The present invention is generically directed on a technique
for so-called "beam forming" on acoustical signals.
The use of directional acoustical/electrical transducers and
especially of such microphones is one of the most efficient
ways for improving signal to noise ratio in audio systems. It
is known to realise directional microphones by using an array
of microphone cells and time delaying and superimposing the
output signals of such cells following up the known "delay and
sum" technique.
With two omnidirectional microphone cells this known principle
is shown in fig. 1. Two omnidirectional microphones, 1 and 2,
are provided with a mutual distance p. The output signal of one
of the microphones according to signal A1 is time delayed by
the time amount τ, the time delayed signal according to A1' is
superimposed at a superimposing unit 3 to the undelayed output
signal A2 of microphone 2. At the output of the superimposing
unit 3 there results the output signal Ar with an amplification
versus impinging angle characteristic, as shown in fig. 2 for
one frequency ω considered. Thereby, it is customary to select
as delay time τ as the quotient of distance p and velocity of
sound c. With this arrangement there results, as shown in fig.
2, a first order cardoid characteristic. It may be shown that
the amplitude of the resulting signal Ar is proportional to the
sine of the signal frequency ω and to the distance p. The
maximum gain in target direction (180°) occurs at the frequency
fr = c/(4p). For a distance p of 12 mm, fr becomes approx. 7
kHz.
By staggering more than one of the fig. 1 double-cell arrangements
and superimposing the resulting signals Ar of the more
than one double-cell arrangements, higher order cardoid characteristics
may be realised.
In fig. 3 a known arrangement to realise second order cardoid
characteristics according to fig. 4 is shown. Thereby, a narrower
beam can be achieved. The higher the order of the directional
microphone arrangement, the higher becomes the directivity
index and the gain at fr, but the higher will also be the
roll-off for low and high frequencies and the number of unwanted
side-lobes. With respect to the definition of the directivity
index please refer to speech communication 20 (1996),
229 - 240, "Microphone array systems for hands-free telecommunications",
Garry W. Elko.
In fig. 5 there is shown the gain versus frequency characteristic
of the first and second cardoid characteristics for an impinging
angle = 180°. Therefrom, high and low frequency roll-offs
are clearly evident.
Such techniques for beam forming are well-known and have been
realised using analogue signal processing, as e.g. shown in the
US-A-2 237 298, US-A-4 544 927, US-A-4 703 506, US-A-5 506 908
or using digital signal processing, both in time or in frequency
domain, as shown in the EP-A-0 381 498 (time domain) or
in the US-A-5 581 620 (frequency domain).
Beam formings realised with any of these principles has the
following drawbacks:
a) The resulting signal is dampened at low frequencies, which
results in a bad signal to noise ratio. b) The directivity index is very sensitive to matching of the
individual microphone cells, especially at low frequencies. c) The distance p between the microphone cells should be large
(> 12 mm) for audio range. d) The frequency band with a high gain in target direction is
rather small, as may clearly be seen from fig. 5. e) The directivity largely depends upon the number of microphone
cells and thus on the complexity of the overall arrangement. f) As one aims for a high directivity by increasing the number
of cells, more unwanted side-lobes are introduced.
Several techniques have been proposes to overcome some of these
drawbacks:
In the WO 95/20305 (E. Lindemann) an adaptive noise reduction
system for use in binaural hearing aid is proposed. It detects
the power of the received signals to separate the desired from
unwanted signals.
There is proposed a "broad side" microphone-cell array, i.e.
target direction is perpendicular to the axis from one microphone
to the other, in contrary to the arrangement according
e.g. to Fig. 1 and the principles of the present invention,
which is "in line".
The disclosed apparatus is bulky (>> 5 cm), so that it may not
be implemented for one ear hearing aid.
Two equal beam lobes are generated in target and in opposing
directions.
In such a hearing aid a connection between the left and right
ear system must be present, making the apparatus for hearing
aid unhandy. Furthermore, as described by the same author in
"Two microphone non-linear frequency domain beam former for
hearing aid noise reduction" 1995, IEEE ASSP Workshop on Applications
of Signal Processing to Audio and Acoustics, October 15
- 18, Mohonk, New Paltz, New York, such beam forming is efficient
only up to about 2 kHz and leads to distortions of the
desired signals.
The US-A-4 653 102 proposes the use of two directional microphones
aimed in target direction and of a third microphone
aimed in opposite direction. The signal of the third microphone
supposedly only containing noise is used to shape the response
of the two primary microphones. This technique obviously has
the drawback within reverberating rooms, where the desired signal
is reflected on walls, floor, ceiling and furniture and is
therefore considered as noise by the system. This technique is
further unhandy as making use of at least three microphones.
Attention is further drawn to US-A 5 400 409 and 5 539 859.
It is an object of the present invention to provide a method
for electronically forming a predetermined characteristic of
amplification in dependency of direction from which acoustical
signals are received at at least two spaced apart acoustical/electrical
transducers and a respective acoustical sensor
apparatus, with which only a small number of microphones or microphone
cells has to be used and which is thus enabling small
and compact directional transducer or microphone realisation.
Thereby, the preferred apparatus according to the present invention
is a hearing aid apparatus, and especially a one ear
hearing aid apparatus.
It is a further object to provide such method and apparatus
with good frequency response in the audio band, i.e. between
approx. 0,1 and 10 kHz.
Still a further object of the present invention is to provide
such method and apparatus which allow high signal to noise ratio
realisation without unwanted side-lobes and with easily
variable beam form, e.g. for acoustical zooming.
These and other objects are realised by the inventive method,
which comprises the steps of repetitively determining from signals
dependent from the acoustical signals a respective mutual
delay signal according to reception delay at the at least two
transducers; subjecting a signal dependent from the output signal
of at least one of the at least two transducers to filtering
with a filtering transfer characteristic; and of controlling
the filtering transfer characteristic in dependency of the
mutual delay signal; further exploiting a signal dependent from
the output signal of the filtering as electrical reception signal.
To fulfil the above mentioned objects the inventive acoustical
sensor apparatus comprises at least two acoustical/electrical
transducers, arranged at a predetermined mutual distance in
target direction, a time delay detection unit, which has at
least two inputs and an output, the inputs thereof being respectively
operationally connected to the outputs of the two
transducers, whereby the time delay detection unit generates an
output signal in dependency of the time delay of acoustical
signals, impinging on the at least two spaced apart transducers,
preferably a time domain to frequency domain converter
unit generating the output signal of said time delay detection
unit in frequency domain; a weighing unit with a predetermined
weighing characteristic and with an input and with an output,
whereby the input thereof is operationally connected to the
output of the time delay detection unit and preferably receiving
the signal at said output of said time delay detection unit
in frequency domain mode; with a filter unit with a controllable
transfer characteristic, which has at least one input, a
control input and an output and whereat the input is operationally
connected to at least one of the outputs of the at least
two transducers, preferably via at least one time domain to
frequency domain converter, the control input is operationally
connected to the output of the weighing unit, the filter unit
generating an output signal in dependency of its input signal
and of its transfer characteristic which is controlled by the
signal - preferably a spectral signal - which is applied to the
control input of the filter unit, this weighing - preferably
spectral weighing - result signal being dependent from the output
signal of the time delay detection unit and the weighing
characteristic of the weighing unit.
Other objects, advantages and specific embodiments of the present
invention shall be exemplified with the help of further
figures. The figures show:
- Fig. 1:
- A functional block diagram of a two-cell directional
microphone arrangement according to the prior art
principle of "delay and sum";
- Fig. 2:
- the first order cardoid amplification characteristic
of prior art arrangement according to fig. 1;
- Fig. 3:
- departing from the prior art arrangement of fig. 1 a
further arrangement following up the technique of
"delay and sum" for realising second order characteristic;
- Fig. 4:
- the second order amplification characteristic as realised
by the prior art arrangement according to fig. 3;
- Fig. 5:
- in dependency of frequency the amplification characteristic
of the arrangement according to fig. 1 or 3
at maximum amplification impinging angle of acoustical
signals;
- Fig. 6:
- a simplified functional block diagram of an inventive
apparatus operating according to the inventive method
and further showing the sequence of process signals;
- Fig. 7:
- in a representation according to fig. 6 a first preferred
realisation form of an inventive apparatus operating
according to the inventive method;
- Fig. 8:
- in an inventive apparatus operating according to the
inventive method according to fig. 6 a further preferred
form of realisation of a time delay detection
unit;
- Fig. 9:
- a polar diagram of signals as realised by the embodiment
of fig. 8 for explaining operation of a comparator
unit as provided in the fig. 8 embodiment;
- Fig. 10:
- the course of comparison results in dependency of impinging
angle of an acoustical signal and as realised
by the embodiment according to fig. 8;
- Fig. 11:
- a preferred form of realising superimposing result
signal dependency from impinging angle of an acoustical
signal at an embodiment according to fig. 8;
- Fig. 12:
- in a representation according to fig. 10 the course of
comparison results as realised with a preferred embodiment
resulting in the fig. 11 dependency;
- Fig. 13:
- in polar diagrammatic representation the dependency of
superimposing result signals from impinging angle of
acoustical signals and from frequency as realised by
the embodiment according to fig. 8;
- Fig. 14:
- a preferred realisation form of the fig. 8 embodiment,
additionally counteracting frequency dependency as
shown in fig. 13;
- Fig. 15:
- in a representation according to fig. 13 the dependency
of the superimposing result signal with normalisation
as realised by the embodiment of fig. 14 with a
first preferred normalisation frequency function;
- Fig. 16:
- a representation according to fig. 15, realised with a
second preferred normalisation frequency function at
the embodiment of fig. 14;
- Fig. 17:
- a first (rigid line) and second (dashed line) preferred
realisation form of amplitude filter characteristic
at the embodiment of fig. 6 or 7;
- Fig. 18a:
- the effect of the amplitude filter amplitude versus
amplitude transfer characteristic according to fig. 17
(rigid line) on the output signal of the delay detection
unit as provided in the embodiment of fig. 6 or
7;
- Fig. 18b:
- the representation of the output signal of a time delay
detection unit passed through the amplitude filter
with a transfer characteristic according to fig. 17
(rigid line) and as realisable by the embodiment of
fig. 6 or 7;
- Fig. 19:
- the spectrum of an acoustical signal converted into
electrical and input to a controllable frequency filter
as provided by the present invention according to
fig. 6;
- Fig. 20:
- the electrical reception result signal realised by amplitude
filter characteristic according to fig. 17
(rigid line) and reception signal as exemplified in
fig. 19 at the inventive embodiment according to fig.
6;
- Fig. 21:
- the resulting dependency of amplification from impinging
angle of an acoustical signal as realised by the
fig. 17 amplitude filter characteristic (rigid and
dashed lines);
- Fig. 22:
- the amplification versus impinging angle characteristic
as realised by the fig. 6 or fig. 8, 14 embodiments
of the invention, making use of an amplitude
filter characteristic with a maximum to minimum spectral
amplitude transfer behaviour;
- Fig. 23:
- in a simplified signal/functional block diagram a further
preferred embodiment of the invention;
- Fig. 24:
- in a signal flow, functional block representation, a
further mode of realisation of the time delay detection
unit as shown in fig. 6 and
- Fig. 25:
- in a signal flow, functional block representation, a
further mode of realisation of the time delay detection
unit following the technique as shown in fig. 8
or fig. 14.
In fig. 6 there is shown in form of a functional block diagram,
together with principle signal processing diagrams, the principle
of the inventive method and apparatus.
At least two acoustical/ electrical transducers 1 and 2, as especially
of microphones or microphone cells, are provided with
a predetermined mutual distance p along axis a. Acoustical signals
IN are received by the transducers 1 and 2 as they impinge
from different spatial directions . The acoustical signals IN
have frequency spectra which vary in time. Output signals of
transducer 1, S1(t,ω) and of transducer 2, S2(t,ω), are formed
as electrical signals at the output of the transducers 1 and 2.
Due to the mutual distance p of the two transducers 1 and 2 -
which is preferably smaller than 5 cm, preferably between 0,5
and 1,5 cm, especially for the inventive sensor being a one ear
hearing aid apparatus - and as shown with the two respective
pointer diagrams below the functional block diagram of fig. 6,
the acoustical signals IN impinge on the transducers 1 and 2
with a time delay dt, which may be expressed by the phase difference
Δϕω at each spectral frequency ω according to
Δϕω = ω.dtω, where dtω = p c cos ω.
If the source of acoustical signal IN is a point source, then
the time delay dtω becomes equal for all spectral components at
the different ω. The output signals S1 and S2 of the transducers
1 and 2 are operationally connected to the respective inputs
of a time delay detection unit 10, which generates an output
signal A10 according to the spectral distribution of time
delays dtω, which are, as was explained, a function of the impinging
angle at which the respective frequency components
impinge on the transducers 1 and 2 and thus in fact of ω.
Purely as example a possible spectrum of output signal A10 is
also shown in fig. 6. This spectrum varies in time according to
the time variation of impinging acoustical signal IN. The output
signal A10 of time delay detection unit 10 is input to a
weighing unit 12. As the spectrum of dtω with respective spectral
amplitudes of A10 is input to the weighing unit 12 with
the preselected weighing transfer characteristic W, there results
at a certain moment in time, as an output signal A12, a
spectral signal W(ω), as also shown as an example in fig. 6.
A12 results from respectively weighing the spectral amplitudes
of A10 according to the characteristic W. As A10 indicates according
to dtω from which direction ω each frequency component
of the acoustical signal IN impinges, its specific weighing by
means of function W is nothing else than predetermining which
impinging directions ω shall be amplified or attenuated. Thus,
the weighing unit 12 determines with its characteristic W the
beam shape.
The output signal A12 is applied to a filter unit 14 with a
controllable transfer filter characteristic. There, each spectral
line of the time varying spectrum of the output signal
S1(t,ω) is amplified or attenuated according to the controlling
spectrum Wω · A10ω. Thus, unit 14 is a filter unit for input
signal S1 at which the transfer characteristic is varied, as
controlled by A12. In dependency of the kind of filter unit 14
the weighing unit 12, generally spoken, calculates adjustment
of filter characteristic determining coefficients as a function
of A10.
Thus, along the channels 10 and 12 there is predetermined by
the weighing transfer function W which spatial directions
shall be "aimed" at. At the filter unit 14 this beam shaping
information is applied to the electrical analogon S1 of the
acoustical signal IN, thus resulting in an output signal
Sr(t,ω) representing the shaped reception signal.
By adjusting the weighing transfer function W by applying a
control signal CW to a control input C12, the beam form can be
adjusted and thus acoustical zooming is realised.
As shown in dashed line, it may be advantageous to subject both
transducer output signals to a controlled filtering at unit 14.
In fig. 7 there is shown a first preferred form of realisation
of the inventive principle according to fig. 6. Thereby, the
output signals S1 and S2 are first converted from analogue to
digital form in respective analogue/ digital converters 16 and
17. The digital output signals of the respective converters 16
and 17 are input to respective complex time domain/ frequency
domain converters 18 and 19.
The output spectra S
1(t,ω) and S
2(t,ω) of
converters 18, 19 are
input to the spectral time delay detection unit 10'. Unit 10'
computes according to formula (1) the phase difference spectrum
Δϕ
ω divided by the respective frequency ω to result in an output
signal spectrum A
10' according to the time delay dt
ω as was
explained in connection with fig. 6. The output signal of the
time delay detection unit 10', A
10', is further treated, as was
explained in connection with fig. 6, by the weighing
filter
unit 12 and the
controllable filter unit 14. In the following
table there is exemplified how the unit 10' operates. Out of
the spectral phase distribution ϕ
1n of signal S
1 and ϕ
2n of signal
S
2 the time delay dt
ω is calculated for each spectral line
within an interesting spectral band.
| ω1 | ω2 | ω3 | | | ωn |
S1(ω) | A11 | A12 | A13 | | | A1n |
ϕ11 | ϕ12 | ϕ13 | | | ϕ1n |
S2(ω) | A21 | A22 | A23 | | | A2n |
ϕ21 | ϕ22 | ϕ23 | | | ϕ2n |
|
dtωn | ϕ11-ϕ21/ω1 | ϕ12-ϕ22/ω2 | ϕ13-ϕ23/ω3 | | | ϕ1n-ϕ2n/ωn |
So as to extract the phase information ϕ out of the two signals
S1, and S2 the time domain to frequency domain conversion units
18 and 19 perform complex (real and imaginary) operation.
A second preferred realisation form of the present invention,
and especially as concerns realisation of the time delay detection
unit 10, shall be explained with the help of fig. 8 and 9.
The output signal of one of the transducers, as shown e.g. of
transducer 1, S1(t,ω) is fed to a time delay unit 20, wherein,
in a first form of this realisation, signal S1 is time delayed
by a predetermined frequency independent time delay τ.
Looking back on fig. 1, signal S1 accords thus with signal A1.
The output signal of time delay unit 20 thus accords with signal
A1' of fig. 1.
The time delay signal according to A1' is superimposed to the
output signal S2(t,ω) from transducer 2 at a superimposing unit
23 according to unit 3 of fig. 1, thus resulting in an output
signal according to Ar(t,ω) of fig. 1. As is known, and as was
explained in connection with fig. 1, the output signal Ar(t,ω)
depends from the impinging acoustical signal direction according
to the first order cardoid beam of fig. 2, which cardoid
function nevertheless varies with frequency ω. The output
signal Ar of superimposing unit 23 and e.g. the output signal
S2(t,ω) from transducer 2 are input to a ratio unit 25, as a
comparator unit.
For understanding the functioning of ratio unit 25, attention
is drawn to fig. 9. In fig. 9 the cardoid attenuation characteristic
of output signal Ar at a specific spectral frequency
ω1 is shown. At a specifically considered impinging angle o
the output signal Ar of superimposing unit 23 is Aro(ω1) with an
amplitude value as indicated in fig. 9. Simultaneously, at that
frequency ω1 considered and at that impinging angle o considered,
the amplitude of signal S2 is A2o(ω1) as also shown in
fig. 9. It must be emphasised that as the amplitude A2o varies,
the amplitude of Aro varies proportionally. Thus, the ratio of
Aro to A20 according to fig. 9 is indicative of the impinging
angle o. In the division unit 25 of fig. 8 there is formed for
each spectral component amplitude the ratio of Ar to Ar, wherefrom
there results a signal spectrum at the output of division
unit 25 with a ratio spectrum. The spectrum of A10 according to
fig. 6 thus becomes the spectrum of an amplitude ratio which
nevertheless is indicative of the impinging angle , at which
each frequency component of the spectrum of the acoustical signal
impinges with respect to the axis a of the two transducers
(see fig. 6). In fig. 8 the dotted line block indicates the delay
detection unit 10 according to fig. 6. Further signal processing
is performed as was explained by means of fig. 6, i.e.
via weighing unit 12 and controllable filter unit 14.
In this embodiment it is possible to perform a time domain to
frequency domain conversion at the output side of comparator
unit 12.
Thus, the output ratio signal of unit 25 is a measure for the
time delay dtω and is input to the weighing unit 12.
In fig. 10 there is shown the course of the ratio Ar to Ao as a
function of at a specific frequency ω1.
This amplitude ratio is shown for τ at unit 20 of fig. 8 being
selected to be
τ = p/c,
wherein p is the distance of the transducers 1 and 2 and c is
the velocity of sound.
When selecting τ to be p/c and as may be seen from the cardoid
beam function of fig. 2 signal attenuation or dampening for
near 0° becomes very high.
Thus, in this area of impinging angle any kind of noise in A2
according to S2 of fig. 8 would falsify comparison result
formed at unit 25. This problem can be eliminated by choosing a
delay τ which is different, thereby preferably larger than p/c.
In fig. 11 the resulting cardoid diagram is shown for τ = 1.2
p/c, whereas fig. 12 shows in analogy to fig. 10 the course of
the amplitude of Ar divided by the amplitude A2.
Further, it must be noted that the cardoid function as shown in
the figures 2, 9 and 11 is only valid for one specific frequency
considered. In fact, considering different frequencies,
the cardoid function varies as shown in fig. 13, wherein the
amplitude Ar of the output signal of superimposing unit 23 according
to fig. 8 is shown for p = 12 mm, a delay τ of 42 microseconds
and for frequencies of 0.5, 1, 2, 4 and 7.2 kHz.
From this polar diagram the frequency dependency of the cardoid
amplification function is clearly evident. Although such dependency
may be neglected in a first approximation, in a preferred
form of realising the inventive method principally as
shown in fig. 8, such dependency is taken into account. Thus, a
preferred realisation form of the fig. 8 technique is shown in
fig. 14. Here, the same reference numbers are used as in the
figures 7 or 8. The outputs of the transducers 1 and 2 are converted
into digital form by respective analogue to digital converters
16, 17 and the resulting digital signal of transducer 1
is time delayed by a time delay τ', which is larger than p/c.
The output signal S2 of transducer 2 is further converted into
frequency domain by a linear (not complex) time to frequency
domain conversion unit 18', whereas the output signal Ar of superimposing
unit 23 is converted to frequency domain at a linear
time to frequency domain conversion unit 19'. The frequency
dependent polar diagram according to fig. 13 is taken into account
by a normaliser unit 30 which is in fact a filter. In a
first embodiment the transfer characteristic of the filter is
selected proportional to 1/ω. This results in a frequency dependency
of the pole diagram as is shown in fig. 15 for the
same distance and frequency values as shown in fig. 13.
As may be seen, a good matching is achieved for small angles
and frequencies up to about 4 kHz. At 4 kHz the deviation is
about 10 %, at = 180°.
A further, even improved normalisation function or filter characteristic
at unit 30 of fig. 14 is achieved when the filter
characteristic is selected as a function 1/sin(ω). The result
is shown in fig. 16. The characteristics match well from 0.5 to
4 kHz. A further advantage of this normalisation technique is
improved sensitivity in backwards direction. This improved sensitivity
may be exploited for adaptive beam forming, that is
for selectively eliminating noise sources from the rear side.
It is evident for the skilled artisan, that such normalisation
may be performed also in signal path 1 to 23 and/or 2 to 23.
In this embodiment of fig. 14 it is highly advantageous that
only one-dimensional TFC's 18', 19' have to be used and not
complex TFC's as in the embodiment of fig. 7.
Fig. 24 shows in block diagram form, that the signal A10 (dtω)
may also be generated as the output signal of a comparator unit
60 to which on one hand the output signal of an omnidirectional
transducer 61, having equal amplification of its acoustical/electrical
reception characteristic substantially irrespective
of the impinging angle and the output signal of a directional
transducer 62 with selected, beam shaped reception characteristic
are led to.
According to fig. 25 time delaying τ may also be performed by
one of the transducers itself.
Thereby, in the embodiment of fig. 25 as well as of fig. 8 τ
may be selected to be zero.
With the help of fig. 23 a further preferred embodiment, especially
of realising the time delay detection unit 10 of fig. 6
shall be explained. The output signals of the transducers 1 and
2 are first converted by respective analogue to digital converters
16 and 17 and then by respective time domain to frequency
domain converters 18, 19 finally into frequency domain.
One signal, as an example S2, of the converted output signals
of the transducers, which, after time to frequency domain conversion
may be represented as a spectrum of S2ω pointers, is
converted to its conjugate complex pointers at a conversion
unit 50. At the output of this unit 50, the conjugate complex
pointers S*2ω are generated. This spectrum S2* and the pointer
spectrum S1 are multiplied to form the scalar product spectrum
S3 in a multiplication unit 52. As may be shown, the pointers
S3ω of spectrum S3 have a phase angle with respect to the real
axis, which is Δϕω.
Thus, the imaginary part of the pointers S3ω of S3 become
Im (S3ω) = |S3ω | sin (Δϕω)
with
Δϕω = ω · (p/c) · cos(ω)
According to fig. 23, a conversion unit 53 forms the imaginary
part of the pointers S3ω and a further unit 54 forms the amplitudes
|S3ω| of these pointers.
For small values of Δϕω the sinus in (3) may be approximated by
Δϕω itself, so that there results from (3)
Im (S3ω) = |S1ω · S*2ω| ω(p/c) · cos (ω)
Thus, and as performed by unit 55, dividing the imaginary parts
Im (S3ω) of the pointers S3ω of spectrum S3 by the respective
values of the scalar product according to |S3ω|, there results
an output signal which accords with Δϕω. As was already explained
with the help of fig. 7, Δϕω is further divided in unit
56 by the respective pointer frequency ω. The resulting signal
is A10 according to fig. 6 or A10' according to fig. 7.
All the units 50, 52, 53, 54, 55 and 56 are preferably realised
in one calculator unit.
Let's turn back to the generic block diagram of fig. 6 having
described different possibilities of realising the delay detector
unit 10.
By means of the figs 17 to 22 we will further explain with a
specific example the effect of amplitude filter unit 12 and of
controllable filter unit 14.
In fig. 17 there are shown examples of two weighing signal
characteristics of unit 12. According characteristic I each dtω
spectral line amplitude of signal A10 (see fig. 6) is attenuated
to zero, if such amplitude is below or above predetermined
values dtmin,ω, dtmas,ω and is set to be "one" if such spectral
component amplitude is between these two values.
Such selection of weighing function W results in an output signal
spectrum A12, as shown in the figures 18a and 18b.
The figures 18a and 18b are self-explanatory for the skilled
artisan.
Fig. 19 shows a spectrum example of signal S1. At the controllable
filter unit 14 all spectral lines of S1 (Fig. b) are amplified
by the value 1 according to A12 or are nullified according
to zero values of A12. This results, according to fig.
20, in a spectrum Sr as an output signal spectrum of controllable
filter unit 14 of fig. 6. If the weighing function I of
fig. 17 is applied to the technique according to fig. 7 there
results a beam form as shown in fig. 21 in strong lines. If an
amplitude filter characteristic is applied as shown by II in
fig. 17, there results the characteristic as shown in fig. 21
in dashed line.
Fig. 22 shows the resulting beam if in analogy to fig. 17 and
with an eye on figures 8 and 9 all ratio values which exceed
(Ar/A2)max are discarded. This is realised by the amplitude
filter characteristic as also indicated in Fig. 22.
In fig. 22 the ratio Ar/A2 is denoted with r(ω).
It is clear for the skilled artisan that only examples of the
invention were described with the help of the figures. For instance
more than two transducers or microphones arranged in
linear, planar or spatial array form can be used. Additionally,
directional microphones may be used instead of the omnidirectional.
Beam forming following the inventive principle can also
be made by the combination of the functions of two or more microphones.
As is perfectly clear to the skilled artisan also
the delay detector can be realised in many other ways. Further,
normalisation, which was explained with the help of normaliser
unit 30 in fig. 14, may clearly be done by providing time domain
to frequency domain conversion just after the analogue to
digital converters 16 and 17 and providing frequency-specific
arrays or tables of time delays τω.