Table of Contents:

Definition and Overview

1. History of Echo Cancellation

2. Types of Echo

3. Causes of Echo

4. The Combined Problem on Digital Cellular Networks

5. The Process of Echo Cancellation

6. Controlling Acoustic Echo

7. Controlling Complex Echo in a Wireless Digital Network

8. Room for Improvement in the Handset

9. The Future of Echo-Cancellation Technology

Self-Test

Glossary

PDF of this tutorial

In echo cancellation, complex algorithmic procedures are used to compute speech models. This involves generating the sum from reflected echoes of the original speech, then subtracting this from any signal the microphone picks up. The result is the purified speech of the person talking. The format of this echo prediction must be learned by the echo canceller in a process known as adaptation. It might be said that the parameters learned from the adaptation process generate the prediction of the echo signal, which then forms an audio picture of the room in which the microphone is located. Figure 5 shows the basic operation of an echo canceller in a conference room type of situation.

Figure 5. Operation of an Acoustic Echo Canceller

During the conversation period, this audio picture constantly alters, and, in turn, the canceller must adapt continually. The time required for the echo canceller to fully learn the acoustic picture of the room is called the convergence time. The best convergence time recorded is 50 ms, and any increase in this number results in syllables of echo being detected.

Other important performance criteria involve the acoustic echo canceller's ability to handle acoustic tail circuit delay. This is the time span of the acoustic picture and roughly represents the delay in time for the last significant echo to arrive at the microphone. The optimum requirement for this is currently set at 270 ms—any time below this could result in echoes being received by the microphone outside the ability of the echo canceller to remove them, and hence in participants hearing the echoes.

Another important factor is acoustic echo return loss enhancement (AERLE). This is the amount of attenuation which is applied to the echo signal in the process of echo cancellation—i.e., if no attenuation is applied, full echo will be heard. A value of 65 dB is the minimum requirement with the non-linear processor enabled, based on an input level of -10 dBm white noise electrical and 6 dB of echo return loss (ERL).

The canceller's performance also relies heavily on the efficiency of a device called the center clipper, or non-linear processor. This needs to be adaptive and has a direct bearing on the level of AERLE that can be achieved.