PERCEPTION ORIENTED, DELAY-CONTROLLED ECHO CANCELLATION IN IP BASED TELEPHONE NETWORKS
Wolfgang Brandstätter, Frank Kettler · reposiTUm (TU Wien) · 2003
Modern speech communication reveals a strong trend from the well proven circuit towards upcoming packet switched networks.Beside substantial cost savings and exploitation of synergies through convergent networks, packet switching establishes a basis for new, value-adding services, which arises from the opportunity to combine diverse types of media such as video, text, pictures, and audio.The underlying technology of packet switched telephone networks is inherited from the Internet, which has been originally designed for the reliable transmission of time-insensitive data.Therefore, real-time applications experience significant and unpredictable end-to-end delays, which directly impair the perceived voice quality.Another disturbing factor is the unavoidable echoes, which degrade the user's satisfaction with increasing delays.Hence, echo cancellers are facing stringent requirements, which are only partly met especially under double talk conditions.The delay-controlled enhancement of the conventional principle of echo cancellation improves the perceived voice quality by lessening the extent of influence on the voice signals to be transferred with decreasing echo round-trip times.The signal at the ingress of the echo cancellers is made up of the wanted signals of the local participant and the undesired echo components originated by the distant talker.The control algorithm of the newly developed idea is based on two continually observed parameters.On the one side, the delay of the voice signals from the talker's mouth to the talker's ear determines the overall required echo attenuation according to a predetermined characteristic.On the other side, the measured reduction of the signal levels along the echo path results in-in combination with the aforementioned echo attenuation-the residual echo attenuation needed for the echo canceller.The determination of the echo round-trip delay represents the most demanding challenge within the scope of this work.In this context, the discussed methods of resolution mainly take advantage of the timestamps placed in the transferred voice packets.The results obtained from a listening test confirm that there is room for improvement in the delay-controlled approach.The modeling of two analog subscriber lines, which are connected by an IP based telephone network, acts as basis for the simulation of various network conditions as well as for the optimization of diverse parameters of the echo canceller.Different aspects of the binaural voice samples, which have been created in this manner, have been evaluated by a panel of subjects in a listening test.The results conclude considerable improvements in terms of voice quality compared to the conventional approach.As the delay decreases the voice quality, under certain conditions, increases to a level equivalent to that of an a priori echo-free telephone connection.II