Optimal Polarization Demultiplexing for Coherent Optical Communications Systems

Spectrally-efficient optical communications systems employ polarization division multiplexing (PDM) as a practical solution, in order to double the capacity of a fiber link. Polarization demultiplexing can be performed electronically, using polarization-diversity coherent optical receivers. The primary goal of this paper is the optimal design, using the maximum-likelihood criterion, of polarization-diversity coherent optical receivers for polarization-multiplexed optical signals, in the absence of polarization mode dispersion (PMD). It is shown that simultaneous joint estimation of the symbols, over the two received states of polarization, yields optimal performance, in the absence of phase noise and intermediate frequency offset. In contrast, the commonly used zero-forcing polarization demultiplexer, followed by individual demodulation of the polarization-multiplexed tributaries, exhibits inferior performance, and becomes optimal only if the channel transfer matrix is unitary, e.g., in the absence of polarization dependent loss (PDL), and if the noise components at the polarization diversity branches have equal variances. In this special case, the zero-forcing polarization demultiplexer can be implemented by a 2 ? 2 lattice adaptive filter, which is controlled by only two independent real parameters. These parameters can be computed recursively using the constant modulus algorithm (CMA). We evaluate, by simulation, the performance of the aforementioned zero-forcing polarization demultiplexer in coherent optical communication systems using PDM quadrature phase shift keying (QPSK) signals. We show that it is, by far, superior, in terms of convergence accuracy and speed, compared to conventional CMA-based polarization demultiplexers. Finally, we experimentally test the robustness of the proposed constrained CMA polarization demultiplexer to realistic imperfections of polarization-diversity coherent optical receivers. The PMD and PDL tolerance of the proposed demultipl- - exer can be used as a benchmark in order to compare the performance of more sophisticated adaptive electronic PMD/PDL equalizers.