WO2016030545A2

WO2016030545A2 - Comparison or optimization of signals using the covariance of algebraic invariants

Info

Publication number: WO2016030545A2
Application number: PCT/EP2015/069876
Authority: WO
Inventors: Clemens Par
Original assignee: Clemens Par
Priority date: 2014-08-29
Filing date: 2015-08-31
Publication date: 2016-03-03
Also published as: WO2016030545A3

Abstract

The invention relates to a method for a signal analysis of a first signal and a second signal, said method comprising the steps: determination of selected points in the first signal on the basis of invariants of said first signal; determination of a signal analysis parameter on the basis of the covariance of the selected points of the first signal with the second signal.

Description

COMPARISON OR OPTIMIZATION OF SIGNALS DUE TO THE COVARITY OF ALGEBRAIC INVARIATIONS

Gaussian signals whose degrees of freedom are not known are studied according to the prior art on the basis of statistical models.

In these statistical models, the so-called covariance represents a measure of the similarity. For a series of measurements with random variables X and Y with the probability E, the covariance is defined as

Y = E ((X-E (X)) (Y-E (Y)) = E (XY) -E (X) E (Y) (1) An estimate of Y based on random samples with data pairs (Xi , Yi), · · ·, (X _n, Y _n) allows the sample covariance

C = ^ _{= 1} {X _j -lQ {Y _j -Y) = ^ _i {n ^ ^X i ^Y iW) (2) where X and Y are the means of X _j and Y _j .

The covariance has proven itself, for example, in the coding of spatial audio signals, in which the highest possible similarity of the spatial audio signals with the spatial audio signals of the original should be achieved. For this purpose, the spatial audio signals or samples of these spatial audio signals are used. However, according to the prior art, these samples do not represent algebraic invariants, as first described in WO2012016992. This application will be shared with EP1850639, WO2009138205, WO2011009649, WO2011009650,

WO2012032178 and WO2014072513 are hereby incorporated by reference.

Furthermore, based on the covariance, the correlation for a measurement series with random variables X and Y with the probability E and the variance var is defined as follows:

[Νατ {Χ) νατ {Υ) γΙτ-

An estimate of p, based on random samples with data pairs (Xi, Yi), · · ·, (X _n, Y _n) allows the sample correlation coefficient

In audio technology, the sample

Correlation coefficient for the time interval [-T, T] and the rms values X _eff and Y _{eff of} the so-called short-term cross-correlation, which will be discussed later: TX (t) Y (t) dt

(5 2TX (t) _eff y (t) _eff

WO2012016992 introduces for the first time in signal technology the analysis of Gaussian signals on the basis of algebraic invariants. In algebra for rational integral homogeneous functions f _LR f _m with n variables xi. , , , x _n (so-called basic forms or basic form systems), such invariants represent whole rational homogeneous functions of the coefficients of such functions f _lr ... _r f _m dcLir ·

In particular, David Hilbert defines: "By 'invariant' without further addition we mean in the following always a whole rational invariant, ie such a whole rational homogeneous function of the coefficients a of the basic form or of the Basic form system, which deals only with powers of

Substitution determinants multiplied by the

The coefficients a are replaced by the corresponding coefficients b of the linearly transformed basic form. "David Hilbert proved that these invariants form an algebraic body in 1893. Since signal technology is based on the basic form systems described above, algebraic invariants should also be used for Gaussian signals Problem is solved for the first time with WO2012016992 for signal technology.

As WO2012016992 explains in detail, spatial audio signals, for example a stereo signal according to the transfer functions, can be used

and

R '= ^ (7) according to

Project S = L '+ R' (8) to the complex plane. The goniometer, which is known from the prior art, is an example of this projection. Lissaj ous figures are formed on the complex plane, which in turn can be projected onto algebraic cones.

Consider the general algebraic cone equation a = ax + bx + cx] +2 fx ₂ x ₃ + 2gx ₃ x _l + 2hx _x x ₂ = 0 and considering now projections of the resulting Lissaj ous figures on such algebraic cones, occurs for example for the trivial cone a _x ² = ax ² + bx + cxj = 0 (10 is a uniform change in the amplitude, see Fig. 1st

Now, if all other trivial algebraic cones are first constructed whose invariants satisfy the so-called apolarity condition, ie equal to zero, according to David Hilbert's discovery of the invariant body with the first apolar nonrivial conical equation linear combinations of two invariants can be formed, which in turn are invariants and the

Fulfill apolarity condition:

For the algebraic cone equations

+ bx + ex] + 2x ₂ x ₃ + 2g _{3 j} +2? X _j X ₂

(11) and ιξ, = Au ¹ + Bu + Cif ₃ + 2F u ₂ u ₃ + 2G 'uu _x + 2 H' u _x u ₂ = 0 (12

the expression aα ² 'is an invariant.

For this, the algebraic conic equations are used

dx - i ⁺ 2 ₍₁₃ and a ^, 2 ² _x = x2 + x ₂ 2 ² - ^ X ² = 0 _{(1 4} considered, for which the expression

is an invariant, and its apolarity condition

reads. It is easy to see that there is no third trivial algebraic cone equation.

The first nontrivial conical equation should show the same apolar behavior to the trivial algebraic conic equation (13), namely u ²

^ + ^ + ^ + 2 ^ + 2 ^ + 2 ^ = 0 (17)

9

Their apolarity condition is known and is aA '+ bB' + cC '+ 2 fF' + 2gG '+ 2hH' = 0 (18) or in this case

With

From (16.) and (20.) follows for the case of apolarity g '= g "= i (21) and

The second trivial cone equation is shown in FIG. 2 and the nontrivial cone equation is shown in FIG.

The relation aα ² 'is linear for the coefficients of (13.) and (14.) as well as linear for the coefficients of (13.) and (17.). Thus, the linear combinations represent

(23

according to Hilbert's theorem on the invariant body also apolar invariants, which together form the plane of FIG.

A second equivalent proof of the existence of these invariant apolar linear combinations can be built on Hilbert's construction of the so-called canonical null forms, which will not be discussed here. Disclosure of the invention

The invention is not to determine the covariance on the basis of Gaussian signals or samples of Gaussian signals, but to project these Gaussian signals or these samples of Gaussian signals onto the complex plane, and rather the time-varying algebraic invariants of these Gaussian signals or derive the algebraic invariants of these samples from Gaussian signals, and then determine their covariance. Thus, not the covariance of two signals X and Y or samples of two signals X and Y is examined, but rather the covariance of their non-polar invarian th. Alternatively, the Gaussian signal X or the sample X of a Gaussian signal can be projected onto the complex plane, and the algebraic invariants of this Gaussian signal X occurring over time or the algebraic invariants of the sample X of a Gaussian signal can be derived. Then the function values of the second Gaussian signal Y occurring simultaneously with the particular invariants or the further sample of the Gaussian signals Y are determined, and then their covariance with the first determined invariants. Thus, not the covariance of two signals X and Y or samples of two signals X and Y is examined, but rather the covariance of the apolar invariants of the signal X and the function values of the signal Y. This allows only the algebraic invariants and / or possibly also Function values of a small subset of these two signals X and Y or these samples of two signals X and Y to determine, such as only the beginning of the signal, and thus significantly reduces the computational effort. As practical results show, see also below, the result is far more precise in terms of the totality of available data, even if only a tiny subset of the two signals X and Y are available for determining the algebraic invariants and / or possibly also for functional values. The application of these results to the signal technology thus enables the rapid comparison and the rapid optimization in the signaling technology and in particular in the spatial audio coding.

The example of an arrangement according to the invention is shown in FIG. 15. Description of the Figures

• Fig. 1 shows the first trivial case of an apolar algebraic cone over the complex plane. FIG. 2 shows the second trivial case of an algebraic cone which is apolar to FIG. 1, above the complex plane.

FIG. 3 shows the non-trivial case of a first algebraic cone that is apolary to FIG. 1 and intersects the complex plane.

• Fig. 4 shows the plane on which the entirety of the linear combination of apolar invariants formed above is mapped.

• Fig. 5 shows an ECMA-407 S5 encoder. · Fig. 6 shows an ECMA-407 S5 decoder.

• Fig. 7 shows the parameters of the inverse coding according to ECMA-407 ("Table 3").

FIG. 8 shows the interval steps for the variation of the parameters in FIG. 7. FIG. 9 shows the first flow chart of a signal analysis according to ECMA-407.

• Fig. 10 shows a NHK-22.2 loudspeaker arrangement.

• Fig. 11 shows the optimized interpolation of the algebraic invariants for discrete signals. · Fig. 12 shows the second flowchart of a

Signal analysis according to ECMA-407 with dichotomy.

FIG. 13 shows a multichannel upmix (in particular 3D upmix) on the basis of a measurement series database.

FIG. 14 shows a virtual multichannel upmix (in particular 3D upmix). • Fig. 15 shows by way of example the principle of

Invariant determination for a signal X and the principle of determining the covariance with a signal Y.

FIG. 16 shows the application of FIG. 15 for an ECMA

407 S5 encoder or upmix.

Application example for ECMA-407

ECMA-407 standardizes a system for inverse coding of spatial audio signals. ECMA-407 is hereby incorporated by reference.

Furthermore, this inverse coding is also described in EP1850639, WO2009138205, WO2011009649, WO2011009650, WO2012016992, WO2012032178 and WO2014072513.

FIGS. 5 and 6 cite ECMA-407 (there "Figure 1" and "Figure 2") and show the schematic structure of an ECMA-407 S5 encoder and ECMA-407 S5 decoder:

An audio signal ("f-channel audio source", see (51)) is supplied to an S5 base encoder ("Base S5 encoder", see (52)), which performs signal analysis (see (53) In addition to the configuration data, the parameters ("Inverse Coding parameter data", see (55) and (65)) for the subsequent inverse coding ("") are output and the downmix (see (54)). S5 upmix ", see (63)) in the S5 base decoder (" Base S5 decoder ", see (62)) as well as the downmix (" g-channel audio downmix (g <f) ", see (56) resp (66)).

An application of the invention can thus represent the signal analysis (53) of both the audio signal (51) and the simulated inverse coding of the downmix (63), as will be done later in the S5 base decoder (62): In this case, for each successive window of, for example, 512 samples at a sampling rate of 48 kHz of the audio signal (51), which are not all equal to zero for the respective window, the sample covariances of the algebraic invariants of two channels of the original signal and the function values of the inverse coding of a downmix signal of these two channels (which usually also contains the beyond the 512 samples following subsequent samples of each successive window to at least the maximum possible delay or its subsequent sample) calculated with a respective varied parameter set until the respectively calculated sample covariance the calculated invariants and calculated function values satisfy a previously defined selection or abort criterion (for example, that the associated sample

Correlation coefficient, see above, in a sufficiently small neighborhood of 1).

Also, the output channels of the inverse coding can be interchanged. Such an exchange determines the value of the parameter S5Swap of the inverse coding.

Since the invariants lie on the diagonal of the first and third quadrants intersecting the origin, it suffices to calculate the invariants of one of the two channels S5Left or S5Right, see below. Thus, these algebraic invariants again significantly accelerate the

Computing process.

The parameter set which satisfies a previously defined selection or abort criterion (for example, that for the covariance of the calculated invariants and calculated function values the associated sample correlation coefficient, see above, is in a sufficiently small neighborhood of 1), therefore, part of the transmitted parameters is the inverse Coding (55, 65). This process can be over a run longer time course again in the same way. The process can be performed in total for a variety of channels and channel combinations, all of which should be considered as included in the invention. Fig. 7 shows the parameter set for an inverse coding (corresponding to "Table 3" of ECMA-407) The meaning of this parameter set can be found on page 3 of ECMA-407, the parameters n for the directional characteristic, φ for the angle between Main axis and sound source, α for the fictitious left opening angle, β for the notional right opening angle, s for the time scaling factor and λ for the amplification factor for the side signal of EP1850639, WO2009138205, WO2011009649, WO2011009650, WO2012016992, WO2012032178 and WO2014072513 7 have proven to be particularly favorable for the mentioned parameters, they can be both fallen below and exceeded.All possible values are to be considered included in the invention.

In a preferred embodiment, S5Directivity is first set equal to zero. In a preferred embodiment, S5Incidence is first set equal to zero. In a preferred embodiment, S5Alpha is first set equal to n / 2. In a preferred embodiment, S5Beta is first set equal to n / 2. In a preferred embodiment, S5Scale is first set equal to 100ms. In a preferred embodiment, S5Side is first set equal to 0.05.

In a preferred embodiment, first the parameters S5Incidence, S5Alpha, S5Beta S5Side and S5Swap of the inverse coding are optimized. In a preferred embodiment, the parameter S5Scale is then optimized, starting with the value 29ms.

In a preferred embodiment, the parameter S5Directivity is then optimized, starting with the value 0.

Finally, the parameter S5Side is again optimized in order to achieve for the resulting inverse coded signal those sample correlation coefficients or those short-term cross-correlation which is as equal as possible to the sample correlation coefficient or the short-term cross-correlation of the associated original signal pair:

For this purpose, in a preferred embodiment showing quadratic convergence, first S5Side: = 0.5 is set, and then the covariances are calculated with 1: = 0.25 for S5Sidei = S5Side - 1 = 0.25 and S5Side ₂ = S5Side + 1 = 0.75, respectively. It is now set again S5Side equal to that of the two values S5Sidei and S5Side ₂ whose sample correlation coefficient or its correlation degree based on the short-term cross-correlation is closest to the sample correlation coefficient or the correlation of the original signal determined by the short-term cross-correlation, and then set new 1 equal to the halved value of its old value, 1: = 1/2. In a next step are now each newly of the sample correlation coefficient or newly of specific reference to the short-time cross-correlation degree of correlation for newly S5Sidei = S5Side - 1 and newly S5Side ₂ = S5Side + determined 1, and now new S5Side equal to that of the two values S5Sidei and S5Side ₂ whose sample correlation coefficient or its correlation degree, determined on the basis of the short-term cross-correlation, determines the sample correlation coefficient or the short-term cross-correlation

Correlation degree of the original signal is closest. This The process continues in the same way until the sample data obtained by the inverse coding

Correlation coefficient or determined on the basis of the short-term cross-correlation degree of correlation of that of the original signal sufficiently close or new 1 falls below a predetermined limit.

For the selection or abort criterion for the comparison of two sample covariances, a difference of less than or equal to 0.001 has been proven. The same applies to the selection or termination criterion for the comparison of two sample correlation coefficients.

The flow of one possible embodiment of the signal analysis (53) for ECMA-407 is shown in the flowchart of Figure 9 as well as the arrangement of Figure 16. Specifically, the steps for one of the successive windows include, for example, 512 samples at a sampling rate of 48kHz for every two adjacent channels of the original signal:

1. the determination of the sample covariance and the sample correlation coefficient of the two channels of the

Original signal,

2. the determination of the apolar invariants for the transfer functions (6.), (7.) and (8.) of the two channels of the original signal, see also below, 3. the determination of the downmix signal of the two channels of the

Original signal for the window of, for example, 512 samples and, for example, the following 7680 samples (in order to be able to calculate all delays for the inverse coding of the downmix signal), the determination of the function values of the transfer functions (6.), (7.) and (8.) the two channels of the inverse coded Signal and their commutation for all possible parameter variations of S5Incidence, S5Alpha, S5Beta, S5Side, and the optimization of these inverse encoding parameters including S5Swap in terms of the sample covariance of these function values and the apolar invariants of 2.,

5. the determination of the function values of the transfer functions (6.) _/ (7.) and (8.) of the two channels of the inverse coded signal also their permutation for all possible values for S5Scale, starting with S5Scale: = 29ms, and successively increasing S5Scale (where, for example, the function values for 100ms may already be 4.) until S5Scale: = 146ms, and the optimization of this parameter with respect to the sample covariance of these function values and the apolar invariants from 2.,

6. the determination of the function values of the transfer functions (6.) _/ (7.) and (8.) of the two channels of the inverse coded signal also their permutation for all possible values for S5Directivity, starting with S5Directivity: = 0 (possibly already with 4 .), and successively increasing S5Directivity until S5Directivity: = 2, and optimizing this parameter with respect to the sample covariance of these function values and the apolar invariants from 2 nd, and

7. iterative determination of the optimized parameter S5Side with quadratic convergence, see above.

In general, for suitably selected selection or termination criteria, see above, such a procedure leads already after very few windows to determine the appropriate parameter set. This procedure represents a pure embodiment, and other arrangements which the Calculate parameters of the inverse encoding in a different order, other initial values or other interval steps are therefore easily possible; these should be considered included in the subject of the invention.

For example, a highly complex 3D loudspeaker format such as NHK 22.2, see FIG. 10, can be calculated in terms of its inverse encoding in real time, for example according to WO2014072513.

Interpolation of algebraic invariants for discrete signals

As a rule, discrete time signals X and Y are available for the determination by means of the transfer functions (6), (7) for the determination of the described algebraic invariants, all of which lie on the diagonal passing through the origin of the first and third quadrants of the complex number plane .) and (8.) are available.

For a sample S _{j of} this mapping, where ÃŸ (s, -) denotes its real part and Im (sj) its imaginary part, ße (s) = / m (s) (24) then S is a searched apolar invariant i Apply for two consecutive samples s and s

(25) and

Re (s) </ m (s) (26) resp. Re (s _/ -_ ₁ ) <Im (^ Sj_ ₁ ) (27) and ße (s)> / m (s) (28) an invariant ij can be interpolated on the shortest distance between sji and Sj according to the following formula:

_{■ =} / m (5 Re (5_ ₁ ) - / m (5_ ₁ ) * Re (5)

^ Re {sj_ ₁ ) -Re {sj) -lm {sj_ ₁ ) + lm {sj)

Thus, at the beginning, it is only necessary to determine the position of the first point of the mapping by means of the transfer functions (6.) _/ (7.) and (8.) with respect to the origin-intersecting diagonals of the first and third quadrants of the complex plane. The algorithm can be obtained by considering only those pairs of successive samples (for example, the pairs S_ ₃ , Sj_ ₂ and Sj_ ₂ r Sj-i in FIG. 11), whose distance - in functional dependence on the energy spectrum of the transfer function (6). resulting signal - the shortest distance of its first sample from the going through the origin diagonal of the first and third quadrant, respectively, significantly accelerate.

The interpolation is shown by way of example in FIG. 11.

Semantic distinction between alternative parameter sets of an inverse coding

In fact, the subject invention allows for alternative parameterizations for inverse coding which, considered psychoacoustically, provide equivalent results.

A possible procedure when selecting a parameter set is thus the statistical analysis of occurring parameterizations, for example according to their frequency over time. However, it may happen that semantically different parameter sets, ie parameter sets with differently perceived spatiality, are treated equally. Thus, one problem is the activation of signal analysis (53) in ECMA-407 for semantically different ones

Parameter sets, for example due to changed

Tonmaterial, while suppressing a change between semantically equivalent, alternative parameterizations, and another problem, the stepless transition from one parameter set to the next, the maximum length usually depends on the latency of the entire system remains (for example, about 900ms, if in the Encoder and / or decoder a loudness measurement is performed).

As WO2012032178 states, the parameters of the inverse coding can alternatively be calculated by a weighting of S5Alpha and S5Beta as well as of S5Side on the basis of the following dichotomy:

Select S5Directivity, S5Incidence, S5Scale, a first value for S5Side, and a weight p for the size of S5Alpha = S5Beta (or equivalent α = ß). Calculate the sample correlation coefficient or the correlation level of the channels of the original signal as determined by the short-term cross-correlation.

Define for L _a and Iß (see ECMA-407) a first weighting ^¬ function, for example, g (a) = 2 ^{~ 20 La + L} ß ^ (30) and finally select each of those S5Side, which for a given value for α = ß an inversely coded signal with the sample correlation coefficient R or an inversely coded signal with the degree of correlation r of its short-term cross-correlation, for example, h (a) = SSSide (a _* #O) ^10_p (31) Then the optimized value for α = β is calculated as, for example

There is no determination of possible alternative parameter sets, which would lead to the same semantic results - but with the serious disadvantage that the spatiality is not nearly as precisely described by the thus determined parameter set for the inverse coding, as with the aid of sample covariance algebraic invariants, see above, is possible.

However, this variant allows as an aid the specific discrimination of alternative parameter sets for the inverse coding determined by the sample covariance of the associated apolar invariants and those parameter sets which do not represent a semantically equivalent alternative but rather have semantic differences.

For example, in one embodiment, S5Directivity is set equal to 0, S5Incidence equal to 0 and S5Scale equal to 100ms, and the weight for S5Alpha = S5Beta (or equivalent to α = β) equals 5. As the initial value for α = β, assume n / 2 as the initial value for S5Side the value 0.05. Now, L _a and Iβ are calculated. Subsequently, S5Side, see above, is determined with quadratic convergence.

For S5Alpha = S5Beta and / or S5Side thus determined, its differences are now quantized over time, for example, compared with the first value which occurs simultaneously with a parameter set for the inverse coding determined by means of the sample covariance of apolar invariants and function values. If these are greater than or equal to a specific value (for example 5 ° or 10 ° for S5Alpha = S5Beta), a corresponding parameter set is newly generated by means of the Sample covariance determined by apolar invariants and / or function values.

In general, using the sample covariance of apolar invariants, certain parameter sets are very robust and can remain constant for long time intervals of several minutes. If they change, they are based on a completely different semantics (spatiality). With regard to semantically different sound material, this may mean a short transition (hereinafter called "cut") or a gradual transition (hereinafter called "fade").

Both a "cut" and a "fade" can be visualized by quantizing dichotomy-specific parameters for inverse coding, namely S5Alpha = S5Beta and / or S5Side. While a "cut" is usually around a frame of 512 samples, a "fade" can be up to around 900ms and longer.

It is at the same time part of the invention to distinguish a "cut" or "fade" and to compensate psychoacoustically by short or long fading from the inverse coding of the old parameter set to the inverse coding of the new parameter set. The transition from the old parameter set to the new parameter set with careful quantization of the parameters determined by dichotomy for an inverse coding, namely S5Alpha = S5Beta and / or S5Side, is subjectively no longer perceptible. The maximum length for a "fade" usually corresponds to the latency of the entire system (for example, around 900 ms, if a loudness measurement is carried out in the encoder and / or decoder, etc.) The sequence described in steps 1 to 7 complements each other therefore the following steps: For the window with, for example, 512 samples, for which an optimized parameter set for the corresponding inverse coding was determined by means of the sample covariance of apolar invariants, use dichotomy to determine the parameters S5Alpha = S5Beta and / or S5 Side, see above. Determine the parameters S5Alpha = S5Beta and / or S5 Side for each following window with, for example, 512 samples by dichotomy, see above, and then compare each of these as follows

Parameter with the parameters S5Alpha = S5Beta and / or S5 Side of 8. until the amount of their respective difference exceeds a first smaller, previously determined limit (eg 5 ° for S5Alpha = S5Beta) ("critical limit") the amount of the respective difference determined in 9 exceeds the second larger previously determined limit value (for example 10 ° for S5Alpha = S5Beta), for a "cut" 11. Otherwise, determine the parameters S5Alpha = S5Beta and / or S5 Side for each subsequent window with, for example, 512 samples by dichotomy, and then compare each of these parameters with the parameters S5Alpha = S5Beta and / or S5 Side of FIG. 8. until either the amount of their respective difference exceeds the second larger, previously determined threshold (eg 10 ° for S5Alpha = S5Beta) ("absolute threshold") and thus a "fade" is performed in 11, or until the latency of the overall system is reached (for example, round

900ms, if a loudness measurement is carried out in the encoder and / or decoder, for example), and thus 9 must be carried out again. 11. In the case of a "cut" or "fade", execute now for this window, for example 512 samples and the other windows steps 1 to 7. Crossfades linear or nonlinear (the sum of the amplification factors for the two inverse codings with different parameter sets is constantly equal to 1) in the case of a "cut" from the beginning of the last window before reaching the "critical threshold" with the 8th corresponding inverse coding up to Beginning of the following window, with which the "absolute limit value" is reached, with the inverse coding which results from the just executed steps 1 to 7. Crossfade linear or nonlinear (the sum of the amplification factors for the two inverse codings with different parameter sets always equal to 1) in the case of a "fade" from the beginning of the window defined in Fig. 9 for the "critical limit" with the 8th corresponding inverse coding to the beginning of the window with which the "absolute limit value" is reached Inverse encoding, which results in the just-executed steps 1 to 7.

12. From there, continue for this same window and the following windows 8, 9, 10 and 11 - until the end of the downmix signal is reached.

The entire process is shown in FIG. 12.

Ideally, according to ECMA-407 via S5SyncTag, S5SyncTag-l, S5SyncTag-2, the respective position for the respective new parameter set of the inverse coding to be executed is signaled at (62) or (63).

Application example for a multi-channel-upmix (especially 3D-Upmix) based on a measurement series database In the following, an upmix, that is, obtaining a higher-order audio signal (having a higher number of audio channels) from a lower-order audio signal (having a lower number of audio channels), is to be based on a measurement series database containing the frequency spectrums of room pulses or free-field measurements. respectively.

These space impulses or free-field measurements are first transferred into the Fourier space by means of the Fourier transformation known from the prior art, and thus describe factors x _k from which equivalent factors xi can be derived, with which the same frequencies of the Frequency spectrum of any signal that has been transmitted equally in the Fourier space, multiply (convolve). Such a multiplication (convolution) per se generates a high-quality upmix related to the measured free field or the measured space.

If the Fourier series is determined for the time-dependent signal s ± '(t), then for the synthesis, k = -1, 0, 1,

and for the analysis

In practice, the Fast Fourier Transform (FFT) can be derived directly for the discrete Fourier transformation (DFT), k = 0, N-1 now being used for the analysis

and for the synthesis s (m = ^ = ¹ ₀ S ' _i (ke ^ik (36)

As a rule, the selection of the most suitable measurement series and associated xi and thus such an upmix is manually implemented by a sound engineer who judges the result semantically.

If this selection is automated, statistical models must be used which, however, according to the prior art, do not include the determination of the covariance of algebraic invariants and / or function values. However, if the function values of the transfer functions (6.), (7.) and (8.) are determined at the complex level for at least two adjacent channels of the original signal, according to (23.) their apolar invariants can be considered as the intersections of the cathode ray with the determine the origin-intersecting diagonals of the first and third quadrants of the complex plane, see above, and in particular describe the semantics (spatiality) of these adjacent channels of the original signal algebraically.

Now, for a channel of the original signal or for a signal derived from the original signal, for at least one window of suitable length (for example 512 samples at a sampling rate of 48 kHz or even about 40 ms), in which not all samples are equal to zero, a Fourier transformation Then, based on the factors of different measurement series of room pulses or free-field measurements in Fourier space, see above, a multiplication (convolution) with certain factors xi can be carried out for each window.

If the results or signals derived from these results, see above, are subsequently transformed from the Fourier space into the period, and the covariances become the first determined apolar invariants and the function values of these from the transformed multiplications

(Convolutions) determined with xi-derived signals at the time of occurrence of apolar invariants corresponds to that covariance that meets a previously defined selection or termination criterion (for example, that the associated sample correlation coefficient, see above, in a sufficiently small environment of 1 ), that multiplication (convolution) or those multiplications (convolutions) with xi, their semantics (spatiality) of the semantics

(Spatiality) of the original signal comes closest.

All channels of the original signal or all channels derived from this original signal are now multiplied (convoluted) by the corresponding factors xi of this series of measurements of room impulses or free field measurements and thus result in an upmix whose semantics (spatiality) as far as possible match the semantics (spatiality) of the room Original signal corresponds.

This procedure can optionally be extended to any selection of adjacent channels of the original signal, in order to determine the best measurement series of room impulses or free field measurements on the basis of their semantics (spatiality).

This process is shown in FIG. In parallel, an inverse coding with dichotomy, see above, can be performed according to steps 8 to 12, replacing "optimized parameter set for the corresponding inverse coding" and "inverse coding" by the "optimal convolution xi", and in Step 11. Steps 1 to 7 through FIG. 13.

Application example for a virtual multi-channel-upmix (especially 3D-Upmix) The multiplication (convolution) with the factors xi of an associated measurement series database can be achieved by a virtual upmixing method, for example a linear or nonlinear inverse coding as described in EP1850639, WO2009138205, WO2011009649, WO2011009650, WO2012016992, WO2012032178, WO2014072513 and ECMA-407. replace equivalently. Thus, in the above-described procedure, the function values obtained by means of xi of different measurement series are not used for the determination of the covariance, but equivalently the function values of channels obtained or derived on the basis of these virtual upmix methods. Thus, not that series of measurements of spatial impulses or free-field measurements is selected, but rather that upmixing method and / or that parameterization of this upmix method whose semantics (spatiality) most closely matches the semantics (spatiality) of the original signal.

This process is shown in FIG.

In parallel, an inverse coding with dichotomy, see above, can be performed according to steps 8 to 12, where "optimized parameter set for the associated inverse coding" and "inverse coding" are replaced by "virtual multichannel upmixing" if necessary.

Application example for a virtual multi-channel-upmix (especially 3D-Upmix) in combination with a measurement series database

Both procedures described can also be combined: In a first step, it is possible to determine the measurement series of room impulses or free-field measurements and thus those xi whose semantics (spatiality) corresponds as far as possible to the semantics (spatiality) of the original signal. Subsequently, according to the above "application example for a multichannel upmix (in particular 3D upmix) using a measurement series database" the multichannel upmix (in particular 3D upmix) at least for the times of the specific apolar invariants of the original signal or of the signal derived from the original signal determined by the xi.

Subsequently, the covariance of this multichannel upmix (in particular 3D upmix) with the function values of the virtual multichannel upmix (in particular virtual 3D upmix) equivalent to at least these function values is determined for at least these function values with its associated parameters.

When this multichannel upmix (especially 3D upmix) is fully applied to the original signal or to channels derived from that original signal, the result is in terms of the semantics (spatiality) of the original signal and that obtained by multiplication (convolution) with the xi factors Multi-channel upmix (especially 3D upmix) optimized virtual multi-channel upmix (especially virtual 3D upmix).

Alternatively, the invariants of the multiplication (convolution) with the factors xi achieved multi-channel upmix (especially 3D Upmix) can be determined, and then according to the above "application example for a virtual multi-channel upmix (especially 3D Upmix)" this optimal virtual Multi-channel upmix (especially virtual 3D upmix) with its associated parameters.

Parallel to this, an inverse coding with dichotomy, see above, can be carried out according to steps 8 to 12, where optionally "optimized parameter set for the corresponding inverse coding" and "inverse coding" by the "Optimal folding xi" are replaced, and in step 11, the steps 1 to 7 through Fig. 13.

All described multichannel upmix methods (in particular 3D upmix methods) have the advantage that only one or a few signal windows are sufficient in comparison with statistical models in order to determine the best possible multichannel upmix (in particular 3D upmix) with the lowest possible latency. This allows an automatic upmix in real-time applications and thus, for example, the live processing of multi-channel signals generated in the broadcasting operation of a radio or television broadcaster.

Closing remarks

It is possible that for the second signal Y there is a blurring with respect to the particular invariants of the first signal X. Signals Y having this property are considered included in the subject invention.

It is immediately apparent that other transfer functions, which with appropriate transformation represent the transfer functions (6.), (7.) and (8.), can be used to determine the apolar invariants. These

Variants are considered included in the subject invention.

Also, instead of the covariance, the correlation can be considered equivalently (for the definition of the correlation, see above). These embodiments are considered included in the subject invention.

Also, other embodiments and parameter sets, see EP1850639, WO2009138205, WO2011009649, WO2011009650, WO2012016992, WO2012032178 and WO2014072513, as described in ECMA-407, can be applied for inverse coding. These variants are considered in the

Subject of the invention included.

Also, equivalent dichotomies and / or other weights than those described in (30), (31), and (32) can be formed. These embodiments are considered included in the subject invention.

Also, the factors xi for multiplication (convolution) may change with a signal over time. The resulting embodiments are considered to be included in the subject invention.

A parameter set contains at least one parameter for

parametric determination of an upmix. A signal analysis parameter based on covariance may be the covariance itself or any parameter formed from the covariance, such as e.g. to be the correlation.

Determining selected points in a first signal based on invariants of the signal works in the

Usually so that the points of the first signal are selected

at which the first signal or a function value of the first signal value at this point fulfills an invariant condition. The invariant condition is typically determined as described above based on at least two channels of the first signal (two-dimensional or two-channel)

multidimensional first signal). The selected points can be both selected sample points, e.g. the the

Invariant condition of nearest sample points, as well as selected interpolation points generated by interpolation between the two adjacent sample points closest to the invariant condition, as well as other points lying in an appropriate environment around the point where the invariant condition is met , One Signal has a certain number of sample points in practice. Selecting points based on invariants of the signal may result in a number of selected points that is between zero and the determined number of sample points. In theoretical limit cases, the number of selected points may also assume zero or the certain number of sample points. But, as a rule, the number of selected points is less than the specified number of sample points. In general, the selected points are not evenly spaced or

are randomly / randomly selected, but are selected algebraically on the basis of the invariants. This can lead to different degrees of reduction depending on the signal type. For

Gaussian signals, e.g. This may result in a reduction of more than 80% or 90% or 95%, in some cases an even higher reduction, and lead to consistent or improved quality over prior art statistical methods.

Claims

PATENTA'S TEST

A method of signal analysis of a first signal and a second signal comprising the steps of:

Determining selected points in the first signal based on invariants of the first signal;

Determining a signal analysis parameter based on the covariance of the selected points of the first signal with the second signal.

The method of claim 1, wherein the covariance between the selected points of the first signal and selected points of the second signal is calculated.

3. The method of claim 2, wherein the selected points of the second signal are determined based on the second signal or based on the selected points of the first signal.

The method of claim 3, wherein the selected points of the second signal correspond to the selected points of the first

Correspond to signal.

5. The method according to any one of claims 1 to 4, wherein the first signal based on at least a first

Input signal is determined and the second signal is determined on the basis of the at least one second input signal.

6. The method according to any one of claims 1 to 5, wherein the first signal is two- or multi-dimensional and / or the second signal is two- or multi-dimensional.

7. The method of claim 6, wherein the covariance of the selected points of the first signal with a second Signal is determined on the basis of only one dimension of the first signal and / or only one dimension of the second signal.

8. The method of claim 6, wherein the covariance of the selected points of the first signal with a second

Signal is determined on the basis of at least two dimensions of the first signal and / or at least two dimensions of the second signal.

9. The method of claim 1, wherein the signal analysis parameter is based on the correlation of the selected points of the first signal with a second one

Signal is determined.

10. The method of claim 1, wherein the first signal is determined based on a first base signal and a base signal of the first signal, and the second signal is determined based on a first base signal and a first base signal

Base signal of the second signal is determined.

11. The method of claim 10, wherein the first signal is based on a function of the first

Base signal and the base signal of the first signal and the second signal based on the same function in

Dependence of the first base signal and the base signal of the second signal is determined.

The method of claim 11, wherein the function is - = [x (t) ^■ (-1 + i) + y (t) ^■ (1 + i)] with the first one

V2

Base signal x (t) and the second base signal y (t), where the first dimension is the real space and the second dimension is the imaginary space.

13. A method for determining a target parameter set comprising the steps:

Providing a first provided signal; Determining a first parameterized signal based on a parameter set;

Determining a signal analysis parameter according to the signal analysis method of the first provided

Signal and the first parameterized signal according to any one of the preceding claims;

Determine at least one other first

parameterized signal based on at least one

further parameter set;

Determine at least one other

Signal analysis parameters according to the method for signal analysis of the first signal provided and the at least one further first parameterized signal according to one of the preceding claims;

Compare the signal analysis parameter and the

at least one further signal analysis parameter for determining a target parameter set for determining the first

parameterized signal.

14. The method according to claim 13, comprising the further steps:

Providing a second provided signal;

Determining a second parameterized signal based on the parameter set;

Determining the signal analysis parameter according to the method of any one of claims 10 to 12 with the first

provided signal and the second provided

Signal as first base signal and second base signal of the first signal and with the first parameterized signal and the second parameterized signal as the first base signal and the second base signal of the second signal;

Determining at least one further second parameterized signal based on the at least one further parameter set;

Determining the at least one other

Signal analysis parameters according to the method of any one of claims 10 to 12 with the first signal provided and the second provided signal as the first base signal and the second base signal of the first signal and with the

at least one further first parameterized signal and the at least one further second parameterized signal as the first base signal and the second base signal of the second signal.

15. The method of claim 13, wherein the first parameterized signal and / or the second parameterized signal are based on the parameter set and based on the first provided signal and / or the second parameterized signal

provided signal is / are determined.

16. The method of claim 15, wherein the first

provided signal and / or the second provided

Signal is / are compressed, and the first parameterized signal and / or the second parameterized signal on the basis of the parameter set and on the basis of the compressed first signal provided and / or second

provided signal is / are determined.

17. The method of claim 14, wherein the first provided signal and the second provided signal are processed into a downmix signal, and the first parameterized signal and the second parameterized one

Signal is / are determined on the basis of the parameter set and on the basis of the downmix signal.

18. The method of claim 15, wherein an upmix signal having at least three channels based on the first provided signal and the second provided

Signal is determined, and the first parameterized signal and the second parameterized signal on the basis of two or more channels of the upmix signal is determined.

19. The method according to claim 18, wherein the channels of the

Upmixsignals be determined by convolution of the first signal provided and the second signal provided with different impulse response functions, the

Impulse response functions form the parameter set.

20. The method according to claim 14, wherein the at least one further first parameterized signal and / or the at least one further second parameterized signal are based on the at least one further first provided signal and / or the at least one further second

provided signal is / are determined as the first parameterized signal and / or the second parameterized signal on the basis of the parameter set and on the basis of the first signal provided and / or the second

provided signal is / are determined.

21. A method for determining target parameter sets for a plurality of adjacent signal windows, comprising the steps:

Determining a destination parameter set for a first signal window;

Determining a characteristic value for a signal content of the signal window;

Determine the characteristic value for a

Signal content of a second signal window, the first

Signal window follows;

Comparing the characteristic value of the first signal window with the characteristic value of the second signal window; Determining whether the target parameter set of the first signal window is used for the second signal window or a new one based on the comparison result

Destination parameter set for the second signal window is determined.

22. The method of claim 21, wherein the difference of the characteristic value of the first signal window and the characteristic value of the second signal window is determined, the second signal window being assigned to the parameter set and the new parameter set if the difference is above a threshold value.

23. The method of claim 22, wherein when the difference of the characteristic value of the first signal window and the characteristic value of the second signal window is above a further threshold, but below the

Threshold is and if a third signal window within a number of signal windows the second

Signal window follows, which is smaller than a certain number, then the signal window between the second and third signal windows, the parameter set and a new

Parameter set assigned, wherein the new parameter set is calculated on the basis of the third signal window.

24. The method of claim 22 or 23, wherein the assignment of two parameter sets to one or more signal windows on the basis of fading, in particular of linear fading occurs.

25. The method according to any one of claims 21 to 24, wherein the

Parameter set for each signal window is determined by the method according to one of claims 13 to 20.

26. The method of claim 1, wherein the signals are audio signals.

27. Application of the method according to one of the previous ones

Claims in an audio encoder.

28. Device designed to carry out the method according to one of claims 1 to 26.

29. Computer program formed when executed on a processor to perform a method according to one of claims 1 to 26 from.