Research On And Implementation Of Noncooperative Intelligent Spectrum Sensing Algorithm

In recent years,with the rapid development of radio spectrum related industries in China,the demand for spectrum resources is increasing.Under the traditional spectrum management mode,the spectrum utilization rate is low,so choosing a reasonable and effective spectrum management mode is more conducive to the healthy development of spectrum-related industries.Cognitive radio is an efficient dynamic spectrum allocation management mode.Its implementation can be divided into three parts: spectrum sensing,spectrum sharing and spectrum management.Spectrum sensing is the premise of spectrum sharing and the basis of the whole cognitive radio system.Fast and accurate spectrum sensing plays an important role in improving spectrum utilization.Non-cooperative spectrum sensing algorithms in spectrum sensing is studied,including energy-based spectrum sensing algorithm and depth learning-based spectrum sensing algorithm.At the same time,the corresponding non-cooperative spectrum sensing software system is designed according to the actual application requirements.The main contents and achievements of this paper are as follows:1.To overcome the shortcomings of traditional energy sensing methods,which have poor perception performance and are vulnerable to noise uncertainty,a double threshold energy sensing algorithm based on exponential smoothing prediction is proposed.Based on the traditional dual-threshold energy sensing,this algorithm combines the historical sensing state information to assist in judging the band state,which effectively improves the performance of anti-noise uncertainty and spectrum sensing.According to the correlation of signals and the uncorrelated characteristics between noises,the energy detection is improved to further improve the spectrum sensing performance.2.Aiming at the low performance of traditional spectrum sensing methods under low signalto-noise ratio(SNR),a spectrum sensing algorithm based on twin convolution neural network is proposed.The twin network structure is used to enhance the ability of the neural network to resist over-fitting,and the training mode of separating feature extraction network from classification and recognition network is used to improve the spectrum sensing performance of convolutional neural network effectively.To further improve the spectrum sensing performance,a twin self-encoder based spectrum sensing algorithm is proposed,which can extract more effective sensing features.3.In order to verify the performance of the proposed spectrum sensing algorithm in complex environment,a non-cooperative spectrum sensing software system is designed.The software system has the functions of user-friendly operation and fast spectrum sensing.It also provides a platform for in-depth study of the performance analysis of spectrum sensing algorithm in the actual environment and related function expansion.