Blind Channel Identification Aided Generalized Automatic Modulation Recognition Based on Deep Learning

Hao Gu,Sheng Hong,Yu Wang,Guan Gui

doi:10.1109/access.2019.2934354

Abstract

Automatic modulation recognition (AMR) plays an important role in cognitive radio (CR), which relies on AMR responding to changes in the surrounding environment and then adjust strategies simultaneously. Deep learning based reliable AMR method have been developed in recent years. However, all of their AMR training models are considered in a specialized channel rather than generalized channel. Hence, these AMR methods are hard to be applied in general scenarios. In this paper, we propose a blind channel identification (BCI) aided generalized AMR (GenAMR) method based on deep learning which is conducted by two independent convolutional neural networks (CNNs). The first CNN is trained on in-phase and quadrature (IQ) sampling signals, which is utilized to distinguish channel categories like BCI behaviors. The second CNN is trained by line of sight (LOS) model and non-line of sight (NLOS) model, respectively. Simulation results confirm that our proposed generalized AMR method is significantly better than conventional method.

Highlights

Modulation identification techniques have many potential applications in the field of wireless communications, typically in cognitive radio systems [1], [2]
FORMER convolutional neural networks (CNNs) PERFORMANCE COMPARISONS In this paper, the first CNN we trained is used to identify the channel type of the unknown input signal sampled by in-phase and quadrature (IQ)
We have adopted 5 deep learning or machine learning algorithms to compare the classification accuracy. They are CNN, CNN without batch normalization (BN) layer, RNN, higher order cumulants (HOC) feature extraction followed by Deep Neural Network (DNN) and HOC feature extraction followed by the Random Forest (RF) classification algorithm

Summary

Introduction

Modulation identification techniques have many potential applications in the field of wireless communications, typically in cognitive radio systems [1], [2]. INDEX TERMS Automatic modulation recognition (AMR), deep learning, convolutional neural network (CNN), in-phase and quadrature (IQ) samples, blind channel identification. There are some scholars proposing AMR based on CNN to identify multiple modulation signals. The authors proposed different AMR systems to classify QAM and PSK modulation signals for various channels [18].

Results

Conclusion