kumoh national institute of technology
Networked Systems Lab.

Godwin Brown Tunze, Thien Huynh-The, Jae-Min Lee, and Dong-Seong Kim, Sparsely Connected CNN for Efficient Automatic Modulation Recognition, IEEE Transaction on Vehicular Technology, 2020. (S)
By : Godwin
Date : 2020-07-01
Views : 154

This paper proposes a convolutional neural network (CNN) called SCGNet for low-complexity and robust modulation recognition in intelligent communication receivers. Principally, the network combines two types of sparse convolutional layers—depthwise and grouped in an architecture—to achieve a few trainable parameters while maintaining high recognition accuracy. The network architecture leverages sparsely connected convolutional layers in three principal modules: speed–accuracy tradeoff (SAT), deep feature extraction and processing (DFEP), and generic feature extraction (GFE) data pre-processing module. SAT deploys depthwise convolutional layers to extract features generated by GFE, achieving a tradeoff between complexity and accuracy. In addition to SAT, DFEP employs a cascade of grouped convolutional layers for mining more discriminative features from SAT via a multilayer transformation module. The aim of the cascade is to prevent a loss of essential details of the signal as the network becomes deeper. Additionally, skip connections are deployed between sub-blocks within SAT and DFEP to allow inter-module feature sharing, for handling inter-block features loss. Experimental results indicate that our architecture achieves an overall recognition accuracy of >94.39% at a signal-to-noise ratio of 20 dB on the RadioML2018.10A dataset.

Paper History
Submitted: 01/07/2020
Journal: IEEE Transactions on Vehicular Technology
Date Received: 07/09/2020

Revision
Submitted: 06/10/2020
Current status: With Associate Editor