Fun with ADSL and snmp..and a question..on noise.

Having Finally changed ISP and got an SNMP enabled router so I can be
totally nerdy and peer at MRTG graphs, I decided to also peer at things
like noise and link speed and stuff like that, to develop some long term
graphs of noise versus time.

This is working quite well, BUT there is a slight issue. SNR ratio not
only varies with the noise itself, but with the link speed. If that
drops snr goes up again, to stabilise (in my case, cos I suspect that's
the way the exchange is set) at around 10.5 dB.

What is the formula that more or less represents 'total noise' as a
function of frequency of link, and signal to noise ratio? is it
frequency or square root frequency? Square root springs to mind, so for
a constant SNR every 3dB gain in SNR could be translated into a doubling
of sync speed at the same SNR? that doesn't sound right..

PeeGee wrote:
> The Natural Philosopher wrote:
>> Having Finally changed ISP and got an SNMP enabled router so I can be
>> totally nerdy and peer at MRTG graphs, I decided to also peer at things
>> like noise and link speed and stuff like that, to develop some long term
>> graphs of noise versus time.
>>
>> This is working quite well, BUT there is a slight issue. SNR ratio not
>> only varies with the noise itself, but with the link speed. If that
>> drops snr goes up again, to stabilise (in my case, cos I suspect that's
>> the way the exchange is set) at around 10.5 dB.
>>
>> What is the formula that more or less represents 'total noise' as a
>> function of frequency of link, and signal to noise ratio? is it
>> frequency or square root frequency? Square root springs to mind, so for
>> a constant SNR every 3dB gain in SNR could be translated into a doubling
>> of sync speed at the same SNR? that doesn't sound right..
>
> The figures I have seen with my router(s) indicated the trade-off is:
> 6dB change doubles (or halves) sync speed
> (61dB attenuation :-( )though there other factors, so that isn't exact.
>
OK. That is what my maths said, but it seemed unbelievable.


I've already noticed a couple of dB ripple per day as night falls and
the ionosphere or whatever it is does it thing, plus an interesting rise
in noise at 6pm..presumably everybody using more electrical gadgetry.
The the noise fades out a bit after midnight.

But this gets swamped by the change when the router resynchs. So I want
to remove that factor.

If anyone else is interested in linux/apache/php/snmp monitoring tools,
feel free to ask for source code etc etc.

"The Natural Philosopher" <(E-Mail Removed)> wrote in message
news:hb99us$tao$(E-Mail Removed)...
> Having Finally changed ISP and got an SNMP enabled router so I can be
> totally nerdy and peer at MRTG graphs, I decided to also peer at things
> like noise and link speed and stuff like that, to develop some long term
> graphs of noise versus time.
>
> This is working quite well, BUT there is a slight issue. SNR ratio not
> only varies with the noise itself, but with the link speed. If that drops
> snr goes up again, to stabilise (in my case, cos I suspect that's the way
> the exchange is set) at around 10.5 dB.
>
> What is the formula that more or less represents 'total noise' as a
> function of frequency of link, and signal to noise ratio? is it frequency
> or square root frequency? Square root springs to mind, so for a constant
> SNR every 3dB gain in SNR could be translated into a doubling of sync
> speed at the same SNR? that doesn't sound right..

I think you're confusing cause and effect here.

If the noise level increases, then the SNR Margin increases to accommodate
the higer noise level. The signal-to-noise ratio of the received signal has
actually reduced, meaning that there is less useful information in the
signal, which means a reduction in the capacity of the link to carry data.
The link then renegotiates its speed, which reduces by around 800kbits/sec
for a 3dB increase in SNR Margin.

--
Graham J

Graham J wrote:
> "The Natural Philosopher" <(E-Mail Removed)> wrote in message
> news:hb99us$tao$(E-Mail Removed)...
>> Having Finally changed ISP and got an SNMP enabled router so I can be
>> totally nerdy and peer at MRTG graphs, I decided to also peer at things
>> like noise and link speed and stuff like that, to develop some long term
>> graphs of noise versus time.
>>
>> This is working quite well, BUT there is a slight issue. SNR ratio not
>> only varies with the noise itself, but with the link speed. If that drops
>> snr goes up again, to stabilise (in my case, cos I suspect that's the way
>> the exchange is set) at around 10.5 dB.
>>
>> What is the formula that more or less represents 'total noise' as a
>> function of frequency of link, and signal to noise ratio? is it frequency
>> or square root frequency? Square root springs to mind, so for a constant
>> SNR every 3dB gain in SNR could be translated into a doubling of sync
>> speed at the same SNR? that doesn't sound right..
>
> I think you're confusing cause and effect here.
>
> If the noise level increases, then the SNR Margin increases

No, it falls. Signal to noise is signal/noise. If noise gets bigger, S/N
gets smaller, They teach this at primary school. It's called
mathematics. You may even have heard of it?

> to accommodate
> the higer noise level.

No it reflects the noise to signal ratio. If the noise goes up the
signal to noise goes down, (unless the signal goes up as well), or the
in band noise goes down. Using less bandwidth (slower connect) improves
SNR because simply there is more band used per unit information transmitted.

> The signal-to-noise ratio of the received signal has
> actually reduced,

Increased actually.

> meaning that there is less useful information in the
> signal,

BECAUSE there is less information in the signal.

which means a reduction in the capacity of the link to carry data.

You got that bit right tho.

> The link then renegotiates its speed, which reduces by around 800kbits/sec
> for a 3dB increase in SNR Margin.
>

I have found it more than that. At my sort of 4Mbps raw speed.

Anyway, I think that noise power is generally a function of bandwidth,
noise voltage of root bandwidth. But that may not answer my original
question.

Which is in essence how to estimate total noise (variation) from a
function of SNR and link speed.

Well, an hour later I found this which implies that 10 log 10 of 2 to
the power of bits per second equals the total noise for a fixed noise
margin and error rate..for reasonably large values of b(i)

10log10[2^b(i)] = SNR (dB) (plus noise margin plus other fixed stuff
like attenuation etc etc)

My maths is rustier than Labours policies, but to me this implies that a
6dB gain in TOTAL signal to noise implies a doubling in bits per second.


so a 3db improvement in SNR would be a 41.4% gain.

Which would take a 2Mbps circuit to 2.8Mbps. Or in my case on a line at
9dB target as compared with my last line on a 6dB target, my speed now
averages about 4Mbps, and it used to do over 5.5Mbps..albeit bloody
flakily. ;-)

Anyway, it LOOKS like the answer is that if I take reported SNR and
subtract 10log10(synch speed) it should represent the *total* noise on
the line, plus a load of constants I don't care about.

So, if the synch speed is constant, the bigger the reported SNR is, the
less noise there is and the biger the *real* SNR is. So that works. The
faster I connect at a given total noise, the more I subtract ..showing
that to connect faster at a given margin, there has to be less noise on
the line. Yup. I think I finally got there.

REAL noise(db) = (REPORTED SNR (dB)) - 10* Log10(synch speed) +- (some
constants).

On Sun, 18 Oct 2009 09:07:48 +0100, Peter Crosland ate alphabet spaghetti
and shat out:
.....
> I am simply curious to know what you investigations are intended to
> prove or disprove. Whilst I don't disagree with your findings does it
> actually make any practical difference in everyday usage?

Indeed, the question on my mind was 'And the point being?'
--
political correctness: The safety net protecting deaf blind disabled
ethnic minority gays & lesbians with odd religious beliefs from reality

Peter Crosland wrote:
> "The Natural Philosopher" <(E-Mail Removed)> wrote in message
> news:hbdfnr$hcg$(E-Mail Removed)...
>> Graham J wrote:
>>> "The Natural Philosopher" <(E-Mail Removed)> wrote in message
>>> news:hb99us$tao$(E-Mail Removed)...
>>>> Having Finally changed ISP and got an SNMP enabled router so I can be
>>>> totally nerdy and peer at MRTG graphs, I decided to also peer at things
>>>> like noise and link speed and stuff like that, to develop some long term
>>>> graphs of noise versus time.
>>>>
>>>> This is working quite well, BUT there is a slight issue. SNR ratio not
>>>> only varies with the noise itself, but with the link speed. If that
>>>> drops snr goes up again, to stabilise (in my case, cos I suspect that's
>>>> the way the exchange is set) at around 10.5 dB.
>>>>
>>>> What is the formula that more or less represents 'total noise' as a
>>>> function of frequency of link, and signal to noise ratio? is it
>>>> frequency or square root frequency? Square root springs to mind, so for
>>>> a constant SNR every 3dB gain in SNR could be translated into a doubling
>>>> of sync speed at the same SNR? that doesn't sound right..
>>> I think you're confusing cause and effect here.
>>>
>>> If the noise level increases, then the SNR Margin increases
>> No, it falls. Signal to noise is signal/noise. If noise gets bigger, S/N
>> gets smaller, They teach this at primary school. It's called mathematics.
>> You may even have heard of it?
>>
>>> to accommodate the higer noise level.
>> No it reflects the noise to signal ratio. If the noise goes up the signal
>> to noise goes down, (unless the signal goes up as well), or the in band
>> noise goes down. Using less bandwidth (slower connect) improves SNR
>> because simply there is more band used per unit information transmitted.
>>
>>> The signal-to-noise ratio of the received signal has actually reduced,
>> Increased actually.
>>
>>> meaning that there is less useful information in the signal,
>> BECAUSE there is less information in the signal.
>>
>> which means a reduction in the capacity of the link to carry data.
>>
>> You got that bit right tho.
>>
>>> The link then renegotiates its speed, which reduces by around
>>> 800kbits/sec for a 3dB increase in SNR Margin.
>>>
>> I have found it more than that. At my sort of 4Mbps raw speed.
>>
>> Anyway, I think that noise power is generally a function of bandwidth,
>> noise voltage of root bandwidth. But that may not answer my original
>> question.
>>
>> Which is in essence how to estimate total noise (variation) from a
>> function of SNR and link speed.
>>
>> Well, an hour later I found this which implies that 10 log 10 of 2 to the
>> power of bits per second equals the total noise for a fixed noise margin
>> and error rate..for reasonably large values of b(i)
>>
>> 10log10[2^b(i)] = SNR (dB) (plus noise margin plus other fixed stuff like
>> attenuation etc etc)
>>
>> My maths is rustier than Labours policies, but to me this implies that a
>> 6dB gain in TOTAL signal to noise implies a doubling in bits per second.
>>
>>
>> so a 3db improvement in SNR would be a 41.4% gain.
>>
>> Which would take a 2Mbps circuit to 2.8Mbps. Or in my case on a line at
>> 9dB target as compared with my last line on a 6dB target, my speed now
>> averages about 4Mbps, and it used to do over 5.5Mbps..albeit bloody
>> flakily. ;-)
>>
>> Anyway, it LOOKS like the answer is that if I take reported SNR and
>> subtract 10log10(synch speed) it should represent the *total* noise on
>> the line, plus a load of constants I don't care about.
>>
>> So, if the synch speed is constant, the bigger the reported SNR is, the
>> less noise there is and the biger the *real* SNR is. So that works. The
>> faster I connect at a given total noise, the more I subtract ..showing
>> that to connect faster at a given margin, there has to be less noise on
>> the line. Yup. I think I finally got there.
>>
>> REAL noise(db) = (REPORTED SNR (dB)) - 10* Log10(synch speed) +- (some
>> constants).
>
> I am simply curious to know what you investigations are intended to prove or
> disprove. Whilst I don't disagree with your findings does it actually make
> any practical difference in everyday usage? The reason I say this is that,
> as far as I can see, in normal circumstances the throughput of a broadband
> connection between the user and the exchange is determined almost entirely
> by the sync speed established at the time when the connection is made. The
> only other significant factor is the degree of conjestion caused by
> contention. The BT system has such a large amount of tolerance built in that
> quite large variations in SNR once the reconnection is made seem to make
> little or no difference to the line throughput. For example my sync speed is
> currently 4,000 and results in a stable speed set by BT, and notified to my
> ISP, of 3,500. The latter limits the maximum speed my ISP sends data to my
> line. Looking at the data on my line the connection was made 9 days ago and
> at that time the SNR was 9.1dB. During the hours of darkness the
> instantaneous SNR drops down to 3dB but the line throughput measured by my
> ISP's speedtester remains at around 3,250. During the last 72 hours there
> have been 494 CRC errors, and 2109962 FEC errors the latter obviously
> corrected without the need for the data to be resent. I would be interest to
> hear you comments.
>
> Peter Crosland
>
>
Pure curiosity Peter. I am sim0ply curious...too!

I just want to see how the noise varies with time.

One thing I did pick up from the web research is that the major noise
contribution is in fact other broadband users down the same cabling route.

The other is LW/MW radio.

Plus pure thermal noise - that is of course fairly large!.

And, if something does go wrong, and I have the stats to show it, it
might prod BT etc into action if I can say 'look, it rained then, and
the noises spiked 3dB'

or also in my case, tie it in to central heating operation and the like.

Peter Crosland wrote:
> Pure curiosity Peter. I am simply curious...too!
> >
> > I just want to see how the noise varies with time.
> >
> > One thing I did pick up from the web research is that the major noise
> > contribution is in fact other broadband users down the same cabling
route.
> >
> > The other is LW/MW radio.
> >
> > Plus pure thermal noise - that is of course fairly large!.
> >
> > And, if something does go wrong, and I have the stats to show it,
it might
> > prod BT etc into action if I can say 'look, it rained then, and the
noises
> > spiked 3dB'
> >
> > or also in my case, tie it in to central heating operation and the
like.


>Noted and thanks. In my experience trying to get BT to take action is
very
>hard. I would suggest concentrating on the things you can alter yourself
>such as router modem that can cope with noisy and variable line
conditions,
>connecting to the master socket, severing the bell wire or fitting an
Iplate
>and using top quality filters plus screened lead from modem/router to the
>faceplate. The former things can often increase the sync speed by 30%.
>Good
luck with your experiments.

Peter Crosland


Oh..I don't have a problem beyond being vaguely interested as to why
things change..I get at the moment a BRAS of 3Mbps. But I have had it as
high as 4Mbps.

But I have changed ISP AND phone line, and it appears target noise
margin, at the exchange.

This line synchs at 9dB - the old one synched at 6dB...

and none of the above applies. I have a faceplate filter. No bell wires
are used as I have a PABX on the back of it. The interface is
essentially two wire to that IIRC.

the top quality filter did nothing for my synch speed, but it did reduce
the hiss on voice calls..the PABX I suspect does not like ADSL up its
snout - even the smallest amount. I've shoved a toroid on its input lead
there and got it down even more.

No, the whole thing is just about seeing that my synch speed and noise
are variable, and wanting to pull out the fundamental quantity - noise
induced on the line - to see what the pattern was.

Because sometimes code is amusing to write, and the results might be
interesting in some way. Or not.

"The Natural Philosopher" <(E-Mail Removed)> wrote in message
news:hbdfnr$hcg$(E-Mail Removed)...
> Graham J wrote:
>> "The Natural Philosopher" <(E-Mail Removed)> wrote in message
>> news:hb99us$tao$(E-Mail Removed)...
>>> Having Finally changed ISP and got an SNMP enabled router so I can be
>>> totally nerdy and peer at MRTG graphs, I decided to also peer at things
>>> like noise and link speed and stuff like that, to develop some long term
>>> graphs of noise versus time.
>>>
>>> This is working quite well, BUT there is a slight issue. SNR ratio not
>>> only varies with the noise itself, but with the link speed. If that
>>> drops snr goes up again, to stabilise (in my case, cos I suspect that's
>>> the way the exchange is set) at around 10.5 dB.
>>>
>>> What is the formula that more or less represents 'total noise' as a
>>> function of frequency of link, and signal to noise ratio? is it
>>> frequency or square root frequency? Square root springs to mind, so for
>>> a constant SNR every 3dB gain in SNR could be translated into a doubling
>>> of sync speed at the same SNR? that doesn't sound right..
>>
>> I think you're confusing cause and effect here.
>>
>> If the noise level increases, then the SNR Margin increases
>
> No, it falls. Signal to noise is signal/noise. If noise gets bigger, S/N
> gets smaller, They teach this at primary school. It's called mathematics.
> You may even have heard of it?

I beg to differ.

SNR means signal to noise ratio - as you have defined.

SNR Margin is the difference between one SNR value and another.

As it applies to ADSL, the SNR Margin is the difference between the actual
SNR and the SNR required to run at a specific speed. For example, if your
line needs 35dB SNR to run at 8Mbps, and the actual line SNR is 41dB, then
the SNR Margin is 6dB.

Suppose the noise level increases by 6dB and the signal level remains the
same. This means that the SNR drops from 41dB to 35dB. The DSLAM and
router together will try to maintain the SNR Margin at the default value of
6dB, so the available SNR to carry data reduces to 29 dB. At this lower SNR
the achievable speed is lower. Bob Pullen elsewhere in this thread suggests
that a 3dB change in SNR relates to about 800kbits/sec.
http://www.kitz.co.uk/adsl/linestats.htm suggests 1dB relates to
450kbits/sec (i.e 3dB implies 1350kbits/sec)

If the noise level varies the router and DSLAM will re-negotiate the link
speed in order to maintain the SNR margin. If this means that
re-negotiations happen too frequently then the SNR Margin is increased, with
the intention of improving the reliability of the link at the expense of
speed.

--
Graham J

Meanwhile, at the uk.telecom.broadband Job Justification Hearings, The Natural
Philosopher chose the tried and tested strategy of:

> No, the whole thing is just about seeing that my synch speed and noise
> are variable, and wanting to pull out the fundamental quantity - noise
> induced on the line - to see what the pattern was.
>
> Because sometimes code is amusing to write, and the results might be
> interesting in some way. Or not.

I've been doing this for a couple of years. My summary of it would be: noise
margin increases during the day and decreases at night [but only by 1dB]. Be
seem to boost the downstream output power at night to compensate [by about 3dB].
Attenuation changes very rarely. Upstream output power has never changed.

Be* added an option earlier in the year to choose a profile to bias the sync
process towards speed or stability. I chose speed, and I'm currently synced at
10.5Mbps down, compared to 7.2Mbps this time last year.

--
<http://ale.cx/> (AIM:troffasky) ((E-Mail Removed))
12:05:37 up 3:11, 3 users, load average: 0.45, 0.34, 0.25
"Stupid is a condition. Ignorance is a choice" -- Wiley Miller

"Peter Crosland" <(E-Mail Removed)> wrote in message news:-8-(E-Mail Removed)...
> "The Natural Philosopher" <(E-Mail Removed)> wrote in message news:hben9o$k09$(E-Mail Removed)...
>> Peter Crosland wrote:
>> > Pure curiosity Peter. I am simply curious...too!
>> > >
>> > > I just want to see how the noise varies with time.
>> > >
>> > > One thing I did pick up from the web research is that the major noise
>> > > contribution is in fact other broadband users down the same cabling
>> route.
>> > >
>> > > The other is LW/MW radio.
>> > >
>> > > Plus pure thermal noise - that is of course fairly large!.
>> > >
>> > > And, if something does go wrong, and I have the stats to show it,
>> it might
>> > > prod BT etc into action if I can say 'look, it rained then, and the
>> noises
>> > > spiked 3dB'
>> > >
>> > > or also in my case, tie it in to central heating operation and the
>> like.
>>
>>
>> >Noted and thanks. In my experience trying to get BT to take action is
>> very
>> >hard. I would suggest concentrating on the things you can alter yourself
>> >such as router modem that can cope with noisy and variable line
>> conditions,
>> >connecting to the master socket, severing the bell wire or fitting an
>> Iplate
>> >and using top quality filters plus screened lead from modem/router to the
>> >faceplate. The former things can often increase the sync speed by 30%. Good
>> luck with your experiments.
>>
>> Peter Crosland
>>
>>
>> Oh..I don't have a problem beyond being vaguely interested as to why things change..I get at the moment a BRAS of 3Mbps. But I
>> have had it as high as 4Mbps.
>>
>> But I have changed ISP AND phone line, and it appears target noise margin, at the exchange.
>>
>> This line synchs at 9dB - the old one synched at 6dB...
>>
>> and none of the above applies. I have a faceplate filter. No bell wires are used as I have a PABX on the back of it. The
>> interface is essentially two wire to that IIRC.
>>
>> the top quality filter did nothing for my synch speed, but it did reduce the hiss on voice calls..the PABX I suspect does not
>> like ADSL up its snout - even the smallest amount. I've shoved a toroid on its input lead there and got it down even more.
>>
>> No, the whole thing is just about seeing that my synch speed and noise are variable, and wanting to pull out the fundamental
>> quantity - noise induced on the line - to see what the pattern was.
>>
>> Because sometimes code is amusing to write, and the results might be interesting in some way. Or not.
>
> Noted. Apart from obvious thing ling central heating thermostats and the like the only regular thing I have found affecting the
> SNR is medium wave interference after dark. Anything else is likely to be random and as such quite unpredicatable. You are correct
> that PABXs do not coexist with ADSL and if possible they should be on separate lines.
>
> Peter Crosland

Perhaps if the switch manufactures took more care in making POTS
trunks emulate the "hybrid" conditions of a standard telephone (country specific)
this problem wouldn't exist as much.

--
Graham.

%Profound_observation%