Research On Modulation Recognition Algorithm Based On Neural Network And Multiple Features

Modulation recognition is a key technology in the communication field,especially in the field of non-cooperative communication,which can realize the automatic identification and classification of unknown signal modulation modes without prior knowledge.With the rapid development of the modern communication technology,modulation methods become more complex and diverse,and modulation recognition technology is also facing more challenges.Although some related studies were conducted,there are still many problems and research work that need to be resolved and improved.Therefore,in this paper,the modulation recognition technology is deeply analyzed and studied.The specific work content is as follows:(1)In view of the poor recognition effect under multipath fading channels,eight common digital modulation signals of 4ASK,8ASK,2FSK,4FSK,BPSK,QPSK,16 QAM,and 64 QAM are used as research objects.A method of modulation recognition based on multiple features combining cumulant and cyclic spectrum is proposed.In order to overcome the influence of multipath interference,and construct the structural stable performance and the feature parameters with high recognition rate,this paper uses the same order cumulants to construct five new cumulant features as the identification feature parameters,and deduces the anti-multipath performance of each order cumulant theoretically.Since the cumulant features of 2FSK and 4FSK are basically the same and cannot be completely recognized,the cyclic spectrum features are introduced to further improve the signal recognition performance.The projection cross section of the cyclic spectrum on the cyclic frequency axis and the frequency axis are extracted respectively.The two correlation coefficients of the cyclic spectrum on the dual-frequency axis section and the projection section are extracted as the recognition features.At the same time,BP neural network is used as a classifier for recognition to avoid the lack of empirical threshold.The experimental results show that the feature-based recognition algorithm proposed in this paper has better anti-multipath performance,and can achieve better recognition results under low signal-to-noise ratio.(2)Aiming at the problem of multi-user aliasing signal recognition in nonorthogonal multiple-access systems,a modulation recognition method based on convolutional neural network and joint constellation density map is proposed.Because the coincidence of constellation points on the original constellation map cannot effectively indicate the density,which leads to the loss of some information.To overcome this problem,this paper constructs a joint constellation density map of aliased signals and uses it as the input of a convolutional neural network.Convolutional neural network,as a novel algorithm in deep learning,shows good superiority in image recognition.In order to prevent the model from overfitting,speed up the convergence speed,and make the model with better generalization performance,the algorithm of Batch Normalization is added to the network for optimization.The results show the feasibility and effectiveness of the classification algorithm,and the recognition performance is good.