Classification of Modulation Error Rate Measurement using Convolutional Neural Networks in ISDB-T

Abstract

This article proposes to perform detection and recognition of the modulations of the ISDB-T system and its measurement of the Modulation Error Rate (MER) through deep learning. Initially, a data set of 30,000 constellation images of the QPSK, 16-QAM, and 64-QAM modulations with different identified MER values in dB was generated. These data sets are used to train and validate a convolutional neural network based on the transfer learning in AlexNet network architecture, destined to recognize different types of images. The validation results and a test set obtained from the same database were highly satisfactory. Most of them approach 100% accuracy in the classification, which showed a good detection of modulation and especially discrimination of the MER value when evaluating constellations. ISDB-T signals transmitted by broadcast in the city of Quito-Ecuador and by the laboratory were also evaluated. A professional ISDB-T analyzer and a receiver designed with software-defined radio (SDR) ADALM-PLUTO were used for reception. The results in the receiving equipment show an accuracy of 100% of the detected modulation and very close values between the measured MER values and those obtained by the neural network.