JPH07120959B2 - n-ary tree search vector quantizer - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a vector quantizer for high-efficiency signal coding.

[Prior Art] FIG. 2 is a block diagram of an n-ary tree search vector quantizer shown in Japanese Patent Application No. 62-10487, and FIG.
FIG. 4 is a detailed block diagram of a vector quantizer using the inner product operation shown in -10487, and FIG. 4 is an explanatory diagram showing the relationship between the tree search and the index code shown in Japanese Patent Application No. 62-10487.

In FIG. 2, (1) is an input vector whose average value is separated, (2) is a vector quantizer of each stage, (3) is an output vector index, (4) is a codebook of each stage,
(5) is the amplitude gain, and in FIG. 3, (6) outputs the address (12) of the output vector to be sequentially read according to the output vector index (3) output from the vector quantizer (2) in the preceding stage. Address counter, (7) multiplier, (8) inner product value, (9) maximum inner product detector, (1
0) is an amplitude gain latch that latches the inner product value as an amplitude gain, (11) is an index latch that latches the codebook address and outputs it as an index, and (13) is an adder.

Next, the operation will be described.

The input vector (1) whose average value has been separated in FIG. 2 is quantized by the vector quantizer (2) as follows. First, the output vector is an n-ary (n is a power of 2) tree search. FIG. 4 shows an example of a binary tree and a part of a quaternary tree. The code of the index attached to the output vector of each node of the tree structure is determined as shown in FIG. 4, for example.

The codes of the quaternary tree in FIG. 4 (b) are shown in binary numbers in the upper row and quaternary numbers in the lower row. As can be seen from this figure, each time the number of stages in the n-ary tree structure increases by one, log ₂ n bits (1 in n-ary number)
By increasing the code by (digits), and truncating the code by log ₂ n bits from the LSB, the node of the tree structure can be reversely climbed from the lower stage to the upper stage.

The input vector (1) whose average value is separated by searching the output vector codebook as described above up to the final stage is vector-quantized using the inner product as the distortion measure.
The tree search vector quantizer is represented by the connection of vector quantizers as shown in FIG. 2, and the vector quantizer at the m-th stage is shown in the block diagram of FIG.

In FIG. 3, when the output vector index (3) is input to the (m-1) th stage vector quantizer, the address counter (6) compares the input vector (1) with the n-ary tree structure output vector. Outputs the address (12) of the output vector to be performed. When the output vector is read from the codebook (4), the product sum is calculated with the input vector (1). When the inner product value (8) of the vector thus obtained becomes maximum with respect to the input vector, the maximum inner product detector (9) generates a strobe. This strobe causes the amplitude gain latch (10) to have an inner product value (8).
However, the output vector address (12) is taken into the index latch (11), and when the address counter (6) finishes outputting all the output vector addresses (12) for the input vector (1), it is stored in the index latch (11). The fetched value is output as the output vector index (3) of the m-th stage. When the m-th stage is the final stage, the amplitude gain (5) of the amplitude gain latch (10) is also output.

[Problems to be Solved by the Invention] Since the conventional n-ary tree search vector quantizer is configured as described above, the calculation of the inner product is performed by n at each stage of the tree search.
There was a problem that it was necessary to do it once.

The present invention has been made in order to solve the above problems, and it is possible to perform the inner product operation of the same stage of the tree search performed by each vector quantizer at a high speed no more than n times. An object of the present invention is to obtain an n-ary tree search vector quantizer capable of switching whether to perform the inner product calculation in each vector quantizer n times or less or n times.

[Means for Solving the Problem] An n-ary tree search vector quantizer according to the present invention inputs an input vector and an index code, and performs a tree search on an n-ary tree search codebook based on this index code. A plurality of vector quantizers that output the maximum inner product value at that time together with the index code of the above output vector that calculates the inner product of the read next stage output vector and the above input vector and gives the maximum inner product, and a set of vector quantizers High-speed mode in which the inner products of the same stage of the tree search are shared and calculated by each vector quantizer in one set obtained by dividing the above-mentioned plurality of vector quantizers so that there are two or more , A switching signal for switching between the normal mode for calculating the inner product of the same stage of the tree search, the inner product value output from each vector quantizer in the one set, and the index And a comparison selector for outputting the index code from the vector quantizer that has output the maximum inner product value.
Each vector quantizer in the set is calculated by sharing the inner product in the same stage of the tree search, and outputs the maximum inner product value at that time together with the index code of the output vector giving the maximum inner product to the comparison selector, When the switching signal is in the normal mode, one of the vector quantizers calculates the inner product in the same stage of the tree search, and outputs the maximum inner product value at that time together with the index code of the output vector giving the maximum inner product to the next stage of the tree search. It is output to the vector quantizer for processing.

[Operation] In the n-ary tree search vector quantizer according to the present invention, when the switching signal is in the high speed mode, the vector quantizers in the one set share the inner product in the same stage of the tree search. The maximum inner product value at that time, which is calculated and gives the maximum inner product, is output to the comparator / selector, and when the switching signal is in the normal mode, one of the vector quantizers has the same stage of tree search. The inner product in is calculated, and the maximum inner product value at that time is output to the vector quantizer which performs the next stage processing of the tree search together with the index code of the output vector which gives the maximum inner product.

[Embodiment] An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an n-ary tree search vector quantizer showing an embodiment of the present invention. (1) is an input vector whose mean value is separated, and (2) is a vector quantizer at each stage. , (3) is the output vector index, and (4) is the output vector of the node of each stage selected in the process of the tree search using the input vector (1) as the input, and a predetermined value for this output vector. The codebook of each stage that stores the index code that is given in a related relationship,
(5) is the maximum inner product value that is the amplitude gain, (14) is the comparator, and (15) is the output vector index output by the two vector quantizers in the high-speed mode in which the two vector quantizers are used in parallel. An address generator that selects one of (3), (16) uses two vector quantizers in parallel, and each vector quantizer has a high-speed mode in which the inner product operation is performed n / 2 times, and one vector This is a switching signal for switching between the normal mode in which the inner product calculation is performed n times by the quantizer.

Next, the operation will be described.

First, the input mean value separation vector is quantized, but the output vector is an n-ary (n is a power of 2) tree search, vector quantization is performed using the inner product as a distortion measure,
The operation in the normal mode is the same as that of the conventional n-ary tree search vector quantizer, and the vector quantizer at the m-th stage may be the same as that of the conventional one.

Hereinafter, based on FIG. 1, the first stage vector quantizer (2)
A high-speed mode in which two of the second-stage vector quantizers (2) are used in parallel will be described.

In this high speed mode, the processing of each stage of n-ary tree search is performed using two vector quantizers. At this time, the first stage,
The second stage codebook (4) stores half the output vectors of each stage search, and the largest inner product value (5) in the first half is output from the first stage vector quantizer (2), for example. And the largest inner product value in the latter half of the search (5)
Is output from the second stage vector quantizer (2).

For example, if the tree search is a quaternary tree,
Four output vectors (0 ₍₄₎ , 1 ₍₄₎ , 2 ₍₄₎ , 3 ₍₄₎ ) to be compared and an input vector (1) in the first stage of FIG. 4 (b)
The inner products of and are # 0, # 1, # 2, and # 3, respectively.

At this time, the inner product # from the first stage vector quantizer (2)
The larger inner product value of 0 and # 1 is output from the second stage vector quantizer (2), and the larger inner product value of inner products # 2 and # 3 is output.

At this time, output vectors of 0 ₍₄₎ and 1 ₍₄₎ are input to the first stage codebook (4), and 2 ₍₄₎ and 3 are output to the second stage codebook (4).
The output vector of ₍₄₎ is stored in advance.

Also, the output vector index (3) from the first stage vector quantizer (2) to the second stage vector quantizer (2) is cut off by a switch by the mode selection signal (16).
The inner product value (5) output from the first and second stage vector quantizers (2) is compared by a comparator (14), and the result is output to an address generator (15).

The address generator (15) receives the comparison output of the comparator (14) and the output vector index (3) from the first and second stage vector quantizers (2), and outputs the larger inner product value. The output vector index (3) from the quantizer (in the case of a quaternary tree, the most significant 2 bits) is output. The output vector index (3) sent to the third stage vector quantizer (2) is the mode selection signal (1
By 6), the output from the second stage vector quantizer (2) is selected in the normal mode, and the output from the address generator (15) is selected in the high speed mode.

By performing the tree search as described above to the final stage, the input vector whose average value has been separated is vector-quantized by using the inner product as the distortion measure.

In the above embodiment, the case where the tree search is a quaternary tree and two vector quantizers are used in parallel has been described, but the tree search may be an n-ary tree (n is a power of 2). Further, the number of vector quantizers may be three or more, and the same effect as that of the above-described embodiment can be obtained.

[Effects of the Invention] As described above, according to the n-ary tree search vector quantizer of the present invention, two or more vector quantizers in one set are the same as those in the tree search when the switching signal is in the high speed mode. When the switching signal is in the normal mode, one vector quantizer calculates the inner product in the same stage of the tree search, so that the inner product calculation performed by each vector quantizer is performed n times or less. Can be performed at high speed, and the inner product operation in each vector quantizer can be switched between n times or less or n times.

[Brief description of drawings]

FIG. 1 is a block diagram showing an n-ary tree search vector quantizer according to an embodiment of the present invention, FIG. 2 is a block diagram showing a conventional n-ary tree search vector quantizer, and FIG. 3 is a conventional distortion. FIG. 4 is a detailed block diagram of the vector quantizer using the inner product operation for the operation, and FIG. 4 is an explanatory diagram showing the relationship between the tree search and the index code. (1) is an input vector whose average value is separated (2) is a vector quantizer (3) at each stage is a vector index,
(4) is a codebook of each stage, (5) is an amplitude gain which is a maximum inner product value, (6) is an address counter which sequentially outputs a read address (12) according to the output vector index of the preceding-stage quantizer, ( 7) is a multiplier, (8) inner product value,
(9) is the maximum inner product detector, (10) is the amplitude gain latch,
(11) is an index latch, (13) is an adder, (14)
Is a comparator, (15) is an address generator, and (16) is a mode selection signal. The same reference numerals in the drawings indicate the same or corresponding parts.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display location H04N 1/41 Z 7/24

Claims (1)

[Claims] 1. K input signal sequences (K is an integer of 2 or more)
The output vector of the node at each stage selected in the process of reaching the final stage of the tree search with the K-dimensional vector blocked for each input as an input, and the index code given in a predetermined relationship with this output vector The n-ary tree search codebook to be stored (n is a power of 2), the input vector and the index code are input, and the n-ary tree search codebook is tree-searched and read based on the index code. A plurality of vector quantizers for calculating the inner product of the output vector of the next stage and the input vector and outputting the maximum inner product value at that time together with the index code of the output vector giving the maximum inner product, and a set of vector quantizers. So that each vector quantizer in the set obtained by dividing the plurality of vector quantizers is 2 or more, the inner product of the same stage of the tree search is shared. A switching signal for switching between the high-speed mode for calculation and the normal mode for calculating the inner product of the same stage of the tree search by one vector quantizer, the inner product value output by each vector quantizer in the one set, and the index code And a comparator / selector that outputs the index code from the vector quantizer that outputs the maximum inner product value, and when the switching signal is in the high-speed mode, each vector quantizer in the set is In each case, the inner product at the same stage of the tree search is shared, and the maximum inner product value at that time is output to the comparison selector together with the index code of the output vector that gives the maximum inner product. The two vector quantizers calculate the inner product in the same stage of the tree search, and give the index vector of the output vector that gives the maximum inner product. And an n-ary tree search vector quantizer which outputs the maximum inner product value at that time to the vector quantizer which performs the next stage processing of the tree search.