Adaptive multi-stage vector quantization

Claim 2

In a multi-stage vector quantization encoder comprising a plurality of consecutive vector quantization stages wherein each stage comprises a vector quantizer and an associated codebook containing a plurality of codevectors, there being a unique address associated with each codevector in the codebook of each stage, a method comprising the steps of: p1 (a) receiving a data vector indicative of a block of data to be compressed; (b) selecting a codevector from the codebook at an initial stage that most closely resembles the data vector, generating a residual vector, and proceeding directly to step (e) if a measure of quantization error at that stage satisfies a threshold value, but otherwise first performing the following steps (c) and (d); (c) at a next succeeding stage, selecting a codevector from the codebook at that stage that most closely resembles the residual vector from the immediately preceding stage, and generating another residual vector; (d) repeating said step (c) until a measure of quantization error at a particular stage satisfies the threshold value; and then (e) transmitting addresses of each codevector selected in said steps (b) and (c) to a remote location.

Claim 3

A method according to claim 1 wherein the addresses are transmitted at a fixed data rate, further comprising the steps of storing said addresses to be transmitted in a buffer, periodically obtaining a measure of unused capacity of the buffer, and automatically adjusting the threshold value based upon the measure of unused buffer capacity.

Claim 4

A method according to claim 1 wherein said step (a) further comprises determining a mean value of the data vector and removing the mean value from the data vector to obtain a mean-removed vector, and wherein the mean-removed vector is employed in step (b) to select the codevector from the codebook at the initial stage.

Claim 5

A method according to claim 3 wherein said step (e) further comprises transmitting the mean value.

Claim 6

A method according to claim 3 wherein said steps (b) through (d) are only performed if a measure of distortion between the mean value and the data vector exceeds the threshold value, otherwise the process proceeds directly from said step (a) to step (e).

Claim 7

A method according to claim 1 wherein the residual vector generated in said step (b) at the initial stage is generated by subtracting the selected codevector from the data vector, and wherein the residual vector generated at each succeeding stage in said step (c) is generated by subtracting the codevector selected at that stage from the residual vector generated by the immediately preceding stage.

Claim 8

A method according to claim 1 wherein the measure of quantization error at the initial stage is defined as a measure of distortion between the codevector selected at that stage and the data vector, and wherein the measure of quantization error at each succeeding stage is defined as a measure of distortion between the codevector selected at that stage and the residual vector generated by the immediately preceding stage.

Claim 9

A method according to claim 1 wherein said step (e) further comprises constructing and transmitting a data header that specifies the number of codevector addresses being transmitted in said step (e) for a particular data vector.

Claim 10

A method according to claim 1 wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at the remote location, and wherein the following steps are performed at the remote location: a') receiving each address transmitted in step (e) for said data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the reproduced codebooks; and c') summing the retrieved codevectors to obtain a substantial reproduction of the data vector.

Claim 11

A method according to claim 4 wherein the codebooks of each stage of the multi-stage encoder are reproduced at the remote location, and wherein the following steps are performed at the remote location: a') receiving the mean value and each address transmitted in said step (e) for said data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the reproduced codebooks; c') summing the retrieved codevectors to obtain a substantial reproduction of the mean-removed vector; and d') adding the mean value to the reproduced mean-removed vector to substantially reproduce the data vector.

Claim 12

In a multi-stage vector quantization encoder comprising a plurality of consecutive vector quantization stages wherein each stage comprises a vector quantizer and an associated codebook containing a plurality of codevectors, there being a unique address associated with each codevector in the codebook of each stage, a method comprising the steps of: (a) receiving a data vector indicative of a block of data to be compressed; (b) providing the data vector to an initial stage of the multi-stage vector quantization encoder, said data vector defining an input vector for the initial stage, said initial stage defining a current stage; (c) selecting a codevector from the codebook at the current stage that most closely resembles the input vector and generating a residual vector based on the input vector and the selected codevector; (d) determining a measure of distortion between the input vector and the selected codevector; (e) performing the following additional steps if the measure of distortion exceeds a threshold value, but otherwise proceeding to step (f): (i) passing the residual vector to a next succeeding stage as an input vector to that stage, said next succeeding stage becoming the current stage; and (ii) repeating said steps (c) through (e) until the measure of distortion calculated in step (d) at a particular stage does not exceed the threshold value; and then, (f) transmitting the address of any codevectors selected in said steps (c) and (e) to a remote location, and repeating said steps (a) through (f) for a different data vector.

Claim 13

In a multi-stage vector quantization encoder comprising a plurality of consecutive vector quantization stages wherein each stage comprises a vector quantizer and an associated codebook containing a plurality of codevectors, there being a unique address associated with each codevector in the codebook of each stage, a method comprising the steps of: (a) receiving a data vector indicative of a block of data to be compressed; (b) providing the data vector to an initial stage of the multi-stage vector quantization encoder, said data vector defining an input vector for the initial stage, said initial stage defining a current stage; (c) selecting a codevector from the codebook at the current stage that most closely resembles the input vector and generating a residual vector based on the input vector and the selected codevector; (d) determining a measure of distortion between the input vector and the selected codevector; (e) performing the following additional steps if the measure of distortion exceeds a threshold value, but otherwise proceeding to step (f): (i) passing the residual vector to a next succeeding stage as an input vector to that stage, said next succeeding stage becoming the current stage; and (ii) repeating said steps (c) through (e) until the measure of distortion calculated in step (d) at a particular stage does not exceed the threshold value; and then, (f) transmitting the address of any codevectors selected in said steps (c) and (e) to a remote location, and repeating said steps (a) through (f) for a different data vector.

Claim 14

A method according to claim 11 wherein the addresses are transmitted at a fixed data rate, further comprising the steps of storing addresses to be transmitted in a buffer, periodically obtaining a measure of unused capacity of the buffer, and automatically adjusting the threshold value based upon the measure of unused buffer capacity.

Claim 15

A method according to claim 11 wherein said step (a) further comprises determining a mean value of the data vector and removing the mean value from the data vector to obtain a mean-removed vector, and wherein said step (f) further comprises transmitting the mean value, and further wherein the mean-removed vector is provided to the initial stage of the encoder as an input vector thereto in said step (b) .

Claim 16

A method according to claim 13 wherein said steps (b) through (e) are only performed if a measure of distortion between the mean value and the data vector exceeds the threshold value, otherwise the process proceeds directly from said step (a) to step (f).

Claim 17

A method according to claim 11 wherein the codebook of each stage comprises a tree-structured codebook and wherein the vector quantizer of each stage performs vector quantization in accordance with a distortion adaptive tree-search vector quantization technique.

Claim 18

A method according to claim 11 wherein the codebook of each stage comprises a full-search codebook and wherein the vector quantizer of each stage performs vector quantization in accordance with a full-search technique.

Claim 19

A method according to claim 11 wherein the codevectors of the codebook at each stage are stored in a memory device and wherein each address is a memory address.

Claim 20

A method according to claim 11 wherein said step (f) further comprises constructing and transmitting a data header that specifies the last stage reached while performing said steps (c) through (e) for the data vector, said data header thereby indicating how many addresses were transmitted for the data vector.

Claim 21

A method according to claim 11 wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at the remote location, and wherein the following steps are performed at the remote location: a') receiving each address transmitted in said step (f) for a particular data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the reproduced codebooks; and c') summing the retrieved codevectors to obtain a substantial reproduction of the data vector.

Claim 22

A method according to claim 13 wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at the remote location, and wherein the following steps are performed at the remote location: a') receiving the mean value and each address transmitted in said step (f) for a particular data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the reproduced codebooks; c') summing the retrieved codevectors to obtain a substantial reproduction of the mean-removed vector; and d') adding the mean value to the reproduced mean-removed vector to substantially reproduce the data vector.

Claim 23

In a multi-stage vector quantization encoder comprising a plurality of consecutive vector quantization stages wherein each stage comprises a vector quantizer and associated codebook containing a plurality of codevectors, there being a unique address associated with each codevector in the codebook of each stage, a method comprising the steps of: (a) receiving a data vector indicative of a block of data to be compressed; (b) providing the data vector to an initial stage of the multi-stage vector quantization encoder, said data vector defining an input vector for the initial stage, said initial stage defining a current stage; (c) selecting a codevector from the codebook at the current stage that most closely resembles the input vector and generating a residual vector based on the input vector and the selected codevector; (d) determining a measure of distortion between the input vector and the selected codevector; (e) performing the following additional steps if the measure of distortion exceeds a threshold value, but otherwise proceeding to step (f): (i) passing the residual vector to a next stage of the encoder as an input vector to that stage, said next stage becoming the current stage; and (ii) repeating said steps (c) through (e) until the measure of distortion calculated in step (d) at a particular stage does not exceed the threshold value; and then, (f) inserting the address of any codevectors selected in said step (c) into a buffer, and repeating said steps (a) through (f) for a different data vector; and (g) transmitting the addresses from the buffer at a fixed data rate, periodically obtaining a measure of unused capacity of the buffer, and automatically adjusting the threshold value based upon the measure of unused buffer capacity.

Claim 24

In a multi-stage vector quantization encoder comprising a plurality of consecutive vector quantization stages wherein each stage comprises a vector quantizer and associated codebook containing a plurality of codevectors, there being a unique address associated with each codevector in the codebook of each stage, a method comprising the steps of: (a) receiving a data vector indicative of a block of data to be compressed; (b) providing the data vector to an initial stage of the multi-stage vector quantization encoder, said data vector defining an input vector for the initial stage, said initial stage defining a current stage; (c) selecting a codevector from the codebook at the current stage that most closely resembles the input vector and generating a residual vector based on the input vector and the selected codevector; (d) determining a measure of distortion between the input vector and the selected codevector; (e) performing the following additional steps if the measure of distortion exceeds a threshold value, but otherwise proceeding to step (f): (i) passing the residual vector to a next stage of the encoder as an input vector to that stage, said next stage becoming the current stage; and (ii) repeating said steps (c) through (e) until the measure of distortion calculated in step (d) at a particular stage does not exceed the threshold value; and then, (f) inserting the address of any codevectors selected in said step (c) into a buffer, and repeating said steps (a) through (f) for a different data vector; and (g) transmitting the addresses from the buffer at a fixed data rate, periodically obtaining a measure of unused capacity of the buffer, and automatically adjusting the threshold value based upon the measure of unused buffer capacity.

Claim 25

A method according to claim 21 wherein said step (a) further comprises determining a mean value of the data vector and removing the mean value from the data vector to obtain a mean-removed vector, and wherein the mean value is inserted into the buffer and transmitted therefrom in said steps (f) and (g), respectively, and further wherein the mean-removed vector is provided to the initial stage of the encoder as an input vector thereto in said step (b).

Claim 26

A method according to claim 22 wherein said steps (b) through (e) are only performed if a measure of distortion between the mean value and the data vector exceeds the threshold value, otherwise the process proceeds directly from said step (a) to step (f).

Claim 27

A method according to claim 21 wherein said step (f) further comprises inserting a data header into the buffer ahead of the addresses, said data header specifying the last stage reached while performing said steps (c) through (e).

Claim 28

A method according to claim 21 wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at a remote location, and wherein the following steps are performed at the remote location: a') receiving each address transmitted in step (g) for a particular data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the reproduced codebooks; and c') summing the retrieved codevectors to obtain a substantial reproduction of the data vector.

Claim 29

A method according to claim 22 wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at a remote location, and wherein the following steps are performed at the remote location: a') receiving the mean value and each address transmitted in said step (f) for a particular data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the reproduced codebooks; c') summing the retrieved codevectors to obtain a substantial reproduction of the mean-removed vector; and d') adding the mean value to the reproduced mean-removed vector to substantially reproduce the data vector.

Claim 30

In a data compression system employing a multi-stage vector quantization encoder comprising a plurality of vector quantization stages coupled consecutively for compressing each of a plurality of data vectors wherein each stage has a codebook containing a plurality of codevectors and each codevector has a unique address, and wherein the number of stages through which vector quantization proceeds in the multi-stage vector quantization encoder is controlled by obtaining a measure of quantization error at each successive stage and proceeding through the successive stages only until the measure of quantization error at a particular stage satisfies a threshold value, and wherein the addresses of codevectors selected from the codebooks at each stage reached during compression of a particular data vector are transmitted to a remote location, and further wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at the remote location, a method for substantially reproducing the data vector at the remote location comprising the steps of: a') receiving the codevector addresses transmitted by the multi-stage vector quantization encoder for that data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the codebooks at the remote location; and c') summing the retrieved codevectors to obtain a substantial reproduction of the data vector.

Claim 31

In a data compression system employing a multi-stage vector quantization encoder comprising a plurality of vector quantization stages coupled consecutively for compressing each of a plurality of data vectors wherein each stage has a codebook containing a plurality of codevectors and each codevector has a unique address, and wherein the number of stages through which vector quantization proceeds in the multi-stage vector quantization encoder is controlled by obtaining a measure of quantization error at each successive stage and proceeding through the successive stages only until the measure of quantization error at a particular stage satisfies a threshold value, and wherein the addresses of codevectors selected from the codebooks at each stage reached during compression of a particular data vector are transmitted to a remote location, and further wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at the remote location, a method for substantially reproducing the data vector at the remote location comprising the steps of: a') receiving the codevector addresses transmitted by the multi-stage vector quantization encoder for that data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the codebooks at the remote location; and c') summing the retrieved codevectors to obtain a substantial reproduction of the data vector.

Claim 32

A method according to claim 27 wherein, prior to said vector quantization in the multi-stage vector quantization encoder, the mean value of the data vector was removed therefrom and transmitted to the remote location along with the codevector addresses, said method further comprising the steps of receiving the mean value and adding the mean value back to the data vector reproduced in step (c').

Claim 33

In a data compression system employing a multi-stage vector quantization encoder comprising a plurality of vector quantization stages coupled consecutively for compressing each of a plurality of data vectors wherein each stage has a codebook containing a plurality of codevectors and each codevector has a unique address, and wherein the number of stages through which vector quantization proceeds in the multi-stage vector quantization encoder is controlled by obtaining a measure of quantization error at each successive stage and proceeding through the successive stages only until the measure of quantization error at a particular stage satisfies a threshold value, and wherein the addresses of codevectors selected from the codebooks at each stage reached during compression of a particular data vector are transmitted to a remote location, and further wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at the remote location, a method for substantially reproducing the data vector at the remote location comprising the steps of: a') receiving the codevector addresses transmitted by the multi-stage vector quantization encoder for that data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the codebooks at the remote location; and c') summing the retrieved codevectors to obtain a substantial reproduction of the data vector.

Claim 34

In a data compression system employing a multi-stage vector quantization encoder comprising a plurality of vector quantization stages coupled consecutively for compressing each of a plurality of data vectors wherein each stage has a codebook containing a plurality of codevectors and each codevector has a unique address, and wherein the number of stages through which vector quantization proceeds in the multi-stage vector quantization encoder is controlled by obtaining a measure of quantization error at each successive stage and proceeding through the successive stages only until the measure of quantization error at a particular stage satisfies a threshold value, and wherein the addresses of codevectors selected from the codebooks at each stage reached during compression of a particular data vector are transmitted to a remote location, and further wherein the codebooks of each stage of the multi-stage vector quantization encoder are reproduced at the remote location, a method for substantially reproducing the data vector at the remote location comprising the steps of: a') receiving the codevector addresses transmitted by the multi-stage vector quantization encoder for that data vector; b') for each received address, retrieving the codevector associated with that address from a respective one of the codebooks at the remote location; and c') summing the retrieved codevectors to obtain a substantial reproduction of the data vector.

Claim 35

A multi-stage vector quantization encoder comprising: a plurality N of vector quantization stages arranged consecutively and each having a vector input, a residual vector output and a codevector address output, the residual vector output of each stage being coupled to the vector input of a succeeding stage, the vector input of an initial stage being coupled to receive a data vector indicative of a block of data to be compressed, each stage comprising: a vector quantizer having an associated codebook containing a plurality of codevectors, each codevector having an address associated therewith, said vector quantizer performing vector quantization on an input vector to said stage; a vector subtractor coupled to said vector quantizer for generating a residual vector based on a codevector selected by the vector quantizer and the input vector; a distortion calculator for determining a measure of distortion between the input vector and the selected codevector; a comparator for comparing the measure of distortion to a threshold value; and a gating device responsive to the comparator for allowing the residual vector to pass from the residual vector output of the stage to the vector input of a next succeeding stage only when the measure of distortion exceeds the threshold value, said encoder further comprising a transmitter for transmitting the addresses of codevectors selected at each stage to a remote location.

Claim 36

A multi-stage vector quantization encoder comprising: a plurality N of vector quantization stages arranged consecutively and each having a vector input, a residual vector output and a codevector address output, the residual vector output of each stage being coupled to the vector input of a succeeding stage, the vector input of an initial stage being coupled to receive a data vector indicative of a block of data to be compressed, each stage comprising: a vector quantizer having an associated codebook containing a plurality of codevectors, each codevector having an address associated therewith, said vector quantizer performing vector quantization on an input vector to said stage; a vector subtractor coupled to said vector quantizer for generating a residual vector based on a codevector selected by the vector quantizer and the input vector; a distortion calculator for determining a measure of distortion between the input vector and the selected codevector; a comparator for comparing the measure of distortion to a threshold value; and a gating device responsive to the comparator for allowing the residual vector to pass from the residual vector output of the stage to the vector input of a next succeeding stage only when the measure of distortion exceeds the threshold value, said encoder further comprising a transmitter for transmitting the addresses of codevectors selected at each stage to a remote location.

Claim 37

An encoder according to claim 30 further comprising: a data buffer for temporarily storing the addresses of selected codevectors prior to transmission; and a threshold calculator coupled to the buffer and to each stage for providing each stage with the threshold value and for periodically adjusting the threshold value based upon a measure of the un-used capacity of the buffer.

Claim 38

An encoder according to claim 31 wherein the transmitter transmits data from the buffer at a fixed data rate.

Claim 39

An encoder according to claim 30 wherein a last stage comprises only a vector quantizer and a codebook.

Claim 40

An encoder according to claim 30 further comprising a mean value calculator coupled to the initial stage for determining a mean value of the data vector and for removing the mean value from the data vector to produce a mean-removed vector, said initial stage receiving the mean-removed vector.

Claim 41

An encoder according to claim 34 further comprising means for obtaining a measure of distortion between the mean value and the data vector and for allowing the initial stage to receive the mean-removed vector only if the measure of distortion between the mean value and the data vector exceeds the threshold value.

Claim 42

An encoder according to claim 30 further comprising means coupled to each stage for generating a data header that identifies, for a particular data vector, a last stage reached during compression of that data vector, said data header being transmitted along with the codevector addresses, if any, generated for that data vector.

Claim 43

An encoder according to claim 30 wherein the codebook of each stage comprises a tree-structured codebook and wherein the vector quantizer of each stage performs vector quantization in accordance with a distortion adaptive tree-search vector quantization technique.

Claim 44

An encoder according to claim 30 wherein the codebook at each stage comprises a full-search codebook and wherein the vector quantizer of each stage performs vector quantization in accordance with a full-search technique.

Claim 45

A vector quantization decoder for use at a remote location in conjunction with the encoder of claim 30, said vector quantization decoder comprising: a plurality of codevector codebooks, each codebook being identical to the codebook of a respective one of the stages of said encoder; a receiver for receiving each of the codevector addresses transmitted by the encoder for a particular data vector; means for employing each received codevector address to retrieve the codevector associated with that address from a respective one of the codebooks; and an adder for summing the retrieved codevectors to obtain a substantial reproduction of the data vector.

Claim 46

A decoder according to claim 39 wherein the receiver is further operative to receive a data header transmitted by the encoder along with the codevector addresses, wherein said data header specifies the number of codevector addresses transmitted for a given data vector, said vector quantization decoder further comprising a header decoder for processing the data header to determine the number of codevector addresses to be received for the data vector.