Thesis description

Future wireless systems may evolve beyond isolated devices toward a Super-UE: a coordinated cluster of smartphones, AI glasses, smart watches, earbuds, and wearables that pools their antenna, communication, computing, and sensing capabilities as one logical terminal. Together they enhance connectivity, enable distributed AI inference, and improve user experience. Cooperation, however, is neither free nor always beneficial: partner selection requires measurements and signaling; information exchange consumes radio resources, energy, and synchronization overhead, while capabilities, channels, and application requirements change dynamically. The key challenge is therefore intelligently determining when, with whom, and to what extent devices should cooperate, ensuring that the benefits outweigh the costs.

This thesis targets an AI-native framework addressing two coupled challenges. First, compact, progressively refinable semantic representations that let heterogeneous devices exchange task-relevant information at capability-matched levels of detail. Second, an uncertainty-aware value-of-cooperation controller learning whether to cooperate, with which devices, and how much information and computation to exchange, priced by probing, communication, processing, and coordination costs, named as the semantic token tax, to jointly optimize the value of exchange against the net benefit of cooperation. Through modeling, learning-based design, and system-level simulations, the thesis will quantify inference accuracy, user experience, communication efficiency, latency, and energy, identify when cooperation is beneficial, and deliver a reproducible framework and proof-of-concept demo.

Qualifications

• Master’s student in Computer Science, Electrical Engineering, Mathematics, or a related field.

• Good knowledge of AI/ML with interest in signal processing and wireless communications.

• Python programming skills, with experience in PyTorch or TensorFlow.

Preferred starting date

ASAP in 2027.