LEO satellites enable close to global communications coverage and are a key technology for future communication systems. When combining SDMA and precoding, we increase spectral efficiency while managing Inter-User-Interference. Due to the high velocities of LEO satellites and positioning imperfections, finding the optimal precoder can prove challenging.

Model-free reinforcement learning is a technique that enables finding approximately optimal solutions for complex decision problems without any a-priori assumptions on the structure of the problem, instead learning through "trial-and-error".

It has recently proven it's capability to perform on highly demanding tasks.

We offer a selection of projects to study the suitability of combining reinforcement learning and precoding for satellite communications.

Prerequisites:

- Understanding of communication fundamentals and mathematical skills

- Prior experience with Python programming required 

- Prior experience with Python machine learning toolkits recommended