Blog Entries
Another Machine Learning project?
05 November 2022 by Eva Lagunas
If you follow a bit the latest trends in technology, you will have noticed that Machine Learning (ML) is the new topic in hype. There has been numerous articles, research works and real-world applications where ML is being discussed / used. Research in ML has been active for several decades… Then, you may wonder, why another Machine Learning project now?
FNR funded project SmartSpace focuses on the benefits that ML can bring to the Satellite Communications (SatCom) systems. The non-terrestrial “world” has been tremendously evolving in the past years, and is progressing towards having all the ingredients to make ML suitable for particular satellite related use-cases. The SatCom industry is witnessing a revolution regarding both space and ground segment, with the full payload digitalization and the multi-gateway ground system. The first is arguably one of the most exciting developments to date, since enables more sophisticated payload designs able to adapt the satellite resources to the real and moving traffic conditions. The latter, has been imposed by the trends of non-geostationary systems as well as the feeder links being moved to higher frequency bands.
SmartSpace would like to jump into the ML bandwagon and investigate what ML can bring to satellite communications. Some ideas that we would like to explore are listed below:
- [Algorithm Acceleration] With ML, complex and complicated problems typically encountered in satellite communications can be addressed and efficiently solved. We usually have algorithms that solve these problems but unfortunately, its complexity is too high preventing its success in real applications.
- [Estimate unknown or inaccurate system model] Satellite payloads are known by its nonlinearities effects, which cause significant headaches to engineers in order to countermeasure their impact on the final performance. Furthermore, it is impossible to assume an accurate CSI estimate at the gateway side. To deal with the channel model drawbacks, ML can be exploited to either predict the channel directly from experience or to estimate the parameters to fine-tune the already available channel model.
- [Predict Network Load] Humans tend to behave according to patterns that are predictable (e.g. browsing internet and watching TV-on-demand during evenings). Even the aeronautical traffic follows a schedule with most of the traffic taking place during day-time. In general, the analysis of such human patterns can help in predicting the satellite data traffic, which at the same time can be used to better distribute the satellite resources where needed.
Do you have more ideas? Feel free to drop us a message !