Reinforcement Learning with Imitation for Cavity filter tuning
Tid: Må 2019-06-24 kl 14.00 - 15.00
Plats: Seminar room (Rumsnr: A:641), Malvinas väg 10, Q-huset, våningsplan 6, KTH Campus
Respondent: Simon Lindståhl
Opponent: Jonas Hongisto
Handledare: Alexandre Proutiere
Examiner: Elling Jacobsen
Abstract: Cavity filters are vital components of radio base stations and networks. After production, they need tuning, which has proven to be a difficult process to do manually and even more so to automate. Previously, attempts to automate this process with Reinforcement Learning have been made but have failed to reach consistent performance on anything but the simplest filter models. This Master thesis builds upon these results and aims to improve them. Multiple methods are tested and evaluated, including introducing a pre-processing step, tuning hyperparameters and dividing the problem into multiple sub-tasks. In particular, by using Imitation learning as an initial phase, a semi realistic filter model with 13 tuning screws is tuned, fulfilling both insertion loss and return loss requirements. On this problem, this algorithm has a greater efficiency than any previously published results on Reinforcement Learning for Cavity filter tuning.