Study On Modulation Signal Detection Method Based On Cyclic Spectral Correlation

With the continuous development and innovation of modern electronic communication technology, communication signal modulation automatic identification technology has been widely used in various fields. As an important segment between the signal acquisition and demodulation, whether in military or civilian areas, signal modulation automatic identification technology plays an important role. The research in software-defined radio and cognitive radio gives it a new significance. How to effectively improve the accuracy of automatic recognition and reduce the computational complexity has become the main research directions in the field. In this paper, based on the help of theoretical analysis and computer simulation, we have researched the cyclic-spectrum estimation algorithm, extracted characteristic parameters and designed the classifier on the base of Radial Basis Function, at last put forward several improved algorithms.Firstly, we have analyzed the cyclic-spectrum estimation algorithms from the basic principle of signal cyclic-spectrum. The noise issues were improved by using the wavelet transform theory of smoothing algorithm in the frequency domain, and several circulation common digital modulation signal spectral correlation characteristics were analyzed.Different modulation signals have different spectral correlation characteristics, five classification features of the modulating signals with great distinction degree and stable antinoise performance were extracted through the analysis of characteristics of the a and f profiles of cyclic-spectrum including the number of spectrum lines, the relative strength between spectrum lines, the normalized area of f profiles, zero center normalized instantaneous amplitude and so on.Finally, through analyzing the basic theory of the RBF neural network, the classifier was designed and constructed on the basis of the RBF neural network, which can automatically recognize and classify the modulation signal. Some improved algorithms were proposed for the classifier slow convergence and low correlation in identification accuracy. The effectiveness and stability of the improved algorithm was proved by the computer simulation.