Distributed source coding with context
AssigneeMicrosoft Corporation
InventorsJin Li, Yong Sun
Life of this patent
Estimating the term…
Abstract
A process for distributed source coding (DSC) using context is presented. By forming contexts from prior coded bitplanes of both a reference and DSC coded peer, the DSC virtual channel is split into several virtual sub-channels of different characteristics. The selection of the sub-channel becomes side information that is known to the receiver. As a result, the DSC coding bitrate is reduced.
Claims15
A computer-implemented process for distributed source coding of a set of to-be coded symbols with the aide of a set of reference symbols and context comprised of a set of prior coded symbols, comprising the following actions:
- establishing a virtual channel between the set of to-be coded symbols and a set of reference symbols, wherein the set of to-be coded symbols will be distorted by the virtual channel into the set of reference symbols, and the virtual channel reflects the difference between the set of to-be coded symbols and the set of reference symbols;
- establishing a prescribed number of context categories for the virtual channel using the set of prior coded symbols;
- dividing the virtual channel into multiple sub-channels each dedicated to one of the context categories;
- calculating a channel error characteristic of each sub-channel;
- calculating a combined channel error rate of the virtual channel as a whole;
- and
- channel encoding the set of to-be coded symbols into syndrome or parity bits using a channel encoder which are capable of protecting against the combined channel error rate.
A computer-implemented process for distributed source decoding of a set of channel encoded coded symbols with the aide of a set of reference symbols and a set of prior coded symbols, using syndrome and priority bits generated by a channel encoder that protects from channel error in a virtual channel, and wherein the virtual channel has been divided into context-based sub-channels, comprising the following process actions:
- inputting the set of reference symbols;
- inputting the set of prior coded symbols;
- establishing context categories for the virtual channel based on the context categories used to divide the virtual channel into sub-channels;
- dividing the virtual channel into multiple sub-channels each dedicated to one of the context categories;
- calculating a channel error characteristic of each sub-channel;
- computing a log likelihood of each coded symbol using the reference symbols and the corresponding virtual sub-channel error characteristic;
- and
- decoding the coded symbols using a channel decoder to remove the errors of the virtual channel.
Dependencies are inferred from the claim text.