GB2109207A - Improvements in or relating to interference cancellers - Google Patents

Abstract

The invention is a two-stage echo canceller arrangement including a first echo cancelling stage (20) and a second echo cancelling stage (18). The first a. echo canceller adapts an input signal (x(k)) to synthesize a first estimate of an echo replica signal (@1(k)). A combiner (28) subtracts this first echo replica signal from the actual echo signal to generate a first echo cancelled signal (e1(k)). A control unit (30) monitors the change in value of the echo cancelled signal, and locks the adaptation process of the first echo canceller when the first echo cancelled signal is sufficiently less than the input signal, usually 6 dB of the input signal, to allow a near-end speech detector (23) included in the second stage to function correctly, ie to reduce the echo to a level below that at which it might be wrongly identified as near end speech. The second echo canceller, which is disposed after the first echo canceller, functions in a similar manner as the first echo canceller, and, unless inhibited by the detection of near end speech, continues to operate after the first echo canceller is locked until the echo cancelled signal produced by the second echo canceller is approximately equal to zero. <IMAGE>

Description

SPECIFICATION Improvements in or relating to interference controllers This invention relates to interference controllers, and finds application, for example, as an echo canceller for providing a solution to the network balancing problem.

Echoes commonly occur in a communication system when electrical signals on a receive signal path meet an imperfectly matched impedance at a hybrid junction and are partially reflected back to the distant source over a transmit signal path and, as a result, the reflected signal, or echo, is heard at the far end of the transmit path some time after the original signal has been transmitted. As the distance between the talking and the listening parties is increased, the echo takes longer to reach the talking party, and as a result, the echo becomes, at least subjectively, more annoying to the talking party.

An attempt is therefore generally made to control echoes. One echo controlling arrangement which is disclosed in U.S. Patent 4,005,277 includes a speech signal operated device known as an echo suppressor.

Typcially, echo suppression involves some form of selective attenuation, performed in response to voice levels in the transmission path, so that the echo that would otherwise be returned to the talker is suppressed.

An arrangement such as this is usually satisfactory for terrestrial communication paths in which the echo delay on the round-trip propagation time between the source of the signal and the return of the echo is not long. In communication paths via satellite links, however, the transmission delays are much longer and the echo is more disturbing and may even disrupt conversation by chopping the return signal during intervals where both parties are talking i.e., double-talking.

On the other hand, rather than interrupt the outgoing path, another echo controlling arrangement, known as an echo canceller, typically synthesizes a replica signal of the echo signal and algebraically subtracts the estimate from the outgoing signal to obtain an echo cancelled signal. Most conventional echo cancellers, as for example the arrangement disclosed in U.S. Patent 3,499,999, synthesize the replica by using a taped delay line with adjustable multipliers in an adaptive feedforward arrangement also called a transversal filter.

The multipliers are automatically adjusted by a control signal derived from the difference between the echo and the replica signal.

Ideally, the hybrid included in the 4-wire/2-wire interconnection between the signal paths and the transmitter/receiver would allow, at most 6 dB of the received signal to leak onto the transmit signal path. In practice, however, it has been found that a hybrid may be misconnected, or lacking its associated balancing network, thereby resulting in a network imbalance where a substantial portion of the received signal will leak through to the transmit signal path. Often, prior art echo cancellers will mistake this large echo signal for near-end speech and, therefore, recognizes this signal as near-end speech and hence, not cancel the echo signal.

According to this invention there is provided an interference controller for substantially cancelling an interference signal of a first signal propagating along a transmit signal path of a near-end transmitter/ receiver, wherein, in operation, an interference replica signal is subtracted from the first signal to produce an interference cancelled signal, the interference replica signal is produced in response to the interference cancelled signal and an input signal ion a receive path of the transmitter/receiver, adjustment of the production of the interference replica signal is inhibited when the first signal reaches a predetermined level, and the first signal is a preliminary interference cancelled signal produced in response to an actual return interference signal from the transmitter/receiver and the input signal.

The present invention as an echo canceller, may thus have two-stages. The first stage, or pre-canceller, comprises a conventional echo canceller not including a near-end speech detector and functions to eliminate enough of the return echo signal to allow the near-end speech detector of the second stage to operate properly. Once the first stage eliminates a predetermined amount of the echo signal, the near-end speech detector of the second stage will be capable of correctly distinguishing between near-end speech and large echo, and, therefore the first stage should be inhibited. The second stage, therefore, will function to eliminate the echo signal without incorrectly adapting in the presence of near-end speech.

The present invention may thus overcome the problem associated with prior art echo cancellers which would incorrectly recognize large echo signals as near-end speech. If hybrid loss is less than, for example, 6 dB, pre-cancelling ensures the correct operation of the echo canceller, and, therefore, the echo cancelling process is broken into two stages which provide a partial cancellation and a final cancellation of the echo signal. Upon obtaining a sufficient partial cancellation, the near-end speech detector of the second stage will be able to distinguish between near-end speech and echo, and thereby be able to cancel the echo signal in accordance with the normal operation of an echo canceller.Further, since near-end speech will be uncorrelated with the input signal, near-end speech and echo can still be distinguished and the echo cancellation process inhibited in the presence of near-end speech.

The invention will now be described by way of example with reference to the accompanying drawings in which like references denote like parts and in which: Figure 1 illustrates a two-stage echo controlling arrangement, including a first and a second echo canceller, embodying the invention; Figures 2 and 3, which form Figure 4, show an exemplary embodiment of the invention including a near-end speech detector in the second echo canceller and a "rate-of-change ' control unit capable of inhibiting the action of the first echo canceller; and Figures 5 and 6, which form Figure 7, show another embodiment of the invention which includes a single controller for monitoring the operation of both the first and the second echo canceller.

The following description and accompanying illustrations describe the invention in terms of an echo canceller arrangement disposed at the 2-wire/4-wire interconnection of a telephone circuit. Generally, however, the invention may be used to control any sort of analog of digital interference signal appearing in communication systems in general, including radar, satellite systems, sonar, radio, television, etc.

Therefore, it will be understood that the description which follows hereinafter is exemplary only and for the purposes of exposition and not for the purposes of limitation since the present invention is applicable in any situation wherein an interfering signal is present.

In the ideal, an echo canceller could be constructed which uses only the correlation properties between an incoming signal x(t) originating at the far end and the echo return signal y(t) to control the adaptation process of an included transversal filter, where the correlation signal (t) may be defined by

When using a correlation detector, the time required for the canceller to give a prescribed degree of cancellation will be determined by the incoming signal's bandwidth or bit rate, the resulting maximum sample rate of the incoming signal, the relative level of noise, and near-end speech. When using the correlation process, the adaptation time becomes excessive from a telephone user's point of view in the presence of high levels of noise and typical levels of near-end speech.Conventional prior art echo cancellers, therefore, use a near-end speech detector to inhibit the adaptation process when near-end speech is present.

However, these near-end speech detectors cannot distinguish high level echo signals from near-end speech, and hence, prior art echo cancellers will function incorrectly in the presence of high level echo signals.

The present invention is capable of solving this problem by dividing the cancellation process into stages.

An exemplary arrangement which illustrates the interference-cancelling properties of the invention as it applies in the specific example of an echo canceller is illustrated in Figure 1. A single transmission terminal is basically illustrated for interconnecting a near-end transmitter/receiver 10 with a receive signal path 12 and a transmit signal path 14 by way if a hybrid network 16. A balancing network 17 is connected to hybrid 16 in order to match, as nearly as possible, the impedance of transmitter/receiver 10.

As illustrated in Figure 1, an echo controller primarily includes a tandem arrangement of a second echo canceller 18 and a first echo canceller 20, where first echo canceller 20 is positioned between second echo canceller 18 and hybrid 16. Second echo canceller 18, which may include any form of that known in the prior art, where one example is the previously cited U.S.Patent 3,499,999, receives samples of the signal x(t) approaching near-end transmitter/receiver 10 on receive signal path 12 and processes these samples to form, as an output, an estimate of the echo signal, defined as a second echo replica signal 92(t). Combiner 24, which is disposed in transmit signal path 14, subtracts the second echo replica signal 92(t) produced by second echo canceller 18 from the signal e1(t) travelling along transmit signal path 14to form a difference signal. This difference signal, defined as a second echo cancelled signal e2(t), is thereafter propagated along the remainder of transmit signal path 14 for reception by the far-end transmitter/receiver (not shown).This same second echo cancelled signal e2(t) is also fed back to second echo canceller 18 to supply second echo canceller 18 with the information necessary to improve the estimate of the impulse response of the echo signal, second echo replica signal 92(t), it produces as an output.

Additionally, a near-end speech detector 23 is included with the embodiment of second echo canceller 18 of Figure 1, which functions to inhibit the operation of echo canceller 18 when it deems that near-end speech exists. Near-end speech, that is, speech originating at transmitter/receiver 10 is unwanted noise as far as second echo canceller 18 is concerned, and would diverge second echo replica signal 92(t) from actual return echo signal y(t) if second echo canceller 18 continued to adapt when near-end speech was present.

Therefore, near-end speech detector 23 inhibits second echo canceller 18 when it deems that near-end speech exists by comparing stored values of the input signal with the signal e1 (t) propagating along transmit signal path 14. As discussed above, the problem associated with prior art arrangements is that near-end speech detector 23 would often falsely recognize near-end speech and inhibit second echo canceller 18 in the presence of high level echo signals.

Therefore a first echo canceller 20 functions to make a preliminary cancellation of the echo signal and reduce the signal transmitted along transmit signal path 14 to be somewhat below the threshold of near-end speech detector 23, which is typically 6 dB below the input signal; Basically, first echo canceller 20 includes a correlation means 26 which is responsive to both the input signal x(t) and a first echo cancelled signal e1(t).

Correlation means 26 functions in accordance with equation (1), and is capable of producing as an output a first echo replica signal 91(t), which is, in accordance with equation (1), the cross correlation of the input signal x(t) and first echo cancelled signal el(t) which are applied as separate inputs thereto. Combiner 28, which is disposed in transmit signal path 14 between combiner 24 and transmitter/receiver 10, subtracts this first echo replica signal y1(t) from the actual return echo signal y(t) which leaks through hybrid 16, and forms as the output of combiner 28 the first echo cancelled signal e1 (t) which thereafter propagates along transmit signal path 14 towards combiner 24.In addition, as described hereinabove, the first echo cancelled signal e(t) is fed back to be applied as a first input to correlation means 26.

The first echo cancelled signal e1(t) propagates along transmit signal path 14 and is received by combiner 24, which as described hereinbefore, subtracts the second echo replica signal 92(t) from first echo cancelled signal e1(t) to form the output of the echo cancelling arrangement, second echo cancelled signal e2(t), which is received by the far-end transmitter/receiver. The first echo cancelled signal e1(t) is applied as an input to near-end speech detector 23 and is compared to the input signal x(t) to control the operation of second echo canceller 18. Since first echo cancelled signal e1(t) will be below the threshold of near-end speech detector 23 near-end speech detector 23 will function correctly and will not incorrectly acknowledge the presence of near-end speech when, in fact, large echo signal is present.Also, since any large echo signal will be correlated with the input signal while near-end speech will be uncorrelated with the input signal, near-end speech detector 23 will function only in the presence of a correlated version of the input signal and not, therefore, mistakenly alter any near-end speech that may be present.

Continued simultaneous operation of first echo canceller 20 and second echo canceller 18 in tandem may cause dynamic instabilities. When these instabilities are objectionable to the listener, circuitry may be added to inhibit first echo canceller 20 from further adaptation after a prescribed level of initial echo reduction has been achieved.

Therefore a control unit 30 may be, but does not necessarily have to be, included in first echo canceller 18 to monitor the operation of correlation means 26 by tracking, for example, the rate of change of first echo cancelled signal e1 (t) and lock the updating process of transversal filter 26 when the rate of change reaches a predetermined value. Correlation means 26, subsequent to being inhibited by control unit 30, will continue to produce a value of first echo replica signal y1(t) capable of combining with actual echo y(t) to result in a value of e1(t) which is below the threshold of near-end speech detector 23. This function of control unit 30 will be described in greater detail hereinafter in association with the remaining figures.The locking function of control unit 30 does not inhibit correlation means 26, but rather maintains the state of correlation means 26 at its predetermined acceptable value associated with the predetermined rate of change which triggers control unit 30. Therefore, first echo canceller 20 will continue to produce first echo cancelled signal e1(t), which will accurately represent the echo signal content of y(t) as long as the characteristics of the transmission path do not vary with time. If, however, variations do occur and first echo cancelled signal e1(t) has become too large to allow near-end speech detector 23 to operate properly, control unit 30 may reactivate first echo canceller 20, which would then function to produce an improved first echo cancelled signal e1(t).

An exemplary detailed embodiment of the present invention as it may be applied in association with digital signals is illustrated in Figures 2 and 3, which form Figure 4. Here, an exemplary tapped delay line structure of second echo canceller 18 is shown in detail in Figure 2, where transversal filter 22 includes a plurality of (N-1) delay elements denoted 321-32N-lr N multipliers 331 -33N, N tap weight generators 341 -34N and an accumulator 36.The signal travelling along receive signal path 12, x(k), is introduced into transversal filter 22 and propagates through the (n- 1) delay elements 321 -32N-1 to form, in association with the present input signal value x(k), an N-length sequence comprising the values {x(k), x(k-1), x(k-2) x(k-(N-1))}, as shown in Figure 2.

As described hereinabove in association with Figure 1, second echo cancelled signal, in digital form denoted e2(k), is fed back to transversal filter 22 to assist in improving second echo replica signal 92(k)'s model of the actual return echo signal e1(k) which, in turn, reduces the level of second echo cancelled signal e2(k). lecond echo cancelled signal e2(k) is not, however, suitable by itself for improving second echo replica signal 92(k). Accordingly, second echo cancelled signal e2(k) is passed through the plurality of N tap weight generators 341 -34N which function individually to adjust second echo cancelled signal e2(k) to form a plurality of tap weight values ho(k) to hN1 (k), where these values may be employed in connection with the plurality of sampled and delayed input signal values x(k) to x(k-(N-1)) to form second echo replica signal y2(k).As illustrated in detail in association with tap weight generator 341, which includes a multiplier 311 and an integrating network 351, an exemplary tap weight value ho(k) is formed from second echo cancelled signal e2(k) by multiplying second echo cancelled signal e2(k) with its associated delayed input signal sample value x(k) in multiplier 311 and averaging the resultant composite signal in integrating network 351 to produce tap weight value ho(k). The polarity and magnitude of tap weight value ho(k) thereby formed indicate the appropriate correction necessary for the input signal sample value x(k).

Tap weight values ho(k) to hN-l(k) are subsequently multiplied in a one-to-one manner with their respective delayed sample values of the input signal by multipliers 331-33N, respectively, to form a plurality of N products, ho(k)x(k) to hN~1(k)x(k-(N-1)). That is, ho(k) is multiplied with x(k) via multiplier 331, h1(k) with x(k-1) via multiplier 332, and so on, with hN-l(k) multiplied with x(k-(N-1)) via multiplier 33N The plurality of N products forming the outputs of multipliers 331-33N are applied as separate inputs to accumulator 36 which sums the N product values to produce the second echo replica signal y2(k) produced by transversal filter 22. As stated hereinabove, this signal 92(k) and the first echo cancelled signal e1(k) are applied as inputs to combiner 24 which subtracts 92(k) from e1(k) to produce the second echo cancelled signal e2(k), which is propagated along the remainder of transmit signal path 14 and received by the far-end listener.

An exemplary first echo canceller 20 is illustrated in greater detail in Figure 3. In a similar arrangement as transversal filter 22 of Figure 2, correlation means 26 of Figure 3 is also in the form of a transversal filter which comprises a series of (M - 1) delay elements 401 -40M-1, where the number M is usually, but does not necessarily have to be, equal to N. As can be seen by reference to Figure 3, correlation means, or in this digital example, transversal filter 26 is also coupled to receive signal path 12, where the signal present thereon propagates through delay elements 40140M-1 to form in association with the present value of the input signal, an M-length sequence (x(k),x(k-l),..., x(k-(M-1))}.First echo canceller 20, in a manner similar to the N components of second echo canceller 18, includes a plurality of M tap weight generators 421 -42M (where only tap weight generator 42a, including a multiplier 411 and an integrating network 45a, is illustrated in detail in Figure 3 to avoid complicating the illustration) and a plurality of M multipliers 431-43M. Tap weight generators 421-42M, in response to first echo cancelled signal e1(k) and through the same process as described hereinabove in association with tap weight generator 341, produce a plurality of M tap weights, denoted Io(k)-iMl(k), where each tap weight value is multiplied with its associated delayed value of the signal x(k) produced by delay elements 40i 40M-1 in its associated one of the M multipliers 43i 43M in the same manner as described herein before in association with corresponding components of second echo canceller 18. The M products (jo(k)x(k)-jN1--l (k)x(k-(M-1)) produced by multipliers 431 43M are applied as separate inputs to an accumulator 44 which sums the M elements to produce first echo replica signal 91(k) as the output of transversal filter 26.As decribed hereinabove in association with Figure 1, the signal 91(k) produced by transversal filter 26 and the return echo signal y(k) from hybrid 16 are applied as inputs to combiner 28, which subtracts first echo replica signal y1(k) from return echo signal y(k) to form first echo cancelled signal e1(k) which will travel along transmit signal path 14 and become an input to combiner 24.

in this particular embodiment illustrated in Figures 2 and 3, first echo cancelled signal e1(k) is also applied as an input to control unit 30, which will lock, or maintain, the values of jo(k) to IM-1(k) when the equation

is satisfied for a predetermined rate of change A. In other words, when the rate of change A of first echo cancelled signal e1(k) reaches a value acceptable to the user (e.g. a value corresponding to the time at which first echo cancelled signal e1 (k) is below the threshold of near-end speech detector 23), the first echo canceller 20 has achieved a sufficient echo replica signal to allow near-end speech detector 23 to function correctly, and the tap weights jo(k) to jM-1 (k) associated with that rate may be fixed by near-end speech detector 23.

Therefore, once first echo canceller 20 brings the level of the echo signal down to a level acceptable to near-end speech detector 23, second echo canceller 18 complete the echo cancelling process based on the fixed first echo cancelled signal e1(k) which will continue to be produced by first echo canceller 20. At the completion of the conversation between the near-end and far-end speakers, the ji(k) tap weights must be unlocked, i.e. reset to zero, and control unit 30 reset, in order that the process may be started over again at the initiation of the next conversation. This completion may be recognized by the absence of signal on receive signal path 12 and transmit signal path 14, or in response to some external (on-hook/off-hook) signal (not shown), or any other method known in the art to recognize the absence of conversation.

Another embodiment of the invention as illustrated in Figures 5 and 6 forming Figure 7, corresponds to the arrangement of Figure 4 except that control unit 30 of Figure 3 is incorporated into near-end speech detector 23 to form a near-end speech detector/controller 38. In operation, second echo canceller 18 functions in precisely the manner described hereinabove in association with echo canceller 18 of Figure 2. However, first echo canceller 20 is locked in relation to first echo cancelled signal e1 (k) itself instead of the time derivative of first echo cancelled signal e1(k), as employed in conjunction with the embodiment shown in previous Figures 2 and 3.Specifically, near-end speech detector/controller 38, in addition to inhibiting tap weight generators 341 -34N in accordance with the solution of equation (1), will also lock the tap weight value jo(k) to JM-1 (k) associated with tap weight generators 42142M in accordance with the relation Ie1(k) M < B max(x(k),x(k-l ),...x(k-(M-l))). (3) The coefficient B is a predetermined value that will allow near-end speech detector/controller 38 to lock the Jj(k) values upon reaching a sufficient value of first echo cancelled signal e1(k). Thus, in accordance with equation (3), first echo cancelled signal e1 (k) is significantly small enough to be deemed equal to approximately, for example, 6 dB of the input signal x(k) when it is less than B times the maximum sample value of the input signal, allowing near-end speech detector/controller 38 to function properly and second echo canceller 18 to complete the approximation process. Alternatively, near-end speech detector/controller 38 may be responsive to the actual return echo signal y(k) instead of the input signal x(k), as shown by the dotted line in Figure 7. Thus, instead of applying equation (3), near-end speech detector/controller 38 will lock the tap weight values jO(k)-ju-l(k) associated with tap weight generators 42i 42M in accordance with the relation Ie1(k) < Cy(k) C|y(k)| (4) The coefficient C is a predetermined value which will allow near-end speech detector/controller 30 to inhibit first echo canceller 20 when sufficient cancellation has occurred that will allow second echo canceller 18 to function correctly. As with the arrangement illustrated in Figures 2 and 3, the Jj(k) values must be reset to zero at the completion of the conversation by any method known in the art capable of recognizing the absence of conversation.

Claims (10)

1. An interference controller for substantially cancelling an interference signal of a first signal propagating along a transmit signal path of a near-end transmitter/receiver, wherein, in operation, an interference replica signal is subtracted from the first signal to produce in interference cancelled signal, the interference replica signal is produced in response to the interference cancelled signal and an input signal on a receive path of the transmitter/receiver, adjustment of the production of the interference replica signal is inhibited when the first signal reaches a predetermined level, and the first signal is a preliminary interference cancelled signal produced in response to an actual return interference signal from the transmitter/receiver and the input signal.

2. A controller as claimed in claim 1 including means for subtracting from the first signal the interference replica signal to produce the interference cancelled signal, means responsive to the input signal and the interference cancelled signal for producing the interference replica signal, means responsive to the input signal and the first signal for inhibiting adjustment ofthefirst-mentioned responsive means when the first signal reaches the predetermined level, means for subtracting from the actual return interference signal another interference replica signal to produce the first signal, and means responsive to the input signal and the first signal for producing the other interference replica signal.

3. A controller as claimed in claim 2 and adapted to receive a digital input signal and transmit a digital interference cancelled signal, wherein the means for producing the interference replica signal is responsive to the input signal for generating a plurality of N increasingly delayed sample values of the input signal and so combining the plurality of N values with the interference cancelled signal to generate the interference replica signal and the means for producing the other interference replica signal is responsive to the input signal for generating a plurality of M increasingly delayed sample values of the input signal and so combining the plurality of M values with the first signal to generate the other interference replica signal.

4. A controller as claimed in claim 3 wherein the inhibiting means is responsive to the input signal for generating a plurality of N increasingly delayed sample values of the input signal and comparing the plurality of N values with the first signal to effect inhibition when the first signal is greater than one-half the value of a maximum sample value of the plurality of N values.

5. A controller as claimed in claim 3 or 4, wherein the means for producing the interference replica signal includes a plurality of (N -1) delay elements forming a transversal filter responsive to the input signal for generating the plurality of N values of the input signal, a plurality of N tap weight generators each being responsive to the interference cancelled signal for generating a respective one of a plurality of N tap weight values, a plurality of N multipliers, each being responsive to a respective one of the plurality of N sample values and a respective one of the plurality of N tap weight values, for generating respective products of the values, and means for summing the N product values to generate the interference replica signal, and the means for producing the other interference replica signal includes a plurality of (M- 1 ) delay elements forming a transversal filter responsive to the input signal for generating the plurality of values of the input signal, a plurality of M tap weight generators each being responsive to the first signal for generating a respective one of a plurality of M tap weight values, a plurality of M multipliers, each being responsive to a respective one of the plurality of M sample values and a respective one of the plurality of M tap weight values, for generating respective products of the values, and means for summing the M product values to generate the other interference replica signal.

6. A controller as claimed in claim 2,3,4 or 5 including means for inhibiting adjustment of the means for producing the other interference replica signal at a predetermined degree of interference cancellation.

7. A controller as claimed in claim 2,3,4 or 5 including means responsive to the other interference replica signal for performing a time derivative thereon to generate an interference replica rate, and for comparing the rate with a predetermined value to inhibit adjustment of the means for producing the other interference replica signal when the rate is less than the predetermined value.

8. A controller as claimed in claim 2,3,4 or 5 including means responsive to the input signal and the first signal for generating a plurality of M increasingy delayed sample values of the input signal, for comparing that plurality of values with the first signal, and for inhibiting adjustment of the means for producing the other interference replica signal when the first signal is less than a predetermined weighted value of a maximum sample value of that plurality of values.

9. A controller as claimed in claim 2,3,4 or 5 including means responsive to the actual return interference signal and the first signal for inhibiting adjustment of the means for producing the other interference replica signal when the first signal is less than a predetermined weighted value of the actual return interference signal.

10. An interference controller substantially as herein described with reference to Figure 1, or to Figures 2 to 4, or to Figures 5 to 7 of the accompanying drawings.