Research On Algorithms Of Spectrum Analysis Based On Big Data

With the rapid development of wireless communication services,radio spectrum resources are facing severe problems of tight demand. As spectrum resources are limited nowadays, it is an important issue for us to reasonably plan the limited spectrum resources and improve the utilization of spectrum in spectrum management. Through the use of specific technical means and professional equipment, radio spectrum monitoring computes, counts and analyzes the received radio signals in dimensions such as time, frequency, space, and so on, in order to obtain information related to spectrum management. Therefore, with the wild increase of spectral data collected by spectrum monitoring, the spectrum analysis method combining data mining and spectrum analysis is proposed.This paper uses data mining algorithms to extract the potential valuable information from massive spectrum monitoring data, which can assist the decision-making in spectrum management. The paper focuses on the algorithms related to spectral anomaly-related mining and spectral occupancy prediction based on spectrum monitoring data, and achieves the following results:First, the paper proposes the methods to detect the abnormal state of the spectrum by using Logistic Regression and Naive Bayesian classification respectively. In the implementation of logistic regression algorithm, the paper presents a logistic regression learning model, which introduces the dynamic factor of the least squares method into the weight updating process to accelerate the convergence rate of the model parameters and improve the learning effect of the model.Second, the paper designs a method to predict the time of spectrum anomaly. Based on the training data of anomaly detection, the probability of the anomaly in each time period is obtained by the Naive Bayesian classification algorithm, thus the time period that is most likely to be abnormal is predicted.Third, in this paper, a spectrum occupancy prediction method is proposed. By using BP neural network, the paper takes the 24 hours every day as the 24-dimensional time characteristic attribute, and the corresponding spectrum occupancy of each hour as the characteristic parameters. Through the training of neural network, the paper achieves the spectrum occupancy prediction.