Abstract
Channel coding for unequal error protection (UEP), which can provide different protecting efforts with respect to differentprotection requirements, has recently attracted significant research interests, since it can make the best use of the channelbandwidth for many practical applications. This thesis investigates convolutional codes, one of the most popular codingschemes in digital communication systems, for the application of UEP. From an algebraic theory point of view, we make a fullexploration of the UEP capability of convolutional codes, based on which general properties about combining the optimality,canonicity, and minimality of generator matrices are clarified. We also focus on the construction of powerful UEP schemes comprising convolutional codes for practical applications. A new class of path-compatible pruned convolutional (PCPC) codes with the flexible and powerful capabilities of UEP are invented. Besides, a concatenated super-imposed rate-compatible punctured convolutional (SI-RCPC) coder is constructed for UEP channel coding in wireless speech communications.For the study of UEP convolutional codes, necessary and sufficient conditions together with a procedure for obtaining an optimal andbasic generator matrix with the smallest external degree are given. We also demonstrate the existence of a canonical generator matrix which has the greatest separation vector among all canonical generator matrices for every convolutional code. Necessary and sufficient conditions as well as a procedure are presented for obtaining such a canonical generator matrix of the best UEP capability. By a counterexample, we show that the set of optimal generator matrices for a given convolutional code may contain no systematic generator matrices. A constructive proof is then provided to show the existence of a systematic generator matrix with the greatest separation vector, based on which a procedure is invented for obtaining such a best systematic generator matrix.For the construction of powerful UEP schemes, path pruning is proposed to achieve free distance enlargement for convolutional codes. Through path pruning, every convolutional code can be used for UEP, no matter whether it is originally a UEP code. To avoid the undesired path discontinuity and reduce the possible path distance loss, a cascaded implementation together with a path-compatible criterion is proposed for path pruning, under which PCPC codes are constructed. Necessary and sufficient conditions are derived for a sub-class of PCPC codes whose decoding can be done by a single decoder for the parent code, and some of this kind of PCPC codes with good UEP capabilities found by computer search are given as well.For wireless speech communications, a general methodology is presented for the combined design of the speech and channel coding, based on which a combined variable-rate code-excited linearly predictive (QCELP) speech coding and UEP channel coding system is constructed. In contrast to the conventional schemes, our system employs a concatenated SI-RCPC channel coder which can provide UEP with respect to not only the bit-significance of speech packets but also the speech activity and local channel characteristics. Verified by the simulation results, the combined system achieves an average transmission rate less than 8 kbps as well as an average 2 dB signal-to-noise ratio gain over the conventional equal error protection system.