Modulation Recognition Based On Spectral Correlation And Neural Network

The purpose of modulation recognition of communication signals is to identify the correct modulation type of communication signals and provide the signal information that further signal processing required. With the rapid development of communication technology in today's society, the wireless communication environmentis getting more and more complicated. In order to improve bandwidth utilization, new modulation types come out one after other, leading to the coexistence of multiple communication systems. This makes the modulation recognition in the field of non-cooperative communication more and more important. Integrating the advantages of cyclic spectrum and the advantages of neural network in classification, this dissertation studied the modulation recognition for digital modulated signals. The main contents include: to propose characteristic parameters for each modulated signal according to their cycle spectral, to improve the neural network training algorithm and its simulation, to design the digital modulation recognition system and its simulation.First, this dissertation introduced the cyclic spectral theory, analyze the features of cyclic spectral and its noise suppression performance, presented the basics of neural networks, discussed BP algorithm and the points to design BP networksSecondly, this dissertation analyzed the cyclic spectrals of 2ASK, 4ASK, 4PSK, 2FSK, 4FSK and MSK modulated signals, constituted the feature vector space for these modulated signals accoding to the new characteristic parameters that cyclic spectral theory proposed and the amplitude characteristics of these modulated signals.Finally, the neural network recognition system was constructed. The number of neurons in different hidden layers were simulated and compared. Standard BP algorithm and its improved algorithms were simulated and compared. This dissertation made an improvement on the additional momentum method. Simulation results showed that this mothod speeded up convergence speed, reduced the recognition time, improved the system performance. The correct recognition rate is more than 90% when the SNR is greater than 5dB.