Deep Learning-Based Spectrum Sensing in Space-Air-Ground Integrated Networks

To complement terrestrial connections, the space-air-ground integrated network (SAGIN) has been proposed to provide wide-area connections with improved quality of experience (QoE). Spectrum management is an important issue in SAGIN due to the explosive proliferation of wireless devices and services. While the progress on enabling dynamic spectrum access shows promise in advancing increased spectrum sharing, the issue of reliable spectrum sensing under low signal-to-noise ratio (SNR) remains one of the key challenges faced by the spectrum management. As artificial intelligence can provide wireless networks intelligence through learning and data mining, deep learning-based spectrum sensing is proposed in order to improve the spectrum sensing performance, where a deep neural network-based detection framework is built to extract features in a data-driven way based on the covariance matrix of the received signal. To eliminate the impact of noise uncertainty, a blind threshold setting scheme is proposed without using the system prior information. Numerical analyses on simulated and real-world signals show that the detection performance of the proposed scheme is improved under a low SNR regime.