DK1648197T4 - Method and apparatus for reducing feedback in an acoustic system - Google Patents

Abstract

A feedback signal (RS) is detected in an incoming signal (ES), which is processed by relying on a detected feedback signal in an outgoing signal (AS), which is modulated (MO) so that the feedback signal is also modulated correspondingly. This modulation detects the feedback signal and has to be unheard by hearing-aid wearers. An independent claim is also included for a signal-processing device for an acoustic system.

Description

The present invention relates to a method for reducing feedback in an audio system by detecting a feedback signal in an input signal and processing the input signal on the basis of the detected feedback signal to produce an output signal. The present invention also relates to an appropriate signal processing apparatus for an audio system. By way of example, the audio system is a mobile radio, a headset, an auditorium sound system and particularly a hearing aid or middle ear implant.

Audio feedback, called feedback below, frequently arises in hearing aids, particularly when they are high-gain devices. This feedback is expressed as severe oscillations at a particular frequency and can be heard as whistling. This “whistling” is usually very unpleasant both for the hearing aid wearer himself and for people who are relatively close by. Feedback can arise, for example, when sound which is picked up via the hearing aid’s microphone, amplified by a signal amplifier and output via the earphone gets back to the microphone and is amplified again.

The simplest approach to feedback reduction is to reduce the hearing aid’s gain on a permanent basis, so that the loop gain remains below the critical limit value even in adverse situations. However, the crucial drawback is that this limitation means that it is no longer possible to achieve the gains which are required for more severe hardness of hearing. Other approaches measure the loop gain during the hearing aid adjustment and reduce the gain specifically in the critical range using “notch filters” (narrowband rejection filters). Since the loop gains can change constantly in everyday life, however, as outlined above, the benefit is likewise limited.

To reduce feedback dynamically, a series of adaptive algorithms have been proposed which automatically adapt themselves to the respective feedback situation and effect appropriate measures. These methods can be roughly divided into two classes:

The first class comprises the “compensation algorithms”, which use adaptive filters to estimate the feedback component in the microphone signal and to neutralize it by subtraction and hence do not adversely affect the hearing aid’s gain. However, these compensation methods presuppose uncorrelated, i.e. ideally white, input signals. Tonal input signals, which always have a higher level of time correlation, result in incorrect estimation of the feedback path, which can lead to the tonal input signal itself being subtracted by mistake.

The second class includes the algorithms which do not become active until feedback whistling is present. They generally include a mechanism for detecting the feedback whistling which continuously monitors the microphone signal for feedback oscillation. If oscillations typical of feedback are detected, the hearing aid’s gain is reduced at the appropriate point until the loop gain drops below the critical limit. The gain reduction can be effected by lowering a frequency channel or by activating a suitable narrowband rejection filter (notch filter), for example. A drawback is that the oscillation detectors cannot in principle distinguish between tonal input signals and feedback whistling. The result is that tonal input signals are thought to be feedback oscillations and are then inadmissibly lowered in level by the reduction mechanism (e.g. notch filter).

In summary, it can be stated that the manner of operation of all of the adaptive feedback reduction methods is adversely affected by input signals which have a tonal character shaped by dominant sinusoidal signal components (e.g. sounds from a triangle, alarm signals). This frequently results in unacceptable tone impairments in the input signal. This is the starting point of the present notification of invention.

The compensation algorithms frequently involve delay elements with a decorre-lating effect being introduced into the signal processing chain in order to prevent tonal signal sections with a length which is characteristic of voice signals from being noticeably attacked. However, echo effects and irritations by desynchronized visual and audio information mean that only delays in the millisecond range are admissible. It is therefore not possible to avoid reducing music signals, for example, which are frequently correlated over a much longer period.

Another countermeasure is to slow down the filter’s adaptation such that all relevant tonal ambient signals are not attacked. However, a consequence of this is also that the compensation filter is no longer able to follow rapid changes in the feedback path fast enough, which means that feedback whistling is produced for a certain time and does not disappear again until the feedback path has stabilized and the filter is adapted with sufficient accuracy again.

The negative consequences of incorrect detection by oscillation detectors are countered by virtue of the resultant gain lowering being effected only to a limited extent, which means that tonal useful signals (e.g. alarm signals) which have been mistaken for feedback oscillations, for example, continue to remain audible. However, this holds the risk that in a feedback situation the gain is not lowered sufficiently to drop below the critical limit, and hence the feedback whistling is not eliminated.

The document WO 2001/06746-A2 discloses step size control for the compensation filter, where the feedback detector operates on the basis of the principle of bandwidth detection. If the bandwidth detector recognizes a narrow bandwidth for the hearing aid’s input signal in the frequency band which is susceptible to feedback whistling, it is assumed that there is feedback whistling. However, it is not possible to distinguish natural, narrowband signals with spectral components in this frequency band, such as music. In addition, the feedback whistling must represent a dominant signal component in order to be recognized.

Also, the document EP 1 052 881-A2 discloses an oscillation detector for detecting feedback. In this case too, the feedback whistling needs to be very distinctly pronounced in order to be recognized.

The document WO 2001/95578-A2 describes detection of feedback whistling by estimating the variance in the frequency estimation of the hearing aid’s input signal. This method also has the drawbacks cited above.

In addition, the document DE 199 04 538-C1 proposes the selective attenuation of individual frequency bands. In this case, frequency bands in which there is feedback whistling are subjected to a greater level of attenuation by an added attenuation element than could be expected for useful signals. The intervention in the forward signal path is sometimes audible to the hearing aid wearer and in addition the detection is probably slow, since the bands are ideally examined in succession.

Another method for reducing feedback in audio systems is known from the document US 6,347,148 B1. In this case, the spectrum of an input signal is estimated and a psychoacoustic model is used to generate a control signal. The control signal is used to actuate a noise source which can be used to produce an inaudible noise signal on the basis of the noise signal. This document also describes the option of impressing short noise signals of a prescribed duration onto the output signal. The noise signals in the input signal are used to reduce feedback signals.

The document US 5 748 751 A describes echo reduction by means of decorrelation of the input signal and the output signal. The decorrelation is performed with a phrase modulator that is operated at a particular frequency.

In addition, the document US 5 412 734 and WO 00/44113 A1 shows feedback reduction by means of AM, FM or QPSK. The accordingly modulated signal is added to the fundamental signal and detected again upon feedback.

The object of the present invention is therefore to improve the reduction of feedback in a hearing aid further.

The invention achieves this object by means of a method according to Claim 1 and a signal processing apparatus according to Claim 7.

The underlying idea is to impress features which the hearing aid wearer cannot perceive onto the output signal from the audio system and particularly from the hearing aid. This makes it possible to use appropriate analysis of the input signal to determine whether the input signal is feedback or a “normal” external input signal (useful signal). Determining the form of the feature in the input signal also allows inferences about corresponding ratios of feedback to useful signal. This can then be used directly to control feedback reduction algorithms.

Advantageously, it is thus possible to determine, in the course of operation and totally inconspicuously or inaudibly, the extent to which a microphone or the hearing aid’s microphone is hearing feedback signals, which allows a significant improvement in the control and action of the known feedback reduction algorithms.

Preferably, the input signal is processed using an adaptable filter whose adaptation speed and/or level of action is dependent on the quantity of the detected feedback signal. In particular, it is advantageous if the adaptation speed rises in proportion to the quantity of the detected feedback signal. If the feature analysis of the input signal is then negative, for example, i.e. it does not contain a feedback signal, the adaptation speed of the aforementioned compensation filter can be slowed down such that the filter is not adjusted by tonal input signals and these signals are not attacked. If the feature is detected in the input signal, on the other hand, the level of action and/or speed of the feedback compensator is set to the value at which feedback is rejected in optimum fashion.

If a feedback signal is detected then at least one notch filter for processing the input signal can be activated.

The phase modulation has no particular susceptibility with regard to incorrect detection for narrowband signals. A feedback situation can actually be recognized before the feedback whistling becomes dominant in the signal mix.

Feedback can be detected separately in a plurality of subbands. It is thus possible to adjust the gain, but also the reduction of feedback, individually in the individual subbands. A closed loop in the signal processing apparatus can be used for signal modification. In this case, the modulated signal passes through the loop a plurality of times, so that the corresponding signal modification is brought about.

The present invention is now explained in more detail with reference to the appended drawings, in which: FIGURE 1 shows a hearing aid system based on the prior art; FIGURE 2 shows a hearing aid system based on a first embodiment of the present invention; FIGURE 3 shows a hearing aid system based on a second embodiment of the present invention; and FIGURE 4 shows a feedback detector with a filter bank.

The exemplary embodiments outlined in more detail below are preferred embodiments of the present invention. To provide a better understanding of the invention, the prior art is first of all explained in more detail with reference to FIGURE 1. FIGURE 1 shows a hearing aid HG, whose input is formed by a microphone M. The signal picked up is forwarded as input signal ES to a processing unit V. There, it is processed and possibly amplified. The resultant output signal AS is sent to an earphone H. A feedback path RP is used to feed back the output sig- nal from the earphone H to the microphone M. When the supply is open, there is primarily an audio feedback path. Generally, electromagnetic, electrical, magnetic and other feedback loops are also conceivable, however. The feedback signal RS resulting from the feedback path is added to a useful signal NS, and the summed signal is picked up by the microphone M.

The signal path from the microphone M via the hearing aid processing V, the earphone H, the feedback path RP back to the microphone M is a loop. If the loop gain, i.e. the gain to which a signal is subjected when it passes through this loop, has a value of at least 1.0 at at least one frequency and if the phase condition is satisfied then feedback whistling occurs. Even if the loop gain is just below this limit, audible feedback effects occur, e.g. tone changes.

One successful method for rejecting the feedback effects is digital simulation of the feedback path RP. This feedback path is simulated by an adaptive filter AF to which the output signal from the processing unit V is supplied. An appropriate compensation signal KS coming from the compensating, adaptive filter AF is subtracted from the input signal ES for the microphone M, and the resultant difference signal is supplied to the processing unit V.

There are thus two paths, first the outer feedback path RP and secondly the digital compensation path simulated by means of the adaptive filter AF. The resultant signals on both paths are subtracted from one another at the input to the appliance, as shown in FIGURE 1 by the two addition units. Ideally, this cancels the effect of the outer feedback path RP.

An important component in the adaptive algorithm for determining the feedback path is its step size control. This indicates the speed at which the adaptive compensation filter adapts itself to the outer feedback path RP. Since there is no appropriate compromise for a permanently set step size, this needs to be adapted to the respective present situation which the system is in.

In principle, a large step size is desirable for rapid adaptation of the adaptive compensation filter AF to the outer feedback path RP. A drawback of a large step size, however, is the production of perceptible signal artifacts.

If a feedback situation is not present, the step size should be extremely small. In this context, a feedback situation is denoted as that situation in which the loop gain is just below 1 or is greater than or equal to 1 and the phase condition is satisfied at at least one frequency. If a feedback situation occurs, on the other hand, the step size should be or become large. This ensures that the algorithm adapts the adaptive compensation filter AF only when the latter’s characteristic differs significantly from the characteristic of the feedback path RP, i.e. when readaptation is required. For this purpose, a feedback detector is provided.

To be able to detect feedback reliably, the invention provides a modulation device MO which is connected between the processing unit V and the earphone H, as shown in FIGURE 2. This device modulates the output signal AS to produce a modulated output signal AS’. The modulation of the output signal AS is not perceptible. In a feedback situation, a significant component of the sound signal which is output by the earphone FI gets back to the microphone M and is picked up by the appliance together with the ambient signal. FIGURE 2 indicates that the feedback path RP can basically be in any form. That is to say that it is not necessary to have an audio feedback signal RS, as indicated in FIGURE 1, which is added to an audio useful signal NS before the microphone M. Rather, the feedback into the microphone M may also be effected by means of structure-borne noise or electromagnetic interference, for example.

The input signal ES for the microphone M is analyzed by a feedback detector RD. This allows the feedback signal RS to be detected on the basis of its modulation. A downstream controller S actuates the adaptive compensation filter AF in line with the detection result from the feedback detector RD. This changes the adaptation speed of the adaptive filter AF, for example.

The exemplary embodiment in FIGURE 3 essentially corresponds to that in FIGURE 2. In this case, the feedback path is of purely audio nature as in the example in FIGURE 1, which means that the feedback signal is added to the useful signal before the microphone M.

Another difference from the circuit in FIGURE 2 is that the signal for the feedback detector RD is tapped off not directly after the microphone M but rather after subtracting the compensation signal from the adaptive filter AF at point A. The level of signal modulation produced at point A is a depiction of the difference between the action of the feedback path RP and the action of the adaptive compensation filter AF. However, there is no fundamental difference from the embodiment shown in FIGURE 2, in which the signal to be analyzed is tapped off directly after the microphone M.

In addition, FIGURE 3 indicates that a step size controller can be incorporated into the feedback detector RD, which means that it is possible to dispense with a separate control chip. The other components of the exemplary embodiment in FIGURE 3 correspond to those of the exemplary embodiment in FIGURE 2. In this regard, reference is thus made to the description relating to FIGURE 2.

In the exemplary embodiment shown in FIGURE 3, the phase of the output signal AS is modulated, since the human ear is largely insensitive toward phase changes. In a specific example, the phase of the output signal AS is linearly rotated forward and backward between two phase values at a particular frequency, in this case called the modulation frequency f_mod. By way of example, the phase values are a and α+π/2, where a is any fixed phase. In the feedback situation, a detectable tremolo component at a frequency of f_mod develops in the signal loop.

The tremolo component can be detected using a frequency demodulator in the feedback detector RD. In this case, it is beneficial to design the feedback detector RD to have a filter bank, as shown in FIGURE 4, which splits the input signal ES into subbands using a plurality of bandpass filters BP1, BP2, ..., BPn, for example. Downstream of each bandpass filter there is respectively arranged an analysis unit AE and a threshold value switch SW. The output signals from the signal paths for each subband are optionally supplied to an OR gate OR. The respective analysis units AE and threshold value switches SW may have the same design as one another. Hence, in this example, the analysis in each subband path takes place in the same way. If the analysis result in a band exceeds a certain threshold, the associated threshold value switch SW responds, i.e. a feedback situation is recognized for this band.

This information can be used for an adaptive compensation filter AF adapting in subbands for the purpose of step size control. If an adaptive filter AF is used in the whole band, on the other hand, the results of the subband detection operations need to be combined into a whole-band detection statement using a logic OR function. Even the special instance in which the whole band is analyzed as one, with n = 1, results in an operable system. However, the error detection rate is lower for a larger n, e.g. n = 16.

The step size control of the adaptive filter AF can also be effected in more differentiated fashion besides the simple threshold value decision as shown in FIGURE 4, where only the presence or absence of feedback is detected. By way of example, the step size can be ascertained by virtue of proportional recalculation of the estimated level of the signal modulation at point A. This may also be done using a subband approach again. The greater the signal modification recognized, the higher the need for readaptation would then be, i.e. the higher the necessary step size would need to be selected. The step size can thus be continually adapted to the signal modulation. In the case of a pure threshold value decision, the step size is, by contrast, stepped up for a certain firmly prescribed time or for the time frame in which feedback is detected. Otherwise, it assumes a small value.

Claims (12)

1. A method for reducing the feedback in an acoustic system (HG) and in which: - detecting a feedback signal (RS) in an input signal (ES), - processing the input signal (ES) to produce an output signal (AS) ) on the basis of the detected feedback signal (RS), and - modulation (MO) of the output signal (AS) by applying a feature of the output signal (AS) so that the feedback signal (RS) is similarly modulated, thereby detecting - the feedback signal (RS) on the basis of the presence of the applied feature, characterized in that the modulation (MO) is provided by phase modulation and the phase is thereby rotated back and forth between two phase values at a predetermined frequency. The method of claim 1, wherein the processing of the input signal (ES) is performed by an adaptive filter (AF), whose adaptation rate and / or effect level depends on the amount of feedback signal detected (RS). The method of claim 2, wherein the rate of adaptation and / or the level of action increases proportionally to the amount of feedback signal detected (RS). The method according to any one of the preceding claims, wherein, in the case where a feedback signal is detected, at least one Notch filter for processing the input signal (ES) is activated. A method according to any one of the preceding claims, wherein the detection is performed separately in a number of sub-bands. The method of claim 9, wherein an adaptive filter (AF) is individually adapted in the subbands. 7. An acoustic system (HG) signal processing device having - a processing device (V, AF) for generating an output signal (AS) from an input signal (ES) having regard to a feedback signal (RS), - a modulation device for modulation (MO) of the input signal by applying a feature to the output signal (AS) so that upon feedback a corresponding modulated feedback signal (RS) is present, and - a detection device (RD) for detecting the modulated feedback signal (RS), whereby - the feedback signal can be detected by the detection device due to the presence of the printed feature, characterized in that - the output signal (AS) can be modulated by the modulation device (MO) in phase modulation, and - the phase is rotatable back and forth between two predetermined values. frequency. The signal processing device of claim 7, wherein the processing device (V, AF) has an adaptable filter (AF), whose adaptation rate and / or effect level depends on the amount of feedback signal (RS). The signal processing device of claim 8, wherein the adaptation rate and / or the level of action increases proportionally to the amount of feedback signal (RS). The signal processing device according to any one of claims 7 to 9, wherein the processing device can be used to activate at least one Notch filter if a feedback signal is detected. Signal processing device according to one of claims 7 to 10, which at all times has a multi-band detection device (RD). Signal processing device according to one of claims 1 to 11, which utilizes a closed loop to produce a signal modification.