Classificação automática de modulação baseada em aprendizagem discriminativa

Abstract

The main purpose of the algorithms for automatic modulation classification in an in- telligent receiver for cognitive radio or military applications is to identify the modulation of a signal so that the receiver can subsequently carry out the process of demodulation and then processing the received information. The lack of knowledge of parameters such as, for exam- ple, signal power, frequency and phase of carrier and timing; added to the channel effects such as multipath fading, make the automatic modulation classification problem challenger and fairly investigated currently. In this context, this thesis presents an investigation of automatic modulation classification algorithms based on discriminative learning and evaluates their per- formance in different usage scenarios. The thesis presents a new classification proposal based on front end called HISTO. Classification is performed on the received symbols, where the magnitude and phase histogram’s are computed. The results obtained using the proposed front end and the Support Vector Machine algorithm (SVM) are compared with other tech- niques. In the literature, some of the proposed algorithms are evaluated assuming knowledge of some modulation parameters and using different scenarios, which makes difficult to compare the algorithms. In this thesis, algorithms representative of the prior art are compared with the proposed ones in a systematic and uniform manner, including non-ideal scenarios such as with phase jitter and frequency offset. The simulation results show that the proposed classification based on histograms is effective with a relatively low computational cost.