Amplitude-adaptive vector quantization system
First Claim
Patent Images
1. An amplitude-adaptive vector quantization system comprising:
- a mean value separating circuit which receives K-dimensional vectors derived from a continuous series of input signals divided into blocks of K-samples and separates out a mean value of said input signal vectors;
a predictive differential quantizer which implements DPCM quantization for said separated mean value;
an amplitude component calculation circuit which calculates the magnitude of amplitude components of mean separated input vectors;
a first amplitude vs. quantization stage correspondence table which determines a quantization state of tree-search vector quantization in correspondence to the magnitude of said calculated amplitude components; and
a tree-search vector quantizer which calculates the inner product between each of said mean separated input vectors and normalized output vectors with zero mean arranged to have a tree structure, implements a tree-search for an output vector providing a maximum inner product sequentially up to a last stage, provides a maximum inner product at each stage as an amplitude gain at each stage along with an output vector index, and outputs selectively an amplitude gain of a quantization stage prescribed by said first amplitude vs. quantization stage correspondence table;
an index buffer which holds the output vector index of each quantization stage obtained during the quantization of said tree-search vector quantizer;
an amplitude gain DPCM quantizer which implements DPCM quantization for said output amplitude gain to provide a DPCM-quantized differential amplitude gain and implements DPCM decoding to provide a reproduction of DPCM-quantized amplitude gain;
a second amplitude vs. quantization stage correspondence table which provides a true quantization stage selection signal from said DPCM-decoded reproduction amplitude gain;
a coding section which has an index selector for selecting an output vector to be outputted finally from among output vector indices of all stages held in said index buffer, and transmits or records three elements, including said DPCM-quantized mean value, DPCM-quantized amplitude gain and selected output vector, as a coding output; and
a decoding section including a mean value DPCM decoder and amplitude gain DPCM decoder which implement DPCM decoding for the DPCM quantized mean value and amplitude gain coded by said coding section respectively, an amplitude vs. quantization stage correspondence table which has the same content as said second amplitude vs. quantization stage correspondence table to determine from the DPCM-decoded amplitude gain a quantization output of which stage of coding section is the output vector index decoded by a vector quantization decoder, said vector quantization decoder implementing decoding for said output vector index by switching output vectors adaptively in response to an output of said amplitude vs. quantization stage correspondence table, an amplitude reproducing circuit which implements amplitude reproduction of said decoded normalized output vector on the basis of said DPCM-decoded amplitude gain, and a mean value adding circuit which adds said DPCM-decoded mean value to said amplitude reproduced vector to produce an output signal vector.
1 Assignment
0 Petitions
Accused Products
Abstract
An amplitude-adaptive vector quantization system intended for efficient signal coding and decoding. A mean value separation circuit (2) separates the mean value from an input signal which has been divided into blocks, and a tree-search vector quantizer (6) implements tree-search inner product vector quantization so that index information is truncated depending on the detected vector amplitude, thereby reducing the volume of information for transmission.
-
Citations
4 Claims
-
1. An amplitude-adaptive vector quantization system comprising:
-
a mean value separating circuit which receives K-dimensional vectors derived from a continuous series of input signals divided into blocks of K-samples and separates out a mean value of said input signal vectors; a predictive differential quantizer which implements DPCM quantization for said separated mean value; an amplitude component calculation circuit which calculates the magnitude of amplitude components of mean separated input vectors; a first amplitude vs. quantization stage correspondence table which determines a quantization state of tree-search vector quantization in correspondence to the magnitude of said calculated amplitude components; and a tree-search vector quantizer which calculates the inner product between each of said mean separated input vectors and normalized output vectors with zero mean arranged to have a tree structure, implements a tree-search for an output vector providing a maximum inner product sequentially up to a last stage, provides a maximum inner product at each stage as an amplitude gain at each stage along with an output vector index, and outputs selectively an amplitude gain of a quantization stage prescribed by said first amplitude vs. quantization stage correspondence table; an index buffer which holds the output vector index of each quantization stage obtained during the quantization of said tree-search vector quantizer; an amplitude gain DPCM quantizer which implements DPCM quantization for said output amplitude gain to provide a DPCM-quantized differential amplitude gain and implements DPCM decoding to provide a reproduction of DPCM-quantized amplitude gain; a second amplitude vs. quantization stage correspondence table which provides a true quantization stage selection signal from said DPCM-decoded reproduction amplitude gain; a coding section which has an index selector for selecting an output vector to be outputted finally from among output vector indices of all stages held in said index buffer, and transmits or records three elements, including said DPCM-quantized mean value, DPCM-quantized amplitude gain and selected output vector, as a coding output; and a decoding section including a mean value DPCM decoder and amplitude gain DPCM decoder which implement DPCM decoding for the DPCM quantized mean value and amplitude gain coded by said coding section respectively, an amplitude vs. quantization stage correspondence table which has the same content as said second amplitude vs. quantization stage correspondence table to determine from the DPCM-decoded amplitude gain a quantization output of which stage of coding section is the output vector index decoded by a vector quantization decoder, said vector quantization decoder implementing decoding for said output vector index by switching output vectors adaptively in response to an output of said amplitude vs. quantization stage correspondence table, an amplitude reproducing circuit which implements amplitude reproduction of said decoded normalized output vector on the basis of said DPCM-decoded amplitude gain, and a mean value adding circuit which adds said DPCM-decoded mean value to said amplitude reproduced vector to produce an output signal vector. - View Dependent Claims (2)
-
-
3. An amplitude-adapted vector quantization system comprising:
-
a mean value separating circuit which receives K-dimensional vectors derived from a continuous series of input signals divided into blocks in K-samples units (K is an integer greater than one) and separates out a mean value of said input signal vectors; a mean value predictive differential (will be termed "DPCM") quantizer which implements DPCM quantization for said separated mean value; a tree-search vector quantizer of n'"'"'s notation system (n is a power of
2) which receives the mean separated vectors provided by said mean value separation circuit as input vectors, calculates the inner product between each of the mean value separation vectors and normalized output vectors read out of a tree-search code book of n'"'"'s notation system, implements tree-search for an output vector providing a maximum inner product sequentially up to a last-stage, and provides a then maximum inner product value as an amplitude gain along with an output vector index providing a maximum inner product at a last-stage quantization, and an amplitude gain DPCM quantizer which implements DPCM quantization for the last-stage quantization amplitude gain provided by said vector quantizer to produce a DPCM-quantized amplitude gain through DPCM quantization, implements DPCM decoding, and produces a DPCM-decoded amplitude gain;a coding section including an index bit length control circuit which receives an index code of the last-stage quantization output vector provided by said vector quantizer and truncates the index code for a specified number of bits from the lowest bit position (will be termed "LSB") in correspondence to the magnitude of said decoded amplitude gain value, and tree-search code book of n'"'"'s notation system constructed to express the index code of output vector at nodes of each stage selected during the search up to the last stage through truncation in log2 n bit units from LSB for the code of last-stage output vector index with the intention of use by said vector quantizer; and a decoding section including a mean value DPCM decoder and amplitude gain DPCM decoder for respectively decoding the mean-value and amplitude gain coded by said coding section, a vector quantization decoder which recognizes the index code length of the output vector to be decoded from the amplitude gain decoded by said amplitude gain DPCM decoder, an amplitude reproduction circuit which multiplies the decoded amplitude gain to the decoded output vector and a mean value adding circuit which adds the DPCM decoded mean value to the multiplication result. - View Dependent Claims (4)
-
Specification