In telecommunications, there are a couple factors that influence the perception of echoes. The first is the round trip delay. The longer the round trip delay, the more the echo is perceived by the far-end user. In other words, the amount of attenuation required by an echo canceller to make the residual echo imperceivable increases with the delay. Therefore, echo cancellers can make their own job easier by simply minimizing the latency it introduces into the communication system.

The second factor influencing the perception of echoes is the bandwidth of the voiced data. In public switch telephone networks (PSTN) applications, the audio data is sampled at 8kHz, and contains frequency content between 300 Hz to 3.4kHz. Wideband (16kHz sampling frequency) and Super Wideband (>20kHz sampling frequency) transmissions have gained attention in Telepresence and VoIP applications because of the improved sound quality. Wideband transmissions contains frequencies from 50 Hz to 7kHz. The frequency content between 50 to 300 Hz adds a sense of fullness and presence to the voice, and the frequency content between 3.4 kHz to 7 kHz adds a sense of intelligibility and clarity. This upper frequency range can be problematic in the sense of echo cancellation because this range makes the echoes more perceivable. Therefore, more echo attenuation would be required in this frequency region.

Fortunately, the increased attenuation required for the upper frequency range is aided by the fact that the higher frequencies are less reflective than lower frequencies. In other words, the echo tail length is shorter and the echo return loss is greater for higher frequencies. This information can be used in reducing the added computational complexity of the higher sampling rate. In sub-band or frequency domain adaptive filters, since the echo tail length for higher frequencies is shorter, the complexity can be reduced by shortening the filter lengths for those frequency bands.

More Information