Francisco Rau

Incremental Learning in Network Traffic Management: A Fixed-Representation Rehearsal Approach

F. Rau, P. Georgieva, C. Herranz, I. Val, J. Perez

2026 15th International Symposium on Communication Systems, Networks and Digital Signal Processing (CSNDSP), Edinburgh, United Kingdom, 15–17 Jul. 2026 — DOI: 10.1109/CSNDSP68462.2026.11654486

Traffic classifiers must learn new applications without forgetting old ones or retraining from scratch. This paper freezes the core of the neural network and retrains only its final layer, using new data plus a small sample of old data. The method keeps accuracy above 87% while cutting training time by up to 92% and storage needs by up to 77%.

A SDN-Based Access Control Framework for Powerline Communications Networks

L. M. Giraldo, J. Perez, R. Garcia, A. Pouralizadeh, V. Jungnickel, F. Rau, C. Herranz-Claveras

2026 15th International Symposium on Communication Systems, Networks and Digital Signal Processing (CSNDSP), Edinburgh, United Kingdom, 15–17 Jul. 2026 — DOI: 10.1109/CSNDSP68462.2026.11654376

Powerline networks lack a simple way to control which devices are allowed to connect. This paper adds a Software-Defined Networking layer that blocks unauthorised devices at the network edge, tested on a hardware setup combined with a simulated powerline network. The approach stops unwanted devices without reducing network throughput.

LSTM-Based Traffic Prediction for 5G Mobile Network Operator

F. Rau, C. Herranz, I. Val, J. Perez

2025 International Conference on Software, Telecommunications and Computer Networks (SoftCOM), Split, Croatia, 18–20 Sep. 2025 — DOI: 10.23919/SoftCOM66362.2025.11197406

An LSTM-based predictive method for traffic modeling in 5G networks is proposed, using real mobile operator data. Results show that LSTM outperforms traditional RNNs in accuracy and computational stability, with inference times ranging from 2.2 to 2.8 seconds.

Implementing a Traffic Classifier on Embedded SoC Systems with Deep Learning Processor Unit

F. Rau, L. M. Giraldo, C. Herranz, J. Perez, I. Val, R. Garcia

2025 International Conference on Software, Telecommunications and Computer Networks (SoftCOM), Split, Croatia, 18–20 Sep. 2025 — DOI: 10.23919/SoftCOM66362.2025.11197409

A traffic classifier is designed and deployed on a Zynq UltraScale MPSoC platform using a Deep Learning Processor Unit, classifying flows from applications like YouTube, Netflix, and Teams. A Fully Connected Neural Network trained with Vitis AI achieves 80.5% accuracy at 700 flows per second, validating efficient AI inference on resource-constrained hardware.

A Novel Flow-Based Online Network Traffic Classification Using Machine Learning

F. Rau, C. Herranz, I. Val, P. Georgieva, J. Perez

TechRxiv, 2025 — DOI: 10.36227/techrxiv.175459507.75164419/v1

A flow-based traffic classification model using machine learning extracts 29 statistical and temporal features from 5-second windows, enabling real-time classification while preserving user privacy. Evaluated across multiple applications and six public datasets, the model consistently exceeds 96% accuracy, outperforming state-of-the-art approaches and proving suitable for modern network environments.

VLC Data: A Multi-Class Network Traffic Dataset Covering Diverse Applications and Platforms

F. Rau, C. Herranz, I. Val, J. Perez

EU Open Research Repository (Zenodo), April 2025 — DOI: 10.5281/zenodo.15121418

Publicly available multi-class network traffic dataset collected across diverse applications and platforms, enabling reproducible research in traffic classification and network management.

A Novel Traffic Prediction Method Using Machine Learning for Energy Efficiency in Service Provider Networks

F. Rau, I. Soto, D. Zabala-Blanco, C. Azurdia-Meza, M. Ijaz, S. Ekpo, S. Gutierrez

Sensors, vol. 23, no. 11, p. 4997, 2023 — DOI: 10.3390/s23114997

A systematic methodology using recurrent and sequential neural networks is proposed to address energy efficiency prediction in telecom data centers, comparing RNN, LSTM, GRU, and OS-ELM models. OS-ELM outperforms the others in accuracy and computational efficiency, achieving potential energy savings of up to 12.2% in a single day on real traffic data.

Energy Efficiency Using Prediction Techniques and Artificial Intelligence

F. Rau — M.Sc. in Electrical Engineering, Universidad de Santiago de Chile (USACH), 2023

Development of energy efficiency strategies for Internet Service Provider networks using traffic prediction techniques based on artificial intelligence.

Remote Detection of COVID-19 Using 5G and AI

R. Zamorano-Illanes, I. Soto, W. Alavia, V. Garcia, P. Adasme, F. Rau

13th International Symposium on Communication Systems, Networks and Digital Signal Processing (CSNDSP 2022), Porto, Portugal, 20–22 Jul. 2022, pp. 395–400 — DOI: 10.1109/CSNDSP54353.2022.9907995

Novel solution for SARS-CoV-2 detection from gel electrophoresis images of sewage samples, leveraging 5G connectivity and AI-based image processing for noise reduction and identification of characteristic virus bands for pandemic control.

MIMO QAM Indoor VLC Using Polar Codes for Low-Cost Emitters and FPGA Receiver

R. Zamorano-Illanes, M. C. Estela, I. Soto, M. Ijaz, F. Rau

4th West Asian Symposium on Optical and Millimeter-Wave Wireless Communications (WASOWC 2022), Tabriz, Iran, 12–13 May 2022 — DOI: 10.1109/WASOWC54657.2022.9798443

Indoor visible light communications system combining MIMO QAM modulation with polar codes, implemented on low-cost emitters and an FPGA receiver platform.

Forecasting Mobile Network Traffic Based on Deep Learning Networks

F. Rau, I. Soto, D. Zabala-Blanco

IEEE Latin-American Conference on Communications (LATINCOM 2021), Santo Domingo, Dominican Republic, 17–19 Nov. 2021 — DOI: 10.1109/LATINCOM53176.2021.9647788

Deep learning techniques are compared for predicting traffic peaks in a real 5G mobile core EPC node in Chile, addressing the growing challenge of network burst management. LSTM outperforms GRU in prediction accuracy by a factor of 0.4, while both exhibit identical computational cost.

Network Traffic Prediction Using Online-Sequential Extreme Learning Machine

F. Rau, I. Soto, P. Adasme, D. Zabala-Blanco, C. Azurdia-Meza

3rd South American Colloquium on Visible Light Communications (SACVLC 2021), Toledo, Brazil, 11–12 Nov. 2021 — DOI: 10.1109/SACVLC53127.2021.9652247

LSTM and OS-ELM neural network techniques are compared for forecasting real traffic demand of a Chilean ISP, addressing the challenge of cost-efficient bandwidth management. OS-ELM outperforms LSTM in computational cost by a factor of 2300 while remaining competitive in prediction accuracy, making it a highly efficient alternative for traffic forecasting.

Study of the Soiling Effect on the Performance of Photovoltaic Solar Panels in a Residential Area

F. Rau — B.Sc. in Industrial Engineering, Universidad Técnica Federico Santa María (UTFSM), 2017

Investigation of the soiling effect on photovoltaic solar panel performance in a residential environment, analyzing the impact of dust and particle accumulation on energy generation efficiency. Includes field measurements and performance degradation modeling under real-world conditions.

Project Proposal for LAN Network Installation for a Small Enterprise with Open-Source Security

F. Rau — B.Eng. in Electrical Engineering (Electronics & Telecom), Universidad de Santiago de Chile (USACH), 2009

Design of a LAN network installation project for a small enterprise incorporating open-source security solutions. Covers network architecture design, hardware selection, IP addressing, and access control policies using open-source firewall and intrusion detection tools.