Promoteur : Prof. Patrice Mégret

Résumé :

With the aid of flex grid technology, the spectrally efficient superchannel transponders can be
realized with either multiband single carrier or multicarrier modulation techniques. This thesis shows
that multicarrier signals would be a good choice for enabling such terabit-scale transponders because
they yield some advantages over the single carrier signal such as data rate adaptation and efficient
frequency domain based signal processing. Nonetheless, the multicarrier modulations also have
several issues that need to be addressed before deploying for next generation network. One obvious
drawback is the susceptibility to phase noise of multicarrier signals because of its long symbol
duration. Furthermore, system performances with multicarrier modulations are also sensitive to fiber
nonlinear effects due to narrow frequency spacing and high peak-to-average power ratio.
The thesis firstly provides a proof of concept of the multicarrier applications for realizing the
superchannel transmissions, and demonstrates them numerically. Two potential candidates among
multicarrier modulations, i.e. orthogonal frequency division multiplexing (OFDM) and filter bank
multicarrier (FBMC), are taken into account for performance evaluations and comparisons. Secondly,
several low complexity and high performance DSP algorithms are proposed to tackle the laser phase
noise vulnerability of both OFDM and FBMC-based transmissions. Finally, one chapter of this thesis is
focused on fiber nonlinear mitigation technique for OFDM-based systems by employing machine
learning approaches, more specifically, the support vector machine. Compared to other existing
techniques, these above-mentioned DSP proposals provide superior performances and/or lower
complexities. Thus, they should be partly considered while designing the next generation transport
networks.