JPH0219480B2 - - Google Patents

Abstract

PURPOSE:To attain coding with high efficiency and a high speed by separating a block average value and normalizing the average value by means of the amplitude in the block and using a set of different output vectors according to the change in the amplitude in a block so as to apply vector quantization. CONSTITUTION:The average value in the block is separated by an average value calculating device 14 from an input vector X formed by blocking an input signal train at each K trains, a standard deviation component is obtained by an amplitude calculator 18 as the amplitude in the block and an average value separation normalizing input vector is obtained by dividing the component by a divider 20. The average value separation normalizing output vector yi is divided in plural code tables 23. A maximum element distortion detector 8 obtains a distortion di of the X and yi from a parallel subtractor 6 and a parallel absolute value operating device 7 and a minimum distortion detector 9 obtains a minimum value (d) of the distortion di. The vector indiating the minimum distortion is a vector quantization output.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to vector quantization, which blocks a plurality of input signal sequences and quantizes them in a multidimensional signal space. First, the principle of vector quantization will be briefly explained. Now, K information source input signal sequences are put together into an input vector X = [x ₁ , x ₂ , . . . , x _K ]. At this time, the K-dimensional Euclidean signal space R ^K ( X ∈
R ^K ) N representative points (i.e. output vector)
Set y _i = [y _i1 , y _i2 , ..., y _iK ] as Y = [ y ₁ , y
₂ ,... _, yN ]. Each division of R ^K with the output vector y _i as the representative point (for example, the center of gravity) is R ₁ , R ₂ ,...,
If R _N , vector quantization Q is defined by the following equation. Q: R ^K →Y Here, R _i =Q ^-1 ( y _i )={X∈R ^K :Q ( X )= y
_i } _N ∪ ⁱ⁼¹ R _i =R ^K , R _i ∩R _j =φ (i≠j) The above vector quantization Q is expressed as a cascade connection of encoding C and decoding D. Encoding C is the mapping of the set Y = [ y ₁ , y ₂ , ..., y _N ] of the output vectors of R ^K to the index set I = [1, 2, ..., N], and the decoding D is This is mapping from I to Y. That is, C: R ^K → I, D: I → Y and Q = D・C
It is. In vector quantization, since the encoded output I is transmitted or recorded, the encoding efficiency is extremely high. Vector quantization is the mapping of an input vector to the output vector y _i that is the shortest distance (resulting in the least distortion). Specifically, if the distance (distortion) between the input and output vectors is d( X , yi ₎ , then the following equation is obtained. _{if d} ( X , yi ₎ < d ( X , yj _{) for} all _j A set Y of output vectors y _i as shown in Figure 1 is obtained by clustering (selection of representative points and division of signal space until the sum of distortions is minimized) using an information source input signal sequence that becomes a training model. ). The configuration of a conventional vector quantizer will be described below with reference to FIGS. 2 and 3. In the figure, 1 is an input vector, 2 is a register, 3 is a code table address counter, 4 is an output vector code table memory, 5 is a code table output vector register, 6 is a parallel subtracter, 7 is a parallel absolute value calculator, 8 9 is a maximum element distortion detector, 9 is a minimum distortion output vector detector, 10 is an index signal, 11 is an index latch, 12 is an encoder output signal, and 13 is an output vector. Next, the operation will be explained. K input signal sequences of the encoder are collectively converted into blocks and taken into the input vector register 2 as input vectors X = {x ₁ , x ₂ , . . . , x _K }. At this point, the address counter 3 is sequentially counted up to i=1, 2, ..., N, and the output vector y _i =[y _i1 ,
y _i2 ,..., y _iK ] from code table 4,
Latch to output vector register 5. For each output vector y _i , the parallel subtracter 6, parallel absolute value calculator 7, and maximum element distortion detector 8 calculate the maximum element distortion d _i of the input/output vector by the following calculation. ^d _i = d _{( _ _ _} _ _{_} . The minimum distortion d is d = min ⁱ d( X , y _i ) = min ⁱ [ max ^j |x _j −y _ij |]. The minimum distortion detector 9 calculates the distortion d ( X , yi _{) of} the output vector yi and the input vector At each time, this is stored as a new minimum distortion, and a strobe signal is sent to the index latch 11 each time, and the index signal 10, which is the code table address of the output vector, is taken into the index latch 11. If the above procedure is continued until all output vectors y _i are read out from the code table 4 (i = 1 to N), the index i of the output vector resulting in the minimum distortion will remain in the index latch 11 when the full search is completed. This becomes the encoder output 12. Next, the operation of the decoder shown in FIG. 3 will be explained. The encoder output signal 12 sent from the encoder is taken into the index latch 11, the address is specified by the index signal, and the output vector y _i is read out from the code table 4.
It is latched into the output vector register 5 and an output vector 13 is obtained as the decoder output. Since the conventional vector quantizer is constructed as described above, it has the following drawbacks. That is, each element x _j (j=1,
2,...,K) takes a characteristic distribution in each input sequence when the information source model is different in the K-dimensional signal space. In addition, when the number of output vectors is increased in order to perform high-performance vector quantization with less quantization noise, the search time required for each vector also increases. This invention was made to eliminate these drawbacks.The block average value is separated from the input vector X , and the vector ^X ' is normalized by the intra-block amplitude. By performing vector quantization using multiple code tables, quantization efficiency is high.
The present invention also aims to provide an adaptive vector quantizer that reduces processing time (search time). An embodiment of the present invention will be described below with reference to the drawings. FIG. 4 shows an example of the configuration of an encoder. In the figure, 1 is an input vector, 14 is an average value calculator, 15 is a block average value calculated by the average value calculator 14, 16 is a subtracter that subtracts the block average value 15 from the input vector 1, and 17 is a subtracter. The mean value separation vector 18 is the mean value separation vector 1 from which the block mean value 15 is subtracted by the subtractor 16.
an amplitude calculator that calculates the amplitude within the block from 7;
19 is the intra-block amplitude obtained by the amplitude calculator 18; 20 is a divider that divides the average value separation vector 17 by the intra-block amplitude 19; 21 is the average value separation normalized vector that is the output of the divider 20;
is a register, 22 is a code table switch, 23
is a code table switching signal, 3 is a code table address counter, 24 is code table memory 1, ..., T (T is plural), 5 is a code table output vector register, 6 is a parallel subtracter, 7 is a parallel absolute value calculator , 8 is the maximum element distortion detector, 9 is the minimum distortion vector detector, 10 is the index signal, 25
is a latch, and 12 is an encoder output signal. Further, FIG. 5 shows an example of the configuration of the amplitude calculator 17. In the figure, 26 is an absolute value calculator;
7 is an average value calculator. Further, FIG. 6 shows an example of the configuration of the decoder. In the figure, 28 is a mean value separation normalized output vector, 29 is a multiplier,
30 is an average value separation output vector, 31 is an adder,
32 is an output vector, 33 is an output vector register, and 34 is a decoder output signal. Next, the operation will be explained. First, the operation of the encoder will be explained with reference to FIGS. 4 and 5. The input signal sequence of the encoder is grouped into K blocks and the input vector X {x ₁ , x ₂ , ..., x _K }
becomes. For input vector 1, average value calculator 1
4 is the block average value m15 by m=E(x _j )
Calculate. Subtractor 16 subtracts block mean value m from input vector 1 to obtain mean value separation vector 17 represented by x _j -m. The amplitude calculator 18 is
Calculate the intra-block amplitude σ from the mean value separation vector. There are various methods to calculate σ, but the following two
-3 examples are shown below. σ=E(|x _j −m|) σ=[E(x _j −m) ² ] ^1/2 σ= max ^j x _j − min ⁱ x _jFigure 5 shows an example of the configuration using the above method. Give an example. The absolute value of the mean value separation vector 17 is determined, and the mean value calculator 27 determines σ. The divider 20 normalizes the input vector by dividing the mean value separation vector 17 by the intra-block amplitude σ to obtain the mean value separation normalized input vector X' . That is, X′ _j =(x _j −m)/σ In this invention, the mean value separation normalized vector is vector quantized. By this, the vector is K
It is randomly distributed on a sphere centered on the origin of the dimensional signal space, increasing the quantization efficiency for different sequences. In other words, it is a mixture with scalar quantization. Vector X′ and quantization vector at this time
The relationship with Y' is shown in Figure 7. In addition, there is a correlation between the pattern shown by the mean value separation normalized vector and the intra-block amplitude; vectors with small amplitudes often have gentle slope vectors, whereas vectors with large amplitudes have many steep rising components. , the basic structure patterns of the signals are different in both cases. Therefore, the average value separated normalized output vector is divided into a plurality of code tables according to the intra-block amplitude. This is code table 1~T
It is 24. The code table switch 22 compares the intra-block amplitude 19 with a preset threshold value and determines which code table to refer to. Next, the code table address counter 3 is
From the selected code table, the mean value separation normalized output vector y′ _i is sequentially
is read and latched into register 5. Since only the selected code table is searched, processing time is shortened. The maximum element distortion detector 8 obtains the distortion d _i of X' and y' _i from the parallel subtracter 6 and the parallel absolute value calculator 7 as follows. d _{i =} d ( X′ , y′ _{i )} = max ^j | I don't mind. Next, the minimum distortion detector 9 sequentially reads out y′ _i
Detect the minimum value of distortion d _i between and X′ . That is, the minimum distortion d is d=min ⁱ [max ^j |x _j −y′ _ij |]. The vector with this minimum distortion is the vector
This is the vector quantization output of X′. When the minimum distortion vector is detected, a strobe signal is sent to latch 25, which takes in index signal 10, which is the address of the vector. The latch 25 also takes in the block average value 15 and the intra-block amplitude 19, and outputs them as the encoder output signal 12. Next, the operation of the decoder shown in FIG. 6 will be explained. The encoder output signal 12 is taken into a latch 25 of the decoder, and a code table switch 22 outputs a code table switch signal 23 according to the intra-block amplitude 19. The latch 25 also receives the block average value 15 and the index signal 10. The average value separated normalized output vector y'i read from the code table selected by the code table switching signal 23 according to the address _indicated by the index signal 10 is latched in the register 5. The vector y' _i 28 is multiplied by the intra-block amplitude 18 in a multiplier 29, and then added to an adder 3.
1, the block average value 15 is added, resulting in an output vector y _i 32, which is latched in a register 33 to obtain a decoder output signal 34. That is, y _j =σ・y′ _j +m The coding efficiency η of the adaptive vector quantizer according to the present invention is as follows: One code table has N vectors, and the block average value and intra-block amplitude each have a bits. , b bits are allocated, then η=(log ₂ N+a+b)/K (bits/sample). As described above, the adaptive vector quantizer according to the present invention separates the block average value, normalizes it by the intra-block amplitude, and then converts the vector quantizer using different sets of output vectors according to changes in the intra-block amplitude. By doing so, it is possible to realize highly efficient and high-speed encoding that adaptively follows changes in the amplitude distribution characteristics of the input signal.

[Brief explanation of drawings]

Figure 1 is an explanatory diagram showing the arrangement of output vectors in a conventional vector quantizer, Figure 2 is a configuration diagram of a conventional vector quantization encoder, and Figure 3 is the configuration of a conventional vector quantization decoder. 4 is a block diagram showing an embodiment of the vector quantization encoder according to the present invention, FIG. 5 is a block diagram showing an embodiment of the amplitude calculator in FIG. 4, and FIG. FIG. 7 is a block diagram showing an embodiment of a vector quantization decoder according to the invention. FIG. 7 is an explanatory diagram showing an arrangement of output vectors in vector quantization according to the invention. In the figure, 1 is an input vector, 2 is a register, 3 is a code table address counter, 4 is an output vector code table memory, 5 is a code table output vector register, 6 is a parallel subtracter, 7 is a parallel absolute value calculator, 8 9 is a maximum element distortion detector, 9 is a minimum distortion output vector detector, 10 is an index signal, 11 is an index latch, 12 is an encoder output signal, 13 is an output vector, 14 is an average value calculator, and 15 is a block. Average value, 16 is subtractor, 1
7 is a mean value separation vector, 18 is an amplitude calculator, 1
9 is an amplitude within a block, 20 is a divider, 21 is an average value separation normalization vector, 22 is a code table switch, 23 is a code table switch signal, 24 is a code table memory 1, ..., T (T is plural), 25
is a latch, 26 is an absolute value calculator, 27 is an average value calculator, 28 is an average value separation normalized output vector, 2
9 is a multiplier, 30 is an average value separation output vector, 3
1 is an adder, 32 is an output vector, 33 is an output vector register, and 34 is a decoder output signal. Note that the same or corresponding parts in the figures are indicated by the same reference numerals.

Claims (1)

[Claims] 1. An average value separation vector is obtained by subtracting the average value within a block from an input vector formed by dividing the input signal sequence into K blocks (K is a plurality of blocks). A mean value separation and normalization circuit that obtains the standard deviation component of an element as an intra-block amplitude, normalizes the mean value separation vector with this intra-block amplitude, and forms a mean value separation normalized vector; Based on the distribution, a plurality of sets of mean value separation normalized output vectors with the minimum distortion are obtained in advance and stored in a plurality of code table memories, and the in-block amplitude obtained by the mean value separation normalization circuit is a code table switcher that selects one of the plurality of code table memories based on the selected code table memory; a code table address counter that sequentially reads an average value separation normalized output vector from the selected code table memory; and an average value separation normalized input. a distortion calculator that calculates and compares distances in a K-dimensional signal space between the vector and the vectors read out sequentially to detect an average value separation normalized output vector having the shortest distance; a code table address of the vector having the shortest distance; an encoder that encodes the intra-block average value and the intra-block amplitude; and based on the encoder output signal, reads out the average value separation normalized output vector at the shortest distance from the plurality of code table memories. a decoder, a multiplier that multiplies the average value separated normalized output vector by the intra-block amplitude, and an adder that adds the intra-block average value to the vector multiplied by the intra-block amplitude to obtain an output vector. An adaptive vector quantizer characterized by: 2. An amplitude calculator which determines the intra-block amplitude to be the average of the absolute values of the deviations from the intra-block average value of the elements constituting the vector when average-separating and normalizing the vector in adaptive vector quantization; 2. The adaptive vector quantizer according to claim 1, further comprising a code table memory prepared according to the definition of intra-block amplitude. 3. An amplitude calculator that determines the intra-block amplitude as the difference between the maximum and minimum values of the elements constituting the vector when average-separating and normalizing a vector in adaptive vector quantization, and a definition of the intra-block amplitude. 2. An adaptive vector quantizer according to claim 1, further comprising a code table memory prepared according to the following. 4 In vector quantization, 2 are compared sequentially
The present invention is characterized by being equipped with a distortion calculator that searches the code table by approximating the distance in the K-dimensional signal space between two vectors by the largest value among the absolute values of the differences between each element. An adaptive vector quantizer according to claims 1, 2, and 3.