US 5,313,552 AGranted

Apparatus for quantizing an input group of data samples into one of N quantized groups of data via a process on less than N/2 reference groups of data samples

AssigneeUnisys Corporation

InventorRobert A. Lindsay

Life of this patent

Estimating the term…

Loading patent…

Abstract

A quantizer, which converts an input group of data samples into one of N quantized groups of data samples, does so by performing several dot products and compares. Each dot product is between the input group of data samples and a reference group of data samples u.sub.x, and each compare is with the dot product result and a constant k.sub.y. All of the reference groups of data samples and all of the constants are stored in a memory. This memory holds only N-1 constants and less then N/2 reference groups of data samples. A memory saving of several thousand percent is achieved by limiting the memory to hold no more then 2r reference groups of data samples where r is the number of samples in the input group, or no more then 2 log.sub.2 N reference groups of data samples.

Claims22

Dependent claims

A sampled data quantizer for converting an input group of data samples, having substantially more than N possible values, to the centroid of one of N regions into which said values are divided--where N is a predetermined positive integer;

said system comprising:
an input circuit for receiving said input group of data samples;
a storage means which stores multiple reference groups of data samples u and multiple constants k which respectively do not exceed N/2 and N-1, with each u and k being selected such that the product uk defines a vector which is orthogonal to and reaches a boundary between at least two of said N regions;
a control means, coupled to said storage means, which selects one of said reference groups of data samples u.sub.x and one of said constants k.sub.y ;
an arithmetic means, coupled to said input circuit and said storage means, which performs a dot product on said input group of data samples with u.sub.x ;
a comparing means, coupled to said arithmetic means, which compares the dot product result from said arithmetic means to k.sub.y ;
said control means being further coupled to said comparing means and adapted to repeatedly select a different u.sub.x and k.sub.y for said arithmetic means and said comparing means, with each u.sub.x and k.sub.y for being selected in accordance with a selection tree which has multiple nodes that are interconnected by branches--with each node specifying the u.sub.x for said arithmetic means and the k.sub.y for said comparing means, and indicating which branch is taken based on the compare by said comparing means;
and,
a logic means, coupled to said comparing means, which identifies said centroid from each and every one of the compares by said comparing means.

6 more elements

said system comprising:
an input circuit for receiving said input group of data samples;
a memory means which stores N-1 constants k;
a register means which stores no more than N/2 reference groups of data samples u;
each u and k being selected such that the product uk defines a vector which is orthogonal to and reaches a boundary between at least two of said N regions;
an arithmetic means, coupled to said input circuit and said register means, which performs in parallel respective dot products on said input group of data samples with all of said reference groups of data samples u;
a selecting means, coupled to said arithmetic means and said memory means, which selects the results R.sub.x of one of said dot products and one of said constants k.sub.y ;
a comparing means, coupled to said selecting means, which compares the selected dot product result R.sub.x to the selected constant k.sub.y ;
a control means, coupled to said selecting means, which repeatedly causes said selecting means to select a different R.sub.x l and k.sub.y, with each R.sub.x and k.sub.y being selected in accordance with a selection tree which has multiple nodes that are interconnected by branches--with each node specifying the R.sub.x and k.sub.y for said selecting means, and indicating which branch is taken based on the compare by said comparing means;
and,
a logic means, coupled to said comparing means, which identifies said centroid from each and every one of the compares by said comparing means.

8 more elements

Dependencies are inferred from the claim text.

Facts

Application: 08/031,076Filed 11 Mar 1993 · continuation
Priority: 27 Mar 1991
Granted: 17 May 1994
Term ends: Not estimatedThe source records don’t support an estimate.

Classification6 codes

G10L 19/038
Quantisation or dequantisation of spectral components: vector quantisation, e.g. TwinVQ audio
H03M 7/3082
Compression ; Expansion; Suppression of unnecessary data, e.g. redundancy reduction: vector coding
H04N 19/94
Using coding techniques not provided for in groups H04N19/10-H04N19/85, e.g. fractals: vector quantisation
G10L
Speech analysis techniques or speech synthesis