Fiber optical transmission systems are approaching their capacity limits [Winzer2017]
. In recent years, several techniques have been proposed for increasing spectral efficiency in order to keep up with traffic demands. Within this context, constellation shaping in general—and PS in particular—have received considerable interest. PS uses uniformly spaced constellation points occurring with different probabilities, which can theoretically provide shaping gains up to 1.53 dB SNR for the AWGN channel[Forney1984]. Even higher gains for the nonlinear fiber optical channel have been reported [Dar2014]. Consequently, the development of constellation shaping algorithms integrated with coded modulation has been the subject of intense research [Bocherer2015, Fehenberger2019, Buchali2016, Fehenberger2016].
PAS [Bocherer2015] integrates a shaping algorithm into an existing BICM system [AlexBICMBook], as shown in LABEL:fig:shaping_block_diagram. CCDM was introduced in [Bocherer2015] and is one of the most popular ways of implementing PAS. CCDM is based on a DM which requires long blocklengths in order to reach optimum performance [Schulte2016]. Short blocklengths can be used to reduce the implementation challenges associated with the required arithmetic coding, however, this results in relatively large rate losses [Fehenberger2019].
In this paper, an alternative probabilistic shaping algorithm based on ESS is experimentally validated for the first time. ESS was introduced in 1993 [Willems1993] and has been recently considered for wireless communications [Gultekin2019]. Very recently, ESS has been introduced to the optical community in [Amari2019]. The main result of [Amari2019] is that short blocklength ESS outperforms CCDM at the same blocklength, long blocklength CCDM, and uniform. This gain is due to the combination of linear shaping gain and nonlinear tolerance. The results in [Amari2019], however, are only based on numerical (split-step Fourier) simulations.
In this paper,