Embedded source coding

In many applications an embedded source code is useful. An embedded source code produces a sequence of source descriptions with the property that early descriptions are contained (embedded) within later descriptions. Each source description, except the first, may be decomposed into the prior source description and an incremental refinement. The source descriptions can be decoded by a receiver to obtain a sequence of source reconstructions with decreasing expected distortion. This allows a source reconstruction to be refined by transmitting an incremental description and conditioning upon the prior source description. The goal in design of embedded source codes is to make the sequence of source descriptions efficient in a rate-distortion sense.;A key question relating to embedded source codes is to characterize the cost of embedded quantizers in an operational context. By the cost of embedded quantizers we mean the excess distortions (rates) incurred by an embedded quantizer compared with the equivalent distortions (rates) of a set of non embedded quantizers designed for the same set of rates (distortions). This dissertation explores this question for a variety of different types of quantizers and different operational scenarios.;First embedded scalar quantizers are studied; both fixed-rate and variable-rate embedded quantizers are developed. In addition, the inclusion of complexity constraints in the design is considered, and their effects studied. The embedded quantization methodology is then extended to develop embedded trellis coded quantizers, which provide improved rate-distortion performance, relative to scalar quantizers. As in the scalar case, both fixed-rate and variable-rate embedded trellis coded quantizers are developed, and complexity constrained designs are also considered. Finally, the combination of embedded source codes with embedded channel codes is considered. A combination of a tree structured vector quantizer and rate-compatible punctured convolutional codes is developed which provides rate distortion performance superior to that of a channel-matched tree structured vector quantizer alone. This combination is investigated for tree structured vector quantizers designed in both a channel-matched and non channel-matched fashion.