On constellation shaping for short block lengths

Gaussian channel inputs are required to achieve the capacity of additive white Gaussian noise (AWGN) channels. Equivalently, the n-dimensional constellation boundary must be an n-sphere. In this work, constellation shaping is discussed for short block lengths. Two different approaches are considered: Sphere shaping and constant composition distribution matching (CCDM). It is shown that both achieve the maximum rate and generate Maxwell-Boltzmann (MB) distributed inputs. However sphere shaping achieves this maximum faster than CCDM and performs more efficiently in the short block length regime. This is shown by computing the finite-length rate losses. Then the analysis is justified by numerical simulations employing low-density parity-check (LDPC) codes of the IEEE 802.11 standard