Scalable predictive coding method and apparatus
AssigneeVoicecraft, Inc.
InventorKenneth Rose
Life of this patent
Estimating the term…
Abstract
A scalable predictive coder in which the current frame of data is predicted at the enhancement-layer by processing and combining the reconstructed signal at: (i) the current base-layer (or lower layers) frame; and (ii) the previous enhancement-layer frame. The combining rule takes into account the compressed prediction error of the base-layer, and the parameters used for its compression.
Claims13
A method for predicting the current frame of data in a digital coding system wherein a signal is segmented into frames of data that are sequentially encoded, said system including a base layer and an enhancement layer, said base layer including a base encoder and a base decoder, said base encoder including a quantizer and a predictor, said enhancement layer including an enhancement encoder and an enhancement decoder, said base decoder producing prediction data and quantized residual data, said enhancement decoder producing an enhanced reconstructed signal, said method comprising:
- predicting the current frame of data at the enhancement-layer by evaluating the conditional expected value of the driving process model for each sample of the frame where said conditional expected value depends on (a) the reconstructed data representing the previous enhancement layer frame and (b) the quantization interval within which the corresponding original base layer residual sample lies.
A method for scalable predictive coding of a signal, comprising:
- encoding data representing said signal with a base layer predictive coding system that provides a first prediction of said signal and information indicative of a decoded base layer approximation to said signal;
- and
- encoding data representing said signal by a first enhancement layer which performs predictive coding with a second prediction of said signal derived from a combination of information from the current base layer and information from the past decoded approximation to said signal generated in said first enhancement layer;
- wherein said step of encoding data representing said signal by a first enhancement layer comprises providing to said first enhancement layer time evolution statistics derived either by off-line computation or by computations using quantized parameters of said signal.
A method for scalable predictive coding of a signal, comprising:
- encoding data representing said signal with a base layer predictive coding system that provides a first prediction of said signal and information indicative of a decoded base layer approximation to said signal;
- and
- encoding data representing said signal by a first enhancement layer which performs predictive coding with a second prediction of said signal derived from a combination of information from the current base layer and information from the past decoded approximation to said signal generated in said first enhancement layer;
- wherein said coding system includes a second enhancement layer and wherein said second enhancement layer performs predictive coding with a third prediction of said signal derived from a combination of information from said first enhancement layer and information indicative of the past decoded signal approximation generated in said second enhancement layer.
A method for scalable predictive coding of a signal, comprising:
- encoding data representing said signal with a base layer predictive coding system that provides a first prediction of said signal and information indicative of a decoded base layer approximation to said signal;
- and
- encoding data representing said signal by a first enhancement layer which performs predictive coding with a second prediction of said signal derived from a combination of information from the current base layer and information from the post decoded approximation to said signal generated in said first enhancement layer;
- wherein said second prediction at predetermined intervals is derived exclusively from information from the base layer and at all other times is derived by combining information from the base layer and information indicative of the past decoded signal approximation generated in said first enhancement layer.
An apparatus for scalable predictive coding of a signal, comprising:
- means for encoding data representing said signal with a base layer predictive coding system that provides a first prediction of said signal and information indicative of a decoded base layer approximation to said signal;
- and
- means for encoding data representing said signal by a first enhancement layer which performs predictive coding with a second prediction of said signal derived from a combination of information from the current base layer and information from the past decoded approximation to said signal generated in said first enhancement layer;
- wherein said means for encoding said signal data with said first enhancement layer comprises means for providing to said first enhancement layer time evolution statistics derived either by off-line computation or by computations using quantized parameters of said signal.
An apparatus for scalable predictive coding of a signal, comprising:
- means for encoding data representing said signal with a base layer predictive coding system that provides a first prediction of said signal and information indicative of a decoded base layer approximation to said signal;
- means for encoding data representing said signal by a first enhancement layer which performs predictive coding with a second prediction of said signal derived from a combination of information from the current base layer and information from the past decoded approximation to said signal generated in said first enhancement layer;
- and
- a second enhancement layer, wherein said second enhancement layer performs predictive coding with a third prediction of said signal derived from a combination of information from said first enhancement layer and information indicative of the past decoded signal approximation generated in said second enhancement layer.
An apparatus for scalable predictive coding of a signal, comprising:
- means for encoding data representing said signal with a base layer predictive coding system that provides a first prediction of said signal and information indicative of a decoded base layer approximation to said signal;
- and
- means for encoding data representing said signal by a first enhancement layer which performs predictive coding with a second prediction of said signal derived from a combination of information from the current base layer and information from the past decoded approximation to said signal generated in said first enhancement layer;
- wherein said second prediction at predetermined intervals is derived exclusively from information from the base layer and at all other times is derived by combining information from the base layer and information indicative of the past decoded signal approximation generated in said first enhancement layer.
A method for scalable predictive coding of a signal, comprising:
- encoding data representing said signal with a base layer predictive coding system that provides a first prediction of said signal and information indicative of a decoded base layer approximation to said signal;
- and
- encoding data representing said signal by a first enhancement layer which performs predictive coding with a second prediction of said signal derived from information from the past decoded approximation to said signal generated in said first enhancement layer which is combined in a combining operation with the decoded signal approximation generated in the current base layer frame and at least one compression parameter associated with the base layer or one statistical parameter associated with either the base or enhancement layers;
- wherein said combining operation employs a statistical time evolution model to calculate said predictive coding;
- wherein said combining operation is performed in the transform domain per coefficient and comprises computing an expectation of the corresponding transform coefficient of the original signal, given that it lies in the interval determined by said quantization interval employed at the base layer;
- and
- wherein said computing of an expectation is taken over the statistics provided by said time-evolution model.
A method for predicting the current frame of data within a channel of a digital coding system wherein a signal is temporally segmented into consecutive frames of data that are sequentially encoded, said system including a base layer and an enhancement layer, said base layer including a quantizer and a predictor, said enhancement layer including an enhancement encoder and an enhancement decoder, said base decoder producing a prediction data and quantized residual data, said enhancement decoder producing an enhanced reconstructed signal, said method comprising:
- predicting the current frame of data at the enhancement-layer by processing;
- and
- combining the reconstructed data from the previous enhancement layer frame, the current base layer prediction data, and time evolution statistics derived either by off-line computation or by computations using quantized parameters of said signal.
Dependencies are inferred from the claim text.