Partially Block Markov Superposition Transmission of Gaussian Source with Nested Lattice Codes
This paper studies the transmission of Gaussian sources through additive white Gaussian noise (AWGN) channels in bandwidth expansion regime, i.e., the channel bandwidth is greater than the source bandwidth. To mitigate the error propagation phenomenon of conventional digital transmission schemes, we propose in this paper a new capacity-approaching joint source channel coding (JSCC) scheme based on partially block Markov superposition transmission (BMST) of nested lattice codes. In the proposed scheme, first, the Gaussian source sequence is discretized by a lattice-based quantizer, resulting in a sequence of lattice points. Second, these lattice points are encoded by a short systematic group code. Third, the coded sequence is partitioned into blocks of equal length and then transmitted in the BMST manner. Main characteristics of the proposed JSCC scheme include: 1) Entropy coding is not used explicitly. 2) Only parity-check sequence is superimposed, hence, termed partially BMST (PBMST). This is different from the original BMST. To show the superior performance of the proposed scheme, we present extensive simulation results which show that the proposed scheme performs within 1.0 dB of the Shannon limits. Hence, the proposed scheme provides an attractive candidate for transmission of Gaussian sources.