretical "single line" behaviour.

Alongside that, I notice that their "red line" average seems to sit below the majority of the solid blue area - especially in the first 400-500 metres. Surely there must be a roughly similar number of samples below as above. I see a strange straight-line effect around 52Mbps and again around 62Mbps ... and wonder if that represents the Dutch equivalent of "banding", and that there are a fair number of lines caught in there.

Anyway, the problem with variability is something that caught the eye of the Dutch too. They annotated the chart with these points...
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: WWWombat on September 25, 2016, 11:14:45 AM: Note that the spread they describe as "15Mbps" is between the 16th and 84th percentiles, making it very equivalent to the ranges used by BT in the BTW estimator (which are supposedly the 20th and 80th percentiles).
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: WWWombat on September 26, 2016, 02:32:20 PM: The next installment, focussing on the question asked in the last chart: "Why is the spread still so large?" Here's the apparent answer: Bad modem/chipset and/or bad firmware in the CPE.

This chart appears to include data from two-thirds of the connections in the first chart (120,000 results out of 183,000). It looks like perhaps 15% of chipset/firmware combinations give a bad performance.
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: S.Stephenson on September 26, 2016, 03:32:44 PM: No blame for the metallic path?
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: Weaver on September 26, 2016, 04:47:37 PM: Regarding the question of why G.Vector doesn't always get a chance to deliver as you might hope, sections 4.3, 4.4 of this doc discuss this :
https://www.broadband-forum.org/marketing/download/mktgdocs/MR-257.pdf
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: WWWombat on September 26, 2016, 06:15:51 PM: Quote from: S.Stephenson on September 26, 2016, 03:32:44 PM
No blame for the metallic path?

No analysis of the quality of any of the pairs. Presumably such problems get averaged over a few thousand users.
Title: Re: Vectoring - Dutch results from 180,000 lines
Post by: ejs on September 26, 2016, 06:29:51 PM: I found the source of all these graphs in a PDF from here (http://4gbb.eu/index.php/publications/white-papers), in case anyone else is interested.