Constellation Design and Joint Optimization of Analog Joint Source-Channel Coding

Recent studies have shown that analog joint source-channel coding (AJSCC) is capable of approaching the optimum performance theoretically attainable (OPTA). However, it would be costly to transmit the continuous output of AJSCC. This paper proposes a two-dimensional (2D) quantization-based modulation scheme to map the discrete-time real-valued source encoder output into a finite set of 2D constellation points for modulated transmission. To minimize the resultant quantization distortion, the constellation is optimized by constructing a centroidal Voronoi diagram that can best match the distribution of the source encoder output under power transformation. Based on a comprehensive analysis of effects of encoder configuration and power transformation in connection with the channel condition, an algorithm is presented for jointly optimizing the source encoder and the constellation, with the objective of maximizing the system performance in terms of the end-to-end signal-to-distortion ratio (SDR) with respect to the channel condition in terms of signal-to-noise ratio (SNR). Simulation results show that under joint optimization, the modulated AJSCC system improves as the modulation order increases, and it tends to approach the benchmark performance of the unmodulated system that is close to the OPTA. More practically, the modulated AJSCC system is robust to tolerate imperfect channel estimation to some extent. This work provides a promising way of enabling a source-oriented AJSCC system to access digital communication networks.