Research On Spectrum Sensing Technology In Cognitive Radio Based On Machine Learning

In recent years,with the continuous development of communication technologies,wireless communication applications are exploding.The demand for spectrum resources is getting higher and higher,and limited spectrum resources are becoming increasingly scarce.Cognitive radio technology is one of the effective methods to solve this problem.It can realize the reuse of idle spectrum resources through the unauthorized users'(ie,secondary users')perception of the surrounding electromagnetic environment,thereby improving the utilization efficiency of the spectrum.As the basis for cognitive radio systems to make reasonable frequency decisions,spectrum sensing is the most important part of cognitive radio systems.In order to ensure that the access behavior of secondary users does not interfere with authorized users,the spectrum sensing algorithm should have higher perceptual accuracy.At the same time,in order to enable cognitive users to obtain as many spectrum access opportunities as possible,and vacate the spectrum as soon as possible to avoid frequency conflicts when an authorized user reoccupies the spectrum,the spectrum sensing algorithm should also have high sensing efficiency.In addition,blind sensing is also a major challenge in spectrum sensing applications when the priori information of authorized users is not available.In this paper,three classical spectrum sensing algorithms,energy detection,matched filter detection and cyclostationary feature detection,are firstly collated and analyzed.It is found that they do not use historical observation information to help detect the judgment,but the measured data shows that the authorized users use The frequency behavior has certain regularity,and the reasonable use of this law can effectively improve the detection performance of the system.From this point,this thesis studies the application of time series model in machine learning algorithm,hidden Markov model,in the classical spectrum sensing algorithm.Using signal energy as the test statistic,the training and application process of the model are introduced in detail,and the test decision formula is given.The analysis shows that the algorithm dynamically selects the energy decision threshold based on the prior probability,so that better detection results can be obtained.Then,on account of the memoryless property of the distribution of state dwell time in hidden Markov model,which can not describe the high-order time series model well,this thesis proposes an energy detection algorithm based on hidden semi-Markov model.The specific implementation methods are given,including model parameter estimation based on EM algorithm and real-time estimation of spectrum state.Based on the hidden Markov model,this model has a new parameter to represent the duration of the state,so that the state dwell time can be described as obeying an arbitrary distribution of time series models,and a more accurate a priori estimate can be obtained,which leads to more excellent test results.In the end,the simulation experiments of these two algorithms are carried out.The results show that when there are some regularities in the frequency of authorized users,both models can effectively improve the performance of energy detection,but when the law does not satisfy the Markov property When the hidden Markov model's detection performance is affected,the hidden semi-Markov model does not.Its application scope is wider.At the cost,the model is more complicated,and its training overhead is twice of hidden Markov model.Times,but in general it is within acceptable limits.Compared with the energy detection algorithm,the spectrum sensing algorithm based on the hidden semi-Markov model improves the noise margin by 4dB,which can work in a lower signal-to-noise ratio environment;the sensing time is shortened to 1/4,and the spectrum sensing efficiency is higher.Excellent detection performance.Finally,it is concluded that the spectrum sensing algorithm based on hidden semi-Markov model is very practical.