Research And Implementation Of Modulation Recognition Of Digital Communication Signal

With the rapid development and a large number of applications of wireless communication technology,the wireless channel environment is becoming more and more complex.As one of the key technologies of non-cooperative communication,modulation recognition has been widely used in signal detection,interference analysis and spectrum monitoring in military and civil fields.With the development of modulation technology and the complexity of channel environment,pursued features easier to be distinguished and better-performing classifiers have always been pursued by researches.First of all,ten modulation modes are researched,including 2ASK,4ASK,8ASK,2PSK,4PSK,8PSK,2FSK,16 QAM,32QAM and 64 QAM.The principle of higher-order cumulant,instantaneous feature extraction and cyclic spectrum theory are introduced.Based on the cyclostationary characteristics of digital modulation signals,the existential detection of signals is designed based on the value of normalized spectral lines at the zero-cyclic frequency of the cyclic spectrum.The simulation results show that the probability of correct detection of all modulation modes in this paper can reach more than 90% when the signal-to-noise ratio is greater than-10 d B.Secondly,based on the difference in amplitude and phase of different modulation signals,six characteristic parameters are constructed from the instantaneous amplitude,instantaneous phase,and higher-order cumulants to distinguish the ten modulation modes.As it is difficult to distinguish 4PSK and 8PSK from instantaneous information,a new feature,the maximum value of segmented phase mean,based on the instantaneous phase information,is proposed on the premise that the symbol rate is known.The experimental results show that the accuracy of the two signals can reach more than 90% when the signal-to-noise ratio is not less than 4d B.The simulation results show that when the decision tree classifier is used,the correct recognition rate of 8ASK can reach more than 90% when the signal-to-noise rate is not less than 11 d B.To achieve the same correct recognition rate,the signal-to-noise rate that 64 QAM needed is not less than 13 d B.The rest of the modulation methods can reach 90% of the correct recognition rate when the signal-to-noise rate is not less than 4d B.Then,on the basis of the support vector machine classifiers in machine learning and three commonly used classification strategy of multi-classification to two-classification,a multi-classification algorithm combining support vector machine and one-to-one strategy to realize 10 modulation methods is emphatically studied.Aiming at the shortcoming that the one-to-one classification strategy cannot make full use of all the information of the basic classifier,such as the correct detection probability and the probability of misclassification,an improved multiclassification algorithm based on the confidence weight coefficient matrix is proposed.The simulation results show that the average recognition rate of multi-classification algorithm based on the confidence weight coefficient matrix is about 4% higher than that of support vector machine using the one-to-one classification strategy when the signal-to-noise ratio is between-10 d B and-6d B.Finally,for all the instantaneous features used in this paper,the algorithm implementation and verification are completed through the DSP hardware processing platform,and the accuracy of the classification is tested.The test results show that the calculation results of the hardware processing platform and the software simulation are basically consistent with each other within the allowable error,which proves the feasibility of the algorithm in practical applications.