Probabilistic Constellation Shaping
Probabilistic constellation shaping optimizes the probability distribution of symbols within a standard constellation (e.g., QAM) to improve communication system performance, primarily aiming to maximize mutual information and thus data rate or reliability. Recent research focuses on employing machine learning techniques, such as deep learning autoencoders and generative models like diffusion models, to design these optimal symbol probability distributions, often jointly optimizing both the symbol probabilities and the constellation geometry itself. These advancements offer significant potential for enhancing the spectral efficiency and robustness of various communication systems, including wireless and optical fiber networks, by achieving higher data rates or improved performance under challenging conditions.