Spectrum Sensing Scheme Based on Federated Learning and Denoising Autoencoder
Yuebo Li, Xiaoyang Ren, Z G Liu, Xu Han, Junsheng Mu, Xiaojun Jing · 2023
Spectrum sensing is an important technology in cognitive radio networks, which can effectively utilize idle spectrum resources and improve spectrum utilization. However, traditional spectrum sensing methods often require all users' data to be centralized into a central server for processing, which brings data privacy and communication overhead issues. To solve these problems, this paper proposed a spectrum sensing method based on federated learning and denoising auto-encoder. This method takes advantage of the distributed learning advantages of federated learning, so that each user trains a spectrum sensing model locally, and then sends the model parameters to a central server for aggregation. In order to reduce the communication overhead, a denoising auto-encoder is introduced between each user and the central server for parameter compression and dimensionality reduction. Simulation results show that by incorporating joint learning into the spectrum sensing process, we achieve better accuracy compared to traditional non-joint techniques. The application of denoising auto-encoders plays a key role in reducing the impact of noise, resulting in highly reliable and robust spectral sensing.