A New JSCC Scheme for Transmission of Gaussian Sources Over AWGN Channels

This paper is concerned with joint source-channel coding (JSCC) for the transmission of Gaussian sources over additive white Gaussian noise (AWGN) channels. A Gaussian source is first discretized by a ternary embedded quantizer with adjustable quantization levels and then protected by rate-matching block Markov superposition transmission of repetition (BMST-R) codes, a very special class of convolutional low-density generator matrix (LDGM) codes. In contrast to the Lloyd-Max algorithm, which minimizes the distortion, the proposed quantizer minimizes the discrete entropy under the distortion constraint, which can be solved by a simulated annealing algorithm. Also different from the existing schemes, the rate-flexible BMST-R codes allow us to encode ternary symbols in different quantization planes with different rates matching with the (conditional) entropy. Numerical results show that the code rate of the proposed scheme can approach the theoretical bound while the distortion is dominated by the quantizer.