Fiber-induced intra- and interchannel nonlinearities are experimentally tackled using blind nonlinear equalization (NLE) by unsupervised machine-learning-based clustering (MLC) in ~46-Gb/s single-channel and ~20-Gb/s (middle-channel) multichannel coherent multicarrier signals (orthogonal frequency-division multiplexing (OFDM) based). To that end, we introduce, for the first time, hierarchical and fuzzy-logic C-means (FLC)-based clustering in optical communications. It is shown that among the two proposed MLC algorithms, FLC reveals the highest performance at optimum launched optical powers (LOPs), while at very high LOPs, hierarchical can compensate more effectively nonlinearities only for low-level modulation formats. When employing binary phase-shift keying and quaternary phase-shift keying, FLC outperforms K-means, fast-Newton support vector machines, supervised artificial neural networks, and NLE with deterministic Volterra analysis. In particular, for the middle channel of a QPSK wavelength-division multiplexing coherent optical OFDM system at optimum -5 dBm of LOP and 3200 km of transmission, FLC outperforms Volterra-NLE by 2.5 dB in Q-factor. However, for a 16-QAM single-channel system at 2000 km, the performance benefit of FLC over inverse Volterra-series transfer function reduces to ~0.4 dB at a LOP of 2 dBm (optimum). Even when using novel sophisticated clustering designs in 16 clusters, no more than additional ~0.3-dB Q-factor enhancement is observed. Finally, in contrast to the deterministic Volterra-NLE, MLC algorithms can partially tackle the stochastic parametric noise amplification .

A support vector machine (SVM)-based classification nonlinear equalizer (NLE) is demonstrated, for the first time, in coherent optical orthogonal frequency-division multiplexing (CO-OFDM). For a 40-Gb/s 16 quadrature amplitude modulated (16QAM) CO-OFDM system at 400 km of transmission, SVM-NLE reduces the fiber-induced nonlinearity penalty by about 1 dB in comparison to the benchmark artificial-neural-network (ANN)-based and inverse-Volterra-series-transfer-function-based NLEs.

Coherent optical orthogonal frequency division multiplexing (CO-OFDM) has attracted a lot of interest in optical fiber communications due to its simplified digital signal processing (DSP) units, high spectral-efficiency, flexibility, and tolerance to linear impairments. However, CO-OFDM’s high peak-to-average power ratio imposes high vulnerability to fiber-induced non-linearities. DSP-based machine learning has been considered as a promising approach for fiber non-linearity compensation without sacrificing computational complexity. In this paper, we review the existing machine learning approaches for CO-OFDM in a common framework and review the progress in this area with a focus on practical aspects and comparison with benchmark DSP solutions.

Modern fiber-optic coherent communications employ advanced, spectrally efficient modulation formats that require sophisticated narrow-linewidth local oscillators (LOs) and complex digital signal processing (DSP). Self-coherent optical orthogonal frequency-division multiplexing (self-CO-OFDM) is a modern technology that retrieves the frequency and phase information from the extracted carrier without employing a LO or additional DSP. However, a wide carrier guard is typically required to easily filter out the optical carrier at the receiver, thus discarding many OFDM middle subcarriers that limit the system data rate. Here, we establish an optical technique for carrier recovery, harnessing large-gain stimulated Brillouin scattering (SBS) on a photonic chip for up to $116.82\mathrm{Gbit}·{\mathrm{s}}^{-1}$ self-CO-OFDM signals, without requiring a separate LO. The narrow SBS linewidth allows for a record-breaking small carrier guard band of $\sim 265\mathrm{MHz}$ in self-CO-OFDM, resulting in higher capacity than benchmark self-coherent multi-carrier schemes. Chip-based SBS-self-coherent technology reveals comparable performance to state-of-the-art coherent optical receivers while relaxing the requirements of the DSP. In contrast to on-fiber SBS processing, our solution provides phase and polarization stability. Our demonstration develops a low-noise and frequency-tracking filter that synchronously regenerates a low-power narrowband optical tone, which could relax the requirements on very-high-order modulation signaling for future communication networks. The proposed hybrid carrier filtering-and-regeneration technique could be useful in long-baseline interferometry for precision optical timing or reconstructing a reference tone for quantum-state measurements.

Single section quantum dash (QDash) mode locked lasers (MLL) can provide a flat and broadband optical frequency comb with low energy consumption, operational simplicity and large-scale low-cost production possibilities. MLL longitudinal modes can be employed as single carriers with a regular spectral spacing in a dense wavelength division multiplexing (DWDM) link, making them promising components for next generation DWDM transceivers. However, individual modes of a MLL suffer from relatively high phase noise while high capacity coherent transmission requires carriers with low phase noise. Optical feedback, which is a well-known method to reduce the linewidth of a single mode laser, can be used to stabilize comb lasers. This paper reports on the investigation of phase noise properties of comb lines delivered from a single section QDash MLL under optical feedback and shows that each MLL longitudinal mode optical linewidth can be drastically narrowed. It enables coherent transmission with extended link budget over 50 km at Tbit/s line rates.

40-Gb/s/$\lambda$ multi-band carrierless amplitude and phase (CAP) modulation long reach passive optical networks (LR-PONs) were demonstrated using 10-G class transceivers only. A major issue of multi-band CAP is that it is vulnerable to timing error and this work has offered quantitative analysis about it in detail for the first time. A novel simple timing recovery approach, partial differential QAM constellation encoding and decoding schemes together with blind multi-modulus algorithms (MMAs) equalization are proposed to address the issue efficiently, enabling zero-overhead signal recovery. Results show that it can offer excellent system tolerance to timing error of at least $\pm$0.1 symbol period even for the highest frequency CAP sub-band. The characteristics of the transceiver are measured and optimization of critical system parameters is performed including the CAP sub-band count, 10-G Mach-Zehnder intensity modulator operation conditions, optical launch power, and wavelength offset asymmetrical optical filtering. For downlink using erbium-doped fiber amplifier (EDFA) pre-amplifiers, successful 40-Gb/s multi-band CAP signal transmission over an 80-km (90-km) SMF is achieved with a link power budget of 33 dB (29 dB) considering a FEC threshold BER of $3.8\times10^{-3}$.