Mauricio
aka CigaR
Premium
join:2008-12-06
| reply to Bicephale
Re: In-depth analysis of line stats
said by Bicephale  :Personally, i "chop the top" in order to make the lesser details more visible... If, per say, you manually adjust the height to fifty RxCRC errors per sample then this should preserve most of the curve's shape while the explosive bursts would remain visible as a "saturated" section. I will keep that in mind. 
Just added some RF chokes and the connection seems much more stable. CRC count is very low and SNR is steady at 7-8 dB. Could be a coincidence though. Only time will tell. |
|
Bicephale
join:2005-09-24
 ·TekSavvy Solutions..
| Should you wish to collect "strong" evidence nothing would
please me more than to assist by processing SpreadSheet
records spanning over days, if not weeks!... I suppose two
instances of the John V.W. Lowen 3rd-party utilities will be
necessary to collect complete data: 'RouterStats' for your
Error Rate aka "noise" counters and 'RouterStats-Lite' for
your SNR Margins and BitRates... I must point out that i've
never got as far as to manage to make the logging feature
work for me, though. In any case, if you're up to it, it's no
big issue to combine records as long as there's some time
reference. One requirement which helps greatly with 24 h
records is to begin at least a few minutes before midnight
and to stop at least a few minutes past midnight some day
after. That way every daily record overlaps completely so
to ease synchronization in a SpreadSheet for the purpose
of later curve "averaging" (to enhance daily patterns)...
Of course, nobody knows what's the right number of turns
around the ferrite core so that means you must start with
a given value and explore other configurations from there.
For example, lets suppose ten turns is chosen as the basic
startup value. One would collect a noise curve for at least
a couple hours (to give enough time for the noise to knock
down fragile bins)... Then, he would change the number of
turns and do another test. Unfortunately, it's impractical
to aim at middle values in an attempt to shorten the whole
procedure because of the manipulations this would require:
10 turns at 1st, 15 turns turns out bad, 5 tuns not so bad,
7 turns not so bad, 12 turns just fine... In real life the way
to do this is to begin with a minimal number of turns and to
add one turn at the end of every test. So, if you're going
to proceed meticulously, starting at 5 turns and ending at
15, then you'll need at least 24 h non-stop if each test will
last 2 hours and there's an 11 minutes buffer between them
to account for unforeseable events... Eating, for example!
Evidently, the meticulous method is prohibitive if you're in
a hurry...
 |
|
Mauricio
aka CigaR
Premium
join:2008-12-06
| That's very considerate of you. As you said, patience is a must for this kind of thing and I might not be able to fully test different configurations over the next few days (lots of school work). I will however keep you updated over my findings if the occasion arises.
Here's a quick snapshot of the last 2 hours or so. CRC scale is 30 and SNR scale is 20. It's much better than before. But perhaps the line is not as noisy tonight. It's difficult to say though because the CRC error count is somewhat low at 121 over 2 hours. Yet Impulse Noise Comp. Tones (37) were quite active as per the stats from the modem. Could it be that the RF chokes are actually doing their job ? |
|
Bicephale
join:2005-09-24
·TekSavvy Solutions..
| I have an on-going test which ends this week... If you wish
i'll remove my RF Choke to provide you with a "before"/"after"
comparison the next week-end or so. This should prove quite
interesting as i haven't revisited that issue for a long time!
 |
|
Mauricio
aka CigaR
Premium
join:2008-12-06
1 edit | reply to Mauricio
Just thought I'd share an interesting graph depicting the already-known and intimate relationship between CRC errors and SNR. They seem to complement each other quite nicely. |
|
dslrocker3
join:2002-05-26
Toronto
1 edit | While this is generally true, I have personally seen stuff on my line where connections with lower signal to noise ratios had less errors. |
|
Oinktastic
join:2005-08-24
Scarborough
·TekSavvy Solutions..
 ·3 Web
| said by dslrocker3 :While this is generally true, I have personally seen stuff on my line where connections with lower signal to noise ratios had less errors. +1 |
|
Mauricio
aka CigaR
Premium
join:2008-12-06
| reply to dslrocker3
said by dslrocker3 :While this is generally true, I have personally seen stuff on my line where connections with lower signal to noise ratios had less errors. That's interesting. Was it a modem in specific ? |
|
dslrocker3
join:2002-05-26
Toronto
4 edits | indirectly, yes. directly, no. There are many line factors that can influence noise and different modems handle noise and the types of noise differently.
Also, we need to rememmber that signal to noise ratio is just that - a ratio. A low number does not mean that there is a lot of noise. It could just mean that there is very little signal. Also, a connection with a very high signal to noise ratio can still have a lot of noise but has signal levels that are many times higher than the noise level. A line with a high signal to noise ratio could also be a combination of very little signal but low noise as well.
For example, it is very possible to have a clean wiring pair that has a weak signal. This could cause the low signal to noise ratio but a low noise amount would result in few crc's and fec's. |
|
Mauricio
aka CigaR
Premium
join:2008-12-06
| Makes senses. But there has to be a way to determine the actual noise level at any given time. The 2Wire seems to report a lot of data pertaining to that, such as impulse noise characteristics. The goal would be to compare such stats with SNR and actual errors (FEC and CRC) to portray a more defined picture of what is really happening and why. That would be interesting. I was browsing the 2Wire DSM Elements and Line Statistics and there was lots of stats that haven't been looked at and monitored. |
|
Bicephale
join:2005-09-24
·TekSavvy Solutions..
| reply to Mauricio
%20.GIF "24403 bytes")
ST516v6 - Preliminary Examination (2009-Oct, 1-20) |
Hi Mauricio,
Ma> Was it a modem in specific?
One could even talk about specific FirmWares on the very same unit...
I'm not done yet processing my last record but here's one case where
the error counter(s) slope (not the error rate) sure appeared to follow
the SNR Margin indeed! Was it because of the higher SNR Margin that
i had less noise or was it a consequence of the lesser noise when the
SNR Margin happened to be higher?... The existance of a relationship
seems evident to me but i still wonder which comes 1st.
 |
|
Mauricio
aka CigaR
Premium
join:2008-12-06
| reply to Mauricio
It is my understanding that, typically, the signal is constant. It is sent at a certain power and attenuation ultimately lowers it. From that point it hovers around that value. The signal being relatively constant, the noise becomes the only variable. If the noise increases, errors tend to increase also. |
|
Bicephale
join:2005-09-24 | That's how i see it and that's why i wanted you to have a tool which can
monitor "noise" after i read that you were going to evaluate RF Choking...
 |
|
Mauricio
aka CigaR
Premium
join:2008-12-06
| said by Bicephale :That's how i see it and that's why i wanted you to have a tool which can monitor "noise" after i read that you were going to evaluate RF Choking... Just reasoning for the sake of it lol..  I am preparing another type of monitoring. Stay tuned for it ! |
|
Bicephale
join:2005-09-24 | I'm all ears, you have my undivided attention!
 |
|
Mauricio
aka CigaR
Premium
join:2008-12-06
| said by Bicephale :I'm all ears, you have my undivided attention! Don't know if this makes any sense but here it goes.
I am trying to find a pattern between inpulse noises and CRC errors. The 2Wire modem reports many stats about impulse noises and I found two of them to of interest. One is called Average Period of Inpulse Noise and the other is Standard Deviation Period of Impulse noise. Now, I do not know what they actually mean, but I decided to monitor them nonetheless.
So basically, there are 3 graphs ; 2 monitoring impulse noises and 1 for CRC errors. This is very experimental and could end up meaning nothing after all. At first, I was expecting a "normal" graph for the impulse noises. But somehow, the graphs contain negative values and it almost looks like a random signal with its varying amplitude and period.
Since I have not monitored SNR during that time lapse, it is hard to interpret those graphs. That was a big mistake because the modem lost sync and reconnected at a lower profile speed which evidently increased the SNR; thus making the line more resilient to noise.
So there it is. Still haven't configured out the right variables to monitor this case. This was an improvised approach. |
|
Bicephale
join:2005-09-24
·TekSavvy Solutions..
| I'm only familiar with the RxCRC curves. If the sampling period is
fifteen seconds and there's an RF Choke then your 2Wire seems
as noisy as my Thomson STs at times... Lets add a few turns!
|
|
morphector
join:2009-04-06
Jonquiere, QC
| reply to Mauricio
My line analysis
I'm about 2.6km away from CO.
Havent checked my line since 2-3 months ago, I'll surely have to check or repair something because before it was right around 80-85% and now download is at 100% but it doesn't produce too much errors(yet), it's on a 2wire 2700
A line I would die to get. 500m from slam on a dsl2+ equipement with an inside terminal.. you can't really get better than that.
 |
|
Bicephale
join:2005-09-24
1 edit | Do you have some long-term level-3 snapshots we can admire?
Oups! Level-2, sorry...
 |
|
morphector
join:2009-04-06
Jonquiere, QC | is that what you're looking for?
I know it's a shame I've got errors on my line but I haven't checked it since 2-3 months so I'll probably repair it within a week |
|
