Research On Modulation Format Recognition Based On Neural Networ

Modulation format identification is defined as that the receiver recognizes the modulation format of the received signal.It is very important in both civil and military systems.With the development of optical communication network,the high order modulation format is more widely used in order to improve the spectrum utilization.Therefore,recognition technology of signal modulation format has become a hot topic.At the same time,neural network technology has been widely used in the field of modulation format recognition because of its powerful learning ability.Therefore,the modulation identification technology based on neural network is introduced into the field of optical communication,and the advantages of the two can be combined to achieve better results.Based on neural network,this thesis focuses on photonic assisted modulation format recognition.Firstly,realizing the transmission and processing of modulated signal based on photonics method.Then the signal is preprocessed by digital ways to further highlight the signal characteristics and reduce the impact of noise,further enhance the anti-interference ability of signal.Finally,different neural networks are used to classify the signals and realize automatic recognition of various modulated signals.Firstly,a feature extraction system of modulated signal based on photonics is proposed.After the signal is modulated to the optical carrier,it is divided into two channels through the optical coupler,one of which delays the signal by 1bit.The envelope interference of two optical signals is realized in the coupler,and the frequency information of the signal is converted to the amplitude,so that the statistical characteristics of the amplitude of the signal itself can be used as the signal feature to realize the feature extraction of the modulated signal.Then,the obtained signals are preprocessed by Hilbert transform,smooth filtering,amplitude homogenization.further highlight the characteristics between different modulated signals,and the statistical amplitude histogram is obtained as characteristic parameters.In the Back Propagation Neural Network(BPNN),the database built by using 5-point smoothing filter and 7-point smoothing filter is identified respectively.The effect of smoothing filter points is compared.The convolutional neural network(CNN)and deep neural network(DNN)were used to identify binary amplitude keying(2ASK)signals,binary frequency shift keying(2FSK)signals,binary phase shift keying(2PSK)signals and orthogonal amplitude modulation(16QAM)signals when the signal-to-noise ratio(SNR)was from-1d B to15 d B.The classification accuracy can reach at 99.89%.The recognition accuracy of2 ASK signal and 2FSK signal is 100%.Four neural networks(BPNN,CNN+DNN,LSTM and CNN+LSTM)were constructed respectively,and automatic format recognition was realized for seven modulated signals(2ASK,4ASK,2FSK,4FSK,2PSK,4PSK and 16QAM).The classification results show that the recognition accuracy of the four networks is higher than 98.40% when the SNR is between-1d B and 15 d B.Therefore,the related neural network modulation format recognition system can realize the recognition of various modulated signals,with excellent performance.