(19)
`
`(12)
`
`Europaisches Patentamt
`European Patent Office
`
`Office europeen des brevets peen des brevets
`
`EUROPEAN PATENT S P E C I F I C A T I O N
`
`E P 0 6 1 5 3 4 0 B 1
`
`(45) Date of publication and mention
`of the grant of the patent:
`20.05.1998 Bulletin 1998/21
`
`(21) Application number: 94301499.3
`
`(22) Date of filing: 02.03.1994
`
`(51) mtci e H03H 2 1 / 0 0
`
`(54) Low-delay subband adaptive filter
`Adaptives Subbandfilter mit geringer Verzogerung
`Filtre adaptatif en sous-bandes a faible retard
`
`(84) Designated Contracting States:
`DE FR GB
`
`(30) Priority: 12.03.1993 US 30931
`
`(43) Date of publication of application:
`14.09.1994 Bulletin 1994/37
`
`(73) Proprietor: AT&T Corp.
`New York, NY 10013-2412 (US)
`
`(72) Inventors:
`• Morgan, Dennis B.
`Morristown, New Jersey 07960 (US)
`• Thi, James Chi Huu
`Montgomery, Maryland 20877 (US)
`
`(74) Representative:
`Buckley, Christopher Simon Thirsk et al
`Lucent Technologies (UK) Ltd,
`5 Mornington Road
`Woodford Green, Essex IG8 0TU (GB)
`
`(56) References cited:
`WO-A-88/03341
`
`• 1990 INT SYMP ON CIRCUITS AND SYSTEMS;
`1-3/5/90; NEW ORLEANS (US); P279-282; E.
`HANSLER: ADAPTIVE ECHOCOMPENSATION
`APPLIED TO THE HANDS-FREE TELEPHONE
`PROBLEM
`• PROC OF EUSIPCO-92, 24-27/8/92, BRUSSELS
`(BE); G. EGELMEERS et al: RELATION
`BETWEEN REDUCED DIMENSION TIME AND
`FREQUENCY DOMAIN ADATIVE ALGORITHM;
`P1 065-1 068.
`
`DO
`o
`^ -
`CO
`lO
`CO
`o
`a .
`LU
`
`Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give
`notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in
`a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art.
`99(1) European Patent Convention).
`
`Printed by Jouve, 75001 PARIS (FR)
`
`LGE EXHIBIT NO. 1018
`
`- 1 -
`
`Amazon v. Jawbone
`U.S. Patent 8,467,543
`Amazon Ex. 1018
`
`

`

`1
`
`EP 0 615 340 B1
`
`2
`
`Description
`
`Field of the Invention
`
`The present invention relates generally to the field
`of adaptive filtering techniques and specifically to the
`use of such techniques for adaptive noise cancellation.
`
`Background of the Invention
`
`Adaptive filtering techniques are now in widespread
`use for a number of applications such as adaptive ar-
`rays, adaptive line enhancement, adaptive modeling
`and system identification, adaptive equalization, and
`adaptive noise cancellation, including acoustic echo
`cancellation and active noise control.
`In particular, the adaptive noise cancellation prob-
`lem typically involves the generation of a signal which
`reflects an estimate of a disturbance (i.e., noise) which
`is to be reduced or eliminated (i.e., cancelled) from a
`primary source signal. Once determined, this estimate
`signal may then be subtracted from this primary source
`signal to reduce the effect of the disturbance. Active
`noise control in particular involves the generation of a
`secondary signal (e.g., sound) for the purpose of coun-
`teracting the effect of a preexisting noise disturbance.
`Adaptive filtering techniques are advantageously em-
`ployed in the context of adaptive noise cancellation be-
`cause a source signal from which a disturbance has
`been partially removed may be iteratively tested and
`processed to further reduce (e.g., minimize) the pres-
`ence of the disturbance (see e.g. the document WO-A-
`88 03341).
`Certain adaptive filtering applications involve adap-
`tive filter lengths with hundreds of taps. Examples of
`such applications include wideband active noise control
`for complex mechanical structures and acoustic echo
`cancellation, both of which are characterized by long im-
`pulse responses. The computational burden associated
`with these long adaptive filters precludes their use for
`many low-cost applications. In addition to computational
`complexity, adaptive filters with many taps may also suf-
`fer from long convergence times, especially if the refer-
`ence signal spectrum has a large dynamic range.
`A technique that involves the use of subbands has
`been recently exploited to address the above problems
`(see e.g. the document 1 990 Int. Symposium on Circuits
`and Systems; 1-3/5190; New Orleans (US); pp.
`279-282; E. Hansler: "Adaptive Echo Compensation Ap-
`plied to the Hands-Free Telephone Problem"). Process-
`ing the signals in subbands has a twofold advantage.
`First, the computational burden is reduced by approxi-
`mately the number of subbands, since both the tap
`length and weight update rate can be decimated in each
`subband. Second, faster convergence is possible be-
`cause the spectral dynamic range within each subband
`is greatly reduced as compared to the overall spectral
`range.
`
`5
`
`One disadvantage of existing subband adaptive fil-
`ters, however, is that a delay is necessitated by virtue
`of the bandpass filters used to derive subband signals.
`This delay presents a problem for some applications. In
`active noise control applications, for example, delay se-
`riously limits the bandwidth over which good cancella-
`tion can be achieved. For acoustic echo cancellation ap-
`plications, some transmission systems mandate a very
`low signal path delay. Thus, conventional subband
`10 adaptive filtering techniques may be precluded for ap-
`plications requiring low delay.
`
`Summary of the Invention
`
`is
`
`A technique is provided for generating a distur-
`bance estimate signal for use in, for example, adaptive
`noise cancellation. According to an illustrative embodi-
`ment of the invention, a signal reflecting reference infor-
`mation is filtered by a plurality of subband filters to pro-
`20 duce a plurality of subband reference signals. A signal
`reflecting a disturbance is filtered by a corresponding
`plurality of subband filters to produce a plurality of sub-
`band disturbance reflecting signals. Then, a plurality of
`sets of time domain subband weighting coefficients are
`25 generated, each set being derived based on a corre-
`sponding subband reference signal and a correspond-
`ing subband disturbance reflecting signal. Each set of
`time domain subband weighting coefficients is trans-
`formed into a set of frequency domain subband weight-
`ing coefficients. The frequency domain subband weight-
`ing coefficients are combined into a combined set of fre-
`quency domain weighting coefficients. The coefficients
`of this combined set are then transformed back into the
`time domain. This resultant set of combined time do-
`35 main weighting coefficients is then supplied to a pro-
`grammable filter which filters the reference signal ac-
`cordingly in order to produce the disturbance estimate
`signal.
`
`so
`
`40
`40 Brief Description of the Drawing
`
`Fig. 1 shows a block diagram of a low-delay sub-
`band adaptive filter according to a first embodiment of
`the present invention.
`Fig. 2 shows a block diagram of each conventional
`LMS processor of the system of Fig. 1 .
`Fig. 3 shows the frequency response of subband
`filters in accordance with an illustrative example of the
`system of Fig. 1 .
`Fig. 4 shows the process of frequency stacking in
`accordance with the illustrative example of the system
`of Fig. 1.
`Fig. 5 shows the system of Fig. 1 without a cancel-
`lation path filter.
`Fig. 6 shows a block diagram of a low-delay sub-
`band adaptive filter according to a second embodiment
`of the present invention.
`Fig. 7 shows a block diagram of each conventional
`
`45
`
`so
`
`55
`
`2
`
`- 2 -
`
`

`

`3
`
`EP 0 615 340 B1
`
`4
`
`LMS processor of the system of Fig. 6.
`Fig. 8 illustrates the application of active noise con-
`trol techniques to address an acoustic noise problem.
`Fig. 9 shows the system of Fig. 1 as applied in the
`context of the technique of active noise control as illus-
`trated in Fig. 8.
`Fig. 1 0 shows the system of Fig. 1 as applied to the
`problem of acoustic echo cancellation.
`
`Detailed Description
`
`s
`
`10
`
`Fig. 1 shows a block diagram of a low-delay sub-
`band adaptive filter for use in adaptive noise cancella-
`tion according to a first embodiment of the present in-
`vention. A disturbance signal d(t) reflects a disturbance 15
`to be reduced or eliminated. The reference signal x(t) is
`a signal reflecting reference information which is corre-
`lated with the disturbance to be reduced. For example,
`in an acoustic echo cancellation application it may be
`desirable to cancel certain components of a micro- 20
`phone's output signal. These components may be gen-
`erated as a result of the microphone's proximity to a
`loudspeaker. Such an application is presented by a con-
`ventional speakerphone. In such a case, d(t) may rep-
`resent the microphone's output signal (which includes a 25
`disturbance) and x(t) may represent the loudspeaker's
`input signal (upon which the disturbance is based). Note
`that signal d(t) may or may not include that part of a
`source signal which is not part of a disturbance. The em-
`bodiment of the present invention will remove from the 30
`microphone output signal those portions of the signal
`which are correlated with the disturbance. (See the dis-
`cussion of Fig. 2 below.)
`According to the first embodiment, reference signal
`x(t) is filtered by programmable filter 1 2 having a transfer 35
`function W. This transfer function is derived iteratively,
`and is based on a set of MN (i.e., M times N) weights
`(or coefficients), w, supplied by inverse FFT processor
`24. The resulting filtered signal, which reflects an esti-
`mate of the disturbance signal, is filtered in turn by can- 40
`cellation path filter 26 (having a transferfunction C). The
`result is a disturbance estimate signal d(t) Disturbance
`estimate signal d(t) like the signal output from program-
`mable filter 12, reflects an estimate of the disturbance
`to be removed from disturbance signal d(t). Removal of 45
`d(t) from d(t) is accomplished by summation 14.
`Illustratively, cancellation path filter 26 may repre-
`sent an inherent transfer function applied by the envi-
`ronment to the signal generated by programmable filter
`12. In an active noise control application, for example,
`it may be desired to create a "zone of silence" in a par-
`ticular physical location by producing sound from a loud-
`speaker which will cancel the effect of a given preexist-
`ing disturbance. In this case, transfer function C of can-
`cellation path filter 26 may reflect the effect on the loud- 55
`speaker's input signal by the loudspeaker itself, as well
`as the effect of air (or other medium through which the
`sound may travel) between the loudspeaker and the in-
`
`so
`
`tended "zone of silence". In other words, even though
`the active noise control system may generate a given
`signal to be provided to a loudspeaker (i.e., the output
`of programmable filter 1 2), the actual cancellation result
`achieved will be based on the generated signal as "fil-
`tered" by the effects of the loudspeaker and the air.
`Thus, C represents the electroacoustic (or electrome-
`chanical) transfer function from the input of the loud-
`speaker to the location of the intended zone of silence.
`In other cases, such as in the acoustic echo cancellation
`application described above, there may be no cancella-
`tion path filter 26, since direct access to summation 14
`is available. That is, the output signal of programmable
`filter 1 2 may be directly subtracted from disturbance sig-
`nal d(t) by an electrical implementation of summation
`14.
`Summation 14 removes the disturbance estimate A
`from the disturbance signal by subtracting signal d(t)
`from signal d(t) to produce residual error signal e(t).
`Thus, residual error signal e(t) is based on disturbance
`signal d(t). Since signal d(t) reflects a disturbance, sig-
`nal e(t) is therefore also a signal which reflects a distur-
`bance. The operation of summation 14 may be effectu-
`ated by the physical environment (as in the active noise
`control case) or by an electrical component (as in the
`acoustic echo cancellation case). Note that residual er-
`ror signal e(t) may or may not include a desirable part
`of a source signal which is not part of a disturbance (de-
`pending on whether disturbance signal d(t) does or does
`not include the desirable part of a source signal). How-
`ever, that portion of residual error signal e(t) which is
`part of a disturbance represents the actual residual dis-
`turbance (residual error) which is to be reduced (mini-
`mized). In other words, the actual residual error is the
`remaining portion of e(t) which is correlated with the dis-
`turbance signal (and thus with reference signal x(t)). The
`operation of the illustrative embodiments involves an it-
`erative adjustment of the coefficients w of programma-
`ble filter 12 such that the mean square residual error is
`reduced.
`The technique used for adapting the coefficients of
`programmable filter 12 employs a conventional modifi-
`cation of the complex LMS (least-mean-squared) proc-
`ess well known in the art. The modification compensates
`for the effect of cancellation path transfer function C by
`initially filtering reference signal x(t) by cancellation path
`estimate filter 28 to produce filtered reference signal y
`(t). Filtered reference signal y(t), like reference signal x
`(t), reflects reference information which is correlated
`with the disturbance. Cancellation path estimate filter 28
`has a transfer function C, which is an estimate of transfer
`function C of cancellation path filter 26. In this manner,
`the choice of coefficients to be applied by programmable
`filter 12 will appropriately compensate for the effect of
`cancellation path filter 26. This technique is commonly
`known in the art as the "filtered-x" LMS (or FXLMS) proc-
`ess. Of course, in the case where there is no cancella-
`tion path filter 26 (e.g., in the acoustic echo cancellation
`
`3
`
`- 3 -
`
`

`

`5
`
`EP 0 615 340 B1
`
`6
`
`case) there will also be no need for cancellation path
`estimate filter 28. Insuchacase, both transfer functions
`A
`C and C may be viewed as equalling one (the identity
`function).
`According to the embodiment of the present inven-
`tion, filtered reference signal y(t) is decomposed into a
`set of subband reference signals y0, y-, ... and yM by the
`application of a set of M + 1 single-sideband bandpass
`filters. These bandpass filters, subband filters 16-0,
`16-1 ... and 16-M (hereafter subband filters 16-m) have
`transfer functions H0, H-,, ... and HM, respectively. Sim-
`ilarly, residual error signal e(t) is decomposed into a set
`of subband residual error signals e0, e-,, ... and eM by
`the application of a corresponding set of M + 1 single-
`sideband bandpass filters. These bandpass filters, sub-
`band filters 18-0, 18-1, ... and 18-M (hereafter subband
`filters 18-m), also have (identical) transfer functions H0,
`H-|,...and HM, respectively. These subband filters span
`the frequency range from zero to the sampling rate. In
`each subband, both the subband reference signals and
`the subband residual error signals are appropriately
`decimated (i.e., downsampled) by subband filters 16-m
`and 18-m, respectively, to reflect the reduced frequency
`range. Such downsampling is conventional in subband
`processing techniques, illustratively, M may be 32.
`Once signals y(t) and e(t) have been decomposed
`into sets of corresponding subband signals, a set of N
`adaptive weights (i.e., filter coefficients) is computed for
`each subband individually. As described above, a con-
`ventional complex LMS process is used, implemented
`by LMS processors 20-0, 20-1, ...and 20-M (hereafter
`LMS processors 20-m), respectively. In particular, LMS
`processor 20-0 generates a set w0 comprising N coef-
`ficients, LMS processor 20-1 generates a set w-| com-
`prising N coefficients,... and LMS processor 20-M gen-
`erates a set wM comprising N coefficients. The detailed
`function of each LMS processor is illustrated in Fig. 2,
`and discussed below. Illustratively, N may be 32.
`After each LMS processor 20-m has generated
`adaptive weights w0, w-|,...wM (hereafter wm), respec-
`tively, for each subband, these sets of coefficients are
`then transformed into the frequency domain by FFT
`processors 22-0, 22-1,... and22-M (hereafter FFT proc-
`essor 22-m), respectively. Specifically, each FFT proc-
`essor 22-m produces a set of N frequency domain co-
`efficients. These FFT processors may be implemented
`by conventional fast fourier transformation techniques.
`Next, the M + 1 sets of frequency domain coeffi-
`cients are appropriately stacked and inverse trans-
`formed by inverse FFT processor 24 to obtain the (time
`domain) filter coefficients w for programmable filter 12
`(i.e., the coefficients of the transfer function W). Inverse
`FFT processor 24 may be implemented by a conven-
`tional inverse fast fourier transformation technique. An
`illustrative example of frequency stacking is illustrated
`in Fig. 7 and described below. Note that because the
`(wideband) filter coefficients are real, only half of the
`sampling band is actually processed, corresponding to
`
`5
`
`the positive frequency components of the wideband fil-
`ter response. The other half of the response is formed
`in complex conjugate symmetry.
`The FFTs and the inverse FFT need not be per-
`formed at the decimated sample rate. A substantial re-
`duction in computation can be realized if they are com-
`puted only once every several decimated samples, with
`a corresponding moderate time lag in convergence. It is
`further noted that the wideband filter convolution can be
`10 more efficiently computed either by using a vector co-
`processor or by using orthogonal transform techniques.
`A vector coprocessor is a conventional specialized
`hardware device that is dedicated to fast convolution.
`Alternatively, fast convolution can be realized using con-
`'s ventional orthogonal transform techniques such as the
`FFT. However, some care must be taken to insure that
`no delay is introduced into the signal path. Usually, a
`fast FFT convolution will entail a block delay in through-
`put. However, this can be avoided by splitting the wide-
`20 band filter coefficients into segments of equal length.
`Then processing with the first segment may be imple-
`mented by direct convolution while the remaining seg-
`ments may be processed by fast convolutions in time
`sequence. In this way, the fast convolution part may be
`25 started ahead of time by the number of samples in the
`direct convolution so that the output is available when
`needed. Thus, the total number of computations for the
`wideband convolution may be reduced by approximate-
`ly the number of segments (neglecting the fast part of
`the computation).
`Fig. 2 illustrates the detailed operation of each con-
`ventional LMS processor 20-m. Each LMS processor
`takes as input a corresponding subband reference sig-
`nal ym(t) and a corresponding subband residual error
`35 signal em(t), and produces a set wm of N adaptive
`weights (coefficients) for the given subband. Note that
`except for shift register 30, each of the illustrated devic-
`es in Fig. 2 is to be replicated N times in each LMS proc-
`essor 20-m. Specifically, subband reference signal ym
`(t) is processed by shift register 30, a tap delay line hav-
`ing N taps, to produce a set ym of N values. The complex
`conjugate of each of these values is computed, as indi-
`cated by the asterisk (*) in the illustration, and a product
`of each of these resulting N values and subband resid-
`es ual error signal em(t) is computed by multiplier 32. Final-
`ly, each of the N values which result from this multipli-
`cation is passed through amplifier34 (having gain u.) and
`integrated by adder 36 and delay 38, to produce a cor-
`responding one of the N adaptive weights of set wm.
`so This process ensures that only that part of the subband
`residual error signal which is correlated with the sub-
`band reference signal will be removed by the adaptive
`filtering process of the system of Fig. 1 . Therefore, if dis-
`turbance signal d(t) includes a source signal which is
`55 not part of a disturbance, the technique of the present
`invention will advantageously only remove those por-
`tions of the signal which are, in fact, correlated with the
`disturbance.
`
`30
`
`40
`
`4
`
`- 4 -
`
`

`

`7
`
`EP 0 615 340 B1
`
`8
`
`Illustratively, the low-delay subband adaptive filter
`of Fig. 1 assumes M = 32 and N = 32. A wideband filter
`of MN/2 = 512 taps may then be selected for program-
`mable filter 12 and a 32-subband filter bank may be de-
`signed using a conventional polyphase FFT technique
`with twice-critical sampling (i.e., decimation by a factor
`of 16). Note that the subband filter back actually com-
`prises 33 (M + 1 ) subband filters, where the first and last
`filters reflect two halves of the same subband. Fig. 3 il-
`lustrates the frequency response of the first lvl/2 + 1 =
`1 7 filters, which are the only ones processed due to the
`complex conjugate symmetry in accordance with this il-
`lustrative example. Each subband spans the 512 tap im-
`pulse response using N = 32 taps. Each of the sets wm
`of subband adaptive weights are transformed by a cor-
`responding 32-point FFT processor 22-m to obtain 32
`frequencies per subband. These frequencies are then
`stacked in the manner illustrated in Fig. 4 to form points
`0 to 255 of a 512 point spectrum. In particular, the fre-
`quencies from subband 0 are assigned first, followed by
`those from subband 1 , and so forth. Note that only the
`middle half (16) of the N = 32 frequencies are used for
`the odd numbered subbands, and only the upper and
`lower quarters of the 32 frequencies (i.e., the upper 8
`and the lower 8) are used for the even numbered sub-
`bands. Moreover, these upper and lower quarters of the
`frequencies in the even subbands are transposed be-
`fore assembly. The array is then completed by setting
`point 256 to zero and by using the complex conjugates
`of points 1 to 255 in reverse order to fill in points 257 to
`511. Finally, the 512 point spectrum is then transformed
`by 512-point inverse FFT processor 24 to obtain the
`wideband filter weights w for programmable filter 12.
`Note that the above example is described for illustrative
`purposes only. The technique according to the present
`invention can accommodate an arbitrary number of
`taps, number of subbands, decimation factor, etc., all of
`which may be optimized for a particular application.
`A special case of the system of Fig. 1 results when
`C = 1, that is, when there is no cancellation path filter
`26 (or cancellation path estimate filter 28). In this case,
`the system of Fig. 1 reduces to the low-delay subband
`LMS system illustrated in Fig. 5. This special case con-
`cerns the acoustic echo cancellation problem, where x
`(t) is an electrical line input signal, d(t) is a microphone
`signal, and summation 14 is an electronic subtraction
`circuit that derives the output signal e(t). As with the
`case of Fig. 1 , signal d(t) output from programmable fil-
`ter 12 reflects an estimate of the disturbance signal d(t).
`The low-delay subband systems of Figs. 1 and 5
`can be characterized as closed-loop systems since the
`output residual error signal is fed back to the subband
`residual error filter bank. An alternative embodiment is
`possible for the special case of the low-delay subband
`system of Fig. 5 (where there is no cancellation path
`transfer function). Such an alternative, second embod-
`iment of the present invention may be characterized as
`an open-loop system, and is illustrated by the system
`
`5
`
`10
`
`shown in Fig. 6.
`Specifically, in the open-loop system of Fig. 6, LMS
`processors 20-0, 20-1,... and 20-M are replaced by al-
`ternate LMS processors 40-0, 40-1,... and 40-M (here-
`after 40-m), respectively. In addition to generating a cor-
`responding set wm of N adaptive weights, each of these
`substitute LMS processors also cjenerates^subband dis-
`turbance estimate signals d0, d-,,...and dM (hereafter
`dm), respectively, which are used to derive correspond-
`ing "local" subband residual error signals e0, e-,,...and
`eM (hereafter em), respectively. Each of these local sub-
`band residual error signals is, in turn, supplied back to
`the corresponding LMS processor. Specifically, sub-
`band filters 18-m decompose disturbance signal d(t),
`is which reflects a disturbance, into a plurality of subband
`signals d0, d-,,...anddM (hereafter dm). This is, of course,
`in contrast to the system of Fig. 5, in which subband fil-
`ters 18-m decompose residual error signal e(t) (which
`also reflects a disturbance) into subband residual error
`20 signals em. Then, in the system of Fig. 6, each subband
`summation 42-0, 42-1, ...and 42-M(hereafter 42-m)
`computes the corresponding local subband residual er-
`ror signal em to be supplied to the corresponding LMS
`processor 40-m. Ultimately, programmable filter 12 gen-
`25 erates disturbance estimate signal d(t), which reflects
`an estimate of the disturbance, based on the outputs of
`LMS processors 40-m. Since the (wideband) residual
`error signal e(t) is not fed back to the subband weight
`calculation, the system of Fig. 6 can be characterized
`30 as an open-loop version of the system of Fig. 5.
`Fig. 7 illustrates the detailed operation of each con-
`ventional LMS processor 40-m of the system shown in
`Fig. 4. The operation of LMS processor 40 -m is nearly
`identical to that of LMS processor 20-m as illustrated in
`35 Fig. 2 and described above. The difference is that mul-
`tiplier 44 has been added to LMS processor 40-m to
`generate subband disturbance estimate signal dm(t) for
`the purpose as described above in connection with the
`operation of the system of Fig. 4.
`The convergence of an open-loop system such as
`the one illustrated in Fig. 4 may be initially quicker than
`that of the closed-loop system of Fig. 3. However, after
`an initial convergence phase, the closed-loop system
`may continue to converge at a faster rate than the open-
`45 bop system. For this reason, it may be advantageous
`to provide a system incorporating both techniques. In
`such an embodiment of the present invention, the open-
`loop technique may be used initially, followed by a
`switch over to the closed-loop technique. For example,
`(electrical) switches may be provided at the inputs and
`the outputs of subband filters 18-m. The input switches
`may be designed to supply either disturbance signal d
`(t) or residual error signal e(t) to each subband filter. The
`output switches may be designed to supply the output
`55 of each subband filter either to a corresponding subband
`summation 42-m or directly to a corresponding LMS
`processor 40-m. In this manner, an embodiment having
`combined convergence characteristics may be ob-
`
`40
`
`so
`
`5
`
`- 5 -
`
`

`

`9
`
`EP 0 615 340 B1
`
`10
`
`s
`
`tained.
`Fig. 8 illustrates the application of active noise con-
`trol to address an acoustic noise problem. Primary dis-
`turbance source 50 generates acoustic noise in an en-
`closed room. Specifically, it is desired to create a "zone
`of silence" around microphone 46 by controlling loud-
`speaker 45 to produce a canceling acoustic signal. Mi-
`crophone 46, therefore, obtains the residual error signal
`e(t) which is to be minimized. The reference signal x(t)
`is derived from microphone 48, which is in close prox- 10
`imity to primary disturbance source 50. Thus, reference
`signal x(t) may be reasonably assumed to be highly cor-
`related with the disturbance which is to be eliminated
`from the intended zone of silence. The acoustic transfer
`functions Hd, Hx, and Care transferfunctions over which 15
`no control exists, as they are inherent in the environment
`(i.e., resulting from the acoustics of the room). These
`transferfunctions are described below in the discussion
`of Fig. 9.
`Fig. 9 shows the system of Fig. 1 as applied to the 20
`problem of active noise control as illustrated in Fig. 8.
`The components of the system contained in (dashed)
`box 56 represent that part of the environment over which
`no direct access or control is available, except for the
`reference signal and residual error "observation" ports 25
`x(t) and e(t), respectively, and the control input (the out-
`put of programmable filter 12) which is to be supplied in
`accordance with the technique of the present invention.
`Specifically, filter 52, having transferfunction Hx, reflects
`the effect of the environment on disturbance source 50 30
`as it is travels through the room and is captured by mi-
`crophone 48 to generate reference signal x(t). Filter 54,
`having transfer function Hd, reflects the effect of the en-
`vironment on disturbance source 50 as it travels