JPS6333935A - Gain/shape vector quantizer - Google Patents

Abstract

PURPOSE:To suppress the information transmission speed low by applying vector quantization to gain information of plural input vectors and sending the result. CONSTITUTION:A shape V//Q coder 5 selects a vector (k) lost similar to a vector j' and sends its index (k). On the other hand, gain information is composed into a gain vector g =(g1, g2,-gn) having an optional length B at the 2nd block device 4 and one vector is formed to plural input vectors. The gain vector is inputted to the gain VQ coder 7 and a vector Gm most similar is selected from a gain code book 8 formed in advance and its index (m) is sent. A shape VQ decoder 9 decodes n-set of shape vectors based on the index (k), and a gain VQ decoder 10 uses an amplifier 14 to output a shape vector to an output terminal 15 every time one gain vector is decoded based on the index (m) and its gain vector information.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to gain/shape and
Regarding vector quantizers.

(Prior Art) Recently, vector quantization has been studied as one of the techniques for information processing and information compression of audio/image signals, etc.

FIG. 2 shows the basic principle of a system using such vector quantization techniques.

The signal sequence 5t (t=1.2.
3...) is a block generator 19 composed of buffers, etc.
, where a block V with arbitrary length b
j ”” (31, Sg, . . . sb) and vectorized. This vectorized signal is input to a vector quantization encoder 20, and input vectors vj
The vector Vk that is most similar to is selected. Then, the index k of the vector Vk is transmitted from the transmission system 22 to the vector quantization decoder 23. At this time, the information transmitted by the vector quantization encoder 20 from the transmission system 22 is
Since only the index of the vector is used, the amount of information transmitted is very slow compared to transmitting the original signal.

On the other hand, when the index k of the vector transmitted from the transmission system 22 is input to the vector quantization decoder 23, the vector quantization decoder 23 searches the code block 24 for the vector Vk, and its component ( s,', 82
'.

...sb') are sequentially output from the output terminal 25.

However, in order to select a vector with an appropriate pattern for every input vector from the codebook 21, it is necessary to have a large number of patterns, so recently the gain/shape (Gain/shape) shown in FIG. 5hape
)・Vector quantization is being considered. Here, the term "shave" refers to the shape of the waveform pattern from which the output vector is determined, and the term "gain" refers to the quantization gain that causes minimum distortion from the input vector with respect to the normalized output vector.

FIG. 3 shows a system using gain/shape vector quantization techniques.

This gain/shape vector quantization system includes:
Shape V Q (Vector Quantizer)
An encoder 26 and a shaved VQ decoder 27 are used. For general gain/shuv vector quantization, see e.g. Vector Quantizati
on: Robert M, Gray IEEB AS
SP Magazine April, 1984 (PP4
~29)'.

That is, a gain normalizer 29 is connected to the system shown in FIG. An amplifier 30 that amplifies the restored signal sequence based on gain information transmitted from the gain normalizer 29 is connected. In a system using such gain/shape vector quantization, all vector patterns stored in the shave codebook 31 and 32 are normalized by the gain, so there is no possibility of pattern failure due to the power of the input signal. The codebook is highly versatile.

(Problem to be Solved by the Invention) However, in the gain/Shuib vector quantization technique described above, not only the vector index but also the gain information must be transmitted for each vector, so the amount of information of the vector index is If there is more gain information compared to , there is a problem that the amount of information transmitted as a whole system increases.

(Means for Solving the Problems) The present invention provides means for vector quantizing shave information formed by blocking an input signal sequence, and means for vector quantizing gain information of a plurality of consecutive input vectors. It is equipped with the following.

(Operation) Shape information obtained by dividing the input signal sequence into blocks is vector quantized, and gain information of a plurality of consecutive vectors is also vector quantized.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configurations of a coding system A, a decoding system B, and a transmission system C using a gain/cheve vector quantizer according to the present invention.

Signal sequence S input to input terminal 1, (i=1°2.3
) is the first blockizer 2, and a certain arbitrary length b
) LiV j = (S, +, St,”
"Sb). This vector vj is input to the gain normalizer 3, where the gain information of the input vector is transmitted to the second blocker 4, and the vector vj normalized by the gain is ' is sent to the sive VQ encoder 5. This gain is, for example, the power value of the input vector.

Then, this shape VQ encoder 5 selects this vector vj from among the shape code blocks 6 created in advance.
′, and transmit its index k.

On the other hand, the gain information is obtained by the second blocking unit 4 as a gain vector gl=(g+, gz
,...g++). One gain vector is created for a plurality of input vectors. This gain vector is input to a gain VQ encoder 7, which selects the most similar vector Gm from a gain codebook 8 created in advance and transmits its index m.

Incidentally, the decoding system B includes a shave VQ decoder 9 for restoring the shave vector and a gain VQ decoder 10 for restoring the gain vector. In addition, these decoders 9 and 10 are provided with a shape code book 11 and a gain code book 12 that have been created in advance in the same way as the shave code book 6 and gain code book 8. In the codebook IL, 12,
The vector to be decoded can be selected using the transmitted index m as a clue.

Buffer 13 has different block lengths for sieve and gain (gain is n times longer than the number of input samples)
Therefore, it is a circuit for synchronous input to the amplifier 14.

Therefore, in the shave VQ decoder 9, the index k
n shape hectors are restored based on the gain V
In the Q decoder 10, each time one gain vector is restored based on the index m, the amplifier 14 amplifies the shave vector based on the information of the gain vector and outputs the amplified vector to the output terminal 15.

In addition to the gain/shape vector quantizer described above, there is also a mean value separate quantizer (Separa quantizer), which is a method of encoding the mean value information and shape information in a block separately.
Of course, it can also be applied to ting Mean VQ).

(Effects of the Invention) As described above, according to the present invention, the gain information of a plurality of input vectors is vector quantized and transmitted, so that the information transmission speed can be kept low.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing an embodiment of the gain/shive vector quantizer according to the present invention, Fig. 2 is a block diagram for explaining the principle of vector quantization, and Fig. 3 is a block diagram of a conventional gain/shape vector quantizer. - It is a block diagram showing a vector quantizer. 2...First blocker 4...Second blocker 5...Shape VQ encoder 6...Shape codebook 7...Gain VQ encoder 8...Gain Codebook Figure 7

Claims (1)

[Claims] 1) A gain/shape characterized by comprising means for vector quantizing shape information formed by blocking an input signal sequence and means for vector quantizing gain information of a plurality of consecutive input vectors.・Vector quantizer.