Research On Object Classification Technology Applied To Wireless Communication Signal Modulation Recognition

Automatic modulation recognition in wireless communication is an important part of spectrum resource application and management,and is also one of the hottest research directions in this field.Traditional wireless communication signal modulation recognition research mainly uses modulation recognition based on maximum likelihood ratio decision theory and feature-based modulation recognition.The shortcomings of these two methods limit the accuracy and speed of modulation recognition.improve.In this paper,the object classification technology based on convolutional neural network is introduced into the modulation recognition of wireless communication signals,and combined with the needs of automatic modulation recognition to study and improve it.An improved object classification algorithm is proposed and carried out.Experimental verification,the work is as follows:Firstly,the method of combining decomposition convolution and grouping convolution is used to improve the deep network model VGGNet,and the mixed feature extraction network model DSNet is designed with double Re LU activation function and channel shuffling technology,and the accuracy and speed of feature extraction Explored.Secondly,on the basis of DSNet,the feature extraction technology is used to improve the feature extraction network.A dense feature extraction network Dence-DSNet is proposed,and the feature extraction capability of the network is explored.Finally,the speed and accuracy of the proposed new convolutional neural network object classification are tested through experiments.The experiment self-built a wireless communication signal constellation data set covering 7 kinds of modulation methods,each modulation method of 1200 sheets,according to the ratio of 2: 8 to form a training set and a test set to test on the improved network,the calculation results show that DSNet The classification correct rate is 93.7% and the average recognition speed is 14.6ms;the proposed Dence-DSNet classification correct rate is 95.1% and the average recognition speed is 18.5ms.The improved convolutional neural network algorithm has significantly improved the recognition speed and recognition rate of modulation methods.