MACHINE LEARNING IMPLEMENTATION FOR SOFTWAREDEFINED RADIO-BASED MODULATION CLASSIFIER

• Main Author: Damario Lukito, William
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55549
• Institution: Institut Teknologi Bandung

Automatic modulation classification becomes an interesting problem in communication systems theory, since its useful application in real-world problems. However, in nowadays technological era, the implementation of modulation classification basically is still based on general statistical methods, even though machine learning is frequently used to solve classification problems. Therefore, an idea to implement machine learning in modulation classification problems came up. This research will discuss about the development of machine learning-based modulation classifier prototype. Initially, in order to enable modulation classification, a receiver obviously need a classification-rule model. The classification-rule model will be obtained using a machine learning algorithm, specifically SVM (support vector machine) and implemented using MATLAB's classification learner. For the purpose to acquire a classification-rule model, some classification features will be needed as the predictors of machine learning algorithm. Hence, wavelet transform-based, spectral-based, and higher order statistics-based features are being used as the input predictors to machine learning algorithm. In this research, the modulation types are limited to BPSK, QPSK, 8- PSK, 16-QAM, BFSK, and 8-PAM. After SVM has performed its training, we obtained a classification-rule model with 91.4% of accuracy without any optimization. This model will be used later as a classifier function in the receiver algorithm. The receiver hardware is implemented using an ADALM-PLUTO SDR (software-defined radio).