PAGE CONTENTS
Objectives
This project aims to identify problems in the field of satellite communications that are suitable for the methods and techniques provided by machine learning (ML) approaches.
It is investigated, if ML models can potentially represent an alternative to solve computationally expensive problems or can provide even performance gains w.r.t. traditional approaches in satellite communications.
In the course of this project, potentially adequate ML use cases in satellite communications are collected. A subset of these scenarios is selected for a proof-of-concept (PoC) implementation and performance assessment.
Challenges
The challenge in this project is to identify problems in satellite communications that could benefit from the use of machine learning techniques.
Once a problem in satellite communication is selected, the target is to detect suitable ML algorithms for similar problems in other fields of application (i.e. computer vision, medical applications, etc.) that show a promising performance gain and then adapt them to the selected problem.
The quality of the training, validation and test dataset is crucial for the success of the ML model, since the dataset has to be well prepared in order to avoid over-fitting.
System Architecture
The ML algorithm is developed as a stand-alone model.
Training, validation and test datasets for the use case specific suitable ML models are generated based on measurement data and physical layer simulations.
The performance obtained by well-established approaches in satellite communications represents the baseline reference performance for the ML approach.
A specific ML approach is applicable for a certain problem, if the achieved performance either exceeds the reference, or meets the reference performance but significantly improves the processing speed.
Plan
The project is roughly structured in two steps, where in the course of step 1, potential application areas for machine learning in the field of satellite communication are investigated. From the list of identified use cases, a subset is selected for PoC.
The actual implementation and PoC is executed in step 2, which is concluded by a summary of potential future ML fields of application, guided by the obtained performance gains.
Current Status
The project has been finished in 2020 with the deliverable of the final report and an executive summary. Several conference papers related to this activity have been submitted.
