Source Optimized Channel Decoding and Unequal Error Protection of CELP Encoded Speech

The reliable transmission of Federal Standard CELP 1016 encoded speech over very noisy communication channels is investigated. First, the interframe and intraframe redundancy present in the CELP 1016 parameters is quanti ed using second-order Markov chains. It is shown that over one-quarter of the CELP bits in every frame of speech are redundant. An unequal error protection (UEP) coding scheme, which exploits this residual redundancy, is next proposed for the transmission of the CELP parameters over BPSK-modulated AWGN and independent Rayleigh fading channels. It employs rate-compatible convolutional (RCPC) codes used in conjunction with maximum a posteriori (MAP) soft decision decoding. Experimental results indicate substantial coding gains (up to 7 dB in E b =N 0 ) over conventional systems that utilize equal error protection and maximum likelihood (ML) decoding.