Wideband Spectrum Sensing Algorithm Based On Compressed Sensing

With the rapid development of wireless communication technology and the increasing demand for wireless communication services,spectrum resources are becoming increasingly scarce.In order to improve spectrum efficiency,cognitive radio,which based on the concept of dynamic spectrum sharing,has attracted researchers’ attention.As one of the most basic technologies in cognitive radio,the research related to narrowband spectrum sensing methods has been very mature.At present,the spectrum detection range of cognitive radio has been expanded to wideband frequency bands of several thousand MHz.However,due to bandwidth differences,narrowband spectrum sensing methods cannot be directly applied to wideband spectrum sensing problems.In order to quickly detect spectrum holes in wideband communication systems,wideband spectrum sensing technology has become an important research direction.Based on the Nyquist sampling frequency,wideband sensing methods face sampling bottlenecks and other issues.Compressed sensing provides a theoretical basis for subNyquist sampling implementation.At the same time,multi-band wideband spectrum sensing can be transformed into a multi-label multi-classification problem of whether authorized user signals exist.Therefore,deep learning technology has great potential in the field of spectrum sensing.In this theis,we studys wideband spectrum sensing technology using compressed sensing and deep learning.The theis focuses on the compressed power spectrum estimation algorithm under the compressed sampling framework of single antenna and multichannel.This algorithm does not directly reconstruct wideband signals but reconstructs the power spectrum of wideband signals through compressed sampling samples.The theis proposes a compressed covariance spectrum sensing algorithm based on deep learning.The algorithm constructs the covariance of compressed sampling signals as input and uses a depth-separation convolutional neural network to learn frequency band information.It is also suitable for non-sparse wideband signals.Experimental results show that the proposed algorithm has significant performance advantages over the power spectrum estimation-based spectrum sensing algorithm under this sampling framework.Moreover,this theis uses transfer learning to extend the proposed algorithm to a compressed sampling framework with multiple antenna and single channel.Simulation experiments have shown that the algorithm proposed still performs better than the compressed power spectrum estimation algorithm under the multi-antenna model.Based on the difficulty of deploying a fully neural network framework on terminals,this theis proposes a spectrum sensing algorithm based on end-edge collaboration,which reduces transmission latency through deep learning compression encoders and decoders while ensuring sensing performance.