Enhancement of Spectrum Efficiency in Satellite Communication Applying Prediction Model of Machine Learning Technique

Abstract

Spectrum sensing is critical to the advancement of cognitive radio technology in next-generation wireless communication systems. According to FCC policy, the frequency coexistence of GEO (geostationary) as the main user satellite network and NGEO (non-geostationary) as the secondary user satellite network, NGEO as the interference user system, shall not cause harmful interference to the GEO system. In this situation, spectrum sensing is shown as a viable approach to avoid interference. With the increasing number of NGEO satellites in orbit, a NGEO system detecting signals from one GEO system may be affected by another NGEO system. Several sensing technologies such as energy detection, cyclo-stationary features and matching filters have been introduced in the last decade. However, these methods have several drawbacks. Consequently, all of these strategies require the establishment of a threshold, which requires previous knowledge of the noise distribution. As a result, spectrum sensing dependence in wireless communications research remains an unresolved question. In this study, we present a spectrum sensing approach for cognitive radio networks based on machine learning technique. The spectrum sensing problem is studied systematically, and as a result large-scale, exhaustive data collections are created. This dataset is used to train, validate, and test a variety of machine learning algorithms, including random forests, support vector machines, decision trees, k-nearest neighbors, and others. The model was thoroughly tested and evaluated using matrices such as detection probability, false alarm, and missed detection, sensing time, delay, throughput and classification accuracy. Thus, according to the simulation data, random forest models and neural networks outperform all other machine learning methods. © 2022 IEEE.