Join our Team

About this opportunity:

Telecom 6G challenge focus: Radio Resource Management (RRM) is a critical component of Radio Access Networks (RAN). It consists in allocating Time-Frequency resources for transmissions and is typically performed in many steps: time scheduling pre-selects a subset of relevant users based on their current service needs, frequency scheduling allocates these users to frequency sub-bands depending on the quality of their wireless channel, power control splits the available power among the concurrent transmissions, precoding forms directed beams to focus energy on targeted users while avoiding creating interference for the other, link adaptation selects modulation and coding schemes to maximize the quantity of information conveyed while avoiding failed transmissions. These decisions are complex, coupled together, and need to be taken every Transmission Time Interval (TTI) –every millisecond or every half-millisecond–, hence finding an optimal solution is not possible in general. Thus, RRM functions are typically performed sequentially with well-tuned optimization heuristics. These algorithms do not perfectly handle channel uncertainties when scheduling concurrent transmissions and tend to create a lot of interference at cell edges. Both of these effects significantly limit the efficiency of current networks. Future generation hardware will have more computing capabilities –including GPUs–, making it possible to use more powerful algorithms to tackle RRM decisions in a more efficient way.

Aim of the internship: This internship aims at investigating approaches to jointly optimize scheduling over multiple neighboring RAN cells, which have limited information exchange capabilities and need to coordinate their RRM decisions to avoid generating too much interference to each other.

AI approach: Our project focuses on using artificial intelligence, and in particular Reinforcement Learning (RL), to optimize RRM for 5G/6G networks. The RL approach relies on internal or public simulators to generate many RRM problem instances and learn how to take efficient RRM decisions by exploring the set of possible decisions and evaluating their impact on performance.

What you will do:

• Familiarizing with the problem addressed and the associated simulators.
• Reviewing existing literature and identifying relevant RL techniques.
• Proposing and adapting suitable RL techniques for optimizing RRM decisions.
• Adapting the environments and simulators to the problem addressed.
• Evaluating algorithm performance through rigorous simulations or experiments in realistic scenarios.

The skills you bring:

To apply for this role, you are in your second year of Master or the final year of engineering school and be studying a degree in the following subjects (or related): Computer Science, AL/Machine Learning, Electrical Engineering, Mathematics, or Physics.  To be successful in the role you have:

• good understanding of Machine Learning algorithms (Reinforcement Learning, Generative-AI, Neural Networks, times series, etc.).
• good foundation in practical programming skills, experience in Python and in using scientific Python packages such as PyTorch, Scikit-learn, Numpy, etc.
• ease with reading and implementing recent machine learning research articles from the established conferences (ICLR, ICML, NeurIPS etc).
• background in telecommunication.
• excellent communication skills both in written and spoken English.
• knowledge about stack technologies and cloud architectures (optional but recommended): Git, Docker, Kubernetes.