MXPA97002958A - Elimination of ecos using cross correlation of reception and transmission sample segments with intermediate storage to determine coefficient of cancelac filter - Google Patents

Abstract

An echo detection system comprising data buffers (8, 18) for storing upper and lower channel signals (5, 15) respectively. The signals undergo pre-processing (9, 19) to identify the signal forms characteristic of the dialogue and instruct the measuring unit (10) to perform the comparison using cross-correlation techniques between the signals stored in the buffers (8). , 18) only when such characteristics are detected. This reduces the processing power required and increases the accuracy of the correlations. Parallel processing techniques allow echoes to be detected with longer delay times. The results of the measurement can be used to generate an echo cancellation signal (units 7, 1

Description

ELIMINATION OF ECOS USING CROSS CORRELATION OF SEGMENTS OF RECEPTION AND TRANSMISSION SAMPLE INTERMEDIATE STORAGE TO DETERMINE COEFFICIENT OF CANCELLATION FILTER S "'DESCRIPTION OF THE INVENTION This invention relates to the measurement of signal quality over telecommunications links, and in particular to interference detection. More particularly, this invention relates to the detection of spurious signals generated on a second channel as a result of signals that are transmitted on a first channel, a situation generally known as "crosstalk". The spurious signal, once detected, can be measured and canceled. The invention is particularly suitable for detecting echoes. This occurs in a 2-way telecommunications link. A signal traveling in a first direction gives rise to a spurious signal traveling in the opposite direction. If this spurious signal returns to the original source of the signal, it will appear as an echo. The echo effect can be caused in one of several ways. It can occur as a result of feedback acoustics between the earpiece and the mouthpiece of a telephone.

It can occur as a result of reflections caused by impedance imbalances. It can occur as a result of a cross coupling between the trajectories in hybrid points of 4 to 2 wires; these are the points where the two-way traffic, carried over the connection of two wires of a telephone termination, is separated into two separate channels (a so-called four-wire connection). This invention is suitable for detecting echo effects at points in the system where the signals in the two directions are carried over the two separate channels. The result of any of these echo effects is that an announcer will receive his own speech, delayed for a short period. The magnitude of the delay is largely determined by the distance that the signal has to travel, with a very small contribution of signal processing delays. The distances traveled by the signals in intercontinental calls can introduce delays easily detectable by human observers: the total distance on the surface of the earth between a point on the surface of the earth and its antipodes is of 40000 km (approximately 140 milliseconds light - already that the wire lines do not necessarily follow the shortest route, the practical distance is greater than this). The full stroke distance between two points, on the surface of the earth via a geostationary satellite is approximately 1/2 second light (150000 km). The diversity of international calls and other network services can create even longer trajectories. Delays of this order of magnitude, as well as being annoying, also confuse the speaker who may find it impossible to continue the speech. Therefore, it is desirable to detect when an echo is occurring, so that a remedial action can be taken. This remedial action may involve taking the fault circuit and taking it out of use until it can be repaired, or limiting the use of the fault circuit to uses where the echo causes fewer problems, such as short distance calls (where, the echo delay is too short to be problematic), or a one-way transmission such as facsimile transmissions. There are also methods to cancel the echo signal by artificially combining it with a complementary signal derived from the output signal to generate a zero output. However, all these systems require prior knowledge that there is an echo, and some of its characteristics, notably its time of delay and its attenuation. It is known to transmit test signals on a telecommunications link in order to detect the presence of echoes. This system can only be used on lines that are not currently in use, due to traffic on the line that may interfere with the detection of the test signal echo. It is also known to use trained human observers to verify conversations live, but this method is very laborious, subject to human subjectivity, and also has implications for the privacy of the speakers. In-service non-intrusive measuring systems are known, which use adaptive least-squares (MCM) filter systems to measure the delay and echo resistance from the impulse response transported. Modern digital signal processors can support filter coefficients of approximately 650; at a sample rate of 8 kHz this equals a maximum detectable echo path of approximately 80 ms. To detect longer echo trajectories using this impulse response technique, either the number of filter coefficients beyond these practical limits must be increased, or the sample rate must be reduced, which reduces the likelihood of a response Convergent of any of the samples. U.S. Patent 5,062,102 (Taguchi) discloses an echo canceller wherein echo is detected by identifying cross correlations between signals carried by first and second transmission lines using a cross-correlation technique.

This allows short signal samples to be used, instead of the long passages required by an adaptive filter, and allows the filter coefficients for the echo correction signals to be generated more quickly than through the use of adaptive filters. According to a first aspect of the invention, an interference detection system for a telecommunications link having first and second separate channels, the system comprising first verification means for inspecting the signals traveling on the first channel, second means for verification for inspecting signals traveling on the second channel and comprising comparison means for comparing the signals verified by the first and second verification means for one or more delay periods to identify the presence of interference between the channels, wherein the Comparison means are arranged to identify the cross correlations between the signals inspected by the first and second verification means, characterized in that the first verification means include means for detecting and selecting signal segments on the first channel having predetermined characteristics. The comparison means are adapted to identify the cross correlations between such characteristic signal segments, and the signals verified by the second verification means, the selected signal segments having lengths corresponding to the duration of the predetermined characteristic. According to a second aspect, a method for detecting inter-channel interference over a telecommunications link having first and second channels, the method comprising the steps of verifying the signals traveling on a first channel, verifying the signals traveling on a second channel, and compare the signals for one or more delay periods to identify the presence of interference between the channels where the method comprises the identification of cross correlations between the signals carried by the first and second channels, characterized in that it comprises the steps of detecting signals having predetermined characteristics on the first channel, selecting segments of signals having such characteristics, and identifying the cross-correlations between such characteristic signal segments and the signals carried by the second channel, the selected signal segments having lengths that correspond to the duration of the default feature. By selecting such characteristic signals to analyze the available processing capacity, it can be efficiently used concentrating on signal samples, which are probably produced by strong cross correlations, allowing a larger scale of delays to be verified. False corrections of low-level white noise are also avoided. In a preferred arrangement, signal segments having such characteristics on the first channel are detected and selected for the intended correlation with the signals carried by the second channel, the selected signal segments having lengths corresponding to the duration of the predetermined characteristic, and preferably greater than a predetermined minimum. By developing the sample length to coincide with the duration of the characteristic element, the opportunity to correlate correctly is improved, since the longer the sample is, the less likely it is that a false correlation will occur, without trying to correlate parts of a signal that does not contain the characteristic elements. The signal characteristics that are to be identified may include signal strength or may be characteristics associated with human language. Since the characteristic aspects of the signals are verified and correlated, these aspects can also be used to determine other characteristics of the interference phenomenon.

In a preferred arrangement, the comparison means comprise a plurality of cross-correlation means, each of the cross-correlation means performing a cross-correlation for a different delay period, and delay measuring means for determining, from the outputs of the cross-correlation means *, the magnitude of the delay in the interference signal. The invention can be used to verify the interference between any two channels of a communication system, but is particularly suitable for echo detection as long as the sending and receiving paths are separate, eg conventional four-wire analogue telephony , digital telephony, broadband applications, duplex radio systems (time division or frequency division) or asynchronous transfer mode (MTA). Accordingly, the pair or pairs of channels of the communication system to which the interference detection system is connected, preferably each comprising a two-way communication link, the system being arranged to detect echoes. The embodiments of the invention allow a larger scale of the delay periods to be verified simultaneously by storing several samples in separate storages and processing each separately. In a typical situation, two different periods of echo delay will be found, depending on which the caller is speaking. The system can be used to provide the input of an echo canceller. An echo canceller is added to the return path a cancellation signal corresponding to the signal on the outward path, presenting a delay and attenuation corresponding to those of the echo, but having an opposite phase. A problem encountered with known echo cancellers is that a false correlation can cause a cancellation signal to be inserted where it is not necessary, which creates its own echo effect. The problem can be avoided by determining a balance average from a predetermined number of measurements of the delay and / or attenuation measuring means, which differ from each other by values less than a predetermined value. The effects of the individual false correlations, which will have attenuations and delays different from the true echo, are therefore, reduced to a minimum. In a network management system, there is a plurality of interference detection systems, each associated with a respective pair of channels, and one or more means for introducing a cancellation signal into a channel over which the interference is detected. By arranging the system in this way, a number of cancellers can be reduced, the cancellers being dynamically allocated to those pairs of channels where the interference is detected, or the more serious interference. The system may include language address determination means comprising means for determining on which channel the longest signal segments having the inspected characteristics are occurring. The characteristic aspects of the input signal, therefore, can be used to identify which of the two callers is talking, and therefore, which path should be verified for the echo signals, thus reducing the processing overhead by a additional factor of two. The length of the delay can be used to help locate the source of the echo, since the longer delays are caused by distant equipment or with intermediate processing elements. Until the call route is known, a network operator can identify the failure device in this way. Of course, in some cases, the call may be an interconnection between two operators, and one operator may not know the route in the other operator's network. In this case, a network operator using the echo detection system of the invention, however, you can identify from the length of the echo delay if the echo is caused by your own network or by another, and in this way if the remedy action is within your restriction. An embodiment of the invention will now be described by way of example only, with reference to the drawings, in which: Figure 1 illustrates a simple telephony network including an echo detector according to the invention. Figure 2 shows the various elements of an echo detector mode of Figure 1, incorporating an echo canceller. Figure 3 shows an echo measurement system incorporating an echo detector according to the invention. Figure 1 shows a simplified telephony network having two terminations 1 and 2 connected through hybrids of 4 to 2 respective wires, 3, 4 to a main telephone link having a first path 5, (from hybrid 3 to hybrid) 4) and a second path 15 in the reverse direction. Connected at some point along trajectories 5 and 15 is a non-intrusive measuring device 6, which is described in more detail below. Device 6 is connected to a first path 5 at a point X and a second path 15 at a point Y. Figure 2 shows the echo detector of Figure 1 in more detail. From the branch points X, Y on the trajectories 5, 15, respective signals are respectively fed to respective buffers 8, 18, and therefore to respective pre-processing units 9, 19. The pre-processing units 9, 19 feed a speech or voice address classification unit 11. A measuring unit 10 receives inputs from the data buffers 8, 18, the pre-processing units 9, 19, and the address sorting unit 11, and supplies an output to a post-processing unit 12, which in turn provides an output to one or the other of the two echo cancellation units 7, 17, which also receive an input of the respective data buffers 8, 18. The echo cancellers 7, 17 provide an input to the paths 15, 5, respectively through respective combiners 13, 14, downstream of the branch points X, Y. Figure 3 illustrates a device for calculating loss of echo, which can make use of the output of the device according to the invention. Two signals X, Y, have input to a dialogue classifier 11, which, as in Figure 2, identifies which signal is the incident signal and which is the reflected signal, and controls the switches 36, 37 to feed the incident signal to an input 21 and the signal reflected to an input 30.

The incident and reflected signals are fed by means of the buffers 8, 18 to a processor 10, as shown in Figure 2, and the output of the processor 10 is fed to a bulky delay buffer 22. The signal incident in the input 21 enters the buffer 22 to delay it during a period corresponding to the delay of echoes determined by the postprocessor 12, generating a delayed input 23. Then, both signals are fed to respective modifiers 26, 27, where they are applied charges for generating a modified delayed input signal 28 and a modified reflected signal 31. The loads are derived from an analysis unit 24 by checking the delayed incident signal 23. The modified input signal 28 is then fed to a digital analog filter ( FAD) 29. The output 32 of the filter 29 is compared to the modified reflected signal 30 in a comparator 35 to generate an error signal 33, the c ual is fed back to the FAD 29. The FAD filter values 29 can be deduced from an output 34 to allow calculation of the echo loss through a computer 38. The operation of the invention will now be described. Referring now to Figure 1, it can be made that an echo, when part of a signal traveling along a first path 5, intended for the termination 2, is reflected in the hybrid 4 and returned on the second path. This signal will be heard by the user in Termination 1, who was the original speaker of the pronunciation. Similarly, an echo can be caused by the hybrid 3, by reflecting signals transmitted by the transmission 2 back to the speaker using said termination. The echo can also be caused by acoustic feedback at the far end, between the earphone and the user's mouthpiece. The delay between the output and input signal, perceived by the user of a termination 1, is determined largely by the distance between the termination 1 and the hybrid 4, or another element that causes the echo. Similarly, the delay between the output and the input signal, as perceived by the user of a termination 2, is largely determined by the distance between the termination 2 and the hybrid 3, which is causing the echo. The device 6 is connected to the network by branch connections X and Y connected to the paths 5, 15, respectively, and is used to detect the presence of echo in the system, inspecting both lines for signals and cross-correlating these signals to identify the characteristic signals that have passed on the X connection and subsequently are passing on the Y connection, or vice versa. The X and Y connections are simple low-impedance T-shaped connections that allow the signals transmitted on the paths 5, 15 to be inspected by the device 6. Measuring the delay between these cases, the distance of the echo source can be derivative for example, an echo generated by hybrid 4 produces a shorter echo delay than one of * termination 2. In addition, path 5 or 15, over which the original signal appeared, identifies the direction in which the echo has come, thus establishing whether the echo source is between the device 6 and the first termination 1, or between the device 6 and the second termination 2. In a current network, you can find several elements such as hybrids 3, 4 over any of the sides of device 6, any of these can be the source of the echo. The echo detection device 6 uses a cross-correlation technique to compare the dialogue on the reflected and transmitted trajectories. Cross-correlation is a method of statistical comparison of two signals generally used, in signal processing, to calculate the delay between the input waveform and the output waveform of a system. In the present case, the system in question is the echo path of the telephony circuit, that is, from connection X to connection Y via hybrid 4, or from connection Y to connection X via hybrid 3. The transmitted signal is compared with the reflected signal (normalized in amplitude to correspond to that of the transmitted signal) and the cross-correlation coefficient is calculated. The cross-correlation coefficient has a value of -1 to 1 and describes how the two similar signals are. A value of 1 means a complete cross-correlation and results when the two waveforms are identical. A value of -1 means a complete negative equilibrium, that is, the signals are identical, but for a phase inversion of 180 °. The human ear is not sensitive to the phase, so for the purpose of the present, a negative correlation is as important as a positive one, since the human ear will detect either as an echo. Consequently, in this way the absolute magnitude of the correlation is used. The transmitted signal is then delayed by a unit of time and the cross correlation coefficient is recalculated. A balance between the two signals (that is, the magnitude of the cross correlation coefficient being closer to the unit) will occur when the delayed transmission signal is equal to the reflected signal. The echo detector pre-processes the dialogue signal before performing the cross-correlation. This significantly improves the accuracy and reliability of the device by choosing segments that contain dialogue to cross-correlate. In particular, since only selected segments are analyzed, they can be analyzed in more detail. For example, the elementary delay imposed on the transmitted signal can proceed in smaller increments, improving the accuracy of. the measurement of the delay. To improve the accuracy and reliability of the system, the signals are pre-processed to identify the dialogue segments that are suitable for cross-correlation. This pre-processing also identifies the address of the speaker's dialogue ie, near to far or far to close. Since dialogue is an essentially unidirectional means of communication (one person speaks and the other listens) monitor 6 measures both echo trajectories ('X to Y' via hybrid 4 and 'Y to X' via hybrid 3). To allow real-time measurements (almost), parallel processing is used to divide the echo path into segments. From the verification point X, the original signal is passed to a data buffer 8, which stores the input signals during the time in which the measurements are made with them. The data entering the buffer memory is inspected by a dialogue preprocessing unit 9, which identifies suitable segments for measuring and indicates to a measurement unit 10, which of these segments are present in the buffer memory 8. A second data buffer 18 and dialogue pre-processing unit 19 inspect the signals passing through the verification point Y. The outputs of the dialogue pre-processing units 9, 19 are compared in an identification unit of address 11. This unit compares certain characteristics of the signal such as the signal strength and the duration of the dialogue segment to determine which of the channels carries the original signal. The measurement unit 10 uses the output of the address indication unit 11 and the dialog pre-processing units 9, 19 to select the data of the buffers 8, 18 on which cross-correlation measurements are made. The results of these measurements are transmitted to a post-processing unit 12, which makes use of the cross-correlation results to perform the appropriate action. The post-processing unit 12 can use the cross-correlation measurements to generate an echo cancellation signal. This is done in the canceller 7 or 17, extracting the input signal from buffer 8 or 18, respectively, attenuating and delaying it in amounts equivalent to the echo signal detected as measured in unit 12 and applying to the return path 15 or 5, respectively, a corresponding signal to the result of this procedure, but out of phase with the signal detected by 180 °. This applied signal is combined in the combiners 13, 14, respectively, with the arrival of the echo in the return path 5 or 15 to produce a zero output. It should be noted that the echo cancellation signal must be applied downstream of the measurement points X, Y, to avoid that the same echo cancellation signal forms part of the signal measured in the return path. The post-processing unit 12 can generate information for network management purposes. The delay length can be used, along with knowledge of the route of the call, to identify the component that causes the echo, allowing the remedial action to be carried out. Alternatively, the call can be diverted to another route, or abandoned. The time required to resolve an individual echo and delay measurement depends on the maximum delay that will be resolved, that is, a delay of one second after an appropriate dialogue segment is detected, it takes a second to accumulate the samples and a additional period to perform the processing. Through judicious programming, it is possible to reduce the additional processing time, but finally the processing time still depends on the number of samples required that are to be stored by the echo path. To reduce the processing time, fixing means of the measuring scale are included. Using this technique, the algorithm can run simultaneously on several digital signaling processors (PSD) within the measurement unit 10, each PSD looking for a different scale of measurement. For example, four PSDs can be used to process delay measurements of 1 second. Each PSD looks for a scale of 250 ms for the echo path (0-250, 250-500, 500-750, 750-1000), therefore the limiting factor of the measurement dialog is now only 250 ms. If the algorithm is used for national networks, where the upper limit delay probably does not exceed 60 ms, the scale can be reduced accordingly. This configuration is very suitable for parallel processing that allows the correlation to be extended to several processors, this improves the speed / efficiency of the algorithm. A high-level controller can determine which PSD returns to the correct delay value by examining the cross-correlation coefficient.

This technique of dynamic placement of the algorithm through the resources of PSD increases the number of successful measurements in a given period. The buffers 8, 18 are used to store uncompressed samples of the 2Mbit / s streams in the paths 5, 15. The buffers use a FIFO buffer (first input first output) of two pointers, which has two indicators FULL and EMPTY. A conversation is built from increases and pauses in dialogue. The dialogue increases give the best cross-correlation as the attenuation due to the echo path that will decrease the low energy segments, such as unspoken and noise signals, most. Therefore, it is important that the pre-processing select segments that are likely to give a good cross-correlation. The pre-processing units 9, 19 select dialog segments of the signals for cross-correlation. A minimum segment length (40ms) is required to give a reliable and exact cross-correlation. Reliability is improved more if a longer segment is used, although improvement is not observed above 80 ms. However, if a segment has a fixed length, ie 80 ms, it may contain only a small increase in dialogue at the beginning, and the rest of the segment having noise. If this occurs, the segment is less likely to cross-correlate. A variable segment length ensures that the segment contains mainly dialogue, not noise. The pre-processing selects dialog segments between 40 and 80ms in length. Since conversations are essentially unidirectional, people take turns talking to each other, an address indication unit 11 can be used to detect which party is talking. The echo path delay and loss is then calculated for that address, that is, if the dialog is detected at point 'X', the echo path 'X-4-Y' is calculated; conversely, if the dialogue is detected at point 'Y1, the echo path is calculated? -3-X'. If the dialogue is only present in one direction, then it is not possible to solve the echo path in the opposite direction. The address is found by comparing the length of the dialogue segments on the two channels. The channel with the longest segment of dialogue is taken as the channel with the incident dialogue. A normal cross-correlation algorithm is used to calculate the delay.

If the dialogue echo path delay (RTED) is resolved, the incident signal is giving a delay, equal to RTED, and the echo signal loss is calculated from the difference between the mean square root (rcm) of the signal incident and the rcm of the reflected signal. As mentioned above, the dialogue needs to be present on a channel before the measurement is resolved. The minimum measurement time is 15 seconds. This will increase the probability that an appropriate segment of the dialogue will be present on the channel. Within 15 seconds, several measurements will probably be made, some means are required to choose the measurements that are correct. The method is based on two procedures. First, the cross-correlation produces a correlation coefficient value, or confidence factor. If the signals coincide exactly after the signals have been normalized and properly delayed, an exact match will produce a correlation coefficient of 1. Due to the deterioration of the echo path, the delay will probably be less than 1. Tests have shown that if the delay value is greater than 0.5, then the delay has been calculated correctly. Secondly, if several results are produced it is reasonable to assume that each measurement is within an allowable accuracy between them. An average balance is used so that a value is included in the average if at least two results are within the allowed accuracy between them. Probably, any erroneous cross-correlation will produce random delay estimates and therefore it will not be included in the average result. In the modality described above, cross-correlation is performed in the time domain. Alternatively, it can be performed in the frequency domain using fast Fourier transformations (TRF). This requires more memory, but it is more efficient. A simple method to calculate the cross-correlation only is to use the bit sign of the signal. If the samples of the original and reflected signals are of the same sign, a counter is incremented, if they are of an opposite sign the counter is reduced. For a good match, a large total will be found, its magnitude being related to the length of the sample and its sign depending on whether the echo is in phase or antiphase. The output can be normalized using the length of the sample, giving values in the scale of -1 to +1. This method is not as accurate as other means to calculate the correlation coefficient, but it is reasonably accurate for the low level of loss. It has the advantage of not being computationally intensive and therefore very fast. Said arrangement is suitable for lower cost PSDs, which have limited processing power and are designed to operate on circuits that will have a lower echo loss value. The method is not limited to using dialogue as the circuit stimulus (however, it has been optimized by dialogue). Circuits that already have echo cancellers will not have, under normal operation, an echo present. Even if the echo is not present, the full stroke delay is a useful measurement to obtain. For these circuits, a continuity signal, generated by the signaling system, can be used to perform the "crossed" correlation. A continuity signal is a tone, transmitted over the path of the dialogue from the output switch to the input switch that forms circuits with the return signal. This method presents a measurement of the delay between international switches. Continuity verification tones are generated by the International Telecommunications Union (ITU-T) signaling system number 7 before the ring tone. The method of the invention can be applied to other applications not directly related to voice telephony, and in this specification the term telecommunications link is used in the broad sense to cover any link that carries signals from one point to another, either as part of the switching system or as a dedicated link. The interference detection system of the invention can be used to provide the echo delay input for an echo loss calculator, as will be described later. In Figure 3, a delayed incident signal 28 and a reflected signal 31 enter the digital analog filter 29. The output 32 of the FAD 29 is compared to a reflected signal 31 in a comparator 35 to generate an error signal 33, which enters to FAD 29 Using the unmodified incident dialog 23 (delayed by the voluminous delay 22) and the reflected dialogue 30 as in the inputs, the FAD 29 can converge to produce the impulse response of the echo path. The impulse response of the echo path is effectively; a model of the echo path, however, the model produced will not be accurate as it is dependent upon the characteristics of the dialogue. An FAD will converge to its optimal state if a white noise signal is used as its input. Therefore, to improve the accuracy and speed of convergence, a linear prediction unit 24 is used to perform a pre-emphasis form to modify the delayed incident signal 23 and the reflected signal 30 to the FAD, to the "whiteness" of the signals.

The delayed incident signal 13 is modified in a filter 26 to generate a modified delayed incident signal 28. Similarly, the reflected signal 30 is modified in a filter 27 or generates a modified reflected signal 31. Modified signals 28, 31 are used as inputs to FAD 29. Dialog signals consist of expressed and non-expressed segments. The expressed segments have a high energy and the samples are self-correlated in contrast to the lower energy noise type samples in the non-expressed segments. These characteristics result in a poor convergence regime of the MCM algorithm (mean least squares) used by FAD. Since the non-expressed segments have a low energy, they tend to be corrupted by the echo path noise, so that the properties of the expressed segments of higher energy have been exploited to improve the performance of the MCM algorithm. In order to accomplish this, the delayed incident signal is supplied to a CPL analysis unit (linear predictive coding) 24, which derives the coefficients of a filter H (z) having a frequency response similar to the frequency spectrum of the incident signal. Such analysis is well known in the art. Essentially, a series of coefficients is generated which, when applied to a white noise signal, reproduce the voice sound that was modeled. In this way, it stimulates the effect of the vocal tract on the entry of essentially white noise towards it through the lungs and trachea of the speaker. Applying the inverse function l / H (z) of this in the filters 26, 27, a pseudo white noise signal corresponding to the original dialogue energy can be generated. The linear prediction unit 24 receives an input of the delayed incident speech signal 23. The sequence H (z) generated by the units 24 is transmitted as an output 25 to the filters 26, 27, which apply the inverse form of the sequence H (z) to the delayed incident signal 23, and the reflected signal 30, to generate modified outputs 28, 31, respectively. The delay imposed by the buffer 22 is determined by the correlation technique described above, and this delay is applied to the signal 21 by means of the variable delay buffer 22, so that FAD 29 is centered on the echo path delay. The FAD 29 will then converge on the echo path. If the delay period is predetermined in this way, the FAD 29 can focus on the echo path, thus requiring that the filter have a much shorter length than if the delay imposed by the buffer 22 were only an estimate. It is desirable to test the echo on both channels of a two-way telecommunications link, since the echo can appear on either or both channels. Therefore, it is necessary to identify on which channel the incident signal is located, so that the correct signal is delayed. Instead of operating using a variable delay period between zero and a predetermined maximum, the period may rather be made variable between negative and positive values of the maximum. However, since this will require both positive and negative values of the delay to be tested, it could have the number of delay periods of different magnitude to be tested. Rather, in a preferred arrangement, the channel that actually carries the incident signal is identified in a pre-characterization step. In most cases, a two-way voice link is used by the speakers at the same time. Therefore, it is possible to identify which of these two channels is actually in use and check only the return channel for echoes. This can be done by identifying on which of the two channels the strongest signals are occurring. This channel is identified as the "incident" channel and the other is, therefore, the "reflected" channel.

In the embodiment of Figure 3, the dialogue classification is performed by the voice activity detector 11. The detector 11 identifies on which of the two channels X, Y the strongest signals are found, and controls the switches 36, 37 The switch 36 is arranged to provide either an X channel or a Y channel to the input 21, under the control of the detector 11. Similarly, the switch 37 is arranged to provide either an X channel or a Y channel at the input 30, also under the control of the detector 11. The detector 11 provides outputs so that when the switch 36 is fixed in the channel X, the switch 37 is fixed in the Y channel, and vice versa.

Claims (23)

1. An interference detection system for a telecommunications link having first and second separate channels, the system comprises first verification means for verifying signals traveling on the first channel, second verification means for verifying signals traveling on the second channel and which comprises comparison means for comparing the signals inspected by the first and second verification means for one or more periods of delay, to identify the presence of interference between the channels, wherein the comparison means are arranged to identify the cross-correlations between the signals verified by the first and second verification means, characterized in that the first verification means include means for detecting and selecting signal segments on the first channel having predetermined characteristics, and the comparison means are arranged to identify the cross correlations between such characteristic signal segments, and the signals verified by the second verification means, the selected signal segments having lengths corresponding to the duration? e the predetermined characteristic.

2. The system according to claim 1, characterized in that the selection means are arranged to select parts of the signal having the identified characteristics that have a duration greater than a predetermined minimum.

3. The system according to any of the preceding claims, characterized in that the comparison means comprise a plurality of cross-correlation means, each cross-correlation means performing a cross-correlation during a different period of delay, and delay measurement means to determine , from the outputs of the cross correlation means, the magnitude of the delay in the interference signal.

4. The system according to claim 3, characterized in that it includes means for determining a balance average from a predetermined number of measurements of the delay measurement means different from each other by values less than a predetermined value.

5. The system according to any of the preceding claims, characterized in that it includes means for determining the direction of dialogue comprising means for determining on which channel the longest segments of signal having the verified characteristics occur.

6. The system according to any of the preceding claims, in association with an echo loss calculation device, characterized in that it includes means for generating an echo delay signal in response to the identification of a cross correlation during a given delay period. , and means for transmitting the echo delay signal to the echo loss calculation device.

7. The system according to any of the preceding claims, characterized in that it includes means for introducing a cancellation signal to the second channel.

8. A network management system, characterized in that it includes one or more means for introducing a cancellation signal to a channel over which interference is detected, a plurality of interference detection systems, according to any one of claims 1 to 6. , each associated with a respective pair of channels, and means for selecting the channel with which the cancellation means are associated in response to the interference detected on the channel by the respective detection means.

9. The network management system that includes an interference detection system according to any of claims 1 to 7, characterized in that it comprises means for identifying, the delay measured by the system, the network elements responsible for the interference.

10. The communication system, characterized in that it has a plurality of communication channels, and an interference detection system according to any of claims 7, the first and second verification means of the interference detection system being arranged to inspect one or more pairs of communication channels.

11. The communication system according to claim 10, characterized in that the pair or pairs of channels, each comprising a communication link of two paths, the system being arranged to detect echoes.

12. A method for detecting inter-channel interference on a telecommunications link having first and second channels, the method comprising the steps of inspecting signals traveling on a first channel, inspecting signals traveling on a second channel, and comparing the signals during one or more periods of delay to identify the presence of interference between the channels, wherein the method comprises the identification of cross correlations between the signals carried by the first and second channels, characterized in that the additional steps < ie detecting signals having predetermined characteristics on the first channel, selecting segments of signals having said characteristics, and identifying cross correlations between such characteristic signal segments and the signals carried by the second channel, the selected signal segments having lengths corresponding to the duration of the default feature.

13. The method in accordance with the claim 12, characterized in that the selected segments have a duration greater than a predetermined minimum.

14. The method according to any of claims 12 or 13, characterized in that the cross correlations are related during a plurality of delay periods, to determine the magnitude of the delay.

15. The method in accordance with the claim 14, characterized in that an average equilibrium of the determined delay is recorded, the average being calculated from a predetermined number of delay measurements differing from each other by less than a predetermined value.

16. The method according to any of claims 12 to 15, characterized in that the channel to be verified for the initial dialogue signal is identified by verifying both channels for signals having the predetermined characteristics, and determining on which of the channels the Longer segments that have the default characteristics.

17. The method for measuring the echo path loss, characterized in that the echo delay is determined according to the method of any of claims 12 to 16.

18. The interference cancellation method, characterized in that it comprises detecting the interference by the method according to any of claims 12 to 17, and adding to the second channel a signal complementary to the signal detected on the first channel and having the same delay and attenuation as the detected interference signal.

19. An interference cancellation method in a telecommunications system, characterized in that it comprises a plurality of channel pairs comprising verifying each channel pair for interference by the method according to any of claims 12 to 18, and selecting one or more channel pairs having the strongest interference signals, and applying the second channel of each of said pairs, a signal complementary to the signal detected on the first channel and having the same delay and attenuation as that of the detected interference signal .

20. The method for verifying a telecommunications network, characterized in that it comprises detecting the presence of interference by a method according to any of claims 12 to 19, and determining, from the delay so measured, the location of the element of the network responsible for cause the interference.

21. The method according to any of claims 12 to 20, characterized in that the two channels form a communication link of two paths, the method being such that the interference detected on the second channel is an echo of the signal on the first channel.

22. An interference detection system, substantially as described herein, with reference to the drawings.

23. A method for detecting interference, substantially as described herein, with reference to the drawings.