Dynamic Convolutional Neural Network for Wireless Interference Identification
H. Yan, Gongpu Wang, Feifei Gao, Wanmai Yuan
Abstract
Deep learning has achieved satisfactory performance in the field of wireless interference identification (WII). However, the existing WII methods are all based on static neural networks, which are difficult to capture the changes of interference-to-noise ratio (INR) without increasing the computational complexity and storage overhead. In this letter, we propose a dynamic convolution based WII scheme that can quickly adapt to the changes of INR. Moreover, we apply group convolution and local importance-based pooling to further improve the adaptability of the proposed scheme as well as to reduce the storage overhead. Simulation results show that the proposed scheme boosts the recognition performance compared with the existing static WII methods.