Diagnosing a strange broadband problem - wide variety of sync speeds

Discussion in 'Broadband' started by NY, Jul 15, 2014.

  1. NY

    NY Guest

    I asked to help someone whose broadband had stopped working. The DSL light
    on their router came on but the internet light never came on. Sadly this
    make of router had no logging facility to show how far it was getting - eg
    whether it was even trying to log on using the ADSL username and password or
    whether it was failing to authenticate with those details, and it didn't
    report attenuation, noise margin or sync speed. Pretty useless to have a
    router that either works or doesn't but gives no clues as to why it doesn't
    work :-( After checking for wiring errors (eg unfiltered phone) and trying
    in the test socket of the master socket, I checked what happened with my
    known-good router.

    And that exhibited a strange fault: occasionally it would successfully
    connect, but many times it wouldn't. I kept unplugging the DSL cable to
    force it retry the connection. And there was a correlation between DSL sync
    speed and ability to connect: if the router happened to sync at a fairly
    slow speed (eg 1 Mbps) it would connect fine. If it happened to sync at a
    higher speed (eg 2 or 3 Mbps) it would fail. Attenuation remained the same
    but noise margin was worse as sync speed increased.

    Can anyone think of a line fault that would cause a router to sync at a wide
    variety of different speeds and only to connect successfully if it happened
    to sync at a slow speed? Wouldn't you expect a router always to sync at the
    same speed +/- a fairly small latitude, rather than trying speeds as far
    apart as 1 Mbps and 4 Mbps?

    Intriguingly when a BT Openreach engineer was called to investigate, he
    diagnosed a fault in the customer's router rather than a line fault - so
    maybe the behaviour of my test router was a red herring. I checked my router
    on my own line later and it always synced at the same speed, give or take a
    small latitude, so it hadn't started playing silly buggers.
     
    NY, Jul 15, 2014
    #1
    1. Advertisements

  2. NY

    Peter Able Guest

    It seems quite rational that a low negotiated sync speed would increase
    the likelihood of making a connection, so perhaps the question is
    primarily "why the differing sync speeds?". This is where a
    differential-input oscilloscope comes in handy. It would clarify if the
    problem is (intermittent) HF noise on the line.

    PA
     
    Peter Able, Jul 15, 2014
    #2
    1. Advertisements

  3. NY

    NY Guest

    Yes, I sort-of asked that question (although in a rather roundabout way) in
    my question: why would a router sync at a variety of different speeds rather
    than always syncing at the same speed on successive connection attempts.

    I realise that if the router happens (for whatever reason) to sync at too
    high a speed for the noise on the line, the connection will probably fail.

    A USB oscilloscope that plugged into a laptop and displayed traces on the
    screen would be lovely (please, Santa!) but they are horrendous amounts of
    money for the few occasions when I'd use one.
     
    NY, Jul 15, 2014
    #3
  4. I wouldn't make that my tool of choice.
     
    The Natural Philosopher, Jul 15, 2014
    #4
  5. NY

    Peter Able Guest

    I happen to have one and so can simultaneously observe A, B and A-B, but
    a simple oscilloscope transformer-coupled to the line will suffice.
    Sometime I must try transformer-coupling a spectrum analyser to my line.

    It's all down to the perk of having an office facing onto the corridor
    between the firm's calibration department and the Biffa bin ;-}}

    PA
     
    Peter Able, Jul 15, 2014
    #5
  6. NY

    Graham J Guest


    Normally, if the sync speed increases you would see the SNR margin
    reduce, to settle at about 6dB - depending on whether the service is
    ADSL, ADSL2, or ADSL2+.

    The converse occurs when there is noise on the line. The SNR margin
    increases and the sync speed decreases.

    This is indeed what you saw. The implication is that at the higher
    speeds the noise margin was too small to allow reliable communication,
    and the login process may have been corrupted by the noise.

    It follows that the router should have shown you errors. This is where
    the choice of router as a diagnostic tool becomes important. Some will
    only report SNR margin and loop attenuation for the downstream
    direction; others will report it in both directions.

    Some routers will report uncorrected errors (which require that the data
    packet is retranmitted) and others will also report corrected errors
    (where the Forward Error Correction [FEC] mechanism allows the router to
    identify and correct errors).

    For example the (no longer current) Vigor 2600 and 2800 series routers
    report both corrected and uncorrected errors. I have seen cases where
    the corrected error count rises to about 300 errors per second, and
    stays there; but a re-sync causes the corrected count to drop back to
    nearly zero. In this instance I suspect the processor in the router
    cannot keep up, thus while correcting one packet it misses the start of
    the next packet and so has to correct that packet also. A more modern
    router (the TP-Link TD-W8960N) on the same line shows a high corrected
    error count at times, (maybe 1,000 errors per second) but it always
    drops back to nearly zero as the intermittent noise fades away.

    By contrast the more modern Vigor 2820 and 2830 routers show only
    "errors" and do not identify the cause.

    I think the exchange equipment (DSLAM or whatever) can only recognise
    uncorrected errors - certainly when the router shows a high corrected
    error count the ISP says there are no errors reported at their end.

    When I've watched Openreach engineers monitor a line, they generally
    ignore FEC errors.

    I've never seen the circumstance you describe where a router behaves
    erratically at a customer site yet is reliable at your home site without
    also the Openreach engineer being able to see a problem at the customer
    site. Did you actually watch the engineer working and ask him to
    explain his conclusions? I find this approach is absolutely vital to
    understanding the nature of the problem, both for you and sometimes also
    for the Opnreach engineer!
     
    Graham J, Jul 15, 2014
    #6
  7. NY

    Martin Brown Guest

    There seems to be a trend towards magic black box ADSL zero diagnostics
    routers at the consumer end of the market these days :(
    That is normal as is getting a worse SNR/connect speed after sunset
    (which might be helpful here). You need to track SNR against connect
    speed to see where you stand. A reliable connection should sit at 6dB
    and if you are lucky on fastpath. Rural lines tend to be 9dB and
    interleaved. I favour routerstats light as a first test - there is a
    German one even better but it doesn't work with any modem that I own.
    If there is an intermittent fault like a dry joint where the thing is
    capable of 4Mbps for short intervals long enough to calibrate the link
    but not to sustain meaningful data transfer then this is sort of what
    you expect. But then what should happen is that you enter a spiral of
    speed downgrades due to disconnects until the SNR is sufficiently high
    to maintain a reliable connection with a floor at about 256k.

    A power cycling router reboot is generally a bit more consistent that
    plugging and unplugging the external connection. Some brands are prone
    to getting themselves into strange states after extended periods...
    Local intermittent interference of some sort would be my guess. I don't
    understand why you are not seeing successive retries drop in speed until
    a reliable connection is established (albeit at pathetic data rates).
    That is certainly what happens on lines with dodgy rectifying aluminium
    wire with corroded oxide connections round here.
     
    Martin Brown, Jul 16, 2014
    #7
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.