Research On Wireless Network Signal Recognition Technology Based On Machine Learning

With the development and popularization of modern wireless communication technology,the number and scale of wireless network equipment are rapidly expanding,especially in the ISM frequency band,because it can be used without authorization,its channel space becomes very crowded and busy.Accurate identification of wireless signals can help people use wireless channels more efficiently.At the same time,in the electronic countermeasure,the identification of the wireless network signal can monitor the enemy effectively and make the enemy gain the advantage.Therefore,wireless network signal recognition technology is of great significance in both civilian and military fields.There are two kinds of wireless signal recognition technologies: one is the recognition algorithm based on maximum likelihood discrimination;The other is the recognition algorithm based on feature extraction.The maximum likelihood method USES probability theory and hypothesis testing theory as tools to compare the signal to be recognized with the typical signal,minimize the cost function as the principle,and combine with appropriate threshold for comparison to form the judgment criterion.The feature extraction is based on the classification feature extraction of the signal,from the early amplitude,frequency,phase and other parameters,to the later spectrum related features,are based on this idea.In recent years,many machine learning and deep learning methods have been proposed for signal recognition.This thesis combines the traditional signal recognition method based on feature extraction with the machine learning method,from feature analysis of wireless signals to feature extraction and fusion,and finally based on the recognition and classification of target signals.The main work of this thesis is as follows:(1)This thesis studies the characteristic parameters of current wireless signal recognition used has carried on the induction,the characteristic parameters of existing will have features physical meaning of these parameters is defined as a dominant trait,due to different to identify the target characteristic parameters also have differences,this article is based on condition entropy is proposed for dominant feature parameter extraction and fusion method.(2)In order to achieve a better recognition effect,this thesis uses the one-dimensional convolutional neural network to extract the characteristics of wireless network signals.The extracted characteristic parameters are only used to classify the signals without specific physical significance,so they are called recessive characteristic parameters.The recessive feature parameters are extracted based on the spectrum of the wireless signal rather than the original signal,which can not only extract the stable features of the signal but also eliminate the influence of noise to a certain extent.(3)Is proposed in this thesis the wireless signal recognition model based on support vector machine(SVM),on the basis of the feature extraction and fusion,will become a joint characteristics and its building as the input into the support vector machine(SVM)for training and testing,to optimize the classifier,get effective classifier for the wireless network signal recognition and classification.