Research On Key Technologies Of Spectrum Situation Cognition

In recent years,with the rapid development of mobile Internet,industrial Internet and Internet of things(IoT),the demand for electromagnetic spectrum is increasing,and the problem of spectrum deficit is becoming more and more serious.Facing the increasingly complex electromagnetic spectrum environment and the growing demand for frequency,in order to maintain the order and security of electromagnetic spectrum and improve the overall utilization efficiency of spectrum resources,it is urgent to use the spectrum state data collected by limited sensing nodes to mine the spectrum situation information in wide geographical space.The regional demand for electromagnetic spectrum monitoring is becoming wider and wider,and it is gradually expanding from the surface to the sky.Limited by the number of monitoring nodes,moving path(trajectories)and the constraints of monitoring,storage,computing and communication resources,the spectrum situation cognition of wide area complex electromagnetic environment shows the characteristics of sparse spatial sampling and limited sample data.Starting from the multi-dimensional spectrum situation representation of space,time and frequency,this dissertation studies the spectrum situation generation method of electromagnetic environment based on spatial sparse sampling.The main research contents and contributions include the following four aspects:1.In the aspect of representation and modeling of electromagnetic spectrum situation,this dissertation proposes a multi-dimensional spectrum situation representation method of space,time and frequency based on multivariate function,establishes a spectrum situation representation model suitable for different transmitter characteristics,analyzes the identifiability and recoverability conditions of the spectrum situation model,and provides a theoretical criterion for the reliable generation of spectrum situation under spatial sparse sampling.According to whether the propagation characteristics of space radiation sources meet the situation superposition characteristics,the tensor based parametric model and data-driven nonparametric model are constructed,respectively.For the parametric model,this dissertation analyzes the identifiability conditions required by tensor decomposition method to realize the uniqueness of model parameter estimation under specific sampling patterns and nonnegative matrix decomposition method under specific prior information;For the nonparametric model,based on the correlation,continuity and low rank characteristics of spectrum situation,this dissertation analyzes the recoverability conditions required by tensor completion,interpolation and neural network methods to estimate situation,and gives the relationship between the upper bound of situation estimation error and sampling ratio.2.To solve the problem of spectrum situation generation based on parametric spectrum situation model,this dissertation establishes a tensor LL1 decomposition model with transmitter power spectral density(PSD)and spatial loss field(SLF)as parameters,analyzes the identifiability conditions of model parameters,and designs an efficient situation estimation algorithm.For the scenario of regular sampling paradigm that meets the identifiability conditions,a coupled tensor LL1 decomposition(CLL1)spectrum situation estimation algorithm is proposed;For the scenario with very low spatial sampling ratio and random sampling location,an improved non-negative tensor LL1 decomposition(NNLL1)spectrum situation estimation algorithm is proposed by using the non-negative constraints of SLF and PSD.Simulation results show that the performance of the proposed CLL1 and NNLL1 algorithms are better than that of the benchmark algorithm under both regular and random sampling patterns.3.To solve the problem of dynamic spectrum situation generation based on nonparametric spectrum situation model,a tensor completion model based on tensor canonical polyadic decomposition(CPD)is established,and an online tensor completion algorithm for sequential spectrum situation estimation is proposed.To meet the requirements of online algorithms on the low storage space and low running time,this dissertation proposes dynamic window size CPD(DW-CPD)and incremental CPD(I-CPD)algorithms.In DW-CPD algorithm,only the correlation of data within the window is used to complete the missing data.The length of the window depends on Kruskal theorem in the identifiability condition of tensor CPD;In I-CPD,all data in the past are assigned with exponentially decreasing weight.This dissertation proves that the I-CPD method converges to a stationary point of the original problem.Simulation results show that DW-CPD and I-CPD algorithms have similar estimation accuracy in dynamic spectrum situation generation,and are better than other comparison methods;I-CPD has significant advantages in running time and storage space,especially at low sampling ratio.It is suitable for online spectrum situation generation tasks in practical applications.4.To solve the problem of joint estimation of transmitters' parameters and spectrum situation under spatial sparse sampling,this dissertation comprehensively utilize parametric and nonparametric models to establish a sparse optimization model for transmitters' location and power estimation,and proposes variable Bayesian expectation maximization(VBEM)and TPS interpolation algorithm to estimate parameters and spectrum situation.Considering the sparsity of transmitter spatial distribution,the transmitter location and power estimation problem is modeled as a sparse optimization problem.Aiming at the mismatch between the transmitter position and the candidate point set,the candidate point set is taken as a parameter variable,and the VBEM algorithm is used to alternately estimate the transmitters' position and power information.Further,considering the influence of shadowing fading on the observed data and spectrum situation,the spectrum situation is decomposed into the sum of path loss component and shadowing fading component,and VBEM-TPS algorithm is proposed to estimate the spectrum situation.VBEM-TPS algorithm first uses VBEM algorithm to estimate transmitters' position and power information and calculate the path loss component,then separates the shadow fading component at the sampling position,and uses thin plate spline interpolation(TPS)to smoothly interpolate the shadow fading component.The simulation results show that the proposed VBEM algorithm can accurately estimate the location and power information of spatial transmitter,and the VBEM-TPS algorithm can estimate the high-precision spectrum situation.