• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    estimation and suppression of harmonic loudspeaker nonlinearities is a harmonic echo power estimator configured to estimate the echo power generated by harmonic loudspeaker nonlinearities in user equipment that has an echo path between a speaker input and a microphone output. this estimator includes a frequency range mapper (40) configured to map each frequency range into a set of speaker output frequency ranges (blsp) in a frequency array in the set (b (blsp, k)) , where each frequency range in the set is mapped to several frequency boxes in the corresponding arrangement. a power estimator (42) is configured to determine a power estimate (px (blsp, k))) in each frequency range arrangement in a corresponding estimate (px, ml (blsp)) of high-power input speaker that generates harmonic speaker nonlinearities.
    公开号:BR112013016222B1
    申请号:R112013016222-8
    申请日:2011-02-03
    公开日:2020-12-15
    发明作者:Anders K. Eriksson；Per Áhgren
    申请人:Telefonaktiebolaget L M Ericsson (Publ)；
    IPC主号:
    专利说明:

    Technique Field
    [001] The present invention refers to the echo cancellation in user equipment for communication systems and, in particular, the estimation and suppression of harmonic speaker non-linearities generated in such equipment. Foundations
    [002] An echo subtractor consists of one of the key components of the echo canceller. It is also distinguished from a pure echo suppressor, which only attenuates the signal when echo is present. The main benefit of an echo canceller is the improved performance in situations with simultaneous speech from both ends of the communication (called double talk) and also an increased transparency to the low-level near-end sound, which increases the naturalness of the conversation.
    [003] Echo subtraction is usually implemented using a linear model, mainly due to the fact that a linear model is computationally simple to estimate, but also due to the fact that it is very difficult to find a suitable non-linear model that works in general. For these reasons, echo subtraction cannot generally remove non-linear echoes that originate from non-linearities in the echo path.
    [004] Another key component in an echo canceller consists of a residual echo suppressor, which reduces any residual echoes present at the output from the echo subtractor to such an extent that the echo attenuation requirements imposed by the relevant standards are met. met, and at such a level that the residual echo is not noticeable in the presence of the near end signal. However, since the suppression performed by the residual echo suppressor also affects the desired near-end signal if the frequency content of the near-end signal and the residual echo are overlapping, the suppression performed by the residual echo suppressor should be as small as possible, as the loss of transparency (of the near-end signal) introduced by this component is directly related to the amount of suppression performed.
    [005] The harmonic overtones at the speaker output caused by non-linearities will be picked up by the microphone as non-linear echoes. These echoes also need to be removed by the echo canceller. However, as the echo subtractor is based on a linear model of the echo path, the echo subtractor cannot reduce non-linear echoes. These must therefore be removed by the residual echo suppressor. In order to do this, the residual echo suppressor needs an estimate of the power of the nonlinear echoes. In addition, this estimate has to be accurate, since otherwise the residual echo suppressor needs to perform the extra suppression (plan for the worst possible scenario) in order to compensate for the uncertainty in the nonlinear echo power estimate. This will then result in the reduced echo canceller transparency of the near-end signal, which is undesirable.
    [006] A class of methods [1-4] for modeling harmonic speaker nonlinearities is based on a Volterra model that uses powers of the speaker input signal. This is, however, computationally very complex. Additionally, the harmonics produced by the Volterra model are typically irregular, so an upward / downward sampling scheme is necessary to prevent the irregularity from affecting the power estimate of the harmonic speaker nonlinearities, which makes the solution base from Volterra even more complex. summary
    [007] An objective of the present invention is the computationally simple estimation of echo power that originates from harmonic speaker non-linearities.
    [008] Another objective of the present invention is the suppression of echo power that originates from harmonic speaker non-linearities.
    [009] These objectives are achieved according to the attached claims.
    [0010] According to a first aspect, the present invention involves a method of estimating echo power generated by harmonic speaker non-linearities in user equipment that has an echo path between a speaker input and a microphone output. This method includes the following steps: Each frequency range in a set of speaker output signal frequency ranges is mapped into a corresponding arrangement of speaker input signal frequency ranges, with each range of frequency in the set is mapped into various frequency ranges in the corresponding arrangement. A power estimate is determined for each speaker input signal in each frequency range arrangement. The power estimates determined in each frequency range arrangement are combined into a corresponding estimate of speaker input power that generates harmonic speaker nonlinearities. Speaker input power estimates are transformed through the echo trajectory into echo power estimates generated by harmonic speaker nonlinearities.
    [0011] According to a second aspect, the present invention involves an echo suppression method that uses a frequency selective filter based on the ratio between a power estimate of a near-end signal and an estimate of power of a signal of eco. This method includes the following steps: A power estimate of a residual echo signal from an echo subtractor is determined. An echo power estimate generated by harmonic speaker nonlinearities is determined according to the first aspect. The echo signal power estimate is formed by adding the residual echo signal power estimate to the echo power generated by harmonic speaker non-linearities.
    [0012] According to a third aspect, the present invention involves a harmonic echo power estimator configured to estimate the echo power generated by harmonic loudspeaker nonlinearities in user equipment that has an echo path between a speaker input and microphone output. The harmonic echo power estimator includes the following elements: A frequency range mapper configured to map each frequency range into a set of speaker output signal frequency ranges in a corresponding arrangement of signal frequency ranges speaker input, with each frequency range in the set mapped to various frequency ranges in the corresponding arrangement. A power estimator configured to determine a power estimate for each loudspeaker input signal in each frequency range arrangement. A power estimator combiner configured to combine power estimates determined in each frequency range arrangement into a corresponding speaker input power estimate that generates harmonic speaker nonlinearities. A power estimator transformer configured to transform the speaker input power estimates through the echo path into echo power estimates generated by the harmonic speaker nonlinearities.
    [0013] According to a fourth aspect, the present invention involves an echo canceller that has a residual echo suppressor that uses a frequency selective filter based on the ratio between a power estimate of a near-end signal and an estimate power of an echo signal. The echo canceller includes the following elements: A power estimator configured to determine a power estimate of a residual echo signal from an echo subtractor. A harmonic echo power estimator according to the third aspect configured to determine an echo power estimate generated by harmonic speaker nonlinearities. An adder configured to add the residual echo signal power estimate to the echo power estimate generated by harmonic speaker non-linearities.
    [0014] In accordance with a fifth aspect, the present invention involves user equipment that includes an echo canceller according to the fourth aspect.
    [0015] An advantage of the present invention consists in the fact that it provides the computationally simple estimation of echo power that originates from harmonic speaker nonlinearities with the use of a limited number of parameters.
    [0016] Another advantage of the present invention is the fact that it fits perfectly in range schemes normally used in a residual echo suppressor, which is typically a component in an echo canceller where the estimate of non-linear echo power is used. Brief Description of Drawings
    [0017] The invention, together with its added benefits and objectives, can be better understood by referring to the following description taken in conjunction with the attached drawings, in which: Figure 1 is a block diagram illustrating the general principles of a conventional echo canceller; Figure 2A-B is a diagram illustrating typical frequency responses of the frequency selective filter applied to the residual echo from an echo subtractor; Figure 3 is a diagram that illustrates the phenomenon of harmonic speaker non-linearities; Figure 4 is a block diagram showing an embodiment of an echo canceller in accordance with the present invention; Figure 5 is a flow chart illustrating a modality of a method of estimating echo power generated by harmonic speaker non-linearities in accordance with the present invention; Figure 6 is a block diagram that illustrates a modality of a harmonic echo power estimator according to the present invention; Figure 7 is a flow chart illustrating an embodiment of an echo suppression method according to the present invention; Figure 8 is a block diagram illustrating a modality of user equipment in accordance with the present invention; Figure 9 is a block diagram illustrating a modality of a harmonic echo power estimator according to the present invention; and Figure 10 is a block diagram showing an embodiment of an echo canceller according to the present invention. Detailed Description
    [0018] Figure 1 is a block diagram that illustrates the general principles of a conventional echo canceller. The received signal X (t) is passed to a speaker 10 and forms an unwanted echo signal on a microphone 12. This echo is picked up by the microphone together with a desired near-end signal V (t) as the microphone signal Y (t) An EP echo path is formed between the speaker input and the microphone output. An echo subtractor 14 uses an adaptive model to form an estimate S (t) of the echo signal v 'in an echo predictor 16. The echo in the. , y (t). .. ... _. . ... s (t). y (t) _ _microphone J 'and then reduced by subtracting v' from v 7 in umÍ f) 6 (i) adder 18. Finally, any residual echoes '' present at the output '' from the echo canceller they are suppressed by a residual echo suppressor 20, thus producing the final output signal 0UT 'from the echo canceller.
    [0019] The residual echo suppressor 20 is typically implanted so that the echoes IS In residuals are suppressed using a frequency selective filter. The characteristics of the frequency response & frequency selective filter applied by the residual echo suppressor 20 depend on the estimated spectral characteristics Pv (t, f) .v (t) P (t, f). s (í) _. . . . . . . ,. .d7 d of '' and '' of ''. Typically, if for a given frequency f 'P (t f'] »P (tf] we have vv 7 s' 1, that is, the near-end signal is much stronger than the residual echo signal, then, & would be closer to 1 (almost no attenuation). On the other hand, if we had v 7 '7 s' '', that is, the near-end signal is approximately equal to the residual echo signal, then G (t, f) would be typically chosen to be small (significant attenuation).
    [0020] Generally, it is desired to have a stable continuous behavior from the passage through the signal to significantly suppress the signal. Such behavior will eliminate distortions caused by discontinuities in the course of time. Typically, this is achieved by making 0 proportional to the ratio between ePs (t, f) ePv (t, f)

    [0021] Figure 2A-B illustrates examples of the function that represents the frequency response G (t, f) of the frequency selective filter applied by the residual echo suppressor20. Clearly, the format of the function illustrated in Figure 2B is preferred from a point of view of smoothness.
    [0022] Since divisions are computationally complex to perform, many real-time echo canceller realizations. Computation is typically performed over frequency bands to minimize the number of divisions required to compute ^. If a uniform bandwidth # is used for the range, it can be. So. approximate as:

    [0023] In this way, G (t, f) is approximated by a constant function composed of variables G (t, b).
    [0024] A typical band scheme would consist of using uniformly for a frequency range from 0 to 4000 Hz. To simplify the discussion in this document. This band scheme will generally be assumed. but the present invention is by no means restricted to this particular banding scheme. In this way, it can be greater or less than the example given. Another possibility is to let it vary over the frequency range. As an example. It could be smaller in the middle of the diagram in Figure 2B than near the end points to represent the variation in that region.
    [0025] A common type of non-linearity in speakers generates harmonic overtones at the speaker output. Figure 3 is a diagram that illustrates the phenomenon of such harmonic speaker nonlinearities in an UE. The diagram shows a fundamental (sinusoidal) variation in frequency from 4000 Hz to 0 Hz over the given period of time. For such a sinusoidal speaker input. the speaker output contains harmonic overtones at 4 * • • • times the fundamental frequency. These overtones are generated in a non-linear manner. since the speaker input does not contain any (or very low) power at those frequencies. The power of the different tones has been illustrated by the thickness of the lines in Figure 3. Thus, in the example. the second overtone is stronger than the first and third overtones. but weaker than the fundamental tone.
    [0026] A viable method for computing the harmonics at the speaker output could be based on the harmonics at the speaker input. Solutions based on such an approach would. However. require a complete spectral estimate of the speaker input to estimate the non-linear speaker output. rendering. like this. the computationally complex method.
    [0027] With reference again to Figure 1., the harmonic overtones at the loudspeaker output will be captured by microphone 12 as non-linear echoes. These echoes must be removed by the echo canceller. However, since the echo subtractor 14 is based on a linear model of the EP echo path, the echo subtractor cannot reduce non-linear echoes. These therefore need to be removed by the residual echo suppressor 20. In order to do this, the residual echo suppressor needs an estimate of the power of the nonlinear echoes. Additionally, the estimation of the power of the non-linear echoes needs to be accurate, since otherwise the ecoresidual suppressor needs to perform the extra suppression in order to compensate for the uncertainty in the estimation of the non-linear echo power. This will result in ecoreduced cancellation transparency of the near-end signal, which is undesirable.
    [0028] Figure 4 is a block diagram showing an embodiment of an echo canceller according to the present invention. The echo canceller includes a residual echo suppressor 50 that uses a frequency selective filter, as described above, based on the ratio of a power estimate of a near-end signal to an estimate of power of the echo signal. The difference is that the echo power estimate now also needs to include an echo power estimate generated by harmonic speaker nonlinearities in addition to the linear echo power estimate. In this way, the frequency selective filter ^ is represented as:
    P (t, b) represents the power estimate of the near-end signal, Ps, ni (t, b) represents the linear echo power estimate (represented as Ps (t, b) in equation (2)), that is that is, an estimate of the power of the residual echo signal from the echo subtractor 14, and represents the estimate of echo power generated by harmonic speaker nonlinearities.
    [0029] Again with reference to Figure 4, the echo canceller, therefore, additionally includes a harmonic echo power estimator 30 configured to determine the echo power estimate generated by harmonic speaker non-linearities. From now on, time dependence on power estimates will be reduced to avoid clutter in the equations. However, it must be remembered that the power estimates in the equations given below depend on time, as well as the frequency range.
    [0030] From what has been exposed, it is evident that an important aspect of the invention consists of estimating the power of the non-linear echo in the microphone signal caused by the non-linearity of the harmonic speaker. The estimation is performed on the harmonic echo power estimator 30 in a banded manner, preferably compatible with the band structure of the residual echo suppressor.
    [0031] Figure 5 is a flow chart illustrating a modality of a method of estimating the echo power generated by harmonic speaker non-linearities in accordance with the present invention. The harmonic echo power estimator 30 is configured to implement this functionality.
    [0032] Step S1 maps each frequency range into a set of speaker / sp signal output frequency ranges in a corresponding arrangement of frequency bands b (b., K} of loudspeaker input signal). speaker, where each frequency band in the set is mapped to various frequency bands in the corresponding arrangement. The purpose of this X (The step is to determine which bands in the input signal that can actually b, k = 1 2 3 produce an overtone in the speaker output range 'sp. Here,' '' ••• denotes the overtone number. An example of this mapping (and how it can be done) is given in table 1 in APPENDIX 1. From this table, you can It should be noted that a speaker / sp output range can include overtones generated by several input signal ranges (several), so mapping typically consists of “one to many” ranges, especially for the larger sp / ranges On the other hand, for smaller bands, many frequency bands in the arra Matching njo may actually consist of the same track.
    [0033] Step S2 determines a power estimate Px (b (bisp, K) for each speaker input signal in each frequency range arrangement. Thus, this step determines a power estimate for each signal input in ranges that can generate an overtone in the lsp speaker output range.
    [0034] Step S3 combines the power estimates determined Px (b (bisp, K) in each frequency range arrangement in a corresponding estimate of iΦisp) of speaker input power that generates harmonic speaker nonlinearities . Thus, this step determines a total power estimate of input signal components that generate overtones in the lsp speaker output range.
    [0035] In a preferred embodiment, the step of combining S3 can be based on the combination:
    where denotes the estimated input power of the speaker in the frequency range of the speaker output signal, denotes the mapping, denotes the power estimates determined in the frequency range of the speaker input signal denotes predetermined coefficients, maximum number of terms to be included in each combination.
    [0036] The maximum number of terms corresponds to the maximum number of overtones to be considered, for example, it is in the range 3 to 9. It has been found that it generates reasonable storage and complexity requirements and seems to be sufficient for most speakers that display harmonic nonlinearities. Likewise, relatively few coefficients are needed to specify the behavior, while still retaining good control over the speaker model. The actual values for the coefficients are different for different types of speakers. Typically, the actual values are determined from the speaker input and output spectrogram estimates, where the input consists of a smooth sinusoid.
    [0037] In one mode, only determined power estimates that exceed a predetermined power threshold are combined. Power estimation represents a minimum level below which spectral components do not generate nonlinear harmonics. This modality also implies a reduction of additional complexity. The threshold can be found by smooth sinusoidal frequencies of different levels and observing at which level nonlinearities stop occurring.
    [0038] In another modality, only the terms C (bis, k) Px (b (bis, k)) that exceed another predetermined threshold are included in the sum. In this modality, the power estimates determined weighted by the threshold coefficients, which means that only the most important terms in the sum are retained.
    [0039] Step S4 transforms the speaker input power estimates (^ Asp) through the EP echo path into echo power estimates generated by the harmonic speaker nonlinearities. The transformation can be performed by multiplying the speaker input power estimates with the squared magnitude of an estimate of the frequency response of the EP echo path according to:
    where NBANDS is the number of frequency bands. In an echo canceller, H (b) is typically known from an adaptive filter in the echo trajectory impulse response estimator of echo subtractor 14. If no estimate is available, it can be readily estimated from the characteristics frequency of the speaker input and microphone signal outputs.
    [0040] Again with reference to Figure 4, in the illustrated modality, the estimator of x (7) harmonic echo power 30 uses the speaker input signal and the estimate from the echo subtractor 14 to produce the estimate of nonlinear echo power, as will now be described with reference to Figure 6.
    [0041] Figure 6 is a block diagram that illustrates a modality of a harmonic echo power estimator 30 that implements the described method.
    [0042] A frequency range mapper 40 is configured to map each frequency range into a set of lsp speaker output signal frequency ranges in a corresponding arrangement of speaker input signal frequency ranges b I b ki, where each frequency range in the set is mapped into various frequency ranges in the corresponding arrangement. The frequency range mapper 40 can, for example, be deployed as a predetermined query table, such as Table 1 in APPENDIX 1
    [0043] A power estimator 42 that receives the mapped frequency bands of speaker input signal ^ lspl and speaker input signal X is configured A lb (bISD, k \ to determine an estimate of 77 vx power of each speaker input signal in each frequency range arrangement.
    [0044] A power estimator 44 connected to the IPX power estimator (b (b, SD, k} in each frequency range arrangement in a corresponding estimate dde1 ^ 1 ^ of speaker input power that generates harmonic speaker non-linearities, c (b,, k} for example, according to equation (4). The predetermined coefficients can be stored in a lookup table.
    [0045] A power estimator transformer 46 connected to the power estimator combiner 44 is configured to transform the speaker input power estimates through the EP echo path into echo power estimates generated by the high non-linearities. harmonic speaker. The transformation can be performed according to equation (5). The estimate of the frequency response of the EP echo path can, for example, be obtained from the echo subtractor 14, as shown in Figure 4.
    [0046] As previously described, in one embodiment, the power estimation combiner 44 can be configured to include only determined power estimates that exceed a predetermined power threshold in the combination (4).
    [0047] In another modality, the power estimation combiner 44 can be configured to include only those terms that exceed a predetermined threshold in the sum (4).
    [0048] The power estimates are routed to a residual echo suppressor 50. The residual echo suppressor 50 includes two power estimators 52 and 54. The functionality of the power estimators 52 and 54 will be described only briefly below, since these elements are typically found in conventional residual echo suppressors. X (í 1
    [0049] The power estimator 52 receives the input signal from the speaker and the estimate of the frequency response of the EP echo path. Using these entities, the power estimate is determined. This estimate is sent to an adder 56, which adds it to the power estimates of the 1 ^ echo generated by the harmonic speaker non-linearities.
    [0050] The power estimator 5 receives the signal K W + s W from the echo subtractor 14 and forms a power estimate of the near end signal ''.
    [0051] The output power estimates from the power estimator 54 and the adder 56 are sent to a frequency selective filter 58 represented and (t] by the function in equation (3), such filter produces the output signal 0UT ''
    [0052] Figure 7 is a flowchart illustrating an embodiment of an echo suppression method according to the present invention. Step S10 determines an estimate of the power of a residual echo signal from an echo subtractor. Steps S1 to S4, which are explained in more detail with reference to Figure 5 above, determine an echo power estimate generated by harmonic speaker non-linearities. Step S11 forms the echo signal power estimate by by adding the power estimate of the residual echo signal to the power estimate ^ s, n / (^) of the echo generated by harmonic speaker non-linearities.
    [0053] Figure 8 is a block diagram that illustrates a modality of user equipment in accordance with the present invention. An echo canceller 60 according to the present invention is connected to an antenna on a radio 62 and a speech encoder / decoder 64. The radio performs upward / downward conversion, amplification and conventional channel decoding. Speech encoder / decoder 64 performs conventional speech encoding / decoding. Since both elements 62, 64 are conventional units, they will not be described in further detail.
    [0054] The steps, functions, procedures and / or blocks described in this document in this document can be implanted in hardware using any conventional technology, such as integrated circuit technology or discrete circuit, which includes both set of electronic circuits general purpose as a set of specific application circuits.
    [0055] Alternatively, at least some of the steps, functions, procedure and / or blocks described in this document can be implemented in software for execution by a suitable processing device, such as a microprocessor, digital signal processor (DSP) and / or any suitable programmable logic device, such as a field programmable port arrangement (FPGA) device.
    [0056] It should also be understood that it may be possible to reuse the general processing capabilities of the UE. This can, for example, be done by reprogramming existing software or by adding new software components.
    [0057] As an example of implantation, Figure 9 is a block diagram illustrating an example modality of a harmonic echo power estimator 30 according to the present invention. This modality is based on a processor 100, for example, a microprocessor, which executes a software component 110 for the frequency range mapping, a software component 120 for power estimation, a software component 130 for the combination of power estimates and a software component 140 for power estimate transformation. These software components are stored in memory 150. Processor 100 communicates with memory over a system bus. The speaker input signal X ^ and the estimate of the frequency response of the EP echo path are received by an input / output controller (I / O) 160 that controls an I / O bus, to which the processor 100 and memory 150 are connected. In this mode, the parameters received by the I / O controller 160 are stored in memory 150, where they are processed by the software components. The software component 110 can implement the functionality of block 40 in the modalities described above. The software component 120 can implement the functionality of block 42 in the modalities described above. The software component 130 can implement the functionality of block 44 in the modalities described above. Software component 140 can implement the functionality of block 46 in the modalities described above. The power estimate obtained from software component 140 is output from memory 150 by I / O controller 160 over the I / O bus.
    [0058] Figure 10 is a block diagram illustrating an embodiment of an echo canceller 60 in accordance with the present invention. In addition to the software components 110 to 140 described with reference to Figure 9, memory 150 also includes a software component 200 for echo prediction and subtraction, a software component 210 for adding power and a software component 220 for echo suppression. The software component 200 can implement the functionality of block 14 in the modalities described above. The software component 210 can implement the functionality of block 56 in the modalities described above. The software component 220 can implement the functionality of block 58 in the modalities described above. The speaker input signal% and the microphone signal are received by the controller and (I) from I / O 160 on the I / O bus and the output signal O '7r' from the echo canceller is routed to the encoder on the I / O bus.
    [0059] In the modalities of Figure 9 and 10, it is assumed that other tasks, such as demodulation, channel encoding / decoding and speech encoding / decoding in one UE, are handled elsewhere in the UE. However, an alternative is to allow additional software components in memory 150 to also handle all or part of these tasks.
    [0060] In the case where the UE is a computer that receives voice over Internet protocol (IP) packets, the IP packets are typically routed to the I / O controller x (t} 160 and the input signal from speaker is extracted by the additional software components in memory 150.
    [0061] Non-limiting examples of typical UEs, where the present invention can be used consist of: personal computers (stationary or notebook type), netbooks, tablet PCs, mobile internet devices, smart phones, phones with resources.
    [0062] Part or all of the software components described above can be loaded onto a computer-readable medium, for example, a CD, DVD or hard drive, and loaded into memory for execution by the processor.
    [0063] Since harmonic speaker nonlinearities occur mainly for narrow band type speaker input signals, these types of signals can be detected in order to determine when the method described for nonlinear speaker speaker should be used. In order to do this, various types of signals can be detected, and if any of these types are present, the method is used, otherwise it is not used. Such types of signals consist, for example, of harmonic and non-stationary signals.
    [0064] To detect harmonic signals that include several narrowband components that can trigger non-linearity, the following inspired detection method H (í, / í) pit f) in Cepstrum can be used. The periodogram * of the 32 minor bins of is computed in order to detect the presence of narrow band harmonics:

    [0065] The reason for using only the 32 minor bins for periodogram computation is that harmonics are usually more prominent for these bins, and the inclusion of more bins would result in a less accurate estimate.
    [0066] The uniformity of * is then estimated as the number of bins in * 0.7ma> ^ R (í, k) as it exceeds a threshold of *. If this number exceeds 2, then the signal is considered to include harmonics.
    [0067] This detection scheme for non-stationary signals can be used to capture the beginning of harmonic signals, which are sometimes lost by the above technique. These are characterized by a change in signal statistics and are detected as non-stationary in the signal. The detection technique detects non-seasonality as a significant deviation from the average power and is performed as shown below:

    [0068] Elements skilled in the art will understand that various modifications and changes can be made to the present invention without departing from the scope of the same, which is defined by the appended claims. APPENDIX 1
    [0069] Since an overtone has to be an integer multiple of the fundamental frequency, the overtones in a given speaker output range will originate from an arrangement of speaker input ranges. Table 1 below is an example mapping based on a 250 Hz equidistant bandwidth for each frequency range. Table 1


    [0070] As an additional example, the following MATLAB® code can be used to determine a similar mapping with a sampling frequency of 48 kHz, a band structure that uses 250 Hz / band and 6 overtones: f0 = 0: 250: (24000-250); f0 = f0 '; f1 = 249: 250: (24000); f1 = f1 '; M = 1+ [floor (f0 / 2/250) floor (f1 / 2/250) floor (f0 / 3/250) floor (f1 / 3/250) floor (f0 / 4/250) floor (f1 / 4 / 250), floor (f0 / 5/250) floor (f1 / 5/250) floor (f0 / 6/250) floor (f1 / 6/250) floor (f0 / 7/250) floor (f1 / 7 / 250)]; M = M (:, 1: 2: end); APPENDIX 2
    [0071] The harmonic nonlinearities in the speaker are modeled by the relative amplitude of the harmonics (in relation to the fundamental), denoted, and a gain factor that describes the intensity of the nonlinearities produced by a given frequency.
    [0072] An estimated spectrum of non-linearities is computed from the speaker output signal spectrum as:
    where the vectors are determined from according to the diffusion of the overtones (the frequency mapping is described in APPENDIX 1): f1, [0 -250 [: / Jl (fl) + ... + hk (f,) f2, [250-500 [: / 71 (fl) + ... + MO f3, [500 -750 [: / J1 (f2) + / 72 (f1) + ... + hk (Q f4, [750 -1000 [: / 7l (f2) + / 72 (f2) + / 73 (fl) + / 74 «) + ... + MO f5, [l000 -1250 [: / 7l (/ - 3) + / 72 (/ -2) + / 73 (/ 2) + / 74 (/ - l) + / 75 (fl) + ... + hk (Q f6, [l250 -1500 [: / 7l (f3) + / 72 (f2) + / 73 (/ 2) + / 74 (f2) + / 75 (f2) + / 76 (fl) + ... + / 7JO f7, [1500-1750I: / Jl (f4) + h2 ( f3) + / J3 (f2) + / J4 (/ 2) + / J5 (f2) + / 76 (f1) + ... + hk (V / 8, [l750-2000 [: / Jl (f4) + / 72 (f3) + / J3 (f2) + / J4 (f2) + / 75 (f2) + / J6 (^) + MO + - + ^ (O Therefore:

    [0073] The relative amplitude of the harmonics and the fundamental gain factors Yn should be selected according to the non-linearity produced by the speaker.
    [0074] The coefficients 'p' in equation (4) are formed by products of these parameters. Abbreviations DSP FFT digital signal processor Fast Fourier Transform FPGA Arrangement of programmable ports by I / O field IP Input / Output EU Internet Protocol User equipment References [1] A. Stenger, R. Rabenstein, “ADAPTIVE VOLTERRA FILTERS FOR NONLINEAR ACOUSTIC ECHO CANCELLATION ”, http://www.ee.bilkent.edu.tr/~signal/Nsip99/papers/146.pdf [2] G. Budura, C. Botoca,“ Nonlinearities Identification using The LMS Volterra Filter ”, http://hermes.etc.upt.ro/docs/cercetare/articole/BudBot05.pdf [3] Hongyun Dai, Wei-Ping ZhuI, “Compensation of Loudspeaker Nonlinearity in Acoustic Echo Cancellation Using Raised-Cosine Function”, EEE TRANSACTIONS ON CIRCUITS AND SYSTEMS — II: EXPRESS BRIEFS, VOL. 53, NO. 11, NOVEMBER 2006 [4] H. Schurer, CH Slump, OE Herrmann, “Second Order Volterra Inverses for Compensation of Loudspeaker Nonlinearity”, November 2009, http://doc.utwente.nl/17422/1 / 00482982. pdf
    权利要求:
    Claims (13)
    [0001]
    1. Method of estimating the echo power generated by harmonic speaker non-linearities in user equipment that has an echo path between a speaker input and a microphone output, characterized by the fact that said method includes the steps of mapping (S1) each frequency range into a set of speaker output signal frequency ranges into a corresponding frequency range arrangement (b (blsp, k)] x ", with each range being frequency in the set is mapped into various frequency bands in the corresponding array; determine (S2) an estimate of power 'y / z of each input signal in each array of frequency bands; combine (S3) power estimates determined in each array of frequency ranges in a corresponding estimate (^ xnl ^ lsp ^) of speaker input power that generates harmonic speaker nonlinearities; transform (S4) the power estimates (^ ■ Π / ^ / SP ^) speaker input are of the echo trajectory (EP) in power estimates of the echo generated by the nonlinearities of the harmonic speaker.
    [0002]
    2. Method, according to claim 1, characterized by the fact that the combining step (S3) is based on the combination:
    [0003]
    3. Method according to claim 1 or 2, characterized by the fact that only determined power estimates that exceed a predetermined power threshold are combined.
    [0004]
    4. Method, according to claim 2, characterized by the fact that only terms that exceed a predetermined threshold are included in the sum.
    [0005]
    5. Method according to claim 1, 2, 3 or 4, characterized by the fact that the transform step (S4) multiplies the speaker input power estimates with the squared magnitude of an estimate of the response frequency of the echo path (EP).
    [0006]
    6. Echo suppression method using a frequency selective filter based on the ratio between a power estimate of a near-end signal and an power estimate of an echo signal, characterized by the fact that it includes the steps to determine ( S10) a power estimate of a residual echo signal from an echo subtractor (14); determine (S1-S4), according to any one of claims 1 to 5, an estimate of echo power generated by non-linearities of the speaker; form (S11) the power estimate of the echo signal by adding the power estimate ^ s, / ^ of the residual echo signal to the estimate of echo power generated by harmonic speaker nonlinearities.
    [0007]
    7. Harmonic echo power estimator configured to estimate the echo power generated by harmonic speaker non-linearities in user equipment that has an echo path between a speaker input and a microphone output, characterized by fact that said harmonic echo power estimator includes a frequency range mapper (40) configured to map each frequency range into a set of speaker output signal frequency ranges ^ lsp ^ in a corresponding arrangement of speaker input signal frequency ranges (b [bls., k]] 'x' ', with each frequency range in the set being mapped into various frequency ranges in the corresponding arrangement; a power estimator ( 42) configured to determine a power estimate of each speaker input signal in each frequency range arrangement; a power estimate combiner (44) configured to match (px (bk}) power estimates determined 'x "' in each frequency range arrangement in an estimate corresponding to π / ^ / sp ^ of speaker input power that generates harmonic speaker non-linearities; a power estimator transformer (46) configured to transform the speaker input power estimates through the echo trajectory (EP) into power estimates of the echo generated by the speaker nonlinearities.
    [0008]
    8. Harmonic echo power estimator, according to claim 7, characterized by the fact that the power estimation combiner (44) is configured to generate a combination based on
    [0009]
    9. Harmonic echo power estimator according to claim 7 or 8, characterized by the fact that the power estimation combiner (44) is configured to include only determined power estimates that exceed a predetermined power threshold in the combination.
    [0010]
    10. Harmonic echo power estimator, according to claim 8, characterized by the fact that the power estimation combiner (44) is configured to include only terms that exceed a predetermined threshold in the sum.
    [0011]
    11. Harmonic echo power estimator according to claim 7, 8, 9 or 10, characterized by the fact that the power estimation transformer (46) is configured to multiply the speaker input power estimates with ade echo.
    [0012]
    12. Echo canceller that has a residual echo suppressor that uses a frequency selective filter based on the ratio between a power estimate of a near-end signal and an power estimate of an echo signal, characterized by the fact that said echo canceller includes a power estimator (52) configured to determine a power estimate of a residual echo signal from an echo subtractor; a harmonic echo power estimator (30), according to any of claims 7 to 11, configured to determine an echo power estimate generated by harmonic speaker non-linearities; an adder (56) configured to add the residual echo signal power estimate to the echo power estimate generated by non-harmonic speaker linearities.
    [0013]
    13. User equipment, characterized by the fact that it includes an echo canceller (60) as defined in claim 12.
    类似技术:
    公开号 | 公开日 | 专利标题
    BR112013016222B1|2020-12-15|METHOD OF ESTIMATING ECO POWER, ECO SUPPRESSION METHOD, HARMONIC ECO POWER ESTIMATOR, ECO CANCELER, AND, USER EQUIPMENT
    CN107211063B|2020-07-21|Nonlinear echo path detection
    EP3080975B1|2017-07-12|Echo cancellation
    US9653091B2|2017-05-16|Echo suppression device and echo suppression method
    US20100208908A1|2010-08-19|Echo supressing method and apparatus
    WO2007049643A1|2007-05-03|Echo suppressing method and device
    WO2012153452A1|2012-11-15|Echo erasing device and echo detection device
    US9918163B1|2018-03-13|Asynchronous clock frequency domain acoustic echo canceller
    EP3929919A1|2021-12-29|Voice signal processing method and device, apparatus, and readable storage medium
    Bendersky et al.2008|Nonlinear residual acoustic echo suppression for high levels of harmonic distortion
    US11043228B2|2021-06-22|Multi-microphone signal processing apparatus, method, and program for wind noise suppression
    US9674368B2|2017-06-06|Method and apparatus for audio processing
    JP4879195B2|2012-02-22|Noise reduction device
    JPWO2012157783A1|2014-07-31|Audio processing apparatus, audio processing method, and recording medium recording audio processing program
    WO2013032001A1|2013-03-07|Speech processor, contrl method, and control program thereof
    JP4631933B2|2011-02-23|Echo suppression method
    JP5161838B2|2013-03-13|Echo canceller and acoustic echo canceling method
    Muhammed Shifas et al.2018|Robust acoustic echo suppression in modulation domain
    JP2011170261A|2011-09-01|Speech enhancing device, speech enhancing program
    JP6314608B2|2018-04-25|Echo suppression device, echo suppression program, and echo suppression method
    JP5049256B2|2012-10-17|Echo canceller and echo canceling method
    US20120140939A1|2012-06-07|Method and device for cancelling acoustic echo
    Kar et al.2014|EFficient suppression of Remnant Non-Linear Echo caused due to harmonic distortions
    JP2013005106A|2013-01-07|In-house sound amplification system, in-house sound amplification method, and program therefor
    同族专利:
    公开号 | 公开日
    CN103339671B|2016-05-04|
    EP2671223A1|2013-12-11|
    ES2558559T3|2016-02-05|
    BR112013016222A2|2018-05-15|
    EP2671223B1|2015-10-21|
    US20130287216A1|2013-10-31|
    CN103339671A|2013-10-02|
    US9420390B2|2016-08-16|
    ZA201304705B|2014-09-25|
    WO2012105880A1|2012-08-09|
    EP2671223A4|2014-07-23|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US4629829A|1984-12-14|1986-12-16|Motorola, Inc.|Full duplex speakerphone for radio and landline telephones|
    DE19831320A1|1998-07-13|2000-01-27|Ericsson Telefon Ab L M|Digital adaptive filter for communications system, e.g. hands free communications in vehicles, has power estimation unit recursively smoothing increasing and decreasing input power asymmetrically|
    WO2003010950A1|2001-07-20|2003-02-06|Koninklijke Philips Electronics N.V.|Echo canceller having nonlinear echo suppressor for harmonics calculations|
    JP3753996B2|2002-03-20|2006-03-08|日本電信電話株式会社|Echo suppression device, echo suppression method and program|
    US7894598B2|2004-12-14|2011-02-22|Nuance Communications, Inc.|System for limiting receive audio|
    US20060188089A1|2005-02-18|2006-08-24|Diethorn Eric J|Reduction in acoustic coupling in communication systems and appliances using multiple microphones|
    US7426270B2|2005-08-10|2008-09-16|Clarity Technologies, Inc.|Method and system for clear signal capture|
    CN101346896B|2005-10-26|2012-09-05|日本电气株式会社|Echo suppressing method and device|
    FR2908003B1|2006-10-26|2009-04-03|Parrot Sa|METHOD OF REDUCING RESIDUAL ACOUSTIC ECHO AFTER ECHO SUPPRESSION IN HANDS-FREE DEVICE|
    GB2448201A|2007-04-04|2008-10-08|Zarlink Semiconductor Inc|Cancelling non-linear echo during full duplex communication in a hands free communication system.|
    DE102008039330A1|2008-01-31|2009-08-13|Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.|Apparatus and method for calculating filter coefficients for echo cancellation|
    US8213598B2|2008-02-26|2012-07-03|Microsoft Corporation|Harmonic distortion residual echo suppression|
    US8538749B2|2008-07-18|2013-09-17|Qualcomm Incorporated|Systems, methods, apparatus, and computer program products for enhanced intelligibility|
    CN101778183B|2009-01-13|2013-06-05|华为终端有限公司|Method and device for suppressing residual echo|
    US8634569B2|2010-01-08|2014-01-21|Conexant Systems, Inc.|Systems and methods for echo cancellation and echo suppression|US9065895B2|2012-02-22|2015-06-23|Broadcom Corporation|Non-linear echo cancellation|
    GB2510331A|2012-12-21|2014-08-06|Microsoft Corp|Echo suppression in an audio signal|
    GB2509493A|2012-12-21|2014-07-09|Microsoft Corp|Suppressing Echo in a received audio signal by estimating the echo power in the received audio signal based on an FIR filter estimate|
    GB2512022A|2012-12-21|2014-09-24|Microsoft Corp|Echo suppression|
    US20150003606A1|2013-06-28|2015-01-01|Broadcom Corporation|Detecting and quantifying non-linear characteristics of audio signals|
    GB201321052D0|2013-11-29|2014-01-15|Microsoft Corp|Detecting nonlinear amplitude processing|
    EP3080975B1|2013-12-12|2017-07-12|Koninklijke Philips N.V.|Echo cancellation|
    US20170024495A1|2015-07-21|2017-01-26|Positive Grid LLC|Method of modeling characteristics of a musical instrument|
    JP6537997B2|2016-04-12|2019-07-03|日本電信電話株式会社|Echo suppressor, method thereof, program, and recording medium|
    US10129408B1|2016-10-11|2018-11-13|Google Llc|Residual echo detection based on non-causality|
    US9928847B1|2017-08-04|2018-03-27|Revolabs, Inc.|System and method for acoustic echo cancellation|
    CN107958671B|2017-11-22|2021-04-13|广州市六体智能科技有限公司|Echo cancellation device of law enforcement recorder|
    CN107705799B|2017-11-24|2020-12-11|苏州狗尾草智能科技有限公司|Chat robot echo cancellation method and device|
    DE102018129525A1|2018-11-23|2020-05-28|Harman Becker Automotive Systems Gmbh|NONLINEAR ACOUSTIC ECHOCOMPENSATION|
    US10964305B2|2019-05-20|2021-03-30|Bose Corporation|Mitigating impact of double talk for residual echo suppressors|
    WO2021194859A1|2020-03-23|2021-09-30|Dolby Laboratories Licensing Corporation|Echo residual suppression|
    法律状态:
    2019-01-08| B06F| Objections, documents and/or translations needed after an examination request according art. 34 industrial property law|
    2020-07-07| B15K| Others concerning applications: alteration of classification|Free format text: AS CLASSIFICACOES ANTERIORES ERAM: G10L 21/02 , H04B 3/20 , H04M 9/08 Ipc: G10L 21/02 (2013.01), G10L 21/0208 (2013.01), H04B |
    2020-07-07| B06U| Preliminary requirement: requests with searches performed by other patent offices: suspension of the patent application procedure|
    2020-10-27| B09A| Decision: intention to grant|
    2020-12-15| B16A| Patent or certificate of addition of invention granted|Free format text: PRAZO DE VALIDADE: 20 (VINTE) ANOS CONTADOS A PARTIR DE 03/02/2011, OBSERVADAS AS CONDICOES LEGAIS. |
    优先权:
    申请号 | 申请日 | 专利标题
    PCT/SE2011/050119|WO2012105880A1|2011-02-03|2011-02-03|Estimation and suppression of harmonic loudspeaker nonlinearities|
    [返回顶部]