Development and comparative study of effects of training algorithms on performance of artificial neural network based analog and digital automatic modulation recognition
Journal of Engineering Science and Technology Review
Centre for Telecommunications Access and Services, School of Electrical and Information Engineering, University of The Witwatersrand, Private Bag 3, Wits, Johannesburg, South Africa; Department of Electrical and Electronics Engineering, School of Engineering and Technology, Federal University of Technology, P.M.B. 704, Ondo State, Akure, Nigeria
This paper proposes two new classifiers that automatically recognise twelve combined analog and digital modulated signals without any a priori knowledge of the modulation schemes and the modulation parameters. The classifiers are developed using pattern recognition approach. Feature keys extracted from the instantaneous amplitude, instantaneous phase and the spectrum symmetry of the simulated signals are used as inputs to the artificial neural network employed in developing the classifiers. The two developed classifiers are trained using scaled conjugate gradient (SCG) and conjugate gradient (CONJGRAD) training algorithms. Sample results of the two classifiers show good success recognition performance with an average overall recognition rate above 99.50% at signal-to-noise ratio (SNR) value from 0 dB and above with the two training algorithms employed and an average overall recognition rate slightly above 99.00% and 96.40% respectively at-5 dB SNR value for SCG and CONJGRAD training algorithms. The comparative performance evaluation of the two developed classifiers using the two training algorithms shows that the two training algorithms have different effects on both the response rate and efficiency of the two developed artificial neural networks classifiers. In addition, the result of the performance evaluation carried out on the overall success recognition rates between the two developed classifiers in this study using pattern recognition approach with the two training algorithms and one reported classifier in surveyed literature using decision-theoretic approach shows that the classifiers developed in this study perform favourably with regard to accuracy and performance probability as compared to classifier presented in previous study. © 2015 Kavala Institute of Technology.
Algorithms; Amplitude modulation; Conjugate gradient method; Neural networks; Pattern recognition; Signal to noise ratio; Surveys; Automatic modulation recognition; Comparative performance; Decision theoretic approach; Development approach; Instantaneous amplitude; Modulation recognition; Neural networks classifiers; Scaled conjugate gradients; Modulation