Beyond Exabit Optical Communications: towards transceiver integration
Lead Research Organisation:
UNIVERSITY COLLEGE LONDON
Department Name: Electronic and Electrical Engineering
Abstract
The aim of this fellowship is to develop disruptive approaches through theory and experiment to unlock the capacity of future information systems. To go beyond current optical fibre channel limits is arguably the greatest challenge faced by digital optical communications. To target it, the proposed research will combine techniques from information theory, coding, higher-dimensional modulation formats, digital signal processing, advanced photonic design, and machine learning to make possible breakthrough developments to ensure a robust communications infrastructure beyond tomorrow.
Optical communications have to-date been able to fulfil the ever-growing data demand whilst simultaneously reducing cost and energy-per bit. However, optical communications have now exceeded the fundamental capacity of existing single-mode technology a trend leading to a rapid duplication of line systems which in time will translate in less affordable broadband access. To meet future demands with prospective cost and energy savings and avoid the impending exhaust of fibre capacity, this fellowship offers a scalable path towards parallelism in optical fibre communications resembling the advent of parallel computing using multiple cores to sustain Moore's law - once we were unable to double the number of transistors in a single-core microprocessor. The emergent technology of spatial division multiplexing (SDM) provides much wider conduits of information by offering additional means for transporting channels over one single fibre, using multi-mode and multi-core fibres. The fellow has shown that the internal structure of optical fibres can be optimised to support thousands of different spatial paths, each with full transmission capacity. And, critically, that there are principal launching conditions that allow for full transmission rate over each path with a small fraction of the equalisation cost assumed before. These discoveries offer the potential to foster a revolution in how optical fibre communications networks operate to meet the ever-increasing traffic demand with decreasing cost and energy consumption per bit, enabling ubiquitous and universal broadband access.
This fellowship renewal envisages how to achieve chip-scale integration for multimode SDM transceivers packing intelligent optical beamforming powered by generalised machine learning and principled digital signal processing for highly spatially diverse fibre channels. Moreover, this fellowship renewal will initiate a new class of low crosstalk multi-mode fibres using elliptical cores and a new class of multimode optical fibre amplifiers with adaptive mode gain profile - opening fundamentally new theoretical and experimental possibilities up to now unexplored for SDM systems. These new developments will push multimode SDM technology far beyond that of the standard single-mode fibre infrastructure and bring it to an industry-ready development stage, unlocking decades of capacity growth in future optical networks with sustainable cost- and energy-per-bit.
Optical communications have to-date been able to fulfil the ever-growing data demand whilst simultaneously reducing cost and energy-per bit. However, optical communications have now exceeded the fundamental capacity of existing single-mode technology a trend leading to a rapid duplication of line systems which in time will translate in less affordable broadband access. To meet future demands with prospective cost and energy savings and avoid the impending exhaust of fibre capacity, this fellowship offers a scalable path towards parallelism in optical fibre communications resembling the advent of parallel computing using multiple cores to sustain Moore's law - once we were unable to double the number of transistors in a single-core microprocessor. The emergent technology of spatial division multiplexing (SDM) provides much wider conduits of information by offering additional means for transporting channels over one single fibre, using multi-mode and multi-core fibres. The fellow has shown that the internal structure of optical fibres can be optimised to support thousands of different spatial paths, each with full transmission capacity. And, critically, that there are principal launching conditions that allow for full transmission rate over each path with a small fraction of the equalisation cost assumed before. These discoveries offer the potential to foster a revolution in how optical fibre communications networks operate to meet the ever-increasing traffic demand with decreasing cost and energy consumption per bit, enabling ubiquitous and universal broadband access.
This fellowship renewal envisages how to achieve chip-scale integration for multimode SDM transceivers packing intelligent optical beamforming powered by generalised machine learning and principled digital signal processing for highly spatially diverse fibre channels. Moreover, this fellowship renewal will initiate a new class of low crosstalk multi-mode fibres using elliptical cores and a new class of multimode optical fibre amplifiers with adaptive mode gain profile - opening fundamentally new theoretical and experimental possibilities up to now unexplored for SDM systems. These new developments will push multimode SDM technology far beyond that of the standard single-mode fibre infrastructure and bring it to an industry-ready development stage, unlocking decades of capacity growth in future optical networks with sustainable cost- and energy-per-bit.
Organisations
- UNIVERSITY COLLEGE LONDON (Lead Research Organisation)
- OFS (Collaboration, Project Partner)
- Xtera Communications (Collaboration)
- UNIVERSITY OF NOTTINGHAM (Collaboration)
- Fraunhofer Society (Collaboration)
- Corning Inc. (Collaboration)
- Nokia (Collaboration)
- Montana State University (Collaboration)
- Corning Incorporated (International) (Project Partner)
- NICT (Project Partner)
- Fraunhofer HHI (Project Partner)
Publications
Barbosa F
(2025)
Enhancing the Vertical Resolution of Coherent Optical Receivers
in Journal of Lightwave Technology
Barbosa F
(2026)
Quasi-Single-Mode Transmission Over Ultra-Low-Loss Few-Mode Fibre for Data Centre Interconnects
in IEEE Photonics Technology Letters
Barbosa F
(2025)
Universal Fibre for Coherent Data Centre Interconnect Applications
in Journal of Lightwave Technology
Barbosa F
(2024)
Modal Dispersion Performance of Mode Vector Modulation
Barbosa F
(2025)
Universal Fibre for Coherent Data Centre Interconnect Applications
in Journal of Lightwave Technology
Bessin F
(2026)
Polarization-Insensitive Fiber Optical Parametric Amplifier With Noise Figure Down to 4.4 dB Enabled by Mach-Zehnder Architecture
in Journal of Lightwave Technology
Carniello P
(2025)
Closed-Form Expressions for Nonlinearity Coefficients in Multimode Fibers
in IEEE Journal on Selected Areas in Communications
| Description | Global demand for data is growing extremely rapidly, driven by cloud services, artificial intelligence (AI), and the increasing exchange of information between large data centres. Optical fibres form the backbone of this global digital infrastructure, but existing transmission technologies are approaching fundamental limits in how much information they can carry. To continue scaling capacity while keeping energy consumption manageable, new approaches to optical communication are required. This project explores the use of spatial division multiplexing (SDM), a technique that allows multiple data streams to be transmitted simultaneously through different spatial channels within the same optical fibre. While SDM can dramatically increase transmission capacity, signals travelling in neighbouring channels can interfere with one another, creating a challenge known as inter-mode crosstalk. If not properly controlled, this interference limits performance and increases the complexity and energy consumption of signal processing systems. The research has developed new optical computing and machine learning techniques that operate directly on optical signals to mitigate this interference. In particular, the project investigates optical neural network approaches capable of learning how to suppress crosstalk between channels in real time. By performing part of the signal processing in the optical domain rather than electronically, these techniques offer the potential for much faster and more energy-efficient data recovery. As a result of this work, new experimental and modelling tools have been created to study crosstalk dynamics in spatially multiplexed fibre systems and evaluate the performance of optical neural network processors. Early demonstrations show that optical computing techniques can significantly reduce interference between spatial channels, enabling higher transmission capacity and more scalable fibre systems. These results contribute to the development of next-generation optical communication technologies that will be required to support the continued growth of AI, cloud computing, and global data infrastructure. The findings have been disseminated through international conferences and journal publications, helping to advance research into energy-efficient, high-capacity optical networks. |
| Exploitation Route | The outcomes of this project can inform the development of next-generation optical communication systems using spatial division multiplexing (SDM). The techniques developed to mitigate inter-mode crosstalk using optical neural networks may be used by researchers and industry to design more scalable and energy-efficient fibre transmission technologies. The experimental results and modelling approaches also provide a basis for further research and development in advanced optical networking. |
| Sectors | Aerospace Defence and Marine Digital/Communication/Information Technologies (including Software) Electronics |
| Description | The research is informing the development of next-generation optical communication technologies through ongoing engagement with industry partners and the wider telecommunications ecosystem. Experimental results from the project are helping guide the design of future optical transceivers and fibre transmission systems, and are being discussed with companies working on advanced fibre technologies and optical networking equipment, including collaborators such as Corning and OFS. The work also contributes to the broader technology roadmap for scalable optical networks that will underpin future telecom and data-centre infrastructure, with potential longer-term opportunities for commercialisation and spin-out activity. In addition, the findings support discussions within professional bodies such as the IEEE and provide technical insights to industry and standards communities on emerging optical networking technologies. |
| First Year Of Impact | 2025 |
| Sector | Aerospace, Defence and Marine,Communities and Social Services/Policy,Digital/Communication/Information Technologies (including Software),Electronics,Manufacturing, including Industrial Biotechology,Security and Diplomacy |
| Impact Types | Economic Policy & public services |
| Description | 2024 Regular Fondecyt National Projects Competition Chile - Peer review |
| Geographic Reach | Multiple continents/international |
| Policy Influence Type | Participation in a guidance/advisory committee |
| Impact | No impacts yet |
| Description | Hack Your Research: Open-Source Tools & Lab Automation Event at Conferences |
| Geographic Reach | Multiple continents/international |
| Policy Influence Type | Influenced training of practitioners or researchers |
| Impact | The workshop has driven a tangible shift in research practice within the optical communications community, encouraging early career researchers and students to move away from duplicating existing work and towards openly sharing and reusing tools and resources. Each edition of the event has attracted between 150+ attendees drawn from the international research community, demonstrating sustained and significant reach over approximately 12 years of delivery at ECOC and OFC. The consistent repeat attendance across editions is itself evidence of the value participants place on the event, with many returning year on year and recommending it to peers through word of mouth, social media. And, this event has been included in conference reports. Concrete changes in practice include the adoption of open-source tools by researchers who would otherwise have developed solutions independently, and the wider dissemination of tools and resources shared online following each event. By providing a dedicated space to learn from process rather than polished final results, the workshop has helped normalize a more open and collaborative approach to research, one that is still relatively uncommon in traditional conference settings. Over its 12-year lifespan, the cumulative reach of the event spans thousands of researchers internationally, contributing to a growing culture of openness and knowledge sharing in optical communications research. |
| URL | https://hack-your-research.gitlab.io/post/ |
| Description | Massive MIMO Standardization Working Group |
| Geographic Reach | Multiple continents/international |
| Policy Influence Type | Participation in a guidance/advisory committee |
| Impact | This working group will be working to create and write the new IEEE standard for an architectural framework for a Massive MIMO system. A standard is needed to provide a general framework for the industry to develop interoperable Massive MIMO components and products. The scope cover 6G technologies such as OAM multiplexing and terahertz communications. |
| URL | https://standards.ieee.org/ieee/1942.1/10956/ |
| Description | National Science Centre Poland - Peer review |
| Geographic Reach | Multiple continents/international |
| Policy Influence Type | Participation in a guidance/advisory committee |
| Impact | No impacts yet |
| Description | Raising the Profile of Women in Engineering at UCL Electronic and Electrical Engineering |
| Geographic Reach | National |
| Policy Influence Type | Influenced training of practitioners or researchers |
| Impact | This contribution helped raise the profile of women in engineering by providing authentic, personal accounts of the challenges and rewards of an engineering career. Published on the UCL Engineering website and linked to the official INWED and Women's Engineering Society platforms, the feature reached prospective students, current researchers, and the wider engineering community nationally and internationally. The video produced for Women's Day at UCL further extended this reach through UCL's public communications channels. By sharing honest reflections on issues such as immigration barriers, workplace confidence, and work-life balance, the contribution offered relatable and inspiring content particularly valuable to young women and girls considering engineering. The continued institutional support for INWED at UCL, now in its tenth year since the day's founding in 2014, reflects the sustained cultural impact of such awareness raising activities. |
| URL | https://www.ucl.ac.uk/engineering/news/2023/jun/ong-celebrates-inwed-2023 |
| Description | UKRI Future Leaders Fellowships - Peer review |
| Geographic Reach | Europe |
| Policy Influence Type | Participation in a guidance/advisory committee |
| Impact | No impacts yet |
| Description | High-capacity and adaptive optical fibre networks |
| Amount | £105,000 (GBP) |
| Funding ID | 2930719 |
| Organisation | Engineering and Physical Sciences Research Council (EPSRC) |
| Sector | Public |
| Country | United Kingdom |
| Start | 08/2024 |
| End | 09/2028 |
| Title | Data underpinning "Experimental Demonstration of Zero-Shot Machine Learning Equalisation in Dual Polarisation Coherent Transmission" |
| Description | "graph_data.xlsx" is an excel spreadsheet containing the graph data used in the paper. It contains three sheets for each of the three figures in the paper."Q_fibres" contains a table containing the Q-factor achieved by both conventional and ML-based DSP. The launch power used is contained in the "LP" column. Each other column is labeled as "fibre-dsp", where "fibre" denotes the fibre being used to transmit data, and "dsp" is either "conv" for conventional DSP or "ml" for ML-based DSP."Q_freqs" contains a table structured in the same way as the table in "Q_fibres", except each column is labeled as "freq-dsp" where "freq" denotes the transmission frequency."Q_PMD" contains a table containing Q-factors achieved by ML-based DSP. Each column is labeled by the induced delay used at the transmitter to generate each result, and the column "LP" contains the launch power used. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange via graph data points from respective publication. |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_Experimental_Demonstration_of_Zero-Shot_Mac... |
| Title | Data underpinning "Fully Generalized Machine Learning-Based Equalization in Coherent Optical Transmission" |
| Description | "graph_data.xlsx" contains data displayed in the graphs present in the paper. It contains two sheets for each of the two graphs in the paper."BER_Fibres" contains a table with the bit error rate achieved by both conventional and ML-based DSP. The launch power used is contained in the "LP" column in dBm. Each other column is labeled as "fibre-dsp", where "fibre" denotes the fibre being used to transmit data, and "dsp" is either "conv" for conventional DSP or "ml" for ML-based DSP."BER_PMD" is structured the same as the previous sheet, except containing columns from data with different amounts of polarisation delay. Each column is labeled by the amount of delay used at the transmitter, and launch power is given in the "LP" column in dBm. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange via graph data points from respective publication. |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_Fully_Generalized_Machine_Learning-Based_Eq... |
| Title | Data underpinning "Generalization Capabilities of Machine Learning-based PDM Equalization" |
| Description | "graph_data.xlsx" is an excel spreadsheet containing the graph data. There are two sheets, "Nonlinearities" which contains the data in Fig 2a, and "Dispersion" containing the data in Fig 2b. In each sheet, the first column is the X axis and further columns are the Y values. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2023 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange via graph data points from respective publication. |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_Generalization_Capabilities_of_Machine_Lear... |
| Title | Data underpinning "On a Scalable Path for Multimode SDM Transmission" |
| Description | Data points for figures of the research paper "On a Scalable Path for Multimode SDM Transmission". File format: '.csv (comma-separated values)'. Use a standard text editor or spreadsheet program to visualize the data. The name of the files was defined according to the figures in the paper. Please see below a brief description of the files: Fig3_PMs_GD_vs_PMIndex: Group Delay Spread: Evolution of the group delay (GD) spread across different groups of transmitted principal modes. Each column contains the group delays for a given group of transmitted principal modes. Group delay are sorted in ascending order. This file can be used to reproduce Fig. 3. Fig4a_Fig4b_PMs_XT_f_vs_PMindex0.01s - 10s: These files contain the XT values for each principal mode and modified principal mode, for channels with a drift of 0.01-10s. These files can be used to generate Figs. 4(a) and 4(b). Fig4c_PMs_Inter_vs_PMindex0.01 - 10s: These files contain the number of interfering terms for each principal mode under channels with a drift of 0.01-10s. These files can be used to generate Figs. 4(c), and correspond to the same simulation scenarios used to obtain Figs. 4(a) and 4(b). Fig6_MISOarray_PMindex_tau_env_1_to_0.01_NModes_82PMs: Multiple-input Single-Output array size to guarantee a target SNR of 24 dB under different channel drift conditions when 82 principal modes are transmitted. Each column represents the array size for a given channel drift condition. The values are sorted in ascending order. This file can be used to reproduce Fig. 6. Fig5_PMs_vs_MPMs_tau_env_0.1: Average SNR as a function of the equaliser array size for principal modes and modified principal modes. This file contains results for different numbers of transmitted principal modes, ranging from 22 to 144 principal modes. It allows full reproduction of Fig. 5. Fig6_MISOarray_PMindex_tau_env_1_to_0.01_NModes_144PMs: Multiple-input Single-Output array size to guarantee a target SNR of 24 dB under different channel drift conditions when 144 principal modes are transmitted. Each column represents the array size for a given channel drift condition. The values are sorted in ascending order. This file can be used to reproduce Fig. 6. Fig7_TotalCapacity_vs_NormalisedMISO: Throughput as a function of normalised equaliser array size for different groups of transmitted principal modes, ranging from 22 to 144. A channel drift condition of 0.1s was considered. This file allows full reproduction of Fig. 7. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2026 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_On_a_Scalable_Path_for_Multimode_SDM_Transm... |
| Title | Data underpinning "On equalisation requirements for Scalable SDM Transmission" |
| Description | Data points for the paper "On equalisation requirements for Scalable SDM Transmission". This is an inveted paper at the 2023 SPPCOM. Fig2a_GDspred_vs_PMindex_tenv48s.csv: Group delay spread as a function of principal mode index (sorted in ascending order) for two methods investigate for principal mode selection. The first column contains the group delays obtained when choosing principal modes based on their group delays (lowest values first) and the second column contains the group delays when selection is done based on crosstalk. This file allows reproducing Fig. 2(a). Fig2b_PMs_XT_vs_PMindex_tenv48s.csv: Analysis of crosstalk scaling as a function of the principal mode index for two methods investigated for the selection of principal modes. The first eight columns contain the results obtained when using the group delay for the selection of principal modes. The subsequent eight columns contain the results obtained when the crosstalk is used for the selection of principal modes. This file allows full reproduction of Fig. 2(b). Fig2c_AverageMISOarray_vs_Nt.csv: Average equaliser array size as a function of the number of transmitted principal modes. The first seven columns contain the values for the selection of principal modes based on their group delays, while the subsequent seven columns refer to results for the crosstalk-based selection. This file allows full reproduction of Fig. 2(c).File format: '.csv (comma-separated values)'. Use a standard text editor or spreadsheet program to visualize the data. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2026 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_On_equalisation_requirements_for_Scalable_S... |
| Title | Data underpinning "On the advantages of principal modes for multimode SDM transmission systems" |
| Description | Data points for all figures of the research paper "On the advantages of principal modes for multimode SDM tranmission systems".1) File Fig2a.csv - The multimode fiber channel matrix H(w) at the center frequency. This file correponds to a 2-D matrix showing the absolute value of the coupling coefficient.2) File Fig2b.csv - Residual channel when using the principal modes for transmission. This file correponds to a 2-D matrix showing the absolute value of the coupling coefficient.3) File Fig3a.csv - Group delay per principal mode in the selected group;4) File Fig3b.csv - Crosstalk per principal mode in the selected group. Each column contains the crosstalk observed when transmitting a given number of principal modes.5) File Fig3c.csv - Multiple-input single-output array size required to detect each principal mode in the selected group. Each column corresponds to the array size for a different number of transmitted principal modes.6) File Fig4a.csv - Crosstalk per principal mode when 330 tributaries are transmitted. Each column corresponds to the crosstalk of a given principal mode in Fig. 4(a).7) File Fig4b.csv - Multiple-input single-output array size required to detect each tributary (when 330 principal modes are used for transmission). Each column corresponds to the required array size to guarantee the target SNR of 20 dB.File format: '.csv (comma-separated values)'. Use a standard text editor or spreadsheet program to visualize the data. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2026 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_On_the_advantages_of_principal_modes_for_mu... |
| Title | Data underpinning "Single-Mode Transmission Over Ultra-Low-Loss 0.1400 dB/km Few-Mode Fibre for Data Centre Interconnects" |
| Description | This dataset contains the data points of all figures on the paper "Quasi-Single-Mode Transmission over Ultra-Low-Loss Few-Mode Fibre for Data Centre Interconnects". A description of the files is presented below: nli_ase_fitting.xlsx, snr_vs_lpower_fitting.xlsx & snr_vs_lpower_fibrea_ullfmf.xlsx: These files contain the signal-to-noise ratio as a function of launch power for the full reproduction of Fig. 2(a). filter_taps.xlsx: Filter weights obtained for the utra-low-loss few-mode fibre at optimum launch power. This file allows full reproduction of Fig. 2(b). mpi_vs_wav.xlsx: This file contains the multipath interference results as a function of the wavelength obtained based on power samples and on the modified Gaussian model. It allows full reproduction of Fig. 3. snr_vs_wav_nodbp.xlsx: This file contains the signal-to-noise ratio analysis as a function of the wavelength for fibre A and the ultra-low-loss few-mode fibre. This file allows full reproduction of Fig. 4. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2026 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_Single-Mode_Transmission_Over_Ultra-Low-Los... |
| Title | Data underpinning "Universal Fibre for Coherent Data Centre Interconnect Applications" |
| Description | This dataset contains all the data points required to reproduce the figures from the paper "Coherent DWDM Single-Mode Transmission over Universal Fibre for Data Centre Interconnects". A description of the files is provided below: snr_vs_lp_om2.xlsx, snr_vs_lp_uf_off.xlsx, snr_vs_lp_uf.xlsx & snr_vs_lp_fibre_a.xlsx contain the signal-to-noise ratio for all wavelengths and launch power configurations tested in Fig. 2 for the OM2 fibre, Universal Fibre with 3.4um offset, Universal Fibre and Fibre A, respectively. These files also include the launch power. They allow full reproduction of Fig. 2(a). snr_vs_wav.xlsx: Signal-to-noise ratio as a function of the wavelength for the Universal Fibre with 3.4um offset, Universal Fibre and Fibre A. The wavelength at which performance was measured is also included. This file allows full reproduction of Fig. 2(b). snr_vs_CIR.xlsx: Signal-to-noise ratio as a function of the channel impulse response length for the Universal Fibre with 3.4um offset, Universal Fibre with 2um offset, Universal Fibre and Fibre A. The results were obtained at three different wavelength, namely 1540 nm, 1550 nm and 1560 nm. This file allows full reproduction of Fig. 3(a). dwdm_snr_vs_wav.xlsx: Line rate measured in terms of the generalised mutual information as a function of the wavelength for the Universal Fibre with 3.4um offset, Universal Fibre with 2um offset, Universal Fibre and Fibre A. The wavelength at which performance was measured is also included. This file allows full reproduction of Fig. 3(b). snr_vs_power_NTSsb_64_to_2048_FibreA.xlsx: Signal-to-noise ratio as a function of the launch power obtained with the fibre A for different training sequence lengths (from 64 to 2048 symbols). The first six columns correspond to the launch power at which performance was measured. The subsequent six columns correspond to the signal-to-noise ratio values for the different training sequence lengths. Columns are sorted in ascending order with respect to the training sequence length. This file allows full reproduction of Fig. 5(a). snr_vs_power_NTSsb_64_to_2048_UF.xlsx: Signal-to-noise ratio as a function of the launch power obtained with the universal fibre for different training sequence lengths (from 64 to 2048 symbols). The first six columns correspond to the launch power at which performance was measured. The subsequent six columns correspond to the signal-to-noise ratio values for the different training sequence lengths. Columns are sorted in ascending order with respect to the training sequence length. This file allows full reproduction of Fig. 5(b). Files titled snr_vs_time_UF_chRate and snr_vs_time_FibreA_chRate provide the signal-to-noise ratio obtained with the universal fibre and fibre A, respectively, after processing 4967 consecutive frames under different channel estimation rates (i.e. for every frame, 10th and 100th frame). These files allow full reproduction of Fig. 6 and Fig. 7. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2026 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_Universal_Fibre_for_Coherent_Data_Centre_In... |
| Title | Data underpinning "Zero-shot ML Equalization in Coherent Optical Transmission: a First Experimental Demonstration" |
| Description | All data files are *.txt files in a space-delimited format. The first row are the column titles, and subsequent rows are data.fig5_loss_curve.txt: Data for Figure 5 describing the MSE loss in simulation and experimental Q-factor (at optimal launch power) during the training process. The "epoch" column contains the training epoch, the "loss" column contains the MSE loss in simulation, and "qs" contain the experimental Q-factor.fig6_fibres.txt: Data for Figure 6 showing the experimental Q-factor vs launch power for different fibres. Each column title is in the format of "FIBRE-ml" for the Q-factor achieved by the neural network, and "FIBRE-conv" for the Q-factor achieved by the conventional linear DSP. "FIBRE" is one of: SSMF, NZDSF, PSCF, GI-MMF, each describing the Q-factor achieved over standard single-mode fibre, non-zero dispersion-shifted fibre, pure silica-core fibre, and graded-index multi-mode fibre respectively. Lastly, there is the "LP" column denoting the launch power in dBm.fig7_freqs.txt: Data for Figure 7 showing the experimental Q-factor vs launch power for different transmission frequencies over SSMF. Each column title is in the format of "FREQ-ml" for the Q-factor achieved by the neural network, and "FREQ-conv" for the Q-factor achieved by the conventional linear DSP. "FREQ" is one of 191.556, 192.175, 192.788, 193.414, 194.045, 194.67, and 195.31, each denoting the transmission frequency in THz. Lastly, there is the "LP" column denoting the launch power in dBm.fig8_pol_skew.txt: Data for Figure 8 showing the experimental Q-factor vs launch power for different polarisation skew introduced at the transmitter. Each column is in the format "Xps", where "X" can take the value of 7, 5, 3, 1, 0.7, 0.3, and 0, each denoting the polarisation skew in picoseconds. Lastly, there is the "LP" column denoting the launch power in dBm.fig9_ps_size: Data for Figure 9 showing the experimental Q-factor vs launch power for different training sequence sizes. This folder contains multiple space-delimited text files title "psN", where N denotes the length of the pilot sequence in symbols . Each column is titled with one of SSMF, NZDSF, GI-MMF, or PSCF to denote the different fibres, and contains a "LP" column to denote the launch power in dBm. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2025 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange via graph data points from respective publication. |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_Zero-shot_ML_Equalization_in_Coherent_Optic... |
| Title | Data underpinning Single-Mode Coherent Transmission over Universal Fiber for Data Center Interconnects |
| Description | This dataset contains the data points of all figures in the paper 'Single-Mode Coherent Transmission over Universal Fiber for Data Center Interconnects'. A description of the files is presented below: snr_vs_lpower_dbp.xlsx: This file contains the signal-to-noise ratio values obtained for each fibre with and without digital backpropagation. It also includes the launch power values at which performance was measured. This file allows full reproduction of Fig. 3. sim_snr_vs_lpower_mw.xlsx & exp_snr_vs_lpower_mw.xlsx: These two files contain, respectively, the simulation and experimental results obtained with the universal fibre with and without mandrel wrapping. The launch power at which performance was measured is also included. They allow full reproduction of Fig. 4. snr_vs_ntaps.xlsx: This file contains the signal-to-noise ratio values for each fibre and symbol rate configuration tested in Fig. 5. The number of taps is also included. This file allows full reproduction of Fig. 5. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2026 |
| Provided To Others? | Yes |
| Impact | Knowledge exchange |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Data_underpinning_Single-Mode_Coherent_Transmission_over_Univ... |
| Title | Dataset - Throughput Performance in Mode-Group Division Multiplexing Metro-Scale Data Centre Interconnects |
| Description | Datasets used in the published paper: Throughput Performance in Mode-Group Division Multiplexing Metro-Scale Data Centre Interconnects. DOI: 10.1109/ACP66871.2025.11349984. The attached files is the datacenter locations and edge information in Fig 1. The following numbers is the blue points in Fig.2: Demand List: [70.2, 140.4, 210.6, 280.8, 351.0, 421.2, 491.4, 561.6, 631.8, 702.0, 772.2, 842.4, 912.6, 982.8, 1053.0, 1123.2] Throughput List: [70.2, 140.4, 210.6, 280.8, 351.0, 421.2, 481.5, 518.4, 552.6, 582.0, 613.8, 631.2, 643.2, 646.2, 645.3, 643.2] |
| Type Of Material | Database/Collection of data |
| Year Produced | 2025 |
| Provided To Others? | Yes |
| Impact | This dataset and model estimated the network throughput of applying few-mode fiber (FMF) in a regional datacenter interconnects, using a novel resource allocation algorithm. |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Dataset_-_Throughput_Performance_in_Mode-Group_Division_Multi... |
| Title | Generalized Machine Learning Equalization in Coherent Receivers |
| Description | Dataset underpinning manuscript "Generalized Machine Learning Equalization in Coherent Receivers". "graph_data.xlsx" is an excel spreadsheet containing the graph data. There are two sheets, "Launch Power" which contains the data in Fig 2a, and "Dispersion" containing the data in Fig 2b. In each sheet, the first column is the X axis and further columns are the Y values. "train_data.pt" and "val_data.pt" contain the training and validation data used in this paper (testing data was generated at test time). Both were generated in Pytorch 2.1, and can be accessed using the built in "torch.load" function supplied by Pytorch. Each file contains a list of dictionaries with keys "rx_training_sequece", "rx_data_sequence", and "tx_data_symbols", denoting the received training sequence, received data sequence, and transmitted symbols respectively, with the corresponding values being complex tensors. "training_sequence.pt" contains the pulse-shaped training sequence, generated in Pytorch 2.1 and can be accessed in the same way as above, containing a complex tensor. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2023 |
| Provided To Others? | Yes |
| Impact | Setting the foundation for novel machine learning-based DSP for optical coherent communications. |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Generalized_Machine_Learning_Equalization_in_Coherent_Receive... |
| Title | Modal Dynamics in Multimode Optical Fibers for All Linear Coupling Regimes |
| Description | The uploaded file is the dataset for our manuscript named: "Modal Dynamics in Multimode Optical Fibers for All Linear Coupling Regimes", which is submitted for publication in Optics Express. MATLAB software is used to carry out the extensive simulations as discussed in the manuscript. The data uploaded is used to plot figures in the submitted manuscipt. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | The uploaded file is the dataset for our manuscript named: "Modal Dynamics in Multimode Optical Fibers for All Linear Coupling Regimes", which is submitted for publication in Optics Express. MATLAB software is used to carry out the extensive simulations as discussed in the manuscript. The data uploaded is used to plot figures in the submitted manuscipt. |
| URL | https://rdr.ucl.ac.uk/articles/dataset/Modal_Dynamics_in_Multimode_Optical_Fibers_for_All_Linear_Cou... |
| Description | Bell Labs Internship |
| Organisation | Nokia |
| Department | Nokia Bell Labs |
| Country | United States |
| Sector | Private |
| PI Contribution | Successfully characterised and validated in-house multi-plane-light-converters for mode multiplexing. |
| Collaborator Contribution | Provided training and equipment for device characterisation. |
| Impact | Successfully validated in-house lithography/manufacturing methods for manufacturing high-bandwidth MPLC devices. Manufactured fibre arrays for multimode launching. |
| Start Year | 2025 |
| Description | Collaboration with OFS |
| Organisation | OFS |
| Department | OFS Labs |
| Country | United States |
| Sector | Private |
| PI Contribution | Planned joint experiments |
| Collaborator Contribution | Supply of few-mode fibres |
| Impact | plan to publish in journals and conferences in the future. |
| Start Year | 2022 |
| Description | Improving Operational Efficiency of High-Capacity Transmission Systems |
| Organisation | Xtera Communications |
| Country | United Kingdom |
| Sector | Charity/Non Profit |
| PI Contribution | My team brings expertise and insights into operating and optimizing high-capacity transmission systems, offering proposals for enhancements during the development stage of field systems. |
| Collaborator Contribution | Insight and data for field deployed systems as well as access to high-speed instrumentation. |
| Impact | Currently working towards a joint research paper. |
| Start Year | 2024 |
| Description | Multi-dimensional modulation formats for energy efficient SDM transmission systems |
| Organisation | Montana State University |
| Country | United States |
| Sector | Academic/University |
| PI Contribution | My teams contributes with our expertise in principled transmission schemes for space division multiplexing systems, including theoretical and experimental capabilities. |
| Collaborator Contribution | Montana State University offers expertise in the design of multi-dimensional modulation formats with improved energy efficiency. |
| Impact | Currently working towards a joint research paper. |
| Start Year | 2024 |
| Description | Optimisation and fabrication of advanced multi-mode/-core optical fibres |
| Organisation | Corning Inc. |
| Country | United States |
| Sector | Private |
| PI Contribution | Work on developing advanced optical multi-mode/-core fibres, focused on novel refractive index profiles allowing to scale the number of modes suitable for inter-datacentre applications. |
| Collaborator Contribution | Corning offers expertise on fibre design and optimisation for fabrication. |
| Impact | Currently working towards a joint research paper. |
| Start Year | 2023 |
| Description | Optimisation and fabrication of array integrated high-speed transmitter |
| Organisation | Fraunhofer Society |
| Department | Fraunhofer Heinrich Hertz Institute |
| Country | Germany |
| Sector | Academic/University |
| PI Contribution | Design and optimisation of an array integrated high-speed transmitter. |
| Collaborator Contribution | Optimisation and fabrication of an array integrated high-speed transmitter, |
| Impact | Plan to publish in journals and conferences in the future. Unique (future) testbed to be accessible to UK researchers. |
| Start Year | 2022 |
| Description | Phase light modulation |
| Organisation | University of Nottingham |
| Country | United Kingdom |
| Sector | Academic/University |
| PI Contribution | UCL set up the a novel type of transmission system for high-capacity SDM transmission, carried out fibre characterisation, signal measurement, and signal processing |
| Collaborator Contribution | Provided know-how of phase light modulation. Provided access to prototype devices. |
| Impact | Paper submission to ECOC 2026. |
| Start Year | 2025 |
| Title | OPTICAL SIGNAL CONVERTER AND CONVERSION METHOD |
| Description | It is an aim of the present disclosure to provide a method for converting an optical signal into an electrical signal and a signal converter which at least partially address one or more of the challenges discussed above. It is a further aim of the present disclosure to provide a method for converting an optical signal into an electrical signal and a signal converter with a lower ENoB than conventional converters, in order to allow for a wider analog bandwidth. Accordingly, it is an aim of the present disclosure to provide methods and apparatus for converting optical signals with an improved resolution, for example by 0.5 bits. In particular, it is an aim of the present disclosure to provide improved receiver sensitivity and/or the performance of digital nonlinear compensation techniques such as DBP by providing a signal converter with improved ENoB. The present disclosure may use optical signal and optoelectrical processing rather than conventional electronic processing, in order to deliver broadband signal processing with higher energy efficiency and ultra-low-noise. An embodiment of the present disclosure provides a method for converting an optical signal into an electrical signal. The method comprises receiving the optical signal, splitting the optical signal into a first signal and a second signal, and measuring intensity information of the first signal. The method further comprises processing the first signal using an intensity modulator to generate an inverse intensity waveform, combining the inverse intensity waveform and the second signal to generate a constant intensity waveform, converting the constant intensity waveform into a digital constant intensity signal using an Analog-to-Digital Converter (ADC), combining the intensity information and the digital constant intensity signal, and generating the electrical signal using a Digital Signal Processor (DSP). A further embodiment of the present disclosure provides signal converter for converting an optical signal into an electrical signal comprising an intensity modulator, an Analog-to-Digital Converter (ADC), and a Digital Signal Processor (DSP). The converter is configured to receive the optical signal, split the optical signal into a first signal and a second signal, and measure the intensity information of the first signal. The converter is further configured to process the first signal using the intensity modulator to generate an inverse intensity waveform, combine the inverse intensity waveform and the second signal to generate a constant intensity waveform, convert the constant intensity waveform into a digital constant intensity waveform using the ADC, combine the intensity information and the digital constant intensity signal, and generate the electrical signal using the DSP. |
| IP Reference | |
| Protection | Patent / Patent application |
| Year Protection Granted | 2024 |
| Licensed | No |
| Title | Spatial light modulator based beam manipulation |
| Description | This repo converts a matrix for displaying on a slm. The matrix is converted into different grating patterns for controlling transmitted light into the first order. This is useful for optical matrix multiplication. To check holograms, take Fourier transform to visualise first order diffractions. |
| Type Of Technology | Physical Model/Kit |
| Year Produced | 2023 |
| Impact | No impact yet. |
| URL | https://github.com/beyondexabit/slm_matrix_gen |