Method of estimating a feedback path of a hearing aid and a hearing aid

Abstract

A method of estimating a feedback path of a hearing aid (200). The invention also relates to a hearing aid (200) adapted to carry out said method.

Claims

1. A method of estimating a feedback path of a hearing aid comprising the steps of: storing, in a memory of the hearing aid, at least one of a measure of the energy of a feedback test signal and an autocorrelation matrix based on a feedback test signal or a characteristic of a feedback suppression filter; performing an in-situ feedback test by providing the feedback test signal, represented by an output signal vector x(n), using an output transducer of the hearing aid and measuring the resulting input signal using an input transducer of the hearing aid and hereby providing an input signal vector y(n) representing the measured input signal samples; using an analytical expression to determine a feedback suppression filter vector ĥ based on the output signal vector x(n), the corresponding samples of the input signal vector y(n) and at least one of the measure of the energy of the feedback test signal and the autocorrelation matrix based on the feedback test signal or the characteristic of the feedback suppression filter, wherein the feedback suppression filter vector ĥ comprises the filter coefficients of the feedback suppression filter; operating the hearing aid with a feedback suppressing system comprising the feedback suppression filter that is at least initially set with the determined filter coefficients.

2. The method according to claim 1, wherein the feedback test signal is a white noise signal.

3. The method according to claim 1, wherein the output signal vector represents a Maximum Length Sequence noise signal.

4. The method according to claim 1, wherein the feedback test signal does not comprise parts consisting only of a pure tone.

5. The method according to claim 1, wherein the analytical expression used to determine the feedback suppression filter vector is derived by using a Least Mean Square approach.

6. The method according to claim 2, wherein the analytical expression used to determine the feedback suppression filter vector ĥ is given as:
ĥ=(P).sup.−1Xy.sup.T wherein y.sup.T is the transposed input signal vector, X is an output signal matrix formed by at least one output signal vector and P is the measure of the feedback test signal energy.

7. The method according to claim 1, wherein the analytical expression used to determine the feedback suppression filter vector ĥ is given as:
ĥ=(W.sup.TXX.sup.TW).sup.−1(W.sup.TX)y.sup.T wherein y.sup.T is the transposed input signal vector, wherein X is an output signal matrix formed by at least one output signal vector, wherein W is a warped filter matrix and wherein the feedback suppression filter is a warped filter.

8. The method according to claim 2, wherein the analytical expression used to determine the feedback suppression filter vector ĥ is given as:
ĥ=(P).sup.−1(W.sup.TW).sup.−1(W.sup.TX)y.sup.T wherein the feedback suppression filter is a warped filter, wherein P is the measure of the feedback test signal energy, wherein y.sup.T is the transposed input signal vector, wherein X is an output signal matrix formed by at least one output signal vector, wherein W is a warped filter matrix representing characteristics of a delay line of the warped feedback suppression filter and wherein (W.sup.TW).sup.−1 is the inverse of an autocorrelation matrix of the warped filter matrix.

9. The method according to claim 8, wherein the inverse autocorrelation matrix of the warped filter matrix (W.sup.TW).sup.−1 is expressed in the form of a KMS matrix and is stored in the memory of the hearing aid.

10. A hearing aid comprising an input transducer, a signal processor, an output transducer, a feedback suppression filter inserted in a feedback path, and a non-volatile memory, wherein the non-volatile memory comprises at least one of a measure of the energy of a feedback test signal and an autocorrelation matrix based on a feedback test signal or a characteristic of a feedback suppression filter, and wherein the signal processor is configured to: perform an in-situ feedback test by providing the feedback test signal, represented by an output signal vector x(n), using an output transducer of the hearing aid and measuring the resulting input signal using an input transducer of the hearing aid and hereby providing an input signal vector y(n) representing the measured input signal samples; use an analytical expression to determine a feedback suppression filter vector ĥ based on the output signal vector x(n), the corresponding samples of the input signal vector y(n) and at least one of the measure of the energy of the feedback test signal and the autocorrelation matrix based on the feedback test signal or the characteristic of the feedback suppression filter, wherein the feedback suppression filter vector ĥ comprises the filter coefficients of the feedback suppression filter; and operate the hearing aid with a feedback suppressing system comprising the feedback suppression filter that is at least initially set with the determined filter coefficients.

11. The hearing aid according to claim 10, wherein the feedback test signal is a white noise signal.

12. The hearing aid according to claim 11 wherein the analytical expression used to determine the feedback suppression filter vector ĥ is given as:
ĥ=(P).sup.−1Xy.sup.T wherein y.sup.T is the transposed input signal vector, X is an output signal matrix formed by at least one output signal vector and P is the measure of the feedback test signal energy.

13. The hearing aid according to claim 11, wherein the analytical expression used to determine the feedback suppression filter vector ĥ is given as:
ĥ.sub.w=(P).sup.−1(W.sup.TW).sup.−1(W.sup.TX)y.sup.T wherein the feedback suppression filter is a warped filter, wherein P is the measure of the feedback test signal energy, wherein y.sup.T is the transposed input signal vector, wherein X is an output signal matrix formed by at least one output signal vector, wherein W is a warped filter matrix representing characteristics of a delay line of the warped feedback suppression filter and wherein (W.sup.TW).sup.−1 is the inverse of an autocorrelation matrix of the warped filter matrix.

14. The hearing aid according to claim 10, wherein the analytical expression used to determine the feedback suppression filter vector ĥ is given as:
ĥ=(W.sup.TXX.sup.TW).sup.−1(W.sup.TX)y.sup.T wherein y.sup.T is the transposed input signal vector, wherein X is an output signal matrix formed by at least one output signal vector, wherein W is a warped filter matrix and wherein the feedback suppression filter is a warped filter.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) By way of example, there is shown and described a preferred embodiment of this invention. As will be realized, the invention is capable of other embodiments, and its several details are capable of modification in various, obvious aspects all without departing from the invention. Accordingly, the drawings and descriptions will be regarded as illustrative in nature and not as restrictive. In the drawings:

(2) FIG. 1 illustrates highly schematically a hearing aid according to the prior art; and

(3) FIG. 2 illustrates highly schematically a hearing aid according to an embodiment of the invention.

DETAILED DESCRIPTION

(4) The present idea is based on an improved feedback test wherein the filter coefficients of the adaptive feedback suppression filter is determined based on a simple and very fast measurement. Thus the present idea distinguishes the prior art in that the filter coefficients are determined based on a calculation as opposed to prior art methods that rely on allowing an adaptive feedback suppression filter to adapt in response to a provided audio test signal until a predetermined convergence criteria is fulfilled and then using the filter coefficients that led to this convergence as the result of the feedback test.

(5) Reference is now given to FIG. 2, which illustrates highly schematically a hearing aid 200 according to an embodiment of the invention. The hearing aid 200 is similar to the hearing aid 100 illustrated in FIG. 1 and the components that basically are the same will not be described further and will maintain the numbering given in FIG. 1.

(6) In addition to the previously mentioned components the hearing aid 200 comprises a test signal generator 201, a memory 202, a feedback estimator 203 and a feedback suppression filter 204. The feedback suppression filter 204 distinguishes the corresponding component in FIG. 1 in that it is not an adaptive filter. However in variations the feedback suppression filter 204 may be adaptive and in that case the estimated feedback suppression filter coefficients are just used as a starting point for the adaptive filter.

(7) Consider now a feedback suppression filter vector h=[h(0), h(1), . . . h(K−1)].sup.T that represents filter coefficients of the feedback suppression filter 204, an output signal vector x.sub.n=[x(n), x(n−1), . . . x(n−K+1)].sup.T that represents at least a part of a feedback test signal (and in the following the terms feedback test signal and output signal vector may therefore be used interchangeably) and an input signal vector y=[y(0), y(1), . . . y(N−1)] comprising input signal samples measured by the input transducer 101 in response to the feedback test signal being provided by the output transducer 103.

(8) Assuming that the feedback suppression filter 204 is a linear filter, such as a FIR filter, then the desired filtering function may be expressed as:

(9) $y\left(n\right)=\underset{k=0}{\overset{K-1}{.Math.}}h\left(k\right)x\left(n-k\right)=h^T{x}_n;$

(10) and assuming that a multitude of corresponding feedback test signals and measured input signal samples are determined then the input signal vector y may be given as:
y=h.sup.TX;

(11) wherein X=[x.sub.0, x.sub.1, . . . x.sub.N-1] and wherein X in the following may be denoted the output signal matrix. It follows directly that the output signal matrix is formed by horizontal concatenation of N output signal vectors and according to the present embodiment each of the output signal vectors represent at least a part of the feedback test signal.

(12) Now, the above equations represent the ideal case where the optimum filter coefficient vector is known. However, in reality an estimate of this optimum filter coefficient vector need to be determined and this can be done by minimizing the squared error E between the estimated input signal samples ŷ(n), provided by the estimated filter coefficient vector ĥ, and the real input signal samples y(n):

(13) $E=\frac{1}{2}\underset{N=0}{\overset{N-1}{.Math.}}{\left(y\left(n\right)-\hat{y}\left(n\right)\right)}^2=\frac{1}{2}\underset{n=0}{\overset{N-1}{.Math.}}{\left(y\left(n\right)-{\hat{h}}^T{x}_n\right)}^2;$

(14) Wherefrom the estimated filter coefficient vector ĥ may be determined:

(15) $\frac{\partial E}{\partial \hat{h}}=\underset{n=0}{\overset{N-1}{.Math.}}\left(y\left(n\right)-{\hat{h}}^T{x}_n\right)x_n=0;.Math.\hat{h}={\left({\mathrm{XX}}^T\right)}^{-1}{\mathrm{Xy}}^T;$

(16) Wherein XX.sup.T is the autocorrelation matrix for the output signal vector x.sub.n and wherein Xy.sup.T is a crosscorrelation between the output and input signal vectors.

(17) The output signal vector x.sub.n and hereby also the output signal matrix X are selected and therefore known in advance, whereby the inverse autocorrelation matrix (XX.sup.T).sup.−1 may be calculated off-line and stored in the memory 202 of the hearing aid 200. Preferably the output signal vector x.sub.n is also stored in the memory of the hearing aid 200, whereby the feedback test signal need not be streamed from an external device and to the hearing aid because the hearing aid is capable of generating the desired feedback test signal internally based on the stored output signal vector x.sub.n. Thus, the hearing aid 200 is configured to, in response to a trigger event, activate the test signal generator 201 in order to provide the feedback test signal through the output transducer 103. However, in a variation the feedback test signal may be generated internally in the hearing 200 and in this case the hearing aid is adapted to calculate the inverse autocorrelation matrix (XX.sup.T).sup.−1 internally.

(18) The crosscorrelation between the output and input signal vectors may also be determined in a simple manner by the feedback path estimator 203 based on input signal samples y(n) measured in response to a provided feedback test signal.

(19) By having the inverse autocorrelation matrix (XX.sup.T).sup.−1 stored in the memory 202 the processing resources and time required to determine the feedback suppression filter coefficients may be reduced compared to previously known methods.

(20) The inventors have found that the feedback test may be carried out in less than 3 seconds generally and the duration may be as short as 1 second. in many cases the duration is approximately 1 second.

(21) It is specifically advantageous to apply the present invention, when the feedback suppression filter is a high order filter (i.e. has many filter coefficients), because the relative amount of additional time required to carry out the feedback test using an adaptive algorithm increases with the order of the filter.

(22) According to an especially advantageous embodiment the feedback test signal provided by the output signal vector is white noise such as Maximum Length Sequence (MLS) noise. By applying this type of feedback test signal the resulting autocorrelation matrix XX.sup.T becomes a scaled identity matrix and consequently the estimated filter coefficient vector ĥ may be determined as:
ĥ=(P).sup.−1Xy.sup.T;

(23) wherein P is a measure of the energy of the known white noise feedback test signal as represented by the output signal vectors. Thus according to this embodiment it is only required to store the measure of the energy of the feedback test signal instead of the whole autocorrelation matrix of the output signal vector.

(24) It has been found that the estimated filter coefficient vector ĥ may be determined with a sufficiently high precision based only on a white noise feedback test signal, so that single test tones can be used, which will improve perceived comfort during the feedback test for at least some users.

(25) Generally the linear feedback suppression filter 204 may be of any type, such as an IIR filter.

(26) According to an alternative embodiment the feedback suppression filter 204 is a warped FIR filter, i.e. a filter with a frequency dependent delay and thereby a non-uniform frequency resolution as opposed to the traditional FIR filter that provides a uniform frequency resolution. In this context it is advantageous to apply a warped filter because it allows a good match to the response of the human auditory system. According to a specific embodiment the non-uniform frequency resolution of the warped filter is designed to match the psychoacoustic Bark scale.

(27) A warped filter is characterized in that the transfer function D.sub.k(z) between each node of the delay line is frequency dependent (i.e. dispersive) as opposed to the unit delay provided between the nodes of the delay line of a traditional FIR filter. In the following the warped filter may also be denoted a warped delay line.

(28) Consider now a warped filter matrix W defined as:
W=[w.sub.0,w.sub.1, . . . w.sub.K-1]
wherein the vectors w.sub.k represent the impulse responses of the transfer functions characterizing the delay line of the warped filter. Thus the warped filter matrix is formed by horizontal concatenation of vectors representing impulse responses characterizing the warped filter delay line.

(29) Following the same procedure as outlined above for the FIR filter implementation we find that an estimate of the warped filter coefficient vector may be determined as:
=(W.sup.TXX.sup.TW).sup.−1(W.sup.TX)y.sup.T;

(30) wherein (W.sup.TX)y.sup.T represents a modified crosscorrelation matrix between the output and input signal vectors.

(31) In analogy with the FIR filter embodiment it may be selected to use white noise as feedback test signal whereby the estimate of the warped filter coefficient vector may be determined as:
=(P).sup.−1(W.sup.TW).sup.−1(W.sup.TX)y.sup.T;

(32) The warped filter matrix W is known in advance and it is therefore possible to calculate off-line the autocorrelation matrix of the warped filter matrix W.sup.T W or the inverse of the autocorrelation matrix of the warped filter matrix (W.sup.TW).sup.−1 and store the result in the memory 202 of the hearing aid 200. In an obvious variation the warped filter matrix W itself may also be stored in the memory 202 in order to facilitate the calculation of the modified crosscorrelation matrix.

(33) In another variation the inventors have realized that the autocorrelation matrix of the warped filter matrix can be expressed in the form of a Kac-Murdock-Szegö (KMS) matrix which is particularly simple to invert, whereby the inverse of the autocorrelation matrix of the warped filter matrix can be calculated off-line and stored in the memory 202 of the hearing aid 200 as a relatively simple expression.

(34) It should be appreciated that the disclosed embodiments of the invention are characterized in that an autocorrelation matrix or a measure derived from the autocorrelation matrix are stored in a memory of a hearing aid whereby the filter coefficients for a feedback suppression filter may be determined independently by the hearing aid as part of a feedback test of short duration.

(35) In the present context, an autocorrelation matrix is construed to cover matrices that primarily consists of elements of the discrete autocorrelation function.

(36) In further variations the methods and selected parts of the hearing aid according to the disclosed embodiments may also be implemented in systems and devices that are not hearing aid systems (i.e. they do not comprise means for compensating a hearing loss), but nevertheless comprise both acoustical-electrical input transducers and electro-acoustical output transducers. Such systems and devices are at present often referred to as hearables. However, a headset is another example of such a system.

(37) In still other variations the invention is embodied as a non-transitory computer readable medium carrying instructions which, when executed by a computer, cause the methods of the disclosed embodiments to be performed.

(38) Other modifications and variations of the structures and procedures will be evident to those skilled in the art.