LINEAR-PHASE FIR AUDIO FILTER, PRODUCTION METHOD AND SIGNAL PROCESSOR

Abstract

The invention relates to a method for producing a linear-phase digital FIR filter (4) from two sub-filters (6a, b) for an audio signal (AU, AF), in which method the sub-filters (6a, b) are provided as sub-sets (Ta, b) having numbers (Ma, b) of coefficients, a lower cutoff frequency (Ga, b) of the particular sub-filter (6a, b) being greater than the sampling frequency of the audio signal (AU, AF) divided by the number (Ma, b), the sub-sets (Ta, b) are linearly convoluted with one another so as to form a total set (SU) having a number (L) of coefficients greater than the numbers (Ma, b), and the total set (SU) is symmetrically reduced to a number (Z) less than the number (L), so as to form a reduced total set (SK) of the filter (4). A linear-phase digital FIR filter (4) for an audio signal (AU, AF) is created by the method. A signal processor (2) for an audio signal (AU, AF) contains the filter (4).

Claims

1. A method for generating a linear-phase digital FIR filter (4) for an audio signal (AU, AF), wherein the filter (4) comprises two individual linear-phase digital FIR subfilters (6a,b) for respectively subfiltering the audio signal (AU, AF), in which: the two subfilters (6a,b) are provided as subsets (Ta,b) having a respective number (Ma,b) of coefficients, wherein a lower limit frequency (Ga,b) of the respective subfilter (6a,b) is greater than the ratio of the sampling frequency of the audio signal (AU, AF) divided by the respective number (Ma,b), the two subsets (Ta,b) are linearly convoluted with one another to form a sum set (SU) of which the number (L) of coefficients is the greater of the respective numbers (Ma,b) of the subsets (Ta,b), and the sum set (SU) is symmetrically truncated to a number (Z) which is less than the number (L), resulting in a truncated sum set (SK), in order to generate the filter (4) according to the truncated sum set (SK).

2. The method as claimed in claim 1, wherein the subsets (Ta,b) are convoluted to form the sum set (SU) by means of a direct linear convolution in the time domain, or the subsets (Ta,b) are indirectly linearly convoluted by Fourier-transforming the subsets (Ta,b) into the frequency domain, and complex-multiplying the transforms to form a sum transform, and forming the coefficients of the sum set (SU) as the real component of a subsequent inverse Fourier transform of the sum transform, or the subsets (Ta,b) are indirectly linearly convoluted by Fourier-transforming the subsets (Ta,b) into the frequency domain, and, for the case that the subfilters (6a,b) have the same number (Ma,b) of coefficients, performing scalar multiplication of the complex-valued one transform by the value of the complex-valued other transform to form a sum transform, and forming the coefficients of the sum set (SU) as the real component of a subsequent inverse Fourier transform of the sum transforms.

3. The method as claimed in claim 1, wherein the sum set (SU) is truncated via a window function by means of weighting.

4. The method as claimed in claim 3, wherein a function acting predominantly in a boundary range of a maximum of 20% is used as a window function.

5. The method as claimed in claim 1, wherein a shelving filter is used as a first or second subfilter (6a,b).

6. The method as claimed in claim 5, wherein the shelving filter has an attenuation having a constant value for frequencies below a lower limit frequency (6a,b), and has an attenuation for higher frequencies which is generally not equal to the constant value.

7. The method as claimed in claim 1, wherein a filter having a lower crossover frequency which is higher than the limit frequency (Ga,b) is used as a first or second subfilter (6a,b).

8. The method as claimed in claim 5, wherein a band filter (8a,b) or a high-pass filter is used as such a subfilter (6a,b).

9. A linear-phase digital FIR filter (4) for an audio signal (AU, AF), wherein said filter is generated via a method as claimed in claim 1.

10. A signal processor (2) for an audio signal (AU, AF), wherein said processor comprises a filter (4) as claimed in claim 9.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Additional features, effects, and advantages of the present invention will result from the following description of a preferred exemplary embodiment of the present invention and the attached figures. The following are depicted as schematic diagrams:

[0049] FIG. 1: a digital signal processor according to the present invention during operation;

[0050] FIG. 2: a flow chart for generating the filter according to the present invention from FIG. 1;

[0051] FIG. 3: coefficients of two band-pass filters;

[0052] FIG. 4: various frequency responses of subfilters and sum filters from FIG. 1;

[0053] FIG. 5: frequency responses for untruncated and truncated sum sets according to FIG. 1;

[0054] FIG. 6: phase responses of subfilters and sum filters according to FIG. 1.

DETAILED DESCRIPTION

[0055] FIG. 1 depicts a signal processor 2 during operation. Said processor comprises a linear-phase digital FIR filter 4 in the form of a coefficient set, here, a truncated sum set SK. During operation, an unfiltered audio signal AU is fed into the signal processor 2, filtered in said signal processor by the filter 4, and output as a filtered audio signal AF. The filter 4 is a combination filter and functionally comprises two series-connected individual filter functionalities in the form of linear-phase digital FIR subfilters 6a,b. The subfilter 6a is a frequency response correction and band-pass filter; the subfilter 6b is a shelving filter. Originally, i.e., with respect to their dimensioned and intended filter function, the subfilters 6a,b have respective filter characteristics according to respective full coefficient sets in the form of the subsets Ta,b. The signal processor 2 no longer comprises the subfilters 6a,b in this original form, but rather, said subfilters are already merged into the filter 4 in the form of a sum filter.

[0056] FIG. 2 depicts a flow chart of a method according to which the filter 4 has been generated from the subfilters 6a,b by means of merging.

[0057] In the method, first, the two subfilters 6a,b are provided with their original full-fledged subsets Ta,b of coefficients. The sampling frequency for the audio signal AU, AF, here, 44.1 kHz, divided by the number Ma,b of coefficients (subsets Ta,b), here, 513 in each case, is approximately 86 Hz, and is respectively less than the lower limit frequency Ga,b of the subfilters 6a,b, here, Gb =1 kHz for the shelving filter (subfilter 6b) and Ga=1.8 kHz for the band-pass filter (subfilter 6a).

[0058] In a convolution step SF, the two subsets Ta,b are now linearly convoluted with one another to form a sum set SU. Said sum set now has the length or number (L) of coefficients. In the example, a direct linear convolution of the two filter coefficient sets or subsets Ta,b takes place in the time domain. The number L results in L=number(Ta)+number(Tb)1=513+5131=1025. The filter formed by the sum set SU of coefficients (sum filter, untruncated) now merges the characteristics of the two subfilters 6a,b.

[0059] Subsequently, in a truncation step SZ, a symmetrical truncation of the sum set SU is performed, resulting in a truncated sum set SK having a number Z=513 of coefficients. This takes place by using a Tukey window function having a parameter alpha of at least () 0.9. The truncated sum set SK now describes the filter 4.

[0060] FIG. 3 shows, by way of example and for illustrating the present invention, a diagram of respectively 513 coefficients of two linear-phase FIR band-pass filters 8a,b having the limit frequencies [200 Hz, 2 kHz] for band-pass filter 8b, and [1 kHz, 10 kHz] for band-pass filter 8a. For this comparison, the coefficient sets are normalized to 1. In the case of the band-pass filter 8a having the higher limit frequencies, the values of the outer coefficients converge significantly faster to zero. Therefore, in contrast to band-pass filters having lower limit frequencies for an identical number of coefficients, their influence on the frequency response of the filter is insignificant.

[0061] For filters which were generated with the aid of the method according to FIG. 2 depicted above, frequency and phase responses are depicted in the following FIGS. 4 to 6.

[0062] FIG. 4 depicts the frequency responses (attenuation/dB over frequency/Hz) for the subfilter 6b (shelving filter) having an increase of 12 dB in the upper plateau, for the subfilter 6a (loudspeaker band filter), for the merged filter 4 (sum filter, truncated), and once again for the merged filter 4, now, however, having an attenuation of 12 dB (graphical decrease, only for illustration), as the curve 4. FIG. 4 thus depicts the frequency response of the subfilter 6a (frequency response correction and band filter), the frequency response of the subfilter 6b (shelving filter), and the frequency response of the merged filter 4 (sum filter untruncated). The curve 4 depicts the frequency response of the merged filter 4 reduced by 12 dB, i.e., by the amplification of the subfilter 6b (shelving filter) in the upper plateau. The resulting filter 4 exhibits the same profile as the original subfilter 6a (speaker band filter), with high accuracy in the plateau regions of the subfilter 6b (shelving filter)only said plateau regions can be compared by means of superimposition: the curve for the filter 4 and the curve for the subfilter 6a coincide in the lower plateau region. In the upper plateau region, the shifted or lowered curve 4 for the filter 4 and the curve for the subfilter 6a coincide.

[0063] FIG. 5 illustrates the effect of the coefficient truncation and the coefficient weighting via a Tukey window function, i.e., the effect of the method step truncation step SZ, on the frequency response, on the basis of frequency responses (attenuation/dB over frequency/Hz). FIG. 5 shows that the effects of the coefficient truncation and windowing in the resulting filter 4 (truncated: coefficient set is the sum set SK) with respect to the untruncated merged filter (coefficient set: sum set SU) are minimal. Even in the blocking regions of the resulting filter 4, only the smallest deviations are apparent at an attenuation of 90 dB. These deviations exist at low frequencies, as expected.

[0064] FIG. 6 depicts the phase responses, (phase/degree over frequency/Hz) of the subfilter 6b (shelving filter), the subfilter 6a (loudspeaker filter), and the merged filter 4 (sum filter, truncated).

[0065] The phase response of the merged filter 4 also corresponds highly precisely to that of the original subfilter 6a (loudspeaker band filter). The phase linearity of the original (sub)filter is thus maintained.