arXiv subjects
Publications and source records attributed to Frank Slyne.
We implement a ML-based attention framework with component-specific decoders, improving optical power spectrum prediction in multi-span networks. By reducing the need for in-depth training on each component, the framework can be scaled to multi-span topologies with minimal data collection, making it suitable for brown-field scenarios.
We introduce a ML-based architecture for network operators to detect impairments from specific OSaaS users while blind to the users' internal spectrum details. Experimental studies with three OSaaS users demonstrate the model's capability to accurately classify the source of impairments, achieving classification accuracy of 94.2%.
We demonstrate a real-time, converged 5G-PON through the Cooperative Transport Interface, synchronising 5G and PON-DBA upstream schedulers. This innovative approach, implemented using 5G and PON open network implementations, significantly enhances network resource allocation, reducing latency.
Efficient use of spectral resources will be an important aspect of converged access network deployment. This work analyzes the performance of variable bandwidth Analog Radio-over-Fiber signals transmitted in the unfilled spectral spaces of telecom-grade ROADM channels dedicated for coherent signals transmission over the OpenIreland testbed.
We present a novel ML framework for modeling the wavelength-dependent gain of multiple EDFAs, based on semi-supervised, self-normalizing neural networks, enabling one-shot transfer learning. Our experiments on 22 EDFAs in Open Ireland and COSMOS testbeds show high-accuracy transfer-learning even when operated across different amplifier types.
Ultra-low end-to-end latency is one of the most important requirements in 5G networks and beyond to support latency-critical applications. Cloud-RAN and MEC are considered as the key driving technology that can help reduce end-to-end latency. However, the use of MEC nodes poses radical changes to the access network architecture. As it brings the processing and the networking services closer to the edge, it often requires network functions (for example, the CU/DU stack and the application processing) to be distributed across different MEC sites. Therefore, a novel transport mechanism is needed to efficiently coordinate and connect network functions across MEC nodes. In order to address this challenge, we propose a novel two-tier virtualized PON transport method with schedulers coordination over a virtualised and sliced MESH-PON architecture. While a MESH-PON architecture enables direct communication between MEC nodes that are hosting CU/DU and/or the application processing, our method provides a two tier virtualised PON transport scheme with coordinated schedulers. This approach greatly reduces latency incurred in transporting signals across the different PON tiers, while maintaining the flexibility of the multi-tier methods. We show that our proposed scheme can achieve end-to-end application-level latency below 1ms or 2ms, depending on the network configurations.
SNR margins between partially and fully loaded DWDM systems are estimated without detailed knowledge of the network. The ML model, trained on simulation data, achieves accurate predictions on experimental data with an RMSE of 0.16 dB.
In this work we analyse and demonstrate the coexistence of digital coherent and analogue radio over fibre signals over an access-metro transmission network and field fibre. We analyse how the spectral proximity of the two signals and the non-ideal filter alignment of typical telecomms-grade ROADMs affect the signal performance. Our results show that coexistence is indeed possible, although performance deteriorates with the increase in number of ROADMs in the network topology. Thus, while todays access-metro networks will be able to support future 5.5 and 6G cell densification operating at mmWave and THz frequency, using spectral efficient analogue radio over fibre transmission, there will be trade-offs to be considered. In our experiment setup, we show that the limit for ARoF accessible performance is reached after transmission over 3 ROADMs and a total of 49 km of fibre.
In this paper we propose a hybrid softwarized architecture of Network Function Virtualization (NFV) where Software-Defined Networking (SDN) and Software-Defined Radio (SDR) components are integrated to form a cloud-based communication system. We analyze CPU utilization and power consumption in the OpenIreland testbed for different parameter settings and use case scenarios of this NFV architecture. The experiment results show different behaviour between SDN data plane switching and SDR in terms of CPU utilization and parallelization, which provides insights for processing aggregation and power savings when integrating them together in a cloud-based system. We then propose a power saving scheme with flexible CPU allocation that can reduce the overall power consumption of the system. Our results show that our proposed NFV architecture and its power saving scheme can save up to 20% power consumption compared to conventional scheme where SDN and SDR are separately deployed.
We present a heuristic algorithm for a PON upstream scheduling hypervisor,supporting low latency services with strict service-level agreement. The algorithm achieves near-optimal performance while running in only 3.5 us, thus operating in real-time.
We propose a mechanism to support URLLC Open-RAN ultra-low latency over a MESH-PON, serving dense deployment of small cell and MEC nodes in an access network. We show the possibility, under given assumptions, to achieve application-to-application end-to-end latency below 1ms.
The commercialization of Cloud-RAN, and Open-RAN in particular, is a key factor to enable 5G cell densification, by providing lower cost and more agile deployment of small cells. In addition, the adoption of MEC is important to support ultra-low latency and high reliability required by mission-critical applications, which constitute a milestone of the 5G and beyond vision of a fully connected society. However, connecting antenna site, C-RAN processing and MEC at low cost is challenging, as it requires high-capacity, low latency connectivity delivered through a highly inter-connected topology. While PON is being considered as a solution for providing low-cost connectivity to C-RAN, they only allow data transmission from the endpoints (for example hosting RU at the antenna site) towards a central node (e.g., the central office, hosting computing equipment), thus cannot support traffic from RU towards MEC end nodes that could host DU and possibly CU and network core. This led to research into the evolution of PON architectures with the ability to provide direct communications between endpoints, thus supporting mesh traffic patterns required by MEC installations. In this context, virtualization plays a key role in enabling efficient resource allocation (i.e. optical transmission capacity) to endpoints, according to their communication patterns. In this article, we address the challenge of dynamic allocation of virtual PON slices over mesh-PON architectures to support C-RAN and MEC nodes. We make use of a mixed analytical-iterative model to compute optimal virtual PON slice allocation, with the objective of minimizing the use of MEC node resources, while meeting a target latency threshold (100 $μs$ in our scenario). Our method is particularly effective in reducing computation time, enabling virtual PON slice allocation in timescales compatible with real-time or near real-time operations.
Cloud-Radio Access Networks (Cloud-RANs) are separating the mobile networks base station functions into three units, the connection between the two of them is referred to as the fronthaul network. This work demonstrates the transmission of user data transport blocks between the distributed Medium Access Control (MAC) layer and local Physical (PHY) layer in the radiounit over a Passive Optical Network (PON). PON networks provide benefits in terms of economy and flexibility when used for Cloud-RAN fronthaul transport. However, the PON upstream scheduling can introduce additional latency that might not satisfy the requirements imposed by Cloud-RAN functional split. In this work we demonstrate how our virtual Dynamic Bandwidth Allocation(DBA) concept can be used to effectively communicate with the mobile Long Term Evolution (LTE) scheduler, adopting the well known cooperative DBA mechanism, to reduce the PON latency to satisfactory values. Thus, our results show the feasibility ofusing PON technology as transport medium of the fronthaul for the MAC/PHY functional split, in a fully virtualised environment.Further background traffic is added, so that measurements show a more realistic scenario. The obtained round trip times indicates that using PON at fronthaul might be limited to the distance of 11km for a synchronised scenario, or no compliance for a non-synchronised scenario.