Using a 3Com/USR modem to test the phone line.

by Jeff Liebermann  Version 1.06

Trying to determine whether a POTS phone line is capable of doing 56K has been a never ending challenge.   3Com/USR modems have the ATY11 command which displays the compensated frequency response of the phone line and modem.  The results can be used to determine if the phone line is 56K capable.  Note that this response curve is the results of the modem adaptive equalizer's heroic attempt to obtain the best response from a POTS phone line and not the actual frequency response of the phone line.  Results will vary depending upon modem type and firmware.


How to use the ATY11 command

You do not need a 56K 3Com/USR modem to do this test.  A USR V.34 or 33.6 Sportster or Courier modem will work just fine.  However, non-USR modems will not work as they do not have the ATY11 command.
 

  • Use Hyperterminal or other terminal emulator to connect directly to your modem.  Setup for 57600 and to connect directly to whatever com port your modem is using.
  • Type:  AT<enter>.  The modem should return OK.  If it doesn't, try AT&F1 or AT&F to temporarily reset the USR modem to its defaults.  Don't bother proceeding until it belches OK.
  • Type:  ATDT 123-4567<enter> to dial your ISP.  Please replace the 123-4567 with your ISP's phone number.
  • You should get some kind of a login prompt.  You do not need to login.
  • Type:  +++  and wait a few seconds.  It should return OK after a second or so.
  • Type: ATY11<enter> and a two column table of frequencies and amplitudes should result.
  • Wait 30-45 seconds for the equalizers to settle down.  Then run: ATY11<enter>  for the final values.  3Com recommends 60 seconds, but I've found that 15 seconds is usually sufficient.
  • Type: ATH0 <enter> to hangup the phone line.
  • Save or print the final table of numbers.


    • You can plot your results online at http://www.frostbyte.com.au/public/support/linequality.html
    or you can download the Excel 97 spreadsheet I used to generate the following graphs here (80KB).  For Netscape, hold down the shift key when clicking the "here" to force download.  For MS Internet Exploder, right click and choose "Save target as..."

    What can go wrong?

    Some systems are set to hangup immediately if they see a +++ sent.  If this is the case, login with your network login and password.  You will get some bursts of crud ever few seconds.  Then do the +++ stuff.


    How to tell if you have a chance at 56K.

    Almost the entire 56K performance is determined by the frequency response between 2500Hz and 4000Hz.  An inability to pass these frequencies will result in horrible performance because the V.34/V.90 carrier frequencies attempt to use the highest possible carrier frequencies.  The following is a graph of various calls to Cruzio (a local ISP) from various modems and locations.  Note that each vertical division is 5db which is a fairly large amount of signal level variation (approx 1.8 times).

    The caller SteveP may have an SLC (Subscriber Line Concentrator) on his phone line from Boulder Creek.  Estimated wire line distance is about 3 miles to the CO (central office).  The SLC adds an additional A/D conversion into the downsteam path and mangled the frequency response.  This line has no chance whatsoever of obtaining 56K connectivity.  At 3750Hz, this line has 20db (10 times) more loss than the working lines.

    The caller Ken W. of Bear Creek may have an SLC.  His line obtains a 56K (V.90) connection, but had trouble maintaining connectivity with constant retrains.  I have no clue what is causing the slow rolloff.  Line length is about 3-4 miles.

    The caller Neil S of Branciforte can only do 26400 (to any ISP).  The frequency response and problems are similar to Ken W.  Line length is about 6 miles to the Santa Cruz CO.  Tests calling other ISP's resulted in almost identical frequency response.

    The caller Gary S. of Boulder Creek may have a loading coil in the line.  26K is the best he can do with marginal reliability.  Line length is about 4 miles.

    The callers  Cap P and Charlie are in Santa Cruz approximately 1 mile from the Santa Cruz CO.  Connections are at 49333k.  Jeff L is in Ben Lomond approximately 1.5 miles from the Ben Lomond CO and gets 49333.  Note that the high frequency response of these are almost identical where the most obvious difference to the callers that only get 26k is the 10db difference at 3750Hz.  10db is a substantial amount of difference and is equal to about 3 times the signal level.

    Assuming no SLC, it appears that line length to the CO has a significant effect on performance and frequency response.   However, a larger sample of ATY11results will be necessary to be sure.


    Answering Machines, cordless phones, ad nausium.

    The modem should be the very first device connected to the phone line and before any devices that might present a load to the phone line.  These include answering machines, cordless phones, fax machines, caller ID boxes, digital cable set-top boxes, other modems, TDD boxes, and such.  Any or all of these have a ring detector circuit that is capable of creating waveform distortion and trashing the attempts of the adaptive equalizer to obtain a flat frequency response.  This is illustrated in the following, which shows response curves with and without an answering machine installed.  Note that these devices must be disconnected and not just powered off as the ring detector is in the circuit regardless of whether the device is powered on.
    [Note:  I'm not sure that the aformentioned is correct as the removal of exessive equipment from another callers setup did NOT change his frequency response.  This part of the test will need to be repeated.  Jeff L.]

    #1 are the curves with the answering machine and 2 additional telephone instruments.  #2 are the curves without the extra devices.  Note the decreasing slope of the curves with the answering machine.  The adaptive equalizer is being fooled by the non-linearity of the ring detector and producing a bad attempt at equalization.  Additional testing needs to be done to determine which of the 3 attached devices is the major culprit.


    Loading Coils

    This is what I guess is a bad loading coil in the line.  The best this could do was about 24-26k to any ISP.  Loading coils are used to increase the high frequency response at the expense of lower frequencies.  The high frequencies look good, but everything else (level and flatness) are all wrong.  Without the loading coil, the frequency response would drop like a rock starting at 1350 making the line useless for voice or data.


    Which ISP is best?

    A common misconception is that certain ISP's have better 56k performance than others.  This is not due to the connection between the telephone company frame (office) and the ISP as this connection is pure digital.  Performance is almost entirely determined by the connection between the caller and the local telco frame.  From there, it doesn't matter which ISP receives the call.

    These curves show the results of calling from Boulder Creek to various ISP.  Note that all the curves from a given caller are roughly identical.  The exception is deeptht which is the only "ISP" that is NOT using an all digital connection to the phone company for its modem pool and is thus susceptible to additional frequency response deterioration.


    Readings and reference

    3Com How V.90 works.

  • http://www.3com.com/technology/tech_net/white_papers/500659.html


    • 56k Modem Pages

  • http://www.808hi.com/56k
  • http://www.56k.com


    • 56k diagnostics and analysis

  • http://www.808hi.com/56k/trouble.htm
  • http://www.808hi.com/56k/x2-adconversion.htm
  • http://www.808hi.com/56k/diag3com.htm
  • http://www.frostbyte.com.au/public/support/linequality.html
  • 3Com Troubleshooting for the ISP  (200KB PDF with lots of good info)

    • Latest version can be found here.  Please send all comments and gratuities to jeffl@comix.santa-cruz.ca.us
    Revision History:

  • 08/01/99   1.01  Original inscription using Netscape 4.61 Composer.
  • 08/01/99   1.03  Fixed spelling, GIF's, etc.
  • 08/02/99   1.05  Added Neil, Gary.  Tweaked some conclusions.  Added GaryS graph.
  • 12/01/00   1.06  Fixed bad example phone number.