Effect of training algorithms on performance of a developed automatic modulation classification using artificial neural network
Jide Julius Popoola, Rex van Olst · 2013
As classification has become one of the active areas of applications for artificial neural networks, the major objective of this paper was to verify the impact of training algorithms on classifiers developed using an artificial neural network. This paper presents an algorithm for classifying eight digitally modulated signals using a feature-based approach and a pattern recognition method. The developed automatic modulation classification was trained using two training algorithms often used for supervised neural networks. The performance of the developed automatic modulation classification classifier was evaluated and compared using the two training algorithms. The overall performance evaluation of the classifier using the two training algorithms shows that the developed classifier could successfully classify the eight modulated schemes considered with an average success rate above 97.0% irrespective of the signal-to-noise value. The results of the study also show that training algorithms have an impact on the performance of an artificial neural network classifier. In addition, the result of the comparative analysis carried out between the classifier reported in this paper and the one in surveyed literature shows that the signal recognition rate of this classifier is accurate and reliable.