Key points are not available for this paper at this time.
Cloud-based services such as the training of (AI)/machine learning (ML) models in large-scale data centers as well as smart city applications such as the (IoT), connected driverless vehicles, etc., are key drivers for ever higher data rates, especially in short-range fiber-optic links. Such optical transceivers employ (IM/DD) to reduce the costs and power consumption and, therefore, CD is a nonlinear channel effect with respect to the received signal. With net data rates currently going up to 100 Gbit/s for (PON)s and 400Gbit/s for Ethernet connections, CD is a main limitation in transmission performance at such high data rates. To reconstruct the signal impaired by CD, research has mainly discussed nonlinear equalizers and ML models, while, to the best of the authors knowledge, no research has been conducted on the influence of CD on non-data-aided, digital clock recovery in direct-detection systems. In this work, we investigate analytically and in simulation the effect of CD on oversampled digital clock recovery and find that for some dispersion values, clock recovery fails entirely. Based on our findings, we present two dispersion-tolerant clock recovery algorithms that work for NRZ and digital pulse-shaped signals as well as (FTN) signals. Finally, we validate our findings and algorithms in a 34-GBd (PAM4) transmission experiment for NRZ and (RRC) pulse-shaped signals and different accumulated dispersion values.
Matalla et al. (Tue,) studied this question.