WO2007063467A2

WO2007063467A2 - Noise reduction system and method

Info

Publication number: WO2007063467A2
Application number: PCT/IB2006/054425
Authority: WO
Inventors: Cornelis P. Janse; Rene M. M. Derkx; Marie-Bernadette Gennotte; Guy G. Edou
Original assignee: Koninklijke Philips Electronics N.V.
Priority date: 2005-11-30
Filing date: 2006-11-24
Publication date: 2007-06-07
Also published as: WO2007063467A3

Abstract

A noise cancellation system and method to reduce background noise, whereby; a filter stage comprising a reference microphone for detecting a noise signal and a filter, a noise cancellation stage comprising an acoustic path (12) connected to the output of the filter stage, and an electrical path (11) connected to the output of the filter stage, wherein the acoustic path comprises first amplifier connected to a loudspeaker (16) and an error microphone (18) arranged to detect an anti noise signal from the loudspeaker, and the electrical path comprises an second amplifier (13) connected to a model (15) of the acoustic path, and output of the acoustic path is connected to the output of the error microphone; the error microphone is arranged to provide signal dependant on the anti noise signal from the loudspeaker and on the ambient noise at the microphone position and the model is arranged to provide a signal indicative of the anti noise signal at error microphone position, wherein the model signal and the error microphone signal are combined to provide a signal for updating the filter.

Description

Noise reduction system and method

The present invention relates to a system and method of reducing noise. More specifically it relates to a system and method of adaptive filtering to minimize an error signal between the output of the adaptive filter and a speaker-microphone arrangement which utilizes an adaptation of a filtered X - algorithm for reducing ambient noise using active noise cancellation.

Active noise cancellation (ANC) is based on the known physical principal of destructive interference of acoustic waves. In known ANC systems and methods, reference sensors are placed in the vicinity of noise sources, for example in the engine compartment of a car. A loudspeaker uses a filtered version of the reference sensor signal to produce an anti- noise signal, where the noise cancellation loud speaker is positioned in the vicinity of the person perceiving the noise.

The filter is adjusted in such a way that the anti-noise signal is identical in amplitude, but opposite phase when compared to the ambient noise signal, so as to destructively interfere with the ambient noise signal so as to reduce or cancel it completely. In order to produce the correct amplitude and phase of anti-noise signal an error microphone is used to measure the attenuated noise signal due to the anti-noise signal emitted by the speaker source and the signal is fed back to the filter to adjust the filter accordingly. A gradient-descent algorithm is then applied so that the filter coefficients approach the optimal filter solution, that is, where the error-microphone signal is minimized or converges to a zero signal.

In general, with adaptive filtering one defines a reference signal, a desired signal and an error signal. The reference signal can be the input of the adaptive filter with the output of the filter an estimate of the desired signal. The error signal is the difference between the desired signal and the filter output. The error signal and the reference input signal are fed to the gradient-descent algorithm to minimize the error signal.

Generally, for ANC applications the filter output scheme corresponds to the sound produced by the loudspeaker at the position of the microphone, which is equal to the inverted desired signal. The desired signal is the noise signal at the error microphone when no ANC is active. When the ANC is active, the measured response at the microphone is equal to the error signal.

Generally speaking, the process of adaptive filtering involves the use of a cost function, which is a criterion for optimum performance of the filter for example, minimizing the noise component of a microphone signal, to feed an algorithm, which determines how to modify the filter coefficients and therefore minimize the cost function on the next and subsequent iterations.

Adaptive filters are used in ANC applications because they can accommodate changes over time in the relation between the reference sensor input and the ambient noise signal at the error microphone. This relationship is represented by the adaptive filter, such that filtering reference signals allows estimation of the noise at the error microphone, assuming the ANC is not active.

S.J. Elliot et al, "Active Noise Control", IEEE Signal Processing Magazine, pp. 12-35, Oct.1993 discusses a general ANC scheme known as feed-forward active noise control, which is depicted in Fig.l. As can be seen from Fig.l this scheme deploys a single error-microphone and M references fed to an adaptive ANC filter. The adaptive ANC filter is in turn connected to an anti-noise loudspeaker.

The reference signals can be derived from any appropriate source such as microphones, accelerometers or tachometers. When microphones are used a feedback cancellation algorithm is required in order to remove the feedback from the loudspeaker to the reference microphone. The implementation of the algorithm as depicted in Fig.l is capable of canceling noise at a single point in space and within a particular frequency band. The area in space around this point where noise is substantially cancelled out is known as the "sweet spot". The size of the sweet spot depends on the frequency of the noise that is cancelled and it is well known that the sweet spot corresponds to approximately 1/10^th of the wavelength of the signal to be cancelled. In order to increase the sweet spot it is possible to deploy further error microphones and loudspeakers. For ease of analysis however, the following discussion describes a situation utilizing a single loudspeaker and error microphone.

Referring to Fig.l, it is assumed that the number of reference microphones X_m_ i is equal to the number of noise sources so that it is possible represent the microphone signal x, by the following expression: x= h_nx * n Eqn. 1 where h_nx is a matrix element representing the primary acoustic (or mechanical) paths from the noise sources to the reference sensors x. The microphone signals x are filtered electrically by the adaptive filter, such that the filtered signals y can be described as: y = w^τ _anc * x Eqn. 2 where w^τ _anc are the ANC filters for the noise microphone signals x, and the filtered signal y as computed in Eqn 2, is reproduced acoustically by the loudspeaker. The error microphone picks up a signal e, containing a mixture of signals: e = y * h_ye + h_neT * n Eqn. 3 where h_neT represents the primary acoustic paths from the noise sources n, to the error microphone and where h_ye represents the secondary acoustic path from the loudspeaker to the error microphone.

Using Eqns 1 ,2 and 3 it is possible to determine the optimum ANC filter coefficients by setting e = 0. As shown by: w_opt = -1 [h_neT * (h_ye)^"1 * (h_n^JT Eqn. 4 Knowledge of the acoustic path may not be known due to, for example movement of passengers in the car or changes in temperature. Therefore the error microphone signal can be used to update the coefficients of the adaptive filter to achieve accurate noise cancellation.

To obtain a stable adaptation of the filter for an arbitrary secondary acoustic path hye, an adaptive solution is proposed in D. R. Morgan, "An analysis of multiple correlation cancellation loops with a filter in the auxiliary path", IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 28 no.4, pp.454 -467, Aug.1980, to incorporate a priori knowledge of the secondary path. This solution is shown in Fig. 2 and is known as the filtered-X adaptive filter. Referring to Fig.2, the noise microphone signals x are filtered with a secondary acoustic path estimate h_ye, as follows:

The result, x' of this filter operation is used to update the multi-channel (MC- AF) adaptive filter. Alternatively a single channel mono-adaptive filter can be used in a situation where a single reference signal is used. For the filtered-X algorithm, it is not necessary that the number of reference sensors is equal to the number of noise sources, as was assumed when deriving Eqn. 4 and in general, the secondary path h_ye, should be observed continuously. One possible approach to implement this is to add artificial random noise to y, as discussed in L.J. Eriksson et al, "Use of random noise for on-line transducer modeling in an adaptive attenuation system", J. Acoust. Soc. Am., Vol. 85, no. 2, pp.797- 802, Feb. 1989, where the amount of artificial noise is adjusted in such a way that it remains unperceived by the listener. By using a single channel adaptive filter technique, the estimate of hye is obtained by decorrelating e with the artificial noise added to y. In practice however, problems with the arrangements described above can arise when the filtered-X algorithm has not converged to the optimal filter solution so that complete destructive interference of noise does not occur. In such a case noise can still be perceived and noise amplification may even occur.

For example, such a situation can occur in automobile applications when it is desirable to reduce engine noise in the car by means of, for example, a tachometer reference signal. In such a situation, the RPM of the engine can change frequently and rapidly and the filtered-X algorithm is required to change accordingly, so as to converge with the optimal solution for a specific situation. However, the signal at the error microphone can be larger when compared to a regime where ANC is not used. One possible solution to this problem is to disable the ANC algorithm by switching off the loudspeaker. However, this can cause the filtered - X filter coefficients to continuously increase due to the fact that the feedback loop in the adaptation path is broken.

A further known method and system for providing active noise control is disclosed in US-A-2004/0037431, for example. This disclosure addresses the problem that when an estimate of a physical path C is not equivalent the real physical path C, the output of an error microphone is not equal to a desired signal βN, where N is the noise signal at the error microphone in the case where no ANC is present. Here active noise control is achieved by modifying a spectral shaping path to prevent unbounded growth in errors inherent in the system. A model of the physical path C within the spectral shaping path is biased to encourage the model to overestimate the characteristics of the physical path so that the error between the model and the actual physical path converges to zero. Additionally, the gain in the spectral shaping path is normalized so that the gain decreases as the output signal of the system increases. This gain normalization drives the output to the correct value and the remainder of the algorithm used for noise control is unaffected. However, the solution as provided by US-A-2004/0037431 can prove ineffective in situations when the generated noise signal is not equal in magnitude and in anti-phase with the noise to be cancelled. Situations can arise with this arrangement, dependant on how the C-model is estimated such that if the coefficients of the adaptive filter model that is used to estimate the C-model start from zero, the C-model underestimates the physical path C, that is C is not equal to C, causing ΔC to be negative. In such a situation and when β is large , the system can therefore become unstable causing a large error in the noise signal, resulting in inadequate noise modeling . US-A-2004/0037431 attempts to address this by generating an overestimate of the actual physical path and assumes that the filter coefficients of the filtered-X algorithm are correct. Although such a solution is effective when β is large it does not address the problem of having incorrect filter coefficients for the fϊltered-X adaptive filter.

The present invention seeks to provide for a noise reduction method and system having advantages over known such methods and systems.

According to one aspect of the invention there is provided a noise cancellation system arranged to reduce background noise, comprising; first detecting means arranged to detect a noise reference signal, filtering means arranged to filter the noise reference signal, transducer means arranged to provide a noise cancellation signal, a second detecting means arranged to detect an error signal, and means to enable at least one of an acoustic path and an estimation of the acoustic path so as to produce an updated error signal, wherein the updated error signal is further arranged to update filter coefficients of the said filter

Preferably, means for comparing the error signal with an estimation of the ambient noise signal is provided, wherein the result of the comparison is arranged to further enable the acoustic path or to enable the estimation of the acoustic path so as to produce an updated error signal, wherein the updated error signal is further arranged to update the filter coefficients.

Preferably still, the filter is connected to a bridge stage, the bridge stage comprising the acoustic path and the estimation of the acoustic path, wherein the acoustic path includes a first amplifier arranged to enable said acoustic path and said estimation of the acoustic path includes a second amplifier arranged to enable the estimation of the acoustic path.

In particular the present invention seeks to minimize an error signal utilizing adaptation of a filtered X - algorithm for reducing ambient noise using active noise cancellation. By allowing an acoustic path of a loudspeaker and microphone to be disabled adaptation of the filtered X - algorithm can continue using an estimation, implemented in hardware, of the acoustic path. This prevents the need to halt adaptation of the filtered X - filter and thereby prevent the filtered X - filter coefficients from increasing continuously which can result in unstable filter operation, whilst also allowing filter coefficients to converge to the correct solution for a particular situation.

Preferably, the first amplifier has a gain function of 1-A and the second amplifier has a gain function of A.

Advantageously this allows the acoustic and electric paths to be enabled or disabled accordingly. This allows the system to disable the acoustic path and enable the electric path so that the noise cancellation can be disabled whilst allowing the filter coefficients to be updated based on the estimation of the acoustic path. This prevents the filter coefficients diverging from the correct value when the acoustic path is disabled.

According to a further aspect of the invention there is provided a method of noise cancellation for reducing background noise, comprising the steps of; detecting a noise reference signal to be cancelled, filtering the noise reference signal and producing a noise cancellation signal by means of a transducer detecting an error signal, and enabling at least one of an acoustic path and an estimation of the acoustic path so as to produce an updated error signal , whereby the updated error signal updates filter coefficients of the said filter. Preferably, the error signal is compared with an estimation of an ambient noise signal and the result of the comparison is also employed in enabling at least one of the acoustic path or the estimation of the acoustic path to produce an updated error signal.

Preferably still, enabling the acoustic path comprises the step of enabling a loudspeaker a microphone and enabling the electric path comprises enabling an estimation of the acoustic path.

The invention is described further hereinafter, by way of example only, with reference to the accompanying drawings, in which:

Fig. 1 shows an ANC arrangement according to the prior art; Fig. 2 shows filtered-X adaptive filter according to the prior art;

Fig. 3 illustrates a block diagram ANC arrangement according to a first embodiment of the present invention;

Fig. 4 illustrates a block diagram ANC arrangement according to a second embodiment of the present invention; and Fig.5 illustrates a block diagram of a fϊltered-X adaptive filter and ANC implementation according to the present invention.

In overview the present invention provides active noise cancellation system and method implementing a filtered-X algorithm so as to reduce environmental noise by means of an anti-noise signal. Where the anti-noise signal has opposite phase and identical amplitude to the environmental noise. The noise cancellation circuit can be disabled without influencing the adaptation of the algorithm. An error microphone signal is dependent on an anti-noise signal and in turn dependent on the actual environmental noise to be cancelled. The model of the acoustic path is arranged to provide a signal indicative of the anti-noise signal from a loud speaker.

Typically, the present invention can be applied to active noise cancellation in automotive applications, such as reducing road noise and engine noise in the cabin of a vehicle, whereby the invention can be employed in an "in- vehicle" entertainment system. However, the invention is not restricted to automotive applications. Active reduction of noise and vibrations in medical applications such as Magnetic Resonance Imaging (MRI) is also contemplated and more generally, the invention can be applied to elimination or suppression of low frequency noise in the range 50 to 150 Hz by approximately 1OdB.

Fig.3 illustrates a schematic block diagram of a bridge scheme according to an embodiment of the present invention. The bridge scheme 10 can include an electrical path 11 connected in parallel to an acoustic path 12. The acoustic path can include in series, a first amplifier 14 and a loud speaker 16, and a microphone 18 in spaced opposition to the loudspeaker 16. The electrical path 11 can include in series a second amplifier 13 and an estimation h_ye 15 of the acoustic path between the microphone 18 and the loudspeaker 16. Both the electrical and acoustic paths include a common summation node 20.

The amplifiers 13,14, loud-speaker 16 and microphone 18 implementing the algorithm of the present invention can be any appropriate components as understood by those skilled in the art. The estimation h_ye 15 of the acoustic path can be any appropriate estimation or updateable model implemented in hardware or software, as understood by those skilled in the art. The amplifiers 13, 14 have gain factors A and A-I respectively. Amplifier 13 is a unity gain type amplifier whereby setting the A=I causes the output signal of the amplifier to follow the input signal. Amplifier 14 is an inverting amplifier whereby setting A=I causes the gain to be zero. In this way, it can be seen that setting the value of A to 1 or 0 allows the circuit to be switched between operating the electrical path and the acoustical path respectively.

Indeed, whilst the use of hardwired electronic circuits are contemplated for implementing the present invention, it is also possible to implement the invention using computer hardware and further still it is possible to implement the invention in software using digital signal processing (DSP) software.

In operation, the loud speaker can be switched off by setting the first amplifier 14 coefficient A equal to 1 and as a result no anti-noise signal is generated acoustically by the loud speaker 16.

The error microphone signal ea is dependent on the anti-noise signal emitted by the loudspeaker and on the actual environmental noise. The model of the acoustic path is arranged to provide a signal indicative of the anti-noise signal emitted by the loudspeaker. In the situation where the model of the acoustic path is exactly equivalent to the physical path then there is no change in the error signal e whether A=O or A=I, that is whether the acoustic or the electric path is enabled. To continue the adaptation of the filtered-X algorithm as shown in Fig. 2, the acoustic path is simulated in the electrical path 11, as if the loud speaker 16 had not been switched off. The resultant electrically simulated noise signal -fie is added to the signal e_a generated by the microphone 18 at the summation point 20. The resultant signal e is approximately equal to the microphone signal e_a when A is equal to 0. The resultant signal e is used in the update loop of the filtered-X algorithm. Since the estimation of the acoustic path hye is available from the filtered-X algorithm it is therefore not necessary to carryout additional estimations or provide further hardware or software to implement the filtered-X algorithm, according to the present invention. The estimate of h_ye is normally implemented in software using digital signal processing (DSP). The adaptive filter W (ref), is adapted using a least mean square algorithm as understood by those skilled in the art. The error signal e is the difference between output of the adaptive filter model h_ye and the output of the error microphone is described as: e = e_a - fie Eqn 6 On the basis of this difference signal, the coefficients of the adaptive filter W, are updated so as to reduce the error signal. After iterative adjustments of the adaptive filter coefficients W, (also known as filter weightings or taps) in the multi-channel adaptive filter, the microphone ea, and the filter model n_e signals converge. As these respective signals converge the noise cancellation signal will converge with the noise signal so as to cancel the noise signal. In this way it can be seen that the error microphone signal e updates the filter coefficients W, such that as much noise cancellation as possible is obtained.

Where A=O, e is equal to e_a and consists of environmental noise and the anti- noise signal. After convergence the anti-noise signal will ideally be equal but opposite to the environmental noise and e_a will be zero. Where A=I, ea only consists of the environmental noise and after convergence n_e will ideally be equal to e_a and e will converge to zero. Therefore it can be appreciated that noise cancellation can only take place when A=O. An extension of the implementation as shown in Fig. 3 is provided. As with the previous embodiment the output from the multi-channel adaptive filter (MC-AF) is connected to the inputs of a unity gain and inverting amplifiers, 13 and 14 respectively, and the input of an additional estimation h_ye of the acoustic path, as shown in Fig. 4.

This embodiment specifically provides that the value of A can be determined so as to switch between operation of the acoustic path and operation of the electric path. The value of A is set to 1 when the filtered-X filter is not adapted to the optimal solution, and set to 0 when it is well adapted. As discussed an additional estimation h_ye of the acoustic path is included, together with a subtraction node 44. The subtraction point allows the anti-noise signal -ή to be subtracted from noise signal e ; the resulting signal z, representing an estimate of the ambient noise, can be compared with the noise signal e using a comparator, which is arranged to give a Boolean, that is 1 or 0, decision. Based on a difference signal d from the comparator 45, the requisite value of A is determined so that it is possible to set the A at any value intermediate 1 or 0, that is 0 < A < 1, that is for example A=O.95 or A=O.05 which can lead faster convergence of the filtered-X algorithm should the estimated acoustic path and the actual acoustic path not converge.

In order to measure the performance of the filtered-X filter the signal y is filtered with the estimated feed forward path h_ye' estimation of the acoustic path. This allows a further anti-noise signal -n' to be computed. This anti-noise signal is then subtracted from the signal in order to obtain a signal z. The signal z is the noise signal, which would be present at the error microphone when no active noise controller is present. The signal powers P_z and P_e of the signals z and e are then compared. The powers of each signal are computed by a first order recursive network (not shown) and a misadaptation of the filtered-X algorithm is detected when:

Misadaptation, refers to the case where the filter coefficients of the filtered-X adaptive filter have not converged and where the noise signal e is greater in magnitude then the estimated ambient noise ambient noise signal z. If Eqn 7 is true then the bridge value A is set to 1, thereby disabling the acoustic path. In all other cases the, the bridge value is set to 0. Where, P_z > P_e, A =0 the ANC can be said to be working well, that is it provides good noise cancellation, particularly when compared to the case when no ANC is present

When implementing the present invention as shown in Fig. 5, the output y of the multi-channel adaptive filter (MC-AF) 50 as shown in Fig. 2 is connected to the inputs of the amplifiers 13 and 14 as shown in either Fig.3 or Fig. 4. The output, e of the bridge scheme 10 according to the first and second embodiments of present invention can be connected to the multi-channel adaptive filter, previously depicted in Fig. 2, such that the error noise signal e updates the filter coefficients w by an amount Δw and therefore updates the characteristics of the noise signal emitted from the loud speaker. Of course, in situations where a single reference is used, for example where a tachometer signal provides the reference signal, it is possible to utilize a single channel mono-adaptive filter in place of the MC-AF, as can be understood by those skilled in the art.

Additionally, computations can be saved by arranging the algorithm so as to combine the acoustic path estimation convolution and determination of A, using a single estimation h_ye of the acoustic path.

It is possible to utilize a single loudspeaker and error microphone for ANC. However, to increase the sweet spot, as discussed above one or more loudspeakers, and reference microphones can be arranged as appropriate.

In this way, it can be seen that the present invention provides an ANC arrangement which controls amplifier gains so as to so as to prevent the microphone signal from exceeding the noise signal.

Claims

CLAIMS:

1. A noise cancellation system arranged to reduce background noise, comprising: first detecting means arranged to detect a noise reference signal, filtering means arranged to filter the noise reference signal, transducer means arranged to provide a noise cancellation signal, second detecting means arranged to detect an error signal, and means to enable at least one of an acoustic path and an estimation of the acoustic path so as to produce an updated error signal, wherein the updated error signal is further arranged to update filter coefficients of the said filter.

2. The system of Claim 1, further comprising means for comparing the error signal with an estimation of an ambient noise signal, wherein the result of the comparison is also employed in enabling at least one of the acoustic path and the estimation of the acoustic path, so as to produce the updated error signal, wherein the updated error signal is further arranged to update the filter coefficients.

3. The system of Claim 1 or 2, wherein the filter is connected to a bridge stage, the bridge stage comprising the acoustic path and the estimated acoustic path, wherein the acoustic path includes a first amplifier arranged to enable the acoustic path, and the estimation of the acoustic path includes a second amplifier arranged to enable the estimation of the acoustic path.

4. The system of Claim 3, wherein the first amplifier has a gain function of 1-A and the second amplifier has a gain function A.

5. The system of Claim 1 or 2, wherein the acoustic path includes a loudspeaker and a microphone, for emitting and detecting acoustic signals respectively.

6. The system of Claim 1, 2, 3, 4 or 5 wherein the estimation of the acoustic path comprises an electrical path, which includes a model of the acoustic path.

7. The system of Claim 6, wherein the model is implemented in an electronic circuit.

8. The system of Claims 6, wherein the model is implemented in software.

9. A method of noise cancellation for reducing background noise, comprising the steps of: detecting a noise reference signal to be cancelled, filtering the noise reference signal and producing a noise cancellation signal by means of a transducer , detecting an error signal, and enabling at least one of an acoustic path and an estimation of the acoustic path so as to produce an updated error signal , whereby the updated error signal updates filter coefficients of the said filter..

10. The method of Claim 9, whereby the error signal is compared with an estimation of an ambient noise signal and the result of the comparison is also employed in enabling at least one of an acoustic path and an estimation of the acoustic path, to produce the updated error signal.

11. The method of Claim 9 or 10, wherein said acoustic path is enabled using a first amplifier and said estimation of the acoustic path is enabled using a second amplifier.

12. The method of Claim 11, wherein the first amplifier is enabled by setting a gain function A to 1, and said second amplifier is enabled by setting the gain function A to 0.

13. The method of Claim 9,10, 11 or 12, whereby the estimate of the acoustic path comprises an electric path arranged to estimate the acoustic path.

14. The method of Claim 9, 10, 11, 12 or 13, whereby enabling the acoustic path comprises the step of enabling a loudspeaker and a microphone.

15. A computer program comprising program instructions for implementing the method of any one or more of Claims 9 to 14.

16. A computer readable medium carrying the computer program of Claim 15.

17. An algorithm comprising instructions according to the method of any one or more Claims 9 to 14.