JPH02215230A - Vector quantizer - Google Patents

Abstract

PURPOSE:To select an accurate representative vector by calculating each distortion of a limited number of representative vectors and an input vector and quantizing the input vector with the number of the representative vector with small distortion. CONSTITUTION:A representative vector storage means 1 stores a representative vector Yti together with its number ti. A distortion arithmetic means 2 reads a number tij of a representative vector designated by the number ti of the representative vector Yti selected in the preceding frame and calculates the distortion di with an input vector X this time. A representative vector selection means 3 selects the vector Yi with the least distortion di calculated by the means 2 and selects the number (i) as a quantized value. Since the distortion of the representative vector of the number tij resident in the vicinity for the number ti of the vector Yi selected precedingly, is calculated, the quantization accuracy is improved without increasing quantized bit number of the like.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a vector quantization device that vector quantizes an A/D converted audio signal or the like.

BACKGROUND ART Conventionally, as shown in FIG. 7, in a vector quantization device of this type, in step 71, an input vector X X= (xt ! , frame.), the input vector X and a predetermined representative vector YI (
i=i, 2. ~． N) Yi ” (yU, yiz l ” + 3'i
n) is calculated for all representative vectors Y1, for example, using the following equation (steps 72→73→74→73).

Next, find i that minimizes the distortion d (step 75).
), the representative vector yi of the number i is determined as the representative vector for the input vector X of that frame, and the number i is set as the quantized value of the input vector X (step 76).
.

FIG. 8 shows the relationship between the input vector X when n=2 and the representative vector Yi when N=4, and the illustrated broken line indicates the boundary between the representative vectors yi.

Here, when setting the representative vector yi in advance, a training vector showing a distribution similar to the distribution of the input vector The vectors representing the center of gravity of the training vectors distributed within each region are designated as representative vectors.

Problems to be Solved by the Invention However, in the conventional vector quantization device described above, in order to improve the quantization accuracy of the input vector of each frame, it is necessary to preset as many representative vectors Yi as possible. Therefore, in this case, the number of bits required to indicate the quantized value (quantization bit number) increases, and the amount of calculation increases because distortion di with many representative vectors yi is calculated. There is a problem.

In view of the above conventional problems, an object of the present invention is to provide a vector quantization device that can improve quantization accuracy without increasing the number of quantization bits or the amount of calculations.

Means for Solving the Problems In order to achieve the above object, the present invention limits and selects a plurality of representative vectors in the vicinity of the previously selected representative vector, and combines the limited number of representative vectors with the input vector. Each distortion is calculated, and the number of the representative vector with the smallest distortion is used as the quantization of the input vector.

Effect: With the above configuration, the present invention has a correlation that the input vector is close to the previously selected representative vector, so even if the number of representative vectors to be calculated is limited, an accurate representative vector can be selected. Further, since representative vectors are limited and selected, even if many representative vectors are set in advance, the number of quantization bits and the amount of calculation do not increase.

EXAMPLES Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 1 is a functional block diagram showing an embodiment of a vector quantization device according to the present invention, and FIG. 2 is an explanatory diagram showing an example of representative vectors stored in the representative vector storage means of FIG. 3 is an explanatory diagram of the operation of the vector quantization device shown in FIG. 1, FIG. 4 is a block diagram showing a specific configuration of the vector quantization device shown in FIG. 1, and FIG. 5 is a diagram explaining the operation of the vector quantization device shown in FIG. and RAM
FIG. 6 is a flowchart for explaining the operation of the CPU shown in FIG. 4.

In FIG. 1, 1 is the representative vector YH Yt+ = (y+t + yi2+ ゝ ytN
) (where i=1, 2. to .N) is the representative vector storage means that is stored in advance together with its number 1+, and in this example, the representative vector Yti for n=6 is stored.

2 reads the representative vector candidate number tij specified by the number 11 of the representative vector Yti selected in the previous frame from the representative vector storage means 1, as shown in FIG. The distortion calculation means 3 which calculates the distortion d is a representative vector selection means which selects the representative vector yi with the smallest distortion di calculated by the distortion calculation means 2, and selects its number i as a quantized value.

Here, input signals such as audio signals have a correlation between each sample, and there is a correlation that the input vector X is in the vicinity of the previously selected representative vector yi. number t
For each i, the distortion with the representative vector of number tij in its vicinity is calculated, so even if many representative vectors Yt are stored in the representative vector storage means 1, the number of quantization bits and the amount of calculations do not increase. , quantization accuracy can be improved.

Next, the specific configuration of the vector quantization device will be explained with reference to FIGS. 4 to 6.

In FIG. 4, 11 is an A/D converter that converts an audio signal into a digital signal, and 12 is an A/D converter that converts the digital signal from the A/D converter 11 into an input vector for vector quantization. This is a central processing unit (CPU) that performs vector quantization by processing as shown in FIG.

Further, 13 is a read-only memory (ROM) in which the execution program of the CPU 12 and each component of the representative vector are stored in advance, and as shown in FIG.
The storage areas 131, 132 and 131 respectively include
Each component (7111y12+~+yxn) of each representative vector Yl (i=1.2°~, N) %()'zt,
y22+〜+yzn) 'i〜(ytx, yi
z+~! yjn) are stored in advance, and the numbers (Fl, ttz+ ゝ, ttm) % of a limited number m of representative vectors Yti in the vicinity of each representative vector Y1 are stored in storage areas 1511152 and ~15i, respectively.
(j21+t22+ −+ tzm) N −(
tit, ti2+ −+ jim) are selected and stored in advance.

14, as shown in FIG. 5, an area 14a for storing the number ti of the representative vector Yti selected when the input vector of the previous frame was quantized, and an area of the ROM 13 used when calculating distortion. An area 14b for storing the start addresses of areas 131 to 13i (representative vector storage addresses) and an area 14b for storing the start addresses of areas 151 to 15i of the ROM 13 (table address of representative vector numbers when calculating distortion). It is a random access memory (RAM) having an area 14c.

Next, referring to FIG. 6, we will explain the operation of the above embodiment, especially the CP.
The operation of U12 will be explained.

In step 61, the input vector X'' (
Xi + X2 + ~, xn), and when the input vector
2, the number ti of the previous representative vector stored in the area 14a of the RAM 14 is referenced, and then, after referring to the table address of the area 14c using this number ti, the vector stored in the corresponding area of the areas 15t to 15i of the ROM 13 is By referring to the number tij, a plurality of representative vectors for which distortion is to be calculated are limited and selected.

Next, areas 13t to 13. of the ROM 13 are stored. Find the start address of each representative vector of the above number tij in R
Each component of the representative address yi is stored in the area 14b of the AM14, and each component of the representative address yi is stored in the area 14b of the ROM 13 using this first address.
31-13. , calculate the distortion between the input vector
Steps 63→64→65→63).

Next, find i for which the distortion dl is the minimum (step 66
), the representative vector yi of the number i is determined as the representative vector for the input vector X of that frame, and the number i is set as the quantized value of the input vector X (step 67
).

As described in detail, the present invention limits and selects a plurality of representative vectors in the vicinity of the previously selected representative vector, and calculates each distortion between the limited number of representative vectors and the input vector. However, since the input vector is quantized using the number of the representative vector with the smallest distortion, the number of representative vectors to be calculated is limited by the correlation that the input vector is near the previously selected representative vector. In addition, since representative vectors are limited and selected, even if many representative vectors are set in advance, the number of quantization bits and the amount of calculation do not increase.

[Brief explanation of the drawing]

FIG. 1 is a functional block diagram showing an embodiment of a vector quantization device according to the present invention, and FIG. 2 is an explanatory diagram showing an example of representative vectors stored in the representative vector storage means of FIG. 3 is an explanatory diagram of the operation of the vector quantization device shown in FIG. 1, FIG. 4 is a block diagram showing a specific configuration of the vector quantization device shown in FIG. 1, and FIG. 5 is a diagram explaining the operation of the vector quantization device shown in FIG. 6 is a flowchart explaining the operation of the CPU in FIG. 4, FIG. 7 is a flowchart explaining the operation of the conventional vector quantization device, and FIG. 8 is a flowchart explaining the operation of the conventional vector quantization device. is an explanatory diagram showing input vectors and representative vectors. DESCRIPTION OF SYMBOLS 1...Representative vector storage means, 2...Distortion calculation means, 3...Representative vector selection means, 11...A/D converter, 12...Central processing unit (CPU), 13...・
Read-only memory (ROM), 14...Random access memory (RAM). Mouth Diagram

Claims (1)

[Claims] A representative vector storage means in which a plurality of representative vectors are stored, a plurality of representative vectors in the vicinity of the previously selected representative vector are selected in a limited manner, and this limited number of representative vectors are read from the representative vector storage means. vector quantization comprising: distortion calculation means for calculating each distortion with an input vector; and means for selecting a representative vector with the smallest distortion calculated by the distortion calculation means and using that number as the quantization of the input vector. Device.