Hybrid Digital-Analog strategies for Source-Channel Broadcast

We consider the problem of transmission of information sources with memory over a broadcast channel. This is motivated by the practical application of the design of audiovisual broadcast systems for digital TV/digital radio. Using a parallel Gaussian source model and an additive white Gaussian noise broadcast channel model, we consider the problem of information source transmission with the goal of attaining trade-offs in meansquare error distortion achieved by individual receivers under a transmission power constraint. Our approach is driven by the high-level recognition that analog (uncoded) and digital (coded) transmission methods offer a somewhat complementary set of cost/benefit trade-off options with regard to this problem. Our proposed solution is accordingly motivated at extracting the best of both worlds. Specifically, we invoke a hybrid constructive framework consisting of building blocks derived from a number of tools from multi-user information theory – source coding with side information, channel coding with side information, successive refinement source coding and super-position broadcast channel coding. Our solution delivers interesting quality trade-offs for the delivery of colored Gaussian sources over AWGN broadcast channels. When specialized to the case of a white source with bandwidth mismatched to the channel bandwidth, our scheme performs as well or in some cases strictly better than all currently known strategies