Echo cancellation in cellular networks must be handled in different ways. In a typical communication systems, an acoustic echo canceller is located as close as possible to the loudspeaker microphone enclosure.(i.e. directly before, the digital-analog converter for the loudspeaker signal and right after the analog-digital converter for the microphone). Due to the limited processing power of most mobile devices, the complexity of the acoustic echo cancellation (AEC) system limits the effective tail length and adaptive algorithm that can be used. To overcome this problem echo cancellation can be moved further back from the loudspeaker microphone enclosure in the transmission channel to a centralized network.
With this approach, transmission errors and low bit-rate vocoders such as, GSM-AMR, are now located in the echo path. This lossy communication channel adds variability and possible non-linearities to the echo path, thus limiting the effectiveness of an echo canceller relying solely on a linear adaptive filter. Therefore, most centralized systems either use non-linear adaptive filters or a post-filter after a standard linear adaptive filter.
There are several cellular network architectures which influence the implementation of an echo canceller. In a Tandem-Free Operation Network, the speech coder at both ends of the communication is the same, therefore, the echo canceller can work on coded sample parameters. This topic is discussed further in AEC on Speech Coders. More commonly in a centralized network, the encoded speech inputs from both send and receive directions to the echo canceller are first decoded to pulse-code modulated samples. Then the AEC processed samples are encoded to the desired speech codec of the destination. Figure 1, illustrates AEC in this Transcoding Network architecture.
