Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Weaver]

This morning I gathered some stats on the differences between ADSL2+ mode and ADSL2 mode on one of my lines with a DLink DSL-320B-Z1 modem.

Thanks to a suggestion from Burakkucat I was alerted to the possibility that locking the modem to choose ADSL2 rather than ADSL2+ might be a better setting for my ultra long lines (~65dB d/s attn; 4.55mi long).

These are the results using one modem on one line only, my line #1.

ds sync / us sync
2595 / --     auto mode; no bitswap!
2784 / --     adsl2; bitswap
2804 / --     auto
2827 / --     adsl2

ds sync / us sync / ds snrm

2795 / 557   2.8    adsl2+
2819 / 553    2.8   auto
2830 / 553    2.3   adsl2
2791 / 553   3.1   adsl2+
2848 / 553   2.1   adsl2
2848 / 553    --    adsl2; [ no bitswap!]
2823 / 553    3.0   adsl2
2809 / 553   3.1    adsl2
2784 / 550    3.0   adsl2
2770 / 553   3.3   adsl2+
2827 / 550    3.0   adsl2

Note that ‘auto’ means auto-select mode where the modem chose who-knows what mode itself.

Geometric means of the above are, approx, -
auto    = 2811.5
adsl2+ = 2785.3
adsl2   = 2819.9

[burakkucat]

And the winner is . . .

[roseway]

Have you been at the magic mushrooms again?

[burakkucat]

No, just the cat-mint. 

[Weaver]

It's so very slight, but tentatively Burakkucat possibly wins from those stats. Adsl2 is possibly very slightly better than adsl2+.

If auto actually really means adsl2 in practice, and it never goes to ADSL1 by choice (which would be very surprising, but I can't prove it) then I could lump in the auto results too which would improve the picture in favour of ADSL2.

I need some advice on statistics to know what the significance of this very small data set is. There have not been enough tests done, certainly not enough runs with adsl2+.

Mrs Weaver was wanting to go off and do other things and was waiting around for me to give her the go-ahead to plug the test modem back into the router as normal, instead of it being straight into the main LAN switch (where I could interrogate it) temporarily.

So the current advice is to definitely follow Burrakucat’s wisdom, because it is not disadvantageous, there's no evidence that it is. How long your line has to be though is unknown. You always have to check on the tone spectrum for safety’s sake.

If I had had any wit, I would have captured some stats from the modem to show anyone who might be interested in seeing a ridiculously long but I suspect clean line.

The stats from the modem’s http output are not very detailed. Not as good as many. I haven't tried telnet into it yet, there was a superb earlier thread on this modem which I need to re-read.

[burakkucat]

I have a Netgear DG834Gv5 stored away in "The Grotto". I first used it with my TalkTalk connection, when that service was provided by 20CN equipment, in G.992.1 mode . . . the device was configured in "auto" mode. When TT upgraded their equipment to 21CN, the DG834G, left in "auto" mode, then connected in G.992.5 mode . . . poorly. (Without getting it out, powering it up and checking, I believe it can only be configured as either "G.Dmt" or "auto". Not G.992.3 mode.)

I then used a 2Wire 2700HGV for some time. That device would first attempt to connect in G.992.5 mode, analyse the training response and re-try in G.992.3 mode. The latter mode it found to be viable and so did not take the final step downwards to G.992.1 mode. The 2Wire device also displayed the bit loading per sub-carrier and it was quite clear that nothing in the sub-carrier range of 256 to 511 was being used.

Hence when I finally started to use devices that were fully configurable (initially Huawei HG6xx devices, latterly ZyXEL VMG1312-B10x devices), I have always manually selected G.992.3 mode.

[Weaver]

And Burakkucat’s line is rather faster than mine at 2.7-2.9 Mbps d/s sync , so if his bins / tones don't go high enough, then mine certainly won't.

One line, cwcc@a.4 is slightly faster downstream btw, at least most of the time.

But for some reason the line tested here, cwcc@a.1 is always the fastest upstream by a sizeable amount, the upstream differences being a long-standing mystery. (I can't even cook up any candidate theories for why.)

One other problem with the weak statistics of these tentative results is the variation in the SNRM. The reported figures might well just be strongly correlated with the sync speed. In that case, the question is, what about the variations in SNRM within one mode? (where say lower SNRM is linked with higher sync rate.)

The results presented are in time order. One bizarre outlier, at 200k d/s sync below the normal range was stripped from the dataset in calculating the derived statistics as an anomaly. It was perhaps to do with its being the very first result obtained (soon) after power loss, or perhaps after upsetting BT somehow.

If anyone speaks stats I would appreciate some help. Utterly ashamed to say never got my head around the subject properly, but amazingly Theoretical Physics students at my place never got a course on the subject (well perhaps that is for the experimental lot).

[ejs]

On this type of exchange equipment, I expect auto will connect using ADSL2+. This is the type of exchange equipment where the difference between ADSL2 and ADSL2+ is very small. I think the difference is greater on the older (TSTC) exchange equipment.

[burakkucat]

On this type of exchange equipment, I expect auto will connect using ADSL2+. This is the type of exchange equipment where the difference between ADSL2 and ADSL2+ is very small. I think the difference is greater on the older (TSTC) exchange equipment.

Nods in agreement.

And it would always be worthwhile, initially, to monitor the circuit to determine the active sub-carrier range.

[Weaver]

I re-read some stuff from the old thread that was generously contributed by user G.DMT (what a star!). I haven't been able to see how to look up bits-per-bin assignments.

[rpdmallett]

Interesting that the 4.5 mile upstream rate on ADSL2 is actually slightly faster than what I get on my 1.3 mile FTTC link!

I get 15M down and 530k up from an ECI cabinet.  I’m hoping that the new firmware for G.INP (when it finally gets released) fixes this somehow as I’m convinced it’s a bug of some kind.

[re0]

I have a Netgear DG834Gv5 stored away in "The Grotto". I first used it with my TalkTalk connection, when that service was provided by 20CN equipment, in G.992.1 mode . . . the device was configured in "auto" mode. When TT upgraded their equipment to 21CN, the DG834G, left in "auto" mode, then connected in G.992.5 mode . . . poorly. (Without getting it out, powering it up and checking, I believe it can only be configured as either "G.Dmt" or "auto". Not G.992.3 mode.)

If I can recall correctly, the Netgear DG834Gv5 was the worst out of all the DG834Gs (it had Conexant chipset). v4 was by far the best (Broadcom BCM6348S), and tweakable in debug mode with custom firmware available (DGTeam). The v1, v2 and v3 (Texas Instruments) were apparently OK (I know the v3 was quite solid) while not tweakable, did have custom firmware (I know v3 did).

[Weaver]

I used a Netgear DG834v4 on good lines and found it to be great but never tested one on my own ultra-long line, and that would have been the real serious test. I have never used any of the with-wireless ones.

I used the v1-v3 on my own long line and especially the v3 was utterly superb. It synced high, was ultra aggressive with low snrm and hung on really well with very few resynchs. The v2 and v3 were a little better than the v1 I think.

[burakkucat]

If I can recall correctly, the Netgear DG834Gv5 was the worst out of all the DG834Gs (it had Conexant chipset).

You do, indeed, remember correctly. 

Conexant 

[re0]

I used the v1-v3 on my own long line and especially the v3 was utterly superb. It synced high, was ultra aggressive with low snrm and hung on really well with very few resynchs.

I think the v3 sync'd a bit lower for myself, but it was probably just as stable as the v4. Texas Instruments chipsets weren't actually that bad from the limited sample I had. In fact, in some cases I would say pretty much on par with Broadcom.

Conexant 

My thoughts exactly. Probably comparable to MediaTek. Can't say I've used either the v5 or a MediaTek-based modem, but Infineon chipsets that were used in some of the TalkTalk modem-router combos (D-Link, AFAIK) back in the ADSL days were certainly bad enough (at least on my line).