Simple self-optimization of WDM networks based on probabilistic constellation shaping [Invited]

Abstract

With the advent of novel elastic optical transponders allowing for fine rate granularity, network designers can maximize the throughput of current installed and future wavelength-division multiplexing (WDM) infrastructures. In this sense, data flow can be properly optimized for each optical connection, while avoiding unnecessary margins. Such a connection mode of operation should rely on a simple mechanism. This paper illustrates how a SNR-driven self-optimization of optical connections is relevant for elastic WDM networks, especially when associated with probabilistic constellation shaping, allowing rate tunability while maximizing spectral efficiency. In addition, we study a WDM North American backbone network to quantify the savings brought by this self-optimization in terms of transponder equipment for 10 years, as compared to conventional planning usually relying upon “end of life” assumptions.

Supply-Power-Constrained Cable Capacity Maximization Using Multi-Layer Neural Networks

“…An alternative two-parameter version of Eq. ( 1) that has also been extensively verified experimentally is given by [6],…”

“…Eq. (6). Figure 9 In this paper, we use the MAE for training and evaluation of the NN, since it makes the training more robust to outliers [32].…”

The Role of Parallelism in the Evolution of Optical Fiber Communication Systems

“…This practically relevant noise floor, which is rarely considered in theoretical fiber capacity analyses, caps the system capacity even if the fundamental limitations due to ASE and NLIN would still allow for higher system throughputs. Typical high-speed optical transponders in commercial products operate at an SNR TRX of ∼20 dB [94], [95], record high-SE research results at much reduced symbol rates have achieved an SNR TRX of ∼30 dB [96], [97], and laboratory results exploiting special techniques, such as single-sideband modulation (SSB) in conjunction with image-band rejecting heterodyning at the receiver, have shown to reach an SNR TRX of ∼38 dB, albeit only at low symbol rates [93]. For the system analyses presented in this article, it is reasonable to assume that realistic transponders in the foreseeable future will not exceed an SNR TRX of 30 dB (i.e., κ = 10 −3 ).…”