JPH088500B2 - Vector quantizer - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vector quantizer that encodes audio signals, image signals and the like.

2. Description of the Related Art Conventionally, as shown in FIG. 4, this type of vector quantizer conventionally calculates the difference between the input vector input from the input terminal 1 and the reproduction vector from the codebook (memory) 2 (subtractor 3 ), A strain measure is calculated from this difference vector by a square norm or the like (strain calculation unit 4), and the index of the reproduction vector in the codebook 2 when this strain measure becomes the minimum is searched (minimum value detection unit 5). , This index is output from the output terminal 6 as the output of the quantizer.

However, in the above-mentioned conventional vector quantizer, the difference between the input vector and the reproduction vector stored in the codebook 2 is obtained. However, there is a problem that a large-capacity codebook (memory) 2 is required.

Also, the codebook 2 when the strain measure is the minimum
However, if the number of dimensions of the vector is large, there is a problem that the amount of calculation required for searching also becomes large.

In view of the above conventional problems, it is an object of the present invention to provide a vector quantizer capable of quantizing an input vector having a large number of dimensions with a codebook having a relatively small capacity.

Means for Solving the Problems In order to achieve the above object, the present invention detects a dimension of a peak of an input vector, and rotationally shifts the input vector so that the dimension of the peak becomes a predetermined dimension, A reproduction vector having the least distortion with respect to the rotationally shifted input vector is detected from the codebook, and the dimension of the detected peak and the reproduction vector index are output.

Operation With the above configuration, the present invention only needs to store a reproduction vector having a peak in a predetermined dimension in the codebook regardless of the number of dimensions of the input vector. Therefore, a codebook with a relatively small capacity and a large number of dimensions can be used. The input vector can be quantized.

Embodiments Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a vector quantizer according to the present invention, FIG. 2 is an explanatory diagram showing each vector in the vector quantizer of FIG. 1, and FIG. It is explanatory drawing which shows the representative vector stored in the codebook of the figure.

In FIG. 1, 11 is a terminal to which the vector x is input,
Reference numeral 12 denotes a peak detection unit that detects the peak dimension of the input vector x from the terminal 11, and reference numeral 13 denotes an input vector x from the terminal 11 so that the dimension detected by the peak detection unit 12 becomes a predetermined dimension. Is a rotation shift unit that rotates and shifts.

Reference numeral 14 is a codebook in which reproduction vectors are stored. In this codebook 14, for example, as shown in FIG. 3A, a reproduction vector ₁ having a peak in the first dimension is stored.

Reference numeral 15 is a subtracter that calculates the difference between the input vector x ₁ from the rotation shift unit 13 and the reproduction vector ₁ from the codebook (memory) 14, and 16 is the square norm or the like from the difference vector from the subtractor 15. By the strain calculation unit that calculates the strain measure, 17, the minimum value detection unit that searches the index of the codebook 14 when the strain measure calculated by the strain calculation unit 16 is minimum, 18 is the peak detection unit 21. This is a multiplexer that outputs the detected dimension and the index searched by the minimum value detection unit 17 from the output terminal 19 as the output of the quantizer.

Next, the operation of the above embodiment will be described with reference to FIGS. 2 and 3.

As shown in FIG. 2 (e), when an eight-dimensional vector x having a peak in the fifth dimension is input, the peak detection unit
12 detects this dimension, and the rotation shift unit 13 rotates the input vector x by four dimensions in the leftward direction and shifts it so that the peak is in the first dimension, as shown in FIG. 2 (b). Output the vector x ₁ .

The subtractor 15 has an input vector from the rotation shift unit 13,
The difference between the reproduction vectors from the codebook (memory) 14 is calculated, the distortion calculating unit 16 calculates the distortion measure from the difference vector from the subtractor 15 by the square norm, and the minimum value detecting unit 17 calculates the distortion. The reproduction vector ₁ when the distortion measure calculated by the calculating unit 16 is the minimum is searched for the index of the codebook 14 of {Fig. 2 (c), Fig. 3 (a)}, and the multiplexer 18 detects the peak detecting unit. The dimension detected by 12 and the index searched by the minimum value detection unit 17 are output from the output terminal 19 as the output of the quantizer.

Therefore, the decoder side can reproduce the vector as shown in FIG. 2 (d) by the dimension of the peak and the index of the reproduction vector.

That is, in the above embodiment, when the input vector x is eight-dimensional and the rotation shift is performed so that the first dimension of the input vector x becomes a peak, the codebook 14 has
As shown in FIGS. 3 (b) to 3 (h), it is not necessary to store the reproduction vector whose peak is located in the second to eighth dimensions, and thus the reproduction vector whose peak is located in the first dimension is represented as the representative vector. Therefore, the capacity of the codebook 14 can be reduced to 1/8 that of the conventional example.

As described above, according to the present invention, the peak dimension of the input vector is detected, the input vector is rotationally shifted so that the peak dimension becomes a predetermined dimension, and the rotational shift is performed. The reproduction vector that minimizes the distortion with respect to the input vector is detected from the codebook, and the detected peak dimension and the reproduction vector index are output. It is only necessary to store the peak reproduction vector in the codebook, and therefore it is possible to quantize an input vector having a large dimensionality with a codebook having a relatively small capacity.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a vector quantizer according to the present invention, FIG. 2 is an explanatory diagram showing each vector in the vector quantizer of FIG. 1, and FIG. Explanatory diagram showing a representative vector stored in the codebook of the figure,
FIG. 4 is a block diagram showing a conventional vector quantizer. 11 …… input terminal, 12 …… peak detector, 13 …… rotation shifter, 14 …… codebook, 15 …… subtractor, 16 …… strain calculator, 17 …… minimum detector, 18 …… Multiplexer, 19 ... Output terminal.

Claims (1)

[Claims] 1. A codebook in which a reproduction vector whose peak is a predetermined dimension is stored, a means for detecting the peak dimension of an input vector, and a dimension detected by the detecting means is the predetermined dimension. Means for rotationally shifting the input vector, means for detecting from the codebook a reproduction vector that minimizes distortion with respect to the rotationally shifted input vector, dimensions detected by the detecting means, and the codebook And a means for outputting the index of the reproduction vector detected from the vector quantizer.