Experimental evaluation of Data-Aided and non-Data-Aided dynamic equalization algorithms for low OSNR regimes

Optical transceivers with flexible code rates offer the possibility of reducing the required OSNR at the receiver by using higher-order modulation formats with longer FEC overheads. To work in such low OSNR scenarios, the performance of digital signal processing algorithms become critical. Therefore, we compared an unsupervised and a supervised implementations of the Radially-Directed Estimation Algorithm (RDE), and an unsupervised Decision-Directed Least Mean Square (DD-LMS). Performance was evaluated by comparing the characteristics of SNR versus OSNR curves generated experimentally by offline post-processing, for OSNRs varying from 8 to 35 dB, and Nyquist 4/8/16/32/64-QAM modulation formats, for various symbol rates. The supervised algorithm resulted in expressive gains for high-order modulation formats (32-QAM and 64-QAM), while for low-order modulation formats (4-QAM, 8-QAM and 16-QAM) no noteworthy improvement was observed. The results indicate that, although supervised algorithms reduce spectral efficiency due to the transmission of pilot symbols, in low OSNR regimes (as in heavily-coded systems) they become an interesting option.