Polarization Demultiplexing for Stokes Vector Direct Detection

The conventional intensity modulation with direct detection faces the capacity bottleneck beyond 100-Gb/s per wavelength, due to its 1-dimension (1-D) modulation and detection. To increase the spectral efficiency, there emerge a variety of direct detection subsystems based on the Stokes vector receiver (SV-R). The Stokes space contains the intensity information of three independent polarizations. Therefore, the SV-R enlarges the detection dimensions at receiver to 3. The SV-R can recover any 2-D or 3-D signal modulated in Stokes space. One of the most significant features of SV-R is the DSP-enabled polarization demultiplexing. Instead of a 2 × 2 Jones matrix, in Stokes space, a 3 × 3 rotation matrix (RM) characterizes the polarization variation during fiber transmission. Previously, we have demonstrated a training-assisted polarization demultiplexing algorithm, which uses three basis vectors in Stokes space to estimate the RM. In this paper, we propose a blind polarization demultiplexing using the signal distribution in Stokes space. Combined with the blind adaptive algorithm, the new method converts the two-stage polarization demultiplexing (RM estimation and channel equalization) to a single-stage blind 3 × 2 real-value multiple-input multiple-output equalization, which significantly simplify the DSP procedures. We experimentally demonstrate the SV-R based on either training or blind demultiplexing algorithms, and compare the performance between them.