Wireless Interference Classification in Satellite Communications with the Standardized Variable Distances Learning Algorithm and Adaptive Spectral Domain Segmentation

Wireless Interference Classification in Satellite Communications with the Standardized Variable Distances Learning Algorithm and Adaptive Spectral Domain Segmentation

연구 분야: Infrastructure

학회: 2024 First International Conference on Electronics, Communication and Signal Processing (ICECSP)

Satellite communications has evolved significantly in recent years with the deployment of new communication satellites both in the traditional geostationary orbit (GSO) and non-GSO (NGSO) satellites. Wireless interference to satellite communication links either intentional (jamming) or unintentional (due to network engineering errors or equipment malfunction) can cause service degradation or denial of service. In parallel, machine learning and deep learning is increasingly used in the wireless communication domain and more specifically the satellite communication domain for a variety of tasks. On the other side, deep learning algorithm require significant computational resources and time for inference, which may be an issue for a fast detection and classification of wireless interferences. Then, this paper proposes the application of computing efficient distance-based machine learning (DML) algorithms to this problem. In particular, the Standardized Variable Distances (SVD) DML algorithm is used with an improvement based on the specific aspects of the wireless interference context. Instead of applying SVD directly to the original time domain representation of the satellite signal, this paper proposes a spectral domain approach where the characteristics of the wireless interference seen as outliers of the main communication signal are exploited to adaptively reduce the input data to the SVD algorithm for enhanced classification performance and robustness to noise. The concept is to use an outlier identification algorithm to identify the most discriminating segments in the spectral domain and then reassemble them as input to the SVD algorithm. The approach is applied to a recently published data set where three types of interference are generated for four different modulation schemes. The results show that the proposed approach is able to outperform other machine learning algorithms and the baseline where the entire spectral domain is used in input to the SVD algorith... Show More