Poster Sessions

Amplification optimisation for short-reach high data rate coherent transmission systems

Mingming Tan, Ian D Phillips, Paul Harper, Wladek Forysiak, Aston Institute of Photonic Technologies, Aston University;  Md Asif Iqbal, BT Applied Research; T. T. Nguyen, Infinera Pennsylvania, USA; Paweł Rosa, National Institute of Telecommunications, Poland; Lukasz Krzczanowicz, Techical University of Denmark

We compared the transmission performances of 600 Gbit/s PM-64QAM WDM signals over 75.6 km of single mode fibre (SMF) using a set of representative amplifiers, which are EDFA, discrete Raman, hybrid Raman/EDFA, first-order, or second-order (dual-order) distributed Raman amplifiers. Our numerical simulations and experimental results show that the simple first-order distributed Raman scheme with backward pumping delivered the best transmission performance among all the schemes, particularly, better than the expected second-order Raman scheme which gave a flatter signal power variation along the fibre. Using the first order backward Raman pumping scheme indeed demonstrated a better balance between the ASE noise and fibre nonlinearity and gave an optimum transmission performance over a relatively short distance of 75 km SMF.


Inter-ONU Sample Timing Offset Estimation and Compensation for Spectrally Overlapped Orthogonal Channels in Hybrid OFDM-DFMA PONs

Omaro Gonem, Roger Philip Giddings, Jianming Tang, DSP Centre of Excellence, Bangor University

Upstream transmissions in hybrid OFDM-DFMA PONs with spectrally overlapped orthogonal channels can be highly sensitive to the inter-ONU STO between orthogonal signals from different ONUs thus causing severe performance degradations. In this work, we numerically simulate a hybrid OFDM-DFMA PON with 4 ONUs, we first show that, for a fixed EVM of -17dB (16-QAM), the inter-ONU STO dynamic range without compensation, for orthogonal baseband (passband) signals is only 0.6 (0.2) samples. We then demonstrate a novel OLT-based inter-ONU STO estimation and compensation algorithm with an inter-ONU STO estimation accuracy of ±0.08 (±0.04) sample intervals for baseband (passband) signals. We then show the OLT-based algorithm can successfully compensate for the estimated integer and fractional inter-ONU STO with only integer sample interval timing adjustments in the ONUs, thus avoiding the requirement and complexity of ONU-based fractional sample interval STO adjustments.


Real-time experimental demonstration of computationally efficient a hybrid OFDM DFMA-PON

Tushar Tyagi, R P Giddings, J M Tang, DSP Centre of Excellence, Bangor University

Hybrid OFDM DFMA PONs have been proposed to share a common fiber transmission medium with dynamic bandwidth allocation, adaptive modulation, optimum network resource utilization and dynamic on-demand connections/services. This work presents the first real time experimental demonstration of the aforementioned PON which is critical to technically verify its feasibility. DSP-based, digital orthogonal shaping filters are employed at the ONU transmitter to dynamically manipulate multiple channels, however the corresponding per-channel matching filters at the OLT receiver are all replaced by a single FFT operation to simultaneously recover all channels. A 2-ONU, 4-channel PON using a joint sideband processing technique is fully validated and analyzed in real-time. When compared to a DFMA PON, it is shown to offer increased performance, including 1.2 dBm lower received optical power and 15% higher channel capacities at the adopted FEC limit, better symbol timing offset tolerance, drastically reduced logic clock rate and significantly lower DSP complexity.


125 -µm cladding diameter Multi-core Fibre Design with Artificial Intelligence

Xun Mu, PhD Student; George Zervas, UCL

In the past decade, single-mode multi-core fibre (MCF) has been intensively researched because of its high spatial density, low digital signal processing complexity and compatibility with the standard single-mode fibre (SSMF). The design of the multi-core fibre is multi-dimensional and multi-constrained and therefore becoming time-consuming to optimize all the parameters. To automate and speed up the design process, neural networks are used for fibre optical properties calculation. Particle swarm algorithm is utilised to explore the optimal MCF structure according to various application requirements.
We designed 6-core and 8-core fibre in standard cladding diameter up to covering O to L band.  The designed MCFs have low crosstalk between -45.6 dB/km and –92.1 dB/km and comparable effective mode area to SSMF. The 6-core fibre offers capacity up to 1.2 Pb/s at 1200 km transmission while the 8-core fibre provides around 30% more capacity at the short-haul transmission than the 6-core fibre.

Multi-stage Raman amplifier for ultra-wideband signal amplification      

Pratim Hazarika, Early Stage Researcher, Aston University

To support ever-growing demand of data, there is a need to utilize the unused spectral bands of optical networks[1]. Existing WDM systems amplified using EDFAs can amplify only C and L band of the optical spectrum [2]. In this work, we propose a novel multistage discrete Raman amplifier extending from 1410–1625 nm of the optical spectrum providing a gain bandwidth of 215 nm.
The multistage amplifier uses split combine approach where signals are separated to two spectral bands, 1410-1457 nm in one stage and 1470-1625 nm in other stage. Inverse Dispersion Fibre is used as a gain fibre with optimized pump wavelengths and powers. VPI simulations were carried out for the proposed configuration, obtaining an average net gain of 14 dB and a maximum noise figure of 8 dB.
We believe multi stage Raman amplifier design is a novel approach towards ultra-wideband amplification for future optical communication systems.


Characterization of an interplay between transceiver noise and inter-stimulated Raman scattering on the performance of ultrawideband systems

Henrique Buglia, PhD student, UCL (Anastasiia Vasylchenkova, Eric Sillekens, Polina Bayvel, and Lidia Galdino  Optical Networks Group, University College London, UK)

Ultra-wideband transmission is a promising and cost-effective solution to meet the increasing demand for data traffic in optical fibre systems. However, the system performance and the Quality of Transmission (QoT) are limited by the transceiver noise [1-2] and the fibre nonlinearity, in particular, inter-channel stimulated Raman scattering (ISRS), which leads to a power transfer from short to long wavelengths. As a result of the latter, a per-channel launch power optimisation is required to maximise the system throughput and reduce the ISRS-induced QoT degradation [3-6].

This study shows an interplay between ISRS and transceiver noise. The latter mitigates the impact of the ISRS on the overall data throughput as well as on the per-channel QoT for short and metro networks. For these systems, it is shown that the transceiver noise reduces the impact of the launch power optimisation on the overall system performance and improves the system QoT.

Reduction of SBS effects in fibre-based optical phase conjugation of high-order modulation formats

Sonia Boscolo, Abdallah Ali, Andrew D Ellis, Aston Institute of Photonic Technologies, Aston University; Tu Nguyen, Infinera PA and Aston Institute of Photonic Technologies, Aston University

We overview our recent progress in mitigating the effects of stimulated Brillouin scattering in the fibre-based optical phase conjugation of high-order quadrature-amplitude modulation (QAM) signals, which are sensitive to the phase noise (PN) arising from residual pump dithering, by discussing two different approaches.  The first approach is based on optimising the pump dithering setup in order to achieve the best possible cancellation of the residual phase modulation. We demonstrate an improved counter-dithering scheme which realises a penalty below 0.2dB, thus enabling, for the first time, performance improvement in a 400-km long 256-QAM system. The second approach relies on the use of a suitable PN compensation algorithm in the receiver digital signal processing block to reduce the penalties associated with residual dithering. We demonstrate a new two-stage compensation scheme which achieves large performance improvement relative to conventional PN compensation when it is used with 16-/64-/256-QAM signals at high pump-phase mismatch levels.

RAMP: High-bandwidth, large-scale, all-to-all, nanosecond optical networks and MPI operations for high performance computing and distributed deep learning

Alessandro Ottino, UCL’s Optical Networks Group (ONG); George Zervas, UCL

Large-scale computing and distributed deep-learning systems strongly depend on network performance. Current electronic packet-switched (EPS) network architectures suffer from variable diameter topologies, low-bisection bandwidth and over-subscription affecting communication and collective operations completion times.
We introduce re-arreangeably and strictly non-blocking optical circuit-switched (OCS) architectures called RAMP. RAMP support all-to-all, single-hop, full-bisection bandwidth with increased node capacity (12.8Tbps), scale (65,536 nodes) and system capacity (0.84 Epbs). They allow nanosecond level network reconfiguration (<20ns) using space/wavelength/time-slot division multiplexing.
We developed a novel system-level algorithm and collective strategies (RAMP-x) that enable all MPI collective operations in a schedule-less and contention-less environment with low deterministic latency. The RAMP system achieves a 42-53× reduction in energy consumption with respect to equivalent EPS systems. The combination of system and algorithms leads to a minimum speed-up in collective completion time of 7.6-171× when compared to realistic Fat-Tree and 2D-Torus architectures for each collective operation at maximum scale.

Designing Low Complexity Neural Network Equalizers in Coherent Optical Systems

Pedro J Freire, PhD Student, Aston University of Photonic Technologies

The deployment of artificial neural networks (NN) based optical channel equalizers on edge-computing devices is critically important for the next generation of optical communication systems. However, due to the computational complexity of the NNs, necessary for the efficient equalization of nonlinear optical channels with huge memory, such solutions seem far from realistic hardware implementation. In this work, we address this issue by employing cutting-edge compression techniques on a recurrent NN-based optical channel equalization. We show that it is possible to lower the equalizers computational complexity below the level of the typical digital backpropagation (DBP) with one step per span and still provide the performance similar to that of DBP 3-STPS in the transmission over 20x50km SSMF link; 64QAM, 30GBd. 


Hardware-based NP-hard schedule computation for Optical Circuit Switched DCNs in sub-µs

Joshua Lawrence Benjamin, George Zervas, University College London (UCL)

The scalability and performance of today’s Data Centre networks are limited by over-subscription, long tail latencies, high energy consumption and cost of multi-tiered electronic packet-switched networks. Whilst sub-µs speed re-configurable optical circuit switched (OCS) networks can overcome these limitations and improve overall performance, they rely on (1) slow software-based algorithms, which take milliseconds to compute while packets today are communicated in nanosecond timescales, or (2) Cyclic schedule-less schemes which do not provide a demand-matched quality of service.

We develop a highly parallel and pipelined hardware-based algorithm and network schedule processing unit (NsPU) to compute a NP-hard schedule within 100s of nanoseconds. Moreover, the NsPU performs with over 90% normalized throughput (at 90% input load) for uniform and skewed traffic with a median and tail latency of 1.5-1.9 µs and 16-24 µs respectively, while consuming energy as low as 61 pJ/bit for a 131,072-node network with 12.8 Tbps per-node bandwidth.


Bismuth-doped fibre amplifiers as promising solutions for multi-band transmission

Aleksandr Donodin, Aston AiPT

There are three main current approaches to increase the capacity of fibre-optical transmission systems by the development of: the higher-order modulation formats, the spatial division multiplexing, and the multi-band transmission (MBT). Arguably, the most practical technique is the MBT capable to utilize the huge and still available spectral bandwidth of the existing fibre base. However, it involves a significant upgrade of current networks with novel amplifiers that are yet being developed and optimized.

The number of doped fibre media operating beyond C- and L- bands have been reported: neodymium, praseodymium, thulium, and bismuth. Unlike many other active dopants, Bi active centres allow the broadband amplification in the spectral range from 1150 to 1500 nm.  Such spectral flexibility of Bi-doped fibres makes them one of the most promising amplification tools for the MBT. Here, we present a summary of recent advances in Bi-doped fibre amplifiers and discuss the potential challenges for their implementation in transmission networks.


Experimental implementation of a neural network optical channel equalizer in restricted hardware using pruning and quantization

Diego Arguello Ron, PhD student, Aston University

Practical implementation of real-time hardware channel equalizers to mitigate transmission impairments in fibre-optic communication systems, is critically important for the development of a new generation of high-speed and high-capacity optical networks. A promising approach is using artificial neural networks (NNs), nevertheless computational complexity is the main limiting factor of this strategy. However, we manage to implement low-complexity NN-based optical channel equalizers by using the techniques known as pruning and quantization. Moreover, we demonstrate the feasibility of our solution by testing it experimentally in two types of resource-constrained hardware. Major novelty and advance of our work is that in addition to the complexity and inference time analysis, it also includes the energy consumption and the study of the impact that the characteristics of both the hardware and the model has on these metrics. This way we address the generally overlooked environmental impact.


Intelligent performance inference: A Graph Neural Network approach to modelling maximum achievable throughput in optical networks

Robin Matzner, UCL 

One of the key performance metrics for optical networks is the maximum achievable throughput for a given network. However, its calculation is an NP-hard optimisation problem, often solved via computationally expensive integer linear programming (ILP) formulations. These are infeasible to implement as objectives, even on very small node scales of a few tens of nodes. An alternative are heuristics, in conjunction with sequential loading, although these too require considerable computation time for large numbers of networks. There is, thus, a need for ultra-fast and accurate performance evaluation of optical networks. We propose the use of message passing neural networks (MPNN), to learn the relationship between, node and edge features, the structure and the maximum achievable throughput of networks. We demonstrate significant computational time savings (a reduction of 5 orders of magnitude compared to the ILP) such a model can achieve, whilst maintaining linear correlation values close to one.


Neural network for direct and inverse nonlinear Fourier transform

Egor Sedov, Aston University

We combine the nonlinear Fourier transform (NFT) signal processing with machine learning methods for solving the direct and inverse spectral problem associated with the nonlinear Schrodinger equation. The latter is one of the core nonlinear science models emerging in a range of applications. Our focus is on the unexplored problem of computing the continuous nonlinear Fourier spectrum associated with decaying profiles, using a specially-structured deep neural network. The second part of the work is devoted to the inverse transformation – the restoration of a signal from a continuous spectrum. The Bayesian optimisation is utilised to find the optimal neural network architecture. The benefits of using the neural network as compared to the conventional numerical NFT methods becomes evident when we deal with noise-corrupted signals, where the neural networks-based processing results in effective noise suppression. We show a neural network capable of reconstructing a signal from an already denoised continuous spectrum.


Implementation of Optical Phase Conjugation for the computational complexity reduction of Machine Learning-based equalisers

Karina Nurlybayeva, PhD student, Aston University

We experimentally demonstrate the combined effects of optimised multilayer perceptron (MLP) and optical phase conjugation (OPC) for nonlinearity compensation in an optical communication system. We have shown that this joint technique can significantly reduce the residual nonlinear effects. The performance of the proposed equaliser is demonstrated for a PDM 28 Gbaud 64QAM transmission in a 400 km link. Our results demonstrate a 12x BER reduction (below the 7% overhead HD-FEC threshold) applying an OPC-aided MLP-based nonlinear compensation scheme. Furthermore, we studied the effect of channel memory on the performance of the suggested equaliser by investigating the input vector size of the MLP. We established that the OPC reduces the effective channel memory made evident by the smaller required size and computational complexity of the equaliser. Our work suggests designing and implementing optical techniques as a solution to the pervasive problem of the computational complexity of DSPs in optical communication systems.




Poster session presenters