Modulation Recognition is one of the most important process carrying out at the output stage of communication systems. In this thesis study, an automatic modulation recognition system is developed to limited numbers of modulation formats exposed to white gaussian channel. In this modulation recognition system continuous wavelet transform and higher order cumulants are used as a feature extraction methods. Also, In this proposed algorithm, a three-level multilayer perceptron neural network was applied as a classifier. As a result of this thesis study, the proposed algorithm seemed to have better results in comparison to the other algorithms located in literature section. After that, in this thesis study, effect of the number of input samples on the performance of classification was investigated. The results of this investigation indicate that there is a direct relevance between the number of input sampels and classification performance. Namely, increasing the number of input sampels leads to an increase of operation in classification process. But increasing the number of input sampels lead to increasing classification time too. In this study, because of selected feature extraction methods which are resistant against noise impact, the value of extracted feature were stable partly. In this thesis instead of using all elements of extracted feature vektor we just used a part of them in the training step. Because our extracted features were reasonably stable in front of signal noise rate. So classification time is reduced subsequently.
Key Words: Automatic Modulation Recognation, Artifical Neural Networks, Support Vector Machine Algorithm, Continuous Wavelete Transform, Higher Order Cumulants.