AMRnet: A Real-Time Automatic Modulation Recognition Network for Wireless Communication System

Abstract

In recent years, deep learning has brought new opportunities and challenges to the wireless communication field, owing to its expressive capacity and convenient optimization capability. In this paper, we propose a real-time Automatic Modulation Recognition Network (AMRnet) for signal transmission process, which can be easily applied to mobile and embedded wireless communication applications (e.g., The GNU Radio system based on USRP platform). Our proposed AMRnet consists of a series of convolutional neural networks based on the design theory of lightweight networks. More specifically, depthwise separable convolution structure is used to reduce computational complexity of the whole network. Residual block is included to improve the robustness for the simple backbone network. In addition, channel attention mechanism is added to lay more emphasis on the channel-wise information due to the unique data futures of the I/Q signal. We present numerous experiments on resource and accuracy tradeoffs and show strong performance compared to a number of state-of-the-art methods. This work will have a great impact on the future communication systems.