RU2461079C2 - Simplified vector indexing and deindexing - Google Patents

Abstract

FIELD: information technology.

SUBSTANCE: indexing procedure involves performing specific processing when an input vector is contained in a set of vectors of a predetermined group of one or more sets of vectors which are part of a plurality of sets of vectors, wherein the specific processing is adapted to a property of the sets of vectors in a predetermined group of sets of vectors and is applicable only in case of input vectors contained in the sets of vectors with a defined property. Also, the indexing procedure involves performing general processing if an input vector is not contained in the set of vectors of a predetermined group of sets of vectors. The invention also relates to a corresponding procedure for determining the target vector contained in the set of vectors, which is part of the plurality of vectors, based on an index associated with said target vector.

EFFECT: simple indexing or deindexing procedure.

37 cl, 4 dwg

Description

FIELD OF TECHNOLOGY

This invention relates to indexing and deindexing of vectors.

BACKGROUND

In providing communication via a mobile terminal, a simple role is played by simple speech and audio coding algorithms. Due to the small amount of memory used and the simplicity of the algorithm, along with efficient coding, structured code dictionaries are used, for example, for quantization in several modern speech and audio codecs, such as the adaptive multi-speed wideband (AMR-WB, Adaptive Multi Rate - WideBand) codec or G.EV -VBR, whose standard is currently being developed as part of ITU-T SG16 / Q9.

Code dictionaries used in methods for quantizing speech and audio codecs can, for example, be based on lattice structures, as described in S. Ragot, B. Bessette and R. Lefebvre "Low-complexity multi-rate lattice vector quantization with application to wideband TCX speech coding at 32 kbit / s "(Uncomplicated multi-speed trellis vector quantization used for TCX broadband speech coding at 32 kbps), published in IEEE International Conference on Acoustics, Speech, and Signal Processing, 2004 (ICASSP ' 04, IEEE International Conference on Acoustics, Speech and Signal Processing), M Onreal, Canada, May 17-21, 2004, Volume 1, pp. 501-504, which is incorporated herein by reference in its entirety. This article also describes the quantization process in detail.

If lattice structures are used as code dictionaries, then algorithms are required for indexing lattice nodes.

A code book with a lattice structure can be defined as a union of leader classes, each of which is characterized by a leader vector. A leader vector is an n-dimensional vector with ordered (e.g., in descending order) components. The leader class corresponding to the leader vector includes the leader vector and all vectors obtained by all permutations of the leader vector, taking into account the signs (with some possible restrictions).

The type of algorithm for indexing lattice nodes, defined as a union of leader classes, can, for example, be an algorithm for indexing leader vectors. In this context, among other things, the lexicographic and binomial algorithms described in the article by A. Vasilache and I. Tabus "Robust indexing of lattices and permutation codes over binary symmetric channels" (Reliable indexing of lattices and permutation codes in symmetric binary channels) can be applied, published in Signal Processing, Volume 83, Number 7, pp. 1467-1486, 2003, which is incorporated herein by reference in its entirety.

Binomial algorithms are usually less complex because they simultaneously take into account several vector components.

SUMMARY OF THE INVENTION

It is possible to further simplify the indexing and / or de-indexing process.

In accordance with a first aspect of the present invention, a method is disclosed comprising indexing an input vector contained in a set of vectors included in a plurality of sets of vectors. The indexing procedure involves performing specific processing in the case where the input vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the set of vectors, while the specific processing is adapted to the property of the sets of vectors in a predetermined group of vector sets and is applicable only in the case of input vectors contained in sets of vectors with a specific property. In addition, the indexing procedure includes performing general processing if the input vector is not contained in the vector set of a predetermined group of vector sets.

In accordance with a first aspect of the present invention, a computer program is also disclosed. The computer program contains instructions for the processor to index the input vector contained in a set of vectors included in many sets of vectors. The indexing procedure involves performing specific processing in the case where the input vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the set of vectors, while the specific processing is adapted to the property of the sets of vectors in a predetermined group of vector sets and is applicable only in the case of input vectors contained in sets of vectors with a specific property. In addition, the indexing procedure includes performing general processing if the input vector is not contained in the vector set of a predetermined group of vector sets.

In accordance with a first aspect of the present invention, a computer-readable medium on which such a computer program is stored is also disclosed. A computer-readable medium can be any medium on which digital data can be stored in electrical, magnetic, electromagnetic, or optical form. The media may be fixedly mounted in the device or may be removable media.

According to a first aspect of the present invention, there is also disclosed a device comprising a processing component adapted to index an input vector contained in a set of vectors included in a plurality of sets of vectors. In addition, the processing component is configured to perform specific processing in the case where the input vector to be indexed is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the set of vectors, while the specific processing is adapted to the property of the sets of vectors in a predefined group of sets of vectors and is applicable only in the case of input vectors contained in sets of vectors with a specific property. The processing component is also configured to perform general processing if the input vector to be indexed is not contained in a set of vectors of a predetermined group of sets of vectors.

In accordance with a first aspect of the present invention, there is also disclosed an apparatus including means for indexing an input vector contained in a set of vectors included in a plurality of sets of vectors. Indexing tools include tools for performing specific processing in the case when the input vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the set of vectors, while the specific processing is adapted to the property of the sets of vectors in a predetermined group of sets of vectors and is applicable only in the case of input vectors contained in sets of vectors with a specific property. In addition, the indexing means comprise means for performing general processing in the event that the input vector is not contained in the set of vectors of a predetermined group of sets of vectors.

According to a first aspect of the present invention, the input vector is a vector to be indexed. The input vector is contained in a set of vectors, which, in turn, is contained in a set of sets of vectors. Many sets of vectors can, for example, serve as a codebook for the quantization process. During such a quantization process, an input vector can, for example, be identified as the most accurate correspondence of all vectors in the set of vector sets with respect to the vector to be quantized.

When indexing, an index is assigned to the input vector. The index may, for example, take the form of a waveform. Between two objects, this signal can be exchanged to identify the input vector. Using an index, an input vector can be at least partially identified in the set of vectors in which it is contained. An index can, for example, identify an input vector in a plurality of sets of vectors, or only in a set of vectors in which this input vector is contained. In addition, the index may or may not represent the signs of the components of the input vector.

When indexing the input vector, for example, at least a position index representing the position of the values in the input vector may be output. During indexing, an additional index of a character may be provided that represents the character of the components of the input vector. In addition, during indexing, a final index can be issued, which combines the position index and the index of the sign and can additionally represent the leader class that contains the input vector.

Depending on the set of vectors in which the input vector is contained, various processing procedures are performed. If the input vector is contained in a set of vectors that belongs to a predefined group of sets of vectors, then specific processing is applied. Otherwise, general processing is applied. Both types of processing form at least part of the indexing procedure of the input vector. For example, specific processing and general processing may relate to the entire indexing process, but may relate only to parts of this process.

Specific processing is adapted to the property of a set of vectors, which contains the input vector.

Specific processing is configured for a set of vectors containing the input vector. This allows you to increase processing capabilities and reduce the overall level of complexity of the process of indexing the input vector.

For example, the structure of a specific treatment can be simplified compared to general processing by, for example, performing several actions in a single cycle.

In the same way, the introduction of specific processing can reduce the complexity of the overall processing, for example, by eliminating the processing of input vectors of specific sets of vectors from the overall processing process (and including this procedure in the specific processing process), thereby simplifying the overall processing. Both measures lead to a decrease in the overall complexity of indexing the input vector.

The tuning process may limit the scope of specific processing to the input vectors contained in sets of vectors with a certain property, i.e., specific processing may not be applicable, for example, if the input vectors are contained in sets of vectors that are not included in a predefined group of sets of vectors .

General processing may not be limited to use only if the input vectors are contained in the sets of vectors of a particular group of sets of vectors. For example, general processing may be applicable if the input vectors are contained in any set of vectors included in the set of vectors, or in the case where the input vectors are contained in any set of vectors included in the set of vectors, except for sets of vectors, sets of vectors included in a predetermined group. In the latter case, it is possible to optimize the overall processing with respect to the input vectors contained in the sets of vectors that are not included in a predefined group of sets of vectors.

According to an exemplary embodiment of the first aspect of the present invention, general processing when applied to input vectors contained in sets of vectors with a specific property would be more complex than the specific processing applied to the same input vectors. In this case, complexity may, for example, be related to computational complexity and / or memory requirements.

According to an exemplary embodiment of the first aspect of the present invention, input vectors contained in a vector set of a predetermined group of vector sets may be more likely to be indexed than input vectors contained in sets of vectors outside the predetermined group of vector sets. Given the total complexity, it can be especially useful that specific processing is less complex than general processing (if applied to input vectors contained in sets of vectors with a specific property). The likelihood may depend on the application in which indexing is performed. The probability may, for example, be determined offline.

According to an embodiment of the first aspect of the present invention, the vector sets are leader classes, and each leader class contains a different leader vector and permutations of the leader vector. The combination of these leader classes can then define a lattice that can be used as a structured codebook for quantization, for example, in a speech, audio or video codec, for image encoding, data compression, or in any other area that uses quantization.

Leader classes can have a sign, and when indexing, at least a position index representing the position of the values in the input vector and a sign index representing the signs of the components of the input vector can be returned. Further, within the framework of specific processing, the position index and the sign index can be jointly determined in one cycle. This loop, for example, can be performed for each component of the input vector. On the other hand, within the framework of general processing, the position index and the index of the sign can only be determined one after another. During indexing, the final index can also be issued, which combines the position index and the index of the sign and can optionally represent the leader class that contains the input vector.

Alternatively, leader classes may be unsigned. In this case, when indexing, only the position index representing the position of the values in the input vector can be returned. Depending on the circumstances, the index may also produce a final index, which additionally represents the leader class that contains the input vector.

A property to which specific processing is adapted may consist in the fact that a predetermined group of leader classes consists only of leader classes with leader vectors whose components take only a predetermined number of different values. This predetermined amount may, for example, be two. Adaptation of specific processing to this property, in particular, can be useful in the context of binomial indexing, because, among other things, it allows you to jointly determine the position index and sign index in one cycle, and not separately. In addition, adapting a specific processing to this property can be useful in the context of binomial indexing, because, among other things, the indexing procedure should only process the position of one value (first) without the need for further preparatory operations to check the positions of the second value (its position (positions ) is fixed by fixing the position (s) of the first value).

A predefined group of leader classes may contain a first subgroup of leader classes with leader vectors whose components take only even values, and a second subgroup of leader classes with leader vectors whose components take only odd values. According to the process of specific processing, the actions that can be taken if the input vector is contained in the leader class of the first subgroup of leader classes may differ from the actions that are performed when the input vector is contained in the leader class of the second leader group classes.

Alternatively, the property to which the specific processing is adapted may consist in the fact that a predefined group of leader classes consists only of leader classes with leader vectors whose components take the maximum number of different values. For example, leader classes in a predefined leader group group may contain leader vectors with four values. Processing such leader classes as part of a specific processing process can simplify the overall processing, since then only leader classes that include leader vectors with three or less values can be taken into account during the general processing. This helps to reduce the overall complexity of indexing input vectors.

For optimization purposes, general processing can be applied if the input vectors are contained in any set of vectors included in the set of vectors, with the exception of sets of vectors included in a predefined group of sets of vectors.

Indexing of the input vector can be performed during the quantization process, which is based on a trellis codebook defined by leader classes.

According to an embodiment of the first aspect of the present invention, indexing is binomial indexing. Such a binomial indexing procedure is based on the number of possible distributions of values by position in a vector of a certain dimension.

According to a second aspect of the present invention, a method is disclosed comprising determining a target vector contained in a set of vectors included in a plurality of sets of vectors based on an index associated with the target vector. The determination procedure involves performing specific processing in the case where the target vector is contained in the set of vectors of a predetermined group of one or more sets of vectors included in the set of vectors, and the specific processing is adapted to the property of the sets of vectors in a predetermined group of vector sets and is applicable only in the case of target vectors contained in sets of vectors with a specific property. In addition, the determination procedure includes performing general processing in the event that the target vector is not contained in the vector set of a predetermined group of vector sets.

In accordance with a second aspect of the present invention, there is also disclosed a computer program including instructions for the processor to determine a target vector contained in a set of vectors included in a plurality of sets of vectors based on an index associated with the target vector. The determination procedure involves performing specific processing in the case where the target vector is contained in the set of vectors of a predetermined group of one or more sets of vectors included in the set of vectors, and the specific processing is adapted to the property of the sets of vectors in a predetermined group of vector sets and is applicable only in the case of target vectors contained in sets of vectors with a specific property. In addition, the determination procedure includes performing general processing in the event that the target vector is not contained in the vector set of a predetermined group of vector sets.

In accordance with a second aspect of the present invention, a computer-readable medium on which such a computer program is stored is also disclosed. A computer-readable medium can be any medium on which digital data can be stored in electrical, magnetic, electromagnetic, or optical form. The media may be fixedly mounted in the device or may be removable media.

In accordance with a second aspect of the present invention, there is also disclosed an apparatus comprising a processing component configured to determine a target vector contained in a set of vectors included in a plurality of sets of vectors based on an index associated with the target vector. In addition, the processing component is configured to perform specific processing in the case when the determined target vector is contained in the set of vectors of a predetermined group of one or more sets of vectors included in the set of vectors, while the specific processing is adapted to the property of the sets of vectors in a predetermined group of sets of vectors and is applicable only in the case of target vectors contained in sets of vectors with a specific property. The processing component is also configured to perform general processing if the determined target vector is not contained in the vector set of a predetermined group of vector sets.

In accordance with a second aspect of the present invention, there is also disclosed an apparatus comprising means for determining a target vector contained in a set of vectors included in a plurality of sets of vectors based on an index associated with the target vector. Means for determining include means for performing specific processing in the case when the target vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in many sets of vectors, while the specific processing is adapted to the property of the sets of vectors in advance a given group of sets of vectors and is applicable only in the case of target vectors contained in sets of vectors with a specific property. In addition, the means for determining contain means for performing general processing in the event that the target vector is not contained in the set of vectors of a predetermined group of sets of vectors.

In accordance with a second aspect of the present invention, the target vector is a vector to be determined based on an index associated with the target vector. The target vector is contained in a set of vectors, which, in turn, is contained in a set of sets of vectors. Many sets of vectors can, for example, serve as a codebook for the quantization process. During such a quantization process, the target vector can, for example, be identified as the most exact correspondence of all vectors in the set of vectors with respect to the vector to be quantized, and it can be assigned an index (that is, it can be indexed) in a way such as this, for example disclosed above with reference to the first aspect of the present invention. This index can, for example, at least partially identify the target vector in the set of vectors. In addition, this index may, for example, identify the set of vectors that contain the target vector, and may also indicate the signs of the components of the target vector. Then, based on this index in accordance with the second aspect of the present invention, the target vector is reconstructed. This may require transferring the index from the object that indexes the target vector to the object that performs the determination of the target vector based on the index. In this regard, the index can be considered as a signal that is transmitted between objects. Indexing can, for example, be carried out during quantization, and the determination of the target vector based on the index, which can be considered as "de-indexation", can then be performed during the dequantization process.

In accordance with a second aspect of the present invention, various processing procedures are performed depending on the set of vectors in which the target vector is contained. If the target vector is contained in a set of vectors that belongs to a predefined group of sets of vectors, then specific processing is applied. Otherwise, general processing is applied.

Specific processing is adapted to the property of a set of vectors that contains the target vector.

Specific processing is configured for the set of vectors that contains the target vector, in accordance with the use of the properties of this set. This allows to improve processing and reduce the overall level of complexity of the procedure for determining the target vector. For example, the structure of a specific treatment can be simplified compared to the general treatment. This measure reduces the overall level of complexity. However, the tuning process may limit the scope of specific processing to cases where the target vectors are contained in sets of vectors with a certain property, i.e., specific processing may, for example, not be applicable in cases where the target vectors are contained in sets of vectors that are not included in a predetermined a group of sets of vectors.

General processing may not be limited to use in cases where the target vectors are contained only in a particular group of vector sets. For example, general processing may be applicable if the target vectors are contained in any set of vectors included in the set of vectors, or in the case where the target vectors are contained in any set of vectors included in the set of vectors, except for sets of vectors, sets of vectors included in a predetermined group. In the latter case, there is the possibility of at least some optimization of the overall processing.

According to an exemplary embodiment of the second aspect of the present invention, general processing when applied to target vectors contained in sets of vectors with a specific property would be more complex compared to the specific processing applied to the same target vectors. In this case, complexity may, for example, be related to computational complexity and / or memory requirements.

According to an exemplary embodiment of the second aspect of the present invention, target vectors contained in a vector set of a predetermined group of vector sets can be more likely to be determined than target vectors contained in a set of vectors not included in the predetermined group of vector sets. Given the total complexity, it may be especially useful that specific processing is less complex than general processing (if applied in the case of target vectors contained in sets of vectors with a specific property). The probability may depend on the application in which the procedure for determining the target vector is performed (for example, the dequantization process). Probability can, for example, be determined offline (offline).

According to an embodiment of the second aspect of the present invention, the vector sets are leader classes, and each leader class contains a different leader vector and permutations of the leader vector.

A property to which specific processing is adapted may, for example, consist in the fact that a predetermined group of leader classes consists only of leader classes with leader vectors whose components take only a predetermined number of different values. This predetermined amount may, for example, be two. In this case, one of two different values may, for example, be zero.

The determination of the target vector can be carried out in the process of dequantization, which is based on a trellised code dictionary defined by leader classes.

The property to which specific processing is adapted may also consist in the fact that a predefined group of leader classes consists only of leader classes with leader vectors whose components take the maximum number of different values. For example, leader classes in a predefined leader group group may contain leader vectors with four values. Processing such leader classes as part of a specific processing process can simplify the overall processing, since then only leader classes that include leader vectors with three or less values can be taken into account during the general processing. This helps to reduce the overall complexity of the de-indexation of input vectors.

For optimization purposes, general processing can be applied if the target vectors are contained in any set of vectors included in the set of vectors, except for the sets of vectors included in a predefined group of sets of vectors.

According to an embodiment of the second aspect of the present invention, the index is associated with the target vector through binomial indexing. Such a binomial indexing procedure is based on the number of possibilities for distributing values by position in a vector of a certain dimension.

In accordance with a third aspect of the present invention, a system is disclosed. The system comprises a first processing component configured to index an input vector contained in a set of vectors included in a plurality of sets of vectors, wherein the first processing component is configured to perform specific processing when the indexed input vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the set of sets of vectors, and the specific processing is adapted to the property of the sets of vectors in a predetermined group npe of sets of vectors and is applicable only in the case of input vectors contained in sets of vectors with a certain property, and, in addition, the first processing component is capable of performing general processing if the input vector to be indexed is not contained in the vector set in advance of a given group of sets of vectors. The system also includes a second processing component, configured to determine the target vector contained in the set of vectors included in the set of vectors based on the index associated with the target vector, while the second processing component is also configured to perform specific processing in the case when the determined target vector is contained in the set of vectors of a predetermined group of one or more sets of vectors included in the set of sets of vectors, moreover, the specific processing of hell is applied to the property of vector sets in a predetermined group of vector sets and is applicable only in the case of target vectors contained in vector sets with a specific property, and, in addition, the second processing component is also capable of performing general processing in case the target vector to be determined, is not contained in the set of vectors of a predetermined group of sets of vectors.

According to a third aspect of the present invention, the above description of the characteristics and advantages of the first and second aspects of the present invention is accordingly applied to the system.

The second processing component can, for example, determine the target vector that is associated with the index assigned to the input vector in the process of indexing the input vector by the first processing component, so that the equality of the input and target vectors is respected. The first and second processing components may, for example, exchange an index.

The first and second processing components can, for example, be part of two different devices, in this case, the input vector is indexed in the device containing the first processing component, and the target vector is determined in the device containing the second processing component, which is associated with the index defined first component of processing.

In addition, the first and second processing components can be part of the same device and can, for example, be used respectively for quantization and dequantization.

These and other aspects of the present invention are explained in the following detailed description. The above features of the present invention and examples of its implementation are intended to be disclosed in all possible combinations with each other.

BRIEF DESCRIPTION OF THE DRAWINGS

The following drawings are attached to the description:

figure 1: a schematic block diagram of a coder G.EV-VBR;

figure 2: the implementation algorithm of the method for indexing the input vector in accordance with an example implementation of the first aspect of the present invention;

3: an algorithm for implementing a method for determining a target vector based on an index associated with a target vector, in accordance with an embodiment of the second aspect of the present invention;

4: a schematic block diagram of a device capable of implementing methods corresponding to the first and second aspects of the present invention.

DETAILED DESCRIPTION OF THE EMBODIMENTS

In the following detailed description of the present invention, examples of its implementation are given in the context of indexing and deindexing lattice nodes associated with the trellis quantizer proposed for the G.EV-VBR codec or its derivative codecs. However, the present invention is not limited to this codec, this particular trellis quantizer or indexing / deindexing trellis structures; it can also be used in all other scenarios that require indexing and deindexing of vectors.

The G.EV-VBR codec standardization process is part of ITU-T SG16 / Q9. This codec is a speech and integrated scalable audio codec for wideband signals with coding rates of 8, 12, 16, 24 and 32 kbit / s.

1 is a schematic block diagram of a G.EV-VBR encoder 1. The first two levels are based on Code Excited Linear Prediction (CELP; Code Excited Linear Prediction) and are referred to as the “base codec” (see block 101 of the base codec). The remaining levels are formed by coding the coefficients of the modified discrete cosine transform (MDCT, Modified Discrete Cosine Transform) (see block 104 MDCT) of the differential signal 103 between the original signal 100 and the synthesized signal of the base codec, while the difference signal is generated by the adder 102. To quantize the MDCT coefficients (for a level corresponding to a 24 kbit / s bit rate) a quantizer 105 is used, the operation of which is based on leader classes obtained from the leader classes of the 2RE8 lattice, which is is a rotated version of the lattice E8, scaled with a factor of 2. The first two levels are then combined with the remaining levels in the bitstream generation unit 106.

The lattice commonly used in quantization can be represented as a union of leader classes. As a result, lattice-based quantization structures can be described as combining leader classes.

The leader vector is an n-dimensional integer vector with ordered components (for example, in descending order):

(ν ₁ ,. ..., ν ₁ , ν ₂ ,. ..., ν ₂ , ν _m ,. ..., ν _m ), where ν ₁ > ν ₂ >...> ν _m , ν _i ∈N.

There are n _i values of v _i . The leader class corresponding to the leader vector includes vectors obtained by all the symbolic permutations of the leader class (with some possible restrictions). Limitations, if applicable, relate to the number of negative components of the vector, which may be odd or even. The number of vectors in the leader class is calculated by the formula

where p determines the parity and can take the value 0 (no restrictions on the sign), +1 (even number of negative components) or -1 (odd number of negative components).

As an example of binomial indexing, consider a 4-dimensional leader vector (2, 1, 1, 0). Suppose that permutations are performed without regard to the sign.

The vectors that belong to the leader class (2, 1, 1, 0) are shown below in table 1, moreover, two indexation options are presented in it.

Table 1 Leader class for leader vector (2, 1, 1, 0) Index Option 1 Option 2 0 2, 1, 1, 0 2, 1, 1, 0 one 1, 2, 1, 0 2, 1, 0, 1 2 1, 1, 2, 0 2, 0, 1, 1 3 1, 1, 0, 2 1, 2, 1, 0 four 2, 1, 0, 1 1, 2, 0, 1 5 1, 2, 0, 1 0, 2, 1, 1 6 1, 0, 2, 1 1, 1, 2, 0 7 1, 0, 1, 2 1, 0, 2, 1 8 2, 0, 1, 1 0, 1, 2, 1 9 0, 2, 1, 1 1, 1, 0, 2 10 0, 1, 2, 1 1, 0, 1, 2 eleven 0, 1, 1, 2 0, 1, 1, 2

From the first option, presented in table 1, it can be seen that in the first group (with indices 0-3, that is, the first four vectors separated by a horizontal line), the relative position of the values 1 and 0 is the same, however, the position of the value 2 changes from vector to vector. The same thing happens in the second group (with indices 4-7), etc. The difference between the first and second groups is how the components with a value of 1 are located in three positions that are available after determining the position for a component with a value of 2. The number of possible placements of these two values in three positions is determined by a binomial coefficient

.

.

вектора.Thus, in the first embodiment there are three groups, and in each group only the position of the value 2 changes, therefore, each of the three groups contains

vector.

Therefore, we can conclude that one of the ways to enumerate vectors in the leader class is to examine alternately the various values in the vector and mark their possible positions. In this example (option 1), the possible positions of the value 2 are first determined, then the positions of the components with the value 1 are determined in the remaining 3 positions. The positions of the component with the value 0 depend on the position of the first two values (since this is the last value), therefore they do not need to be considered separately .

A similar approach is used in option 2, presented in table 1. However, the position of the value 2 does not change in one particular group, but changes in different groups. Accordingly, in this variant four groups appear (with indices 0-2, 3-5, 6-8 and 9-11, respectively). In each of these groups, three vectors are distinguished by the location of two values of 1 in three available positions, i.e.

Table 2 below reproduces the leader vectors defined for the lattice quantizer example proposed for the G.EV-VBR codec when standardized in ITU-T SG16 / Q9 (these leader vectors may differ from standardized leader vectors by a scale factor). Leader vectors (and thus the lattice) have a dimension of eight, and there are 36 leader vectors in total. In this case, for leader vectors with components with odd values, a sign restriction is introduced. In particular, if the sum of the absolute values of all eight components is a multiple of four, then the number of negative components must be even, otherwise it must be odd. For example, the sum of the values of the leader vector with the index 0 (1, 1, 1, 1, 1, 1, 1, 1) is 8, that is, a multiple of four, and thus, in the vectors from the corresponding leader class it should be an even number of values -1. The sum of the values of the leader vector with index 3 (3, 1, 1, 1, 1, 1, 1, 1) is 10, that is, not a multiple of four, so the number of negative components must be odd.

The binomial indexing procedure for a vector belonging to the leader class can be described by the following example of the general indexing code written in ANSI-C using (for modeling a universal digital signal processor (DSP, Digital Signal Processor), for example, sub (), add functions (), L_mult0 (), move16 ()) a library of functions that work with fixed-point values described in ITU G.191 ITU-T Software Tool Library 2005 User's Manual (2005 ITU-T Software Library User Guide) . In this example, code comments appear after two consecutive slash characters ("//") or between a combination of the characters "/ *" and "* /". This indexing code is general, that is, applicable to input vectors contained in any leader class presented in Table 2.

In this indexing code, variables and functions / procedures have the following meanings:

- DIM represents the size of the grating, i.e. 8.

- The variable int no_vals_ [36] stores the number of different values for each leader vector.

- The variable int no_vals_ind_ [36] [4] stores the number of each individual value in each leader vector.

- The variable int vals_ [36] [4] stores values that belong to each leader vector.

- The variable int idx_lead represents the index of the leader class to which the indexed input vector belongs.

- The variable int abs_cv [8] stores the absolute values of the input vector to be indexed.

- The find_pos_elim_fx function (int * abs_cv, int dim, int val, int * p) finds the positions of components with a val value in a dim vector with absolute abs_cv values, places them in the p array and removes them from the abs_cv vector to prepare it for the next values.

- The find_pos_fx function (int * abs_cv, int dim, int val, int * p) is similar to the find_pos_elim_fx function, except that it does not remove components. This function is called for the penultimate value in the leader vector, since the subsequent search for the last value is not performed, and thus, it is not necessary to delete the component with the penultimate value.

- The function int c2idx_fx (int dim, int * p, int no_vals) assigns an index to position no_vals components in the total number of dim possible components. For example:

(x x 0 0), index 0

(x 0 x 0), index 1

(x 0 0 x), index 2

(0 x x 0), index 3

(0 x 0 x), index 4

(0 0 x x), index 5,

where no_vals = 2, DIM = 4, and the position of the two “X” components is indexed at four possible positions.

- The function int number_fx (int no_vals, int * no_vals_ind_, int dim) calculates the number of vectors obtained using unsigned permutations of a vector of dimension dim with no_vals values, the number of occurrences of each of which is represented in the array no_vals_ind_. It corresponds to the formula (1) without the factor 2 ^{m- | p |} .

The functioning of the general indexing code presented above will become apparent from the following example. Consider once again the example of the leader class corresponding to option 2 from table 1. The value of the leader vector in this example is (2, 1, 1, 0), and it is assumed that the leader class is associated with the value idx_lead = 5, although in this an example is not essential. In addition, suppose that the input vector to be indexed has a value of (1, 0, 2, 1), that is, its dimension in this example is less than in the example of a trellis quantizer shown in Table 2, which uses 8-dimensional vectors. However, since the above indexing code is general, it does not matter.

The variables in the indexing code presented above are assigned the following values during initialization:

As a result of the above indexing code, the following results are obtained (after all operations of the WHILE cycle are completed):

i R abs_cv idx pi q index 0 {2} {1, 0, 1} 2 3 3 6 one {0, 2} {0} one one one 7

Thus, finally, as a result of indexing the input vector (1, 0, 2, 1), which is contained in the leader class defined by the variable class idx_lead = 5 and which is shown above as option 2 in table 1, we obtain the index (position) 7 which actually corresponds to the vector (1,0,2,1) in leader class 5.

In the above general indexing code, in addition to the position index, which represents the position of the values in the input vector, information relating to the sign in the nonzero values of the input vector must be encoded. The sign index is calculated as follows:

where S is the number of non-zero components in the vector to be indexed, the value of b _i is 0 if the i-th non-zero component is positive, and 1 otherwise. The value of S is the same for all vectors in the leader class.

The final index I (containing information about the position and sign index, as well as information indicating the leader class that contains the input vector) is calculated as follows:

The variables used in this code have the following meanings:

- int idx_sign_max_bit [36] is a variable in which the number of bits in which the sign index is represented is stored. It actually corresponds to the number of non-zero values in the vector if the parity is zero, or the number of non-zero values without a unit if the parity is +/- 1.

- uint offset_lead [36 + 1] represents the offset of the leader vector for each leader class. This variable actually means that if there are n ₁ vectors in the first leader class, n ₂ vectors in the second, etc., then the offset_lead variable takes the following values: 0, n ₁ , n ₁ + n ₂ , ... Thus, information representing the leader class that contains the input vector can be included in the final index I.

- The final index I is actually formed by placing idx_sign in the last idx_sign_max_bits [] bits of the shifted index, that is, by concatenating the position index bits and idx_sign bits, given that the idx_sign variable can use the maximum idx_sign_max_bit bits, and adding an offset for the leader class to which belongs to the current indexed vector.

According to a first aspect of the present invention, to simplify the general indexing procedure described above, it is proposed to use specific processing if the input vector to be indexed is contained in a leader class that belongs to a predetermined group of leader classes, and otherwise, the use of general processing.

2 is a flowchart 2 illustrating a method for indexing an input vector in accordance with an embodiment of the first aspect of the present invention.

In a first step 200, an input vector is received. The input vector can, for example, be determined during the quantization process as the most accurate correspondence to all vectors in the lattice, which is a union of leader classes. The identification of the leader class that contains the input vector is also taken at step 200. As a few examples of identification, we can cite the leader class index or leader-leader vector of the leader class that contains the input vector.

Then, in step 201, it is determined whether the leader class is contained in a predetermined group consisting of one or more leader classes. If so, then at step 202, specific processing is applied. Otherwise, the general processing is applied in step 203.

Next, examples of a predetermined group of leader classes and specific processing used when indexing the input vectors contained in the leader classes of such a predetermined group are discussed in more detail.

According to a first embodiment of the first aspect of the present invention, specific processing is performed only for the leader class that is associated with the leader vector with index 20 (see table 2), i.e., {6,4,2,0,0,0, 0,0}. Thus, a predefined group of leader classes contains only one leader class, identified by index 20.

In particular, the call to the number_fx () function within the framework of the general indexing code described above can be replaced by the following code:

In this case, bin C coefficients whose dimension does not exceed 8 are stored in the variable C [DIM + 1] [DIM + 1].

In this implementation, specific processing is represented by code that executes when idx_lead is 20, and general processing is represented by code that executes when idx_lead is not 20.

In the context of this embodiment of the present invention, its authors have drawn attention to the fact that the leader class with index 20 is the only leader class with four different meanings. Since the leader vector with index 20 is the only vector with four different values, the no_vals_ind_ table has a dimension of 36 × 4 (36 leader classes and a maximum number of different values of four for one leader class), that is, the value of no_vals_ind [idx_lead] [i] is nonzero only for idx_lead = 20 and i = 3 (i is in the range from 0 to 3). However, in the above general indexing code, the only reference in the number_fx () function to this non-zero value no_vals_ind_ appears for idx_lead = 20. Thus, it is useful to separate the number_fx () function into specific processing (for a leader class with an index of 20 leader vector) and general processing (for the remaining leader classes), in this case, to perform general processing, only a reference to the table no_vals_ind_ with dimensions is required 36 x 3. Entering a specific treatment allows you to optimize the overall processing.

In the context of the second embodiment of the first aspect of the present invention, its authors noted that in some cases of practical use, a small subset of leader classes is much more common than other subsets. For example, for data quantized using a trellis quantizer in the G.EV-VBR codec with a trellis structure, which is based on the vectors presented in Table 2, the leader vector with index 1 will most likely be used, i.e. (2.2 , 0,0,0,0,0,0,0), and usually vectors for which the leader vector has two different meanings (for example, leader vectors with indices 1, 2, 3, 4, 5, 6, 9, and t etc.) will most likely be used more often than other vectors. It was decided that it would be advisable to develop an indexing method that, in particular, would be simpler for these vectors.

To calculate the index for a sign, a sign verification cycle of each component of the vector is required. It is much easier to separately / specifically consider leader vectors with only two different values, since the indexing procedure processes the position of only one value (the first) without further preparatory operations to check the positions of subsequent values. In addition, most leader vectors with only two values have one zero value (see, for example, leader vectors with indices 1, 2, 4, etc. in table 2), therefore, creating a position index can be effectively associated with creating a character index in the same loop.

There are two types of leader vectors with two values, one type has only even values (see, for example, leader vectors with indices 1, 2, 4, 5, etc. in table 2), and the other only odd values (see indices 3, 6, 9, etc.). Most components of the first type have a value of 0, while most components of the second type have a value of 1.

The following specific procedure can, for example, be used to replace all of the above general indexing code in order to process two-digit vectors with even components.

Similarly, the second type (two-digit vectors with odd values) can be separately processed using, for example, the following procedure:

In this second embodiment of the first aspect of the present invention, a predetermined group of leader classes thus contains all leader classes that include only two-digit leader vectors (i.e. leader classes with leader vector indices 1, 2, 3, 4, 5, 6, 9, etc.). Specific processing is represented by two code segments that respectively determine the processing of two-digit leader vectors with even and odd components, and general processing is represented by a general indexing code that is used for all other leader classes (for example, leader classes with leader vector indices 0, 7 , 8, 10, etc. in table 2). It should be noted here that the predefined group is divided in this case into two subgroups, the first subgroup of two-digit leader vectors with even components and the second subgroup of two-digit leader vectors with odd components, and that for each of these subgroups, various actions are performed within the framework of specific processing .

As a first modification of the second embodiment of the first aspect of the present invention, it is also possible to dispense with specific processing for two-digit leader vectors with odd components (for example, with indices 3, 6, 9, etc.) and instead execute a general indexing code. The predefined group of leader classes in this case will contain only leader classes with two-digit leader vectors with even components. Alternatively, specific processing for double-digit leader vectors with even components can be dispensed with.

As a second modification of the second embodiment of the first aspect of the present invention, it is also possible to limit the specific processing to only determining the position index ("index") or sign index ("idx_sign"). Of course, this modification can also be applied to the first modification of the second embodiment of the first aspect of the present invention.

The first aspect of the present invention is focused on indexing the input vector, which, for example, is required in the quantization process. The second aspect of the present invention described hereinafter is focused on de-indexing, that is, determining (or restoring) a target vector based on an index associated with the target vector.

The binomial deindexing procedure for the target vector belonging to the leader class shown in Table 2 can be described using the general deindexation code example below. This code is general, that is, it is applicable to the target vectors contained in any leader class of Table 2. The variables and functions used in this de-indexation code correspond to the variables and functions used in the general indexing code presented above in the context of the first aspect of this inventions. The idx_lead variable, which identifies the leader class, and the index_loc variable, which identifies the position index (which represents the position of the values in the target vector), are transferred as input parameters to the de-indexation procedure, and the target vector cv is determined as the output parameter. The sign of the target vector is not considered here; it can be reconstructed in a separate procedure.

In this case, the function idx2c_fx (int dim, int * p, int no_vals, int index_loc) is complementary to the c2idx_fx function discussed above and returns the position p of the components no_vals in a vector of dimension dim in accordance with index_loc.

In addition, using the variable int occ_pos [8], the positions in which the components of already decoded values were placed are tracked. The variable int offset [8] is used to detect the actual position of the values, because if, for example, one value has already been placed in the 8-dimensional vector, the position index for the next value refers to only seven positions. For example, if for a 4-dimensional leader vector (2 1 1 0) when decoding the position of the value 2, the result is (x 2 x x), and the index position for the value 1 is 0, this corresponds to the positions (0, 1) for two unit values, but units are actually placed at positions (0, 2), since the first value 2 is already at position 1.

3 is a flowchart illustrating a method for determining a target vector based on an index associated with a target vector in accordance with an embodiment of the second aspect of the present invention. This method may, for example, be performed during the dequantization process.

At 300, an index associated with the target vector to be determined is received. An index can, for example, be an index that selects a vector from leader-class vectors, wherein vectors in the leader-class can, for example, be indexed binomially. In addition to the index, leader-class identification, which contains the target vector, is accepted. However, this is not a necessary operation, since the identification of the leader class that contains the target vector may also be included in the index associated with the target vector. The index and leader class identification obtained in step 300 can, for example, be obtained from an index that contains all of the following information: leader class identification, information about which vector in the leader class should be defined as the target vector, and sign information for the target vector.

The target vector to be determined during the execution of the algorithm shown in figure 3, can, for example, be selected in the quantization process as the most accurate match to all vectors in the lattice.

Then, at step 301, it is determined whether the leader class is contained in a predetermined group consisting of one or more leader classes. If so, then at step 302, specific processing is applied. Otherwise, at step 303, the general processing is applied.

Further, examples of a predetermined group of leader classes and specific processing used in the de-indexation of target vectors contained in leader classes of such a predetermined group are considered in more detail.

According to a first embodiment of the second aspect of the present invention, specific processing is performed only for the leader class, which is associated with the leader vector with index 20 (see table 2), that is, {6, 4, 2, 0, 0, 0, 0, 0}. Thus, a predefined group of leader classes contains only one leader class, identified by index 20.

In particular, the first FOR cycle (above the horizontal dashed line) of the general de-indexation code presented above can be replaced by the following code:

In addition, the call to the number_fx () function within the general de-indexation code can be replaced by the following code:

In this case, as already explained above, binomial coefficients whose dimension does not exceed 8 are stored in the variable C [DIM + 1] [DIM + 1].

In this implementation, specific processing is represented by code that executes when idx_lead is 20, and general processing is represented by code that executes when idx_lead is not 20.

In the context of this embodiment of the present invention, its authors have drawn attention to the fact that the leader class with index 20 is the only leader class with four different meanings. Since the leader vector with index 20 is the only vector with four different values, the no_vals_ind_ and vals_ tables have a dimension of 36 × 4 (36 leader classes and a maximum number of different values of four for one leader class), that is, no_vals_ind [ idx lead] [i] and vals_ [idx_lead] [i] exist only for idx_lead = 20 and i = 3 (i is in the range from 0 to 3). However, in the general deindexing code presented above, the only reference to the corresponding values no_vals_indx_ and vals_ appears for idx_lead = 20 in the part of the general deindexation code that has been replaced with a specific code. Thus, now in the remaining general de-indexation code, only a link to the no_vals_ind_ and vals_ tables with a dimension of 36 × 3 is required. Entering a specific processing allows you to optimize the overall processing.

According to a second embodiment of the second aspect of the present invention, a predetermined leader class group contains only a leader class with a leader vector having index 1 (see table 2), i.e., {2,2,0,0,0, 0,0,0}. Instead of performing general de-indexation in any case, i.e. de-indexing, which is applicable to the target vectors contained in any leader class (and which was presented above with reference to the general de-indexation code), for example, the following specific processing can be performed if the target the vector is contained in the leader class with the leader vector having index 1, and general processing can be performed only otherwise. It should be noted that, as in the general de-indexation code presented above, as a result of the following specific processing, only an unsigned target vector cv is generated.

In the G.EV.VBR codec (i.e., for a codebook based on leader vectors defined in Table 2 and speech and music data with bit rates of 16, 24, and 32 kbit / s), the leader vector (2,2, 0,0,0,0,0,0) is the most likely, therefore, the use of specific processing for its leader class reduces the overall level of complexity of the de-indexation procedure.

Figure 4 shows a schematic block diagram of a device 4 capable of implementing methods corresponding to the first and second aspects of the present invention. This device may, for example, be a mobile phone, computer, or other electronic device. The device comprises a central processing unit (CPU, Central Processing Unit) 400, which has access to a program memory 401 and a memory 402. The CPU 400 can, for example, execute a computer program that is stored in a program memory 401. In this case, the program memory 401 can be fixedly installed in the device 4 or be removable. For example, program memory 401 may be implemented as a computer-readable medium on which a computer program is stored. Memory 402 may, for example, be implemented as random access memory (RAM, Random Access Memory). Program memory 401 and memory 402 can also be implemented as a single component.

The CPU 400 also controls the microphone driver, which in turn controls the microphone, the speaker driver 403, which in turn controls the speaker, and an interface 405 through which data can be exchanged with other devices. Interface 405 may be implemented, for example, in the form of a radio interface.

Thus, the device 4 contains all the components required to establish audio communication with another object. The audio signals received by the microphone driver 404 may be encoded in the CPU 400 and transmitted through the interface 405 to another device. In addition, audio data from other devices may be received via interface 405, decoded by the CPU 400, and played back using the speaker driver 403.

In this case, to perform encoding and decoding, the CPU 400 of the device 4 launches a codec, for example, the G.EV-VBR codec. This codec may, for example, be implemented as a computer program stored in a program memory 401. Thus, the device 4 can, for example, implement the structural blocks 101, 104, 105 and 106 shown in figure 1.

The CPU 400 can be considered as a processing component that indexes the input vectors (in accordance with the algorithm shown in FIG. 2) and / or determines the target vector based on the index associated with the target vector (in accordance with the algorithm shown in FIG. 3 ), performing for this purpose a specific and general processing according to the first and second aspects of the present invention. In addition, the codec (or part of it) can be made in the form of specialized equipment 406 (for example, in the form of a digital signal processor (DSP, Digital Signal Processor), a user programmable gate array (FPGA, Field Programmable Gate Array), a specialized integrated circuit ( ASIC, Application Specific Integrated Circuit) or any other specialized equipment), and in this case, specialized equipment 406 can be considered as a processing component that indexes the input vectors and / or determines the target vector based on the index xa associated with the target vector, performing for this purpose specific and general processing according to the first and second aspects of the present invention.

The program code executed by the CPU 400 can also be considered as containing such processing components in the form of functional modules.

The present invention has been described above with examples of its implementation. It should be noted that there are alternative methods and options that are obvious to a person skilled in this technical field, which can be implemented without violating the scope and essence of the attached claims.

Claims (37)

1. The method of indexing the input vector, including:
indexing the input vector contained in the set of vectors included in the set of vectors, the indicated sets of vectors being leader classes, each of which contains a different leader vector and permutations of the indicated leader vector, and the indicated leader classes have signs , and with the specified indexing, a position index representing the position of the values in the specified input vector and a sign index representing the signs of the components of the specified input vector are issued, while the specified indexing includes em:
in the case when the specified input vector is contained in the set of vectors of a predetermined group of one or more sets of vectors included in the specified set of vectors, performing specific processing that is adapted to the property of the specified sets of vectors in the specified predetermined group of vector sets and is applicable only in the case of input vectors contained in sets of vectors with the specified property, the specified specific processing includes the joint determination of the indicated ind in one cycle KSA position and the specified character index, and
performing general processing when the specified input vector is not contained in the set of vectors of the specified predefined group of sets of vectors, while the general processing includes determining, one after another, the position index and sign index for said input vector. 2. The method according to claim 1, characterized in that said property, to which said specific processing is adapted, consists in that said predetermined group of leader classes consists only of leader classes with leader vectors whose components are accepted only in advance set number of different values. 3. The method according to claim 2, characterized in that said predetermined group of leader classes contains a first subgroup of leader classes with leader vectors whose components take only even values, and a second subgroup of leader classes with leader vectors whose components are take only odd values, while during the specified specific processing the actions that are taken if the specified input vector is contained in the leader class of the indicated first subgroup of leader classes differ from actions that accepted if said input vector is contained in the leader class of said second sub-group leader classes. 4. The method according to claim 1, characterized in that the input vectors contained in the set of vectors of the specified predefined group of sets of vectors are indexed more likely than the input vectors contained in sets of vectors not included in the specified predefined group of sets of vectors. 5. The method according to claim 1, characterized in that said general processing, if applied to input vectors contained in said sets of vectors with the indicated property, would be more complicated compared to said specific processing applied to the same input vectors. 6. The method according to claim 1, characterized in that said property, to which said specific processing is adapted, consists in that said predetermined group of leader classes consists only of leader classes with leader vectors whose components take the maximum number different meanings. 7. The method according to claim 1, characterized in that to optimize the specified general processing, this processing is applicable if the input vectors are contained in any set of vectors included in the specified set of vectors, except for the specified sets of vectors included in the specified in advance given group of sets of vectors. 8. The method according to claim 1, characterized in that said indexing is binomial indexing. 9. The method according to claim 1, characterized in that said indexing of said input vector is performed in a quantization process, which is based on a trellis codebook defined by said leader classes. 10. Machine-readable medium on which a computer program is stored, comprising:
instructions, when executed, the processor indexes the input vector contained in the set of vectors included in the set of vectors, and these sets of vectors are leader classes, each of which contains a leader vector and permutations of the specified leader vector, moreover, the indicated leader classes have signs, and with the indicated indexation, a position index representing the positions of the values in the specified input vector and a sign index representing the signs of the components of the specified the input vector, while the specified indexing is performed specific processing in the case when the specified input vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the specified set of vectors, moreover, the specific processing is adapted to the property of these sets of vectors in to a predefined group of sets of vectors and is applicable only in the case of input vectors contained in sets of vectors with the specified property, while the specified specific The processing includes the joint determination in one cycle of the indicated position index and the indicated sign index, and if the specified input vector is not contained in the vector set of the specified predefined group of vector sets, the general processing is performed, while the general processing includes determining one after another a position index and a sign index for said input vector. 11. A device for indexing an input vector, comprising:
a processing component configured to index an input vector contained in a set of vectors included in a plurality of sets of vectors, wherein said sets of vectors are leader classes, each of which contains a leader vector and permutations of the indicated leader vector, wherein the indicated leader classes have signs, and with the specified indexation, a position index representing the positions of the values in the specified input vector and a sign index representing the signs of the components of the specified input are issued vector, the specified processing component is also configured to perform specific processing in the case when the specified indexed input vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the specified set of vectors, and the specific processing is adapted to the property specified sets of vectors in the specified predefined group of sets of vectors and is applicable only in the case of input vectors contained in sets of vectors with the specified In this case, the specified specific processing includes the joint determination in one cycle of the indicated position index and the indicated index of the sign, and, in addition, the specified processing component is capable of performing general processing if the specified input vector to be indexed is not contained in a set of vectors of a specified predetermined group of sets of vectors, wherein the general processing includes determining one after another the position index and sign index for said input vector. 12. The device according to claim 11, characterized in that said property, to which said specific processing is adapted, consists in that said predetermined group of leader classes consists only of leader classes with leader vectors whose components are accepted only in advance set number of different values. 13. The device according to p. 12, characterized in that said predetermined group of leader classes contains a first subgroup of leader classes with leader vectors whose components take only even values, and a second subgroup of leader classes with leader vectors whose components take only odd values, while during the specified specific processing the actions that are taken if the specified input vector is contained in the leader class of the indicated first subgroup of leader classes differ from the actions that e are performed if the specified input vector is contained in the leader class of the specified second subgroup of leader classes. 14. The device according to claim 11, characterized in that the input vectors contained in the set of vectors of the specified predefined group of sets of vectors are indexed more likely than the input vectors contained in sets of vectors not included in the specified predefined group of sets of vectors. 15. The device according to claim 11, characterized in that said general processing, if applied to input vectors contained in said sets of vectors with the indicated property, would be more complicated compared to said specific processing applied to the same input vectors. 16. The device according to claim 11, characterized in that said property, to which said specific processing is adapted, consists in that said predetermined group of leader classes consists only of leader classes with leader vectors whose components take the maximum number different meanings. 17. The device according to claim 11, characterized in that to optimize the specified general processing, this processing is applicable if the input vectors are contained in any set of vectors included in the specified set of vectors, except for the specified sets of vectors included in the specified in advance given group of sets of vectors. 18. The device according to claim 11, characterized in that said indexing is binomial indexing. 19. The device according to claim 11, characterized in that said indexing of said input vector is performed during a quantization process, which is based on a trellis codebook defined by said leader classes. 20. The method of determining the target vector, including:
receiving an index associated with said target vector,
the definition of the target vector contained in the set of vectors included in the set of sets of vectors based on the index associated with the specified target vector, while these sets of vectors are leader classes, each of which contains a different leader vector and permutations of the specified leader vector, the index identifying the leader class and position of values in the target vector, while this definition includes:
performing specific processing in the case where the specified target vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the specified set of vectors, while the specific processing is adapted to the property of these sets of vectors in the specified predefined group of sets of vectors and applicable only in the case of target vectors contained in sets of vectors with the specified property, and
performing general processing if the specified target vector is not contained in the specified set of vectors of the specified predefined group of sets of vectors. 21. The method according to claim 20, characterized in that said property, to which said specific processing is adapted, consists in that said predetermined group of leader classes consists only of leader classes with leader vectors whose components are accepted only in advance set number of different values. 22. The method according to claim 20 or 21, characterized in that the target vectors contained in the set of vectors of the specified predefined group of sets of vectors are more likely to be determined than the target vectors contained in the sets of vectors not included in the specified predefined group of sets vectors. 23. The method according to claim 20 or 21, characterized in that said general processing, if applied to the target vectors contained in said sets of vectors with the indicated property, would be more complicated compared to the specified specific processing applied to the same target vectors . 24. The method according to claim 20, characterized in that said property, to which said specific processing is adapted, consists in that said predetermined group of leader classes consists only of leader classes with leader vectors whose components take the maximum number different meanings. 25. The method according to claim 20 or 24, characterized in that to optimize the specified general processing, this processing is applicable if the target vectors are contained in any set of vectors included in the specified set of vectors, except for the specified sets of vectors included in the specified predefined group of sets of vectors. 26. The method according to claim 20 or 21, characterized in that the specified index is associated with the specified target vector through binomial indexing. 27. The method according to claim 20 or 21, characterized in that the specified definition of the specified target vector is performed in the process of dequantization, which is based on a trellised code dictionary defined by the specified leader classes. 28. Machine-readable medium on which a computer program is stored, comprising:
instructions in which the processor receives an index associated with said target vector, determining a target vector contained in a set of vectors included in a plurality of vector sets based on said index associated with said target vector, wherein said vector sets represent a leader -classes, each of which contains a leader vector different from the others and permutations of the indicated leader vector, and the index identifies the leader class and the position of the values in the target vector, m with the specified definition, specific processing is performed in the case when the specified target vector is contained in the set of vectors of a predetermined group of one or more sets of vectors included in the specified set of vectors, and the specific processing is adapted to the property of the specified sets of vectors in the specified predetermined group sets of vectors and is applicable only in the case of target vectors contained in sets of vectors with the specified property, and if the specified target vector does not contain is stored in a set of vectors of a specified predetermined group of sets of vectors, general processing is performed. 29. A device for determining the target vector, containing:
a processing component configured to receive an index associated with said target vector, determine a target vector contained in a set of vectors included in a plurality of vector sets based on said index associated with said target vector, said vector sets being the leader classes, each of which contains a leader vector different from the others and permutations of the indicated leader vector, the index identifying the leader class and the position of the values in the target vector, while the decree This processing component is also capable of performing specific processing in the case when the specified determined target vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the specified set of vectors, and the specific processing is adapted to the property of these sets of vectors in the specified predefined group of sets of vectors and is applicable only in the case of target vectors contained in sets of vectors with the specified property, and, except th, said processing component is adapted to perform overall processing in case said target vector to be determined is not contained in a set of vectors of said pre-defined group of sets of vectors. 30. The device according to clause 29, wherein the specified property, to which the specified specific processing is adapted, lies in the fact that the specified predefined group of leader classes consists only of leader classes with leader vectors whose components are accepted only in advance set number of different values. 31. The device according to clause 29 or 30, characterized in that the target vectors contained in the set of vectors of the specified predefined group of sets of vectors are more likely than the target vectors contained in the sets of vectors not included in the specified predefined group of sets vectors. 32. The device according to clause 29 or 30, characterized in that said general processing, if applied to the target vectors contained in said sets of vectors with the indicated property, would be more complicated compared to the specified specific processing applied to the same target vectors . 33. The device according to clause 29, wherein the specified property, to which the specified specific processing is adapted, lies in the fact that the specified predefined group of leader classes consists only of leader classes with leader vectors whose components take the maximum number different meanings. 34. The device according to clause 29 or 33, characterized in that to optimize the specified general processing, this processing is applicable if the target vectors are contained in any set of vectors included in the specified set of vectors, except for the specified sets of vectors included in the specified predefined group of sets of vectors. 35. The device according to clause 29 or 30, characterized in that the specified index is associated with the specified target vector through binomial indexing. 36. The device according to clause 29 or 30, characterized in that said procedure for determining said target vector is performed in the process of dequantization, which is based on a trellised code dictionary defined by said leader classes. 37. A system for indexing an input vector and for determining a target vector, comprising:
the first processing component, configured to index the input vector contained in the set of vectors included in the set of sets of vectors, while these sets of vectors are leader classes, each of which contains a different leader vector and permutations of the specified leader vector, moreover, the indicated leader classes have signs, and with the indicated indexation, a position index representing the positions of the values in the specified input vector and a sign index representing the signs of the components of the specified the input vector, while the specified processing component is also configured to perform specific processing in the case when the specified indexed input vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in the specified set of vectors, and the specific processing is adapted to property of the indicated sets of vectors in the specified predefined group of sets of vectors and is applicable only in the case of input vectors contained in sets of vectors in with the specified property, wherein said specific processing includes the joint determination in one cycle of the indicated position index and the indicated sign index, and, in addition, the specified first processing component is capable of performing general processing if the specified input vector to be indexed is not contained in the set of vectors of the specified predefined group of sets of vectors, while the general processing includes determining one after another the position index and sign index for the specified input th vector, and
the second processing component, configured to receive an index associated with said target vector, determine a target vector contained in a set of vectors included in said set of vector vectors based on an index associated with said target vector, wherein said vector sets represent a leader -classes, each of which contains a leader vector different from the others and permutations of the indicated leader vector, and the index identifies the leader class and position of values in the target vector, while said second processing component is also capable of performing specific processing in the case when said specified target vector is contained in a set of vectors of a predetermined group of one or more sets of vectors included in said set of sets of vectors, wherein the specific processing is adapted to the property of said sets of vectors in the specified predefined group of vector sets and is applicable only in the case of target vectors contained in sets of vectors with the specified properties ohm, and, in addition, the specified second processing component is also configured to perform general processing if the specified target vector to be determined is not contained in the vector set of the specified predetermined group of vector sets.

Images