Research On Modulation Recognition For Communication Signals Based On Deep Learning

Modulation recognition is to identify the received unknown signal modulation format under the condition of limited or no prior information.Modulation recognition is widely used in military and civilian fields.In the military field,for example,modulation recognition is generally needed to identify the modulation format of non-cooperative signal in ECM,which lays a foundation for the subsequent acquisition of other parameters of the signal;after obtaining the detailed parameters of the signal,the pertinent interference measures can be taken or the signal can be decomposed to get information that it transmits.In the civil field,the monitoring of electromagnetic spectrum needs the support of modulation recognition.For example,when the electromagnetic spectrum is illegally occupied by the unknown signal,modulation recognition is needed to identify the modulation format of the unknown signal to further identify the unknown signal.According to the application of deep learning algorithm,the modulation recognition algorithms can be divided into two categories: classical modulation recognition algorithm and intelligent modulation recognition algorithm based on deep learning.After half a century of development,the classical modulation recognition algorithms have met some bottlenecks,such as the difficulty of recognizing complex modulation format and the difficulty of adapting to complex electromagnetic environment.At the beginning of its development,there are still some loopholes,such as the weak generalization ability of the algorithm,the lack of selecting pre-processing mechanism and so on.This paper mainly focuses on the problems in the modulation recognition.For the problem that the complex modulation format is difficult to recognize and the complex electromagnetic environment is difficult to adapt,this paper uses the deep learning algorithm,using its excellent feature extraction ability,to realize the recognition feature extraction of the complex modulation format and the extraction of the anti-jamming feature;for the problem that the generalization ability of the algorithm is weak,this paper starts from the data and the network,to reduce the influence of parameters on the data and to enhance the generalization ability of the network.For the problem that lacks of selecting pre-processing mechanism,this paper uses the correlation among network input,network first layer output and network output to establish the pre-processing selection mechanism.Based on the above research ideas,the main achievements obtained in this paper are as follows:1.We propose a preprocessing selection algorithm based on image matching.The preprocessing selection algorithm based on image matching solves the problem that there is no certain guiding principle for the selection of signal preprocessing based for the algorithm of modulation recognition based on deep learning,and establishes a preprocessing selection mechanism.Through pre-processing selection,unnecessary pre-processing can be effectively eliminated,thus effectively reducing the computational complexity of modulation recognition algorithm based on depth learning,less redundant information input by the algorithm.2.A modulation recognition algorithm based on improved CLDNN is proposed.The modulation recognition algorithm based on improved CLDNN has a strong generalization ability by anti-over-fitting,and it can still have a high recognition rate when the training data and testing data parameters differ greatly;at the same time,when the signal-to-noise ratio is typical,CPM,a complex modulation format,can be identified accurately,which proves its ability to recognize complex modulation formats.3.A modulation recognition algorithm based on improved SparseNet is proposed.The modulation recognition algorithm based on improved SparseNet achieves high accuracy modulation recognition under low signal-to-noise ratio by choosing some preprocessing with certain anti-noise ability and improved SparseNet.Moreover,the modulation recognition algorithm based on improved SparseNet can realize the complexity of CPM under low signal-to-noise ratio.This proves its ability to identify complex modulation formats at low signal-to-noise ratio.