Bridging AI and Wireless Technologies Toward a World of Connected Intelligence

A method for associating user equipment with base stations of a cellular network, in particular of a heterogeneous network such as a 5G network, uses a multi-agent reinforcement learning (MARL) algorithm. When a user equipment enters the network, the agent associated with this equipment downloads the instance of an association meta model to a coordinating unit, pre-trained by means of reinforcement learning. The agent subsequently constructs a local observation vector, comprising observables relating to the user equipment, and a global observation vector, including observables relating to the environment of this equipment. The local and global observation vectors are projected into the same reference space, the concatenation of the vectors thus projected encoding the state of the agent. It deduces from the model and from this state, an association decision. The model is updated according to a policy gradient method to consider evolution of the environment of the user.

The present invention relates to a conjoint communication and sensing method for detecting the movement of an obstacle in the environment of a node in a wireless telecommunication network and, where applicable, localising this obstacle. The node divides its environment into angular sectors and measures, at a plurality of observation instants, a quantity characteristic of the signal-to-noise-plus-sector-interference ratio in each of these sectors (410-420) to form a sensing matrix. A blind separation of sources is implemented to extract from this matrix a matrix of sector contributions of the interfering signals (430). The movement of any obstacle is detected from the history of the sector contributions (440), which makes it possible, where applicable, to predict a blockage situation (450) and to modify the allocation of the radio resources accordingly (460).

The present invention relates to a method for managing radio resources in a cellular network. For each node of interest of the network, a set of neighbouring nodes is determined. Each neighbouring node performs a local observation of its environment and extracts thereform a plurality of radio characteristics, then encodes each of these radio characteristics in the form of a message which is transmitted to the node of interest. The node of interest then generates a local mapping of each radio characteristic by aggregating the messages encoding this characteristic. Afterwards, the different local mappings are fused using fusion parameters so as to provide a hybrid local mapping of the radio characteristics. The node of interest decides at all times to perform an action amongst a finite set of possible actions, based on said hybrid local mapping and on a radio resource management strategy defined by a conditional probability parameterised distribution of each action. The set of fusion parameters as well as the set of the parameters of the conditional probability distribution undergo a reinforcement learning so as to maximise a reward over time, dependent on an objective function of the network.

Large language models (LLMs) rely on key-value (KV) caches for efficient autoregressive decoding; however, cache size grows linearly with context length and model depth, becoming a major bottleneck in long-context inference. Prior KV cache compression methods either enforce rigid heuristics, disrupt tensor layouts with per-attention-head variability, or require specialized compute kernels. We propose a simple, yet effective, KV cache compression framework based on attention-guided, layer-adaptive composite tokens. Our method aggregates attention scores to estimate token importance, selects head-specific tokens independently, and aligns them into composite tokens that respect the uniform cache structure required by existing inference engines. A global allocation mechanism further adapts retention budgets across layers, assigning more capacity to layers with informative tokens. This approach achieves significant memory reduction while preserving accuracy, consistently outperforming prior structured and semi-structured methods. Crucially, our approach remains fully compatible with standard inference pipelines, offering a practical and scalable solution for efficient long-context LLM deployment.

Root Cause Analysis (RCA) in mobile networks remains a challenging task due to the need for interpretability, domain expertise, and causal reasoning. In this work, we propose a lightweight framework that leverages Large Language Models (LLMs) for RCA. To do so, we introduce TeleLogs, a curated dataset of annotated troubleshooting problems designed to benchmark RCA capabilities. Our evaluation reveals that existing open-source reasoning LLMs struggle with these problems, underscoring the need for domain-specific adaptation. To address this issue, we propose a two-stage training methodology that combines supervised fine-tuning with reinforcement learning to improve the accuracy and reasoning quality of LLMs. The proposed approach fine-tunes a series of RCA models to integrate domain knowledge and generate structured, multi-step diagnostic explanations, improving both interpretability and effectiveness. Extensive experiments across multiple LLM sizes show significant performance gains over state-of-the-art reasoning and non-reasoning models, including strong generalization to randomized test variants. These results demonstrate the promise of domain-adapted, reasoning-enhanced LLMs for practical and explainable RCA in network operation and management.

The increasing adoption of artificial intelligence in telecommunications has raised interest in the capability of Large Language Models (LLMs) to address domain-specific, mathematically intensive tasks. Although recent advancements have improved the performance of LLMs in general mathematical reasoning, their effectiveness within specialized domains, such as signal processing, network optimization, and performance analysis, remains largely unexplored. To address this gap, we introduce TeleMath, the first benchmark dataset specifically designed to evaluate LLM performance in solving mathematical problems with numerical solutions in the telecommunications domain. Comprising 500 question-answer (QnA) pairs, TeleMath covers a wide spectrum of topics in the telecommunications field. This paper outlines the proposed QnAs generation pipeline, starting from a selected seed of problems crafted by Subject Matter Experts. The evaluation of a wide range of open-source LLMs reveals that best performance on TeleMath is achieved by recent models explicitly designed for mathematical or logical reasoning. In contrast, general-purpose models, even those with a large number of parameters, often struggle with these challenges. We have released the dataset and the evaluation code to ease result reproducibility and support future research.

Artificial intelligence (AI) and 5G systems, although traditionally treated as separate entities, are intrinsically interdependent for sustained success. Communication systems serve as the backbone for integrating AI into our societal infrastructure, facilitating the collection, transmission, and processing of vast amounts of data. As 6G technologies emerge, a semantic-aware communication architecture promises to enable AI-native cooperation and distributed operations. However, the coexistence of legacy data-oriented systems with 6G semantic, goal-oriented systems presents significant challenges in terms of interoperability and resource sharing. This chapter explores state-of-the-art solutions addressing these challenges, focusing on standardized protocols, interoperability mechanisms, and semantic alignment strategies. A key aspect is the adoption of compatible ontology and logic to ensure consistency in how semantics are associated with data and knowledge is represented. Various approaches to AI interoperability are discussed, including reinforcement learning-based consensus building, environmental language grounding, curriculum learning combined with reinforcement learning, and semantic compensation for logic mismatches. The chapter details a promising solution that addresses operational complexity and transmission accuracy by modeling the implications of logic mismatches and applying semantic compensation.