Research On Modulation Recognition Of Digital Signal

With the rapid development of social informatization, modern communicationtechnology is also in the rapid change. The communication signal modulation recognition, asan important subject in the field of signal processing also has achieved rapid development.The developments of the signal recognition, signal detection, signal monitoring andinterference identification is direct impacted by signal modulation recognition technology.Therealization of modulation recognition with the development of the modulation signalautomatic identification technology has become more quickly and efficiently. So far,modulation recognition algorithms mainly include two types: based on the statistics theoryand based on the theory of ruling. In this dissertation, the algorithm based on the statisticalpattern recognition has been studied, and been applied the neural network to realizing theautomatic recognition of modulation signal. Then, in this paper, the constellation based onclustering of MQAM signals recognition method within the class also has been studied.In this dissertation, firstly the digital signal modulation mode recognition algorithm andits mathematical theory and basic knowledge has been introduced. In order to overcome thefrequency distortion problems, this dissertation using the method that combinated EMD andHilbert transform to extract the characteristic parameters of instantaneous signal, thenintroducing the wavelet threshold de-noising to reduce the noise that impacted on signalinstantaneous characteristic parameters. According to the signal instantaneous characteristicparameters, this paper selected five characteristic parameters and confirmed the decisionthreshold, then using the method of decision tree identified2ASK,4ASK,2PSK,4PSK,2FSK,4FSK and16QAM signals. Simulation results indicate that compared the improvedalgorithms with the general algorithm, under the same low SNR, signal recognition rateincreased by10%or so.Then according to the completed content, in this dissertation, BP neural network neuralnetwork has been used to recognize these seven signals. In order to speed up the networklearning, this paper designed the system of using of0.1and0.9goal matrix，compared to thecommon set of1s and0s goal matrix，the new goal matrix speed up the network learningspeed; for the slow convergence speed and local convergence of BP algorithm to theminimum, MLP using one of the fastest training algorithm of LM-BP optimization algorithmto network training, and then simulated the network and its algorithm and arrived at aconclusion. Simulation results indicates that using neural networks for signal recognition notonly can make the recognition process intelligent, but also can reduce the influence ofman-made factors to the recognition result, and in the same low SNR, the rate of automaticidentification recognition compared with the rate of general identification recognition can be increased.Finally, in this dissertation MQAM recognition within the class based on Hilberttransform according to the characteristics and the rolling characteristics of MQAMmodulation signal has been researched, and introduced the linear chirp Z transform to realizethe estimation of baud rate; then use constellation clustering algorithm to realize therecognition of the MQAM signals within the class, and then simulate the solution. Simulationresults indicates that as a result of clustering based on blind constellation is associated withthe estimation precision of baud rate, so using CZT algorithm improving the frequencyresolution to estimate the baud rate is closer to the of the ideal value, then identify the signal,in this case, the ratio of recognition can be over95%when SNR is over15dB.