JP2002526961A - Band-limited adaptive feedback canceller for hearing aids - Google Patents

Abstract

(57) Abstract: An improved method for adaptively canceling the acoustic feedback of hearing aids and other audio amplifiers. Feedback cancellation is limited to the frequency band encompassing all unstable frequencies. By limiting the bandwidth of the feedback cancellation signal, the distortion due to the adaptive filter is minimized and limited only to the unstable feedback region. Relatively simple signal processing algorithms are used to produce highly effective results with minimal signal distortion.

Description

DETAILED DESCRIPTION OF THE INVENTION

BACKGROUND OF THE INVENTION (1. Related Application) This application is a concurrent pending provisional application Seri, filed September 30, 1998.
al No. Claim the benefits of 60 / 102,557.

2. Field of the Invention [0002] The present invention relates generally to the field of audio amplification, and in particular, to audio amplification of hearing aids. More specifically, the present invention provides a method for efficiently canceling the acoustic feedback of a hearing aid.

[0003] 3. Acoustic feedback of hearing aids, returning to the microphone from an electroacoustic transducer (usually called a "receiver"), is common and difficult to suppress. The return can create an audible whistle that is irritating to the wearer of the hearing aid, so that the wearer often has to reduce the volume below a desired level, reducing the effectiveness of the hearing aid.

One reason that it is difficult to effectively suppress the acoustic feedback of a hearing aid is that the frequency at which the feedback occurs changes with changes in external circumstances. Therefore, a form of adaptive cancellation is needed to effectively cancel the feedback without undesirably degrading the amplified signal. Various techniques have been proposed to perform adaptive feedback cancellation. Such techniques are described, for example, in US Pat. No. 5,016,280,
Nos. 5,091,952 and 5,259,033.

One of the major design goals of a hearing aid is to reduce its physical volume. Most wearers prefer hearing aids that can be worn entirely in the ear. Advances in microelectronics have greatly improved the signal processing capabilities of inner ear (ITE) hearing aids. Even so, effective sound cancellation remains a practical design challenge. The prior art always requires some compromise. As a result of such a compromise, the hearing aid may exhibit a slight increase in maximum stable gain, slow filter adaptation, distortion, interference, and / or lack of adaptation to the individual wearer.

SUMMARY OF THE INVENTION The present invention provides an improved method for adaptively canceling the acoustic feedback of hearing aids and other audio amplifiers. Feedback cancellation is limited to certain frequency bands that include all of the unstable frequencies. By limiting feedback cancellation in this manner, it is possible to produce highly effective results with minimal signal distortion using relatively simple signal processing algorithms.

In order to implement the present invention, first the unstable feedback frequency must be identified. This is achieved by various techniques for actually measuring the ear, from which it is possible to obtain complex open-loop transfer functions. After the unstable feedback frequency is identified, a band-limited adaptive filter is implemented. By limiting the bandwidth of the adaptation in this way, the adaptive feedback canceller can adapt very quickly in the range of unstable frequencies and the adaptive noise is relatively low. By limiting the bandwidth of the feedback cancellation signal, the distortion due to the adaptive filter is minimized and limited only to the unstable feedback region. Compared with the wideband feedback cancellation, the distortion generated by the band limited feedback canceller of the present invention is small, and therefore, the output sound quality is greatly improved.

DETAILED DESCRIPTION OF THE INVENTION In the following description, for purposes of explanation but not limitation, certain details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods and devices are omitted so as not to obscure the description of the present invention with unnecessary detail.

FIG. 1 is a functional block diagram of a hearing aid 10 worn on the ear of a hearing aid user. The hearing aid 10 comprises a microphone 12 and a microphone preamplifier circuit 14.
, An analog / digital converter 16, a signal processing circuit 18, a digital / analog converter 20, and a receiver 22. The signal processing circuit 18 has a transfer function of K (f) from the point C to the point D. The return path from point D to point C includes a digital-to-analog converter 20, a hearing aid receiver 22, an acousto-mechanical coupling between the receiver and the microphone, an analog conditioning circuit 14, an analog-to-digital converter 16, including.
β (f) is the transfer function of the feedback path.

FIG. 2 shows a simplified hearing aid worn on the ear. The open loop transfer function of the hearing aid is defined as K (f) β (f). The hearing aid has an open-loop transfer function whose Nyquist instability criterion magnitude: | K (f _u ) · β (f _u ) | ≧ 1 phase: ∠K (f _u ) β (f _u ) = n × 360 ° An unstable feedback is generated only at the frequency f _u that satisfies Equation 1. The hearing aid is in a quasi-oscillating state when the above phase condition is satisfied and the magnitude of the open loop transfer function of the hearing aid is close to, but not less than unity gain. Both quasi-oscillations and oscillations are offensive to the hearing aid wearer and must be eliminated. In the following description, both the unstable feedback frequency and the quasi-oscillation frequency are referred to as unstable feedback frequencies. An efficient feedback cancellation algorithm needs to operate in the frequency domain including all _fu . When feedback cancellation is limited to this frequency domain, it has been found that relatively simple algorithms can produce highly effective results with minimal signal distortion. None of the feedback cancellation methods for hearing aid systems developed earlier make this finding.

In order to obtain an open-loop transfer function, the prior art method employs a closed-loop system behind the microphone preamplifier and before the hearing aid processing module, as shown between points C and D in FIG. Open and create a standard two-channel transfer function measurement with a single analyzer. However, this method cannot be used successfully with commercially available hearing aids because it is impractical to disconnect the hearing aid circuit. We have developed the following three modules to obtain an open loop transfer function without disconnecting the hearing aid circuit.

The first method uses a probe microphone and obtains a probe microphone signal near the eardrum with a hearing aid in the ear canal, thereby obtaining an accurate open-loop transfer function K (f) from the closed-loop measurement. Obtain β (f). This method is disclosed in co-owned patent application Ser. No. 08 / 926,320, the specification of which is hereby incorporated by reference.
Incorporated herein by reference. Briefly, the method involves acquiring a probe microphone signal near the eardrum and calculates the open-loop transfer function of the hearing aid with three gain settings G1, G2, and G3. Here, G3 = 0 (that is, the hearing aid is removed). The method and procedure for detecting and digitizing acoustic signals with a probe microphone for use in the present invention is described, in part, in commonly-owned US Pat. No. 5,325,436, Method.
of Signal Processing for Maintaining
Directional Listening with Listening Ai
ds ”.

A complex open loop transfer function K (f) β (f) can be obtained as follows.

(Equation 1) _Where H _1AB (f), H _2AB (f), and H _3AB (f) are the closed-loop transfer functions from point A to point B in FIG. 1 measured at G1, G2, and G3. Point A is at the hearing aid microphone input and point B is the hearing aid receiver output, ie, the probe
In the microphone.

Since the probe microphone must be used for closed-loop measurements, the ventilation system of the hearing aid is partially interrupted by the probe tube. Therefore,
The acquired open-loop transfer function is only an approximation of the open-loop transfer function associated with an open hearing aid vent. Also, closed-loop transfer function measurements are sensitive to head movements and changes in the surrounding environment during the measurement.

The second method does not require a probe microphone measurement. It is necessary to configure an adaptive feedback canceller in the hearing aid and it operates in the wideband mode shown in FIG. An adaptive digital filter (ADF) 30 for the hearing aid includes a digital-to-analog converter 20, a receiver 22, a coupling path between the receiver and the microphone, a return path including the microphone 12, the AGC 14, and the analog-to-digital converter 16. Provide an estimate.

Data is collected for a hearing aid wearer sitting in a quiet room. A white noise signal is generated in a sound field via a loudspeaker. The hearing aid is programmed with a known reference response. The wearer of the hearing aid has been instructed to adjust the hearing aid gain below the discomfort level (UCL). While the white noise is moving in the sound field, the ADF of the hearing aid
Adapt the hearing aid to match the return path. After the ADF has converged, the filter's coefficient vector W (n) is read from the hearing aid with "delay". Feedback transfer function β '
(F) can be estimated as follows. β ′ (f) = e ^−j2πfD × W _n (f) Equation 3 where D is the delay introduced by the “delay” block 32 and W _n (f) is the frequency response of the ADF at the time index n. And the impulse response to it is W (n)
It is. To improve the accuracy of the measurement, it is possible to obtain several coefficient vectors W (n _i ) with different time indexes n _i from the ADF, average them and calculate W _n (f) in equation (3). it can.

To compensate for the individual's hearing loss, the hearing aid reference response used during the measurement is replaced by the desired hearing aid response K (f). The complex open loop transfer function associated with the desired hearing aid response K (f) can be estimated by K (f) β ′ (f). K (f) β ′ (f) = e− ^j2πfD × W _n (f) × K (f) Equation (4)

The unstable feedback frequency can then be determined by the open loop transfer function estimated based on equation 1. Compared to the first method, this method does not require probe microphone measurements. Therefore, the ventilation system of the hearing aid
Not shut off during measurement. The measurement is not sensitive to head movements and changes in the surrounding environment. No additional circuitry needs to be added to the hearing aid to support the measurement. While white noise serves as a test signal for measurement, the signal coming directly from the loudspeaker also acts as interference noise to the ADF since it carries no information about the return path. Therefore,
White noise introduces adaptive noise into the ADF coefficients. To achieve optimal results, the level of the white noise signal should be as low as possible, but above the noise floor of the room. The hearing aid gain should be as high as possible, but the UCL
Must be lower and have no audible feedback. Further, the hearing aid is configured in an open loop during the measurement. Unstable feedback, but not UCL, may be the limiting factor for the hearing aid gain setting.

The third method is similar to the second method described above. The pseudo-random noise generator shown in FIG. 5 is used to replace the external white noise test signal and the hearing aid processing block is removed. This method has two advantages over the second method described above. (1) Adaptive interference introduced by external test signals is eliminated by using an internal pseudo-random noise generator. (2) The hearing aid is operated in an open loop configuration during the measurement. Therefore, the gain of the hearing aid can be set as high as possible without the problem of instability. This method requires adding a noise generator and some control logic to the hearing aid. Open-loop transfer function measurement
Only required during the initial installation process.

FIGS. 4 and 5 both show a general wideband adaptive feedback canceller. The digital feedback cancellation signal is generated by passing the hearing aid digital output signal through the ADF at point B, which approximates the feedback path β (f). The digital feedback cancellation signal is
It is subtracted from the digitized microphone output at point C, removing the feedback signal from the hearing aid.

The ADF 30 has an adjustable filter and uses the filter coefficients to generate a feedback cancellation signal. The coefficient adaptation section 34 adjusts the coefficients of the filter to approximate the feedback path. There are various adaptive filtering methods, with different filtering structures and different adaptive algorithms. Some show very good convergence behavior and accuracy, but require a lot of computation. For example, the least squares (LS) algorithm belongs to this category. For applications of adaptive feedback cancellation for hearing aids, a simple Stochastic-Gradient (SG) algorithm provides well-acceptable performance. The simplicity of the adaptive filtering algorithm is a very important factor for hearing aid applications, as it is desirable for the hearing aid to minimize hardware requirements. Due to this simplicity and sufficient convergence performance, normalized Least Mean Squares (LMS) for ADF coefficient adaptation
Select an FIR (finite impulse response) filter that uses the algorithm. In order to make the adaptive filtering algorithm more suitable for the problem of adaptive feedback cancellation of the hearing aid, a normalized LMS algorithm, with the adaptive step size normalized to the ADF input and error, as discussed below, Adopted.

In general, adaptive filters are time-varying systems, which can track a time-varying feedback path. However, also unwanted changes due to the adaptive noise of the adaptive filter can introduce unpleasant distortion in the output of the hearing aid. Therefore, adjustment of the adaptive filter should be minimized unless it is necessary to eliminate unstable feedback.

Signals coming from an external sound source (external input signals), other than the signal re-supplied to the microphone of the hearing aid from the output of the hearing aid, are generally regarded as interference with the adaptation of the adaptive filter. External input signals usually introduce bias into the optimal solution for adaptive filters. When the characteristics of the external input signal change, the bias changes. Also, external input signals cause misadjustment noise during the adaptation process. External inputs to the adaptation process by removing frequency components that are not likely to have potential oscillation problems (unstable and quasi-oscillation) from the adaptation signals e (n) and x (n) used in the adaptation process Signal interference can be greatly reduced.

For the modified normalized LMS algorithm, the convergence speed of the adaptive filter is controlled by the adaptive step size, which is the combined signal power e (n) and x (n)
). Reducing the combined signal power results in a larger adaptive step size, which increases the convergence speed. Since the components inappropriate for the oscillation problem are removed from the adaptive signal, the combined signal powers e (n) and x (n) are reduced,
Therefore, it is possible to increase the convergence speed by using a larger adaptive step size. The feedback canceller can better track the rapid gain change of the AGC due to the dynamic feedback path and the sharp change of the input signal at a faster convergence speed.

Furthermore, due to the recursive topology used in adaptive feedback cancellation, the periodicity of the external input signal cancels out the original signal (external input signal) at the output of the hearing aid. By removing a frequency component inappropriate for the oscillation problem from the adaptive signal, the periodicity of the adaptive signal derived from the external input signal can be reduced. Thus, the problem of erasing the original input signal at the output of the hearing aid is reduced.

Furthermore, adaptive filters always work better at frequencies where there is a high energy. This is because the adaptive feedback canceller is active at the oscillation frequency only when the energy of the oscillating component present in the adaptive filter input signal and the error signal is equal to or greater than the peak energy of the spectrum of the external input signal. Means to act. As a result, the magnitude of the frequency component at the unstable feedback frequency is built up and then suppressed up and down around the external input spectral peak. This is a so-called residual oscillation component modulation effect.

By removing unnecessary frequency components from the adaptive signal, the peak spectral level is significantly reduced. Therefore, the magnitude of the residual oscillation component is significantly reduced.

This concept of removing unnecessary frequency components from the adaptation signal is very important and advantageous when the external input signal is a speech signal. As is well known, speech contains most of its energy and periodicity at low frequencies where unstable feedback is less likely to occur. Therefore, the low frequency components of the speech are considered unnecessary for adaptive filtering and can be easily removed by adapting a high-pass (or band-pass) filter to the adaptive signal. The cutoff frequency of the high-pass filter is generally set to 200 Hz, which is lower than the lowest unstable feedback frequency.

As mentioned above, unstable feedback occurs only at certain frequencies. In order to effectively suppress oscillation at these frequencies without introducing distortion at other frequencies, the present invention provides an adaptive feedback canceller that adapts the bandwidth of the adaptive signal to the frequency domain where it is known to include the oscillation frequency. The adaptive feedback canceller is configured to limit. By doing so, the adaptive feedback canceller adapts very quickly in the oscillation frequency domain, has much less adaptive noise, and adapts very slowly at other frequencies. As a result, AD
The feedback cancellation signal generated by F is also limited to the same frequency range. Therefore, unlike the wide band method, the band limited feedback canceller causes less distortion and the output sound quality is greatly improved.

FIG. 6 shows one possible structure for implementing a band limited adaptive feedback canceller. B
PF1 and BPF2 are band-pass filters, and pass only a frequency range including all unstable feedback frequencies. BPF1 and BPF2 are e (n) and x (n
) To remove most unnecessary frequency components and improve the adaptation process. The hearing aid input signal is filtered by BPF2 and then uses it as the desired signal for adaptive filtering, so that the adaptive filter only produces a cancellation signal in the passband of BPF2. Select a Δ ₂ delayed linear phase FIR filter as the bandpass filter BPF2, and thus, using a pure delay Δ ₂ ,
Without introducing phase distortion in the main hearing aid signal path, at the cancellation junction SUM2,
It can be ensured that the unfiltered hearing aid input d (n- [Delta] ₂ ) is properly timed with the filtered desired signal d '(n). Additional delay delta ₂ in key hearing aid signal path to increase the number of the oscillation frequency, so to move the oscillation frequency to a lower frequency, the length of the filter BPF2 has to be minimized. Use delta ₁ sample delay in front of the filter BPF1, feedback cancellation signal is combined feedback signal and properly time, therefore, the impulse response of the adaptive filter, the estimated impulse response of the feedback path energy Guarantee to include most. On the other hand, the delay introduced by the filter BPF1 must not be too long, or the cancellation signal generated by the adaptive filter may lag behind the feedback signal. The group delay introduced by each block must satisfy the following conditions:

(Equation 2) Optimum delay delta ₁ of the sample can be in a hearing aid attachment process is determined based on the measurement of the feedback path. In order to use all ADF coefficients, the delay introduced by the adaptive filter should be minimized. Delay Δ ₁ must meet the following conditions.

(Equation 3) Select the next expression.

(Equation 4)

In the above equation, ε is a sample tolerance. Usually ε is chosen equal to two.

The output of the ADF is calculated on one sample at a time. In the time index n, M
The calculation for the ADF output tapped is described as follows.

(Equation 5) {W0 (n), w1 (n),..., W2 (n)} are ADF coefficients obtained by tapping M at the time index n.

The coefficients of the ADF are updated with the modified normalized LMS algorithm.
As in the conventional normalized LMS algorithm, the adaptive step size is reduced when the signal level is high and increases when the signal level is low. However, in feedback cancellation of the hearing aid, ADF
Input signal of the hearing aid processing module is delayed by BPF1 and delta _1. The power of the error signal e (n) is included to calculate the time step size as follows. e (n) = d ′ (n) −y (n), where d ′ is the output of the bandpass filter BPF2. Equation 9 p (n) = ρ · p (n−1) + x ² (n) + e ² ( n) Equation 10

(Equation 6) Where ρ is a forgetting factor, C is a constant value that controls the adaptation speed, and ME is a small number that prevents the singularity of μ (n). The ADF coefficient for the next time index n + 1 is updated as follows. w _k (n + 1) = w _k (n) + μ (n) · e (n) · x (n−k), where 0 ≦ k ≦ M−1 Equation 12 Due to the bandpass characteristics of BPF1 and BPF2, The adaptive signals e (n) and x (n) are
It only contains frequency components in the oscillation frequency range.

FIG. 7 shows another possible structure for implementing an adaptive band limited feedback canceller. BP
F1 is mainly used to limit the bandwidth of the erasure signal and to limit erasure artifacts to the erasure bandwidth. BPF2 and BPF3 are used to limit the bandwidth of the adaptive signal to the region of the oscillation frequency. BPF1, BPF2, and BPF3 are not necessarily implemented as linear phase FIR filters, but this passband must span all oscillation frequencies.

Using the filtered adaptive signal samples, update the time step size and ADF coefficients as follows. p (n) = ρ · p (n−1) + x ′ ² (n) + e ′ ² (n) Equation 13

(Equation 7) w _k (n + 1) = w _k (n) + μ (n) · e ′ (n) · x ′ (nk), where 0 ≦ k ≦ M−1 where x ′ (n) Is the output of the bandpass filter BPF3, and e '(n) is the output of the bandpass filter BPF2.

To stabilize the adaptive system, the phase response of BPF2 and BPF3 must be as close as possible. BPF2 for simplicity and better adaptation
And BPF3, two identical high-pass filters are selected. Unlike what was required for the bandpass filter of Structure 1, as long as the stopband attenuation was sufficient, B
The passband ripple of PF1, BPF2, and PBF3 is not important for good performance. In our experience, a 30 dB stopband attenuation is adequate. Therefore, such 2 ^nd or 3 ^nd elliptic IIR filter of the order of, by using the low order IIR filter to this application, it is possible to reduce the complexity of hardware and computation.

In contrast to the embodiment shown in FIG. 6, the delays of the various groups in FIG. 7 must satisfy the following conditions:

(Equation 8) Optimum delay delta ₁ of the sample can be in a hearing aid mounting process will be acquired by measuring the feedback path. In order to use all ADF coefficients, the delay caused by the adaptive filter should be minimized. Delay Δ ₁ must meet the following conditions.

(Equation 9) Select the next expression.

(Equation 10)

Where ε is the sample tolerance. Usually, ε is chosen equal to two.

This band limited feedback cancellation structure does not introduce any additional delay into the main signal path and does not introduce any additional phase distortion into the output of the hearing aid.

The purpose of the ADF is to estimate the return path. In FIG. 7, BPF1 is used to limit the bandwidth of the feedback cancellation signal. Generally, as shown in FIG. 21, the frequency response of the feedback path has a band-pass characteristic. Therefore, it is possible to approximately match the frequency response of the feedback path using BPF1. In this way, it is possible to mainly use the ADF to track changes in the return path.

The band limited adaptive feedback canceller can be implemented on a platform with a general purpose digital signal processor or a dedicated digital signal processor. Because of the size and power constraints of the hearing aid circuit design, it is desirable to use a fixed point digital signal processor with limited accuracy and word length as the adaptive feedback canceller. Therefore, an efficient digital implementation of a band limited adaptive feedback canceller is critical to the performance of feedback cancellation under the constraints of limited hardware and computational resources. The present invention simplifies computational requirements and addresses issues associated with the effect of limiting accuracy on the adaptation process.

In both structures of the band limited feedback canceller described above, the operation of adaptive filtering is performed in the same manner. For fixed point implementation, some additional modules are needed to maintain the efficiency of erasure under limited accuracy constraints.
The generalized structure of an adaptively filtered band limited feedback canceller is shown in FIG. This generalized structure is applicable to both embodiments described above.
This adaptive filtering configuration has three input ports (x ₁ (n), x ₂ (n),
And e (n)) and one output port (y (n)).

For the feedback canceller structure shown in FIG. 6, x ₁ (n) = x ₂ (n) = x (n).

For the feedback canceller structure shown in FIG. 7, x ₁ (n) = x (n) x ₂ (n) = x ′ (n) e (n) = e ′ (n).

The generalized structure includes an adjustable FIR filtering module, a power estimation module, a coefficient adaptation module, a DC rejection module, a coefficient bandpass filtering (CBF) module, and a control unit. The purpose and detailed implementation of these modules are described below.

1. Adjustable FIR Filtering An adaptive FIR filter is used to approximate the dynamic feedback path and to determine the input signal x ₁ (n)
Is convolved with the current filter coefficient {wk (n): 0 ≦ k ≦ −1} to generate a feedback cancellation signal. However, the return path response has a very large dynamic range. In a fixed-point implementation, the ADF coefficients must be maximized to match the precision of the words assigned to them in order to take full advantage of the internal precision of fixed-point adaptive FIR filtering. Using the ADF scaling gain (G),
Maximizes the F-factor and provides a wide dynamic range for the feedback cancellation signal. Therefore, the calculation of the output of the M-tap adaptive FIR filter is slightly modified as follows.

[Equation 11]

The scaling gain G is chosen as a power of 2 number 2 ^{L and} can be implemented by a left / right shift. Typically, it is sufficient for L to be in the range [-3,3], which provides a dynamic range from -18dB to 18dB. FIG.
FIG. 4 shows a functional block diagram of adjustable FIR filtering with a power of two scaling gain.

[0048] 2. Coefficient adaptation By including the ADF scaling gain, the calculation for the time step size must be modified as follows.

(Equation 12)

As shown in Equations 10 and 13, p (n) is the combined signal e (n) and x _Two This is the power estimation of (n). In the fixed point implementation, when calculating ρ · p (n-1)
In order to abort the word, the limited cycle prevents p (n) from going to zero,
Therefore, the ME in Equation 20 is not required. For simplicity, ME is set to 0.

To avoid splitting when calculating the time step size, approximate p (n) using a number of powers of 2 and convert C, (1-ρ), G, and M to 2 Choose as the number of powers of. C = 2− ^K (1−ρ) = 2− ^J G = 2 ^L M = 2 ^F Equation 20 can be rewritten as follows.

(Equation 13)

K is a positive integer and controls the adaptation speed. The range of K is usually 7 to 10. Smaller values of K provide faster adaptation speed. J is a positive integer and controls the time constant of the power estimation. Usually, J is selected as 6. L is an integer,
Control ADF scaling gain. As mentioned above, the range is from -3 to 3. L is determined based on feedback measurements during the hearing aid installation process, and
The filter coefficient of the adaptive filter is maximized. F is an integer related to the length of the adaptive filter. At a sampling rate of 16000 Hz, choose F = 5 (32 taps), so that the duration of the adaptive filter impulse is about 2 m
s. This is long enough to cover changes in the group delay of the feedback path in the unstable feedback frequency domain.

Q [] is an abort operation. Q [log ₂ (p (n))] can be implemented by searching for the position index of the most significant bit (MSB) of p (n).

FIG. 8 is a functional block diagram of logarithmic quantization with a base of 2. ξ is a positive quantity expressed in an unsigned binary integer format. The position index of the least significant bit (LSB) is zero. Q (log ₂ (ξ)) returns to the position index of the MSB of ξ. When ξ is 0, Q [log ₂ (ξ)] returns to 0.

Power estimation and coefficient adaptation can be further simplified by approximating the error signal using a power of two number. This quantization of the error signal does not affect the performance of the adaptive filter. Therefore, the power estimation is as follows.

[Equation 14] The coefficient adaptation is as follows.

(Equation 15)

Equation 22 shows that only one multiplication is needed for power estimation. The coefficient adaptation process shown in Equation 23 is a non-multiplicative process and can be implemented with move, negate, and add operations.

FIG. 10 shows a functional block diagram of power estimation, and FIG. 11 shows a functional block diagram of a coefficient adaptation process without multiplication. β (n) is the base 2 logarithmic quantization of the error signal e (n), α (n) is the base 2 logarithmic quantization of the power estimate p (n), and ν (n) is the error This is the sign of the signal e (n).

Since the band limited feedback canceller operates very efficiently and effectively in the unstable frequency domain, the adaptive step size used in the normalized LMS adaptive filter is further reduced to minimize misadjustment noise. Thus, it is possible to improve the sound quality of the hearing aid. In a fixed point implementation, it is possible to simplify the normalized LMS adaptation with a very small step size to a code LMS adaptation. Equation 23 can be rewritten as: w _k (n + 1) = w _k (n) + μsign (e (n) · x ₂ (n−k)), where 0 ≦ k ≦ M−1 Equation 23a In the above equation, μ is constant. For example, when w _k (n) is represented by a 12-bit integer, μ can be chosen equal to one. By doing so, the calculations related to power estimation, MSB search, and shift required for the normalized LMS are eliminated.

[0058] 3. Precision-Limiting Effects of Fixed-Point Implementation In fixed-point implementations of adaptive filtering algorithms, both the input and internal algorithm quantities must be quantized to some limited precision. These quantization errors can accumulate without limit until an overflow occurs,
Results in unacceptable performance. For example, resulting word truncation of the original ADC output or band-limited filtering, light DC offset of e (n) and x ₂ (n) accumulates over time, can increase the DC offset of the adaptive filter coefficients There is. Also, the truncation operation of the adaptive filter coefficient adaptation may cause similar DC accumulation in the adaptive filter coefficient, especially when the signal level is low. further,
If the selected band limited filter BPF1 has excessively low frequency attenuation compared to the feedback path, the low frequency gain of the ADF filter response may build up gradually. In either case, the adaptive filter coefficients can overflow and saturate. The ADC removal module periodically removes the DC from the adaptive filter coefficients.
Included to remove offset. Other bandpass filtering modules are provided for filtering the adaptive filter coefficients to reduce the accumulation of low and high frequency responses in the adaptive filter response. This operation is required only when the filter coefficients are saturated.

The DC removal module performs the following operation to estimate the DC offset of the filter coefficient, and subtracts the estimated DC offset from the ADF filter coefficient.

(Equation 16) w _k (n + 1) = w _k (n) −m (n), 0 ≦ k ≦ M−1 Equation 25

FIG. 12 shows a calculation process of DC removal.

For example, DC removal can be scheduled every 256 samples at a sampling rate of 16000 Hz.

In the CBF module, a zero-delay bandpass filtering operation on the ADF coefficient is performed as follows. w _k (n + 1) = w _k (n) / 2−w _{k + 1} (n) / 2, 0 ≦ k ≦ M-3 Equation 26 w _k (n + 1) = w _k (n) / 2, k = M −2, M−1 Equation 27

FIG. 13 shows a calculation process of band-pass filtering. This simple bandpass filtering operation does not provide a completely flat 0 dB magnitude response over the passband and may introduce a slight audible distortion at the output of the hearing aid.
This operation cannot be applied frequently, and is only activated when any one of the ADF coefficients is saturated.

As mentioned above, adaptive filter coefficient adaptation must be performed with DC removal and coefficient bandpass filtering operations. Since the DC removal and coefficient bandpass filtering operations do not need to be performed frequently, only one of these three operations is scheduled to be performed in each sample period. FIG.
Is a flowchart for scheduling a process. In this manner, the DC removal operation is performed periodically, and the coefficient bandpass filtering operation is activated only when the SAT signal from the coefficient adaptation module is on.

FIG. 15 is a logic diagram of a control unit that generates control signals for a coefficient adaptation module, a DC removal module, and a coefficient bandpass filtering module.

Experimental Results The first test of the present invention was performed by computer simulation. The simulation model was developed at SIMULLINK and built with a dynamic feedback path. KE the dynamic return path with the clipboard slowly approaching and leaving the ear
It was measured on the MAR ear. The subject's various hearing aid responses were used as hearing aid processing for the simulation.

Using the hearing aid response and the dynamic feedback path, an open loop transfer function was obtained to identify unstable feedback frequencies. Using an unstable feedback frequency, a feedback canceller,
Specifically, the bandwidth of the band limiting filter was configured. The test was performed with and without the band-limited adaptive feedback canceller. Maximum stable hearing aid gain was recorded under both conditions. The simulated output of the hearing aid was used for the evaluation of the test subject.

The same test was performed on subjects with real-time prototype digital hearing aids. The open loop transfer function of the hearing aid was determined based on closed loop probe tube measurements. Unstable feedback frequencies were identified from the open-loop transfer function and used to construct a band-limited adaptive feedback canceller. The maximum stable insertion gain was recorded with and without the adaptive canceller.

The results showed that the band limited feedback canceller effectively eliminated the oscillation and quasi-oscillation feedback, increased the stable hearing aid insertion gain by 12-15 dB, and minimized distortion to sound quality.

FIG. 21 shows an example of an open loop transfer function measured for a KEMAR ear. There are three unstable feedback frequencies indicated by crosses. They are 2660 Hz, 3260
Hz, and 3060 Hz. Therefore, the cutoff frequency for the band limited filter is set to 2460 Hz. This is about 200 Hz below the lowest unstable feedback frequency.

Alternative Embodiments The embodiments shown in FIGS. 6 and 7 are two specific examples of band limited adaptive feedback cancellers. However, the invention includes other embodiments having the same function of limiting the adaptive bandwidth and the bandwidth of the cancellation signal to the frequency domain known to include the oscillation frequency.

[0072] For example, in FIG. 6, the delay of delta ₂ sample, can be replaced with two exactly complementary filter. One filter is the same as BPF2, and its frequency transfer function is represented as H ₂ (f). The other is a strictly complementary filter SCF2, which has a complementary frequency transfer function

[Equation 17] Having. Such an embodiment is shown in FIG.

Further, by combining two identical BPFs 2, the hearing aid processing can be divided into two processes as shown in FIG. Using hearing aid process 1, B
Process the signal in the frequency domain covered by PF2. Hearing aid process 2 is used to process the signal generated by the complementary filter SCF2 of BPF2. Since the output bandwidth of hearing aid process 1 is the same as BPF2, the need for BPF1 as shown in FIG. 6 is eliminated. A linear phase FIR filter is selected instead of BPF2 and its strictly complementary filter is selected instead of SCF2 to minimize phase distortion in the main signal path of the hearing aid. For some hearing aid applications, the phase distortion of the main signal path can be tolerated or corrected by other hearing aid processing modules. In this case, the strictly complementary condition can be relaxed to another type of complementary condition, such as power being complementary, and thus using an infinite impulse response (IIR) filter, BPF2 and SCF2
And further reduce hardware and computational complexity.

So far, it has been assumed that analog automatic gain control (AGC) is used for the hearing aid between the hearing aid microphone and the analog-to-digital converter. Since the AGC is part of the feedback path, but not the cancellation signal path, sudden changes in AGC gain due to sharp changes in the hearing aid input signal can degrade cancellation performance. One alternative structure shown in FIG. 18 moves the AGC behind the feedback erase junction, so that the AGC is in both the feedback and erase paths. This requires that the AGC be performed digitally. Another alternative solution shown in FIG. 19 is to add the AGC gain digitally to the feedback cancellation signal so that the feedback cancellation signal is
Track gain changes. This method requires converting the AGC gain to a digital format.

The band limited feedback canceller provides optimal sound quality when the erasure bandwidth is minimum.
As is known, the oscillation frequency of a hearing aid is a function of the hearing aid gain, which is generally controlled by the hearing aid volume control. Therefore, the elimination bandwidth of the feedback canceller can also be controlled by controlling the volume of the hearing aid to achieve optimal sound quality. This can be achieved by storing several sets of filter coefficients for the band-limited filter in the hearing aid. At any given hearing aid volume setting, an appropriate set of filter coefficients is used to cover all oscillating frequencies and provide a filter response with minimal bandwidth. This filter selection process is only necessary when the wearer of the hearing aid changes the setting of the volume control.

Since a higher hearing aid gain setting may require a faster adaptation speed to suppress feedback, a different constant value K is selected to control the adaptation speed for different hearing aid volume control settings. It is also possible.

It is to be understood that the invention described above can be implemented in other specific forms without departing from the spirit or essential characteristics of this disclosure. Therefore, it is to be understood that the invention is not limited by the above illustrative details, but is defined by the appended claims.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a hearing aid capable of implementing the present invention.

FIG. 2 is a diagram schematically showing a transfer function of the hearing aid of FIG. 1;

FIG. 3 is a diagram showing a conventional technique for obtaining an open-loop transfer function of a hearing aid.

FIG. 4 is a functional block diagram of a hearing aid incorporating a general wideband adaptive feedback canceller.

FIG. 5 is a functional block diagram of a hearing aid having an internal noise generator that obtains an open-loop transfer function.

FIG. 6 is a functional block diagram of an embodiment of a hearing aid having a band limited adaptive feedback canceller according to the present invention.

FIG. 7 is a functional block diagram of another embodiment of a hearing aid having a band-limited adaptive feedback canceller according to the present invention.

FIG. 8 illustrates base-2 logarithmic quantization used by the present invention for adaptation of filter coefficients.

FIG. 9 is a functional block diagram illustrating signal processing used to implement an adaptive digital filter.

FIG. 10 is a diagram showing processing used for power estimation.

FIG. 11 is a diagram showing processing used for coefficient adaptation.

FIG. 12 is a diagram showing processing used for DC removal.

FIG. 13 is a diagram illustrating processing used for band-pass filtering.

FIG. 14 is a flow chart of functions showing the process scheduling performed by the control unit of FIG. 9;

FIG. 15 is a logic diagram corresponding to the flowchart of FIG.

FIG. 16 is a functional block diagram of an alternative embodiment of the present invention.

FIG. 17 is a functional block diagram of another alternative embodiment of the present invention.

FIG. 18 is a functional block diagram of yet another alternative embodiment of the present invention.

FIG. 19 is a functional block diagram of yet another alternative embodiment of the present invention.

FIG. 20 is a functional block diagram of adjustable FIR filtering with a squared scaling gain.

FIG. 21 illustrates an example of an open loop transfer function and an unstable feedback frequency measured for a KEMAR ear.

──────────────────────────────────────────────────続 き Continuation of front page (81) Designated country EP (AT, BE, CH, CY, DE, DK, ES, FI, FR, GB, GR, IE, IT, LU, MC, NL, PT, SE ), OA (BF, BJ, CF, CG, CI, CM, GA, GN, GW, ML, MR, NE, SN, TD, TG), AP (GH, GM, KE, LS, MW, SD, SL, SZ, TZ, UG, ZW), EA (AM, AZ, BY, KG, KZ, MD, RU, TJ, TM), AE, AL, AM, AT, AU, AZ, BA, BB, BG, BR, BY, CA, CH, CN, CR, CU, CZ, DE, DK, DM, EE, ES, FI, GB, GD, GE, GH, GM, HR, HU, ID , IL, IN, IS, JP, KE, KG, KP, KR, KZ, LC, LK, LR, LS, LT, LU, LV, MD, MG, MK, MN, MW, MX, NO, NZ, (72) PL, PT, RO, RU, SD, SE, SG, SI, SK, SL, TJ, TM, TR, TT, TZ, UA, UG, UZ, VN, YU, ZA, ZW , Shan-Fen United States · 91745 · California · Haschenda Heights · Labrook Drive · 16907 F-term (reference) 5D020 CC06 5J023 DB03 DC01 DC07 5K046 BB00 DD13 EE06 EE57

Claims (56)

[Claims] An adaptive filter, means for combining the output of the adaptive filter with an input of the audio amplifier, an input coupled to the output of the audio amplifier, and an output coupled to the input of the adaptive filter; First having a pass band limited to a frequency band including an unstable frequency
A feedback canceller for an audio amplifying device, comprising: 2. The apparatus of claim 1, wherein the audio amplifier comprises a hearing aid amplifier. 3. The apparatus according to claim 1, wherein the adaptive filter is implemented in a digital signal processor. 4. The apparatus according to claim 3, wherein the adaptive filter is a finite impulse response filter. 5. The apparatus of claim 4, wherein the adaptive filter is implemented with a normalized least squares algorithm. 6. The apparatus of claim 1, wherein the first band limited filter is implemented on a digital signal processor. 7. The apparatus of claim 1, wherein the first band limited filter comprises a high pass filter. 8. The apparatus of claim 7, wherein the passband of the high-pass filter has a cutoff frequency of about 200 Hz below the lowest unstable frequency. 9. The apparatus of claim 1, wherein the adaptive filter comprises means for tapping the input of the adaptive filter with filter coefficients. 10. The apparatus of claim 9, wherein the adaptive filter further comprises means for adding a scaling gain to the filter coefficients. 11. The apparatus of claim 10, wherein the scaling gain is a power of two. 12. scaling gain, from 2 ^-3 in the range of 2 ³ claim 1
An apparatus according to claim 1. 13. An input coupled to an input of the audio amplifier, a second band limited filter having an output, an input coupled to an output of the first band limited filter, and a third band limited filter having an output. Wherein the second and third band limited filters have a pass band limited to a frequency band including an unstable frequency, and the filter coefficient is a function of the outputs of the second and third band limited filters. An apparatus according to claim 9. 14. The apparatus of claim 13, wherein the second and third band limited filters have matching phase responses. 15. The apparatus of claim 13, wherein the second and third band limited filters have substantially identical characteristics. 16. The apparatus of claim 1, further comprising means for delaying the output of the audio amplifier and thereafter inputting the delayed signal to a first band limited filter. 17. The apparatus of claim 16, wherein the output of the audio amplifier is delayed by an amount of time that is a function of the delay of the feedback path. 18. The apparatus of claim 1, further comprising an automatic gain control (AGC) coupled between the combining means and an input of the audio amplifier. 19. An automatic gain control (AGC) coupled to an input of an audio amplifier.
2. The apparatus of claim 1, further comprising: a) wherein the AGC has an output combined with an output of an adaptive filter. 20. The apparatus according to claim 1, wherein the adaptive filter has an adjustable adaptive speed. 21. The apparatus according to claim 20, wherein the adaptation speed is adjusted as a function of the gain of the audio amplifier. 22. The apparatus according to claim 1, wherein the length of the adaptive filter is selectable. 23. The length of the adaptive filter is selected as a function of the sampling rate and the frequency band containing the unstable frequency, so that the duration of the impulse response of the adaptive filter is sufficiently long, 23. The apparatus of claim 22, wherein the feedback path from the output to its input covers an expected range of change in group delay in a frequency band including an unstable frequency. 24. The apparatus of claim 1, wherein the first band limited filter has a frequency response that substantially matches the frequency response of the feedback path from the output of the audio amplifier to its input. 25. A method for canceling feedback in an audio amplifier, comprising: applying an output of the audio amplifier to a first band limited filter having a pass band limited to a frequency band including an unstable frequency. Applying the output of the first band limited filter to the adaptive filter; and combining the output of the adaptive filter with the input of the audio amplifier. 26. An adaptive filter, comprising: estimating a power of a feedback cancellation error signal and an input signal for the adaptive filter; adapting a filter coefficient with the estimated power; and a DC offset from the adapted filter coefficient. 26. The method of claim 25, implemented by a method comprising: removing; band filtering the adapted filter coefficients; and adding the adapted filter coefficients to an adjustable filter. 27. The method of claim 26, wherein removing the DC offset is performed less frequently than adapting the filter coefficients. 28. The step of band filtering the adapted filter coefficients is performed only when the adapted filter coefficients exceed a predetermined threshold.
The method described in. 29. The step of adapting the filter coefficients comprises moving,
27. The method of claim 26, including negation and addition. 30. The method of claim 25, wherein the adaptive filter is implemented by a method including tapping the output of the bandpass filter with filter coefficients. 31. The method of claim 30, wherein the method of implementing an adaptive filter further comprises adding a scaling gain to the filter coefficients. 32. The method of claim 31, wherein the scaling gain is a power of two. 33. The method according to claim ³ , wherein the scaling gain is in the range of ^2-3 to 23.
3. The method according to 2. 34. The method of claim 25, wherein the audio amplifier comprises a hearing aid amplifier. 35. The method of claim 34, further comprising measuring adaptive filter coefficients with a hearing aid inserted in the wearer's ear to identify unstable frequencies. 36. The method according to claim 25, wherein the first band limited filter is a high pass filter. 37. The method of claim 36, wherein the passband of the high-pass filter has a cutoff frequency of about 200 Hz below the lowest unstable frequency. 38. The method of claim 25, further comprising the step of delaying the output of the audio amplifier and then applying the output to a first band limited filter. 39. The method of claim 38, wherein the output of the audio amplifier is delayed by an amount of time that is a function of the feedback path delay. 40. The method of claim 25, further comprising the step of applying automatic gain control (AGC) to the input of the audio amplifier. 41. The method of claim 40, wherein the AGC is applied after the output of the adaptive filter is combined with the input of the audio amplifier. 42. The method of claim 40, wherein AGC is applied to both the input of the audio amplifier and the output of the adaptive filter. 43. The method of claim 25, further comprising adjusting an adaptation speed of the adaptive filter. 44. The method according to claim 43, wherein the adaptation speed is adjusted as a function of the gain of the audio amplifier. 45. The method of claim 25, further comprising selecting a length of the adaptive filter. 46. The length of the adaptive filter is selected as a function of the sampling rate and the frequency band containing the unstable frequency, so that the duration of the impulse response of the adaptive filter is sufficiently long and the output of the audio amplifier 46. The method of claim 45, wherein the feedback path from to the input covers an expected range of group delay variation in a frequency band that includes unstable frequencies. 47. The method of claim 25, further comprising matching the frequency response of the first band limited filter to the frequency response of the feedback path from the output of the audio amplifier to its input. 48. A means for generating a first delay having an output and an input coupled to the audio output of the hearing aid circuit, and a first band limited filter having an output and an input coupled to the output of the first delay means. An adaptive filter having an output and an input coupled to the output of the first band limited filter; means for generating a second delay having an output and an input coupled to a conditioned output of the hearing aid microphone; A first having a non-inverting input coupled to the output of the two delay means, an inverting input coupled to the output of the adaptive filter, and an output coupled to an input of the hearing aid processing module.
A summing node, an input coupled to an input of the second delay means, a second band limited filter having an output, a non-inverting input coupled to an output of the second band limited filter, and coupled to an output of the adaptive filter. A second summing node having an inverted input and a further output; a first input coupled to the output of the first band limited filter; a second input coupled to the output of the second summing node; Means for selecting a first coefficient having an output to be supplied to an adaptive filter, wherein the first and second band limited filters have a pass band limited to a frequency band including an unstable low frequency. Return canceller. 49. A delay generating means having an output and an input coupled to the audio output of the hearing aid circuit; a first band limited filter having an output and an input coupled to the output of the delay means; An adaptive filter having an input coupled to the output of the band-limited filter; a non-inverting input coupled to the conditioned output of the hearing aid microphone; an inverting input coupled to the output of the adaptive filter; and an input of the hearing aid processing module. A summing node having an output coupled to the summing node; an input coupled to the output of the summing node; a second bandlimited filter having an output; and an input coupled to the output and the output of the first bandlimited filter. A first input coupled to the output of the second band-limited filter; a second input coupled to the output of the third band-limited filter; Means for selecting a filter coefficient having an output for supplying a filter coefficient to the adaptive filter, wherein the first, second, and third band-limited filters have a pass band limited to a frequency band including an unstable frequency. A feedback canceller for an audio amplifying device having a. 50. The apparatus of claim 49, wherein the second and third band limited filters have matching phase responses. 51. The apparatus of claim 49, wherein the second and third band limited filters have substantially identical characteristics. 52. The apparatus of claim 1, further comprising means for storing a plurality of sets of filter coefficients for the first band limited filter. 53. The apparatus according to claim 52, wherein the set of filter coefficients is selected as a function of the gain of the audio amplifier. 54. The method of claim 25, further comprising storing a plurality of sets of filter coefficients for the first band limited filter. 55. The method of claim 54, further comprising selecting a set of filter coefficients as a function of a gain of the audio amplifier. 56. The method according to claim 26, wherein the step of band filtering the adapted filter coefficients adds a pass band limited to a frequency band including an unstable frequency.