EP0656737B1 - Hearing aid with cancellation of acoustic feedback - Google Patents

Abstract

The acoustomechanical disturbing feedback in a hearing aid having an amplifying filter section is compensated by means of a compensator (15). In this case, the signals are joined in the frequency range at the compensator (15) and preferably also in the amplifying filter section (5). For this purpose, time range/frequency range transformation units (20, 28), or corresponding back-transformation units (26, 24) are connected downstream of a subtraction unit (13) to which the compensator signal is coupled. <IMAGE>

Description

The present invention relates to a hearing aid according to the Preamble of claim 1.

The problems that arise particularly due to the acoustic Feedback between the electrical-acoustic transducer and the acoustic-electrical converter of such hearing aids are known and are known, for example, in EP-A-0 415 677 discussed in detail.

An attempt was made to address these problems in principle, as in FIG. 1 shown to solve.

1 shows an acoustic-electrical (ak / el) converter 1 with downstream analog / digital (A / D) converter 3, one digital gain filter section 5, which on the output side to a digital / analog (D / A) converter 7, the latter to the electrical-acoustic (el / ak) converter 9 acts.

With block 11 is the acoustic-mechanical interference feedback with the generally time-variant transmission behavior h shown. The feedback signal y (t) is the Useful signal v (t) superimposed and the input of the ak / el converter 1 supplied, the output side, at the times nT, for the digital processing required time-discrete samples d (nT) returns.

To suppress the interference feedback signal y (t) was for example in D.K. Bustamante et al., "Measurement and adaptive suppression of acoustic feedback in hearing aids ", Proc. 1989 IEEE ICASSP, 3: 2017-2020, 1989, proposed a differential unit 13, via a compensator 15, which the output signal of the gain filter section 5 by filtering with an m-stage FIR (finite impulse response) filter estimate y and (nT). Doing so with Using the well-known LMS (least mean square) algorithm Filter coefficients changed iteratively until the output side Difference signal e (nT) no longer with the estimate y and (nT) correlated. The signal e (nT) required for the adaptation is fed to the compensator 15 via the adaptation input A.

Assuming that the useful signal v (t) is uncorrelated or v (nT) and amplified signal u (t) or u (nT), which by suitable choice of the time delay DT in the digital gain filter the route 5 can be reached, it will possible to compare the gain of the gain filter 5 Hearing aids without compensator 15 to 6 to 10dB too increase.

The disadvantage of this procedure is that, if one is adopted Compensator 15 filter length of m steps, 2 m multiplications 3 are required per sample value of the A / D converter, which leads to an extraordinarily complex system. This especially with regard to the required miniaturization Hearing aids.

Furthermore, it is necessary that the stride length µ of the LMS algorithm for the preservation of the speech signal transmission is chosen as small as possible, with which the adaptation of the compensator filter 15 to the interference feedback path 11 accordingly slowly becomes what the possible increase in gain on route 5, limited for reasons of stability.

In further development of the procedure shown in Fig. 1 an attempt was then made to give the system a stationary measurement signal , such as from "Feedback Cancellation in Hearing Aids: Results from a Computer Simulation ", J.M. Kates, IEEE Trans. On Signal Processing, Vol. 39, No. 3, March 1991, or EP-A-0 415 677. It was considered stationary Measurement signal a noise signal fed to the system.

The disadvantage of this procedure is the additional generator for the measurement signal and its necessary amplitude control to ensure a sufficient signal to noise ratio.

With a compensator filter 32nd order was through this Procedure to increase the gain on the amplifier filter section around 17dB possible.

Because of the technology mentioned above with measurement signal coupling The resulting disadvantages finally became a course of action 2 proposed, according to "Integrated Frequency Domain Digital Hearing Aid With the Lapped Transform ", S.M. Kuo and S. Voepel, Electronics Letters, Vol. 28, No. 23, November 1992.

Accordingly, signal processing was performed on both the gain filter section as well as on the compensator in the frequency range the output signal of the A / D converter 3 by means of an overlapping orthogonal transformation (LOT) was transformed into the frequency domain at the unit 17. A corresponding inverse transformation (ILOT) on the unit 19 then supplies the input of the el / ak converter 7 again the required Signal u (nT).

Because with a suitable time domain / frequency domain transformation, in particular with the discrete Fourier transform (DFT) and the discrete Hartley transform (DHT), the folding at the compensator and gain filters 15 _f and 5 _f during the transition into the frequency domain into one Multiplication passes over, this procedure basically results in a reduction in the computing or hardware effort. In order to obtain a finite transformation length that can be realized, a subdivision of the discrete signal d (nT) on the input side of the transformation unit 17 into blocks of a given length is necessary. Unfortunately, the errors associated with this, compared with conventional folding, cannot be eliminated in the arrangement according to FIG. 2 even with an overlapping block division. They lead to a time-variant system, even if, together with the interference feedback h, the compensation filter 15 _{f is} time-invariant or frozen.

Therefore, a compromise had to be made by choosing long block lengths of, for example, 512 samples, which in turn leads to inefficient compensation via the compensating filter 15 _f . Accordingly, the achievable gain increase on amplifier filter section 5 _{f was} limited to less than 10 dB.

It is an object of the present invention to provide a hearing aid to create the type mentioned, in which, with preservation the advantages of signal processing in the frequency domain, time invariance of the system, with time invariant feedback, is guaranteed, in which the computing or hardware effort is minimized to such a degree that the Signal processing easily among those in hearing aids extremely limited space is feasible.

Based on the latter hearing aid device, this becomes achieved that it according to the characterizing part of claim 1 is trained.

The fact that the time domain / frequency domain transformation is not carried out in front of the differential unit 13 _f , as shown in FIG. 2, but rather that the difference is formed in the time domain, can surprisingly be obtained the required time invariance of the system. In particular, when suitably overlapping block division is selected, it is possible to implement the time domain / frequency domain transformations that are still used with significantly smaller block lengths, which in turn increases the compensation efficiency and therefore enables the gain on the gain filter section 5 _f according to FIG. 2 to be increased drastically.

The invention with its specified in the further claims preferred design variants is then for the time being step by step using figures, for example explained and finally using an implementation example presents.

Fig. 1

anhand eines Funktionsblockdiagrammes, vereinfacht, ein bekanntes Hörhilfegerät, bei welchem die Signalverarbeitung zeitdiskret erfolgt;

Fig. 2

in Darstellung analog zu Fig. 1, ein weiteres bekanntes Hörhilfegerät, bei welchem die Signalverarbeitung an Rückkopplungskompensator und Verstärkungsfilterstrecke gemäss Fig. 1 im Frequenzbereich durchgeführt wird;

Fig. 3

in Darstellung analog derjenigen der Fig. 1 und 2, eine erste Ausführungsvariante eines erfindungsgemässen Hörhilfegerätes;

Fig. 4

eine weitere bevorzugte Ausführungsvariante des Hörhilfegerätes nach Fig. 3, dargestellt analog zu den Fig. 1 bis 3;

Fig. 5

ausgehend von dem in Fig. 4 dargestellten Hörhilfegerät, eine weitere bevorzugte Ausführungsvariante des erfindungsgemässen Gerätes in Darstellung analog derjenigen der Fig. 1 bis 4;

Fig. 6

anhand eines vereinfachten Signalfluss-Funktionsblockdiagrammes eine bevorzugte Realisationsform der dem Adaptionseingang und der Verstärkungsfilterstrecke vorgelagerten Transformationseinheit gemäss Fig. 5;

Fig. 7

anhand eines vereinfachten Signalfluss-Funktionsblockdiagrammes eine bevorzugte Ausführungsvariante der Verstärkungsfilterstrecke am erfindungsgemässen Gerät gemäss Fig. 5;

Fig. 8

anhand eines vereinfachten Signalfluss-Funktionsblockdiagrammes eine bevorzugte Realisation des Kompensatorfilters am erfindungsgemässen Gerät gemäss Fig. 5;

Fig. 9

anhand eines vereinfachten Signalfluss-Funktionsblockdiagrammes die Bildung des Schrittgrössensignals in Funktion der erfassten Signalleistung, welches Schrittgrössensignal, wie in Fig. 9 bevorzugterweise gebildet, bei der Realisation des Kompensatorfilters nach Fig. 8 eingesetzt ist;

Fig. 10

eine bei der Realisation des Kompensatorfilters gemäss Fig. 8 bevorzugterweise eingesetzte Einheit in vereinfachter Signalfluss-Funktionsblockdarstellung;

Fig. 11

ausgehend von der Darstellung gemäss Fig. 4 eines erfindungsgemässen Hörhilfegerätes, eine heute besonders bevorzugte Ausführungsvariante anhand des bereits vorgestellten Funktionsblockdiagrammes;

Fig. 12

ausgehend von der besonders bevorzugten Ausführungsvariante gemäss Fig. 11, einen Teil einer Weiterentwicklung mit Modellierung des elektrisch-akustischen Wandlers im Zeit- und/oder Frequenzbereich;

Fig. 13

ein Funktionsblock/Signalflussdiagramm eines elektrischen, im Zeitbereich arbeitenden Lautsprechermodells, wie es vorzugsweise zur Berücksichtigung des Lautsprecher-Uebertragungsverhaltens am erfindungsgemässen Hörhilfegerät gemäss den Fig. 3, 11 oder 12 eingesetzt wird;

Fig. 14

ausgehend von der Darstellung nach Fig. 12, eine Weiterentwicklung des erfindungsgemässen Gerätes, bei welcher die Modellierung und/oder die Amplitudenlimitierung und/oder die Verstärkung in Funktion des IST-Zustandes einer Batterie geführt wird;

Fig. 15

ausgehend von einem Gerät nach Fig. 11, eine weitere Verbesserung durch gegebenenfalls selektiv gesteuerte Rauschsignalaufschaltung im Frequenz- oder Zeitbereich;

Fig. 16

eine bevorzugte Realisation der Rauschaufschaltung gemäss Fig. 15 im Zeitbereich;

Fig. 17

eine bevorzugte Realisationsform der Rauschaufschaltung nach Fig. 15 im Frequenzbereich.

Show:

Fig. 1

based on a functional block diagram, simplified, a known hearing aid in which the signal processing is time-discrete;

Fig. 2

in representation analogous to FIG. 1, another known hearing aid device in which the signal processing on the feedback compensator and amplification filter section according to FIG. 1 is carried out in the frequency range;

Fig. 3

in representation analogous to that of Figures 1 and 2, a first embodiment of a hearing aid according to the invention;

Fig. 4

a further preferred embodiment of the hearing aid device according to FIG. 3, shown analogously to FIGS. 1 to 3;

Fig. 5

starting from the hearing aid device shown in FIG. 4, a further preferred embodiment variant of the device according to the invention in a representation analogous to that of FIGS. 1 to 4;

Fig. 6

on the basis of a simplified signal flow function block diagram, a preferred form of realization of the transformation unit upstream of the adaptation input and the amplification filter section according to FIG. 5;

Fig. 7

based on a simplified signal flow function block diagram, a preferred embodiment variant of the amplification filter section on the device according to the invention according to FIG. 5;

Fig. 8

based on a simplified signal flow function block diagram, a preferred implementation of the compensator filter on the device according to the invention according to FIG. 5;

Fig. 9

on the basis of a simplified signal flow function block diagram, the formation of the step size signal as a function of the detected signal power, which step size signal, as preferably formed in FIG. 9, is used in the implementation of the compensator filter according to FIG. 8;

Fig. 10

a unit preferably used in the implementation of the compensator filter according to FIG. 8 in a simplified signal flow function block representation;

Fig. 11

starting from the representation according to FIG. 4 of a hearing aid according to the invention, a particularly preferred embodiment variant based on the functional block diagram already presented;

Fig. 12

starting from the particularly preferred embodiment variant according to FIG. 11, part of a further development with modeling of the electrical-acoustic transducer in the time and / or frequency range;

Fig. 13

3 shows a functional block / signal flow diagram of an electrical loudspeaker model operating in the time domain, as is preferably used to take into account the loudspeaker transmission behavior on the hearing aid according to the invention according to FIGS. 3, 11 or 12;

Fig. 14

starting from the illustration according to FIG. 12, a further development of the device according to the invention, in which the modeling and / or the amplitude limitation and / or the amplification is carried out as a function of the current state of a battery;

Fig. 15

starting from a device according to FIG. 11, a further improvement by optionally selectively controlled noise signal application in the frequency or time domain;

Fig. 16

a preferred implementation of the noise lock according to FIG. 15 in the time domain;

Fig. 17

a preferred implementation of the noise lock according to FIG. 15 in the frequency domain.

3 is based on a signal flow / functional block diagram a basic principle of the present invention or that of the invention Hearing aid shown. It is in it the reference numerals already used with reference to FIGS. 1 and 2 for the function blocks and signals already described there used.

3 and 4, the discrete-time difference signal r (nT) from the A / D-converted output signal d (t) of the ak / el converter 1 and the output signal of the compensator filter 15 _f educated. Only the difference signal r (nT) on the output side of the difference formation unit 13 is subjected to an overlapping orthogonal transformation LOT.

3, the difference signal r (nT) is converted at a LOT transformation unit 20 into the adaptation control signal E [k], which is fed to the adaptation input A _{f of} the compensator filter 15 _f . Because the time domain / frequency domain transformation takes place in the LOT transformation unit 20 in blocks of a predetermined number of samples from the difference signal r (nT), [k] denotes the number of the signal block appearing on the output side of the transformation unit 20.

3, the difference signal r (nT) is supplied to the amplification filter section 5 in the time domain and fed to the el / ak converter 9 via the D / A converter 7. On the input side, the D / A converter 7 is acted upon by the time-discrete output signal u (nT) of the amplification filter section 5. This output signal u (nT) is fed to a further orthogonal transformation unit 22, where it is converted from the time domain to the frequency domain. The output signal of the transformation unit 22 is fed as an input signal to the input E _{f of} the compensator filter 15 _f . The output signal Y and [k + 1] of said filter 15 _f is transformed back into the time domain at a reverse transformation unit ILOT 24 and its output signal y and (nT) is fed to the difference forming unit 13 as a discrete-time signal.

In addition to the embodiment variant in FIG. 3, according to FIG. 4 not only the signal processing is carried out on the compensation filter 15 _f in the frequency range but also on the amplification filter section 5 _f . For this purpose, the amplification filter section 5 _{f is} preceded by a transformation unit LOT 28 and the D / A converter 7 is a reverse transformation unit ILOT 26; the transformation unit 22 according to FIG. 3 is omitted.

Basically, therefore, and according to the wording of claim 1, as explained with reference to FIGS. 3 and 4, in difference to known procedures according to FIG. 2, the difference formation the difference formation unit 13 in the time domain, whereby the above-mentioned disadvantages regarding time variance of the 2 are resolved.

This results in the possibility of using the LOT transformation units 20, 22, 28 and, accordingly, on the ILOT reverse transformation units 24, 26 with much smaller ones To work block lengths than this in the procedure according to FIG. 2 is possible, for example according to a preferred embodiment the present invention with block lengths of blocks k of 128 samples.

3 shows, as mentioned, a first form of implementation which corresponds to the definition according to claim 2, namely in which a respective transformation unit LOT 20 or 22 is arranged upstream of the signal input E _f and the adaptation input A _{f of} the compensator filter 15 _f .

A preferred embodiment variant is that according to FIG. 4, which corresponds to the definition according to claim 3, according to which the adaptation input A _{f of} the compensating filter 15 _f and the input of the amplification filter section 5 _{f are preceded} by a LOT transformation unit 20 or 28 and the input of the D / A converter 7 a corresponding ILOT reverse transformation unit 26.

For block formation and processing in overlapping orthogonal Transformations are two simple techniques namely the "overlap-save" and "overlap-add" technology for Available. To this end, it can be fully based on the relevant Literature, such as "Signal Processing with Lapped Transforms ", Henrique S. Malvar, Artech House, Boston, 1992.

In a preferred embodiment of the present invention according to the wording of claim 5, as shown in FIG. 4, the gain filter 5 _{f is also} preceded by a LOT transformation unit 28, the input of the D / A converter 7 is an ILOT reverse transformation unit 26, and further the output of the compensator filter 15 _{f is followed by} an ILOT reverse transformation unit 24. These transformation or reverse transformation units 28, 24 and 26 operate in the mentioned preferred embodiment according to the "overlap-save" technique. In contrast, the LOT transformation unit 20 upstream of the adaptation input A _f , in particular according to FIG. 4, preferably works according to the "overlap-add" principle.

In particular, these preferred embodiment variants of the insert block processing techniques lead to another preferred form of realization of the hearing aid according to the invention, as shown in Fig. 5.

In contrast to FIG. 4, the time discrete differential signal r (nT) of a single LOT transform unit 30 is fed to here, from whose output signal both the adaption input A _f supplied adaptation signal E [k] as well as that of the enhancement filter path 5 _f supplied input signal R [k ] is formed.

As mentioned, the overlapping orthogonal transformations are based preferably at the DFT.

FIG. 6 shows a form of realization of the data transmission path between the time-discrete difference signal r (nT) on the output side of the difference forming unit 13 for the adaptation signal E [k] or the input signal R [k] to the amplification filter section 5 _f according to FIG. 5.

Accordingly, an overlapping orthogonal transformation unit 30a based on the DFT follows the output of the difference formation unit 13 with the time-discrete difference signal r (nT). As shown with the indexing OA, it works according to the "overlap-add" principle. To this end, the error block e [k] is formed by dividing r (nT) into sub-blocks of length N, each of which, in the variant preferred here with N = 64, by adding zeros to a total block length, here from 2N = 128 values , be extended, ie e [k] = (0 ... 0, r ((k + 1) NT), r ((k + 1) NT + T) ... r ((k + 2) NT-T)) T .

In the preferred variant according to FIG. 5, its DFT, namely E [k], is fed directly to the adaptation input A _{f of} the compensation filter 15 _f . The following blocks, that is to say the numbers k and k + 1, are made available via a delay unit 32 with corresponding intermediate storage. A partial overlay in the unit 34 then provides the block R [k] directly, but now the "overlap-save" type, which in the implementation previously described as the preferred variant, according to FIG. 5, is fed directly to the amplification filter section 5 _f . The overlay in unit 34 is given by the formula R j [k] = E j [k] + (-1) j E j [k-1], given, where j (from 0 to 2N-1) denotes the number of the block location.

This procedure significantly reduces the necessary hardware and computing power realized.

According to FIG. 7, within the gain filter section 5 _f acted upon by R [k], the actual gain filter 40 follows first, which is followed by a delay unit 42 with corresponding intermediate storage. Here, the parameter d denotes the total delay of the system (from the output of the A / D converter 3 to the input of the D / A converter 7), normalized with the overlap parameter of the partial block length N. Due to the block processing, there is a minimum delay time of N samples , corresponding to a minimum d-value of 1. In the variant preferred here with a partial block length of N = 64 and a total block length of 2N = 128, using a single partial compensator (as will be explained in more detail below with reference to FIG. 8), d set to the value 2.

The block signal U [k + 1] available on the output side is now supplied on the one hand to the input E _{f of} the compensator 15 _f and on the other hand is subjected to an inverse DFT of the "overlap-save" type in the ILOT unit 26. Since the corresponding time signal u (nT) is delayed by a partial block length N, the numbering of U [k + 1] with the block number k + 1 is justified in retrospect.

FIG. 8 shows a preferred expansion variant of the compensator filter 15 _f on the hearing aid according to the invention according to FIG. 5. The block signals U [k + 1] to are thereby buffered with delay units of the type, as shown at 56 U [k + 1-L] provided and, based on this, with the aid of partial compensators, the first of which is referred to in FIG. 8 as unit 50, which generates the partial estimates Y and ₁ [k + 1] to Y and _L [k + 1], which in turn are provided in unit 52 for Overall estimate Y and [k + 1] can be added. As can be seen in FIG. 5, the ILOT unit 24, in the preferred variant via an inverse DFT of the "overlap-save" type, then transforms back into the time domain.

With reference to the first partial compensator, the partial estimate Y and ₁ [k + 1] arises at the output of the multiplication unit 64, on which the block signals U [k + 1] and the block weight H and ₁ [k + 1] act at the input. The multiplication is for each block position according to the formula Y i, j [k + 1] = U j [k + 2-i] H i, j [k + 1] executed, where j denotes the block position from 0 to 2N-1 and i the partial compensator number from 1 to L.

zur Aktualisierung von H₁[k+1] verwendet. Hierbei bezeichnet j wieder die Blockstelle und i die Teilkompensatornummer. Der Index (*) steht für konjugiert komplex.The block weight H _i [k + 1] represents the current estimate in the frequency range for the i-th subrange of length N of the discrete-time impulse response h of the acoustic-mechanical interference feedback 11. The estimate H _i [k + 1] is preceded by the formation of Y and _{i, j} [k + 1] updated using the old estimate H _i [k]. The block signal acts for this purpose, again with reference to the partial compensator 1 U [k + 1-1] and the step size µ [k + 1-1] to the multiplication unit 54, which on the output side is led to the multiplication unit 58 together with the block signal E [k]. The output of unit 58 is then in summation unit 60 according to the formula

used to update H ₁ [k + 1]. Here again j denotes the block location and i the partial compensator number. The index (*) stands for conjugate complex.

Working with partial expansion joints has the advantage of minimal delay D = N can be set by selecting the partial block length N independently of the actual impulse response length of the interference feedback 11. A "trade-off" between delay D and the partial block length N which determines the efficiency of the machining is thus possible. Furthermore, individual sub-areas of the impulse response h can be influenced in a targeted manner by corresponding block weights in the frequency range, for example in accordance with the acoustic near and far ranges.

Basically, any known method for guiding the Step size µ [k] can be used.

FIG. 9 shows a preferred variant today for generating the normalized step size μ [k] according to FIG. 8, which is also used to stop the adaptation process. For this purpose, starting from the block signal U [k], according to FIG. 8, this block signal is used before the supply to the multiplication unit 54 to calculate the current block signal µ [k] by feeding the block signal U [k] to a power acquisition unit 70 is, which in turn on two interpolation filters 72, respectively. 74 acts. On the output side, these interpolation filters control the scaling unit 78, which ultimately supplies the scaling variable S [k] required for the normalization of the reference step size μ ₀ at the input of the multiplication unit 80.

und sind mit γ und c parametrisiert. Der Index j bezeichnet, wie hier üblich, die Blockstelle. In der bevorzugten Realisierung wurde γ = 0.8 und c = 1 für das Filter 72 und γ = 0.995 und c = 0.2 für das Filter 74 gewählt.The interpolation filters work according to the formula

and are parameterized with γ and c. As is customary here, the index j denotes the block location. In the preferred implementation, γ = 0.8 and c = 1 were chosen for the filter 72 and γ = 0.995 and c = 0.2 for the filter 74.

If γ = 1 is selected for the interpolator 74, this interpolator is omitted, and there remains a block signal P _U ^min which is constant over time and which may be sufficient for various applications and which further reduces the hardware and computing outlay.

gebildet, wobei die j wie üblich die Blockstelle bezeichnen.The scaling variable S [k] is now used on the one hand via the output of the filter 72, in FIG. 9 as a block signal P _U [k], to normalize the reference step size μ ₀ , but on the other hand also via the output of the filter 74 in FIG 9 referred to as block signal P _U ^min [k], for freezing the adaptation process of individual frequency components when the power is insufficient. The scaling variable S [k] is according to the formula

formed, the j denoting the block location as usual.

FIG. 10 shows a further preferred variant which, with the use of partial compensators according to FIG. 8, significantly improves the speech quality, with otherwise the same parameters. For this purpose, the estimate H and _i [k + 1] of the partial compensator i, previously the multiplication with U [k + 2-i] in unit 64 of FIG. 8, via a projection unit 62. For this purpose, for example, the block weight H and _i [k + 1] is subjected to an inverse DFT (unit 82), then cleaned by zeroing the block locations with index N to 2N-1 (unit 84) and finally transformed back into the frequency range (unit 86) .

As is known, the electrical-acoustic converter 9 is not linear in the sense that it no longer converts the input signal into the output signal linearly from certain input signal amplitudes. In addition to the acoustic distortions caused by this, it should be taken into account that the signal path via compensation filter 15 _f should be modeled as exactly as possible over the signal path via function blocks 7, 9, 11, 1 and 3 and, according to the previous explanations, the nonlinearities mentioned on converter 9 should not can reproduce. In addition, the maximum output level should also be adjustable in the hearing aid according to the individual needs of the user. The problem arises that the converter 9 is driven into its non-linear range, of course only if the individually set maximum output level can drive the converter in the mentioned range at all.

For this reason, it is further proposed, as dashed lines in FIG. 3 shown, in this embodiment of the invention Hearing aid the gain filter 5 one in Time range working, preferably adjustable limiter unit 90 downstream, which is the output signal of the gain filter 5 limited in amplitude so that the Converter 9 is never driven into its non-linear range and which also allows the maximum output sound level at Converter 9, according to individual needs, in particular also set deeper, like this with the double arrows is indicated.

In the embodiment variant according to FIG. 4, this is achieved in that the amplification filter 5 _f operating in the frequency range is followed by a unit 90 _f , which limits the frequency components of the signal spectrum, taking into account their mutual phase position, in the frequency range in such a way that the output of the conversion unit 26 and the Digital / analog converter 7 produces a time-variable signal u (t) which never drives the converter 9 into the non-linear transmission range and which also allows the maximum individual modulation to be set.

The same procedure is also implemented with the unit 90 _f in the embodiment variant according to FIG. 5.

FIG. 11 shows a further embodiment variant of the hearing aid according to the invention, which largely corresponds to that shown in FIG. 4, with the difference that the reverse transformation unit 26 according to FIG. 4, now 26a, is provided directly on the output side of the amplification filter unit 5 _f and on the input side of the Compensation filter 15 _f a LOT transformation unit 22a of the type already discussed is arranged. Although at first glance such an embodiment hardly seems to bring any advantages compared to that according to FIG. 4, it nevertheless opens up the possibility explained below.

As can be seen from FIG. 4, which, as explained above, as well as FIG. 5, represents a preferred embodiment variant of the inventive hearing aid, the provision of a limiter unit is only possible in the frequency range because such a unit is also in the signal path with the compensation filter 15 _f must be effective.

As shown in FIG. 11 at 90, this enables here functional block structure shown the provision of a Time domain working limiter unit 90, which is essential is easier to implement than one operating in the frequency domain.

This also allows easy expansion with units for Compensation for non-linear effects as described below.

In order to ensure a sufficiently accurate identification of the converter 9 by the compensator filter 15 _f , its modulation is initially limited in order to prevent it from being operated in the non-linear range. This naturally results in a corresponding reduction in the maximum possible signal amplification of the hearing aid according to the invention from transducer 1 to transducer 9.

FIG. 12 shows a preferred embodiment variant of the signal processing on the device according to FIG. 11 upstream of the compensation filter 15 _f or downstream of the amplification filter 5 _f . According to FIG. 12, the non-linearity of the electrical-acoustic transducer 9 is basically simulated, ie modeled, in the signal path with the compensation filter 15 _f . This is implemented by a modeling unit 92, upstream of the transformation unit 22a according to FIG. 11 and thus operating in the time domain, and / or by a modeling unit 92 _f , downstream of the transformation unit 22a and therefore operating in the frequency domain.

This procedure ensures that, depending on the quality of the modeling unit 92, the limits of unit 90 are set higher and thus the output signal by up to 6dB compared to the Design variant in Fig. 11, can be increased. Possibly can also stop the limiter function of the unit 90 will.

The modeling unit 92 can, for example, as in R. Isermann, "Identification of dynamic systems", Springer-Verlag, 2: 238, 1988, proposed as a simplified Wiener model will be realized.

The transformation into the time domain between the gain filter 5 _f and the compensator filter 15 _{f also} allows the addition of a nonlinear correction filter in the signal path with the gain filter 5 _f in the same manner described above. As can be seen from FIG. 12, this is implemented by a modeling unit 94, connected downstream of the transformation unit 26a and thus operating in the time domain, and / or by a modeling unit 94 _f , connected upstream of the transformation unit 26a and therefore operating in the frequency domain.

Of course, it is possible in the preferred implementation variants 11 and 12, the LOT units 20 and 28 by a single LOT transformation unit 30 replace as shown in Figs. 5 and 6.

13 shows the implementation of a method according to the invention Speaker model shown in the time domain. In the inventive Hearing aid is used, according to Fig. 3 and 11 at the location of block 90 and according to FIG. 12 instead of blocks 92 or 90 and 94.

It comprises a pre-filter 100 with the transfer function F ₁ (ω), essentially with a low-pass characteristic. The cutoff frequency ω ₁ in the Bode diagram of the filter characteristic, which is shown qualitatively in block 100, is approximately 0.8 kHz, the gain | F ₁ | at the corner frequency ω ₁ is approx. 0dB. Likewise, the asymptote slope S _{1 is} approximately 0dB / DK.

The identification variables, namely corner frequency ω ₁ and the asymptotic slopes S ₁ and S ₂ , as well as the amplification, for example at the corner frequency ω ₁ , are identified by identifying the loudspeaker or transducer 9 to be modeled.

A linear amplifier unit 102 is provided downstream of the prefilter 100, and the gain factor K is set to this. A non-linear amplification unit 104 is provided downstream of the linear amplification unit 102. The characteristic of the non-linear gain function Y = Q (x) results in: y = x + ax 2nd + bx 3rd + cx 4th + dx 5 .

For small input signals, the gain is nonlinear Gain unit 104 one, with which the gain characteristic has slope one around the origin. For higher ones Input signal deviation x shows the nonlinear gain characteristic, as known from the loudspeaker or converter 9, saturation behavior on.

The coefficients a, b, c, d and the gain K are again based on the actual speaker to be modeled or converter 9 identified.

Downstream of the non-linear amplification unit 104, a linear amplification unit 106 is again provided, by means of which the amplification K of the linear amplification element 102 is compensated for - K ^-1 -. A filter unit 108 is provided downstream of it, essentially with a high-pass characteristic, which, as can be seen, essentially compensates for the frequency response of the pre-filter 100.

This is the loudspeaker modeling unit as in Fig. 13 is shown, essentially from a linear Amplifier part 102, 106, 100 and 108 and a non-linear Reinforcement unit 104.

Signs of saturation or limitation can besides the Two causes already mentioned, namely deliberate limitation the maximum output signal level of the converter 9 according to individual needs or modulation of the converter 9 in its transducer-specific, non-linear saturation range, based on another cause, namely waste the battery voltage, which the device according to the invention feeds. The aging of the battery that powers the device causes a decrease in signal amplification in particular at the D / A converter 7 and a reduction in the headline limit, i.e. the maximum analog modulation range decreases with decreasing Battery voltage lower.

In addition, the battery's output impedance usually appears in series with the impedance of the electrical-acoustic converter 9. This changes the towards the end of the battery life Battery output impedance and thus the latter, the substitute image downstream of the D / A converter 7, m.W., change it the one to be modeled, as explained, on the output side of the converter 7 appearing non-linearities.

3, 4 or 5, the limiter unit 90 in the time domain or 90 _f in the frequency domain by means of the instantaneous battery voltage and / or the instantaneous one To control battery impedance with regard to its limiting effect.

Starting from the variant according to Figs. 11 and 12, this procedure is shown schematically in FIG. 14. At the output of the battery unit 120, which, as indicated schematically with "block powering", feeds the various active components in the blocks of the hearing aid according to the invention, the current battery voltage U _B and / or the current impedance is measured on a measuring unit 122 Z. _B measured, resulting in corresponding measurement signals e (U _B ) or e ( Z. _B ). These measurement signals control the limiter unit 90, analogously in the frequency range the limiter unit 90 _f according to FIGS. 4, 5, 11 or 12, 14 and / or the model units 92, 92 _f or 94, 94 _f from FIGS. 12, 13, 14. Of course, the measurement signals e are preferably used after digitization, for which purpose the measurement unit 122 is provided on the output side with an A / D converter (not shown).

As a result, the limiter limits and / or the Model parameters through the current battery output voltage or track their instantaneous impedance in a controlling sense.

The model parameters on the model units 92 and 92 _f , 94 and 94 _f are modified in the function of the above-mentioned measured variables on the battery 120 or by means of values stored in tables that can be called up and activated by the current measured variables.

As further shown in FIG. 14, as a function of the measurement signals e mentioned, a gain loss on the D / A converter 7 is compensated for due to a decrease in the battery voltage: If the battery voltage decreases and thus the gain on the converter 7, the measurement signal e at block 7 the gain, compensatory, increased accordingly. The battery voltage drop acts simultaneously as a signal limitation by a limiter and is best and preferably simulated by a battery output voltage-controlled limiter block 90 _{b in} front of the loudspeaker model 92 or 92 _f according to FIG. 14.

With the provision of the Limiterblockes 90 according to FIG _b. 14, the blocks 90 can be omitted, as may 92 or 92 _f blocks 94 and 94 _f omitted when provision of the blocks, and there is a relatively low cost a battery voltage-independent, in this realization, stable Feedback suppression achieved according to the invention.

On the other hand, the function of the mentioned block 90 _{b can be taken over} completely by providing the battery output voltage-controlled block 90 or 90 _f according to FIGS. 4 and 5.

Taking into account the signal limitation caused by battery voltage drop by means of the controlled limiter blocks 90, 90 _f and 90 _b is of great importance in order to ensure the stability of the hearing aid with battery voltages which vary within wide limits.

In order to ensure the stability of the feedback suppression or compensation even in a very noisy environment, where, for example according to FIG. 11, the acoustic-electrical converter 1 is overdriven and thus becomes non-linear, a non-linear model of the acoustic-electrical converter 1, possibly also taking into account the behavior of the A / D converter 3, switched between the output of the compensator filter 15 (FIG. 1) or 15 _f (for example FIG. 11) and the subtraction input of the differential unit 13, depending on Arrangement operating in the frequency or time domain, as entered at 91 or 91 _f in FIG. 11. For the model 91, 91 _f , the explanations apply analogously to those which were made with regard to the model 92, 92 _{f of} the electrical-acoustic transducer.

A further improvement in the effect of the compensation filter section 15 _f can be achieved by superimposing, if necessary, noise r in the time domain, as shown schematically in FIG. 15, on the output side of the gain filter 5 _f .

For this purpose, as shown in FIG. 15, the instantaneous signal spectrum on the output side of the amplification filter 5 _{f is} examined on a spectrum detector 125, for example on how very individual spectral lines are superior in terms of performance, i.e. how much the spectrum profile is peaked, maW, generally, for example, the energy density distribution of the spectrum . If the spectrum characteristic monitored at the unit 125 exceeds a predetermined limit profile, such as a predetermined energy distribution from dominant spectral lines to other spectral lines, then digital noise r is preferably coupled into the superimposition unit 129 via a noise generator 127. In order to reduce the audibility of this noise, a filter unit, as shown at 133 in FIG. 16, can preferably be connected downstream of the noise generator 127, which filter controls the noise in such a way that it is sufficiently weak compared to the instantaneous signal transmitted at the converter 9 Useful signal, for example by 40dB is weaker.

As further shown in dashed lines at 131 in FIG. 15, the noise can also be coupled in the frequency domain if necessary will. If the noise is injected in the time domain, for example, the noise generator 127 consists of a BPRN, in the frequency range according to 127a in FIG. 17, for example from a table with noise spectra or a Noise algorithm.

16, starting from the representation of FIG. 15, shows a preferred form of implementation of the noise lock in the time domain. For this purpose, the output signal of the amplification filter 5 _{f is} examined on a spectrum shape detector unit 125a, and when the spectrum shape leaves a predetermined limit characteristic, the output signal of the noise generator 127, which is passed through the linear filter 133, is represented by the signal u ( 15, preferably superimposed on the input side of the limiter unit 90. As shown with the control connection sc, the transmission behavior of the filter 133 is preferably controlled by the current spectrum.

FIG. 17 shows a preferred embodiment variant of the noise lock in the frequency range according to the dashed embodiment variant with block 131 of FIG. 15. The spectrum on the output side of the amplification filter 5 _f is examined on a spectrum shape detector unit 125 _b , analogously to the unit 125 a of FIG. 16. The output signal of a noise generator 127a, which, for example, noise spectra stored in tables and are available is, the output side of the amplifying filter 5 _f then layered over a shaping filter 137 of the spectrum, as shown schematically by the switch 135a, when the spectrum shape detecting unit 125 _b detects a current spectrum shape , which necessitates the noise switching mentioned. The noise in the frequency range is superimposed on an addition unit 129a.

The shaping filter 137 is in turn controlled by the current spectrum, for example on the output side of the gain filter 5 _f .

Claims (34)

1. Hearing aid having an acoustic/electric transducer (1) with an output-side A/D converter (3), and an electric/acoustic transducer (9) with an input-side D/A converter (7), an amplifying filter section (5) between the A/D and D/A converters (3, 7) and an adaptive compensator filter (15_f) whose signal input is operationally connected to the D/A converter input and whose signal output is operationally connected to an input of a subtraction unit (13), the second input of the subtraction unit (13) being operationally connected to the output of the A/D converter (3), while its output acts on an adaptation input (A_f) of the compensator filter (15_f) and on the input of the amplifying filter section (5_f), it being the case, furthermore, that the signals fed to the signal input (E_f) and adaptation input (A_f) of the compensator filter (15_f) are transformed from the time domain into the frequency domain at at least one transformation unit (20, 28) which carries out fast orthogonal transformation, characterized in that the at least one transformation unit (22, 20; 20, 28) is arranged on the output side of the subtraction unit (13), and an inverse transformation unit (24) corresponding to the transformation unit acts between the output of the compensator filter (15_f) and the assigned input of the subtraction unit (13).
2. Aid according to Claim 1, characterized in that one transformation unit (22, 20) each is connected upstream of the signal input (E_f) and of the adaptation input (A_f) of the compensator filter (15_f).
3. Aid according to Claim 1, characterized in that one transformation unit (20, 28) each is connected upstream of the adaptation input (A_f) of the compensator filter (15_f) and of the input of the amplifying filter (5_f), and an inverse transformation unit (26) is connected upstream of the input of the D/A converter (7).
4. Aid according to Claim 1, characterized in that a common transformation unit (30) is connected upstream jointly of the adaptation input (A_f) of the compensator filter (15_f) and of the input of the amplifying filter section (5_f).
5. Aid according to one of Claims 1 to 4, characterized in that a transformation unit (20, 28; 30; 30a, 32, 34) connected upstream of the input of the amplifying filter (5_f), an inverse transformation unit connected downstream of the input of the compensator filter (15_f), and an inverse transformation unit connected upstream (26) of the D/A converter operate using the overlap-save technique.
6. Aid according to one of Claims 1 to 5, characterized in that a transformation unit (30; 30a) connected upstream of the adaptation input (A_f) of the compensator filter (15_f) operates using the overlap-add technique.
7. Aid according to one of Claims 1 to 6, characterized in that connected downstream of the output of the subtraction unit (13) is a transformation unit (30a) which operates using the overlap-add technique, and its output acts on the adaptation input (A_f) of the compensator filter (15_f) and is fed to a block storage arrangement (32), in which successive mutually succeeding signal blocks formed using the overlap-add technique are stored, mutually assigned memory locations for mutually assigned block locations being added with the correct sign at an addition unit (34) in such a way that the output block of the addition unit represents a block in the overlap-save technique, and the output of the addition unit (34) is fed to the input of the amplifying filter section (5_f).
8. Aid according to one of Claims 1 to 7, characterized in that the amplifying filter section (5_f) comprises an amplifying filter (40) and a delay unit (42) connected downstream of the latter.
9. Aid according to one of Claims 7 or 8, characterized in that the compensator filter (15_f) comprises:

   delay stages (56) connected in series downstream of the input (E_f) of the compensator filter,

   a number 1 ≤ i ≤ L of component compensators (50), where component estimation signals Y i[k+1] for 1 ≤ i ≤ L are generated, k denoting the block number, counted for the time domain/frequency domain transformation on the output side of the subtraction unit (13),

   an addition unit (52), where the component estimation signals Y _i[k+1] of all 1 ≤ i ≤ L component compensators (50) are added and whose output forms the output of the compensator filter (15_f).
10. Aid according to Claim 9, characterized in that each component compensator (50) comprises:

    a component compensator input connected to the input (E_f) of the compensator filter (15_f) via a number of delay stages (56), the number of delay stages corresponding to the number of component compensators connected upstream of a component compensator, each delay stage (56) connecting the input and the output of a component compensator (50),

    a first multiplication stage (54) operationally connected to the output of the component compensator,

    connected downstream of the output of the first multiplication stage (54), an input of a second multiplication stage (58), whose second input is operationally connected to the adaptation input (A_f),

    the output of the second multiplication stage (58) acting via an accumulation unit (60) on one input of a third multiplication stage (64), whose second input is operationally connected to the input of the component compensator (50) and whose output acts on the addition unit (52).
11. Aid according to one of Claims 1 to 10, characterized in that connected upstream of the input of the amplifying filter stage is a transformation unit whose output signal, in addition to acting on the amplifying filter section, acts on a power-detection unit (70) whose output signal controls the effectiveness of a signal at the adaptation input whenever the energy of the signal at the output of the transformation unit exceeds a given threshold value.
12. Aid according to Claim 11, characterized in that acting on the second input of the first multiplication stage (54) is the output of a fourth multiplication unit (80), whose one input is fed a signal corresponding to a reference step size (µ₀), and whose second input is fed the output of a scaling unit (78) to which the outputs of two interpolation filters (72, 74) are fed, to both of which the output signal of the amplifying filter section is applied via the power-detection unit (70).
13. Aid according to Claim 12, characterized in that instead of one of the interpolation filters (74) a temporally constant signal is fed to the scaling unit (78) (γ = 1).
14. Aid according to one of Claims 12 or 13, characterized in that an inverse transformation unit (82), a zeroing unit (84) and a transformation unit (86) are connected between the output of the accumulation unit (60) and the input of the third multiplication stage (64).
15. Aid according to one of Claims 1 to 14, characterized in that an amplitude-limiting unit (90, 90_f) is connected upstream of the electric/acoustic (el/ak) transducer (9).
16. Aid according to one of Claims 3 to 14, characterized in that a transformation unit (22a) is connected upstream of the input of the compensation filter (15_f), and an inverse transformation unit (26a) as well as an amplitude-limiting unit (90) are connected downstream of the output of the amplifying filter section (5_f).
17. Aid according to one of Claims 1 to 16, characterized in that at least one unit (91, 91_f, 92, 92_f), which operates in the frequency and/or time domain and models the electric/acoustic transducer (9) and/or the acoustic/electric transducer (1), is connected upstream and/or downstream of the compensation filter (15_f).
18. Aid according to one of Claims 3 or 4, characterized in that a transformation unit (22a) is connected upstream of the compensation filter (15_f), and an inverse transformation unit (26a) is connected downstream of the amplifying filter section (5_f), and a unit (92) modelling the electric/acoustic transducer (9) and/or the acoustic/electric transducer (1) in the time domain is connected upstream of the transformation unit (22a) connected upstream of the compensation filter (15_f), and/or a unit (92_f) modelling the electric/acoustic transducer (9) and/or the acoustic/electric transducer (1) in the frequency domain is connected downstream of the transformation unit (22a).
19. Aid according to one of Claims 1 to 18, characterized in that one unit (92; 94) each which preferably models the electric/acoustic transducer (9) and/or the acoustic/electric transducer (1) in the time domain is provided, connected upstream of the compensation filter (15_f) and downstream of the amplifying filter (5_f).
20. Aid according to one of Claims 1 to 19, characterized in that it comprises at least one limiter unit (90, 90_f, 90_b) operating in the time domain or frequency domain, and is fed electrically by a battery, and in that a measuring device whose output controls the limiter unit is further provided for detecting the ACTUAL battery state (122).
21. Aid according to one of Claims 1 to 20, characterized in that the D/A converter has a gain control input, the aid is fed by battery, and a measuring device (122) for the ACTUAL state of the feed battery (120) is provided whose output is led to the gain control input of the D/A converter.
22. Aid according to one of Claims 1 to 21, characterized in that it comprises at least one unit (91, 91_f, 92, 92_f, 94, 94_f) preferably modelling the electric/acoustic transducer (9) and/or the acoustic/electric transducer (1) in the time domain, is fed by battery, and comprises a measuring device (122) for the ACTUAL state of the battery (120) whose output is led to parameter control inputs on the at least one modelling unit.
23. Aid according to one of Claims 1 to 22, characterized in that, preferably in the time domain, the compensation filter (15_f) is fed (129, 135a) a noise signal (r) on the input side, preferably at least from time to time.
24. Aid according to Claim 23, characterized in that the signal is fed on the output side of the amplifying filter to a detection unit (125, 125_a, 125_b), where the instantaneous form of its spectrum is examined as to whether it fulfils a prescribed condition or not, and in that the output signal of the detection unit controls the connection (135, 135_a) of the noise signal.
25. Aid according to Claim 23, characterized in that the noise signal is fed via a shaping filter which is controlled by the instantaneous spectrum of the output signal of the subtraction unit.
26. Aid according to one of Claims 1 to 22, characterized in that provided in the frequency domain is a noise generator (127a) whose output signal is injected in a superimposed fashion on the output side of the amplifying filter section (5_f).
27. Aid according to Claim 26, characterized in that the output of the noise generator (127a) is led via a shaping filter (137) which is fed, as control signal for its shaping performance, the instantaneous spectrum of a signal on the output side of the subtraction unit (13).
28. Aid according to one of Claims 1 to 27, characterized in that, at least from time to time, a noise signal is superimposed on the signal transmitted electrically thereto, via a linear filter (133) whose response characteristic is controlled by the instantaneous spectrum of the electrically transmitted signal.
29. Aid according to one of Claims 1 to 28, characterized in that a unit (91, 91_f) modelling the response characteristic of the acoustic/electric transducer (1) of the aid is provided in a compensation branch.
30. Aid according to one of Claims 1 to 29, characterized in that an electric transmission unit is provided which simulates the response characteristic of the electric/acoustic transducer and comprises a linear transmission component (100, 102, 106, 108) and a nonlinear transmission component (104).
31. Aid according to Claim 30, characterized in that the linear transmission component comprises linear amplifiers as well as filters.
32. Aid according to Claim 31, characterized in that the linear transmission component comprises an input-side ante-filter which essentially has a low-pass characteristic and downstream of which the nonlinear transmission component (104) is connected, a compensator filter unit (108) having a frequency response which is essentially inverse to the frequency response of the ante-filter being connected downstream of the latter.
33. Aid according to Claim 32, characterized in that a linear amplifying element (102) is connected upstream of the nonlinear transmission unit (104), and a linear amplifying compensation element (106) which compensates the gain of the linear amplifying element is connected downstream of said nonlinear transmission unit (104).
34. Aid according to Claim 31, characterized in that the nonlinear unit has a response characteristic with saturation performance.

Images