GB2245459A - Echo canceller with adaptive voice switch attenuation - Google Patents

Abstract

This invention relates to an echo canceller (EC) such as for use in a full duplex radio. The echo canceller comprises: an output path (10) for outputting desired signals; an input path (12) for receiving desired signals together with undesired echos from the output path; an adaptive filter (14) for filtering signals on the input path; measuring means (18) for measuring echo attenuation by the adaptive filter; a variable attenuator (15) for attenuating signals on the input path; control means (16) for controlling the variable attenuator to provide a degree of attenuation dependent on the difference between attenuation measured by the measuring means and a predetermined desired attenuation. The echo canceller is particularly useful in any system having large transmission delays like GSM cellular radio telephone system.

Description

ECHO CANCELLER WITH ADAPTIVE VOICE SWITCH ATTENUATION Backaround of the Invention This invention relates to an echo canceller (EC) such as for use in a full duplex radio. The echo canceller is particularly useful in any system having large transmission delays like the GSM system.

Summary of the Prior Art A conventional echo canceller calculates a pseudo-echo by convolution (filter operation) with the received signal and the tap coefficients of the adaptive filter (ADF). The ADF will converge to an estimate of the echopath. This pseudo-echo is subtracted from the actual echo, thereby cancelling the echo. The ADF is usually a N tap FIR filter, where N*T equals the duration of the echo and T denotes the time between samples. In order to eliminate the echo quickly, the ADF must converge to the actual impulse response.

The GSM cellular radio system calls for standards in cancellation of echos not reached in previous systems. The echopath can either be of acoustic nature caused by the coupling between a microphone and loudspeaker in handsfree operation or if can be of electrical nature due to imperfections in the 4 to 2 wire hybrids in the fixed network. Due to the inherent transmission delay in GSM (total loop delay around 200 msec) the typical Echo Return Loss (ERL) required will be 56 dB compared to a typical ERL of 20 dB in conventional cellular systems. As the echo tends to become more annoying as the transmission delay increases, the performance of the EC unit is very important to obtain high conversation quality.

An echo canceller designed to the meet the requirements of the GSM cellular radio system is described in UK Patent Application No. 9000525.7, filed on 10th January 1990.

The convergence speed of the ADF is often too low to meet the required performance.

Summary of the Invention According to the present invention, an echo canceller is provided comprising: an adaptive filter, measuring means for measuring echo attenuation by the adaptive filter, a variable attenuator and control means for controlling the variable attenuator to provide a degree of attenuation dependent on the difference between attenuation measured by the measuring means and a predetermined desired attenuation.

Preferably the attenuator provides up to half of the total desired attenuation. Preferably the attenuator provides no more than 10 to 15dB of maximum total attenuation.

It is preferred that the variable attenuator provides low or negligible attenuation when there is at less than 3 dB of signal on the output path (A energy) above noise level on that path.

Preferably the attenuator responds to a signal on the input path, (B energy) to provide a relatively low level of attenuation at a high level of such a signal and higher attenuation at a lower level of such a signal. Preferably, on opening in response to a signal on the input path (b energy) the attenuator opens at a controlled rate. Such a rate may be one millisecond for full opening.

At a low level of input signal (B energy) and a low level of output signal (A energy) the attenuator may be arranged to open slowly - eg taking 200 milliseconds to open when there is no B energy and when A energy is less-than 3 dB over noise.

The invention has the effect of compensating the slow convergence of the ADF by introducing a voice switch mode whenever the performance of the ADF is insufficient. A specific embodiment will be described in which the voice switch does not suffer from problems of speech clipping and background noise switching.

Brief Description of the Drawing Figure 1 shows an echocanceller with adaptive voice switch attenuation in accordance with the invention.

Description of the Preferred Embodiment Figure 2 shows a flow diagram of the ERLE computation routine.

Figure 3 shows a flow diagram of the voice attenuation routine.

Referring to the figure, there is shown an output power 10 having a loudspeaker 11 and an input path 12 having a microphone 13. The loudspeaker and microphone are typically mounted on the interior of a vehicle. The echocanceller comprises a fullband adaptive filter 14, the details of which are known in the art (reference is made, for example, to Y.Itoh et al, "An acoustic echo canceller for teleconference" proceedings of ICC-85,1985 pp1498-1502) or a sub-band adaptive filter. The echocanceller further comprises a variable attenuator 15 and a control unit 16 having computation means 17 and 18. Measurements of energy on the input path are taken at points B and C before and after adaptive filtering and these are entered into computation means 18.

The computation means 18 in effect measures the attenuation achieved by the ADF 14. The control unit 16 controls the attenuator 15 to augment the measured attenuation by the necessary amount to bring the total attenuation to 30 dB. The measured attenuation is only augmented when there is energy on the output path 10. This is measured by computation means 17.

The computation of the ADF attenuation is described in Figure 2. Computation means 18 computes the ERLE factor (which is an estimate of the current echocanceller performance and is computed by the formal: ERLE = (C energy - C noise)/(B energy - C noise) The flow diagram of Figure 2 shows an example of an ERLE factor computation, although this can be computed in various ways. Referring to that figure, step 21 determines whether the ADF has, in fact, provided any attenuation, and if so, step 22 computes the factor ERLE'. In step 23, no further action is taken if ERLE' is greater than 50 dB. Otherwise, ERLE' is filtered in a two-order IIR filter, with fc = 0.1 hz (step 24). The output of this filter is ERLE. In steps 25, 26 and 27, ERLE is decremented or incremented by a predetermined amount, according to whether it is greater or less than ERLE' respectively.Provided ERLE is greater than 0 this provides the ERLE factor. If ERLE is less than 0, it is set to 0 in step 29.

After the ERLE computation routine of Figure 2, the control unit 16 performs a set voice attenuation routine, described in Figure 3. If ERLE is greater than 30 dB, as determined by step 31, no additional attenuation is required and step 32 sets the attenuation of the attenuator 15 at a minimum limit, which is 3 dB. Otherwise, if ERLE is not greater than 30 dB, step 33 sets the attenuation at 30 dB minus ERLE. If the maximum attenuation as set in step 33 is greater than 15 dB (step 34), that maximum attenuation is capped at 15 dB in step 35. Steps 34 and 35 ensure that the attenuator 15 does not in fact provide more than 15 dB.

Steps 36 and 37 set a lower limit on the attenuation of the attenuator 15 at 3 dB.

The B energy and C energy measurements are absolute values of samples at point B and C in Figure 1, respectively, filtered through a two-order IRR filter, with cutoff frequency 50 Hz. The B energy and C energy values represent the summation of speech and noise at these respective points and will follow the envelope of speech on the input path 12.

Values for C noise are absolute values of samples at point c, filtered through a two-order IRR filter with cutoff frequency 0.1 hz. Thus the measurement of C noise is an estimate of the noise floor at point C.

The voice switch attenuation of the attenuator 15 is set to keep the total attenuation/cancellation at the required value, ie 30 dB, and the voice controller attenuator (or voice switch) will in this particular case by limited by 3 and 15 dB, although other limits for this range could be chosen. Thus the attenuation can vary continously between 3 and 15 dB, thereby avoiding any annoying effect of changing from pure ADF attenuation to a voice switched mode. Such annoying effect could be experienced if the circuit switched between pure ADF at ERLE factors greater than 30dB and pure switched mode at ERLE factors below 30dB.

Continuously varying attenuation is achieved because the ERLE monitor routine is run in real time, measuring the desired attenuation for each sample of digital speech - i.e.

every 125us. Preferably, the energy measurement used for calculating ERLE is down-converted 20 times, effectivly providing a calculation every 20 samples.

Claims (6)

1. An echocanceller comprising: an output path for outputting desired signals; an input path for receiving desired signals together with undesired echos from the output path; an adaptive filter for filtering signals on the input path; measuring means for measuring echo attenuation by the adaptive filter; a variable attenuator for attenuating signals on the input path; control means for controlling the variable attenuator to provide a degree of attenuation dependent on the difference between attenuation measured by the measuring means and a predetermined desired attenuation; means for measuring energy on the output path; and means for controlling the variable attenuator to provide greater attenuation of signals on the input path at a high level of energy on the output path than at a low level.

2. An echocanceller according to claim 1, wherein no more than 15 dB of total attenuation is provided by the attenuator.

3. An echocanceller according to claim 1 or 2, wherein the variable attenuator provides low or negligible attenuation when there is at least a predetermined level of signal on the output path above noise level on that path.

4. An echocanceller according to any one of the preceding claims, wherein the attenuator responds to signals on the input path (B energy) to provide a relatively low level of attenuation at a high level of such a signal and higher attenuation at a lower level of such a signal.

5. An echocanceller according to any one of the preceding claims, wherein the attenuator is arranged to open at a controlled rate in response to a signal on the input path (B energy).

6. An echocanceller according to claim 5 wherein said rate is a fast rate and wherein the attenuator is arranged to open at a slow rate at a low level of input signal (B energy) and a low level of output signal (A energy).