Key Signal Processing Technologies for High-speed Passive Optical Networks

Doctoral thesis, 2021

With emerging technologies such as high-definition video, virtual reality, and cloud computing, bandwidth demand in the access networks is ever-increasing. Passive optical network (PON) has become a promising architecture thanks to its low cost and easy management. IEEE and ITU-T standard organizations have been standardizing the next-generation PON, targeting on increasing the single-channel capacity from 10 Gb/s to 25, 50, and 100 Gb/s as the solution to address the dramatic increase of bandwidth demand. However, since the access network is extremely cost-sensitive, many research problems imposed in the physical layer of PON need to be addressed in a cost-efficient way, which is the primary focus of this thesis. Utilizing the low-cost 10G optics to build up high-speed PON systems is a promising approach, where signal processing techniques are key of importance. Two categories of signal processing techniques have been extensively investigated, namely optical signal processing (OSP) and digital signal processing (DSP). Dispersion-supported equalization (DSE) as a novel OSP scheme is proposed to achieve bit-rate enhancement from 10 Gb/s to 25 Gb/s based on 10G class of optics. Thanks to the bandwidth improved by DSE, the non-return-zero on-off keying which is the simplest modulation format is able to be adopted in the PON system without complex modulation or DSP. Meanwhile, OSP is also proposed to work together with DSP enabling 50G PON while simplifying the DSP complexity. Using both DSE and simple feed-forward equalizer is able to support 50 Gb/s PAM-4 transmission with 10G optics. For C-band 50 Gb/s transmission, injection locking techniques as another OSP approach is proposed to compress the directly modulated laser chirp and increase system bandwidth in the optical domain where a doubled capacity from 25 Gb/s to 50 Gb/s over 20 km fiber can be built on top of 10G optics. For DSP, we investigated the advantages of neural network (NN) on the mitigation of the time-varying nonlinear semiconductor optical amplifier pattern effect. In order to reduce the expense caused by the high computation complexity of NN, a pre- equalizer is introduced at the central office that allows cost sharing for all connected access users. In order to push the PON system line rate to 100 Gb/s, a joint nonlinear Tomlinson- Harashima precoding-Volterra algorithm is proposed to compensate for both linear and nonlinear distortions where 100 Gb/s PAM-4 transmission over 20 km fiber with 15 GHz system bandwidth can be achieved.