APPARATUS AND METHOD FOR ADAPTIVELY ATTENUATING NOISE OR VIBRATION
FIELD OF THE INVENTION The present invention relates to active noise and vibration attenuators, and more particularly to adaptive apparatus and method for adaptively removing relatively broadband undesired sound or vibration waves. DESCRIPTION OF PRIOR ART
In modern life, acoustic noise is considered as a significant environmental factor that may yield to health and safety problems. For example, noise generated by airplanes, building ventilation systems or high voltage electrical transformers need to be attenuated to prevent disturbing or harmful effect of noise on people. Moreover, vibration produced in equipment such as manually operated tools, manufacturing machines, robots, vehicles, etc. causes further problems related to equipment operation, such as mechanical stability of operator interfaces or working elements.
Conventional techniques to control acoustic noise or vibration transmitting within a system generally consist in providing the system with mechanical devices to damp undesired waves passively. A first type of passive attenuator devices uses absorption properties of a damping material disposed on the transmission path of the noise or vibration within the system. Attenuation performances of such absorption type attenuators are generally significant only for medium and high frequency noise or vibration, such as higher than 500 Hz. A second type of passive attenuators uses a resonating cavity especially designed to create a progressive opposed phase shift having its maximum at a predetermined central frequency. Although such resonating attenuators can be employed to cancel low frequency noise, they are not currently used in practice due to their large dimension and high manufacturing costs.
In the past years, many different active techniques have been proposed for generating a canceling wave for attenuating an undesired acoustic noise or vibration. In United States patent N≤ 2,043,416 (Lueg), a process for silencing sound oscillations is disclosed, which consists in introducing into a closed or open system within which an undesired acoustic nose is propagating, a cancellation acoustic wave in phase opposition with the undesired acoustic wave, to produce attenuation thereof. Thereafter, many improvements over the process taught by Lueg have been proposed, as reported in many following patents and publications, and particularly by Widrow et al. in "^Adaptive noise canceling: Principles and applications ", IEEE Proceedings vol. 63, No. 12. Dec. 1975, PP 192-1716 and by Warnaka in " Active attenuation of noise: The state of the art *' Noise Control Engineering, may 1982, pp. 100-110. It is pointed out that attenuation techniques for acoustic noise can also be employed for vibration control in a mechanical system with proper adaptation. Active undesired wave attenuation techniques of the prior art can be generally divided in two main classes: feedback control technique and feedforward system identification, as illustrated in Figs. IA to IC, which show known acoustic noise attenuators connected to an acoustic system 10. In the attenuator shown in Fig. IA, the feedback control technique is based on a residual noise sensing microphone 12, a loudspeaker 14 and an electronic controller 16. Particularly, in French patent application N≤ 2595498 (Carme et al.) there is disclosed a design of a controller in an off-line basis using various second order analog active filters. In United States patent N≥ 5,140,640 (Graupe et al.), there is proposed the use of a digital compensator in a feedback control loop. In United States patent N≤ 4,473,906 (Warnaka et al), there is disclosed a feedforward
adaptive system identification for noise attenuation in a duct, as generally illustrated in Fig. IB. This technique consists of sensing a reference noise with a first microphone 18 and driving a transversal filter provided in an adaptive filter 20 that, in turn, provides a canceling signal fed to an output transducer or loudspeaker 22. In order to adaptively identify the unknown acoustic transfer function between the first microphone 18 and the loudspeaker 22, a second microphone sensing residual noise is used for varying the transversal filter weights. Warnaka et al. suggest also the use of a parallel compensator against acoustic feedback from the loudspeaker and the first microphone, as well as Davidson et al. in United States patent N≤ 4,025,724 and Chaplin et al in United States patent N≤ 4,122,303. For the same purposes, as illustrated in Fig. IC, Eriksson et al., in a paper entitled ^ The selection and application of an IIR adaptive filter for use in active sound attenuation '' ,IEEE Transactions on acoustics and signal processing, Vol. ASSP-35, N°- 4, April 1987 in United States patent N≥ 4,677,676 and N≥ 4,677,677, teach the use of an adaptive infinite impulse response (IIR) filter instead of adaptive transversal filter and parallel acoustic feedback compensator. Eriksson et al., particularly in patent N2 4,677,676 , suggests also the use of an other parallel adaptive transversal filter for identifying both loudspeaker feedback and acoustic error transfer function paths, providing on-line system identification. In ~" A unified control strategy for the active reduction of sound " Proceedings of the conference on recent advances in active control of sound and vibration, pp. 271-289, published on April 15 1991, Doelman suggests a generalized minimum variance (GMV) algorithm for acoustic noise attenuation, which consists in a combination of feedforward and feedback control
principles. Simulation results have shown that GMV algorithm theoretically provides noise attenuation efficiency higher than prior feedback or feedforward based attenuation techniques. However, computation complexity involved in the GMV algorithm has made this technique not practical for real-time noise or vibration attenuation.
SUMMARY OF INVENTION It is therefore an object of the present invention to provide an apparatus and method for adaptively attenuating noise or vibration that combine feedforward and feedback control techniques.
An other object of the present invention is to provide an apparatus and method for adaptively attenuating noise or using fast convergence algorithms for real-time applications.
According to the above objects, from a broad aspect, the present invention provides an apparatus for adaptively attenuating undesired wave for use in a system having an input for receiving an undesired wave and an output for emitting an output wave. The apparatus comprises an input transducer for sensing the undesired noise at the system input to produce an input signal, and output transducer for introducing a canceling wave into the system to combine with the undesired wave and produce attenuation thereof. The apparatus further comprises an error transducer for sensing combination of the canceling wave and undesired wave to produce an error signal. There is provided an adaptive filter means responsive to the input signal and the error signal to send a canceling signal to the output transducer, the adaptive filter means including first, second and third interconnected adaptive filter sections. The first adaptive filter section implements a wave transfer function characterizing a wave feedforward path of the
system. The second adaptive filter section implements a wave transfer function characterizing a wave feedback path extending from the output transducer to the system input transducer. The third filter section is a predictive filter section implementing a wave transfer function characterizing a feedback error path of the system.
According to a further broad aspect of the present invention, there is provided in a system having an input for receiving an undesired wave and an output for emitting an output wave, a method for adaptively attenuating the undesired wave. The method comprises steps of:(i) sensing the undesired noise at the system input to produce an input signal; (ii) introducing a canceling wave into the system at a first predetermined location to combine with the undesired wave and produce attenuation thereof; (iii) sensing combination of the canceling wave and undesired wave at a second predetermined location to produce an error signal; (iv) providing adaptive filtering in response to the input signal and the error signal to produce a canceling signal generating the canceling wave, the adaptive filtering consisting of (a) applying a wave transfer function characterizing a wave feedback path of the system,
(b) applying a wave transfer function characterizing a wave feedforward path extending from the predetermined location to the system input; and (c) applying a wave transfer function characterizing a feedback error path of the system.
BRIEF DESCRIPTION OF DRAWINGS
Preferred embodiments of the present invention will be now described with reference to the accompanying drawings in which:
Fig. IA illustrates a feedback noise attenuator according to prior art.
Fig. IB illustrates a prior art noise attenuator using feedforward adaptive system identification technique.
Fig. IC illustrates another prior art feedforward noise attenuator using infinite impulse response filter.
Fig. 2 is a block diagram illustrating a first embodiment of a feedforward-feedback noise or vibration attenuating apparatus according to the present invention.
Fig. 3 is a block diagram showing an arrangement for off-line modeling of the system feedback error path for the embodiments as shown in Figs. 2 and 5. Fig. 4 is a block diagram illustrating a second embodiment of an attenuating apparatus in accordance with the present invention which is provided with a noise generator for on-line modeling of the system feedback error path. Fig. 5 is a block diagram illustrating a third embodiment of an attenuating apparatus according to the present invention.
Fig. 6 is a block diagram of an alternate configuration for the predictive feedback filter provided on the attenuating apparatus as shown in Fig. 2, 4 and 5.
Fig. 7 is a diagram showing an alternate modeling of the acoustic system to be employed in accordance with the present invention.
DESCRIPTION OF PREFERRED EMBODIMENTS
Referring now to Figs. 2, a first preferred embodiment of the apparatus according to the invention is illustrated, which uses an Infinite Response Filter (IIR) for wave feedback compensation. Figure 2 shows an adaptive undesired wave attenuating apparatus generally
designated at numeral 31, which is realized according to a first model of the system generally designated at numeral 30, in which transmission paths 32, 34 and 36 respectively associated with wave transfer functions H0 , H, and H2 are shown. In this model, l(z) represents the Z transform of the undesired wave before reaching the system input transducer 38. H0 is the upstream or feed-forward wave transfer function characterizing a wave feedforward path of the system extending between the system input transducer 38 and an output transducer 40. Hj is the wave transfer function characterizing a wave feedback path extending between the output transducer 40 and the system input transducer 38. H, is dependent on directionality of the input and output transducers 38 and 40. Finally H2 defines the wave transfer function of the path extending from the output transducer 40 to an error transducer 42. The system output wave represented by 0(z) , as an error wave consisting of combination of the canceling wave represented by Y(z) and undesired wave l(z) that as to be attenuated, is measured by the error transducer 42. Therefore, system output wave can be represented by the following mathematical relation:
E{z) = H2H0I(z) + H2Y(z) (1)
The adaptive attenuating apparatus 31 further comprises an adaptive filter unit 44 responsive respectively at 46 and 48 to input signal and error signal respectively coming from the input and error transducers 38 and 42, to send a canceling signal to the output transducer 40. The adaptive filter unit 44 includes first, second and third interconnected adaptive filter sections, respectively designated at numerals 50, 52 and 54. The first adaptive filter section 50 implements H0 while the second section 42 implements H, . The third section 54 is
a predictive filter section implementing a wave transfer function characterizing a feedback error path He of the system, which can be represented by the following relation:
He = SF2H2F3 (2)
wherein S represents a transfer function characterizing the output transducer 40, Fλ represents the transfer function of a first prefiltering section 56 receiving the input signal from input transducer 38, F3 represents the transfer function of a second prefiltering section 58 receiving the error signal from the error transducer 42, and F2 represents the transfer function of a converting section 60 coupled to a main output 62 provided on the second adaptive filter section 52, for sending a canceling signal y(n) to the output transducer 40. In the preferred embodiment as shown in Fig. 2, the prefiltering sections 56 and 58 are composed individually of an amplifier, an anti-aliasing filter and an analog to digital converter respectively coupled to main inputs 64 and 65 provided on the filter unit 44, while the converting section 60 comprises a digital to analog converter, a reconstruction filter and an amplifier. Sections 56, 58 and 60 may be implemented using known Butterworth filters of various orders or equivalents. Experience has shown that the filter order is preferably chosen as lower as possible to maximize wave attenuation. In order to compensate for the error path He , a plurality of adaptive filter having transfer function M are provided, which filters are designated at numerals 67, 69 and 71 in the embodiment as shown in Fig. 2. Filters 67, 69 and 71 are replica versions of a filter trained on an off-line basis as will be now described with reference to Fig. 3.
In the arrangement shown in Fig. 3, a random signal generator 65 is used to produce an output random signal V(n) sent through the converting section 60 having transfer function F2, the output transducer 40 having transfer function S , the transmission path 36 characterized by the wave transfer function H2 , the error transducer 42 and the second prefiltering section 58 characterized by transfer function F3 . It is to be understood that if any of the transfer functions associated with the converting section 60, the output transducer 40 or the second prefiltering section 58 is set to be substantially equal to unity, the corresponding resulting transfer He function is simplified accordingly. The output random signal is also sent to the adaptive filter having transfer function M and designated at numeral 66, which could be either a finite or infinite impulse response filter whose coefficients are adaptively trained through an adaptation mixer 68 fed by the output random signal and an error signal e(w) produced at an adder 70 receiving the filter M output signal at an inverting input 72 thereof, and the prefiltering section output signal at a second input 74 provided on the adder 70. The adaptation mixer 68 implements an adaptation algorithm such as lean mean square (LMS) or recursive least mean square (RLMS) algorithms, as well known in the art. For example, in a case where a finite impulse response filter is used, we have the following relation for the LMS algorithm:
nik) = W{k)τXM{k) (3)
wherein nik) is the output signal of the filter M, W(k) being current coefficients vector of the filter M and XM{k) being an input vector according two the following relations:
XM{k) = [xnik),xnik- \),xnik- 2),...xnik - L)]τ ( 5 )
wherein L is dimension of the filter M . It can be shown that updated coefficients vector of the filter M are given by:
wherein JH is an adaptation step size having a scalar value ensuring algorithm convergence and e(k) is the associated error signal. It could be shown that equation (6) is still applicable if an infinite impulse response filter is chosen as filter M, using a recursive least mean square algorithm (RLMS).
Returning to Fig. 2, the resulting filter M, embodying an estimation of the transfer function He , is copied into the filter unit 44 to obtain adaptive filters 67, 69 and 71. The input signal x(n) produced at the first prefiltering section 56 is sent to input 64 connecting to the first filter section 50, which is composed of a finite impulse response filter having transfer function A at 76 and receiving the input signal x(n) at a first input 78 thereof. The filter A is used mainly for forward identification of the wave transfer function H0. Through a second input 79, the coefficients of the filter A at 76 are adapted in function of the M 67 filtered version of the prefiltered input signal x(n) feeding a first input 80 provided on a an adaptation mixer 82 receiving prefiltered error signal e(n) at a second input 81 thereof, which mixer implementing an adaptation algorithm as will be described later in more detail. The filter A output signal ya(n) is sent through output 84 toward a first input 85 of an adder 86 having
an inverting output 88 coupled to a positive input 91 of an adder 90 provided on the second adaptive filter section 52, which includes a finite impulse response filter having a transfer function B at 92, and receiving the output canceling signal y(n) at a first input 94 thereof. The filter B is incorporated mainly to compensate for feedback path H,. Through a second input 93, the coefficients of filter B at 92 are also adapted in function of the M 69 filtered version of the canceling signal y(n) feeding a first input 96 provided on an adaptation mixer 98 receiving prefiltered error signal e(n) at a second input 99 thereof, which mixer implementing an adaptation algorithm as will be described later in more detail. The canceling signal y{n) is also sent to the digital to analog converter and the reconstruction filter of the converting section 60 characterized by transfer function F2. Filter B output at 95 is sent to an inverting input 93 provided on the adder 90, in a feedback loop. In the preferred embodiment as shown in FIG. 2, the second adaptive filter section 52 forms an infinite impulse response filter characterized by transfer function HB according to the following relation:
The canceling wave represented by Y(z) as produced by the output transducer 40 is combined with the undesired wave I{z) within the system 30, the system output wave is measured by the error transducer 42 and the prefiltered error signal e(n) is produced as explained before.
In order to introduce feedback control according to the present invention, the prefiltered error signal e(n) is applied to a third filter section 54 which is a predictive filter section implementing a wave transfer
function characterizing a feedback error path of the system. The filter section 54 includes an adder 100 having an error signal input 102 and an adaptive finite impulse response filter P designated at numeral 104 receiving at a first input 101 a signal xp(ri) from an output 105 provided on the adder 100. Through a second input 103, the coefficients of filter P at 104 are also adapted in function of the M 71 filtered version of the signal xp(n) feeding a first input 106 provided on an adaptation mixer 108 receiving prefiltered error signal e(«) at a second input 110 thereof, which mixer implementing an adaptation algorithm as will be described later in more detail. A second input 112 provided on the adder 100 is coupled in a feedback loop to the output 105 of the filter P at 104, through a filter Mp designated at 113, which is, in this preferred embodiment, a copy of the filter M 66 as earlier explained with reference to Fig. 3. In an alternate design of the predictive filter section as shown in Fig. 6, the transfer function of the filter Mp is chosen as being substantially equal to unity, which is equivalent to directly connect adder output 112 with filter output 105. The resulting transfer function of the predictive filter section 54 is represented by the following relation:
Hp = , P v (8)
(l- PMp)
Output 105 of the predictive filter section 54 is fed to a second input 114 provided on the adder 86. It is pointed out that control signal from the predictive filter section 54 must be fed at the input of the second filter section 52, to provide feed-forward feedback wave attenuation according to the present invention.
We can express the adaptation algorithm for updating coefficients of a resulting weighting vector w(w) for the attenuating apparatus 31 by the following relations:
w(w+1)= W(ri) + 2 μe{n)U(n) (9)
(10) w(w) =[ao(»),-az.-i(n),fti(n),-*it(n),/»i(»),.••-?£,(»)] (11)
wherein w(«+l) is the updated coefficients vector associated with filters A , B and P at 76, 92 and 104 respectively; La , L and Lp are respective dimensions of these filters and μ is an adaptation step size.
Referring now to Fig. 4, a second preferred embodiment of an attenuating apparatus in accordance with the present invention is shown, which is provided with an integrated random signal generator for on-line modeling of the system feedback error path, as opposed to off-line pretraining as explained before with reference to Fig. 3, in which same elements as found in the embodiment shown in Fig. 2 are designated with same numerals. The adaptive undesired wave attenuating apparatus 33 provided with an adaptive filter unit 45 is realized according to the first system model 30 and comprises all elements of the first embodiment as described before with reference to FIG. 2. The attenuating apparatus further comprises a random signal generator 115 producing an output random signal V(n) sent to a third input 116 provided on the adder 86, through the second filter section 52, the converting
section 60 having transfer function F2, the output transducer 40 having transfer function S , the transmission path 36 characterized by the wave transfer function H2 , the error transducer 42 and the second prefiltering section 58 characterized by transfer function F3. As mentioned before, the corresponding resulting transfer He function can be simplified if any of the above cited transfer functions is chosen or estimated to be substantially equal to unity. The output random signal is also sent to the adaptive filter M designated at numeral 118, which could be either a finite or infinite impulse response filter whose coefficients are adaptively trained through an adaptation mixer 120 fed by the output random signal and an residual error signal er(n) produced, after random wave signal cancellation, at an adder 122 receiving the filter M output signal at an inverting input 124 thereof, and the prefiltered output signal e(w) at a second input 126 provided on the adder 122. As mentioned before, the adaptation mixer 68 implements an adaptation algorithm such as lean mean square (LMS) or recursive least mean square (RLMS) algorithms, as well known in the art. The adder 126 also sends in parallel the residual error signal er n) to the second inputs 81, 99 and 110 respectively provided on the mixers 82, 98 and 108. It is pointed out that the filter M adaptation can be realized either in absence or presence of an undesired wave passing through the system 30.
Attenuation performance will now be discussed with reference to the embodiment of FIG. 4. It can be readily shown that the resulting system output wave θ(z) is given by the following relation:
0 z) = I{z)H. Ho - PM) + SF
2H
2V{z)
( 12 )
It can be seen from equation (12) that the attenuating apparatus as shown in FIG. 4 theoretically offers a reduction factor of (l - PM) over a conventional feedforward attenuator not provided with the proposed predictive filter section for achieving feedback control. Turning now to FIG. 5, a third preferred embodiment of an undesired wave attenuating apparatus in accordance with the present invention will now be described, in which same elements as found in the embodiment shown in FIG. 2 are designated with same numerals. The adaptive undesired wave attenuating apparatus 35 is provided with an adaptive filter unit 45 and is realized according to the first system model 30. It is pointed out that equation (1) as defined before is still applicable in this case. As for previously described embodiments, the undesired wave attenuating apparatus 35 further comprises an adaptive filter unit 47 responsive respectively at 46 and 48 to an input signal and an error signal respectively coming from input and error transducers 38 and 42, to send a canceling signal to the output transducer 40. The adaptive filter unit 47 also includes first, second and third interconnected adaptive filter sections, respectively designated at numerals 51, 53 and 54. The first adaptive filter section 51 implements H0 while the second section 42 implements H,. As in previously described embodiments, the third section 54 is a predictive filter section implementing a wave transfer function characterizing a feedback error path He of the system, which can be represented by equation (2) as defined before. As for the embodiment shown in FIG. 2, a random signal generator as
illustrated in FIG. 3 is used to obtain a resulting filter M embodying an estimation of the transfer function He , which is copied into the filter unit 47 to obtain adaptive filters 73, 75 and 71, as described before. The prefiltered input signal x(n) produced at the first prefiltering section 56 is sent to a positive input 130 of an adder 132 provided on the second filter section 53, which is composed of a finite impulse response filter having transfer function B and designated at numeral 134. The filter B is incorporated mainly to compensate for feedback path H,. The output 133 of filter B at 134 is sent to an inverting input 136 provided on the adder 132, which produces at an output 138 thereof a prefiltered input signal xa(n) for the first filter section 51, which is composed of a finite impulse response filter having transfer function A at 140 and receiving the input signal xa(n) at a first input 142 thereof. The filter A is used mainly for forward identification of the wave transfer function H0. Through a second input 144, the coefficients of the filter A at 140 are adapted in function of the M 73 filtered version of the prefiltered input signal xa(n) feeding a first input 146 of an adaptation mixer 148 receiving prefiltered error signal e(«) at a second input 150 thereof, which mixer 148 implementing an adaptation algorithm as will be explained later in more detail. The filter A output signal ya(n) is sent through output 152 toward a first input 85 of an adder 86 having an inverting output 88 coupled to an input 154 provided on the second filter section 53, which output signal ya(n) is sent to a first input 156 provided on the filter B at 134. Through a second input at 158, the coefficients of filter B are also adapted in function of the M 75 filtered version of the canceling signal y{n) feeding a first input 160 provided on a mixer 162 receiving prefiltered error signal e(n) at a second input 164
thereof, which mixer implementing an adaptation algorithm as explained later in more detail. The canceling signal y(n) is also sent to the digital to analog converter and the reconstruction filter of the converting section 60 characterized by transfer function F2. The canceling wave Y(n) produced by the output transducer 40 is combined with the undesired wave X(n) within the system 30, the system output wave is measured by the error transducer 42 and the prefiltered error signal e(n) is produced as explained before.
In order to introduce feedback control according to the present invention, the prefiltered error signal e{n) is applied to a third adaptive filter section 54 which is identical with the filter section as described before with reference to Figs. 2 and 6, and equation (8) as defined before is still applicable. Output 105 of the predictive filter section 54 is fed to a second input 114 provided on the adder 86. It is pointed out that control signal from the predictive filter section 54 must be still fed at the input of the second filter section 52, to provide feed-forward feedback wave attenuation according to the present invention. We can express the adaptation algorithm for updating coefficients of a resulting weighting vector w(w) for the undesired wave attenuating apparatus 35 using equation (9) as defined before in view of the following relations:
(13)
W(π)= [ 0(«),... lB_,(w),b0(w),...bZt_1(«), 7I(w),... ?L (/ϊ)] (14)
wherein w(«+l) is the updated coefficients vector associated with filters A , B and P at 140, 134 and 104 respectively; La , Lt, and Lp are respective dimensions of these filters. Referring now to FIG. 7, an alternate model of the system 30 to be employed in accordance with the present invention is illustrated, in which the feedforward path 37 associated with wave transfer fucntion K0 extends from the system input transducer 38 to the error transducer 42. Transmission paths 39 and 41 are respectively associated with wave transfer functions K, and K2. With this alternate system model, the undesired wave attenuating apparatus can be realized using same configurations as described above with reference to FIGS. 2 to 6.
It is to be understood that the present invention can be applied either to attenuate acoustic noise associated with an acoustic system or to attenuate undesired vibration wave associated with a mechanical system. For the former case, transducers are microphones or loudspeakers are employed as transducers, and in the latter case, vibration sensing and generating devices are used. Wave transfer functions implemented in the adaptive filter will characterize either acoustic or vibration transmission paths, using similar adaptive algorithms.