Constellation Design for Deep Joint Source-Channel Coding

Deep learning-based joint source-channel coding (JSCC) has shown excellent performance in image and feature transmission. However, the output values of JSCC encoder are continuous, which makes constellation modulation complex dense. It is hard expensive to design radio frequency chains for transmitting such full-resolution points. In this paper, two methods mapping finite proposed real system implementation. The results correspond regular irregular constellation, respectively. We apply existing deep models evaluate them on AWGN channels with different signal-to-noise ratios (SNRs). Experimental show that outperform traditional uniform quadrature amplitude (QAM) method by only adding a few additional parameters.