Master Thesis: A Learned World Model for Downlink Link Adaptation

Join our Team

About this opportunity:

We are seeking a talented Master's student to develop an action-conditioned world model for downlink link adaptation in AI-native 5G/6G radio access networks. The thesis will combine real radio and baseband trace data, predictive modeling, and offline reinforcement learning to investigate whether synthetic model-generated trajectories can enable safer and more sample-efficient policy training. 

What you will do:

•  Characterize available 5G cell and baseband trace data, including radio conditions, mobility, interference, and traffic load. 

•  Preprocess traces into state, action, next-state, and key-performance-indicator tuples for model training and evaluation. 

•  Design and train a compact latent, action-conditioned world model that predicts short-horizon throughput, block error rate, channel-quality indicator, and spectral-efficiency trajectories. 

•  Evaluate single-step and multi-step prediction accuracy and study how well the model separates the effect of modulation-and-coding actions from external channel variation. 

•  Integrate the learned world model into an offline reinforcement-learning pipeline to generate synthetic rollout data. 

•  Compare rule-based outer-loop link adaptation, logged-data-only offline reinforcement learning, and world-model-augmented reinforcement learning. 

•  If time permits, investigate calibrated uncertainty estimates to restrict policy exploration to regions where predictions are reliable. 

•  Document methods, results, and recommendations in the thesis report and present the work at the final defense. 

•  Collaborate with supervisors and radio, AI, and baseband experts to ensure technical relevance and sound evaluation. 

The skills you bring:

Required Skills and Qualifications 

•  Enrolled in or recently admitted to a Master’s program in Electrical Engineering, Computer Engineering, Computer Science, Machine Learning, Wireless Communications, or a related field. 

•  Strong foundation in machine learning and data analysis. 

•  Programming experience in Python and familiarity with a deep-learning framework such as PyTorch. 

•  Basic understanding of wireless communications, radio access networks, or link-level performance metrics. 

•  Ability to work with time-series or sequential data and design reproducible experiments. 

•  Solid technical writing and communication skills. 

•  Independent, analytical, and collaborative problem-solving mindset. 

Preferred Qualifications 

•  Experience with reinforcement learning, offline reinforcement learning, model-based reinforcement learning, or sequence modeling. 

•  Familiarity with latent dynamics models, recurrent state-space models, transformers, probabilistic models, or uncertainty estimation. 

•  Knowledge of 5G/6G link adaptation, modulation and coding schemes, channel-quality reporting, block error rate, or radio scheduling. 

•  Experience with MATLAB for signal-processing, trace preprocessing, or validation. 

•  Experience handling large experimental datasets, simulation traces, or performance-counter logs.