Technical Field

The present invention relates to a filtering procedure of a kind stated in the ingress to claim 1. The invention relates even to a digital filter arrangement comprising a known time discrete filter of the transversal type, also named Finiry Impulse Response filter (FIR-fHter), of a kind as stated in the ingress to the claim 8.

Background Art

One serious drawback with present-day filter for use in sound studios is that it is not possible to create filter characteristics with total freedom. For instance, it is not possible to create filter characteristics which have the same resonances as the soundboard of a violin. Neither is it possible to obtain a filter which will produce a reverberatory effect having the sound of an actual concert hall or an echo effect having the sound of an echo in natural surrounds. There exists no filter which has the same frequency characteristics as the loudspeaker cabinet of a pleasing guitar amplifier. The possibilities of creating imaginary or fantasy effects are still limited at present.

It is previously known through EP-A-248527, page 3, that an approximation to the response from a concert hall can be recorded and the effect of the response on an input signal can accurately be simulated by affecting this input signal with the measured response, but at the same time it is pointed out in the same literature that it is a well-known fact that a device for that purpose is not practically realizable, since it needs a computer with a computing capacity that is two to three orders of mag¬ nitude greater than the real-time capacity for computers of an acceptable size and price. Therefore EP-A-248 527 suggests that, although with an inferior result, the resonance model should be smplified and to make that simplified resonance model better suited to computer calculations.

US-A-3992582 reveals a filter device aiming at giving a "Concert Hall" character to a signal by delaying the signals lower frequencies longer. With that known device, a

number of parallel connected part-filters with a large number of delay stages and whose sampling frequencies are variable, is indicated. That known filter device aims for giving the reverberation course a "softer tail", where that reverberation course is primarily determined by the choice of sampling frequencies. Furthermore, even in the known device a further connection is required (column 4, last paragraph) which gives a very limited control over the produced reverberation course.

Against that background the aim of the invention is to assign a process and a filter device that offers full freedom to create filter characteristics and that is possible to realize with today's computer technology at a cost that is low enough to make the filter arrangement acceptable to the industry.

A further purpose of the invention is to show a filter device that can be supplied with characteristics that have the same resonances as a model, such as another filter device, a particular violin sound board, a particular acoustic environment or as a preferred guitar amplifier's loudspeaker. The aim includes assigning a filter device that can be given such characteristics by feeding a set of coefficients from a data file. That data file is stored in a registering medium that either contains artificially created characteristics or a true reflexion of characteristics from a real concert hall or similar. The aim includes also that the invented procedure and the invented filter device will be able to very accurately give an input signal characteristics that conform with the set of coefficients contained in the data file. Instead of feeding the set of coefficients from a data file, the set of coefficients can be fed as output data from a program execution or another way.

As examples for artisticialry created data files one could mention nature sounds (e.g. water splash), synthetic sounds (e.g. noise generated by a synthesizer), mathemati¬ cally defined functions (e.g. a sine sweep), hand-drawn functions (e.g. with a mouse on a computer monitor), impulse response to virtual environment (e.g. a CAD- designed concert hall) and various combinations of both the artistically created and the real measured impulse response (e.g. a sound collage).

While designing data files representing the impulse response of the inventive device, the sound passing through the inventive device inherits the properties of the sound

that the data file's impulse response represents. That means for example a musical instrument with a dry and sterile sound can receive sound properties from another instrument or from a whole orchestra or from a fantasy sound. In practice, this can be most easily done using a very short piece of recording of a well sounding orchestra on a compact disc (CD), e.g. 1000 samples following one after another, as coefficients for the inventive device impulse response.

Disclosure of the invention

The above mentioned aims and other stated or implied advantages will be achieved with the procedure and the filter device stated in the enclosed claims 1 and 8. Embodiments of the above mentioned procedure and the inventive filter device are stated in the enclosed non-independent claims.

The inventive filter device is described here in the form of interconnected elements or structural blocks but the professional would realize that the inventive device can be realized in any form, e.g. as a super computer whose program would build the described structure, or as a special hardware that is optimized for this aim, or as a combination of those options described above, or as a general computer working in non-realtime.

In those embodiments of the invention where sampling converters are included, their function is to see to it that the time discrete digital signal is transmitted free of distortion. That means that the time discrete signal at the input of the transversal filter shall have a sampling frequency that the transversal filter is ready to receive. By the same token, the signal at the input of the summation circuit shall have a sampling frequency that the summation circuit can receive.

Sampling frequencies for the input signal and output signal of the part-filters can vary. For example a DAT-player sends a digital audio-signal with a sampling frequen¬ cy of 48kHz to the inventive device and the processed sound is then sent with a new sampling frequency of 44.1kHz from the inventive device's output to another DAT- player. This way the sound is processed and at the same time the sampling frequency is converted from the DAT-player's usual 48kHz to CD-player's standard 44.1kHz.

Therefore it is advisable to arrange one or two sampling converters also in the upper filter branch.

The mixing device to mix the signals from the part-filters can be a pure adder or can have the form of a mixer with several outputs. To give an example, the branches responsible for the so called early reflections are mixed to one output and the so called cluster of the reverberation is mixed to another output to achieve a deeper spatial experience with the basis of one single impulse response. Signals from the various branches are then weighted by individual factors to the various mixer out- puts. Alternatively, the part-filters' output signals can run loudspeakers, whose sounds are summarised in the listening room.

The description of the invention refers to a mono-design. In a mono-design only one single impulse response is used. With mono-design, the various branches give origin to various time sectors of the device's impulse response. These time sectors can overlap each other partly but not entirely.

In a multi-channel system a number of impulse responses is used. This can be achieved by connecting a number of devices of mono-design to an input signal, where outputs from those devices make up the various channels' output signals. For example, in a stereo system the input signal is connected to two devices which deliver the signals for the left and right channel respectively. A stereo reverberation is achieved by providing the two devices with impulse responses from the same hall that are taken between different points for each channel respectively.

Evidently, it is also possible to construct a device containing twice as many branches than what is required for a mono-design. These branches are then mixed to right and left channel respectively. In this case, some branches can have the same passage times since they can give origin to the same time sectors of the device's impulse response.

The signal can be delayed in one or more steps in one, a few or all part-filter branches. Alternatively, the first taps in the transversal filters can be omitted.

Through this technique it is possible to reduce the number of coefficients at the transversal filters and thus also reduce the cost of the device.

To achieve the desired character to the sound process, external connections can be arranged between points of free choice in the structure of the filter device. For example, a feedback can be introduced from the device's output to its input. Another, more particular way is to introduce feedback in the delay at the part-filter branch that is responsible for the last part of the filter device's impulse response. Through repeating the last part of the impulse response with a decreasing volume, the device's impulse response is extended without the impulse response's beginning being affected by the feedback. When required, the signal in this filter branch may require to be delayed in several steps.

In embodiments with sampling converters, the sampling converters for the part-filter branch that is controlled by the highest sampling frequency can be moved outside the parallel connection of the part-filter branches, whereby lower requirements can be made for the other sampling converters.

Sampling frequencies belonging to a particular part-filter branch are obtained preferably through multiplication or division of the sampling frequency for the filter device's input signal, with a constant that preferably equals to a fractional number. For example a part-filter can work with a sampling frequency that equals to V3 of the sampling frequency of the filter device's input signal.

The sampling frequencies for all part filters but one can be lower than the sampling frequency of either the filter device's input signal or output signal.

The total number of utilized taps or coefficients for all part-filter branches' filtering of transversal filter type amounts to at least 5000. Preferably, the total number of coefficients is at least 15000 and most preferably in the magnitude of 50000.

The number of part-filters is at least 2, preferably 3 and most preferably at least 4

The set of coefficients can be obtained in the following way: A file with sampled impulse response is read while the first sample is used as the first coefficient to the first part-filter, the second sample is used as the second coefficient to the first part-filter, and so on until all coefficients in the first part-filter have been defined. The following samples, read from the file, undergo a conversion of sampling fre¬ quency to second part-filter's sampling frequency after which the converted samples are assigned, in the same way as in the case of the first part-filter, to the second part-filter's coefficients. Obtaining coefficients for the Ibird and following part-fil¬ ters is analogous to obtaining of coefficients for the second part-filter.

Evidently, the procedure and the filter device according to the invention makes it possible to register the response characteristics of a real environment such as a specific concert hall with a very high accuracy, while the registering is in the form of a set of coefficients of the number mentioned above. Such a set of coefficients enables the inventive filter device to give a signal a character that is typical for the registered environment/concert hall with a very high accuracy.

Another appUcation for the filtering procedure and the filter device according to the invention is the extinction of feedback in a sound system. It is a well-known problem that the sound from the loudspeaker is acoustically fed back into the microphone. When the amplification of the sound system exceeds a certain value, the sound system starts oscillating and one or several tones can be heard from the loudspeaker. The inventive device can be adjusted to suppress the feedback sound from the loudspeaker and thus even permit a higher amplification to the sound system without a risk of feedback.

A procedure respectively a filter device according to the invention can for example be used for the following alternatives:

a) Copying the characteristics of another filter, by measuring the frequency charac¬ teristics of the filter and setting the coefficients so that the transversal filter will obtain the same frequency and phase properties as said other filter (imitation copy).

b) Executing a program for calculating coefficients which are then used in the transversal filter, thereby creating a certain sound effect.

c) Using the whole or parts of a set of coefficients taken from another system, such as A/D converter, diskette unit, CD-player, digital synthesizer, for instance.

The invention will be described in the following text in the form of embodiment example with reference to the enclosed drawing.

Figure 1 shows a filter device according to the invention.

Figure 2 shows an arrangement for copying an original filter with the filter device according to the invention.

Figure 3 shows the filter device according to the invention supplemented with a loudspeaker-microphone unit.

Figure 4 shows a device for eliminating acoustic re-coupling (feedback and ringings) with the filter device according to the invention.

Figure 5 shows filter device according to Figure 1 and a digital unit.

Figure 6 shows another realization of the filter device according to Figure 1.

Figure 7 shows a feedback in a part-filter branch included in the inventive filter device.

Figure 8 shows another variant of feedback in a part-filter branch included in the inventive filter device.

Figure 9 illustrates a summation device in the form of a mixer for the filter device's output signal.

Preferrerl ernhp rnftπts

Fig. 1 shows a filter device according to the invention, at the input I of which the signal can be regarded to be given a certain sampling frequency. Connected to the input is a number of parallel branches 11-13, that also connect to a mixing or summation device 2. Each branch 11-13 is shown to include a transversal part-filter 31-33 and a sampling converter 21-23 and 41-43 respectively is connected in each branch before and after part-filter 31-33 respectively. The filter and its converters in each branch 11-13 are controlled by a sampling frequency converter 3 which receives the sampling frequency at the input and converts it to other frequencies for branch 11-13 respectively. Converter 3 is preferably arranged to create sampling frequencies for each branch respectively that equals to a fractional number multi¬ plied by the sampling frequency of the input signal.

One, several or all transversal filters 31-33 can be realized as a number of cascade- coupled transversal filters.

Branches 11-13 connect to summation device 2 which is shown to have an output U. But it should be made clear that the summation device 2 can consist of a mixer, as Figure 9 schematically illustrates, where mixer 2' has several outputs Ul, U2. E.g. branches responsible for the so called early reflections can be mixed to an output Ul and the so called cluster of the reverberation is mixed to another output U2 to achieve a deeper spatial experience. Signals from the various branches 11-13 are weighted with individual factors or numbers to the various outputs.

With regard to Figure 1 it is understood that the sampling converters in branch 11 can be omitted if transversal filter 31 is controlled by the sampling frequency at the input of the device and summation device 2 receives the same frequency. The embodiment according to Figure 1 makes it possible to use other sampling frequen- cies for the input signal and output signal of the part-filters.

The entire impulse response from device 1 in Figure 1 is divided into a number of parts. The first part of the impulse response is initiated in the first part-filter 31, the second part is initiated in the second part-filter 32, etc. To ensure that the different

parts of the impulse response do connect to each other correctly, it is advisable to introduce delay units in the second part-filter branch and those following. However these delay units are not compulsory, since the device can work correctly without them. The purpose of delays is merely to reduce the number of taps at the filters in the lower branches of the device and thus reduce the cost of the device. The delays can be placed at any point in respective branch. The delays can even become a part of the transversal filter or the sampling converters, for example by not utilizing a number of initial taps of the part-filters.

A device in accordance with Figure 1, which is extended to four branches (where the uppermost branch works with the highest sampling frequency and those underneath work with gradually lower sampling frequencies) and that are supplemented with delay units in all branches but the uppermost one can, as a practical example, work with the following data.

branch 1 branch2 branch3 branch4

From the above data it is evident that the delay is somewhat shorter than it should have been theoretically since consideration has been taken for delays arising in the sampling converters.

The number of taps (sample values of the impulse response) for achieving a rever¬ beration effect should be chosen as follows:

1st branch: no lower limit 2nd branch: >500 remaining branches: > 1000 the sum of taps in all branches: >5000

In Figure 2, filter device 1 as per Figure 1 is shown connected to a registering device 20. An original filter 21 that is to be copied has its input connected to an output U2 of registering device 20 and its output connected to an input 12 of registering device 20.

Registering device 20 emits a measuring signal, preferably a short pulse, giving a response at the output of filter 21 (at 12). If filter 21 is excited with an impulse emitted from registering device 20, then filter 21 leaves a so called impulse response at its output. In the most simple case the curve shape of the impulse response is converted to digital sample values that are used as the impulse response values of the inventive device. Sample no. 1 will then become impulse response value no. 1, sample no.2 will be value no.2 etc.

If required, the registered sound can be processed in some way before it is used as the impulse response of the inventive device, such as being filtered, weighted with a window, time expanded or distorted. If filter 21 is excited with a measuring signal that is not equal to an impulse and if it is desirable that the inventive device emits a true image of the original filter 21, the response from filter 21 is first transformed to an impulse response before its sample values are used as the impulse response values of the inventive device.

The conversion of filter 21 response to sample values (sampling) can be done with sampling frequencies fl, £2, £3 of the various part-filter branches in order to be immediately used as the impulse response of the various part-filters. Alternatively, filter 21 response is resampled using for example a computer program for adjusting the response to the various filters' sampling frequencies.

In Figure 3, filter device 1 as per Figure 1 is shown connected to a registering device 20. Furthermore, a unit with loudspeaker 30 and microphone 31 is added, with the loudspeaker input connected possibly via an power amplifier to output U2 of the registering device and microphone 31 output connected to input 12 of the registering device.

This device can be used if one, for example, desires to capture the acoustic pattern of a concert hall. A short impulse from the registering device is emitted via loudspeaker

30 and is affected by the acoustics in the concert hall. The pulse response is captured by microphone 31, then processed in the way described earlier and results in a number of digital sample value sets representing a number of part-impulse respon¬ ses belonging to each particular part-filter branch in filter device 1.

Since reflection patterns occuring in nature are usually characterized by a course where the upper limit frequency declines with time, it is most common that the first part-filter branch, responding to the beginning of the impulse response, works with the highest sampling frequency and the following part-filter branches work with gradually lower sampling frequencies, thus fl >f2... >fn. In another case, a sound effect may be desired that is soft in the beginning and the end but markedly hard in between. Then the sampling frequency should be chosen according to pattern f 1 < £2 < f3 > f4 > f5 > f6. By choosing different sampling frequencies in the various part-filter branches a long impulse response is achieved at a low price and with a high quality at the same time.

Figure 4 shows a method for how the device according to the invention can be used for suppressing acoustic feedback in a sound system. Here, filter device 1 as per Figure 1 is used to simulate the feedback signal that is transmitted from loudspeaker 40 to microphone 41 and then subtract that artificial feedback signal (which is found at output Ul) from the real feedback signal coming from microphone 41 which is mixed with the actual sound to be amplified, for example speech or music. The subtraction is done by means of inverter 45 and balance control 42. All that remains after this subtraction is just the speech or music.

Setting of the impulse response for the device as per Figure 4 is done as follows. Registering device 20 sends a measuring signal via switch 46 in position 1 to loud¬ speaker 40. The impulse response for the whole acoustic system (loudspeaker- room-microphone) is fed from microphone 41 to input 12 of the registering device, after which the curve shape of the impulse response is converted, in the manner described earher, to the digital sample values for the impulse response which is used in filter device 1 where the characteristics of the acoustic system are reconstructed.

During the normal function of the sound system, acoustic sound from both the sound source 47 and from loudspeaker 40 is converted to an electric signal in the micro¬ phone 41. This is then fed to subtraction circuit 42, which then passes a patch point 43 and a volume control 44. The output of the volume control is connected to filter device 1 via connection II and to loudspeaker 40 via switch 46 in position 2. Filter device 1 creates a feedback signal that is, unlike the signal from the microphone, free from the direct signal of the sound source and that is inverted in inverter 45 and then used in the balancing circuit to extinct the acoustic feedback of the sound system. Inverter 45 can be omitted if the sign of the impulse response of filter device 1 is reversed.

A so called effect processor, such as a chorus machine or an echo unit, can be connected to the patch point 43 if required.

Filter device 1 and registering device 20 can use common parts, such as the same A/D and D/A converters. E.g. outputs Ul and U2 can physically be one single output.

The device as per Figure 5 with a filter device 1 as per Figure 1, has a digital unit 50, for example a mass storage, a computer, an interface or an A/D converter arranged to feed out data in the form of sample values for part-impulse responses in each particular part-filter branch of filter device 1 to this device when required. Conse¬ quently, the mentioned sample values are intended for successive control of the functions of filter device 1 as the outer signal is fed into the filter device via input terminal II.

In some context it is desirable to alter the impulse response of the filter device as time goes on for the reason of a smooth changeover from one impulse response to another or to achieve a chorus effect, pitch shift, some sort of modulation or another time variant filter effect. That is achieved by digital unit 50 transferring to filter device 1 new values for the momentary pattern of the impulse response at any point of time.

The device as per Figure 6 shows another design'of the device as per Figure 1. It is commonly known that a transversal filter can be arranged in such a way that it can leave beside the filtered (convoluted) signal itself also a purely delayed signal. This delayed signal is used in the uppermost branch of transversal filter 61 as the input signal to the second branch. In this way, the need for a separate delay unit (shown dotted in Figure 1) is elmiinated. In the same fashion, the delayed signal from the second branch transversal filter 62 is used as the input signal to the third branch. The delay in the third branch is thus distributed over several units, namely transversal filters 61 and 62.

The summation of the output signals of the part-filter branches can be distributed, too. In Figure 6 this is done in two summation devices 64 and 65.

The conversion of the sampling frequency of the time discrete signal can be also done in several steps. For example the third part-filter branch in Figure 6 uses sampling converters 21, 22 and 23 to achieve a signal with a suitable sampling frequency at the input to filter 63.

In the design as per Figure 7, a particular option for feedback in the inventive device is illustrated, which is a feedback in the delay at the branch that is responsible for the last part of the impulse response of the device. By repeating the last part of the impulse response with a decreasing strength, the impulse response of the device is extended without its beginning being affected by the feedback. This feedback can be shaped in several various ways and when required, the signal in this branch may need to be delayed in several steps.

Figure 7 shows the lowermost branch 13 (between sampling converters 23 and 43) which contains an adder 71 before transversal part-filter 63. A feedback 70 takes the signal from the delay output of transversal part-filter 63 and carries it to summation device 71 via a feedback unit 72 which, in the simplest case, can be an attenuator with gaina, 0<a< l.

Figure 8 shows another realization of the above mentioned feedback, through the introduction of an IIR (Infinity Impulse Response) filter 84 which is between

transversal filter 33 and sampling converter 43 in the branch that is responsible for the last part of the impulse response. The UR filter 84 can in fact be inserted at any point in that branch. The UR filter consists of an adder 81 and a feedback branch 80 containing an attenuator 82 and a delay 83.

Figure 9 shows mixing or summation device 2' from Figure 1, extended with a number of outputs, two in this case. The signal from the various part-filter branches is weighted with attenuators al-a3 and bl-b3 respectively and then added in adders Σ to output signals Ul and U2 respectively.