In optical fiber communication, optical and electrical components introduce nonlinearities, which require effective compensation to attain highest data rates. In particular, in short reach communication, components are the dominant source of nonlinearities. Volterra series are a popular countermeasure for receiver-side equalization of nonlinear component impairments and their memory effects. However, Volterra equalizer architectures are generally very complex. This article investigates soft deep neural network (DNN) architectures as an alternative for nonlinear equalization and soft-decision demapping. On coherent 92GBd dual polarization 64QAM back-to-back measurements performance and complexity is experimentally evaluated. The proposed bit-wise soft DNN equalizer (SDNNE) is compared to a 5th order Volterra equalizer at a 15% overhead forward error correction (FEC) limit. At equal performance, the computational complexity is reduced by 65%. At equal complexity, the performance is improved by 0.35dB gain in optical signal-to-noise-ratio (OSNR).

We report an industry leading optical dense wavelength division multiplexing (DWDM) field trial with line rates per channel.1.66 Tb/s using 130 GBaud dualpolarization probabilistic constellation shaping 256-ary quadrature amplitude modulation (DP-PCS256QAM) in a high capacity data center interconnect (DCI) scenario. This research trial was performed on 96.5 km of field-deployed standard single mode G.652 fiber infrastructure of Deutsche Telekom in Germany employing Erbium-doped fiber amplifier (EDFA)-only amplification. A total of 34 channels were transmitted with 150 GHz spacing for a total fiber capacity of 56.51 Tb/s and a spectral efficiency higher than 11bit/s/Hz. In the single-channel transmission scenario 1.71 Tb/s was achieved over the same link. In addition, we successfully demonstrate record net bitrates of 1.88 Tb/s in back-to-back (B2B) using 130 GBaud DP-PCS400QAM.

High speed optical transmission systems suffer from intersymbol interference (ISI) and colored noise induced by nonlinear bandwidth limited optical and electrical components. As a countermeasure, this article investigates deep neural network soft-demappers. In particular, we propose a bidirectional recurrent neural network soft-demapper (BRNN-SD) and benchmarked its performance against a time delay neural network soft-demapper (TDNN-SD) and a reference digital signal processing (DSP) scheme consisting of a Volterra nonlinear equalizer accompanied by a symbol-spaced whitening filter and a BCJR detector. On coherent 92GBd dual polarization (DP)-32QAM back-to-back measurements, the proposed soft-demapper matches the performance of the reference DSP. In 800 Gb/s 96GBd DP-32QAM 32-channel dense wavelength division multiplexing (DWDM) transmission over a 600 km fiber link the proposed approach outperforms the reference DSP.

End-to-end learning for amplified and unamplified links including binary mapping is proposed to improve the performance of optical coherent systems. 1.0dB and 1.2dB gains are demonstrated on coherent 92GbaudDP-32QAM 800Gb/s and 82GbaudDP- 128QAM 1Tb/s measurements, respectively.

Nearest-neighbour clustering is a simple yet powerful machine learning algorithm that finds natural application in the decoding of signals in classical optical-fibre communication systems. Quantum k-means clustering promises a speed-up over the classical k-means algorithm; however, it has been shown to not currently provide this speed-up for decoding optical-fibre signals due to the embedding of classical data, which introduces inaccuracies and slowdowns. Although still not achieving an exponential speed-up for NISQ implementations, this work proposes the generalised inverse stereographic projection as an improved embedding into the Bloch sphere for quantum distance estimation in k-nearest-neighbour clustering, which allows us to get closer to the classical performance. We also use the generalised inverse stereographic projection to develop an analogous classical clustering algorithm and benchmark its accuracy, runtime and convergence for decoding real-world experimental optical-fibre communication data. This proposed ‘quantum-inspired’ algorithm provides an improvement in both the accuracy and convergence rate with respect to the k-means algorithm. Hence, this work presents two main contributions. Firstly, we propose the general inverse stereographic projection into the Bloch sphere as a better embedding for quantum machine learning algorithms; here, we use the problem of clustering quadrature amplitude modulated optical-fibre signals as an example. Secondly, as a purely classical contribution inspired by the first contribution, we propose and benchmark the use of the general inverse stereographic projection and spherical centroid for clustering optical-fibre signals, showing that optimizing the radius yields a consistent improvement in accuracy and convergence rate.

In this article, we discuss challenges and options for scaling IM/DD transceivers towards 800Gbps. Our focus is CWDM4 PAM4 transmission and our target distance is 10km in O-band, which is a most urgent use case for next generation optical short reach systems like data centre interconnects and networks. At this reach and rate, chromatic dispersion (CD) becomes the main challenge. Its mitigation is essential and primarily done with digital signal processing. State of the art techniques, however, make transceivers quickly too complex. We show upon measurement results how neural network equalization can meet Volterra equalization performance with 30% less hardware multiplier complexity. When also applying magnitude weight pruning, an additional 43% reduction is possible without performance loss across all CWDM4 lanes. If needed, an added MLSE stage can further push performance in both cases. In any of these configurations, a key enabler against strong CD penalties is duobinary training, which is applicable to all feed-forward equalization architectures.