8900AX-2400 S/N continuosly drops over time

[gatekeeper]

mognog,

Concerning your first comment, I'm assuming, from the pictures he's posted, that ardsar is using a separate third-party piece of software to analyse the performance of his line. Like you, I've never heard of any sort of graphical analyser being available in the Billion GUI. So, playing devil's advocat here, I would ask: which came first? Did ardsar first notice his SNR problem by noting changes in the SNR as presented in the normal GUI (the 8900's browser inteface), or did he go straight to his app and saw the weird SNR effect there? If the latter, then what the app is showing might not be trustworthy.

You yourself, mognog, have presented us with just one snapshot (spot value) of your SNR. The thing to do is to have a look again at that value at a completely different time of day and to repeat the exercise for at least another day. Then let us know whether your SNR has significantly moved in that period, eg by 3dB or more.

It's early days yet but ardsar seems to be the only person so far who has observed this odd SNR behaviour with the 8900.

[mognog]

Thanks gatekeeper. I shall send another snapshot tomorrow when I get time and see if there is any change.

[gatekeeper]

Do look at least four or five times, sometimes in the morning and other times at night. Make a note of the SNR each time, and then in a day or two's time tell us the values you found.

[ardsar]

For your information, the software used to capture the line stats is called MyDSLStats and can be found https://www.mydslwebstats.co.uk/. It basically uses telnet to extract all the line stats every minute. Its a great tool for understanding your line stats and identifying issues with your line.

I was in discussions with Netgear for over 3 months on the issue, who in turn contacted Broadcom regarding this issue. After waiting for several weeks and only getting a suggestion from Broadcom that I should turn on the router when interference is at its highest, I gave up. I do think that the issue is with Broadcom and not Billion and also agree that for the majority, the 8900 will work perfectly fine. It might be down to the fact that the interference is frequency dependant meaning at night particular tones can not be used and as this interference is there for several hours, the bins are marked as unusable.

In answer to another question, yes the S/N would continue to drop over time, but at a reduced rate. The most significant drop happens over the first day - say 2 dB, but after a week this could be upto 4 dB. My line got banded when this was happening, hence the reason why my s/n at sync is around 8dB (capped at 22400).

I am currently back to the HH5a, which has a S/N of approx. 8dB each morning (drops to about 7dB in the evenings (after dark).

My concern is - why are the new broadcomm devices acting like this on my line?!? The 8800AXL was the best product I had, but that had an older generation Broadcom solution.

[mognog]

Ok, sorry to be brief but I don't have much time. Here are my stats for this morning that don't appear to have changed.

status-1.png

snr-1.png

[billion_fan]

Here are the screen grabs from the my test 8900ax-2400

The SNR is holding stable at around the 9db, (when I first setup the device yesterday it was 9.1db, now this morning it is 9.2db)

[ardsar]

I do suspect that its a combination of my line and Broadcom implementation. Noticed that you have an up time of 12 days - the most significant decrease in S/N happens over the first day or two, if its going to happen at all. Also, i'm guessing the problem will only be present if you suffer from REIN at night. If you do not see a drop in S/N during dark hours, then you will not see this problem at all. On the Billion 8800AXL i would usually see a 1 - 1.5dB drop during late evening / night but it would recover during daylight hours. The 8900 does not recover.

[ardsar]

I've found an old graph that clearly shows the problem. This is a comparison between the 8800AXL and the Netgear D7000. The start and end of the graph is using the 8800AXL - as you can see the S/N drops at night by around 1dB and recovers the next day. The D7000 was switched in around the 24th ish July and as you can see S/N started at 6dB but continually dropped. 3 line re-syncs happened which momentarily recovered the S/N to 6dB, but it then started to drop again. 8800AXL was then put back around 13th August.

The 8900AX performs the same as the D7000 on my line.

s_n billion&netgear.png

[gatekeeper]

(This is pre your latest graph).

ardsar,

Referring to the last paragraph in your latest posting above, are you saying that this problem was not present at all when you used the 8800AXL? On the 8800AXL, were you using ADSL or VDSL? Presumably, VDSL?

I must say that, based on the evidence you've presented, it does look as though the issue is being caused by the way in which the modem deals with noise and crosstalk.

Here are some thoughts for you to address. In offering these, please excuse any ignorance I might display about the detailed technical factors of VDSL operation, as to date I've no VDSL experience on which to rely, only ADSL. As before, I'm just trying to play devil's advocat in the hope of perhaps getting you to investigate something that you might be overlooking:

1. Just how long is your copper connection to the street cabinet? Unless you yourself already know the length, I believe the actual cable length is displayed when you run the ubiquitous Openreach online speed estimator (using tel no. rather than postcode). My understanding is that VDSL is suitable for copper loops of about 800m or less and that Openreach assumed at the outset that the average length in the UK would be 500m. Clearly, if someone's on 1km or more and the line is also subject to a high level of noise, performance could be iffy, to say the least.

2. Presumably, inside your house you have absolutely nothing else attached to the line?

3. I notice that Interleave Depths in VDSL are totally different nos. to those used in ADSL. Are you convinced that those showing in the Billion GUI are within range?

4. Presumably, the firmware version of the 8900 you're using is 2.50a.d32?

5. Is PhyR relevant for VDSL and, if so, are you using it? What about the other settings under Configuration > DSL, like Profile, Inner/Outer pair, Bitswap? Have you checked they're appropriately chosen?

6. Do you have a neighbour or a friend in the same street who's on an FTTC connection and who'd be prepared to temporarily use your 8900 to see whether the same phenomenon results?

7. How about giving us some screenshots of your actual Billion GUI xDSL Status page for the next couple of days, so that we can see, independently of your Webstats application, what's going on? Do this without Webstats running.

Again, I have to confess that, from all you've pointed out thus far, the issue does seem to point to a problem specifically in the modem section of the 8900. So, together with your finding re the Netgear D7000, I can quite understand why you suspect the Broadcom chippery as the culprit. But could this simply be something pertaining to your line that's out of range, as far as Broadcom performance goes?

(Just seen your latest submission, the old graph).

[gatekeeper]

Just visually analysed your latest graph, and I've looked back at the first graph you posted here.

In the latest graph, the 8800 part of the trace has a starting or target SNR of 7.5dB, with the margin dropping overnight to 6.5dB. So the excursion is 1dB. As regards the 8900's part, it's difficult for me and others to know the starting SNR of the 8900 trace in that same graph, because it attaches directly to that of the 8800, but if you reckon the 8900's starting SNR is 6dB, then over the first week it continuously drops to 4dB, an excursion of 2dB. You can then see it trying to recover to something around 5 or 6dB, but it then drops away again. By the end of the second week (or a bit more), it's dropped to 3.5dB.

Yep, the trend is indeed distinctly downward in the 8900 case. The 8800 trace and the 8900 trace are about as different from each other as chalk from cheese. (I note that the earlier graph, submitted some days ago, was on a daily scale, whereas this one is on a weekly scale).

Could it be that, with the 8900 and maybe also the Netgear D7000, Broadcom have made some further improvements, ie. have rendered faster speeds and therefore lower SNRs, but at the expense of a much longer training period? I mean, from the graph we've seen what happens over two weeks, but then what happens after three weeks, or even four? Does the SNR still continue downwards, or does it level out? I don't think you've had the 8900 long enough to know that yet, have you?