A Learnable Distortion Correction Module for Modulation Recognition

Abstract

Modulation recognition is a challenging task while performing spectrum sensing in cognitive radio. Recently, deep learning techniques, such as convolutional neural networks (CNNs) have been shown to achieve state-of-the-art accuracy for modulation recognition. However, CNNs are not explicitly designed to undo distortions caused by wireless channels. To improve the accuracy of CNN-based modulation recognition schemes, we propose a signal distortion correction module (CM). The proposed CM is also based on a neural network that can be thought of as an estimator of carrier frequency and phase offset introduced by the channel. The CM output is used to shift the signal frequency and phase before modulation recognition and is differentiable with respect to its weights. This allows the CM to be co-trained end-to-end in tandem with the CNN used for modulation recognition. For supervision, only the modulation scheme label is used and the knowledge of true frequency or phase offset is not required for co-training the combined network (CM+CNN).