KR100414911B1 - Method and apparatus for echo canceller during in wireless communication system - Google Patents

Abstract

The present invention relates to an echo canceling device and method having a dual filter structure for canceling echoes during double talk in a mobile communication system. In particular, according to an embodiment of the present invention, if the double talk is not detected, the filter coefficients are updated to the sample period for performing the adaptive processing. If double talk is detected, the filter coefficients are updated to the previous filter coefficients updated by the adaptive processing of the previous frame. An echo canceller and method for recovering were implemented. Therefore, by simplifying the double filter structure, it is possible to reduce the burden of the calculation amount and to obtain the effect of performing a stable operation during double talk. The reduced computation also reduces the cost of implementing echo cancelers per channel by increasing the number of channels of echo cancellers that can be implemented per digital signal processing (DSP).

Description

Echo cancellation device and method in mobile communication system {METHOD AND APPARATUS FOR ECHO CANCELLER DURING IN WIRELESS COMMUNICATION SYSTEM}

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and method for echo cancellation in a mobile communication system, and more particularly, to an echo cancelation apparatus and method having a double filter structure for canceling echoes during double talk.

In general, echo in a communication system is caused by a resistance mismatch of a four-to-two-wire converter called a hybrid in a public switched telephone network (PSTN). This reverberation acts as a major factor in reducing call quality in systems with high data delay, such as mobile communication systems.

Accordingly, in a communication system, an echo canceller is typically used to eliminate echoes caused by a hybrid. The echo canceller uses an adaptive filter to remove echo generated in the hybrid. The adaptive filter removes echoes by estimating echoes by subtracting the echo signals from echo signals transmitted from the hybrid and echo signals generated from the hybrid signal. That is, if A voice is used as the input of the adaptive filter, and the coefficients of the adaptive filter are continuously trained so that the output at this time is the same as the echo signal, the adaptive filter eventually has an input / output relationship like a hybrid. The echo of A is then passed through this filter, and the echo signal is predicted. If the predicted echo signal is subtracted from the original echo, the echo signal is removed.

Therefore, in order for the adaptive filter to function properly as an echo canceller, it is necessary to constantly update the coefficient to maintain the optimum state. In addition, the correlation between the echo source delivered to the hybrid and the echo generated in the hybrid and transmitted to the echo canceller should be high. To do this, the echo canceller should operate by accepting only the echo source and echo. However, this is impossible due to the characteristics of the communication system. This is because it is impossible to input only echo in the double talk state where the echo is mixed with the sound source of the caller and the other party.

Therefore, if the other party can only listen to the pure echo can be input, but when the other party starts talking, the echo is mixed with the voice of the other party. In this case, the echo mixed with the echo and the voice of the other party is provided as an input of the echo canceller, and the echo canceller should remove only the echo using this signal. However, due to the characteristics of the adaptive filter, in this case, the echo may not be eliminated, but rather, the echo may be generated to further reduce the call quality. That is, when the coefficient of the adaptive filter is continuously updated in the double talk state, the adaptive filter diverges.

In order to prevent this phenomenon, when the echo is mixed with the other party's voice (Double Talk), the adaptive filter should be stopped to prevent the degradation of the call quality. If so, it's important to distinguish between pure reflections and echoes mixed with the other's voice. In other words, how to distinguish whether or not it is a double talk situation is an important problem.

Conventionally, a method of comparing the levels of voice signals is used to determine double talk. The method described above is still widely used. As another method, a method of determining double talk by estimating a correlation coefficient between an echo source and an echo has been proposed. The method described above is used in some high performance echo cancellers. As another method, a method using a double filter structure has been proposed. This method compares the result of the adaptive filter with the result of the fixed filter by using the adaptive filter and the fixed filter in pairs, and makes a better result the output of the echo canceller. The structure of the echo canceller according to the method using the double filter structure is as shown in FIG. As shown in FIG. 1, it can be seen that the conventional dual filter structure is a structure for selecting a better result by comparing the results of the fixed filter and the adaptive filter. The method proposed in FIG. 1 is described in detail in patent application No. "P97-7674" filed by the present applicant.

However, looking at the problems that may be caused by the above-described methods, the method of comparing the level of the voice signal to determine the double talk is often a wrong decision. Because the level of the voice signal is not maintained at any level, it changes considerably fast and its characteristics are different depending on whether it is voiced or unvoiced. Therefore, it is impossible to judge accurately by simply comparing the levels. The method of determining the double torque by obtaining the correlation coefficient between the echo source and the echo designed to compensate for this disadvantage has a high accuracy, but a large amount of calculation to calculate the correlation coefficient. It is also difficult to determine the criteria for judging that the correlation coefficient is high. On the other hand, the method using the dual filter structure is excellent in performance, but it is a burden on the calculation amount because the two filters must be operated together every time the signal is input.

Accordingly, an object of the present invention to solve the above problems is to provide an apparatus and method for effectively canceling echo in a double talk state.

It is another object of the present invention to provide an apparatus and method having a double filter structure that reduces the amount of calculation for eliminating echo in a double talk state.

It is still another object of the present invention to provide an echo canceller and a method using a simple level comparison and a double filter structure together without any burden on calculation amount and excellent performance.

According to the present invention for achieving the above object, the filter coefficient is updated in a sample period in which the adaptive processing is performed when the double talk is not detected, and the filter coefficient is updated by the adaptive processing of the previous frame when the double talk is detected. We propose an echo cancellation device and method for restoring the previous filter coefficients.

That is, the present invention according to the first aspect for achieving the above object is a double talk detector for detecting the occurrence of a double talk in response to each of a plurality of samples constituting one input frame, and by the predetermined filter coefficient Adaptive processing of each sample constituting one input frame is performed to update the predetermined filter coefficient to each sample period. When the double talk is detected by the double talk detector, the predetermined filter coefficient is updated by a previous frame. An echo cancellation device including a filter for restoring the filter coefficients is implemented.

In addition, the present invention according to the second aspect for achieving the above object is a process of detecting a double talk corresponding to each of a plurality of samples constituting one input frame, and constitutes the input frame by a predetermined filter coefficient Performing the adaptive processing of the respective samples to update the predetermined filter coefficients in the respective sample periods, storing the predetermined filter coefficients updated in the respective sample periods in the predetermined periods, and when the double talk is detected, the predetermined filter. An echo cancellation method including a method of restoring coefficients to filter coefficients stored at the predetermined period is implemented.

1 is a view showing the configuration of an echo canceller having a conventional double filter structure.

2 is a view showing the configuration of an echo canceller having a double filter structure according to an embodiment of the present invention.

3 is a graph showing an experimental value of the echo canceller having a double filter structure according to an embodiment of the present invention.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

2 is a view showing the configuration of an echo canceller having a double filter structure according to an embodiment of the present invention. That is, FIG. 2 illustrates an echo canceller using a method of detecting double torque using a level comparison and a method using a dual structure of a fixed filter and an adaptive filter.

Referring to FIG. 2, the echo canceller according to an exemplary embodiment of the present invention, the hybrid 26 transmits a first voice signal (U signal) input from the A side to the B side, and is input from the B side. The second voice signal S is transmitted to the A side. At this time, the signal input to the echo canceller in the hybrid 26 includes the echo signal and the B-side near and echo signal.

The fixed filter 22 copies the filter coefficients stored in the adaptive filter 20 by the detection of the double torque from the level comparison double talk detector (not shown). That is, the fixed filter 22 provides the stored filter coefficients to the adaptive filter 20 when the double torque from the level comparison double talk detector is detected to restore the filter coefficients of the adaptive filter 20. However, if the double talk from the level comparison double talk detector is not detected, the operation of restoring the filter coefficients of the adaptive filter 20 is not performed. Notification based on whether or not the double talk is detected from the level comparison double talk detector is performed in units of samples. Therefore, the operation of copying the filter coefficients stored by the fixed filter 22 to the adaptive filter 20 may be generated in units of samples. In addition, the fixed filter 22 receives new filter coefficients from the adaptive filter 20 in units of frames (160 samples) and updates its filter coefficients. As described above, the fixed filter 22 stores the filter coefficients of the adaptive filter 20 and restores the filter coefficients of the adaptive filter 20 when necessary. Therefore, the fixed filter 22 according to the embodiment of the present invention has a relatively simple structure since it does not play a role of filtering the first voice signal (U signal), that is, the input voice, shown in the drawing. In this case, the calculation amount is reduced.

The adaptive filter 20 inputs the first audio signal (U signal) to perform an adaptive operation on a 160-by-sample basis for 160 samples constituting a frame, and is echoed by the adaptive operation. The prediction signal y1 is generated. The adaptive filter 20 adapts predetermined filter coefficients to the first audio signal U signal in order to generate the echo prediction signal y1. The filter coefficients of the adaptive filter 20 are restored to the filter coefficients from the fixed filter 22 or updated according to the NLMS adaptive algorithm. Restoration of the filter coefficients takes place in a sample period as previously described. On the other hand, when the adaptation of the frame is completed, the adaptive filter 20 transmits the updated filter coefficients according to the NLNS algorithm to the fixed filter 22 so that the filter coefficients of the fixed filter 22 can be updated. do. That is, the adaptive filter 20 updates the filter coefficients of the fixed filter 22 every one frame.

The canceller 24 inputs a reflection prediction signal y1 output from the adaptive filter 20 and a signal output from the hybrid 26, and outputs the echo prediction signal from the signal output from the hybrid 26. The echo cancellation signal e1 cancels y1).

Hereinafter, an operation according to an embodiment of the present invention will be described in detail with reference to the above-described drawings. Operation according to an embodiment of the present invention to be described later will be described by distinguishing between the case of the double talk and the case that is not inspected.

1. If double talk is not checked

First, the filter coefficients of the adaptive filter 20 and the fixed filter 22 are initialized to zero. The adaptive filter 20 then adapts the initialized filter coefficients to the first sample of one frame of the first audio signal (U signal) input from the A side. In this case, the adaptive filter 20 updates its filter coefficient in units of samples according to the NLMS adaptation algorithm. Thus, the filter coefficients updated by the previous sample are adapted for the next sample. By the adaptation, the adaptation filter 20 outputs the echo prediction signal y1 to the canceller 24. The adaptive filter 20 provides the updated filter coefficients to the fixed filter 22 when the adaptive operation for one frame is completed by the above-described operation. The fixed filter 22 stores filter coefficients provided from the adaptive filter 20. The operation of the above-described adaptive filter 20 has described the adaptive operation in the case where the occurrence of the double talk is not detected by the level comparison double talk detector as described above.

As described above, the filter coefficients of the fixed filter 22 in the case where the double talk does not occur are updated in one frame period in which the adaptive operation is performed by the adaptive filter 20 through repeating the above-described operation. That is, the filter coefficients stored in the fixed filter 22 will be referred to as filter coefficients updated by the previous frame from the adaptive filter 20. Therefore, as can be seen from the foregoing description, the adaptive filter 20 updates its filter coefficients with a sample period constituting one frame, and the fixed filter 22 performs the filter coefficients it has. Update at frame cycle.

2. When double talk is detected

Although not shown in the drawing, the level comparison double talk detector detects whether or not double talk is performed for each sample unit of the frame in which the adaptive operation is performed by the adaptive filter 20. If double talk is detected for a particular sample, the level comparison double talk detector provides to the fixed filter 22 that a double talk has been detected.

When the fixed filter 22 receives the double talk detection notification from the level comparison double talk detector, the fixed filter 22 provides the filter coefficient currently stored to the adaptive filter 20. The filter coefficients provided from the fixed filter 22 are filter coefficients updated by the previous frame from the adaptive filter 20.

The adaptive filter 20 will be provided with the filter coefficients from the fixed filter 22 while performing the adaptive operation on the sample by the filter coefficients updated according to the above-described NLMS adaptation algorithm. The adaptive filter 20 receiving the filter coefficients from the fixed filter 22 restores the current filter coefficients to the filter coefficients provided from the fixed filter 22. That is, the adaptive filter 20 performs an adaptive operation by restoring the filter coefficients updated by the sample period to the filter coefficients updated by the previous frame. The adaptive filter 20 reconstructing the filter coefficients performs an adaptive operation on the sample by the reconstructed filter coefficients. Of course, in performing the adaptive operation based on the reconstructed filter coefficients, the adaptive filter 20 performs the filter coefficient update in the sample period. The adaptive filter 20 provides filter coefficients of the fixed filter 22 by providing the filter coefficients updated at the completion time to the fixed filter 22 when the adaptive operation for all the samples constituting one frame is completed. Update

3 is a simulation of the performance of the double talk using the method proposed in the present invention, it is a simulation based on the TEST3B of ITU-T G.168. According to the requirements of ITU-T G.168, there is a provision that a double-talk situation occurs after the adaptive filter fully converges to an input signal of 0 dBm0, and the output signal should be below -20 dBm0 after a certain time. On the other hand, this method maintains -60dBm0, which is far more than the ITU-T G.168 specification.

As described above, an embodiment of the present invention proposes a method of differently applying filter coefficients of an adaptive filter based on whether double talk is detected. That is, when double talk is detected, it is proposed to use the filter coefficient updated by the previous frame as the filter coefficient of the adaptive filter.

Therefore, by simplifying the dual filter structure according to the embodiment of the present invention, it is possible to reduce the burden of the calculation amount and obtain an effect of performing a stable operation at the time of double talk. The reduced computation also reduces the cost of implementing echo cancelers per channel by increasing the number of channels of echo cancellers that can be implemented per digital signal processing (DSP).

Claims (6)

In the echo canceller which performs adaptive echo cancellation on a sample basis for a transmission speech frame in consideration of occurrence of double talk in the received speech frame, A double talk detector for detecting the occurrence of the double talk on a sample basis for the received speech frame composed of a plurality of samples; A filter for adapting an adaptive filter coefficient updated in each sample unit constituting the transmission speech frame to the transmission frame and outputting an echo prediction signal for removing the echo; The filter updates the fixed filter coefficients to the updated adaptive filter coefficients when the adaptation is completed for one transmission speech frame, and restores the adaptive filter coefficients to the fixed filter coefficients when the occurrence of the double talk is detected. Echo cancellation device characterized in that. The method of claim 1, wherein the double talk detector, The echo canceller, characterized in that for detecting the double talk by comparing the input voice signal level. The method of claim 1, wherein the filter, Outputs adaptive filter coefficients at the time when the adaptive echo cancellation is performed on the basis of the transmission speech frame, and when the fixed filter coefficients are provided while performing the adaptive echo cancellation, the adaptive filter coefficients at that time An adaptive filter for restoring to the fixed filter coefficients; And a fixed filter updating the fixed filter coefficients by the adaptive filter coefficients provided from the adaptive filter in the transmission speech frame period and providing the fixed filter coefficients to the adaptive filter when the double torque is detected. Echo Canceller. In a method of performing adaptive echo cancellation on a sample basis for a transmission speech frame in consideration of occurrence of double talk in a reception speech frame, Detecting generation of the double talk on a sample-by-sample basis for the received speech frame consisting of a plurality of samples; Outputting an echo prediction signal for the echo cancellation by adapting an adaptive filter coefficient updated in each sample unit constituting the transmission speech frame to the transmission speech frame; Updating a fixed filter coefficient with the updated adaptive filter coefficient at the time when echo cancellation for one transmission speech frame is completed by the echo prediction signal; And restoring the adaptive filter coefficients to the fixed filter coefficients when the occurrence of the double talk is detected. The method of claim 4, wherein And the double talk is detected by comparing the input voice signal levels. delete