| Multi-band spectrum sensing technology is facing the pressure of broadband sampling,which requires under sampling by compression sensing theory,and then the spectrum usage is obtained.It has the requirements of reliability,effectiveness and real-time.The traditional model-driven algorithm has made great progress in the exploration of prior knowledge of signal and the application of optimization algorithm.However,the performance of the traditional model-driven algorithm is not satisfactory in the case of incomplete agreement between signal and model and low SNR(lower than 0d B)environment.Since the rise of machine learning,data-driven algorithms have made great success in many fields.This paper selects some mature models to apply them to multi-band spectrum sensing,which greatly improves the performance of the algorithm.Because of the weak correlation of communication signals with different frequency support,the pure data-driven algorithm needs large-scale training sets of the same distribution,which affects the practical application.We think that the data and model double drive algorithm can combine the advantages of the two.This paper studies the typical dual drive algorithm,and draws lessons from the achievements in the image field.A multi-band spectrum sensing algorithm based on the optimal depth prior framework is proposed.The specific research contents and achievements are as follows:Firstly,this paper presents the signal model of multi-band spectrum sensing,and discusses three specific sensing indicators.Then,sparse Bayesian algorithm is selected as a typical model-driven algorithm.One example is the pattern coupled sparse Bayesian algorithm,which pays more attention to mining the sparse characteristics of signal blocks.It uses a slightly more complex model but is closer to the sparse characteristics of signal blocks.Another example is the fast sparse Bayesian algorithm,which uses the general probability model,but solves the problem of large computational complexity through relaxation technology.The simulation results show that the two algorithms can accurately restore the support set at a certain SNR,and the performance of the fast algorithm is slightly worse because of the approximation.We also verify the results on the actual sampled communication signals.Finally,because the data-driven algorithm needs large-scale training set,which increases the complexity of training and the difficulty of application,this paper studies the data-driven and model-driven algorithm,and synthesizes the advantages of the two.Firstly,taking the classic soft threshold iterative algorithm and its machine learning version as an example,this paper analyzes how to expand the model-driven algorithm into a dual driven algorithm.Then,the successful optimal depth prior framework in image processing is applied to multi-band spectrum sensing.The framework still starts from the probability model,uses neural network to extract prior information,and then solves various signal processing tasks.In this paper,the loss function is modified,and the problem that affects the complexity of the algorithm is solved by eigenvalue decomposition.Finally,the multi-band spectrum sensing algorithm under the optimal depth prior framework is proposed.The simulation results show that the performance of the algorithm is slightly lower than that of the data-driven algorithm,but it is still significantly higher than that of the model-driven algorithm,and greatly reduces the size of the support set and the complexity of the algorithm. |
