JP3154781B2 - Vector quantizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vector quantization apparatus for vectorizing a plurality of input signals collectively and quantizing the input signals in a multidimensional signal space.

[0002]

2. Description of the Related Art Generally, in a vector quantization apparatus, a vector x = (x ₁ , x ₂ ,
..., from the total set X of x _K), L pieces of K-dimensional representative vector _{y l = (y l1, y} l2, ..., a set of _y lK) Y =
{Y _l ; l = 1,..., L} can be expressed as Q Q: X → Y (Equation 1). Here, the finite area division S _{i of} X
(I = 1,..., L)

(Equation 1) Shall be satisfied.

Accordingly, an input vector x is mapped to y by a mapping Q and is represented by the following equation.

[0004]

(Equation 2) Hereinafter, a specific configuration example of the conventional vector quantization device will be described.

FIGS. 8 and 9 show the schematic configurations of an encoding unit and a decoding unit, respectively. The encoding unit shown in FIG. 8 includes a vector configuration circuit 81, a distortion calculation circuit 8
2. It is composed of a representative vector dictionary 83, a minimum distortion search circuit 84, and a code word extraction circuit 85, and the decoding device shown in FIG.

Thus, in the encoding unit shown in FIG.
When the input signals are sequentially input from the input terminal 80 to the vector configuration circuit 81, the K-dimensional input vector x = {x _k ; k = 1,..., K} (K =
When the value is 1, a scalar is formed, and the input vector x is supplied to the distortion calculation circuit 82.

The distortion calculation circuit 82 first obtains a distortion d _l between the input vector x and the first representative vector y _l of the representative vector dictionary 83. Here, a square distortion is used as a distortion measure.

In the same manner, the distortion calculation circuit 82 calculates the distortion d _l between the input vector x and the second and subsequent representative vectors y _l ... Of the representative vector dictionary 83. Then, the minimum distortion search circuit 84 searches for the minimum distortion d _l ^* among the d _l obtained by the distortion calculation circuit 82, and outputs a corresponding representative vector y _l ^* to the codeword extraction circuit 85. Is done. The codeword extraction circuit 85 extracts a codeword _l ^* from the representative vector yl ^*, and outputs this from an output terminal 86.

On the other hand, in the decoding unit shown in FIG. 9, is sent from the encoding unit shown in FIG. 8, using a codeword l ^* inputted from the input terminal 90, the representative vector y _l corresponding thereto
^* The searched from the representative vector dictionary 91, so that this representative vector y _l ^* is output from the output terminal 92 as an output signal.

By the way, the performance of the vector quantizer is as follows.
It is determined by the representative vector dictionaries 83 and 91,
The design algorithm of these representative vector dictionaries 83 and 91 is an important problem.

In order to solve such a problem, generally, Lind
The LBG-algorithm reported by E et al. is used. The details of the design algorithm are described in Y.
Linde, A .; Buzou and R.S. M. Gra
y, “An Algorithm for Vecto
r Quantizer Design ”, IEEE
rans. Comm. , COM-28-1, PP84-
95, 1980, and will be briefly described here.

In this case, a sufficiently large training sequence T = T that reflects the statistical properties of the data to be quantized
{T _n ; n = 1, 2,..., N} is prepared, and the training system T and the representative vector _yl set Y =
{Y _l ; l = 1, 2,..., L}

(Equation 3) Is iteratively calculated to find a representative vector dictionary Y that minimizes. When the total distortion value at the m-th time is D _m ,

(Equation 4) A set of Y when the following condition is satisfied is defined as a representative vector dictionary.

On the other hand, separately from such an LBG-algorithm, a self-organizing neural network algorithm (Se
If-Organization Neutral Ne
A representative vector dictionary design algorithm using a Twork (SONN-algorithm) has been reported in the following two documents.

Here, one of the documents is described by T.W. Kohone
n, "Self-Organization and
Associative Memory, "secon
detion, Springer-Verlag, and the other reference is F.S. H. Wu and
K. Ganesan, "Comparative St.
udy of Algorithms for VQ
Design using Conventional
and Neural-Net basedAppr
oaches "Proc. ICASPS; 1989, p.
p751-754.

Since the details of the SONN-algorithm are described in these documents, they will be briefly described here.

In this case, K weights w _l = ｛w _lk ; k
= 1, 2,..., K}, and K input vectors x =
Consider a neuron in which the product sum of {x _k ; k = 1, 2,..., K} and the weight w _l is used as the output η _{l of the} following equation (FIG. 10).

[0017]

(Equation 5) Then, for all input vectors x at this time,
An algorithm for obtaining a set of weight vectors W = {w _l ; l = 1, 2,..., L} such that a particular neuron responds greatly is a SONN-algorithm.
It is realized according to the following procedure.

In this case, consider a situation in which the above-mentioned neurons are regularly arranged. Further, here, for simplicity, a two-dimensional plane shown in FIG. 11 will be described as an example.

A training sequence T = ｛t _n reflecting the statistical properties of the data to be quantized; n = 1, 2,...
N｝ is prepared, and the total distortion value D of W and T is obtained from the following equation.

[0020]

(Equation 6) Then, the change rate of the total distortion values D _i at time _{i (D i-1 -D i} ) / D i <ε ( Equation 8) (where, epsilon: _{constant, i ≧ 1, D 0 =} ∞) the When satisfied, W at this time ends as a weight vector to be obtained. On the other hand, if not satisfied, i = i + 1 (Equation 9)

(Equation 7) Let w _l ^* and w _l (l: codeword of a representative vector included in the neighborhood area NE (i)) be

(Equation 8) Update according to. Such procedure is performed for all the training sequence T, freshly Zen'ibitsuchi D _i and Zen'ibitsuchi rate of change is determined. Then, when the rate of change of the total distortion value satisfies the above (Equation 8), the procedure ends.

Here, NE (i) and α (i) include:

(Equation 9) There is a property.

The set W of weight vectors obtained in this manner is regarded as a representative vector dictionary Y. That is, considering Y = W (Equation 14), the SONN-algorithm can be regarded as an algorithm for designing a representative vector dictionary.

As described above, conventionally, as a method of designing a representative vector dictionary, an LBG-algorithm or a SONN
Algorithm is considered, and according to the representative vector dictionary designed by the LBG-algorithm, it is guaranteed that the training sequence used in the design has a sub-optimal performance with respect to the bias of the statistical distribution. , According to the representative vector dictionary designed by the SONN-algorithm,
Performance similar to that of the representative vector dictionary designed by the BG-algorithm can be obtained.

Accordingly, if the bias of the statistical distribution of the input signal sequence is close to the bias of the statistical distribution of the training sequence, and if the bias does not fluctuate over time, the fixed values designed by these algorithms are used. Can have excellent quantization performance.

[0025]

However, a fixed representative vector dictionary designed by such an LBG-algorithm or a SONN-algorithm is applied to the encoding unit and the decoding unit shown in FIGS. In such a case, when the spectrum of the audio signal is given as the input signal of the encoding unit, the spectrum of the audio signal may have a temporal variation due to the bias of the statistical distribution of the signal sequence due to its property. For this reason, there is a disadvantage that not only the excellent quantization performance by the representative vector dictionary cannot be exhibited, but also a quantization error becomes large.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a vector quantization apparatus capable of realizing high-performance vector quantization with a small quantization error.

[0027]

According to the present invention, there is provided a representative vector dictionary storing a plurality of representative vectors, a vector constructing means for constructing an input vector from an input signal, an input vector constituted by the vector constructing means and the representative vector. A minimum distortion search means for searching for a minimum distortion from each distortion with each representative vector stored in the dictionary; extracting a code word of a representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion search means; Codeword extracting means,
A teacher signal setting unit that sets an input vector configured by the vector configuration unit as a teacher signal, and determines an update area of the representative vector dictionary based on a representative vector corresponding to a codeword extracted by the codeword extraction unit. An encoding unit comprising: an update region determining unit; and an updating unit that updates a representative vector included in the update region determined by the update region determining unit using the teacher signal set by the teacher signal setting unit. ing.

Further, the present invention provides a representative vector dictionary storing a plurality of representative vectors, vector constructing means for constructing an input vector from input signals, and an input vector constructed by the vector constructing means and the representative vector dictionary. Minimum distortion search means for searching for the minimum distortion from the respective distortions with each representative vector, and codeword extraction for extracting the codeword of the representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion search means Means, a teacher signal setting means for setting a representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion searching means as a teacher signal, and a representative corresponding to the codeword extracted by the codeword extracting means. Update area determining means for determining an update area of the representative vector dictionary based on the vector; Representative vectors included in the determined update region is configured from the encoding unit provided with an update means for updating with the teacher signal set from the teacher signal setting means.

Further, according to the present invention, a representative vector dictionary storing a plurality of representative vectors, vector constructing means for constructing an input vector from an input signal, an input vector constructed by the vector constructing means and the representative vector dictionary stored in the representative vector dictionary. Minimum distortion search means for searching for a minimum distortion from each distortion with each representative vector, codeword extraction means for extracting a codeword of a representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion search means Update / initialization / continuation determining means for determining any one of updating, initialization, and continuation in the current state of the representative vector dictionary, and determining the update of the representative vector dictionary by the update / initialization / continuation determining means. A representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion search means is set as a teacher signal. The update area of the representative vector dictionary is updated based on the representative vector corresponding to the codeword extracted by the codeword extracting means by the determination of the representative vector dictionary by the master signal setting means and the update / initialization / continuation determining means. Update area determining means for determining; updating means for updating a representative vector included in the update area determined by the update area determining means using the teacher signal set by the teacher signal setting means; A representative vector dictionary initializing means for initializing the representative vector dictionary by determining the initialization of the representative vector dictionary by the determining means; and a current state of the representative vector dictionary by determining the continuation of the representative vector dictionary by the updating / initializing / continuation determining means. And a means for maintaining the state.

[0030]

As a result, according to the present invention, the teacher signal is set by the input vector composed of the input signal or the representative vector of the representative vector dictionary which minimizes the distortion with respect to the input vector. An update region of the representative vector dictionary is determined by a representative vector corresponding to the code word of the minimum distortion representative vector searched from the representative vector dictionary with the smallest distortion, and the representative vector included in this update region is determined using a teacher signal. Since the update can be performed, it is possible to always maintain a state in which the contents of the representative vector dictionary can follow the temporal change of the bias of the statistical distribution of the input signal sequence.

Further, merely determining whether to update the representative vector dictionary, initialize the initial state to the initial state, or continue the present state, the updating of the representative vector dictionary is started based on this determination. A reset can also be performed.

[0032]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a schematic configuration of an encoding section to which the embodiment is applied. In this case, the encoding unit includes the vector configuration circuit 11, the distortion calculation circuit 12, the representative vector dictionary 1
3, a minimum distortion search circuit 14, a code word extraction circuit 15, a teacher signal setting circuit 17, an update area determination circuit 18, and a representative vector dictionary update circuit 19.

When the input signals are sequentially input from the input terminal 10 to the vector configuration circuit 11, the K-dimensional vector x = ｛x _k ; k when K input signals are obtained.
= 1,..., K｝ (scalar when K = 1)
It is supplied to the distortion calculation circuit 12 and the teacher signal setting circuit 17.

In the distortion calculation circuit 12, the vector configuration circuit 1
Then, a distortion d _l between the input vector x obtained from ₁ and each representative vector y _l of the representative vector dictionary 13 is obtained. Here representative vector dictionary 13 in has a representative vector y _l to # 1~ # L, each representative vector y _l has a K number of elements. Further, as a distortion measure, a square distortion shown by the following equation is used.

[0036]

(Equation 10) When the distortion calculation circuit 12 obtains the distortion d _l for all the representative vectors y _l of # 1 to #L of the representative vector dictionary 13, the minimum distortion search circuit 14 calculates the distortion d _{l in} the distortion calculation circuit 12. Search for the minimum distortion d _l ^* at
The corresponding representative vector _yl ^* is output from the representative vector dictionary 13 to the codeword extraction circuit 15.

The codeword extraction circuit 15 extracts a codeword l ^* corresponding to the given representative vector y _l ^* , outputs the extracted code word l ^* from the output terminal 16, and supplies the code word l ^* to the update area determination circuit 18. Note that the output codeword l ^{* in} this case is

[Equation 11] It is represented by

On the other hand, when the input vector x is given from the vector construction circuit 11 to the teacher signal setting circuit 17, the teacher signal setting circuit 17 outputs the teacher signal τ = ｛τ _k ;
By substituting the input vector x into 1,..., K}, τ = x (Equation 17) is obtained.

Then, the teacher signal τ is supplied to the representative vector dictionary updating circuit 19. On the other hand, the update area determination circuit 1
In 8, the update area is determined using the output codeword l ^* supplied from the codeword extraction circuit 15.

In this state, the representative vector dictionary updating circuit 1
9, the update area determined by the update area determination circuit 18 is updated using the teacher signal τ supplied from the teacher signal setting circuit 17.

If the update end condition is satisfied, the representative vector dictionary update procedure ends, and the process for the next input signal is started. If the condition is not satisfied, the representative vector dictionary update procedure is further continued.

[0042] In this case, the determination of the above-mentioned update region, for example the respective representative vector y _l of #. 1 to # L contained in the representative vector dictionary 13, when arranged on a two-dimensional plane as shown in FIG. 2 , the representative vector y _l ^* having an output codeword l ^* to the code word, and so as to determine a representative vector y _l included in the y _l ^* area near NE as update region.

As a means for determining the update area, a hypersphere centered on a representative vector y _l ^* having an output code word l ^* as a code word in a K-dimensional vector space is considered, and is included therein. a method of the other representative vector y _l and update region, updated in K-dimensional vector space, the output codeword l ^* representative vector y _l distantly close to the ^* with the code word, the representative vector y _l a certain number A method of setting the area is also conceivable.

The time i for determining such an update area is
Assuming that the function in is NE (i), NE (i) has the following properties.

[0045]

(Equation 12) The update of the representative vector y _l (i) included in the update area at time i is represented by the following equation, where y ₁ (i + 1) after the update is set assuming that the teacher signal used for the update is τ.

[0046]

(Equation 13) Here, alpha (i) is a function that gives the ratio of the interior division point between y _l (i) and τ in time i, there are the following properties:.

[0047]

[Equation 14] The representative vector dictionary update procedure ends when one of the following two conditions is satisfied.

[0048]

(Equation 15) Therefore, in this case, the contents of the representative vector dictionary 13 are sequentially updated in accordance with the input signal. Even when a change occurs, the contents of the representative vector dictionary 13 are updated according to the change, so that excellent quantization performance with a small quantization error can be obtained.

Next, a second embodiment of the present invention will be described with reference to FIGS.

FIGS. 3 and 4 show schematic configurations of an encoding unit and a decoding unit to which the second embodiment is applied.

In this case, the encoding unit shown in FIG. 3 includes an input terminal 30, a vector configuration circuit 31, a distortion calculation circuit 32, a representative vector dictionary 33, a minimum distortion search circuit 34, a code word extraction circuit 35, an output terminal 36, The decoding unit shown in FIG. 4 includes a teacher signal setting circuit 37, an update area determination circuit 38, and a representative vector dictionary update circuit 39.
0, a representative vector dictionary 41, a teacher signal setting circuit 42, an update area determination circuit 43, a representative vector dictionary update circuit 44,
An output terminal 45 is provided. The circuits having the same names as those in FIG. 1 have the same functions as those described above, and thus the description thereof will not be repeated.

Here, the encoding unit shown in FIG.
The difference from the first embodiment described above is that the signal used for setting the teacher signal τ is different. That is, in the first embodiment described above, the input vector signal x is used for setting the teacher signal τ, whereas in the second embodiment, the minimum distortion representative vector y _l output from the minimum distortion search circuit 34 is used.
^{* Is} used. Thus, a teacher signal common to the teacher signal in the encoder can be used in the decoder.

In this case, the function of the teacher signal setting circuit 37 is expressed by the following equation.

Τ = _yl ^* (Equation 22) Even in this case, the contents of the representative vector dictionary 33 are sequentially updated according to the input signal, as described with reference to FIG.

[0055] On the other hand, the decoding unit shown in FIG. 4, the codeword l ^* is given corresponding to the minimum distortion above representative vectors y _l ^* to the input terminal 40, the representative vector dictionary by using the code word l ^* 41 representative vectors yl of # 1 to _#L
Are selected from the minimum distortion representative vectors _yl ^* . Then, the minimum distortion representative vector _yl ^* is output from the output terminal 45 as an output signal, and is also provided to the teacher signal setting circuit 42. Then, the teacher signal setting circuit 42
Sets the minimum distortion representative vector y _l ^* to the teacher signal τ, and provides this to the representative vector dictionary updating circuit 44. Also, in the update area determination circuit 43, the data is input from the input terminal 40 in the same manner as the update area determination circuit 18 described above. An update area is determined using the input codeword l ^* , and the updated area is provided to the representative vector dictionary update circuit 44.

Thus, the representative vector dictionary updating circuit 44
Also, the representative vector dictionary 41 is updated until the end condition is satisfied in the same manner as the representative vector dictionary updating circuit 19 described above.

Next, a third embodiment of the present invention will be described with reference to FIGS.

FIGS. 5 and 6 show a schematic configuration of an encoding unit and a decoding unit to which the third embodiment is applied.

In this case, the encoding unit shown in FIG. 5 includes an input terminal 50, a vector configuration circuit 51, a distortion calculation circuit 52, a representative vector dictionary 53, a minimum distortion search circuit 54, a codeword extraction circuit 55, and an output codeword conversion. It is composed of a circuit 56, an output terminal 57, a frame counter 58, a switch circuit 59, a teacher signal setting circuit 60, an update area determination circuit 61, a representative vector dictionary update circuit 62, and a representative vector dictionary reset circuit 63, as shown in FIG. The decryption unit includes an input terminal 62
1, input code word inverse conversion circuit 631, representative vector dictionary 6
4. Frame counter 65, switch circuit 66, teacher signal setting circuit 67, update area determination circuit 68, representative vector dictionary update circuit 69, representative vector dictionary reset circuit 70
It consists of. The circuits having the same names as those in FIGS. 3 and 4 have the same functions as those described above, and thus the repeated description will be omitted.

Here, the encoding unit shown in FIG.
The difference from the second embodiment described in
An output codeword conversion circuit 56 for converting an output codeword output from the output codeword into another codeword is provided, a representative vector dictionary update circuit 62 for updating the representative vector dictionary 53, and the representative vector dictionary 53 are reset (initialized). A representative vector dictionary reset circuit 63;
The update / initialization / continuation determination function for determining whether to update, initialize, or continue using the current state is added.

However, in this case, when the minimum distortion representative vector y _l ^* output from the minimum distortion search circuit 54 is given to the code word extraction circuit 55, the minimum distortion representative vector y _l ^*
* Code word l is output. Then, this code word l
^{The *} is given to the output codeword conversion circuit 56, converted by the conversion function H in the output codeword conversion circuit 56, and output from the output terminal 57 as an output codeword h ^* .

H ^* = H (l ^* ) (Equation 23) On the other hand, the minimum distortion representative vector _yl ^* output from the minimum distortion search circuit 54 is sent to the frame counter circuit 58. When the minimum distortion representative vector _yl ^* is given, the frame counter circuit 58 increments the counter by 1 and generates an output indicating whether or not the counter value is an integral multiple of the set value.

If the counter value of the frame counter circuit 58 is not an integral multiple of the set value, the minimum distortion representative vector y _l ^* output from the minimum distortion search circuit 54 is given to the teacher signal setting circuit 60. Also, an ON signal is supplied to the switch circuit 59. As a result, the output code word l from the code word extraction circuit 55 via the switch circuit 59
^* It is supplied to the update area determination circuit 61, similarly representative vector dictionary 53 as described in Figure 1, so that its contents are updated sequentially according to the input signal. That is, during a period when the counter value of the frame counter circuit 58 is not an integral multiple of the set value, the teacher signal setting circuit 60 and the update area determination circuit 6
1. The representative vector dictionary update circuit 62 is operated, and the representative vector dictionary 53 is updated.

Note that, during a period in which the counter value of the frame counter circuit 58 is not an integral multiple of the set value, the switch circuit 59
, The output code word l ^* is not supplied to the update area determination circuit 61, and the representative vector dictionary 53 can be continuously used while maintaining the current state.

On the other hand, when the count value of the frame counter circuit 58 becomes an integral multiple of the set value, an ON signal is given to the representative vector dictionary reset circuit 63 and an OFF signal is given to the switch circuit 59. As a result, the supply of the output code word l ^* from the code word extraction circuit 55 to the update area determination circuit 61 is stopped, and in this state, the representative vector dictionary 53 is reset to the initial state by the representative vector dictionary reset circuit 63. Become.

Here, the output code word conversion circuit 56 will be described with reference to FIG. In this case, if the output codeword before conversion has a 4-bit length, for example, the representative vectors # 1 to #L included in the representative vector dictionary 53 are arranged on a two-dimensional plane as shown in FIG. In this case, the codewords are numbered with 4 bits in order from the upper left to the lower right. On the other hand, the output codeword after the conversion is the case of the 4-bit codeword shown in FIG. The first two bits divide the two-dimensional plane into four parts as shown in FIG. 3C to represent each area, and the next two bits further divide each area into four parts as shown in FIG. Thus, each area is expressed by using the above-described method, and an output codeword after conversion shown in FIG.

On the other hand, in the decoding section shown in FIG. 6, when the input code word h ^* is given to the input terminal 621,
The input codeword inverse conversion circuit 631 performs an inverse conversion of the codeword to obtain l ^* .

L ^* = H ⁻¹ (h ^* ) (Equation 24) Then, using this code word l ^* , the representative vector dictionary 64
, The minimum distortion representative vector _yl ^* is selected and output from the output terminal 71 as an output signal. At the same time, the minimum distortion vector y _l ^* is
5 given.

In this case, the frame counter circuit 65
When the minimum distortion representative vector _yl ^* is given, the counter is incremented by 1. If the counter value of the frame counter circuit 65 is not equal to an integral multiple of the set value, an ON signal is given to the switch circuit 66, and the The minimum distortion representative vector _yl ^* is given to the signal setting circuit 67. When receiving an ON signal from the frame counter circuit 65, the switch circuit 66 supplies the output code word l ^* to the update area determination circuit 68. As a result, the contents of the representative vector dictionary 64 are sequentially updated. Will be done. That is, while the counter value of the frame counter circuit 65 is not an integral multiple of the set value, the teacher signal setting circuit 67, the update area determination circuit 68, and the representative vector dictionary update circuit 69 are operated to update the representative vector dictionary 64. become.

During a period in which the counter value of the frame counter circuit 65 is not an integral multiple of the set value, the switch circuit 66
, The code word l ^* is not supplied to the update area determination circuit 68, and the representative vector dictionary 64 can be continuously used while maintaining the current state.

On the other hand, when the counter value of the frame counter circuit 65 becomes an integral multiple of the set value, an ON signal is given to the representative vector dictionary reset circuit 70 and an OFF signal is given to the switch circuit 66. As a result, the supply of the output code word l ^{* to} the update area determination circuit 68 is stopped, and in this state, the representative vector dictionary 64 is reset to the initial state by the representative vector dictionary reset circuit 70.

The present invention is not limited to the above-described embodiment, but can be implemented with appropriate modifications without departing from the scope of the invention.

[0073]

According to the present invention, the representative vector dictionary can be updated successively. Therefore, in a case where the bias of the statistical distribution of the signal sequence has a temporal variation such as the spectrum of a voice signal. Also, the contents of the representative vector dictionary can be constantly maintained so as to follow these changes, and high-performance vector quantization with a small quantization error can be realized. In addition, by using a common teacher signal for the encoding unit and the decoding unit, the present invention can be applied to the communication field. Furthermore, by simply determining whether to update the representative vector dictionary, initialize to the initial state, or continue in the current state, the representative vector dictionary can be updated and reset based on this determination. In particular, by periodically resetting the representative vector dictionary, it is possible to prevent the quantization performance from deteriorating due to a communication path error.

[Brief description of the drawings]

FIG. 1 is a diagram showing a schematic configuration of a first embodiment of the present invention.

FIG. 2 is a diagram for explaining update area determination in the first embodiment.

FIG. 3 is a diagram showing a schematic configuration of a second embodiment of the present invention.

FIG. 4 is a diagram showing a schematic configuration of a second embodiment of the present invention.

FIG. 5 is a diagram showing a schematic configuration of a third embodiment of the present invention.

FIG. 6 is a diagram showing a schematic configuration of a third embodiment of the present invention.

FIG. 7 is a diagram for explaining input codeword conversion used in the third embodiment.

FIG. 8 is a diagram illustrating an example of a conventional quantization device.

FIG. 9 is a diagram for explaining an example of a conventional quantization device.

FIG. 10 is a diagram for explaining a design algorithm of a representative vector dictionary.

FIG. 11 is a diagram for explaining a design algorithm of a representative vector dictionary.

[Explanation of symbols]

11, 31, 51... Vector configuration circuits. 12, 32, 5
2 ... distortion calculation circuit, 13, 33, 41, 53, 64 ... representative vector dictionary, 14, 34, 54 ... minimum distortion search circuit, 1
5, 35, 55 ... code word extraction circuit, 17, 37, 42,
60, 67: teacher signal setting circuit, 18, 38, 43, 6
1, 68... Update area determination circuit, 19, 39, 44, 6
2, 69 ... representative vector dictionary update circuit, 56 ... output code word conversion circuit, 58, 65 ... frame counter circuit, 5
9, 66 ... switch circuit, 63, 70 ... representative vector reset circuit, 631 ... output code word inverse conversion circuit.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-59-183542 (JP, A) JP-A-62-166655 (JP, A) JP-A-2-186836 (JP, A) JP-A-1- 232829 (JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) H03M 7/30

Claims (5)

(57) [Claims] 1. A representative vector dictionary storing a plurality of representative vectors, vector constructing means for constructing an input vector from an input signal, and an input vector constructed by the vector constructing means and each of the representative vectors stored in the representative vector dictionary. Minimum distortion search means for searching for the minimum distortion from the respective distortions with the representative vector; codeword extraction means for extracting the codeword of the representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion search means A teacher signal setting unit that sets an input vector configured by the vector configuration unit as a teacher signal; and an update area of the representative vector dictionary based on a representative vector corresponding to a codeword extracted by the codeword extraction unit. Update area determining means for determining the update time, and a representative vector included in the update area determined by the update area determination means. A coding unit comprising: an updating unit that updates a file using a teacher signal set by the teacher signal setting unit. 2. A representative vector dictionary storing a plurality of representative vectors, vector constructing means for constructing an input vector from an input signal, and an input vector constituted by the vector constructing means and each of the representative vectors stored in the representative vector dictionary. Minimum distortion search means for searching for the minimum distortion from the respective distortions with the representative vector; codeword extraction means for extracting the codeword of the representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion search means Teacher signal setting means for setting a representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion search means to a teacher signal; and a representative corresponding to the codeword extracted by the codeword extracting means. Update area determining means for determining an update area of the representative vector dictionary based on a vector; An encoding unit comprising: an updating unit that updates a representative vector included in the determined update area using the teacher signal set by the teacher signal setting unit. A representative vector dictionary storing a plurality of representative vectors; and a teacher signal for setting a representative vector of the representative vector dictionary corresponding to the codeword extracted by the codeword extracting means of the encoding unit as a teacher signal. Setting means; updating area determining means for determining an updating area of the representative vector dictionary based on a representative vector corresponding to a codeword extracted by the codeword extracting means of the encoding section; and determining by the updating area determining means. 3. The vector quantization apparatus according to claim 2, further comprising: a decoding unit including: an updating unit configured to update the representative vector included in the updated update region using the teacher signal set by the teacher signal setting unit. . 4. A representative vector dictionary storing a plurality of representative vectors, vector constructing means for constructing an input vector from an input signal, and an input vector constructed by the vector constructing means and each of the representative vectors stored in the representative vector dictionary. Minimum distortion search means for searching for the minimum distortion from the respective distortions with the representative vector; codeword extraction means for extracting the codeword of the representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion search means Update / initialization / continuation determination means for determining any one of update, initialization, and continuation in the current state of the representative vector dictionary; and updating of the representative vector dictionary by the update / initialization / continuation determination means. The instruction for setting the representative vector of the representative vector dictionary corresponding to the minimum distortion searched by the minimum distortion searching means to the teacher signal by the determination. A signal setting unit, and an update area of the representative vector dictionary is updated based on a representative vector corresponding to a codeword extracted by the codeword extracting unit by the update / initialization / continuation determining unit. Updating area determining means for determining; updating means for updating a representative vector included in the updating area determined by the updating area determining means using a teacher signal set by the teacher signal setting means; A representative vector dictionary initializing means for initializing the representative vector dictionary by determining the initialization of the representative vector dictionary by the / continuation determining means; and a representative vector by determining the continuation of the representative vector dictionary by the update / initialization / continuation determining means. A vector quantization apparatus, comprising: an encoding unit including means for maintaining a current state of a dictionary. 5. A representative vector dictionary storing a plurality of representative vectors, update / initialization / continuation determining means for determining any one of update, initialization, and continuation in the current state of the representative vector dictionary; A teacher signal for setting a representative vector of the representative vector dictionary corresponding to the codeword extracted by the codeword extracting means of the encoding unit as a teacher signal by the update / initialization / continuation determining means determining the update of the representative vector dictionary. Setting means; and determination of updating of the representative vector dictionary by the update / initialization / continuation determining means, based on the representative vector corresponding to the codeword extracted by the codeword extracting means of the encoding unit. Update area determining means for determining an update area; and a representative vector included in the update area determined by the update area determining means is set to the teacher signal setting. Updating means for updating using a teacher signal set by a stage; and a representative vector dictionary initializing means for initializing the representative vector dictionary by determining the initialization of the representative vector dictionary by the updating / initializing / continuation determining means. 5. The vector quantum according to claim 4, further comprising: means for maintaining the current state of the representative vector dictionary by determining the continuation of the representative vector dictionary by the update / initialization / continuation determining means. Device.