Flávio Geraldo Coelho Rocha

5G networks impose stringent requirements, such as low latency and high throughput, that challenge traditional traffic classification mechanisms within the UPF. These approaches often become bottlenecks due to the overhead and latency of user-space processing. To address the challenges, this work proposes a methodology for real-time traffic classification using In-Kernel Machine Learning with eBPF/XDP. The implementation employs C headers and leverages native eBPF mechanisms to manage verifier constraints. Experimental results demonstrate that In-Kernel ML significantly outperforms user space execution, achieving over 36× faster inference speed, reduced CPU usage from 15.79% to 12.87%, and maintaining 99.91% of accuracy.

Este trabalho propõe um arcabouço estocástico para o planejamento de fatias URLLC em redes B5G, fundamentado no Cálculo de Rede Estocástico (SNC). A partir da modelagem probabilística dos processos de chegada e de serviço, é derivada uma expressão analítica em forma fechada que relaciona requisitos de latência, probabilidade de violação e, o mais importante, número máximo de usuários admissíveis por célula e fatia. O modelo permite dimensionar recursos de rádio com garantias probabilísticas de QoS, evitando o superdimensionamento de abordagens determinísticas. Resultados numéricos evidenciam ganhos significativos de capacidade quando comparados aos resultados obtidos utilizando Cálculo de Rede Determinístico (DNC). O arcabouço proposto fornece subsídios relevantes para o planejamento eficiente de fatias URLLC em ambientes multicelulares heterogêneos.

Network Slice as a Service (NSaaS) is a key enabler of Beyond Fifth Generation (5G) and Sixth Generation (6G) networks, supporting next-generation applications such as extended reality (XR), immersive services, and the tactile Internet. These networks must provide native support for slice-aware services across the entire Radio Access Network (RAN), including the Radio Unit (RU), Distributed Unit (DU), Central Unit (CU), and transport segments (fronthaul, midhaul, and backhaul). However, uplink slicing identification in shared Open-RUs (O-RUs) presents a fundamental challenge because the Open-DU (O-DU) handles scheduling, and the O-RU does not inherently know which uplink data belongs to which slice. In MultiPoint-to-MultiPoint (MP2MP) fronthaul scenarios, this limitation is further exacerbated by synchronization and timing constraints, which necessitate that the O-RU process control messages and the encapsulated data be delivered with ultra-low latency. To address this issue, we propose a slicing agent embedded in the O-RU that identifies slices and segregates uplink data into slice-specific enhanced Common Public Radio Interface (eCPRI) packets. Our design employs a pipeline architecture with dedicated paths for time-sensitive, flexible slicing, enabling slice isolation and prioritization. When implemented on a Field-Programmable Gate Array (FPGA), the agent processes each packet in 2 clock cycles, supporting up to 3822 slices per slot. Experimental results validate the approach, showing its feasibility, scalability, and high-performance capabilities for real-time, slice-aware uplink processing in Beyond 5G and 6G Open RAN deployments.

In this article, we propose a novel formulation that jointly considers the Virtualized Network Function (VNF) placement at the Radio Access Network (RAN) nodes and the resource allocation for the Transport Network (TN) of sliced and disaggregated RANs. Unlike most works in the literature that address these network aspects separately, we propose a joint approach. Our proposal ensures an end-to-end delay bound for the Ultra-Reliable and Low-Latency Communications (URLLC) use case in an Industry 4.0 scenario, while simultaneously considering the number of admitted User Equipments (UEs), the transmission rate allocation per slice, the functional split of RAN nodes, and the routing paths in the TN. We use Network Calculus (NC) theory to calculate delay along the TN connecting disaggregated RANs deploying network functions at the Radio Unit (RU), Distributed Unit (DU), and Central Unit (CU) nodes. The maximum end-to-end delay is modeled as a constraint in the optimization-based formulation that maximizes the number of admissible flows meeting strict Quality of Service (QoS) requirements, while also taking into account the deployment and operational costs of disaggregated RANs. In this approach, we propose a strategy based on real data from one of the world’s leading Mobile Network Operators (MNOs) to derive coherent estimations to the weights of the proposed objective function. The optimization model leverages a Flexible Functional Split (FFS) approach to provide a new degree of freedom to the resource allocation strategy. Simulation results reveal that, due to its non-linear nature, there is no trivial solution to the proposed optimization problem. Simulation results also show that our proposal guarantees a maximum delay for URLLC use cases in Industry 4.0 while satisfying bandwidth requirements for enhanced Mobile Broadband (eMBB) services.

Deployment of low‐cost Wi‐Fi sensing applications may impose strict constraints on available computational power, thereby making the reduction of time required to process channel state information (CSI) measurements a matter of interest. Most works on passive sensing for classification tasks achieve this via dimensionality reduction of CSI data prior to the training of machine learning models, which in itself still imposes some computational burden. Subcarrier selection, which is a faster approach and is widely used in another application domain (i.e., vital signal monitoring), is seldom considered; in the few works where it is used, only a variance‐based unsupervised strategy is applied. In this letter, a supervised linear‐complexity subcarrier selection strategy is proposed for enhanced sensing classification accuracy. The approach is validated through practical experiments whose results show that the classification performance can approach or even surpass that obtained via dimensionality reduction, with substantial time savings.

Resumo-Este trabalho propõe uma abordagem com Deep Reinforcement Learning (DRL) para otimizar a alocação de recursos para nós inteligentes em redes Open RAN. O foco está na seleção eficiente de nós near-RT-RICs considerando tipos distintos de fatias da rede. O agente DDPG decide quais nós utilizar, como alocar recursos computacionais para o processamento das tarefas nos near-RT

Resumo-Este artigo investiga o desafio conjunto de admissão e associação de usuários, bem como o posicionamento otimizado de unidades funcionais na arquitetura Open RAN com suporte ao fatiamento de rede. Trata-se de um modelo de otimização formulado como um problema de Programação Linear Inteira, que incorpora restrições de Qualidade de Serviço e limitações de recursos computacionais. A escalabilidade do modelo é avaliada por meio de técnicas que combinam abordagens de solução exata e heurísticas aproximadas

Resumo-O monitoramento de sinais vitais é tradicionalmente realizado por meio de sensores corporais ou dispositivos vestíveis, o que pode ser incômodo e resultar em baixa adesão, especialmente entre pessoas que necessitam de acompanhamento contínuo, como idosos e pessoas com doenças crônicas. Além disso, as soluções atuais costumam exigir

Resumo-As redes Open RAN, com sua proposta de desagregação e interfaces abertas, são fundamentais para viabilizar a inovação e a interoperabilidade nas gerações pós-5G. Dentre suas interfaces, o fronthaul (O-FH) se destaca por sua sensibilidade à latência, sendo crítico para o desempenho global da rede. No entanto, há uma lacuna na literatura quanto a avaliações experimentais dessa interface em cenários realistas