 |
Description  |
|
|
Cross Reference to Related Application Ser. No. 925,060, filed Oct. 30,
1986.
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a noise canceling system, and more
particularly to a noise canceling system which cancels a plurality of
background noises that infiltrate into a voice receiver through different
transmission paths.
2. Description of the Prior Art
The common noise canceling system for removing (canceling) from the output
of the voice receiver noises generated from a plurality of noise sources
and received by the voice receiver is such that the frequency transmission
characteristics such as impulse response and transmission functions of
noise transmission paths from the noise sources to the voice receiver, are
estimated, and the noises are produced via the estimated frequency
transmission characteristics, linearly added up together, and are
subtracted from the output of the voice signal receiver so as to be
canceled.
According to the above-mentioned conventional noise canceling system,
however, the amount of operation becomes essentially very great.
That is, in the above typical noise canceling system, frequency
transmission characteristics of noise transmission paths from noise
sources to a voice receiver are estimated by some means, filters such as
transversal digital filters having transmission functions that offer the
above frequency transmission characteristics are constituted as equivalent
noise-producing filters, and noises generated by the noise sources are
produced via the equivalent noise-producing filters, added up together
linearly, and are subtracted as an equivalent superposed noise of the
plurality of noise sources from the output of the voice receiver so as to
be canceled. Therefore, how efficiently to estimate the coefficients of
transversal filters that constitute an equivalent noise-producing filter,
is very important for preventing the amount of processing from greatly
increasing.
The filter coefficient of such an equivalent noise-producing filter is
estimated as described below. That is, when there exists a single noise
source, the filter coefficient which minimizes the electric power of
noise-canceled residual waves after the output of the transversal filter
is subtracted from the output of the voice receiver, is determined by
widely known methods such as solving an inverse matrix of a row number and
a column number determined by the tap number of the filter or searching
relying upon a maximum inclination method. Where there exist a plurality
of noise sources, the coefficients of a plurality of equivalent
noise-producing filters must be determined by taking the effects among the
noise sources into consideration. Even when there exists only one noise
source, however, the amount of processing and operation becomes
essentially very great. The amount of processing and operation becomes
tremendously great when a plurality of noise sources have to be treated by
giving attention to the effects among the noise sources.
According to another method for estimating the filter coefficient of the
equivalent noise-producing filter, the filter coefficient which minimizes
the electric power of noise-canceled residual waves, is set over a
considerably long period of observation time by forming an automatic
control loop and by effecting the adaptive control. However, since the
observation time is considerably long, the processing response tends to be
considerably delayed even when there exists only one noise source. In
particular, this method exhibits poor follow-up performance for the noise
that changes with time.
SUMMARY OF THE INVENTION
An object of the present invention is, therefore, to provide a noise
canceling system capable of canceling noises generated from a plurality of
noise sources.
Another object of the present invention is to provide a noise canceling
system capable of remarkably reducing the calculation amount for
estimating the filter coefficients.
According to the present invention, under the condition where a plurality
of background noise sources exist, there are arranged a first receiver,
primarily receiving desired voice, and a plurality of second receivers
each primarily receiving noise from a corresponding noise source. Filter
coefficient of equivalent noise-producing filters each having a frequency
transmission characteristics equivalent to that of transmission path from
its corresponding noise source to the first receiver are estimated based
upon mutual-correlation coefficients among the outputs of the first and
second receivers and auto-correlation coefficients of the respective
outputs of the second receivers. The noise signals from the equivalent
noise-producing filters are subtracted from the output of the first
receiver, thereby canceling the background noise. The filter coefficients
may be estimated by using a maximum value of the mutual-correlation
coefficients between the outputs of the first receiver and the respective
second receivers.
Other objects and features will be clarified by the following explanation
with reference to the attached drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a block diagram which illustrates a first embodiment and a second
embodiment of the present invention in combination;
FIG. 2 is a diagram which illustrates a fundamental principle for canceling
the noise according to the embodiment of FIG. 1;
FIG. 3 is a diagram illustrating the cancelation of noise utilizing the
estimated impulse responses of the noise transmission paths;
FIG. 4 is a diagram illustrating the estimation of transfer functions of
the equivalent noise-producing filters according to the embodiments of
FIG. 1;
FIG. 5 is a diagram showing the fundamental method of estimating the
transfer function of the noise transmission path; and
FIG. 6 is a diagram illustrating the efficient estimation of coefficients
of the equivalent noise-producing filter.
PREFERRED EMBODIMENTS OF THE INVENTION
FIG. 1 is a block diagram which explains first and second embodiments
according to the present invention, wherein portions indicated by dotted
lines are blocks that are related to the second embodiment.
The first embodiment shown in FIG. 1 comprises sound receivers of a number
P, i.e., 1-1, 1-2, 1-3, 1-4, - - - , 1-P, a delay circuit 2 formed by
connecting L unit delay elements in cascade, a silence detector 3,
mutual-correlation coefficient calculators 4-12, 4-13, - - - , 4-1P,
auto-correlation coefficient calculators 5-2, 5-3, - - - , 5-P, a
coefficient determining unit 6, equivalent noise-producing filters 7-2,
7-3, 7-4, - - - , 7-P, and adders 8-1, 8-2, 8-3, 8-4, - - - , 8-P.
The sound receiver 1-1 chiefly receives voice signals together with noise
generated from a plurality of noise sources. The receivers 1-2, 1-3, 1-4,
- - - , 1-P of a number (P-1) chiefly trap noises generated from a
plurality (P-1) of noise sources. If the frequency transmission
characteristics such as impulse response characteristics are found for
each of the transmission paths from the plurality of noise sources to the
sound receiver 1-1, the noise produced via the impulse response
characteristics can be subtracted from the ouput of the sound receiver 1-1
during silence to cancel the noise. This is based upon the fact that the
output of the sound receiver 1-1 during silence, i.e., the output of mixed
noise from the plurality of noise sources can be regarded to be equal to
the superposition of linear combinations of the noises.
The impulse response can be easily constituted as a transversal filter
having a transfer function that exhibits the impulse response
characteristics. Even in this embodiment, a desired impulse response is
obtained in the form of a transversal filter.
FIG. 2 is a diagram of a fundamental principle for canceling noise
according to the embodiment of FIG. 1.
A voice signal and an undesired noise signal are superposed and added up
together via an input terminal 100-1, and are supplied to a delay circuit
2.
The delay circuit 2 consists of unit delay elements that are combined in L
stages, and imparts a predetermined time delay to the inputs that are
introduced via an input terminal 100-0. By taking into consideration the
relationships among the sound receiver that sends voice signals inclusive
of noise to the input terminal 100-0 and a group a sound receivers that
send noises to input terminals 100-1 to 100-P (P=2, 3, 4, - - - ), the
delay time is so selected that the addition in an adder 40-1 maintains
nearly the same phase with respect to the same noise.
Equivalent noise-producing filters 30-1 to 30-P have impulse responses
h.sub.1 (t) to h.sub.P (t) of noise transmission paths between each of P
noise sources and the sound receiver that traps voice signals. Noises
generated by P noise sources are received by P equivalent noise-producing
filters, superposed and added up together through adders 40-1, 40-2, - - -
, reversed for their polarities, and are added to the output of the delay
circuit 2 through an adder 40-0. That is, the noises are subtracted from
the output of the delay circuit 2 so as to be canceled. That is, the
fundamental requirement for canceling the noise is how efficiently to
determine the impulse responses h.sub.1 (t) to h.sub.P (t) of the
transmission paths for the noises generated from the noise sources.
Described below in detail is a fundamental method of canceling the noise
utilizing the impulse responses of the noise transmission paths.
FIG. 3 is a diagram explaining the cancelation of noise utilizing the
estimated impulse responses of the noise transmission paths. FIG. 3 shows
the case where the noises are to be canceled from the two noise sources.
Symbols N.sub.1 (Z) and N.sub.2 (Z) denote noises by Z-conversion notation
produced by two noise sources, an adder 12-1 represents a function of the
sound receiver which receives a voice signal S(Z), and adders 12-2 and
12-3 represent functions of sound receivers that chiefly trap noises
N.sub.1 (Z) and N.sub.2 (Z).
To the adder 12-1 are input the voice signal S(Z) as well as undesired
signals consisting of noises N.sub.1 (Z) and N.sub.2 (Z), and transmission
paths 11-1 and 11-2 thereof are denoted by transfer functions H.sub.1 (Z)
and H.sub.2 (Z). An adder 12-2 chiefly receives noise N.sub.1 (Z). To the
adder 12-2 is also input an undesired signal consisting of noise N.sub.2
(Z). Transmission paths 11-3 and 11-4 thereof are denoted by transfer
functions H.sub.3 (Z) and H.sub.4 (Z). Further, an adder 12-3 chiefly
receives noise N.sub.2 (Z) as well as undesired noise N.sub.1 (Z).
Transmission paths 11-6 and 11-5 thereof are denoted by transfer functions
H.sub.6 (Z) and H.sub.5 (Z). If the transfer functions surrounded by a
dotted line are known, there are obtained the following adder outputs:
S(Z)+N.sub.1 (Z)H.sub.1 (Z)+N.sub.2 (Z)H.sub.2 (Z) (1)
N.sub.1 (Z)H.sub.3 (Z)+N.sub.2 (Z)H.sub.4 (Z) (2)
N.sub.1 (Z)H.sub.5 (Z)+N.sub.2 (Z)H.sub.6 (Z) (3)
The above equations (1) to (3) represent outputs of the adders 12-1 to
12-3.
The desired voice signals S(Z) only can be obtained if undesired noise
N.sub.1 (Z)H.sub.1 (Z) input via the transfer function H.sub.1 (Z) and
undesired noise N.sub.2 (Z)H.sub.2 (Z) input via the transfer function
H.sub.2 (Z) are subtracted from the output of the adder 12-1 represented
by the equation (1). Namely, the output of the adder 12-2 represented by
the equation (2) and the output of the adder 12-3 represented by the
equation (3) are converted into N.sub.1 (Z)H.sub.1 (Z) and N.sub.2
(Z)H.sub.2 (Z), respectively, to reverse the signs, and are added to the
output of the adder 12-1 represented by the equation (1). In effect, S(Z)
only is left by the subtraction. The above-mentioned conversion can be
applied to the outputs of the adders 12-2 and 12-3 in various ways. In any
case, the operational method can be fundamentally put into practice by the
combination of folding multiplication of the transfer functions and the
addition as well as subtraction.
In the case of FIG. 3, the output of the adder 12-2 is once supplied to
equivalent noise-producing filters 13 and 14 having transfer functions
H.sub.6 (Z) and H.sub.5 (Z), and the output of the adder 12-3 is supplied
to equivalent noise-producing filters 15 and 16 having transfer functions
H.sub.4 (Z) and H.sub.3 (Z). The output of the equivalent noise-producing
filter 15 is subtracted by a subtracter 19 from the output of the
equivalent noise-producing filter 13, and the output of the equivalent
noise-producing filter 14 is subtracted by a subtracter 20 from the output
of the equivalent noise-producing filter 16. The outputs of these
subtracters are given by the following equations (4) and (5):
N.sub.1 (Z)(H.sub.3 (Z)H.sub.6 (Z)-H.sub.4 (Z)H.sub.5 (Z)) (4)
N.sub.2 (Z)(H.sub.3 (Z)H.sub.6 (Z)-H.sub.4 (Z)H.sub.5 (Z)) (5)
The noises N.sub.1 (Z) and N.sub.2 (Z) converted into the forms of folding
multiplications relative to the transfer functions indicated by common
parentheses, are converted into equivalent noises N.sub.1 (Z)H.sub.1 (Z)
and N.sub.2 (Z)H.sub.2 (Z) through equivalent noise-producing filters 17
and 18 having transfer functions as given by the following equations (6)
and (7):
##EQU1##
An adder 21 obtains the desired output S(Z) from which the noise is erased
by adding up together the outputs of the equivalent noise-producing
filters 17 and 18 while inverting their signs.
By combining the transfer functions H.sub.1 (Z) to H.sub.6 (Z) as described
above, there is produced equivalent noise from which are removed the
effects among the noises. The equivalent noise is then subtracted from the
output of the voice signal receiver to fundamentally cancel the noise.
There can be contrived a variety of other methods to utilize the transfer
functions for canceling noises. What is important is how to use the
transfer functions of the equivalent noise-producing filters in order to
simplify the contents of processing.
Here, the transfer functions H.sub.1 (Z) to H.sub.6 (Z) that will be used
in the aforementioned noise canceling means are all unknown values and
must, hence, be estimated before being used. Further, the above-mentioned
embodiment has dealt with the case where there existed two noise sources.
However, the processing can be effected in the same manner even when there
exist two or more noise sources.
Transfer functions of the noise transmission paths can fundamentally be
estimated as described below. To simplify the description, it is now
presumed that there exists only one noise source.
FIG. 5 is a diagram showing a fundamental method to estimate the transfer
function of a noise transmission path.
The noise generated by a noise source is superposed on and added to the
voice signal in an undesired form. This is depicted by an adder 52. The
output is supplied to a subtracter 53. On the other hand, an equivalent
noise-producing filter 51 is constituted as a transversal filter which
traps the noise generated by the noise source and supplies an output
thereof to the subtracter 53. Under this condition, the output of the
equivalent noise-producing filter 51 is supplied as an argument to the
subtracter 53, and the filter coefficient of the equivalent
noise-producing filter 51 is so selected that the output of the subtracter
53 becomes minimum when the voice signal is zero, i.e., so that the
electric power of the noise-canceled residual waves becomes minimum. Then,
the transfer function H.sub.2 (Z) almost converges into H.sub.1 (Z). As
mentioned earlier, the filter coefficient is estimated by arithmetic
operation such as solving the inverse matrix having row and column numbers
determined by the tap number of the equivalent noise-producing filter 51,
or searching based upon the maximum inclination method, or by the adaptive
control using an automatic control loop which minimizes the electric power
of noise-canceled residual waves. Even when there exists only one noise
source, the amount of operation becomes very great to determine the
transfer function of the transmission path, or the response time becomes
so long that follow-up performance is deteriorated for the noise that
change with the lapse of time. When there exist a plurality of noise
sources, therefore, the amount of operation becomes tremendously great,
and the follow-up performance is inevitably deteriorated greatly.
To solve this problem, there can be contrived an efficient method as
described below. FIG. 6 is a diagram which illustrates the fundamental
processing for efficiently estimating the filter coefficient of the
equivalent noise-producing filter. FIG. 6 deals with the case where there
exists only one noise source.
When the voice signal is silent, a sound receiver 54 receives noise
generated by the noise source in an undesired form. A waveform that is
detected is denoted by S.sub..mu. (t). A sound receiver 55 also receives
noise generated by the noise source. A waveform thereof detected is
denoted by S.sub.n (t). Since S.sub..mu. (t) can be regarded to be a
linear combination of S.sub.n (t), the noise can be canceled by the
subtraction between these two noises.
Here, it is presumed that the filter coefficient of the equivalent
noise-producing filter 59 formed as a transversal filter is set at a tap
position that is delayed by one, and other coefficients are all zero. In
this case, the noise-canceled residual waveform U(t) produced by a
subtracter 60 is given by the following equation (8):
U(t)=S.sub..mu. (t)-aS.sub.n (t-.tau.) (8)
If the number of observation sections is N, and the electric power U(t) of
the equation (8) is E, then E is given by the following equation (9):
##EQU2##
From the equation (9), a coefficient a that minimizes the electric power E
at the tap .tau. is obtained to make the following equation (10) zero,
i.e.,
##EQU3##
That is, the coefficient a is found from the following equation (11):
##EQU4##
A numerator on the right side of the equation (11) represents a
mutual-correlation coefficient .phi.(.tau.) of S.sub..mu. and S.sub.n at
the tap .tau., and the denominator denotes an auto-correlation coefficient
R(o) of S.sub.n at the tap zero. Using these symbols, the equation (11)
can be expressed as the following equation (12):
a=.phi.(.tau.)/R(o) (12)
If the coefficient a is determined, U(t) is determined from the equation
(8). The thus obtained U(t) is regarded to be S.sub..mu. (t), and a filter
coefficient which minimizes the noise-canceled residual waveform is
estimated. The above operation is repeated until the noise-canceled
residual waveform becomes smaller than a predetermined level. This method
of repetitive processing helps greatly reduce the amount of operation
required for estimating the filter coefficient compared with the method
described with reference to FIG. 5. However, the present invention effects
the following processing in order to further reduce the required amount of
operation.
If now a mutual-correlation coefficient between U(t) and S.sub.n (t) is
denoted by .phi..sub.1 (v), then .phi..sub.1 (v) is given by the following
equation (13):
##EQU5##
That is, when there exists only one noise source, a mutual-correlation
coefficient .phi.(v) between S.sub..mu. and S.sub.n at a tap v is once
determined, and is corrected by an auto-correlation coefficient sequence
aR (.tau.-v) which includes a, in order to successively estimate .phi.(v)
for each of maximum values. A filter coefficient is obtained if the
mutual-correlation coefficient .phi..sub.1 (v) is divided by R(o) and is
normalized. The correcting processing is thus effected successively to
easily determine the filter coefficients. A mutual-correlation coefficient
calculator 56, a auto-correlation coefficient calculator 57 and a
coefficient determining unit 58 of FIG. 6 work to offer necessary
coefficients and to determine filter coefficients relying upon the
above-mentioned idea for processing.
In the foregoing was described the case where there was no time delay
between the noise entering into the sound receiver which mainly traps the
voice signals and the noise entering into the sound receiver which mainly
traps the noise. Even when there exists a time difference, however, the
invention can be easily put into practice by imparting a corresponding
time delay to the noise that is in advance.
In the above-mentioned embodiments of FIGS. 5 and 6, there existed only one
noise source. When there exist a plurality of noise sources, however,
effects among noises become a problem, and correction must be effected by
taking this fact into consideration. Described below are the contents of
correction when there are a plurality of, for example, two noise sources
as shown in FIG. 3.
A noise that has entered into the sound receiver which traps voice signals
and is detected, is denoted by S.sub..mu. (t) and noises that are detected
after having entered into the sound receivers that trap noises from the
first and second noise sources are denoted by S.sub.n1 (t) and S.sub.n2
(t), respectively. It is now presumed that a filter coefficient of the
equivalent noise-producing filter of the type of transversal filter has
been determined at a tap .tau. only, the equivalent noise-producing filter
having a transfer function that exhibits an impulse response to a
transmission path that is to be estimated for the second noise source. In
this case, mutual-correlation coefficients that have to be taken into
consideration include S.sub..mu. (t), S.sub.n1 (t) and S.sub.n2 (t) as
well as mutual-correlation coefficients of a combination of S.sub.n1 (t)
and S.sub.n2 (t). The auto-correlation coefficient S.sub.n1 (t) and
S.sub.n2 (t) also affect the system. This is explained below. That is, the
filter coefficient of the equivalent noise-producing filter for the second
noise source has been set only with respect to the tap .tau.. In this
case, a noise-canceled residual waveform U(t) is given by the following
equation (14):
U(t)=S.sub..mu. (t)-aS.sub.n2 (t-.tau.) (14)
If U(t) is regarded to be an input noise of the second time instead of
S.sub..mu. (t), mutual-correlation coefficients .phi..sub.1 (v) and
.phi..sub.2 (v) of the input noise and the two detected noises S.sub.n1,
S.sub.n2 are given by the following equations (15) and (16):
##EQU6##
In the equation (15), .phi..sub.n1 (v) denotes a mutual-correlation
coefficient of S.sub..mu. (t) and S.sub.n1 (t), and .phi..sub.12 (.tau.+v)
denotes a mutual-correlation coefficient of S.sub.n1 (t) and S.sub.n2 (t).
Similarly, .phi..sub.2 (v) is given by the equation (16):
##EQU7##
In the equation (16), .phi..sub.n2 (v) denotes a mutual-correlation
coefficient of S.sub..mu. (t) and S.sub.n2 (t), and R.sub.n2 (.tau.+v)
denotes an auto-correlation coefficient of S.sub.n2 (t).
What is meant by .phi..sub.1 (v) and .phi..sub.2 (v) of the equations (15)
and (16) is that the mutual-correlation coefficient of S.sub..mu. (t) and
S.sub.n1 (t) should be corrected by the mutual-correlation coefficient of
S.sub.n1 (t) and S.sub.n2 (t), and that the mutual-correlation coefficient
of S.sub..mu. (t) and S.sub.n2 (t) can be corrected by the
auto-correlation coefficient of S.sub.n2 (t).
The above-mentioned contents include the case where there are two noise
sources. The same idea can be applied even to a case where there are a
plurality of noise sources as described below.
It can be considered that the filter coefficient that has been determined
in advance of the equivalent noise-producing filter for the second noise
source, is a first and a sole filter coefficient which minimizes the
noise-canceled residual waveform U(t). From a different point of view,
this is a filter coefficient of an equivalent noise-producing filter for
the noise output of a noise receiver that exhibits a maximum correlation
with respect to the noise output of the sound receiver that traps voice
signals. The maximum correlation is denoted by .phi..sub.1P where a
postscript 1 denotes an output noise of the voice signal receiver and a
postscript P denotes an output noise of the noise receiver that exhibits
the maximum correlation.
When U(t) is regarded to be an input, .phi..sub.1P can be corrected by d
and R.sub.p as illustrated in conjunction with the equation (16), and
.phi..sub.1j (j.noteq.P) other than the maximum correlation can be
corrected by .phi..sub.Pj. If now .phi..sub.1P is .phi..sub.13, then
.phi..sub.13 can be corrected by a and R.sub.3 for the next U(t), and
.phi..sub.12 can be corrected by a and .phi..sub.32 as meant by the
contents of the equations (15) and (16). In this case, the coefficient a
can be found from the aforementioned equation (12). Namely, the
coefficient a is that of a filter for a noise which produces a maximum
correlation, and is obtained by retrieving a maximum mutual correlation
coefficient .phi..sub.1P and normalizing it with the self-correlation
coefficient R.sub.P (o).
In effect, a maximum mutual-correlation coefficient is corrected by an
auto-correlation coefficient sequence of noise that produces the maximum
value, and the sequence of mutual-correlation coefficients that are not
the maximum value is corrected by the consequence of mutual-correlation
coefficients corresponding to noise that exhibit the maximum value. The
above processing is cyclically repeated until the level of the
noise-canceled residual waves becomes smaller than a predetermined level,
thereby to estimate the filter coefficients. Thus, the filter coefficients
can be estimated while greatly reducing the amounts of operation.
In the cyclical processing, the coefficient of the same tap of the
equivalent noise-producing filter may often be subjected to the estimation
processing a plural number of times. This, however, presents no problem,
and the plural number of the coefficients thus obtained should simply be
added up together.
FIG. 4 is a diagram for explaining the estimation of transfer functions of
the equivalent noise-producing filters in the embodiment of FIG. 1.
The equivalent noise-producing filters 23 and 24 are constituted as
transversal filters having transfer functions given by the equations (17)
and (18). In the case of the equivalent noise-producing filters of FIG. 3,
the filter coefficients are estimated based upon a prerequisite that the
transfer functions H.sub.1 (Z) to H.sub.6 (Z) of noise transmission paths
are all determined. In the case of this embodiment, however, the filter
coefficients of the equivalent noise-producing filters 23 and 24 are
determined by retrieving a maximum mutual-correlation coefficient of noise
output during silence of the sound receiver which chiefly receives voice
signals and noise outputs of a plurality of sound receivers which chiefly
receive noises generated from a plurality of noise sources, by so setting
the filter coefficient of a transversal filter that it exhibits an impulse
response which equivalently expresses the maximum mutual-correlation
coefficient, by successively correcting the maximum mutual-correlation
coefficient and other mutual-correlation coefficients by the
above-mentioned means, and cyclically repeating the processing a required
number of times.
Transfer functions of the equivalent noise-producing filters 23 and 24 are
given by the following equations (17) and (18),
##EQU8##
If outputs of the adders 12-2 and 12-3 are added up together through the
adder 21 via transfer functions given by the equations (17) and (18),
there is obtained an output N.sub.1 (Z)H.sub.1 (Z)+N.sub.2 (Z)H.sub.2 (Z)
which is free from the effect caused by the interference among the noises.
If this output is added with its signs reversed to the output of the adder
12-1 through the adder 22, the noise component can be canceled The
principal object of the embodiment of FIG. 1 is to set the coefficient of
the transversal filter having such a transfer function by the
above-mentioned correction estimated means.
Reverting to FIG. 1, the embodiment will be described below.
The sound receiver 1-1 chiefly receives voice signals together with
undesired noise.
The noise receivers 1-2 to 1-P chiefly trap noses generating by noise
sources of a number (P-1).
The delay circuit compensates the time differences of noise inputs that
stem from the arrangements of the sound receiver 1-1 and the sound
receivers 1-2 to 1-P. Therefore, the delay circuit 2 has been set in
advance by taking into consideration the arrangement and the mode of
operation.
The silence detector 3 detects the silent condition of voice signals input
to the sound receiver 1-1, and sends the data to the coefficient
determining unit 6.
The mutual-correlation coefficient calculators 4-12, 4-13, - - - , 4-1P
calculate mutual-correlation coefficient sequences .phi..sub.12,
.phi..sub.13, - - - , .phi..sub.1P between the noise output of the sound
receiver 1-1 during silence and each of the noise outputs of the sound
receivers 1-2 to 1-P.
The auto-correlation coefficient calculators 5-2, - - - , 5-P calculate
auto-correlation coefficient sequences R.sub.2, R.sub.3, - - - , R.sub.P
of noise outputs of the respective sound receivers 1-2 to 1-P. The
mutual-correlation coefficient sequences .phi..sub.1j (j=2, 3, - - - , P)
and the auto-correlation coefficient sequences R.sub.k (k=2, 3, - - - , P)
are all supplied to the coefficient determining unit 6.
The coefficient determining unit 6 retrieves a maximum value related to the
thus supplied mutual-correlation coefficient sequences .phi..sub.1j
between the noise output of the sound receiver 1-1 during silence and each
of the noise outputs of the second receivers 1-2 to 1-P. Among these
sequences .phi..sub.1j, it is now presumed that a maximum value
.phi..sub.1j, it is now presumed that a maximum value .phi..sub.1q is
retrieved with j=q and having a delay time T.
Next, a filter coefficient of the equivalent noise-producing filter in the
form of a transversal filter having an impulse response hq(T) is
determined to be .phi..sub.1q (T)/R.sub.q (O). If q is 3, it means that
the filter coefficient which determines the impulse response h.sub.3 (t)
of the equivalent noise-producing filter 7-3 is calculated to be
.phi..sub.13 (T)/R.sub.3 (O). This operation is carried out by using the
aforementioned equation (12) to determine the coefficient a in compliance
with the equation (12). The coefficient a obtained by .phi..sub.13 (T)
being normalized with R.sub.3 (O) is offered as an optimum coefficient of
a tap T of the equivalent noise-producing filter 7-3. The noise output of
the sound receiver 1-3 is added to the adder 8-1 with its sign being
inverted via equivalent noise-producing filter 7-3, and adders 8-3 and
8-2, thereby to minimize the noise which offers a maximum
mutual-correlation coefficient sequence. Further, the remaining noise
component is sent to the coefficient determining unit 6 as a
noise-canceled residual waveform.
The coefficient determining unit 6 retrieves a maximum value again for the
noise-canceling residual waveforms that are input to repeat the same
processing cyclically until the electric power of the noise-canceled
residual waveforms becomes smaller than a predetermined level. The adders
8-2 to 8-P add up the outputs of the equivalent noise-producing filters
7-2 to 7-P, and second them to the adder 8-1.
In the foregoing were described the processing contents according to the
first embodiment.
A second embodiment is to further increase the efficiency of the process
for estimating the filter coefficients of the first embodiment. The second
embodiment is constituted by adding mutual-correlation coefficient adders
4-23 to 4-2P, 4-34 to 4-3P, - - - indicated by dotted lines to the
aforementioned first embodiment.
The mutual-correlation coefficient calculators find mutal-correlation
coefficients .phi..sub.ij (i=2, 3, - - - , (P-1), j=3, 4, - - - , P)
without superposition in a way that the mutual-correlation coefficient
calculators 4-23 to 4-2P find mutual-correlation coefficients between the
output of the sound receiver 1-2 and each of the outputs of the sound
receivers 1-3 to 1-P, and the mutual-correlation coefficient calculators
4-34 to 4-3P find mutal-correlation coefficients between the output of the
sound receiver 1-3 and each of the outputs of the sound receivers 1-2 to
1-P (except 1-3).
The coefficient determining unit 6 retrieves a maximum value .phi..sub.1q
out of the sequence .phi..sub.1j, and determines the filter coefficient at
the tap T of the equivalent noise-producing filter that has impulse
response hq(T) to be .phi..sub.1q /Rq(O).
The mutual-correlation coefficient .phi..sub.1q is corrected by Rq, and
.phi..sub.1j (j.noteq.q) other than .phi..sub.1q are all corrected by
.phi..sub.qj among .phi..sub.ij. If now Q is 3, .phi..sub.13 is corrected
by R.sub.3, and .phi..sub.ij other than .phi..sub.13 are all corrected by
.phi..sub.3j among .phi..sub.ij. The above correction processing is based
upon the contents explained in conjunction with the equations (14) to
(16). The feature of the second embodiment resides in that .phi..sub.1j
(j.noteq.q) are generally corrected by .phi..sub.qj among .phi..sub.ij,
and the coefficient estimating process starting from the retrieval of a
maximum value is cyclically performed by utilizing .phi..sub.12,
.phi..sub.13, - - - , .phi..sub.1P that are corrected, until the
noise-canceled residual waveform becomes smaller than a predetermined
level. By adapting this method, the coefficient estimating process of the
first embodiment can be further simplified. The coefficients are estimated
by utilizing the processing idea of FIG. 4 in order to greatly reduce the
amount of operation.
* * * * *
|
|
 |
|
 |
Description |
|
