Deep Learning based Automatic Modulation Classification for Varying SNR Environment

Abstract

Automatic modulation classification (AMC) is a crucial task for various communications applications. Deep learning (DL) based classifier is emerging as a prevalent choice for AMC. Previous research on DL based AMC usually assumes an environment of fixed signal to noise ratio (SNR). This paper considers DL based AMC for varying SNR environment. Two algorithms, including M2M4-aided algorithm and multi-label DL based algorithm, are proposed to combat the varying SNR. The former utilizes an M2M4 estimator to estimate SNR, according to which a proper trained DL model can be selected for AMC. The latter exploits multi-label DL to train a model, with which SNR scenario and modulation type can be inferred simultaneously. Experiment results show that the performance of both algorithms is fairly close to that of DL based AMC under fixed SNR environment.