| With Low Earth Orbit(LEO)mega satellite constellations coming into operation,the available spectrum resources are more crowded.To improve spectrum utilization,cognitive satellite communication technology has become an important candidate technology.As the most critical step in cognitive satellite networks,spectrum sensing is an important technology to realize the dynamic allocation of spectrum resources since its accuracy can ensure that authorized satellite systems are not disturbed.Currently,most spectrum sensing algorithms are based on model-driven,and the detection performance depends largely on the predetermined statistical model.However,for some complex satellite communication environments,it is usually difficult to accurately model a sensing algorithm through a specific statistical model,which will increase the difficulty of modeling and deployment in the actual environment.In recent years,the application of data-driven deep learning technology in the field of wireless communication has become a hot research topic.Deep learning algorithm does not need any prior knowledge,has strong representation ability,and can learn complex hidden features directly from sensing signals.Therefore,this thesis applies deep learning to the spectrum sensing of cognitive satellite communication scenarios.The main research contents are as follows:(1)Based on the characteristics and functions of cognitive communication networks and satellites with different orbital altitudes,a cognitive satellite network topology architecture based on Geosynchronous Earth Orbit(GEO)relay satellites and LEO satellites is constructed.Additionally,the effects of transmission loss between the satellite-ground link,cloud attenuation,atmospheric absorption,and the antenna orientation of the GEO satellite ground station on the channel environment are analyzed.(2)Based on the proposed dual-satellite cognitive satellite network topology,a spectrum sensing algorithm based on the combination of bidirectional long short-term memory networks(Bi-LSTM)and the Bayesian likelihood ratio test is proposed.The algorithm consists of two stages: offline training and online testing.First,sensing LEO satellites receive spectrum information data from the ground stations of authorized satellites,and send them to the spectrum sensing model to complete offline training.Secondly,the Bayesian likelihood ratio is calculated as a test statistic at the output of the network,and a threshold-based detection method is designed to sense the spectrum of the signal samples received online.The model does not need any prior knowledge about the authorized signals,and the simulation verifies the effectiveness and reasonableness of the proposed algorithm.(3)Based on the proposed dual-satellite cognitive network topology,a centralized cooperative spectrum sensing algorithm based on a hybrid network of the convolutional neural network,self-attentive mechanism,and long short-term memory network is proposed,considering that the large propagation delay and fading between sensing satellites and ground stations will lead to the problem that the sensing results of a single sensing satellite are likely to be outdated or unreliable.The algorithm not only improves the sensing accuracy but also greatly reduces the communication overhead of the reporting link by sending the sensing soft features from spatially separated multiple sensing satellites to the fusion center for final judgment.In addition,a mechanism that can adaptively adjust the detection threshold,inspired by the Neyman-Pearson criterion,is designed to maximize the detection probability for a given false alarm probability.Simulations verify the effectiveness and reasonableness of the proposed algorithm.From the above research work,the following conclusions can be drawn:(1)the signal-to-noise(SNR)fluctuation of the received signal during the LEO satellite transit reaches nearly 60 dB;(2)The spectrum sensing algorithm based on the combination of bidirectional long short-term memory network andBayesian likelihood ratio test can still achieve more than 83% detection performance in the environment of the SNR of-14 dB,and it is always better than the traditional energy detection algorithm.(3)The proposed cooperative spectrum sensing algorithm based on a hybrid network further overcomes the shortcomings of non-cooperative spectrum sensing.In the environment of a SNR of-20 dB,the detection probability of the algorithm can still reach more than 90%,and it has the advantages of low communication overhead and short running time. |
