Note: Descriptions are shown in the official language in which they were submitted.
CA 02427845 2003-05-02
- 1 -
METHOD FOR VERIFYING THE AVAILABILITY OF A SIGNAL COMPONENT
AND DEVICE FOR CARRYING OUT SAID METHOD
The present invention is related to a method according to
the pre-characterizing part of claim l, to a usa of the
method as well as a device to perform the method.
The detection of a narrow band signal component, as e.g. a
sinusoidal signal in a noise signal, is a problem to be
solved very often. To solve this problem different known
methods are available. A first method is using correlation
calculations, a second is a method based on parametrizing
followed by peak picking, and a third is using a number of
zero crossing counters.
All these known methods bear the drawback that high
computer power is necessary because of the complex
algorithms which must be applied. In pa=ticular, this is
the case if speech signals are being processed. Possible
fields of application are telecommunication products, audio
products or hearing devices, whereas in the following under
the term "hearing device" so-called hearing aids, which are
used to correct an impaired hearing of a person as well as
all other acoustical communication systems, as for example
radio sets, must be understood.
" CA 02427845 2003-05-02
- 2 -
The present invention therefore has the object to provide a
method which does not incorporate the above-mentioned
drawbacks.
This object is obtained by the el~mer_ts provided in the
characterizing part of claim 1. Advantageous embodiments of
the method according to the present invention, a use of the
method as well as a device to perform the method are given
in further claims.
The method according to the present invention is
characterized by a number of very simple method steps,
which can be performed by using little computer power.
Therefore, the method according to the present invention
qualifies in particular for the use in systems having
restricted access to energy supply, as for example for
mobile devices which must be power line independent, or for
systems in which the occurrence of a signal component must
be determined very quickly.
In further embodiments of the present invention it is
proposed to use the method for the detection and
elimination of signal feedback. Signal feedback is a known
problem in hearing devices, in mobile telephones and other
telecommunication products. A number of solutions have beer.
elaborated by the telecommunication industry. It is known
to attenuate the signals in the signal feedback path by
corresponding adjustment of the attenuation in the transfer
P202978 - 1074020.doc
' CA 02427845 2003-05-02
- 3 -
function in the feedback path. Furthermore, the use of
auto- ar_d/or cross correlations schemes has been envisioned
by which the correlation of the input signal and the outpu
signal are calcula=ed in the time domain or in the
frequency domain. The results of the calculations are used
to adjust the transfer function in the signal feedback
path, using the LMS-(Least Nlear_ Square)-algorithm (feedback
canceller). Alternatively, the results of the calculations
are used to adjust the transfer function in the forward
path, whereby the loop gain is reduced at the critical
frequencies.
For further information on the known methods it is referred
to the following printings: US-5 680 467, EP-0 656 737, WO
99/26453, WO 99/51059, DE-197 48 079.
The known methods have been used successfully but have the
drawback that again a high computer power is necessary to
obtain useful results. The use of the known algorithms in
hearing devices leads to an increased energy usage. As a
result thereof, the operating time until the next recharge
or replacement of the batteries is reduced, which reduction
is basically undesired.
In case the loop gain reaches a value which is greater than
one in a given frequency range, and in case the magnitude
of signal components is some decibels lower at other
frequencies than the frequency of the feedback signal if
P202978 - 1079020.doc
CA 02427845 2003-05-02
- 4 -
the gain is increased in the forward path, then a notch
filter according to the present invention can be used to
reduce the signal feedback. In case that different critical
frequencies lie too far apart, several notcr. filters can be
used according to a further embodiment of the present
invention.
In order that a notch filter can be adjusted to the
critical frequency, i.e. the feedback frequency, the
critical frequency must be detected first. According to the
invention this is performed by the calculation of the
variance of the measure for the frequency of the input
signal, whereas signal feedback is being detected if the
variance lies within a predetermined range in relation to a
predetermined limit value.
The invention will be further explained in the following by
referring to drawings which show exemplified embodiments,
wherein:
Fig. 1 shows a magnitude spectrum of an input signal
having a superimposed narrow bandwidth signal
component;
Fig. 2 shows a block diagram of a circuit arrangement for
checking of an occurrence of a signal component in
the input signal;
P202979 - 1074020.dcc
' CA 02427845 2003-05-02
- 5 -
Fig. 3 shows a block diagram of a circc,it arrangement for
the detection and elimination of a signal feedback
component, and
Fig. 4 shows a further specific embodiment of the circuit
arrangement according to fig. 3.
fig. 1 shows a magnitude spectrum, i.e. the magnitude of an
input signal x in function of the frequency f. In a
frequency range B, which is limited by the upper and lower
frequency fB~l and f~~~, respectively, a narrow bandwith
signal component s with a middle frequency fkrit is
identifiable. The magnitude at the frequency Brit lies some
dB (Decibel) higher than the rest of the input signals x in
the frequency range B. In a first embodiment of the present
inventicn it is provided to detect the occurrence of the
signal component s. A circuit arrangement, which can be
used therefore, is schematically shown in Fig. 2. In a
second embodiment of the present invention, it is provided
to eliminate a detected signal component s from the input
signal x, which signal component s emerged e.g. from a
signal feedback. If the elimination of the signal component
s cannot be reached, it is at least possible to attenuate
the signal component s in a desired measure. Possible
circuit arrangements, which can be used to perform this
task, are schematically shown in figs. 3 and 4.
P202978 - 1074020.doc
CA 02427845 2003-05-02
- 6 -
According to fig. 2, a number of functional units are
connected in series, starting with a band pass filter 1, an
estimator unit 2, a variance unit 3 and a comparator unit
4. The input signal x, which consists either of an
exploitable signal a or of an exploitable signal a and a
superimposed signal component s, is fed tc the band pass
filter 1 having upper and lower limit frequencies f9~1 and
fap2 according to fig. 1, whereas it is assumed that the
signal component s, if it exists, lies within the frequency
range B (fig. 1). The band-limited signal, i.e. the output
signal of the band pass filter 1, is fed to the estimator
unit 2, in which a measure feet for the frequency of the
input signal s is determined.
The term measure felt for the frequency of the input signal
x basically means any frequency-dependent function.
It is proposed that as a first function y1 of the expected
value of the magnitude of a low-pass filter is used. In
time-discrete format, such a function can be stated as
follows:
y,~n~=Ej~.r~n~+x~n-1~~
and in the z-plain, respectively,
P202978 - 1074020.doc
CA 02427845 2003-05-02
Y, (r)=E~l+=-'I ~IX(_~},
whereas a normalization is preferably performed using the
level of the input signal x in order that the level itself
does not have an influence o~ the measure fe3~ for tre
frequency. For the last mentioned reason, two functions are
necessary, of which at least one is frequency-dependent.
As second function y2, a corresponding high-pass filter, or
much easier, merely the expected value of the magnitude of
the input signals x, is chosen:
YZ = E~~~n
By dividing the function y1 by the function y~ the desired
measure felt for the frequency of the input signal x, which
is now magnitude-independent, is obtained, namely:
f,~r(n)-E~x~n~+x(n-l~j- 2.~1+cosco
E x(n),
whereby co refers to the angular frequency.
P202978 - 1074020.doc
CA 02427845 2003-05-02
- c3 -
The determination of the erected value can also be
approximated by an moving averages of first order, which
can be described by the following equation:
whereas
r
and whereas T corresponds to the sample interval and i
corresponds to a time constant having a value of approx. 20
ms.
Whether a signal component s in the input signal x exists,
can be determined by calculating the variances v of the
measure felt for the frequency. Therefore, the variances
unit 3 according to fig. 2 is provided. If the variance v
lies below a given limit value LT, it can be concluded that
a narrow band-width, frequency-stable signal component s
exists in the frequency range B (fig. 1). As a prerequisite
it is mandatory that signal component s, if it exists,
bears a certain stability and that the exploitable signal a
is stable in this sense. Information regarding the
calculation of the variance can be obtained, for example,
from the standard work of Athanasios Papoulis entitled
"Probability, Random Variables, and Stochastic Processes"
(P~lcGraca-Hill, 19$4, page 108 ff. ) .
p2o297a - 10~~o2o.ao~
' CA 02427845 2003-05-02
- g _
The mentioned comparison of tze calculated variance v and
the predetermined limit value LT takes place in the
comparator unit 4 raving an output signal of either zero or
one, depending on whether the variance v is larger than the
limit value LT or vice versa.
The method according to the present invention described
along with fig. 2 can be used in particular for the
detection of a punch of a pushbutton of a telecommunication
terminal supporting frequency dialing. As is generally
known, each of the twelve pushbuttons of such a terminal is
coded by two of a total of seven sinusoidal signals,
whereas the frequencies of the signals are known. The
detection of punching one cf the pushbuttons is therefore
limited to check the occurrence of signals having
corresponding frequencies. According to the two detected
frequencies the pushbuttons being punched can be
identified, whereas the circuit arrangement according to
fig. 2 can be used for each possible signal. Thereby, the
band-pass filter is adjusted in such a way that only one
signal can pass through the band-pass filter. Naturally,
there exists the possibility that a filter bank consisting
of seven band-pass filters to select each of the single
possible signals is provided and that the further
processing of the signals in the estimator unit 2, in the
variance unit 3 and in the comparator unit 4 is dealt with
in. a time multiplex process.
e~oL~~a - 1074020.doc
CA 02427845 2003-05-02
- 1G -
Fig. 3 shows a further block diagram of a Further
embodiment whicr: is based or. tha one shcwn in fig. 2. The
block diagra~r shown above the dashed line ir. fig. 3 is
exactly the same as the one shown in fig. 2. Below the
dashed line according to fig. 3 a filter unit 6, a
coefficient ca?cuiation unit 5 and a switching unit 7 are
provided.
The input signal x fed to the band-pass filter 1 is further
connected to the filter unit 6 and to one of the two
switching contacts S2 of the switching unit 7. The output
signal of the filter unit 6 is connected to the further
switching contact S1. Furthermore, the measure felt for the
frequency of the input signals x is fed to the coefficient
calculation unit 5, in which the coefficients of the filter
implemented in the filter unit 6 are calculated in a way
yet to describe. The calculated coefficients will be
transferred to the filter unit 6 as soon as the
coefficients are determined. The determination of the
measure felt for the frequency can be provided in a way
described along with fig. 2.
Finally, either the input signal x directly or the output
signal of the filter unit 6 will be switched to the output
z in the switching unit 7 according to a control signal
generated in a comparator unit 4. Ir_ other words, the input
signal x is either filtered in the filter unit 6 or the
input signal x will be passed to the output z without being
processed. The switching is advantageously done in a
P2029?3 - 1074020.doc
CA 02427845 2003-05-02
- 11 -
"so=tly", which means tha transition from one stage to
ancther is done is a smooth way.
As a consequence, the method according to the invention and
the device according to the invention, respectively, can be
advantageously used to detect and eliminate a feedback
signal, to be precise, for telecommunication products as
well as for hearing devices, whereas the computer power
necessary for the calculations is diminishing. As a result,
in particular when using the method according to the
invention in a hearing device, the energy consumption can
be held at a low level for the additional computational
efforts.
In case that the signal component s must be suppressed by
the filter unit 6, or at least attenuated, the filter unit
6 is realized as notch filter, whereas the maximum
attenuation of the notch filter must lie in the middle of
the frequency fkrit to be suppressed (fig. 1) . A notch
filter can be realized according to the following equation:
H~z~=1+b' ~z-' +bz ~z-Z
whereas
b' =-2~r~cosr~
and
b,=rz.
P202978 - 1074020.doc
" CA 02427845 2003-05-02
- 12 -
The notcr, filter according to the above meraioned equation
features one single zero having a distance of radius r to
the origin. It is proposed tc fix the radius r, for eaa~nple
by giving it the value 0.98, whereas only cos a~ has to be
determined in order to determine the coefficient b,. This
value can be derived according to tre present invention
from the measure feet for the frequency of the input signal
x by solving the above mentioned equation for the measure
felt for cos w. One can obtain the following equation:
b' _ _2 . r , .f2st _ 1
In a further embodiment of the present invention, it is
provided to determine the notch filter according to the
following equation:
H~w~ _ 1+b' ~~-' +b, .r-z
_~ .
1+a' ~z-' +a: ~z
whereas
a' _-2~rP ~cosco ,
b'=-2~rZ~cosr.~,
and
a~=rP,
P20297$ - 1074020.doc
CA 02427845 2003-05-02
- 13 -
bi - rz .
The equations mentioned above can again be solved far cos
cu in an analogous way. Hereby, the following two equations
can be obtained:
av =-2.rP. .f2r _1
and
2
b, _ -2 . rz _ .f2s' _ 1
The equations mentioned above describe thereby an algorithm
for the estimation of a narrow bandwidth signal component s
and, at the same time, allow obtaining coefficients for the
notch filter to suppress the signal component s.
Fig. 4 shows a specific embodiment of the schematic
representation of the present invention according to the
block diagram of fig. 3. The processing units designated in
fig. 3 are identified by dashed lines in fig. 4, whereby
the same reference signs are used as in fig. 3.
In the estimator unit 2, the block diagram according to the
equations, which have been described in connection with
fig. 2, is shown. Besides the units resulting directly from
the above mentioned equations and which units are not
P202978 - 1074020.doc
CA 02427845 2003-05-02
- 14 -
further explained, two decimation. units 10 and 11 are
providzd in addition, which are provided before a quotient
unit 12 and which reduce the data rate in order to reduce
the already reduced computational effort even further.
Methods for the data rate reduction are generally known and
are further explained, ~cr~example, ir_ the standard work of
R.E. Crochiere et al. entitled "Multirate Digi~al Signal
Processing" (Prentice-Hail Signal Processing Series,
Prentice-1-iall, Inc. , Engl ewood Cl iffs, DTew tersey, 1983) .
Sufficient anti aliasi.ng filters are implicitly provided
before the actual decimation.
Without the decimation units 10 and 11 the measure felt for
the frequency of the input signal x can be obtained at the
output of the estimator unit 2, as has been explained along
with fig. 2:
.fesan~= E~~~n~+r~n-l~i
E x(n))
Considering tha above mentioned explanations, in particular
the one made in connection with the block diagram shown in
figs. 2 and 3, a probability measure fbprQb for the feedback
can be determined from the input signal x according to the
following equation in the variances unit 3 or in the
comparator unit 4, respectively:
P202978 - 1074020.doc
CA 02427845 2003-05-02
- 15 -
J "prob = max 1- k ~ E E{fefr I - fest ~O
J
whereas k represents a sensitivity parameter through which
the amount of ir_fluence of the control mechanism is
determined. According to fig. 4 the probability measure
fbp=ob is not yet the output signal of the comparator unit 4
since it is necessary to change the data rate in
interpolator unit 13, in which a data rate reduction is
performed analogously to the data rate increase in the
decimation units 10 and 11, i.e. in the interpolation unit
13 the data stream is readjusted to the original data rate
of the input signal x.
In the above mentioned equation for the probability measure
fbProb the expected value E{...} is again realized, in the
simplest embodiment of the method according to the present
invention, as a moving averager with a short time constant
for a signal follow-up towards larger signal values, but
with a long time constant for the signal follow-up towards
smaller signal values. Such a moving averager is also
called a fast attack - slow release averager. A
corresponding moving averager 14 is connected to the output
of the comparator unit 4. Thereby, the control behavior of
the closed loop control circuit is further improved.
P202978 - ~074020.doc
' CA 02427845 2003-05-02
- 16 -
The expected value designated E~..} is a symmetric moving
averages which means that the attaci and release time
constants are equal.
In tre filter unit 6 a notch filter according to the
follocaing equation is realized:
HtZ) 1 ~' fbprob ~ \b1 ~ ~ ~ '~w b' . ~pnob ~ Z '
whereas the coefficients b1 and b2 are determined as
follows in the coefficient calculation unit 5:
b~ _ -2 ~ r ~ 'fe$' -1 and
2
bz=rz.
The radius r is again the distance from the zero to the
origin in the z-plane and is preferably fix. It could have
been shown that it is advantageous to choose a value of
0.98 for the radius r. Instead of the above mentioned
specific transfer function for the notch filter the general
form is shown in the following, which is preferably used:
P202978 - 1074020.doc
CA 02427845 2003-05-02
- 17 -
H ~- ~ _ _1 + ~prob ' \b! ' .. -~- b, ' l uprob
fi ~_1 ~, __, r
1 -~- J bprob ' al ' " -~- a2 ' J "prob '
whereas
z
bt - -7 , rl , -Pest - 1
7 r
0
al = -7 . rp . fesr - 1
2
b2=rz,
a, = rP and
~prob =lnax 1-k'E E~f'~t~-,fesr ~~
With r, a constant is referenced having a value of
preferably 0.98; k is a sensitivity parameter for the
adjustment of control characteristics, whereas the value
for k is preferably equal to 10.
P202978 - 1074020.doc