In this paper, by adaptively partitioning and precoding the subcarriers, we proposed a practical and effective entropy loading (EL) scheme for single sideband discrete multi-tone (SSB-DMT) systems. To reveal the practical performance, information bits per symbol (IBPS) is used to identify the optimal probabilistically shaped quadrature amplitude modulation (PS-QAM). Under the constraint of normalized generalized mutual information (NGMI) of the off-the-shelf forward error correction (FEC), we obtain the PS-QAMs that achieve the maximum IBPS using different constellations for different signal-to-noise ratio (SNR). Based on the result, we proposed two adaptively partitioning methods, equally partitioned precoding (EPP) and optimally partitioned precoding (OPP), to perform hybrid constellation entropy loading (HCEL). The HCEL with OPP significantly reduces the number of distribution matcher to 3 from generally several hundred of conventional EL with negligible loss of net data rate (NDR). As demonstrated by experiments, the HCEL with OPP achieves 4.4 dB receiver sensitivity gain compared to conventional bit and power loading, Levin-Campello (LC), and 1.2 dB receiver sensitivity gain compared to HCEL with EPP at the NDR of 60 Gb/s after 80 km standard single mode fiber transmission, making it a competitive and practical solution for EL in the short-to-medium reach transmission systems.
