Research On Application Of Support Vector Machine In Frequency Estimation Algorithm

If authorized signal is interfered, dynamical spectrum aggregation and utilizing the idle spectrum resource by opportunity will be failed. So real-time spectrum sensing to a fixed range is the principal task. Namely, we need to search and analysize this spectrum continually, and judge whether authorized users exist. If the authorized signals are detected, we also want to derive the their parameters, e. g., modulation mode, carrier frequency and bandwidth. Therefore, we can evaluate this spectrum’s characteristics and usage condition, find the idle part and make use of it for opportunity communication. Where the frequency estimation of authorized signal is the keypoint of the hole spectrum sensing process, its results will directly reflect the usage condition of authorized users and describe the utilization of this spectrum.However, as the reason of influence by the time-variation of parameter, geographical position, transmitted distance and so on, the performance of spectrum sensing is restricted by the conditon of small sample and low Signal-to-Noise Ratio(SNR). Statistical Learning Theory(SLT) is specializing in the research of small sample condition, and it will offer a novel resolution to this problem. As its concrete implement, Support Vector Machine(SVM) shows the soul of Structure Risk Minimization(SRM) sense. At the same time, it has excellent capabilities in generalization, high-dimensional processing and nonlinear processing:Also, it overcomes the over-fitting and local minimum problems currently existing in artificial neural network.This paper has applied SVM to solve the frequency estimation problem. Also, the Discrete Fourier Transform(DFT)-based, phase-based frequency estimation algorithm, and the frequency estimation problem under the condition of chirp signal and non-Gaussian noise have been mainly researched. And in the end, the scheme of spectrun sensing system based on SVM has been designed. The main work and innovations in the article are as follows:1. The computational complexity and estimation performance of DFT-based frequency estimation algorithm are two contradictories. This paper regard Least Squares Support Vector Regression(LS-SVR) as a linear interpolator. Moreover, the interpolation range is shrunk between the left and the right neighbor of the maximum value of discrete amplitude spectrum, and a LS-SVR-based DFT interpolating frequency estimation algorithm is proposed.2. There are two important factors which impact the performance of phase-based frequency estimation algorithms remarkably:the approximations of noise phase model and imperfections in phase unwrapping process. From the viewpoint of linear relationship between absolute phase and time series, this paper absolutely makes use of excellent learning and generalization capabilities of SLT to small sample, and presents a Support Vector Regression(SVR)-based robust phase unwrapping and frequency estimation algorithm.3. Aim to the frequency estimation problem under the unknown nosie distribution, algorithms based on ML are ineffective. From the perspective of limited symbol characteristics of modulation information, this paper make use of SLT to construct a SRM function with respect to frequency and convert the estimation problem into deriving the extremum value of a classification function. Ultimately, a Least Squares Support Vector Classification(LS-SVC)-based constellation algorithm is proposed to solve the frequency estimation problem.4. As the reason of the quadratic relationship between absolute phase and time series, this paper select quadratic polynomial kernel function to unwrapping phases. Besides, SVR’s outstanding nonlinear processing capability is utilized to estimate the Instantaneous Frequency(IF), Instantaneous Frequency Rate(IFR) and initial phase of chirp signal.5. According to the spectrum sensing requirements of sub-project ’Research on IMT-A Spectrum Aggregation Technology’in National Significant Science and Technology Project’A New Generation of Wireless Broadband Mobile Network’, this paper designs a spectrum sensing system scheme based on SVM.