This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TI TEC performance - Echoes heard

Hi 

  We are running TEC echo canceller on TI 665x DSP processor. 64 msecs Tail Length with NLP Fixed mode. Please advise. Enclosed are the waveforms. Sin, Sout, Rin and Rout. 

Regards,

Hari

TEC_Waves.zip

  • Hi Hari,

    I've notified the experts. They should respond directly here.

    Best Regarda,
    Yordan
  • One of the noticeable scenario is
    Once TEC stabilizes and if the External phone mutes, then IPPhone in between talkspurts echoes are heard.
    It indicates the TEC is diverging if the other party is muted or in silence.

    IP Phone  Gateway  FXO (Analog Trunk)  PSTN  External phone

    Speech--------------------
    Echo
    Echo heard -------------

    The waveforms are captured at the Gateway -> FXO interface.
  • Additional questions:
    1. How do we set ERL parameter before echo cancellation? Typically ERL is supposed to -6dB, however I do not see TEC kicks in till ERL is approximately <= -18dB.
    2. When I connect a call on the line and disconnect the call without clearing registers, there seems to be divergence of the coefficients before the next call begins, even though the signals are zeros. Is there a way to avoid clear registers for every call?
    3. From Clear registers to convergence for every call varies from 4 seconds to 20/40 seconds of convergence time. This is mostly call dependent. How do we address this issue? What more configuration is required?
  • >>How do we set ERL parameter before echo cancellation? Typically ERL is supposed to -6dB, however I do not see TEC kicks in till ERL is approximately <= -18dB.
    Minimum ERL can be set in channel Control bit field A with 0 and 3dB instead of default 6dB.

    >> When I connect a call on the line and disconnect the call without clearing registers, there seems to be divergence of the coefficients before the next call begins, even though the signals are zeros. Is there a way to avoid clear registers for every call?
    The divergence is expected as the echo path varies in each call. You may need to open/close TEC channel for each call.

    >>From Clear registers to convergence for every call varies from 4 seconds to 20/40 seconds of convergence time. This is mostly call dependent. How do we address this issue? What more configuration is required?
    Do you measure the time starting from call is connected and register is cleared? Or starting from far end signal being present? The convergence time should not be more than one second unless the echo path has significant non-linear component.

    -Garrett