AI4C1 – AI/ML Solutions for Communications
Wednesday, 3 June 2026, 8:30-10:00, room Sala de Conferencias 2 (2.2) (1st floor)
Session Chair: Daniel Kilper (Trinity College Dublin & CONNECT Centre, IE)
Development and Verification of Data-Driven Models for Traffic Forecasting to Support 6G Network Orchestration and Management
Hamid Asgari (Thales UK Research Technology Innovation, United Kingdom (Great Britain) & Thales UK Research, Technology, Solution & Innovation (RTSI), United Kingdom (Great Britain)); Gabriele Incorvaia (Thales UK, United Kingdom (Great Britain)); Darryl Hond (Thales UK, Research, Technology and Innovation, United Kingdom (Great Britain))
Future Open Networks (FONs) are envisaged to be AI-native and heterogeneous data-driven network systems capable of providing fast communication and a wide range of complex services and diverse applications, including autonomous vehicle relevant applications. The orchestration and management of such networks at access, at the edge, and at the core needs to incorporate intelligent processing for more informed and optimized operational decision-making. Cognitive and AI/ML functions such as service and traffic prediction, service profiling and analytics, and anomaly detection will be needed to enhance network operation, and their evolution and integration over time will enable autonomous network orchestration and management. In this paper, we briefly introduce an architectural framework for the orchestration and management of AI-native networks. We then focus on traffic forecasting as one of the functions that supports orchestration and management, providing the ability to predict future volumes of traffic at any part of the network (access, edge, and core). We develop a range of ML models for traffic forecasting, which requires time series processing, and compare their performance. Then, we consider the trustworthiness of these models, focusing on the verification of ML model performance for unseen operational data.
Geometry-Driven Neural Candidate Beam Selection for Overhead-Efficient mmWave Systems
Marco Ibáñez Véliz (Universitat Politècnica de València, Spain); Cesar Montaner (Universitat Politecnica de Valencia, Spain); Hugo Beltrán (Universitat Politècnica de València, Spain); Narcis Cardona (The Polytechnic University of Valencia, Spain)
Beam training overhead is a critical challenge in millimeter-wave (mmWave) multiple-input multiple-output (MIMO) systems, particularly in dense deployments with multiple distributed access points (APs) and mobile users. Exhaustive beam sweeping across large antenna arrays incurs significant signaling cost and latency, motivating the need for efficient candidate-beam selection strategies. We propose GeoResMLP, a lightweight neural framework for transmit-side beam selection that leverages low-dimensional geometry-derived features to predict a compact ranked set of candidate beams per AP. The approach formulates beam selection as a supervised classification problem over a fixed discrete Fourier transform (DFT) codebook, using wideband beam scores generated from ray-tracing-based channel simulations. A shared neural predictor augmented with learnable AP embeddings enables generalization across multiple APs while maintaining a small model footprint. Robustness to localization errors is explicitly addressed through position-noise augmentation during training. Numerical results obtained on a realistic multi-AP mmWave dataset demonstrate that the proposed method concentrates most of the optimal beam under exhaustive sweeping within very small candidate sets, achieving near-optimal beamforming performance with a measurement budget of only a few beams per AP. In particular, substantial reductions in beam training overhead are achieved while maintaining low average loss and favorable tail behavior under practical positioning uncertainties.
Radio Environment Mapping with World Models for Active Measurement Control: Should Networks Dream of Optimal Control?
Jernej Hribar and Ljupcho Milosheski (Jozef Stefan Institute, Slovenia); Ryoichi Shinkuma (Shibaura Institute of Technology, Japan)
Radio Environment Maps (REMs) have the potential to serve as an important enabler for intelligent modeling and control in emerging AI-native 6G networks. Despite significant progress, most REM construction methods remain passive, relying on interpolation or static uncertainty models and lacking an explicit mechanism to reason about how future measurements will affect reconstruction quality under a limited measurement budget. In this paper, we formulate REM construction as a sequential decision-making problem and propose a world-model-inspired framework for active Received Signal Strength Indicator (RSSI) map reconstruction. By learning an internal representation of the radio environment and employing a dreaming mechanism to simulate the impact of candidate measurements, the proposed approach actively selects measurement locations under a limited budget. Experimental results on real indoor RSSI data demonstrate that the proposed method significantly outperforms Gaussian Process-based interpolation in the few-shot regime, achieving up to a fivefold reduction in RMSE with the same number of measurements. These results highlight the potential of world models as a powerful paradigm for sample-efficient radio environment mapping and intelligent model-based sensing in 6G and beyond networks.
Inter-Cell Interference Rejection Based on Ultrawideband Walsh-Domain Wireless Autoencoding
Rodney Martinez Alonso and Cel Thys (KU Leuven, Belgium); Cedric Dehos (CEA, France); Yuneisy Esthela Garcia Guzman (Silicon Austria Labs, Austria); Sofie Pollin (KU Leuven, Belgium)
This paper proposes a novel technique for rejecting partial-in-band inter-cell interference (ICI) in ultrawideband communication systems. We present the design of an end-to-end wireless autoencoder architecture that jointly optimizes the transmitter and receiver encoding/decoding in the Walsh domain to mitigate interference from coexisting narrower-band 5G base stations. By exploiting the orthogonality and self-inverse properties of Walsh functions, the system distributes and learns to encode bit-words across parallel Walsh branches. Through analytical modeling and simulation, we characterize how 5G CP-OFDM interference maps into the Walsh domain and identify optimal ratios of transmission frequencies and sampling rate where the end-to-end autoencoder achieves the highest rejection. Experimental results show that the proposed autoencoder achieves up to 12 dB of ICI rejection while maintaining a low block error rate (BLER) for the same baseline channel noise, i.e., baseline Signal-to-Noise-Ratio (SNR) without the interference.
Cross-Domain Query Translation for Network Troubleshooting: A Multi-Agent LLM Framework with Privacy Preservation and Self-Reflection
Phuc N Tran, Sr and Brigitte Jaumard (Concordia University, Canada); Karthikeyan Premkumar (Data Scientist, Canada); Salman Memon (Ericsson, Canada)
This paper presents a hierarchical multi-agent LLM architecture to bridge communication gaps between non-technical end users and telecommunications domain experts in private network environments. We propose a cross-domain query translation framework that leverages specialized language models coordinated through multi-agent reflection-based reasoning. The resulting system addresses three critical challenges: (1) accurately classify user queries related to telecommunications network issues using a dual-stage hierarchical approach, (2) preserve user privacy through the anonymization of semantically relevant personally identifiable information (PII) while maintaining diagnostic utility, and (3) translate technical expert responses into user-comprehensible language.
Our approach employs ReAct-style agents enhanced with self-reflection mechanisms for iterative output refinement, semantic-preserving anonymization techniques respecting k-anonymity and differential privacy principles, and few-shot learning strategies designed for limited training data scenarios. The framework was comprehensively evaluated on 10,000 previously unseen validation scenarios across various vertical industries.
