Research
Over twenty years chasing the same question: how to distribute resources and sustain mobility when the network itself has to make the decision.
Mission
Putting the decision inside the network: allocating resources, choosing routes, and performing handover through AI-assisted decision-making and AI-native intelligence embedded in network operations.
An ordinary controller executes what it is told. An intelligent agent observes, decides and acts, and the result comes back as the next observation.
It is this loop that separates executing orders from orchestrating the network autonomously: slice, route and handover decided at every layer of the 5G and beyond (NextG) stack.
We get there through analytical modeling, optimization and machine learning, evaluated in simulation and on testbeds.
Areas
The network is a stack, and we work at every layer of it. Each one has its own deadlines, its own scarce resources and its own decision to make.
Radio
Open interfaces and xApps running on the near-real-time RIC, with machine learning deciding handover, energy consumption and spectrum use. This line of work goes back a long way: choosing when to sense the spectrum in cognitive radio, compensating on the device for a channel quality indicator that arrives late, predicting trajectory and signal to switch base stations before the drop.
Transport
The path between the antenna and the core. Critical traffic crosses several routers between the user and the server, and that is where the latency is lost, even with the edge right there. Segment Routing defines the path, and a bandit model chooses which one, minimizing loss and latency at the same time.
Core
The core treated as a system that learns: the NWDAF observing the data and control planes, agents that plan and carry out actions on the network functions, and opportunistic slices using what the primary ones leave free. Defense comes with it, because a neighboring slice is a new threat, detected in the programmable data plane.
Edge
Processing close to the user. The decision to offload a task is made by a classifier, using the context of the device and of the network, and the environment mixes virtual machines and containers to meet a hard deadline without losing elasticity. When the edge node is a drone, weight and battery enter the equation.
Application
Services that only exist if the network delivers. The Internet of Vehicles brings software-defined networking, edge and virtualization together in a single architecture. Computer vision moves off the device and onto the edge. The smart city runs into the problem of making different wireless technologies talk to each other.
Trajectory
Each generation of mobile network brought a different scarce resource to manage. The sequence matters: it is what explains how the group arrived at the agent.
2003 · 2007
Admission control and QoS in IP mobile networks
Time series prediction to admit a call without breaking the guarantees of those already on the network.
2008 · 2012
Handover in heterogeneous networks
WiMAX, WLAN and cellular in the same session, with mobile MPLS and PMIPv6. This is where quality of experience enters as a decision criterion, and not throughput alone.
2013 · 2016
Cognitive radio and the spectrum as a scarce resource
Sensing with multi-objective genetic optimization and virtual network mapping over a cognitive substrate.
2017 · 2020
The network becomes software
SDN, NFV and the edge: offloading decided by machine learning, integrated NFV/SDN architectures and real-time 5G slicing defense.
2021 · 2023
5G in operation: MEC, URLLC and vehicles
Resource allocation for critical traffic in hybrid MEC environments, including a drone-mounted base station, and the Internet of Vehicles as the use case that stresses everything at once.
2024 · 2026
Open RAN and AI-orchestrated 6G core
xApps that decide mobility, multi-agent reinforcement learning choosing whether a vehicle uses sidelink or the base station, Segment Routing optimized by multi-armed bandits, and AI agents operating the mobile core.
Projects
2025 · present
Three fronts: open interfaces and rApps/xApps to manage mobility and QoS, Segment Routing in the transport network, and an AI-orchestrated 5G/6G core with NWDAF evolution and the use of agentic AI.
2024 · present
CNPq call 09/2023, research productivity fellowship. Handover and resource management for the Internet of Vehicles and of drones, in software-defined networks with Open RAN support.
2022 · present
Project with the Brazilian Ministry of Development, Industry, Trade and Services, CIn/UFPE and INES, for research and innovation in 5G software solutions aimed at the national ecosystem.
2019 · present
A flexible, low-cost 5G testbed, with software-defined radios, commodity servers and antennas, built on OpenAirInterface, Open5GS, srsRAN and Magma.
Method
First understand why, then show that it works
Admissions
Topics in 5G and beyond (NextG) at the Graduate Program in Computer Science at CIn-UFPE, with scholarships through CNPq, CAPES and FACEPE.
See the open topics