How to choose a neural network architecture? – A modulation classification example
Authors:
- 1 September 2020
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 413-417
Abstract
Which neural network architecture should be used for my problem? This is a common question that is encountered nowadays. Having searched a slew of papers that have been published over the last few years in the cross domain of machine learning and wireless communications, the authors found that several researchers working in this multi-disciplinary field continue to have the same question. In this regard, we make an attempt to provide a guide for choosing neural networks using an example application from the field of wireless communications, specifically we consider modulation classification. While deep learning was used to address modulation classification quite extensively using real world data, none of these papers give intuition about the neural network architectures that must be chosen to get good classification performance. During our study and experiments, we realized that this simple example with simple wireless channel models can be used as a reference to understand how to choose the appropriate deep learning models, specifically neural network models, based on the system model for the problem under consideration. In this paper, we provide numerical results to support the intuition that arises for various cases.Keywords
This publication has 11 references indexed in Scilit:
- Deep Learning in Mobile and Wireless Networking: A SurveyIEEE Communications Surveys & Tutorials, 2019
- An Introduction to Deep Learning for the Physical LayerIEEE Transactions on Cognitive Communications and Networking, 2017
- Deep neural network architectures for modulation classification2017
- Densely Connected Convolutional Networks2017
- Learning to decode linear codes using deep learning2017
- Convolutional Radio Modulation Recognition Networks2016
- Deep Residual Learning for Image Recognition2016
- Deep learningNature, 2015
- A Blind Pre-Processor for Modulation Classification Applications in Frequency-Selective Non-Gaussian ChannelsIEEE Transactions on Communications, 2014
- Blind identification of ARMA channels with periodically modulated inputsIEEE Transactions on Signal Processing, 1998