Time-Frequency Component-Aware Convolutional Neural Network for Wireless Interference Classification

• DOI: 10.1109/lwc.2022.3204756
• Publication Date: 2022-09-06T19:25:42Z
• Authors: Pengyu Wang; Yufan Cheng; Gaoyang Shang; Jun Wang; Shaoqian Li
• Keywords: Convolution (computer science); Component (thermodynamics)

In wireless communication systems, interference classification (WIC) is considered as one of the most effective technologies to address challenges brought by electromagnetic in military and civilian scenarios. Recently, deep learning (DL) based methods have dominated progress field WIC. However, existing do not consider redundancy input samples, nor ability adaptively allocate computational resources conditioned on inputs. To this end, we propose time-frequency component-aware convolutional neural network (TFCCNN), it allows convolution calculation be performed only at locations where components or important parts exist image signals, leading reduce superfluous computation. Furthermore, further complexity, introduce a novel adaptive forward propagation (AFP) algorithm, can determine depth according difficulty sample during inference. Experimental results demonstrate that proposed method reduces complexity about 75% when recognition accuracy slightly improved compared traditional CNNs.