Adaptive Echo Cancellation

A very successful speech signal processing application is the adaptive echo cancellation used to reduce a common but undesirable phenomenon in most telecommunications systems, called echo. Here, when mismatch impedance is present in any telecommunications system, a portion of the transmitted signal is reflected to the transmitter as an echo, which represents an impairment that degrades the system quality. In most telecommunications systems, such as a telephone circuit, the echo is generated when the long distant portion consisting of two one-directional channels (four wires) is connected with a bidirectional channel (two wires) by means of a hybrid transformer, as shown in Figure 1. If the hybrid impedance is perfectly balanced, the two one-directional channels are uncoupled, and no signal is returned to the transmitter side. However, in general, the bridge is not perfectly balanced because the required impedance to properly balance the hybrid depends on the overall impedance network. In this situation part of the signal is reflected, producing an echo. To avoid this problem, an adaptive filter is used to generate an echo replica, which is then subtracted from the signal to be transmitted as shown in Figure 2. Subsequently the adaptive filter coefficients are updated to minimize, usually, the mean square value of the residual echo. To obtain an appropriate operation, the echo canceller impulse response must be larger than the longer echo path to be estimated. Thus, assuming a sampling frequency of 8kHz and an echo delay of about 60ms, an echo canceller with 256 or more taps is required. Besides the echo path estimation, another important problem is how to handle the double talk, that is, the simultaneous presence of the echo and the near speech signal. The problem is that it is necessary to avoid if the adaptive algorithm modifies the echo canceller coefficients in a domed-to-fail attempt to cancel it.A critical problem affecting speech communication in teleconferencing systems is the acoustic echo shown in Figure 3. When a bidirectional line links two rooms, the acoustic coupling between loudspeaker and microphones in each room causes an acoustic echo perceivable to the users in the other room. The best way to handle it appears to be the adaptive echo cancellation. An acoustic echo canceller generates an echo replica and subtracts it from the signal picked up by the microphones. The residual echo is then used to update the filter coefficients such that the mean square value of approximation error is kept to a minimum. Although the acoustic echo cancellation is similar to that found in other telecommunication systems, such as the telephone ones, the acoustic echo cancellation presents some characteristics that present a more difficult problem. For instance the duration of the acoustic echo path impulse response is o f several hundred milliseconds as shown in Figure 4, and then, echo canceller structures with several thousands FIR taps are required to properly reduce the echo level.

Besides that, the acoustic echo path is non-stationary, because it changes with the speaker's movement, and the speech signal is non-stationary. These factors challenge the acoustic echo canceling, presenting a quite difficult problem because it requires a low complexity adaptation algorithms with a fact enough convergence rate to track the echo path variations. Because conventional FIR adaptive filters, used in telephone systems, do not meet these requirements, more efficient algorithms using frequency domain and subband approaches have been proposed.