United States Patent (19)
`Ikeda
`
`54) NOISE CANCELER
`75 Inventor: Shigeji Ikeda, Tokyo, Japan
`73 Assignee: NEC Corporation, Tokyo, Japan
`
`21 Appl. No.: 09/015,525
`22 Filed:
`Jan. 29, 1998
`30
`Foreign Application Priority Data
`Jan. 29, 1997
`JP
`Japan .................................... 9-014.410
`(51) Int. Cl. ................................................ G06F 17/10
`52 U.S. Cl. .............................................................. 708/322
`58 Field of Search ..................................... 708/322-323;
`375/232–233; 381/7112
`
`56)
`
`References Cited
`
`U.S. PATENT DOCUMENTS
`
`... 708/322
`4,649,505 3/1987 Zinser, Jr. et al. ...
`... 708/322
`5,278,780
`1/1994 Eguchi .............
`... 708/322
`5,608,804 3/1997 Hirano ..
`5,644,596
`7/1997 Sih .......................................... 708/322
`FOREIGN PATENT DOCUMENTS
`O 661 832 7/1995 European Pat. Off..
`0 730 262 9/1996 European Pat. Off..
`0 751 619 1/1997 European Pat. Off..
`8-56180 2/1996 Japan.
`OTHER PUBLICATIONS
`Jin-ichi Nagumo, et al., “A Learning Method for System
`Identification', IEEE Transaction. On Automatic Control,
`vol. 12, No. 3, pp. 282-287, Jun. 1967.
`
`USOO5978824A
`Patent Number:
`11
`(45) Date of Patent:
`
`5,978,824
`Nov. 2, 1999
`
`Bernard Widrow, et al., “Adaptive Noise Cancelling: Prin
`ciples and Applications”, Proceedings of the IEEE, pp.
`1692–1716, Dec. 1975.
`
`Primary Examiner Tan V. Mai
`Attorney, Agent, or Firm-Foley & Lardner
`57
`ABSTRACT
`
`A noise canceler of the present invention includes a signal
`to-noise power ratio estimator to which a main Signal and a
`reference Signal are input. The estimator 10 estimates the
`Sinal-to-noise power ratio of the main Signal from the mean
`power of a desired signal contained in the main Siganl and
`a mean power of a noise Signal also contained in the main
`Signal. In addition, the estimator estimates the Signal-to
`noise power ratio of the reference Signal from the mean
`power of a desired signal contained in the reference Signal
`and the mean power of a noise Singal also contained in the
`reference Signal. An adptive filter for estimating the noise
`Signal of the main Signal has its Step size for coefficient
`updating controlled in accordance with the estimated Signal
`to-noise power ratio of the noise Signal. On the other hand,
`an adptive filter for estimating the desired Signal of the
`reference Signal has its Step size for coefficient updating
`controlled in accordance with the estimated Signal-to-noise
`power ratio of the reference Signal. Delay circuits are
`provided for compensating for a delay ascribable to a power
`averaging procedure which a Signal-to-noise power ratio
`estimator executes to calculate the estimated Siganl-to-noise
`power ratioS.
`
`6 Claims, 6 Drawing Sheets
`
`ADAPTIVE
`FER
`
`
`
`
`
`
`
`
`
`SIGNAL-TO-NOISE
`POWER RATIO
`ESTIMATOR
`
`STEP SIZE
`OUTPUT CITCU
`
`
`
`
`
`SNRB(k)
`
`LGE EXHIBIT NO. 1003
`
`- 1 -
`
`Amazon v. Jawbone
`U.S. Patent 8,019,091
`Amazon Ex. 1003
`
`

`

`U.S. Patent
`
`Nov. 2, 1999
`
`Sheet 1 of 6
`
`5,978,824
`
`
`
`Fig.1
`
`ADAPTIVE
`FILTER
`
`SIGNAL-TO-NOISE
`POWER RATIO
`ESTIMATOR
`
`STEP size
`TPUT CTCU
`OUTPU
`
`SNRB(k)
`
`- 2 -
`
`

`

`U.S. Patent
`
`Nov. 2, 1999
`
`Sheet 2 of 6
`
`5,978,824
`
`Fig.2
`
`16
`
`PSA(k)
`
`2O
`
`CIRCUIT
`
`DVIDER
`
`
`
`POWER MEAN
`CIRCUIT
`
`
`
`
`
`
`
`ADAPTIVE
`FILTER
`
`ADAPTIVE
`FILTER
`
`- 3 -
`
`

`

`U.S. Patent
`
`Nov. 2, 1999
`
`Sheet 3 of 6
`
`5,978,824
`
`Fig.3
`
`INPUT SNRA(k) TO MONOTONE
`DECREASING FUNCTION
`
`
`
`
`
`OUTPUT FUNCTION VALUE OUT1(k)
`
`OUT(k)< u 1min
`
`
`
`
`
`
`
`33
`COMPARE OUT1(k)
`& u 1 max & Li 1 min
`
`OUT(k) > u 1max
`
`pu 1 minsOUT1(k)s u 1 max
`34
`
`OUTPUT u1 min
`
`OUTPUT OUT1(k)
`
`OUTPUT u1max
`
`- 4 -
`
`

`

`U.S. Patent
`
`Nov. 2, 1999
`
`Sheet 4 of 6
`
`5,978,824
`
`Fig.4
`
`
`
`INPUT SNRB(k) TO MONOTONE
`DECREASING FUNCTION
`
`OUTPUT FUNCTION VALUE OUT2(k)
`
`OUT2(k)< u2min
`
`COMPARE OUT2(k)
`& u 2max & u 2min
`
`OUT2(k) > u2max
`
`
`
`
`
`
`
`
`
`
`
`1.2mins OUT(k)s u2max
`44
`
`OUTPUT u2min
`
`OUTPUT OUT2(k)
`
`OUTPUT u2max
`
`- 5 -
`
`

`

`U.S. Patent
`
`Nov. 2, 1999
`
`Sheet 5 of 6
`
`5,978,824
`
`Fig.5
`(Prior Art)
`
`ADAPTIVE
`FILTER
`
`
`
`SIGNAL's
`SOURCE '
`
`HB(z)
`'
`
`NOISE
`SOURCE
`
`- 6 -
`
`

`

`U.S. Patent
`
`Nov. 2, 1999
`
`Sheet 6 of 6
`
`5,978,824
`
`Fig.6
`(Prior Art)
`
`XA(z)
`
`O
`
`Ei(z)
`
`FGE)
`
`6
`
`--
`E2(z)
`GD
`vs. F2(z) 5
`APAVE
`O
`
`
`
`ADAPTIVE
`FILTER
`
`- 7 -
`
`

`

`5,978,824
`
`1
`NOISE CANCELER
`
`BACKGROUND OF THE INVENTION
`
`The present invention relates to a noise canceler and,
`more particularly, to a noise canceler for canceling, by use
`of an adaptive filter, a background noise signal introduced
`into a speech signal input via a microphone, a handsetor the
`like.
`
`A backgroundnoise signal introduced into a speech signal
`input via, e g., a microphone or a handset
`is a critical
`problem when it comes to a narrow band speech coder,
`speech recognition device and so forth which compress
`information to a high degree. Noise cancelers for canceling
`such acoustically superposed noise components include a
`biinput noise canceler using an adaptive filter and taught in
`B. Widrowet al. “Adaptive Noise Cancelling: Principles and
`Applications”, PROCEEDINGS OF IEEE, VOL. 63, NO.
`12, DECEMBER 1975, pp. 1692-1716 (Document
`1
`hereinafter).
`The noise canceler taught in Document 1 includes an
`adaptive filter for approximating the impulse response of a
`noise path along which a noise signal input to a microphone
`assigned to a reference signal (reference signal microphone
`hereinafter) to propagate toward a microphoneassignedto a
`main signal (main signal microphone hereinafter). The adap-
`tive filter is capable estimating noise introduced into the
`main signal microphone. The estimated noise signal
`is
`subtracted from a main signal (combination of a desired
`signal and a noise signal) input to the main signal micro-
`phone.
`The filter coefficient of the above adaptive filter is cor-
`rected by determining a correlation between anerror signal
`produced by subtracting the estimated noise signal from the
`main signal and a reference signal derived from the refer-
`ence signal microphone. Typical of an algorithm for such
`coefficient correction, 1.e., a convergence algorithm is “LMS
`algorithm” describe in Document 1 or “LIM (Learning
`Identification Method) algorithm” described in EEE
`TRANSACTIONS ON AUTOMATIC CONTROL, VOL.
`12, NO. 3, 1967, pp. 282-287.
`A conventional noise cancellation principle will be
`described with reference to FIG. 5. As shown, a noise
`canceler includes a main signal microphone 1, a reference
`signal microphone2, an adaptivefilter 3, a subtracter 4, and
`an output terminal 5. A desired signal S(z) spoken by a
`speaker (signal source) is input to the main signal micro-
`phone 1 adjoining the speaker’s mouth by way of a path
`having an acoustic transfer characteristic HA(z); z is
`expressed as:
`
`z=exp(2aj/FS)
`
`Eq. (1)
`
`where FG denotes a sampling frequency.
`Onthe other hand, noise N(z) issuing from a noise source
`is input to the main signal microphone 1 via a path having
`an acoustic transfer characteristic GA(z). At the same time,
`the noise N(z) is input to a reference signal microphone 2
`remote from the speaker by wayof a path having an acoustic
`transfer characteristic GB(z). The adaptive filter 3 estimates,
`based on the main signal XA(z) and reference signal XB(z),
`the acoustic transfer characteristic (noise path) P(z) of an
`acoustic path along which noise output from the noise
`source N(z) and then input to the reference signal micro-
`phone 2 will propagate to the main signal microphone 1
`whenthe desired signal S(z) is not input.
`The acoustic transfer characteristic P(z) to be estimated is
`produced by:
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`P(z)=GA(z/GB(z)
`
`Eq. (2)
`
`The adaptive filter 3 therefore constitutes a filter having a
`transfer characteristic W1(z)
`identical with the transfer
`function P(z) and operates to generate an estimated noise
`signal F1(z) identical with the noise signal contained in the
`main signal. The subtracter 4 subtracts the estimated noise
`signal F1(z) output from the filter 3 from the main signal
`XA(z), thereby producing an output E1l(z). Whenthe desired
`signal S(z) is not input, the output signal E1(z) is expressed
`as:
`
`El(g) = XA(z) - FI)
`
`Eq. (3)
`
`= XA(Z) — WI(@) XB)
`
`= GA@)N(@) — WI@)GBR)N(]@)
`
`= GA(Z)N(z) — 1GA(z)/ GB@)}gb(2N(Z)
`=0
`
`In this manner, the adaptivefilter 3 is capable of estimat-
`ing the acoustic transfer characteristic P(z) by updating the
`coefficient such that the output signal E1(z) is zero when the
`desired signal S(z) is not contained. The output signal E1(z)
`is referred to as an error signal because it is representative
`of an error in the learning operation of the adaptivefilter.
`
`After the convergence of the adaptive filter 3 the output
`signal E1(z) is expressed as:
`
`El(g) = XA(z) - FI)
`
`= XA(Z) — WI(@) XB)
`
`Eq. (4)
`
`= GA@)N(@)+ HA@)S(Z) — WI(@)GBR)N(@) + HB@)S(@)
`
`= GA(Z)N(z) + HA(Z)S(Z) — WI(Z)GB)N(Z) — WI@)HBE)S(Z)
`
`= HA@)S(@) — WI(@)HB()S (2)
`
`= HA(2)S()[1 — {HB@) / HA} WI)
`
`As the Eq. (4) indicates, the output signal E1(z) does not
`contain any noise signal N(z), i.e., noise has been canceled.
`However,
`the problem is that when the reference signal
`microphone 2 contains the desired signal component S(z),
`iLe., when the acoustic transfer characteristic HB(z) from the
`desired signal S(z) to the reference signal microphone 2 is
`not zero, a signal distortion represented by [1-{HB(z)/HA
`(z)}W1(z)] occurs.
`To solve the above problem, an adaptive filter for cor-
`recting the signal distortion contained in the output signal
`S1(z) may be added, as taught in Japanese Patent Laid-Open
`Publication No. 8-56180. FIG. 6 shows a noise canceler
`
`including such an additional adaptive filter. As shown, the
`noise canceler has an adaptive filter 6 for the above correc-
`tion and a subtracter 7 in addition to the structural elements
`
`shown in FIG. 5. When the main signal XA(z) contains the
`desired signal S(z) and if noise is absent is of less than
`certain level, the adaptive filter 6 performs learning such that
`the output E2(x) of the substracter 7 decreases. Assuming
`that the adaptive filter 6 has a transfer characteristic W2(z),
`then the filter 6 performs the above learning based on,e.g.,
`the LIM scheme such that when N(z) is zero or negligible,
`E2(z) has the following value:
`
`- 8 -
`
`

`

`5,978,824
`
`E2(z) = XA () - F2(3)
`
`Eq. (5)
`
`= 0
`
`Therefore, the transfer characteristic W2(z) of the adap
`tive filter 6 is produced by:
`
`The output F2(z) of the adaptive filter 6 derived from the
`learning is expressed as:
`
`As a result, a desired signal HA(Z)S(Z) free from Signal
`distortion is output.
`AS Stated above, the conventional noise canceler updates
`the coefficient of the adaptive filter 3 and learns the acoustic
`characteristic of noise in Sections where the noise Signal
`N(Z) is present and the desired signal component S(Z) is
`absent or negligibly Small. Further, the noise canceler
`updates the coefficient of the adaptive filter 4 and learns a
`Signal distortion correction filter in Sections where the
`desired signal component S(z) is present and the noise
`component N(z) is absent or negligibly Small. It is therefore
`necessary to detect the above Sections where the desired
`Signal component S(Z) is absent (or little) and the Sections
`where the noise signal component N(Z) is absent (or little)
`and to command the adaptive filters to perform leaning in
`Such Sections from the outside.
`However, it is, in many cases, difficult to command the
`adaptive filters to perform learning from the outside in
`accordance with the level of the desired signal and that of the
`noise Signal, depending on the situation in which the noise
`canceler is located. With the conventional noise canceler, a
`Sufficient noise canceling ability and a Sufficient distortion
`correction characteristic are not achievable unless adequate
`learning Sections are indicated to each adaptive filter for the
`learning purpose.
`SUMMARY OF THE INVENTION
`It is therefore an object of the present invention to provide
`a noise canceler capable of achieving a Sufficient noise
`canceling ability and reducing Signal distortion even when
`adequate learning Sections cannot be indicated from the
`outside.
`A noise canceler of the present invention includes a first
`delay circuit for delaying a main Signal containing a desired
`Signal and a noise Signal by a preselected period of time to
`thereby output a delayed main signal. A Second delay circuit
`receives the noise Signal as a reference Signal and delaying
`it by the preselected period of time to thereby output a
`
`1O
`
`15
`
`25
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`4
`delayed reference Signal. A first Subtracter Subtracts a first
`estimated noise Signal from the delayed main Signal to
`thereby generate a first desired Signal output. A Second
`Subtracter Subtracts a first estimated desired Signal from the
`delayed reference Signal to thereby generate a first noise
`Signal output. A first adaptive filter receives the first noise
`Signal output and adaptively estimates a noise Signal con
`tained in the delayed main Signal to thereby output the first
`estimated noise signal. A Second adaptive filter receives the
`first desired signal output and adaptively estimating a
`desired signal contained in the delayed reference Singal to
`thereby output the first estimated desired signal. A signal
`to-noise power ratio estimator receives the main Signal and
`reference Signal and calculates desired Signal power and
`noise Signal power of the main signal and desired signal
`power and noise Signal power of the reference Signal to
`thereby output an estimated value of a power ratio of the
`main Signal to the noise Signal and an estimated value of a
`power ratio of the reference Signal to the noise signal. A Step
`Size output circuit receives the estimated values from the
`Signal-to-noise power ratio estimator to thereby output a first
`and a Second Step Size representative of an amount of
`correction of a filter coefficient of the first adaptive filter and
`an amount of correction of a filter coefficient of the Second
`adaptive filter, respectively.
`
`BRIEF DESCRIPTION OF THE DRAWINGS
`The above and other objects, features and advantages of
`the present invention will become apparent from the fol
`lowing detailed description taken with the accompanying
`drawings in which:
`FIG. 1 is a block diagram Schematically showing a noise
`canceler embodying the present invention;
`FIG. 2 is a block diagram Schematically showing a
`Signal-to-noise power ratio estimator included in the
`embodiment;
`FIGS. 3 and 4 are flowcharts demonstrating the operation
`of a step size output circuit 11 also included in the embodi
`ment,
`FIG. 5 shows the principle of a conventional noise can
`celer, and
`FIG. 6 is a block diagram Schematically showing a
`Specific configuration of a conventional noise canceler.
`
`DESCRIPTION OF THE PREFERRED
`EMBODIMENT
`Referring to FIG. 1 of the drawings, a noise canceler
`embodying the present invention is shown. AS shown, the
`noise canceler includes a first microphone 1 for a main
`Signal, a Second microphone 2 for a reference Signal, an
`output terminal 5, adaptive filters 3 and 6, Subtracters 4 and
`7, delay circuits 8 and 9, a signal-to-noise power ratio
`estimator 10, and a Step size output circuit 11. The operation
`of the adaptive filters 3 and 6 will be described first.
`A main signal XA(z) is delayed by the delay circuit 8 by
`D samples to turn out a delayed main signal XA(z)Z
`where Z denotes a delay by D samples. The signal
`XA(Z)ZP is applied to the subtracter 4. On the other hand,
`a reference signal XB(z) is delayed by the delay circuit 9 by
`D samples to turn out a delayed reference signal XB(z)Z
`and then applied to the subtracter 7. The delay by D samples
`compensates for a delay ascribable to the calculation of a
`Signal-to-noise power ratio to be effected by the Signal-to
`noise power estimator 10, as will be described later specifi
`cally. Because the delayS provided at the main Signal Side of
`
`- 9 -
`
`

`

`5,978,824
`
`S
`the adaptive filter 3 and the reference signal side of the
`adaptive filter 6, respectively, are equal, they have no
`influence on the relation between the main Signal and the
`reference Signal. Therefore, let D be assumed to be Zero
`hereinafter.
`The adaptive filter 3 operates to estimate a noise Signal
`included in the main signal XA(z) while the adaptive filter
`6 operates to estimate a desired signal included in the
`reference signal XB(z). To allow the filter 3 to estimate the
`noise Signal, the desired Signal estimated by the filter 6 is
`subtracted from the reference signal by the subtracter 7, and
`the resulting noise Signal is input to the filter 3. Likewise, the
`noise signal estimated by the filter 3 is subtracted from the
`main Signal, and the resulting desired signal is input to the
`filter 6. For this purpose, the two filters 3 and 6 are
`cross-coupled, as illustrated.
`ASSume that the Subtracters 4 and 7 produce output
`Signals E1(z) and E2(z), respectively, that the adaptive filter
`3 has a transfer characteristic W1(z) and produces an output
`F1(z), and that the adaptive filter 6 has a transfer character
`istic W2(z) and produces an output F2(z). Then, E1(z) and
`E2(z) are expressed as:
`
`By using the desired signal S(Z), noise N(Z) and acoustic
`transfer characteristics HA(Z), HB(z) and GB(z) described
`with reference to FIG. 5, the main signal XA(z) and refer
`
`The above equations give El (Z) and E2(z), as follows:
`
`As a result, the output E1(z) of the subtracter 4 is the desired
`Signal from which noise has been cancelled.
`Now, for the adaptive filter 3 to estimate a noise Signal
`contained in the main signal accurately, it is necessary to
`increase the amount of updating of the filter coefficient when
`the desired signal of the main Signal obstructing the estima
`
`6
`tion is Smaller than the noise Signal to be estimated.
`Conversely, when the desired signal of the main Signal is
`greater than the noise signal, it is necessary to reduce the
`above amount because the Signal obstructing the estimation
`is greater than the noise Signal.
`On the other hand, for the adaptive filter 6 to estimate the
`desired signal of the reference Signal accurately, it is nec
`essary to increase the amount of updating of the filter
`coefficient when the noise signal contained in the reference
`Signal obstructing the estimation is Smaller than the desired
`Signal. Conversely, when the noise signal of the reference
`Signal is greater than the desired signal, it is necessary to
`reduce the above amount because the Signal obstructing the
`estimation is greater than the desired signal.
`The coefficient of each adaptive filter can be controlled to
`meet the above requirement if the Step size of the learning
`algorithm of the filter is controlled, as follows.
`A method of updating the coefficient will be described,
`assuming the LIM Scheme as a learning algorithm and the
`adaptive filter 3 by way of example. ASSume that the main
`Signal XA(z)is denoted by Xa(k) in time domain, that E2(Z)
`input to the filter 3 is denoted by e2(k) in time domain, that
`F1(z) output from the filter 3 is denoted by f1(k) in time
`domain, and that E1(z) output from the subtracter 4 is
`denoted by e1(k) in time domain; k is an index representa
`tive of time.
`Assuming that the j-th coefficient of the filter 3 at a time
`k is w1j(k), then an estimated noise signal f1(k) output from
`the filter 3 is expressed as:
`
`N-
`f1(k) = X colick) e2(k-j)
`
`Eq. (18)
`
`15
`
`25
`
`35
`
`where N denotes the number of taps of the filter 3.
`A coefficient w1j(k+1) at a time (k+1) is produced on the
`basis of an error Signal e1(k) determined by the Subtracter 4:
`
`40
`
`pu 1(k), e1(k). e2(k - i)
`coli(k+1) = coli(k) +
`
`Eq. (19)
`
`45
`
`50
`
`55
`
`60
`
`65
`
`where u1(k) is the Step size for updating the coefficient of the
`filter 3.
`A greater step size u1(k) promotes rapid convergence
`because the coefficient is corrected by a greater amount.
`However, when components obstructing the updating of the
`coefficient are present, the greater amount of updating is
`noticeably influenced by Such components and increases the
`residual error. Conversely, a Smaller Stepwise u1(k) reduces
`the influence of the above obstructing components and
`therefore the residual error although it increases the con
`verging time. It follows that a trade-off exists between the
`“converging time” and the “residual error” in the Setting of
`the Step size.
`Likewise, as for the filter 6, assume that the reference
`signal XB(z)is denoted by xb(k) in time domain, that E1(z)
`input to the filter 6 is denoted by e1(k) in time domain, that
`F2(z) output from the filter 6 is denoted by f2(k) in time
`domain, and that E2(z) output from the substracter 7 is
`denoted by e2(k) in time domain. Then, an estimated noise
`Signal f2(k) output from the filter 6 is expressed as:
`
`- 10 -
`
`

`

`5,978,824
`
`7
`
`N-1
`P= DY) wrth) elk =p
`JO
`
`Eq. (20)
`
`A coefficient w2j(k+1) at the time (k+1) is produced on
`the basis of an error signal e2(k) determined by the sub-
`tracter 7:
`
`w2jk + 1) = w2j(k)+
`
`H2(kK)-e2(k)-el(k -f)
`
`Eq. (21)
`
`y e2(k —m)*
`m=0
`
`where #2(k)is the step size for updating the coefficient of the
`filter 6.
`As stated above, the coefficient can be variably controlled
`by controlling the step size of the adaptivefilter.
`The operation of the signal-to-noise powerratio estimator
`10 will be described hereinafter. As shown in FIG. 2, the
`estimator 10 is made up of adaptive filters 12 and 13
`subtracters 14 and 15, power mean circuits 16, 17, 18 and
`19, and dividers 20 and 21. The adaptive filters 12 and 13
`and subtracters 14 and 15 are cross-coupled in exactly the
`same manneras in FIG. 1. The difference is that step sizes
`#3 and ywassigned to the adaptive filters 12 and 13,
`respectively, each is fixed and great enough to promote
`convergence. For example, when the LIM schemeis used,
`the step sizes u3 and “4 are selected to be between about 0.2
`and about 0.5. Suchrelatively great step sizes promote rapid
`convergence although they will increase the residual error.
`Assume that
`the adaptive filters 12 and 13 both are
`converged. Then,
`the filter 12 produces an output {3(k)
`whichis the noise signal contained in the main signal. The
`subtracter 14 produces an output e3(k) which is the desired
`signal also contained in the main signal. The power mean
`circuit 16 squares the output e3(k) of the subtracter 14 so as
`to determine its time mean and thereby outputs desired
`signal power PSA(k) particular to the main signal. The
`power mean circuit 17 squares the output £3(k) ofthe filter
`12 so as to determine its time mean and thereby outputs
`noise signal power PNA(k) particular to the main signal.
`The otherfilter 13 produces an output £4(k) which is the
`desired signal contained in the reference signal. The sub-
`tracter 15 produces an output e4(k) whichis the noise signal
`also contained in the reference signal. The power mean
`circuit 19 squares the output e4(k) of the subtracter 15 so as
`to determineits time mean and thereby outputs noise signal
`power PNB(k) particular to the reference signal. Likewise,
`the power meancircuit squares the output f4(k) of the filter
`13 so as to determine its time mean and thereby outputs
`desired signal power PSB(k) particular to the reference
`signal.
`The divider 20 divides the desired signal power PSA(k)
`output from the power meancircuit 16 by the noise signal
`power PNA(K) output from the power mean circuit 17. As
`a result, an estimated signal-to-noise power ratio SNRA(k)
`of the main signal is output from the divider 20. Likewise,
`the divider 21 divides the desired signal power PSB(K)
`output from the power meancircuit 18 by the noise signal
`power pnb(K) output from the power meancircuit 19. As a
`result, an estimated signal-to-noise power ratio SNRB(k) of
`the reference signal is output from the divider 21.
`When the averaging operation of the power meancircuits
`16-19 is implemented by, e.g.,
`the method of moving
`average, the calculated power mean values involve a delay
`of AAV dependent on the numberof times of averaging with
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`8
`to the actual power variation. The illustrative
`respect
`embodimentincludes the delay circuits 8 and 9, FIG. 1, in
`order to compensate for the above delay AAV.It is therefore
`desirable that the delay Z~? of the delay circuits 8 and 9 be
`equal to AAV.
`With the above configuration, the signal-to-noise power
`ratio estimator 10 implements rapid convergence by provid-
`ing the cross-coupled adaptive filters 12 and 13 with a
`relatively great step size. The estimator 10 outputs, by use of
`the converged signals, the estimated signal-to-noise power
`ratio SNRA(k) of the main signal and the estimated signal-
`to-noise power ratio SNRB(k) of the reference signal.
`Reference will be made to FIGS. 3 and 4 for describing
`the operation of the step size output circuit 11. First, the
`estimated SNRA(k) of the main signal output from the
`signal-to-noise powerratio estimator 10 is input to a mono-
`tone decreasing function (step 31). Assuming that f(Q) is the
`monotonedecreasing function for SNRA(k),then the output
`OUT1(k) of the function is produced by (step 32):
`
`OUT1(=f(SNRA®)
`
`Eq. (22)
`
`By use of the above value OUT1(k), the step size, 41(k)
`of the adaptive filter 3 is calculated as:
`
`#1(k)=clip[OUT1(&), wlmax, 11min]
`
`Eq. (23)
`
`whereclip[a, b, c] is a function for setting the maximum
`value and minimum value and defined as:
`
`clipla, b, c]ka(c Sab)
`
`clip[a, b, c]-b(a>b)
`
`clipla, b, c]kc(a<c)
`
`Eq. (24)
`
`Limiting the step size by use of the maximum value
`imax and minimum value “lminis desirable for the stable
`operation of the adaptivefilter. As for the adaptive filter 3 the
`function value determined by inputting the estimated signal-
`to-noise power ratio SNRA(k) to the monotone decreasing
`function is used as a step size, as stated above. It followsthat
`the step size is reduced whenthe signal-to-noise powerratio
`is great, or it is increased when the ratio is small (steps
`33-36).
`The estimated signal-to-noise powerratio of the reference
`signal is also input to a monotone increasing function (step
`41). Assuming that g() is the monotone decreasing function
`for SNRB (k), then the output OUT2(k) of the function is
`produced by (step 42):
`
`OUT2(k)=g(SNR B(A))
`
`Eq. (25)
`
`Byuse of the above value OUT2(k),the step size “2(k) of
`the adaptive filter 6 is calculated as:
`
`#2(k)=clip[OUT2(&), “2max, “2min]
`
`Eq. (26)
`
`As for the adaptive filter 6, the function value determined
`by inputting the estimated signal-to-noise power ratio SNRB
`(K) to the monotone decreasing function is used as a step
`size, as stated above. It follows that the step size 1s increased
`when the signal-to-noise power ratio is great, or
`it
`is
`decreased when the ratio is small (steps 43-46).
`As described above, the step size output circuit 11 con-
`trols the step size to be fed to the adaptive filter 3 in
`accordance with the estimated signal-to-noise power ratio
`SNRA(k)of the main signal. Also, the circuit 11 controls the
`
`-11-
`
`- 11 -
`
`

`

`5,978,824
`
`step size to be fed to the adaptive filter 6 in accordance with
`the estimated signal-to-noise power ratio SNRB(k) of the
`reference Signal.
`Alternatively, an arrangement may be made Such that the
`Step sizes u1(k) and u2(k) assigned to the adaptive filters 3
`and 6, respectively, are compared, and Smaller one of them
`is Set to be Zero in order to interrupt the learning function of
`the filter whose step size is determined to be zero. This kind
`of control Successfully reduces interference between the two
`filters 3 and 6 and thereby promotes more accurate learning.
`In Summary, it will be seen that the present invention
`provides a noise canceler capable of estimating the noise
`Signal of a main Signal and the noise Signal of a reference
`Signal accurately. The noise canceler therefore insures rapid
`convergence and allows a minimum of Signal distortion to
`occur without resorting to a command customarily input
`from the outside for commanding the learning operation of
`filters. These advantages are derived from a unique configu
`ration in which a relation in size between a desired Signal,
`which is an interference Signal for an adaptive filter used to
`estimate the noise Signal of the main signal from the
`estimated Signal-to-noise power ratio of the main Signal, and
`the noise signal to be canceled is determined. This relation
`is used to control a step Size to be fed to the adaptive filter.
`This is also true with an adaptive filter for estimating the
`desired Signal of the reference signal from the estimated
`Signal-to-noise power ratio of the reference Signal; the noise
`Signal is an interference Signal while the desired signal is a
`Signal to be canceled.
`Various modifications will become possible for those
`skilled in the art after receiving the teachings of the present
`disclosure without departing from the Scope thereof.
`What is claimed is:
`1. A noise canceler comprising:
`first delaying means for delaying a main Signal containing
`a desired Signal and a noise Signal by a preselected
`period of time to thereby output a delayed main Signal;
`Second delaying means for receiving the noise Signal as a
`reference Signal and delaying the reference Singal by
`the preselected period of time to thereby output a
`delayed reference Signal;
`first Subtracting means for Subtracting a first estimated
`noise Signal from Said delayed main Signal to thereby
`generate a first desired Signal output;
`Second Subtracting means for Subtracting a first estimated
`desired signal from Said delayed reference Signal to
`thereby generate a first noise Signal output;
`a first adaptive filter for receiving Said first noise signal
`output and adaptively estimating a noise Signal con
`tained in Said delayed main signal to thereby output
`Said first estimated noise Signal;
`a Second adaptive filter for receiving Said first desired
`Signal output and adaptively estimating a desired Signal
`contained in Said delayed reference Singal to thereby
`output Said first estimated desired signal;
`Signal-to-noise power ratio estimating means for receiv
`ing Said main Signal and Said reference Signal and
`calculating desired signal power and noise Signal power
`of the main Signal and desired signal power and noise
`Signal power of the reference signal to thereby output
`an estimated value of a power ratio of the main Signal
`to the noise Signal and an estimated value of a power
`ratio of the reference Signal to the noise Signal; and
`Step Size outputting means for receiving Said estimated
`values from Said Signal-to-noise power ratio estimating
`means to thereby output a first and a Second Step size
`
`5
`
`15
`
`25
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`10
`representative of an amount of correction of a filter
`coefficient of Said first adaptive filter and an amount of
`correction of a filter coefficient of Said Second adaptive
`filter, respectively.
`2. A noise canceler as claimed in claim 1, wherein Said
`Signal-to-noise power ratio estimating means comprises:
`third Subtracting means for Subtracting a Second estimated
`noise Signal from the main Signal to thereby generate a
`Second desired Signal output;
`fourth Subtracting means for Subtracting a Second esti
`mated desired signal from the reference Signal to
`thereby generate a Second noise signal output;
`a third adaptive filter for receiving Said Second noise
`Signal output and adaptively estimating a noise Signal
`contained in the main Signal to thereby output Said
`Second estimated noise Signal;
`a fourth adaptive filter for receiving Said Second desired
`Signal output and adaptively estimating a desired Signal
`contained in the reference Signal to thereby output a
`Second estimated desired signal;
`first power averaging means for receiving Said Second
`desired signal output and producing a Square mean of
`Said Second desired signal output to thereby output
`desired signal power of the main Signal;
`Second power averaging means for receiving Said Second
`estimated noise Signal and producing a Square mean of
`Said Second estimated noise Signal to thereby output
`noise Signal power of the main Signal;
`third power averaging means for receiving Said Second
`estimated desired Signal and producing a Square mean
`of Said Second estimated desired signal to thereby
`output desired signal power of the reference Signal;
`fourth power averaging means for receiving Said Second
`noise Signal output and producing a Square mean of
`Said Second noise Signal to thereby output noise Signal
`power of the reference Signal;
`first dividing means for dividing Said desired Signal power
`of the main sign