Ensemble Wrapper Subsampling for Deep Modulation Classification

Subsampling of received wireless signals is important for relaxing hardware requirements as well the computational cost signal processing algorithms that rely on output samples. We propose a subsampling technique to facilitate use deep learning automatic modulation classification in communication systems. Unlike traditional approaches pre-designed strategies are solely based expert knowledge, proposed data-driven strategy employs neural network architectures simulate effect removing candidate combinations samples from each training input vector, manner inspired by how wrapper feature selection models work. The subsampled data then processed another classifier recognizes considered 10 types. show not only introduces drastic reduction time, but can also improve accuracy dataset. An herein exploiting transferability property networks avoid retraining and obtain superior performance through an ensemble wrappers over possible relying any one them.