dtmf fails sometimes

Discussion in 'General' started by vtofa, May 6, 2009.

  1. vtofa

    Joined:
    Oct 21, 2008
    Messages:
    67
    Likes Received:
    0
    On inbound calls over PSTN with a Sangoma A200, sometimes the dtmf sequence is not decoded properly. Specifically, when dialing 20 or 21, sometimes it results in a call to 22. The reverse has not been observed. I thought that maybe there was an echo issue where the first character was repeated. But testing by pressing only 2 never seemed to result in a 22. It seems to be worse during the beginning of an outgoing message. This is Elastix 1.3. Any ideas?
     
  2. dicko

    Joined:
    Oct 24, 2008
    Messages:
    4,099
    Likes Received:
    0
    An interesting forensic problem here. DTMF over ZAP is of course INBAND, there is a known and resolved issue with DTMF repetitions with SIP in some recent builds of asterisk so first off I suggest you update asterisk to 1.4.24.1. (I assume the extensions are SIP so there is a possibility of pre-transcoding these repetitions into INBAND issues.) There is another interesting issue with 2-4 wire hybrids and echo cancellation, if the the hybrid is not balanced then it is possible that "far end reflection" can confuse the DTMF audio to be reflected in the time domain as dupes especially with non-repetitious digits , I have found that local echo-cancellation can in fact exacerbate the problem rather than reduce it (especially with a "long tail" ) so try turning the echo cancellation off first (re-enable it later when you are satisfied that the underlying connection is normalised and it improves the quality of the calls) . Perhaps the most successful results I personally have achieved is by tuning rxgain and txgain in the /etc/asterisk/zapata.conf file with the help of "milliwatt testing" and ztmonitor (I have posted previously here on this topic) this will not only systemically reduce the echo but often enhance the "user experience" of mis-balanced rx/tx levels. To diagnose perhaps turning DTMF debugging on for the channel you are working with will help.
    Obviously I can't say this is your problem, but these are the steps I have learnt to take before any further diagnostics.
    Another highly undervalued tool when diagnosing analog lines is a good old fashioned "butt set", just clip it on and listen.
     
  3. ramoncio

    Joined:
    May 12, 2010
    Messages:
    1,663
    Likes Received:
    0
    Have you run fxotune to supress the main echo in your line?
     
  4. vtofa

    Joined:
    Oct 21, 2008
    Messages:
    67
    Likes Received:
    0
    I ran ztmonitor and it reuired rxgain=5.0 to get about 13000. For tx I used the Milliwatt function in Asterisk but was only ablt to get about 4600 at a txgain=15.0 but it was distorting at that level. It seemed quite loud.

    Any ideas?
     
  5. vtofa

    Joined:
    Oct 21, 2008
    Messages:
    67
    Likes Received:
    0
    I think fxotune does not work for Sangoma, no?
     
  6. ramoncio

    Joined:
    May 12, 2010
    Messages:
    1,663
    Likes Received:
    0
    Try, you have nothing to loose, and much to win.

    fxotune -i 4
     
  7. dicko

    Joined:
    Oct 24, 2008
    Messages:
    4,099
    Likes Received:
    0
    I believe the wancfg util from sangoma allows for some gain setting adjustemnt when run interactively
     
  8. vtofa

    Joined:
    Oct 21, 2008
    Messages:
    67
    Likes Received:
    0
    What do you look for in a "balanced" system? I can't get ztmonitor to produce more than about 4600 before distorting on tx.
     
  9. dicko

    Joined:
    Oct 24, 2008
    Messages:
    4,099
    Likes Received:
    0
    It comes down ultimately to a subjective acoustical balance, ztmonitor was written for digium hardware I believe we are talking sangoma here so maybe you can't rely on the scale readings. (as I said sangoma have some concept of scaling in their driver and maybe they can help us all here). You need to hear a well modulated voice at both ends without clipping, the send and receive path should be indistinguishable in level. you can use the silentmonitor ext (I believe it is 555) to eavesdrop when you think you are close.
     

Share This Page