Device and method for processing a real subband signal for reducing aliasing effects

Abstract

In order to process a subband signal of a plurality of real subband signals which are a representation of a real discrete-time signal generated by an analysis filter bank, a weighter for weighting a subband signal by a weighting factor determined for the subband signal is provided to obtain a weighted subband signal. In addition, a correction term is calculated by a correction term determiner, the correction term determiner being implemented to calculate the correction term using at least one other subband signal and using another weighting factor provided for the other subband signal, the two weighting factors differing. The correction term is then combined with the weighted subband signal to obtain a corrected subband signal, resulting in reduced aliasing, even if subband signals are weighted to a different extent.

Claims

1. An audio processor for processing an audio signal, the audio signal comprising a plurality of real audio subband signals, the plurality of real audio subband signals being a representation of a real discrete-time audio signal, and the plurality of real audio subband signals being generated and output by an analysis filter bank, comprising: a weighter configured for weighting a real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank by a weighting factor determined for the real audio subband signal generated and output by the analysis filter bank to obtain a real weighted audio subband signal; a correction term determiner configured for calculating a real correction term, the correction term determiner being implemented to calculate the real correction term using at least one other real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank and using an other weighting factor provided for the at least one other real audio subband signal generated and output by the analysis filter bank, the other weighting factor differing from the weighting factor; and a combiner configured for combining the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal as a processed audio signal, wherein at least one of the weighter, the correction term determiner, and the combiner are embodied in hardware.

2. The audio processor according to claim 1, wherein the correction term determiner is implemented to calculate a difference between the weighting factor for the real audio subband signal generated and output by the analysis filter bank and the other weighting factor for the at least one other real audio subband signal generated and output by an analysis filter bank, and to generate the real correction term in dependence on the difference.

3. The audio processor according to claim 1, wherein the correction term determiner is implemented to calculate the real correction term in dependence on the real audio subband signal, and in dependence on the at least one other real audio subband signal and the other weighting factor.

4. The audio processor according to claim 1, wherein the correction term determiner is implemented to use as the at least one other real audio subband signal a specific real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank, the specific real audio subband signal comprising a frequency range index differing from a frequency range index of the real audio subband signal by 1.

5. The audio processor according to claim 1, wherein the correction term determiner is implemented to determine a further real correction term dependent on a third real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank and a third weighting factor associated to the third real audio subband signal, the third weighting factor differing from the weighting factor.

6. The audio processor according to claim 5, wherein the combiner is implemented to combine the further real correction term and the real weighted audio subband signal and the real correction term.

7. The audio processor according to claim 5, wherein the correction term determiner is implemented to use as the third real audio subband signal of the plurality of real audio subband signals a specific real audio subband signal generated and output by the analysis filter bank, the specific real audio subband signal comprising a frequency range index differing from a frequency range index of the real audio subband signal and from a frequency range index of the at least one other real audio subband signal.

8. The audio processor according to claim 1, further comprising the analysis filter bank configured for filtering the real discrete-time audio signal and for outputting the plurality of real audio subband signals, wherein the analysis filter bank is implemented by a transform of a block of samples of the real discrete-time audio signal to a spectral representation of the block of samples of the real discrete-time audio signal, wherein each real audio subband signal of the plurality of real audio subband signals comprises subband samples comprising spectral coefficients of the same frequency index from a sequence of successive spectral representations.

9. The audio processor according to claim 1, further comprising the analysis filter bank configured for filtering the real discrete-time audio signal to for output the plurality of real audio subband signals, wherein the analysis filter bank is a decimated filter bank comprising filters defined by a modulation of a prototype filter.

10. The audio processor according to claim 8, wherein each real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank is a signal comprising several samples, wherein N/M samples are generated for each signal from a quantity of N values of the discrete-time signal, and wherein M is a number of real audio subband signals of the plurality of real audio subband signals generated and output by the analysis filter bank.

11. The audio processor according to claim 1, wherein the combiner is implemented to perform an addition of the real weighted audio subband signal and the real correction term.

12. The audio processor according to claim 1, further comprising: a provider for providing the different weighting factors associated to the real audio subband signal and to the at least one other real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank, wherein the provider is implemented to determine the different weighting factors due to an equalizer function or an echo suppression function or a bandwidth extension function or a parametric multi-channel encoding function for the real discrete-time audio signal.

13. A method for processing an audio signal, the audio signal comprising a plurality of real audio subband signals, the plurality of real audio subband signals being a representation of a real discrete-time audio signal, and the plurality of real audio subband signals being generated and output by an analysis filter bank, the method comprising: weighting, by a weighter, a real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank by a weighting factor determined for the real audio subband signal generated and output by the analysis filter bank to obtain a real weighted audio subband signal; calculating, by a calculator, a real correction term using at least one other real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank and using an other weighting factor provided for the at least one other real audio subband signal generated and output by the analysis filter bank, the other weighting factor differing from the weighting factor; and combining, by a combiner, the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal as a processed audio signal, wherein at least one of the weighter, the correction term determiner, and the combiner are embodied in hardware.

14. An analysis filter bank processor for processing a real discrete-time audio signal, the analysis filter bank processor comprising: a filter bank configured for generating and outputting a plurality of real audio subband signals from the discrete-time signal; and a device configured for processing the plurality of real audio subband signals generated and output by the filter bank to obtain a plurality of corrected real audio subband signals, the device comprising, for each real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank: a weighter configured for weighting the real audio subband signal by a weighting factor determined for the real audio subband signal generated and output by the analysis filter bank to obtain a real weighted audio subband signal; a correction term determiner configured for calculating a real correction term, the correction term determiner being implemented to calculate the real correction term using at least one other real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank and using an other weighting factor provided for the at least one other real audio subband signal generated and output by the analysis filter bank, the other weighting factor differing from the weighting factor; and a combiner configured for combining the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal of the plurality of corrected real audio subband signals, wherein at least one of the filter bank, the device for processing, the weighter, the correction term determiner, and the combiner are embodied in hardware.

15. A synthesis filter bank processor for generating a real discrete-time audio signal, the synthesis filter bank processor comprising: for each real audio subband signal of a plurality of real audio subband signals a device for processing the real audio subband signal of the plurality of real audio subband signals to obtain processed real audio subband signals, the device comprising: a weighter configured for weighting the real audio subband signal by a weighting factor determined for the real audio subband signal to obtain a real weighted audio subband signal; a correction term determiner configured for calculating a real correction term, the correction term determiner being implemented to calculate the real correction term using at least one other real audio subband signal of the plurality of real audio subband signals and using an other weighting factor provided for the at least one other real audio subband signal of the plurality of real audio subband signals, the other weighting factor differing from the weighting factor; and a combiner configured for combining the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal, whereby processed real audio subband signals are obtained; a plurality of synthesis filters configured for filtering the processed real audio subband signals to obtain synthesis-filtered audio subband signals; and an adder configured for summing the synthesis-filtered audio subband signals to obtain a discrete-time audio signal, wherein at least one of the device for processing, the weighter, the correction term determiner, the combiner, a synthesis filter of the plurality of synthesis filters, and the adder are embodied in hardware.

16. A method for analysis processing a real discrete-time audio signal, comprising: generating and outputting a plurality of real audio subband signals from the discrete-time signal using an analysis filter bank; and processing, by a device for processing, the plurality of real audio subband signals generated and output by the analysis filter bank to obtain a plurality of corrected real audio subband signals, the processing comprising, for each real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank: weighting, by a weighter, a real audio subband signal by a weighting factor determined for the real audio subband signal to obtain a real weighted audio subband signal; calculating, by a correction term determiner, a real correction term using at least one other real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank and using an other weighting factor provided for the at least one other real audio subband signal generated and output by the analysis filter bank, the other weighting factor differing from the weighting factor; and combining, by a combiner, the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal of the plurality of corrected real audio subband signals, wherein at least one of the analysis filter bank, the device for processing, the weighter, the correction term determiner, and the combiner are embodied in hardware.

17. A method of synthesis processing for generating a real discrete-time audio signal, the method of synthesis processing comprising: for each real audio subband signal of a plurality of real audio subband signals, processing the real audio subband signal of the plurality of real audio subband signals to obtain real processed audio subband signals, the processing comprising: weighting, by a weighter, the real audio subband signal by a real weighting factor determined for the real audio subband signal to obtain a real weighted audio subband signal; calculating, by a correction term determiner, a real correction term using at least one other real audio subband signal of the plurality of real audio subband signals and using an other weighting factor provided for the at least one other real audio subband signal of the plurality of real audio subband signals, the other weighting factor differing from the weighting factor; and combining, by a combiner, the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal, whereby the processed real audio subband signals are obtained; filtering, by a plurality of synthesis filters, the processed real audio subband signals to obtain synthesis-filtered audio subband signals; and summing, by an adder, the synthesis-filtered audio subband signals to obtain a discrete-time audio signal, wherein at least one of the device for processing, the weighter, the correction term determiner, the combiner, a synthesis filter of the plurality of synthesis filters, and the adder are embodied in hardware.

18. A non-transitory storage medium having stored thereon a computer program comprising a program code for performing, when the computer program runs on a computer, the method for processing an audio signal, the audio signal comprising a plurality of real audio subband signals, the plurality of real audio subband signals being a representation of a real discrete-time audio signal, and the plurality of real audio subband signals being generated and output by an analysis filter bank, the method comprising: weighting a real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank by a weighting factor determined for the real audio subband signal generated and output by the analysis filter bank to obtain a real weighted audio subband signal; calculating a real correction term using at least one other real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank and using an other weighting factor provided for the at least one other real audio subband signal generated and output by the analysis filter bank, the other weighting factor differing from the weighting factor; and combining the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal.

19. A non-transitory storage medium having stored thereon a computer program comprising a program code for performing a method for analysis processing a real discrete-time audio signal, comprising: generating a plurality of real audio subband signals from the discrete-time signal using a filter bank; and processing the plurality of real audio subband signals generated and output by the analysis filter bank to obtain a plurality of corrected real audio subband signals, the processing comprising, for each real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank: weighting a real audio subband signal by a weighting factor determined for the real audio subband signal to obtain a real weighted audio subband signal; calculating a real correction term using at least one other real audio subband signal of the plurality of real audio subband signals generated and output by the analysis filter bank and using an other weighting factor provided for the at least one other real audio subband signal generated and output by the analysis filter bank, the other weighting factor differing from the weighting factor; and combining the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal of the plurality of corrected real audio subband signals.

20. A non-transitory storage medium having stored thereon a computer program comprising a program code for performing a method of synthesis processing for generating a real discrete-time audio signal, the method of synthesis processing comprising: for each real audio subband signal of a plurality of real audio subband signals, processing the real audio subband signal of the plurality of real audio subband signals to obtain real processed audio subband signals, the processing comprising: weighting the real audio subband signal by a real weighting factor determined for the real audio subband signal to obtain a real weighted audio subband signal; calculating a real correction term using at least one other real audio subband signal of the plurality of real audio subband signals and using an other weighting factor provided for the at least one other real audio subband signal of the plurality of real audio subband signals, the other weighting factor differing from the weighting factor; and combining the real weighted audio subband signal and the real correction term to obtain a corrected real audio subband signal, whereby the processed real audio subband signals are obtained; synthesis filtering the processed real audio subband signals to obtain synthesis-filtered audio subband signals; and summing the synthesis-filtered audio subband signals to obtain a discrete-time audio signal.

21. The audio processor according to claim 1, wherein the correction term determiner is configured for calculating the real correction term such that the corrected real audio subband signal has a reduced aliasing when being combined with the real correction term.

22. The audio processor according to claim 1, wherein the real audio subband signal has a sequence of samples, wherein the weighter is configured for weighting each sample of the sequence of samples by the weighting factor determined for the real audio subband signal to obtain the real weighted audio subband signal, and wherein the correction term determiner is configured by filtering using a filter having more than 8 filter taps.

23. The audio processor according to claim 1, wherein the weighter is configured for weighting a further real audio subband signal neighboring the real audio subband signal by a further weighting factor, the further weighting factor being different from the weighting factor; wherein the correction term determiner is implemented to generate the real correction term in dependence on a difference of the weighting factor of the real audio subband signal and the further weighting factors for the further real audio subband signal from the plurality of real audio subband signals, and wherein the further weighting factor is a further amplification factor provided and applied for all samples of the further subband signal and wherein the weighting factor is an amplification factor provided and applied for all samples of the real audio subband signal.

24. The audio processor of claim 1, being implemented in at least one of a consumer electronics device, a communications electronics device, an audio equalizer, an audio echo suppressor, an audio encoder, an audio decoder, and an audio post-processor.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Embodiments of the present invention will be detailed subsequently referring to the appended drawings, in which:

(2) FIG. 1 is a block circuit diagram of an inventive device for processing a real subband signal according to an advantageous embodiment of the present invention;

(3) FIG. 2 is a detailed illustration of the correction term determiner of FIG. 1;

(4) FIG. 3a is a schematic illustration of the inventive device according to an advantageous embodiment of the present invention;

(5) FIG. 3b is a more detailed illustration of the filter part of FIG. 3a;

(6) FIG. 3c is a schematic illustration of the inventive device according to an alternative embodiment of the present invention;

(7) FIG. 3d is a more detailed illustration of the device schematically shown in FIG. 3c;

(8) FIG. 4 shows an analysis filter bank/synthesis filter bank device having a subband-wise device for processing;

(9) FIG. 5 shows a real-value analysis/synthesis filter bank device having an equalizer stage;

(10) FIG. 6 shows cascading of a real synthesis filter bank with a complex analysis filter bank and a complex synthesis filter bank;

(11) FIG. 7 is a schematic illustration of multiband filtering;

(12) FIG. 8 is a more detailed illustration of filter operations for the multiband filtering of FIG. 7;

(13) FIG. 9 is a tabular illustration of the filters for subband signals having even and odd indices;

(14) FIG. 10 shows an exemplary comparison of magnitude frequency responses for filters for determining a correction term;

(15) FIG. 11 shows subband filtering of an impulse;

(16) FIG. 12 shows subband filtering of a sine tone at 1% above the band limit;

(17) FIG. 13 shows subband filtering of a sine tone at 5% above the band limit;

(18) FIG. 14 shows subband filtering of a sine tone at 10% above the band limit;

(19) FIG. 15 shows subband filtering of a sine tone at 20% above the band limit;

(20) FIG. 16 shows subband filtering of a sine tone at 30% above the band limit;

(21) FIG. 17 shows subband filtering of a sine tone at 40% above the band limit; and

(22) FIG. 18 is a schematic illustration of the aliasing reduction for an MDCT with a sine tone at 10% above the band limit.

DETAILED DESCRIPTION OF THE INVENTION

(23) FIG. 1 shows an inventive device for processing a real subband signal x(k) of a plurality of real subband signals which are an illustration of a real discrete-time signal x(n) generated by an analysis filter bank (50 in FIG. 5). The inventive device includes a weighter 10 for weighting the subband signal x.sub.k by a weighting factor c.sub.k determined for the subband signal to obtain a weighted subband signal 11. The weighter is advantageously implemented to perform a multiplication. In particular, subband samples which are samples of a bandpass signal or spectral coefficients of a transform spectrum are multiplied by the correction factor. Alternatively, instead of multiplication, addition of logarithm values may also be performed, namely an addition of the logarithm of the correction value and the logarithm of the subband sample x.sub.k.

(24) The inventive device for processing further includes a correction term determiner for calculating a correction term, the correction term determiner being implemented to calculate the correction term using at least another subband signal x.sub.i and using another weighting factor c.sub.i which is provided for the other subband signal, the other weighting factor differing from the weighting factor c.sub.k. This differentiation of the two weighting factors is the cause for aliasing in a real filter bank application, even when analysis and synthesis filters have a perfectly reconstructing characteristic. The correction term at the output of the means 12 is fed to a combiner 13 as is the weighted subband signal, the combiner being implemented to combine the weighted subband signal and the correction term to obtain a corrected subband signal y.sub.k.

(25) The combiner 13 is advantageously implemented to perform a combination sample by sample. Thus, there is a correction term sample for every sample of the weighted subband signal x.sub.k such that a 1:1 correction can be performed. Alternatively, however, for implementations less complicated as to calculating, a correction can be performed such that, for example, a single correction term is calculated for a certain number of weighted subband samples, which is then added in a smoothed or an unsmoothed manner to every sample of the group of samples associated to the correction term sample. Depending on the implementation, the correction term can also be calculated as a factor and not as an additive term. The combiner would in this case perform a multiplication of a correction term by the weighted subband signal to obtain a corrected subband signal y.sub.k.

(26) It is to be pointed out that aliasing occurs when two subband signals have been generated by filters having overlapping pass characteristics. In special filter bank implementations, there are such overlapping filter characteristics comprising an overlapping region which is significant for neighboring subband signals.

(27) Advantageously, the correction term determiner is thus implemented as is illustrated in FIG. 2. The correction term determiner includes a first correction term determiner portion 12a and a second correction term determiner portion 12b. The first correction term determiner portion considers the overlapping of the current subband signal of the index k and the next higher subband signal of the index k+1. Additionally, the correction term determiner portion 12a, apart from the subband signal x.sub.k+1, also receives the weighting factor c.sub.k+1 of the higher subband signal. Advantageously, the correction term determiner will also receive the difference of c.sub.k+1 and c.sub.k, which in FIG. 2 is illustrated by q.sub.k.

(28) The second correction term determiner portion 12b considers the overlapping of the subband signal x.sub.k with the subband signal x.sub.k1 lower by 1 with regard to its index. The correction term determiner portion 12b thus, apart from the subband signal x.sub.k1, also receives the weighting factor c.sub.k1 for this subband and advantageously the difference of the weighting factor c.sub.k1 and the weighting factor c.sub.k, which in FIG. 2 is referred to by c.sub.k.

(29) On the output side, the first correction term determiner portion 12a provides a first correction term q.sub.ku.sub.k and the second correction term determiner portion 12b provides a second correction term p.sub.kL.sub.k, wherein these two correction terms are added to then be combined with the weighted subband signal c.sub.kx.sub.k, as will be described referring to FIGS. 3a and 3b.

(30) An advantageous implementation which is shown in greater detail in FIGS. 8 and 3a will be detailed subsequently.

(31) The series connection of a real-value synthesis filter bank and a complex-value analysis filter bank is approximated in multiband filtering. Here, an imaginary part for each real subband sample is formed by overlapping three filter output signals. The three respective filters are applied in the respective subband and in the two neighboring bands.

(32) Correspondingly, the conversion from complex to real (c2r) approximates the series connection of a complex-value synthesis filter bank and a real-value analysis filter bank. Here, the real part is formed as a mean value of the original real subband sample and the overlapping of three filter output signals. The three respective filters are applied to the imaginary parts in the respective subband and the two neighboring bands.

(33) The series connection of r2c and c2r has to reconstruct the original subband signal as precisely as possible in order to avoid audible interferences in the output signal. Thus, the corresponding filters may have relatively great lengths.

(34) The approach presented here is based on the idea of subdividing the series connection of r2c, gain control and c2r into signal portions forming when using equal amplification factors, and signal portions forming due to differences between amplification factors of neighboring subbands.

(35) Since the first signal portion is to correspond to the original subband signal, the respective operation can be omitted.

(36) The remaining signal portions are dependent on the differences of the respective amplification factors and only serve the reduction of aliasing components, as would take place in the usual r2c and c2r conversion. Since the respective filters do not influence the reconstruction of unchanged subband signals, they may comprise considerably shorter lengths.

(37) Subsequently, the procedure will be described in greater detail.

(38) The imaginary part in the subband k is calculated from the real-value subband samples of the subbands k, k1 and k+1 to form:
I.sub.k(z)=H.sub.m(z)X.sub.k(z)+H.sub.u(z)X.sub.k1(z)+H.sub.l(z)+X.sub.k+1(z).

(39) The differentiations between H and H may be useful due to the mirroring of the subbands having odd indices.

(40) If every subband is multiplied each by an amplification factor c.sub.k, the result for the reconstructed signal in the subband k, considering an additional normalization factor of 0.5, will be:
Y.sub.k(z)=0.5(c.sub.kX.sub.k(z)+c.sub.kG.sub.m(z)I.sub.k(z)+c.sub.k1G.sub.u(z)I.sub.k1(z)+c.sub.k+1G.sub.l(z)I.sub.k+1(z))

(41) If c.sub.k1 is replaced by c.sub.k+p.sub.k, with p.sub.k=c.sub.k1c.sub.k, and if c.sub.k+1 is replaced by c.sub.k+q.sub.k, with q.sub.k=c.sub.k+1c.sub.k, the result will be:
Y.sub.k(z)=0.5c.sub.k(X.sub.k(z)+G.sub.m(z)I.sub.k(z)+G.sub.u(z)I.sub.k1(z)+G.sub.l(z)I.sub.k+1(z))+0.5(p.sub.kG.sub.u(z)I.sub.k1(z)+q.sub.kG.sub.l(z)I.sub.k+1(z)).(2)

(42) Here, the first term corresponds to the subband signal which is reconstructed when using the same amplification factors in all subbands, and thus equals the original subband signal except for the factor c.sub.k, and/or should be like that. However, the second term represents the influence of different amplification factors and can be considered as a correction term for the subband k of the complex processing compared to the real processing. It is calculated as follows:
C.sub.k(z)=0.5p.sub.kG.sub.u(z)(H.sub.m(z)X.sub.k1(z)+H.sub.u(z)X.sub.k2(z)+H.sub.l(z)X.sub.k(z))+0.5q.sub.kG.sub.l(H.sub.m(z)X.sub.k+1(z)+H.sub.u(z)X.sub.k(z)+H.sub.l(z)X.sub.k+2(z)).(3)

(43) The following connections result from the characteristics of the polyphase filter bank and the mirroring of the subbands having odd indices:
H.sub.m(z)=H.sub.m(z),H.sub.l(z)=H.sub.u(z),H.sub.u(z)=H.sub.l(z),
G.sub.m(z)=H.sub.m(z),G.sub.u(z)=H.sub.l(z),G.sub.l(z)=H.sub.u(z),
G.sub.u(z)=H.sub.l(z)H.sub.u(z)0,G.sub.l(z)H.sub.l(z)=H.sub.u(z)H.sub.l(z)0.(4)

(44) A substitution has the following result:

(45) $\begin{array}{cc}{C}_k\left(z\right)=0,5p_k{H}_l\left(z\right)\left(H_l\left(z\right)X_k\left(z\right)-H_m\left(z\right)X_{k-1}\left(z\right)\right)+0,5q_k{H}_u\left(z\right)\left(H_u\left(z\right)X_k\left(z\right)-H_m\left(z\right)X_{k+1}\left(z\right)\right).& \left(5\right)\end{array}$

(46) Since the reconstruction is no longer dependent on the filters used with amplification factors constant over the subbands, they can be replaced by shorter ones, wherein a respective product filter can also be approximated so that two correction terms can then be calculated instead of the imaginary part:
L.sub.k(z)=0.5(H.sub.ll(z)X.sub.k(z)H.sub.lm(z)X.sub.k1(z))
U.sub.k(z)=0.5(H.sub.uu(z)X.sub.k(z)H.sub.um(z)X.sub.k+1(z))
with H.sub.ll(z)H.sub.l.sup.2(z),H.sub.lm(z)H.sub.l(z)H.sub.m(z),H.sub.uu(z)H.sub.u.sup.2(z),H.sub.um(z)H.sub.u(z)H.sub.m(z).(6)

(47) The desired subband signal including aliasing compensation is obtained by a weighted overlapping of the original subband signal and the two correction signals:
Y.sub.k(z)=c.sub.kX.sub.k(z)+p.sub.kL.sub.k(z)+q.sub.kU.sub.k(z).(7)

(48) However, in practical realizations it may be kept in mind that a delay compensating the delay in the respective signal paths including filtering has to be introduced into the signal paths without filtering.

(49) For checking the overall performance, the following pictures show the output signals after filter bank analysis, attenuation of a subband by 20 dB and subsequent filter bank synthesis for different input signals.

(50) The approach described may also be combined with the MDCT instead of the filter bank used in EBCC.

(51) Suitable filter coefficients for filters having a length of 5 have been generated for this. This corresponds to the uncut filters resulting when sequentially applying the corresponding transforms and/or re-transforms. Compared to the r2cc2r technology, however, the new method is of advantage in that it does not generate approximation errors as long as the MDCT spectrum remains unchanged. With r2cc2r, however, errors would result since only the two respective neighboring bands are considered in the approximation.

(52) The resulting signal spectra for a sine tone which is 10% above a band limit show that the aliasing components are also reduced very efficiently in connection with MDCT. Here, too, the neighboring band has been attenuated by 10 dB.

(53) Thus, the equalizer functions and/or echo suppression methods can be integrated directly in an audio decoder, such as, for example, MPEG-AAC, by the inverse MDCT before re-transforming.

(54) FIG. 8 shows a schematic illustration of the filter operations for real to complex (r2c) and complex to real (c2r). The imaginary component I.sub.k of the band x.sub.k is generated by a subband signal x.sub.k1 filtered by the filter H.sub.u and by the subband signal of subband x.sub.k+1 generated by the filter H.sub.l. In addition, a component of the subband signal x.sub.k filtered by the filter H.sub.m contributes to the imaginary component I.sub.k. Because the portion of the subband signal x.sub.k1 overlapped by the filter k has a low-pass characteristic, the filter H.sub.u is a low-pass filter. In analogy, the portion of the upper subband signal x.sub.k+1 overlapped by the filter for x.sub.k is a high-pass signal, so that H.sub.1 is a high-pass filter. As has already been explained, H and H are differentiated to consider the mirroring of the subbands having odd indices. This inflection of H and H is illustrated in FIG. 9 for the imaginary parts of the subbands I.sub.k+2 to I.sub.k2. In addition, the index m stands for mid and refers to the contribution of the center subband signal. Furthermore, the index l stands for low and considers the contribution of the lower subband shown in FIG. 8 to the current subband, i.e. the subband having an index lower by 1. In analogy, u stand for up and refers to the contribution of the subband illustrated at the top in FIG. 8 for the current subband, i.e. to the subbands having an index higher by 1.

(55) The synthesis filters G corresponding to the individual analysis filters H are illustrated in FIG. 8. G.sub.l has a high-pass characteristic, whereas G.sub.u has a low-pass characteristic. Thus, as has been described before, the product of G.sub.u and H.sub.u is the same as the product of H.sub.l and H.sub.u or the product of G.sub.l and H.sub.l is the same as the product of H.sub.u and H.sub.l and nearly equals 0, since here a respective high-pass filter is multiplied by a low-pass filter and the resulting frequency response of a high-pass filter and a low-pass filter having similar cutoff frequencies equals 0 and/or approximates 0. Even for cases in which the cutoff frequencies are not identical but are apart, the resulting frequency response equals 0. If the cutoff frequency of the low-pass filter is smaller than the cutoff frequency of the high-pass filter, the resulting frequency response will also equal 0. Only in the case in which the cutoff frequency of the low-pass filter is higher than the cutoff frequency of the high-pass filter, the approximation given above would not hold true. However, such a situation does not occur in typical polyphase filter banks and/or would, if occurring, only result in slight interferences which would result in a somewhat more imprecise correction term. Due to the fact that the correction term is advantageously weighted by the difference of the two weighting factors concerned, this error would also decrease with a decreasing difference.

(56) FIG. 3 shows a schematic illustration of the advantageous filters derived above which are implemented by the inventive correction term determiner 12. It becomes obvious from FIG. 3a that the entire device includes a filter part 30 and a weighting part 31. The weighter 10 of FIG. 1 symbolized in the weighting part 31 of FIG. 3a by c.sub.k is in the weighting part 31. The combiner 13 of FIG. 1 corresponds to the adder 13 in FIG. 3a. The correction term determiner 12 includes the filter actions with the four filters H.sub.lm, H.sub.ll, H.sub.uu and H.sub.um. In addition, the correction term determiner also includes the weighting of the unweighted correction terms L.sub.k and U.sub.k by the difference of the respective two weighting factors concerned, i.e. by q.sub.k and p.sub.k, respectively, as is indicated in the weighting part 31. A more detailed implementation of the filter part of FIG. 3a is illustrated in FIG. 3b. The subband signal x.sub.k1 is fed to the low-pass filter H.sub.lm 32. In addition, the subband signal x.sub.k is fed to the low-pass filter H.sub.ll 33. In addition, the subband signal x.sub.k is fed to the high-pass filter H.sub.uu 34 and in addition the next subband signal x.sub.k+1 is fed to the filter H.sub.um 35 which may also be implemented as a high-pass filter. The output signals of filters 32 and 33 are combined in an adder 34 and represent a first unweighted correction term l.sub.k. Additionally, the output signals of the filters 34 and 35 are added in an adder 35 and represent a second unweighted correction term u.sub.k. In addition, the delay of the filters occurring when the filters are implemented as digital filters, i.e. FIR or IIR filters, is considered for the subband signal x.sub.k which is weighted by the weighting factor c.sub.k provided for this subband signal. This considering of delays of the filters 33 to 35 takes place in a delay stage 38 and can occur before or after weighting. In order for such an implementation to achieve maximum quality, it is advantageous for all the filter lengths 32, 33, 34, 35 to be identical and for the delay 38 to be adjusted to the filter length of the filters 32 to 35. If, for example, the filters 32 to 35 each have a filter length of 11, the delay 38 has to provide a delay magnitude of five subband signal samples.

(57) While FIGS. 3a and 3b illustrate the situation in which the filters 32, 33, 34, 35 are represented as product filters, i.e. as filters for calculating the terms l.sub.k, u.sub.k which then only have to be weighted, FIGS. 3c and 3d show an embodiment of an implementation of the present invention where the correction term is not calculated by 4 product filters, but all in all 6 individual filters 320, 330, 340, 350, 381, 382.

(58) As is illustrated in particular in FIG. 3c, the signal L.sub.k is calculated by filtering X.sub.k1 by the filter H.sub.m and by adding the filtered signal X.sub.k having been filtered by H.sub.l. Again, the normalization factor of 0.5 has been introduced. This normalization factor, however, can be omitted, as is the case in the first embodiment, or be set to a different value, including 1. Furthermore, the other component U.sub.k is calculated by filtering X.sub.k by H.sub.u, wherein X.sub.k+1.Math.H.sub.m is subtracted from X.sub.k.Math.H.sub.u. In contrast to the equations shown under FIG. 3a, where the products have already been considered in the filters, the signals in FIG. 3c are filtered individually. The results L.sub.k and U.sub.k will then, as is shown in FIG. 3a, be weighted by p.sub.k and q.sub.h, respectively. In addition to this weighting, filtering by H.sub.l and H.sub.u is performed.

(59) In contrast to FIG. 3a, there is a first filter part and additionally a second filter part which may be integrated and/or combined with the weighting part. The weighting factors can thus already be considered in the filter coefficients or can be applied separately before or after filtering by the digital filter H.sub.l and/or H.sub.u. Thus, the delays z.sup.d consider the delay caused by the filtering in the first filter part of the two components X.sub.k1 and/or X.sub.k+1 and additionally consider the delays in the second filter part caused by the filtering of L.sub.k and/or U.sub.k which are filtered by the filters H.sub.l and/or H.sub.u.

(60) Although, depending on the filter bank implemented, any filter characteristics can be used for the filters H.sub.n, H.sub.l, H.sub.u, it is advantageous to use a low-pass filter for H.sub.l, it is advantageous to use a high-pass filter for H.sub.u and/or it is also advantageous to use a bandpass filter for H.sub.m. The filter H.sub.l has a similar form as FIG. 10, since H.sub.ll 100 in FIG. 10 equals the square of the filter H.sub.l. The filter H.sub.u implemented as a high-pass filter results by mirroring the left part in FIG. 10 at a vertical axis at /2, i.e. about in the center of FIG. 10. The filter H.sub.lm which is no longer there in FIG. 3c since it is a product filter of a bandpass filter and a low-pass filter, could be mirrored at the line at /2 to obtain the filter H.sub.um 35 in FIG. 3b, although this product filter in an assembled form is no longer there in FIG. 3c, but is first calculated implicitly before then combining the components by the combiner 13.

(61) Whereas in FIG. 3b the correction term determiner 12 of FIG. 3a is implemented by the filter part 30 and the weighting of the components L.sub.k, U.sub.k by the weighting factors p.sub.k and q.sub.k, the correction term determination according to FIGS. 3c and 3d takes place in a kind of double filter stage, wherein at first the signals L.sub.k, U.sub.k at the output of the summers 360 and/or 370 are calculated, not using product filters, but using the individual filters, wherein in the second filter part the weighting by p.sub.k and/or q.sub.k is then performed with subsequent individual filtering.

(62) The weighting of the subband signal X.sub.k by the weighter 10, however, takes place in FIG. 3d like in FIG. 3a.

(63) In the embodiment shown in FIG. 3c and FIG. 3d or put generally, two filters are not united to form a product filter. Instead, they are implemented as individual filters. Even if there is no uniting in the product filter, there isapart from the implementationstill the advantage of shortened filter lengths. Thus, the delay compared to a direct recalculation from real to complex and/or complex to real is reduced. The swung dash above the filters in blocks 320, 330, 340, 350, 381, 382 means that the filters, as are schematically indicated in FIG. 10 for the product filters, are reduced in their filter lengths compared to a subband filter of a normal filter bank. It is advantageous to use filter lengths which are smaller than a filter length of a subband filter to generate the subband signals x.sub.k1, x.sub.k and/or x.sub.k+1. In addition, it is advantageous for the filter lengths of the filters h.sub.u, h.sub.m, h.sub.l after the approximation, i.e. after shortening, to be, like in the other case, at most 50% of the length of a filter which has been used to generate a subband signal by applying several such filters in a subband filter bank.

(64) Advantageously, filter lengths of <21 are advantageous, wherein the delay of such filters is <10. The implementation shown in FIG. 3d provides, in comparison to the implementation shown in FIGS. 3a and b, advantages with quickly time-variable attenuation factors. With regard to the time form, the implementation shown in FIG. 3d is more similar to the real/complex-complex/real implementation, whereas in the product filter realization, no more filtering takes place after applying the amplification factors.

(65) Irrespective of whether a realization with individual shortened filters or a summary with product filters is chosen, aliasing-reduced quick real filter banks are implemented according to the invention. In specially advantageous embodiments, the filter lengths in FIG. 3d are even reduced compared to the filter lengths in FIG. 3b in that the entire calculation in FIG. 3d has a similar delay as the entire calculation in FIG. 3b. An implementation similar to FIG. 3b would then in FIG. 3d be for the filters in the first filter part to have a filter length of 7 coefficients, which would correspond to a delay magnitude of 3 samples of a subband signal. In this case, the second delay 383 and/or the subsequent filters 381, 382, for example, would have a filter length of 4 to implement a delay of 2. It is pointed out here that somewhat longer or somewhat shorter filters and/or an implementation in FIG. 3d will also bring about advantages when the overall delay is somewhat greater than the product filter delay of FIG. 3b.

(66) FIG. 4 shows the usage of the device for weighting described in FIGS. 1 to 3d in an analysis filter band and/or synthesis filter bank. It becomes obvious from FIG. 4 that for every filter channel 0 to N1, one device as shown in FIG. 1 may be used. Advantageously, however, every device for processing has, when implemented like in FIG. 3b, the same four filters 32 to 35 so that only the same four filters have to be calculated and/or optimized irrespective of the number of subband signals and/or filter channels of an analysis/synthesis filter bank.

(67) The actual calculation of the filters may be performed either by a direct calculation from the analysis/synthesis prototype filters or by a numerical optimization which typically takes place in a computer-aided manner. In such a numerical optimization of filters 32 to 35, a filter length is preset so that a set of filters for different filter lengths can be obtained. As is in particular illustrated in FIG. 10, the marked low-pass characteristic is obtained for the filter with the filter transfer function 100, i.e. H.sub.ll, or for the filter with the filter transfer function 101, i.e. H.sub.lm. However, it can be seen that these filter with a very marked attenuation in the blocking region can be approximated by considerably shorter filters, namely by the filters 102 or 103. The filters 102 and 103 have a filter length of only 11 and thus approximate the filters 100 and 101. However, it can be seen that in the region of low frequencies, the deviations are very low and only increase at higher frequencies. However, on the other hand, a blocking attenuation of greater than 40 dB is ensured by the filters 102 and 103 so that these very short filters already cause good aliasing suppression.

(68) FIG. 11 shows a response of the filter bank for an impulse at a position 8 in a subband sample period. A real filter bank provides a form indicated at 110. A complex filter bank provides a form indicated at 112. A real filter bank including correction according to the present invention provides the form indicated at 111. It can be seen that the real filter bank including correction has nearly the same form as the complex filter bank, however can be implemented considerably cheaper. Only directly at the band limit between k1 and k does the real filter bank including correction exhibit a wavy form, which can be attributed to the fact that only filter lengths of 11, as is illustrated in FIG. 10, instead of complete filters, as were illustrated in FIG. 10, are used. It is apparent that the deviation between the real filter bank including correction and the complex filter bank which is aliasing-resistant is negligible although short filters 32 to 35 of FIG. 3b are already used. The deviation between the forms 111 and 112 becomes greater with shorter filter lengths, however, for a delay-optimized variation, filter lengths smaller than 5 may be used, wherein the deviation between the curves 111 and 112 still remains reasonable.

(69) Subsequently, a filter bank response when a sine tone at 1% above the band limit is considered is shown in FIG. 12. The input signal 121 represents the sine tone. A real filter bank would produce aliasing, as is illustrated by a curve 122. Aliasing becomes noticeable by the secondary peak 125, wherein this secondary peak is caused by the fact that the neighboring bands k1 and k have been weighted by different weighting factors. Again, it is apparent that a complex filter bank does not have such a secondary peak, i.e. does not generate such aliasing, and that the complex filter bank is approximated optimally by a real filter bank including correction, wherein a deviation of the real filter bank to the complex filter bank is only in the region 126. The real filter bank provides a greater attenuation than the complex filter bank, which in turn can be attributed to the fact that the filter lengths of the filters 32 to 35 have been shortened to 11.

(70) It is to be pointed out that in the example in FIG. 12, and in the examples in FIGS. 13, 14, 15, 16, 17, there has been an attenuation of a subband by 20 dB compared to the other subband.

(71) FIG. 13 shows a similar case as FIG. 12, however with a sine tone at 5% above the band limit. Again, a real filter bank would generate the secondary peak 125. This secondary peak, however, is attenuated nearly completely by the real filter bank including correction 124. Only a very small deviation at 127 is still to be seen. When reducing the filter length of the filters 32 to 35, this peak 127 would continue increasing. Even with degenerated filters, i.e. filters only performing weighting by a weighting factor, the peak 127 would still be smaller than the secondary peak 125. By inventively weighting the filtered values by the difference of the two weighting factors, however, at least for the case of identical or nearly identical weighting factors, almost no interference at all would be introduced, despite the rather rudimentary filtering by degenerated filters.

(72) FIG. 14, FIG. 15, FIG. 16 and FIG. 17 show similar scenarios, wherein the sine tone, however, is in an ever greater distance to the band limit. All the pictures clearly show the aliasing component which would be generated by a real filter bank if no inventive correction was performed. In addition, all the pictures show even smaller deviations at 127 between the real filter bank including correction according to the present invention and an aliasing-resistant complex filter bank 123.

(73) FIG. 18 shows a scenario similar to that of FIGS. 12 to 17, however for a transform in the form of the MDCT. Again, the clear aliasing component 125 can be recognized, occurring at a frequency of 127.88. By means of the inventive correction, i.e. by filtering corresponding MDCT coefficients of successive MDCT peaks by the filters 32 to 35, the aliasing component is reduced except for the small deviation at 127 in FIG. 18. If filter lengths of 11 are used for the filters 32 to 35, the entire assembly will only be settled after around 10 MDCT spectra. Thus, in the implementation of FIGS. 3a and 3b, a delay of 5 MDCT blocks may be used until sensitive output values are generated.

(74) Depending on the circumstances, the inventive method may be implemented in either hardware or software. The implementation can be on a digital storage medium, in particular on a disc or a CD having control signals which may be read out electronically, which can cooperate with a programmable computer system such that the corresponding method will be performed. Generally, the invention is thus also in a computer program product having a program code stored on a machine-readable carrier for performing the inventive method when the computer program product runs on a computer. Put differently, the invention may thus also be realized as a computer program having a program code for performing the method when the computer program runs on a computer.

(75) While this invention has been described in terms of several embodiments, there are alterations, permutations, and equivalents which fall within the scope of this invention. It should also be noted that there are many alternative ways of implementing the methods and compositions of the present invention. It is therefore intended that the following appended claims be interpreted as including all such alterations, permutations and equivalents as fall within the true spirit and scope of the present invention.