The new era of satellite constellations in a low earth orbit (LEO) will most probably lead to additional interference situations The objective of the project is to validate the benefits of using artificial intelligence (AI) and/or machine learning (ML) algorithms to improve existing satellite interference mitigation techniques and develop new applications to predict satellite interference before they even occur.
The search for interferences in a provided spectrum plot can be very tricky as every spectrum plot is different as well as the nature of the interference signal. For example, an interference signal can be hidden behind another signal, it can change in frequency, in bandwidth and/or in power.
A huge number (millions) of high-quality sample spectrum plots need to be generated to train the AI.
The intended project investigates in the feasibility of the following new technologies:
- Automatic identification of interference situations via spectrum analysis
- Prediction of satellite link quality degradation caused by bad weather scenarios
- Automatic identification of interference situations caused by non-GSO satellites
Automatic identification of interferences in several steps:
1. Blind scan identification by AI/ML. Within this step an AI model is trained such that carrier shapes can be identified within a spectrum by the AI model.
2. Anomaly detection, 1D model. This step identifies anomalies compared to a predefined mask from a single spectrum measurement
3. Anomaly detection, 2D model. This step identifies anomalies compared to a predefined mask from a time series of spectrum measurements.
4. Interference classification using 1D supervised model. This step classifies the interference within a single spectrum measurement
5. Interference classification using 2D supervised model. This step classifies the interference within a time series of spectrum measurements
Within this project several models for AI/ML are developed. The architecture is split into two parts, one part for data generation and the second part for AI/ML development, training und verification. Data generation is performed with SkyMon and exported. With a special developed analysis tool, the generated data will be annotated for AI/ML into interference free and interfered recordings. Further the analysis tool allows adding interference for AI/ML development and training. The last step is the AI/ML development, training and verification.
The project plan consists of four phases. Within the first phase a feasibility study is performed to identify if the proposed performance parameters can be reached in theory. This phase is the scope of this project. The second phase consists of the design of the solution. The third phase is the implementation of the solution and the fourth phase is the test phase.
Feasibility studies are ongoing. First results look promising.