Research On Automatic Modulation Recognition Of Communication Signals Based On Deep Learning

Modulation recognition is a process to identify the modulation scheme of the received signal,which is one of the crucial technologies in the field of cognitive radio.Modulation recognition plays a key role whether in the use of engineering or military applications.In terms of engineering applications,modulation recognition technology can improve spectrum utilization and complete more timely and efficient communications.In terms of military applications,modulation recognition can be used to obtain intelligence interception and electronic jamming.Due to the high cost and low efficiency of manual modulation recognition,automatic modulation recognition has drawn more and more attention.In recent years,with the rapid development of artificial intelligence,more and more approaches based on machine learning and deep learning have been studied to solve the problem of automatic modulation recognition.As a backbone of deep learning models,convolutional neural networks are widely used in the field of automatic modulation classification.However,we speculated that the received signal samples are not suitable for directly feeding into CNNs.On the one hand,the shape of the signal samples would greatly limit the structural of the CNN;on the other hand,feeding the signal samples into CNN directly is likely to cause a waste of features.To address these issues,we proposed a novel data preprocessing method to improve the recognition accuracy of the automatic modulation recognition systems based on CNN.We conducted multiple sets of comparative experiments,and the experimental results showed that the proposed preprocessing method improved the recognition accuracy of a modulation recognition model based on CNN proposed in recent years on the famous Radio ML2016.10 a dataset by approximately 7%.Subsequently,based on the analysis of CNN and the proposed data preprocessing method,we designed a neural network with residual blocks.The experimental results showed that the combination of the proposed preprocessing method and the designed neural network achieved the highest recognition accuracy rate of 93.72% on the Radio ML2016.10 a dataset,which outperforms the stat-of-the-art automatic modulation recognition classifiers.Finally,according to the test results on another dataset with different degrees of channel fading,the proposed preprocessing method and the designed automatic modulation recognition model showed great robustness.