Advanced coding techniques for multicasting and broadcasting in wireless communications

The thesis addresses some open problems in the area of efficient transmission of loss-sensitive and delay-sensitive data over wireless channels. The thesis mainly deals with coding techniques for multicast systems. Multicast differs from the information theoretic broadcast channel in that only common information is sent. In point-to-point transmission, reliability is achieved by means of Automatic Retransmission reQuest (ARQ). Forward Error Correcting (FEC) codes and ARQ are combined together in order to optimize the trade-off between reliability and efficiency. This approach is called Hybrid ARQ, (HARQ). We consider HARQ schemes for point-to-point transmission with modern coding techniques (Low Density Parity Check codes, LDPC). The theoretical analysis shows that these codes ideally achieve optimal performance in terms of throughput. However, for practical finite-length codes, the scheme exhibits a loss in performance. Two different solutions are shown to recover most of this performance gap. In a multicast setting, however, HARQ protocols are inefficient. Strictly speaking, they are not fully scalable. This motivates us to study the throughput per user of these protocols. In particular, HARQ based on Selective Repeat (SR) or Incremental Redundancy (IR) can be defined to be fully scalable if we allow for a fraction $x > 0$ of users that do not decode successfully. While in the first part of the thesis we have considered data communications, for which the relevant performance measure is error probability, in the second part we consider the transmission of an analog source (for example an image). Existing practical solutions, mainly based on Shannon's separation theorem, are highly inefficient and in particular they are not robust to channel errors. In a multicast setting, moreover, it is important to design a scheme that guarantees good performance over a wide range of signal to noise ratio. Different users with different channel conditions can decode the source with acceptable reconstruction quality. Joint source-channel coding is a viable solution for robustness and efficiency in this context. In this multicast environment we analyze and optimize three well-known strategies. The first is based on time sharing; the second on a superposition coding scheme. These two fully digital schemes are compared with an optimized Hybrid Digital Analog (HDA) Finally, the problem of code construction for the HDA system is addressed in the last part of the thesis. Two schemes are proposed. In the first case all the complexity relies on the quantizer scheme. The quantizer is defined in such a way that its performance is based on bit error rate (and not frame error rate) at the output of the channel decoder. We consider an embedded Multistage Trellis Quantizer (MTQ), based on standard binary convolutional codes. The second considers a very simple quantizer scheme. Data compression and channel coding are combined and accomplished with a linear code. Here a multilevel compression scheme based on linear codes (Turbo Codes) is considered.