2026 Poster Sessions

Topology Architect: AI for Optical Network Design
Akanksha Ahuja PhD Student University of Cambridge
Automating the design of optical networks is an open challenge. Existing optimisation methods are computationally intensive beyond 100-nodes. Deterministic and probabilistic graph generators are overly simplistic and cannot exploit historical designs. Public datasets remain underutilised for machine-learning-based design. We introduce Topology-Architect, an AI-based graph generative model for optical networks that learns and reproduces complex patterns from 105 physical topologies (Topology-Bench, 1970s–present). It captures structural and spectral properties while enforcing biconnectivity and sparsity. We show three findings. First, generated networks match structural (84%) and spectral (95%) graph properties of real topologies and remain within the empirical range of real networks. Second, sub-second generation of 100-node graphs demonstrates scalability. Third, high-dimensional network embeddings from our model result in much more distinct clusters when projected to two dimensions than embeddings derived from graph features. This indicates an order-of-magnitude increase in topological diversity measured by inter-cluster distances in the embedding space.

Distributed Parametric Amplifier in Standard Single-Mode Fibre with Gain up to 44 dB and bandwidth up to 30 nm in O-band
Munira Binte Mostafa, Aston University
Fibre optical parametric amplification (FOPA) utilises the third-order nonlinearity of the fibre to amplify signals via four-wave mixing. Parametric amplification can be achieved in any optical fiber having a nonlinear response, including standard single-mode fibers(SSMF) employed for signal transmission. This implies that distributed FOPA is also possible however, this area has been little studied and thus merits further investigation. Distributed FOPA has been challenging in the O-band due to the lack of high-power, low-noise pumps. For the first time, we have demonstrated distributed parametric amplification over 25 km of SSMF, with on/off gain of 44 dB over narrow bandwidth and ~20 dB over 30 nm in the O-band. We further demonstrate the detection of a single 50 Gb/s PAM-4 channel at ~80 mW pump power, with an optimal on/off gain of ~14 dB. These results demonstrate the simultaneous transmission and amplification of the signal with greater efficiency than Raman amplifiers.
Efficient Optical Image Transmission: A Deep Learning Approach
Xiaomin Cai , PhD Student, University of Warwick
This project implemented an end-to-end optical fibre communication system enhanced by machine learning. I developed a Convolutional Neural Network (CNN)-based autoencoder to handle dimensionality reduction, compressing raw image data into low-dimensional latent features. To mimic realistic transmission conditions, I employed an Enhanced Gaussian Model to simulate the optical fibre channel. The compressed features were transmitted through this simulated link and accurately recovered by the decoder at the receiver. The results validated that integrating deep learning with optical communications significantly reduces bandwidth usage while maintaining high image reconstruction quality.
Can we make a hop-by-hop architecture viable for BT’s core networking using 400ZR?
Danny Capp, Research professional, Research and Commercialisation (RaC), BT
400ZR/ZR+ optical pluggable modules are being implemented within the BT core network to reduce cost, power-consumption, and footprint of traditional transponders. Hop-by-hop 400ZR/ZR+-based networks can exploit a reconfigurable operation at the IP level without the limitations of the wavelength continuity constraint at the optical layer. This would allow for full dynamic adaptation in response to traffic demands with potential savings in network capacity. However, current cost and reach limits of 400ZR/ZR+ modules hamper their deployment in the core. In this work, we investigate the conditions needed for a hop-by-hop 400ZR/ZR+-based core network to become cost-effective. Numerical results over a 10 year horizon indicate that, assuming 12% annual traffic growth, a hop by hop 400ZR/ZR+ based BT core network becomes more cost effective than a ROADM based alternative only if the reach of 400ZR/ZR+ optics increases threefold and their cost decreases by the same factor compared to current values.
Asymmetrically Intelligent PON with Simplified ONU Design for 100Gb/s in C-band
Safiya Dabwan, Aston University
Asymmetric learning is used in point-to-multipoint PON systems to centralize digital signal processing (DSP) at the Optical Line Terminal (OLT), which reduces the complexity of the Optical Network Units (ONUs). A multi-task learning approach is introduced to optimize training, enhancing the OLT’s ability to handle various tasks simultaneously. This strategy improves overall performance, effectively doubling the OLT’s service coverage range. By centralizing processing and utilizing advanced learning techniques, the system achieves greater efficiency, providing broader and more reliable coverage while keeping the complexity at the ONU end to a minimum.
End-to-End Throughput Maximisation for SCLO Ultra-Wideband Optical Networks
Bohua Deng, PhD student, Optical Networks Group, UCL 
The rapid expansion of artificial intelligence and cloud services is driving a surge in data traffic, pushing standard transmission windows towards their capacity limits. Ultra wideband operation across the S+C+L+O spectrum upgrades scale capacity by exploiting the low loss spectral windows of already deployed single mode fibre, offering a more infrastructure efficient alternative to rolling out parallel fibres or migrating to spatial division multiplexing solutions like multicore fibre. It provides a promising route to scale capacity in current core optical networks, but requires tight coupling between network optimisation and accurate physical layer constraints. This poster presents a comprehensive framework that reproduces and extends a doctoral baseline model for throughput maximisation. We introduce a structured configuration module for multiple bands, defining wavelength grids and specific noise or gain profiles, and report initial comparative results between single band and SCLO operation. The study addresses questions relevant to industry such as trade-offs between capacity and complexity, upgrade planning, and the sensitivity of throughput estimates to physical assumptions. This enables reliable design exploration for practical deployment decisions and supports future data driven optimization strategies. 
Performance and Cost Analysis of Prepare and Measure and Entanglement Based QKD protocols
Damian Hedges, University College London
The quantum threat has accelerated the development of Quantum Key Distribution (QKD). However, scaling from point-to-point links into practical networks raises questions about the relative suitability of Prepare and Measure (PM) and Entanglement-Based (EB) QKD implementations. In this work, the BB84 and E91 protocols are simulated across multiple network topologies. Performance is evaluated in terms of the maximum secure distance and raw key exchange rate. A cost model is also included to account for the scaling of quantum hardware and trusted node requirements. Results show that for point-to-point links, EB-QKD achieves a higher raw key exchange rates and extends the maximum communication distance; however, PM-QKD is generally the more cost-effective option. Nonetheless, in a networked scenario, EB-QKD becomes the more cost-effective option if the entangled photon source (EPS) is placed as the central hub in a star topology.
Agentic AI – Intelligent Network Planning Agents
Oluwole Ilesanmi, Software Engineer, Ex Accenture
Intelligent Network Planning Agents represent a transformative application of Agentic artificial intelligence in optical network management. These autonomous systems function as continuous planning assistants that monitor network performance metrics and align them with business objectives in real-time. Unlike conventional network management tools that require human intervention, these agents proactively identify capacity bottlenecks and performance degradations before they impact service delivery. The agents employ sophisticated simulation capabilities to evaluate multiple network expansion scenarios, analysing trade-offs between cost, performance, and implementation complexity. Through seamless integration with enterprise inventory management systems, they verify equipment availability and compatibility for proposed solutions. Most notably, these agents can autonomously generate procurement requests when infrastructure upgrades are necessary, streamlining the traditionally manual process of network capacity planning. This Agentic approach enables optical networks to evolve from reactive maintenance models to predictive, self-optimising infrastructures that continuously adapt to changing traffic demands and business requirements, significantly reducing planning cycles and operational costs.
Deuterated Silicon-rich Silicon Nitride Waveguides for Broadband All-Optical Wavelength Conversion
Polyxeni Kolka, PhD student, University of Southampton
We design, fabricate, and characterise deuterated silicon-rich silicon nitride (SRSN:D) waveguides that enable broadband, all-optical wavelength conversion for advanced telecommunication systems. By substituting hydrogen with deuterium during low-temperature PECVD, N–H absorption is strongly suppressed, yielding low C-band propagation losses of 1.4–1.9 dB/cm while preserving a high Kerr nonlinearity. Based on these characteristics, we design low-dispersion waveguides that are essential for efficient and broadband four-wave mixing. We demonstrate tuneable wavelength conversion across the S-, C-, and L-bands on this platform, achieving 3-dB conversion bandwidths exceeding 60 nm, and highlighting the strong potential of SRSN:D for compact and broadband parametric photonic devices.
Analog Radio-Over-Fiber Transceivers For 5G mmWave
Krishna Kumar, Avirata Defence Systems Ltd
Authors: Manas Srivatsava, Pradnya Kalkuldrikar, Krishna Kumar
We present the design, development and performance characterization of millimeter wave (mmWave) analog radio over fiber (ARoF) transceivers operating in the N258 band (24.25 GHz – 27.5 GHz) of the 5G NR spectrum, to be deployed in the fiber-based fronthaul link in 5G networks. The transceivers are realized in a 2T2R configuration to realize 2-channel DWDM full duplex link, and are developed as optical line terminal (OLT) and optical network unit (ONU), to be deployed at the central office and the radio site respectively. The OLT and ONU are designed to be compliant to the ITU-T G.9803 standard and 3GPP rel.17 recommendations. We have demonstrated the integration of our OLT and ONU with a multicore fiber link. We present the complete analog performance analysis of the ARoF system in the N258 band and the error vector magnitude (EVM) performance of the 2 km ARoF fronthaul link with 5G NR signal formats.
Compact Entanglement Sources for Satellite-Mediated Quantum Networking
Sarah McCarthy, PhD Student, Fraunhofer Centre for Applied Photonics
In this poster I will present the two strands of work within my PhD research. The first is the development of polarisation-entangled photon pair sources that have been designed for deployment on SWaP platforms in challenging environments, such as on small satellites. I have developed a compact and stable source architecture that is a folded variation of the linear displacement interferometers also used for Bell state generation, which ensures tolerance against misalignment and gives very high brightnesses when compared to similar implementations. I have used the properties of the sources developed to then inform a comprehensive model of dual-downlink entanglement-based quantum key distribution. A generalised approach to define our ground station locations and orbit allows us to model the achievable secret key lengths, considering finite key effects, for any geometry. I will present results on the achievable annual secret key length, the effect of background light and range limitations. 
Ultra-Low Loss (ULL) Interconnects – Supporting advancements in Quantum Photonics
Guy Nurse, Diamond SA
Quantum photonics has emerged as a rapidly advancing field with significant potential in quantum computing, cryptography, secure communication, and precision sensing. As the demand for efficient and reliable transmission of quantum information over fiber-based networks increases, the performance of optical interconnects becomes increasingly critical. A review is presented here on recent developments at DIAMOND in a new class of Ultra-Low Loss (ULL) fiber connectors that reliably achieve coupling losses well below 0.1 dB. The connector architecture and the new manufacturing approach enable both systematic loss reduction and exceptional stability, supporting the stringent performance requirements of modern quantum photonic systems.
Automated packaging of microlens arrays with integrated turning mirrors to PICs for connectorized fiber optic pluggability applications, From Lab to Fab
Josue Parra Cetina, Senior Research and Development Engineer, ficonTEC Ireland
The integration of micro-optical components in photonics integrated packages is a crucial enabler for co-packaged optics, allowing the advancement of technologies across numerous applications such as high-speed data communications in data centers, AI, high-performance computing, lidar, amongst others. The performance of these systems depends strongly on precise alignment and integration of the micro-optical components which can be done with a combination of passive and active alignment techniques. Automated photonic assembly systems, such as those developed by ficonTEC, are critical for achieving the accuracy and scalability required to develop efficient devices. By streamlining microlens placement and bonding, ficonTEC’s solutions support the reliable, high-throughput manufacturing of next-generation Telecom devices. ficonTEC will present an assembly example of a microlens-array featuring an integrated turning mirror to achieve out-of-plane optical coupling to a PIC on a glass interposer. The approach also supports direct applications in connector pluggability. This assembly was developed within an EU-funded project.
Triplets-based Viterbi Algorithm in 16-QAM Optical Transmission
Sina Seyfadini Kuhbanani PhD Student Aston University
We propose a symbol-triplet-based Viterbi decoder for 16-QAM optical communication that models channel memory induced by dispersion and nonlinearity. We improve the detection by including memory information across three adjacent symbols we term triplets. Each Viterbi state corresponds to a symbol-triplet, and the observation likelihoods are modelled by fitting 2D Gaussians to the data generated for every possible triplet, given the symbols before and after it. the data was generated by a non-linear Schrödinger equation (NLSE) split-step emulator: single-polarization 16-QAM at 34.4 GBd over fifteen 80-km SMF spans with EDFAs (4.5 dB noise figure). Compared with hard-decision detection, the proposed probabilistic decoder reduces BER by about 19% across the tested power range. Complexity scales linearly with sequence length and the measured CPU/GPU runtimes grows quadratically/linearly with the number of processed symbols, respectively. The approach is compatible with GPU execution and can be extended to higher-order QAM by adjusting state design.
Kerr Microring Resonator Frequency Combs for Optical Reservoir Computing
Negar Shaabani Shishavan, Egor Manuylovich, Morteza Kamalian-Kopae, Auro M. Perego, Aston Institute of Photonic Technologies, Aston University, Birmingham, UK
This work presents a scalable photonic platform based on Kerr microring resonator frequency combs for implementing optical reservoir computing. The system enables GHz-speed, low-power processing demonstrated through theoretical time-series forecasting and benchmark tasks such as MNIST and Iris flower classification. Our proposed frequency-comb-based reservoir computer could also be applied for telecom signal equalisation tasks and offers a promising solution towards integrated photonic AI accelerators for next-generation communication and data-centre systems.
Characterisation of 400nm Ultra-Wide Band Lithium Niobate Dual-Polarisation Modulator
Shekhar, Aston University
In the current era, data-intensive applications have become widespread, causing rapid growth in data traffic. Modern single-mode optical fibre can support a bandwidth of approximately 400 nm from the O-band to the U-band, enabling the fulfilment of rising data demands. Nevertheless, capacity challenges remain and can be addressed through various methods, including considering other bands.
One of the primary challenges in simultaneously exploring all bands is the unavailability of a single ultra-wideband modulator. Earlier research demonstrated ultra-wideband operation using a dual-polarisation IQ modulator for the SCLU bands and a polarisation emulator for the O/E band, resulting in a complex transmission setup.
This paper introduces a set of measurements and characterises a commercially available modulator, focusing on key parameters including modulation-induced loss, insertion loss and extinction ratio. These measurements enable a rapid evaluation of a modulator’s suitability for multi-band transmission in all available optical bands.
Ultra-wideband Optical Communications
Dr Mingming (Tommy) Tan, Research Fellow/Royal Society Short Industry Fellow, Aston Institute of Photonic Technologies (AiPT)
We demonstrate several optical amplifier technologies for ultra-wideband optical transmission beyond the conventional C band, which extends the lifespan of billions of kilometres of existing optical fibre.
Measurement-Driven Digital Twin with AI-based Data Extrapolation for Deep Reinforcement Learning Optical Networks
Yiran Teng, PhD student, University of Bristol
We propose a digital twin (DT) construction framework that integrates sampled laboratory emulation, AI-based data extrapolation, and interactive DT formation. According to a target network, links with similar transmission characteristics are grouped for lab emulations using limited testing facilities and purposefully sampled measurements. Then, artificial neural network (ANN) models are trained using partially collected measurement data to augment the sample space. An interactive DT is then established to emulate end-to-end Q-factors for all routing paths under different wavelength configurations in the target network, enabling the training and validation of a deep reinforcement learning (DRL) agent. Experimental results show that the proposed DT achieves an average Q-factor estimation error of only 0.05 dB compared with real measurements, while the DRL agent improves the Q-factor of each service by approximately 0.4 dB relative to a heuristic approach, after only 1 hour of training. The proposed approach could reduce the overall required samples. Moreover, compared to online DRL training, the proposed approach significantly reduces training time while maintaining agent performance.
Pulse-Based Optical ISAC Using PSS-PPM for Joint Communication and Ranging
Fengyuan Tian, PhD student, University of Warwick
Integrated Sensing and Communication (ISAC) enables communication and environmental sensing to be performed using shared physical resources. While most existing ISAC implementations focus on radio-frequency platforms, optical systems offer compelling advantages in terms of bandwidth, spatial resolution, and robustness to electromagnetic interference. This poster demonstrates how sensing functionality can be embedded into optical communication signals through signal-processing-level reuse, without introducing dedicated sensing waveforms. As a representative example, we present a Gaussian pulse-based Pulse Sequence Sensing and Pulse Position Modulation (PSS-PPM) optical ISAC system, where communication and ranging are jointly supported within a unified signal structure. Through simulation under atmospheric turbulence, we show that appropriate pulse shaping combined with correlation-based processing enables high-precision distance estimation while maintaining low bit-error rates. The proposed framework remains effective in both single-user and multi-user scenarios, highlighting its potential for practical optical wireless and sensing-enabled communication applications.
AI-Driven Network Optimisation: Practical Barriers, Workforce Skills, and Product Innovation Gaps
Ruky Toje, AI Product Manager, ProteusAI
As AI-driven optimisation becomes essential to the performance and resilience of next-generation optical networks, organisations face a growing challenge: the technology is advancing faster than the workforce, operational processes, and product ecosystems required to deploy it. This presentation explores the human, organisational, and product-innovation barriers that hinder effective AI adoption in telecom and optical environments. Drawing from cross-functional experience in AI product management, digital innovation, a degree in innovation technology management, and technical recruitment, the session highlights the skills gap in network automation roles, the lack of AI-readiness within traditional engineering teams, and the product-development bottlenecks that delay deployment. It also outlines strategies for building a future-proof workforce pipeline, improving cross-team collaboration, and integrating AI tools that enhance network performance without disrupting critical infrastructure. Attendees will leave with actionable insights to accelerate AI adoption while strengthening organisational capability.
Reconfigurable Optical Transfer Functions via Four-Wave Mixing for Photonic AI Acceleration
Wanqi Yang, PhD student ,Optoelectronics Research Centre, University of Southampton
Optical power transfer functions enable quantisation, pulse reshaping and nonlinear limiting, and can also implement nonlinear activations for optical neural networks. In a proof-of-principle experiment, we demonstrate a reconfigurable transfer-function generator based on degenerate four-wave mixing (FWM) in cascaded highly nonlinear fibres. By exploiting the intrinsic nonlinear input–output characteristics of multiple FWM products, and linearly weighting and combining them using a programmable wavelength-selective switch, we approximate ReLU, Softplus, and Sigmoid-like transfer functions after photodetection and power summation, and achieve NMSEs of 4.1×10−3, 7.2×10−5 and 9.0×10−4 respectively. The measured transfer functions are deployed as lookup-table activations in a MNIST MLP, reaching >97% test accuracy, within 0.45 percentage points of the digital baseline and with a best-case gap of 0.14 percentage points. These results suggest Kerr nonlinearity could offer a promising route to ultrafast, low-latency activation operations for photonic AI accelerators.
Acceleration of Coupled-Mode Theory Computations for Estimation of Inter-Core Crosstalk in Multicore Fibres
Brandon Yee, PhD student, Optical Networks Group at UCL
Multicore fibres (MCFs) have arisen as a promising candidate for implementing spatial division multiplexing in next-generation optical networks. Transceivers will need to implement dynamic equalisation to compensate for inter-core crosstalk (ICXT), with applicable models being required to develop specialised digital processing techniques. In this work, we examine Coupled Mode Theory (CMT), which describes signal propagation in bent and twisted MCFs. While the relevant nonlinear Schr?dinger equation (NLSE) can be solved using the Split-Step Fourier Method, full accuracy requires step sizes several orders of magnitude smaller than is needed in an equivalent single-mode fibre. Based on phase-matching conditions present in CMT, we present a method for solving the NLSE that selectively applies the mode coupling term, displaying good accuracy using a drastically smaller number of steps. This method allows for faster computation of signal propagation that can be readily applied in conjunction with other propagation effects, particularly nonlinearities. 
Intelligent Resource Allocation via Hybrid Reinforcement Learning in 5G Network Slicing
Zahraa Zakariya Saleh, Doctoral Researcher, Brunel University London
Manufacturers are prioritising reconfigurable and resilient environments for Industry 5.0, leveraging applications like digital twins and mobile robots that demand advanced communication networks. Beyond 5G (B5G) networks offer enhanced performance and flexibility through virtualisation and network slicing, crucial for meeting application-specific requirements. This paper introduces a framework for optimising resource allocation in 5G network slicing by integrating Double Deep Q-Network with Prioritised Experience Replay (DDQN-PER) and Pointer Network-based Long Short-Term Memory (PtrNet-LSTM). The framework adjusts the attention coefficient to balance Service Satisfaction Level (SSL) and Quality of Experience (QoE), thereby enhancing system and spectrum efficiency and user connectivity. Experimental results show that the PtrNet-LSTM framework, combined with DDQN-PER, outperforms baseline methods in spectrum efficiency and user connectivity, highlighting its scalability for addressing dynamic wireless network challenges.
Terahertz quantum communication with superconducting circuits
Mingqi Zhang, University of Glasgow
Terahertz (THz) quantum communication is attractive as a platform for ultrafast, high-bandwidth wireless links between quantum nodes. Superconducting THz quantum technology presents a transformative platform for scalable, high-speed quantum networks, exploiting the ultrafast dynamics, coherence, and low dissipation of superconducting circuits. We are developing a new generation of superconducting THz devices, including coherent sources, ultrafast modulators, and broadband detectors, engineered for operation in free-space environments ranging from cryogenic to room temperature. These components are optimised for high-frequency performance with minimal loss, available for photon-level control with easy integration from individual elements to chip-scale integrated systems. Our objective is to establish practical THz quantum links for both on-chip interconnects and short-range distributed quantum networks. This presentation highlights our THz quantum key distribution protocols, device design, fabrication, and experimental validation in hybrid cryogenic-open-air systems for a broad range of applications, including next-generation communication, quantum wireless networks, and space applications.References
[1] M. Zhang, S. Nakagawa, Y. Enomoto, Y. Kuzumi, R. Kikuchi, Y. Yamauchi, T. Hattori, R. A.
Klemm, K. Kadowaki, T. Kashiwagi, and K. Delfanazari, “Tunable terahertz source on a chip with decade-long stability using layered-superconductor elliptical microcavities,” Physical Review Applied, vol. 24, no. 5, p. 054012, 2025.
[2] Zhang, Mingqi, and Kaveh Delfanazari. “Orthogonal Frequency Division Multiplexing
Continuous Variable Terahertz Quantum Key Distribution.” arXiv preprint arXiv:2506.22985 (2025).
[3] M. Zhang, S. Pirandola and K. Delfanazari, “Millimeter-Waves to Terahertz SISO and MIMO
Continuous Variable Quantum Key Distribution,” in IEEE Transactions on Quantum Engineering, vol. 4, pp. 1-10, 2023.
[4] Delfanazari, K. Chip-scale electrically driven superconducting coherent photon sources for
quantum information processing. Nat. Photon. 19, 1163–1177 (2025).

 

 

 

 

 

Best Poster Award 2026

Once again there will be an award for the Best Poster, as reviewed by the TOP Committee. The award will be presented after the panel session on Wednesday 25 February.