基于残差注意力密集网络的协作频谱感知方法  

Cooperative spectrum sensing method based on residual attention dense network

作　　者：王安义[1] 朱涛 龚健超 WANG Anyi;ZHU Tao;GONG Jianchao(School of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710600,China)

机构地区：[1]西安科技大学通信与信息工程学院,陕西西安710600

出　　处：《电信科学》2025年第2期84-94,共11页Telecommunications Science

基　　金：国家自然科学基金资助项目(No.62471384)。

摘　　要：针对基于卷积神经网络(convolutional neural network,CNN)的协作频谱感知算法存在的网络结构简单、特征提取能力不足和感知性能下降等问题,提出了一种基于残差注意力密集网络(residual attention dense network,RADN)的协作频谱感知算法。该算法通过改进基础残差块,从感受野、通道和空间3个维度引入注意力机制,结合残差连接和密集连接,构建了强大的深层特征提取结构——密集残差(residual in dense,RID),显著提升了网络的特征提取能力和频谱感知性能。实验结果表明,相较于传统深度学习方法,RADN算法在低信噪比(signal-to-noise ratio,SNR)条件下表现出显著的性能提升。该方法不仅能够适应多种调制方式,还具备较高的检测概率和良好的鲁棒性。To address the limitations of cooperative spectrum sensing algorithms based on convolutional neural network(CNN),including simple network structures,insufficient feature extraction,and reduced sensing performance,a cooperative spectrum sensing algorithm based on residual attention dense network(RADN)was proposed.The basic residual block was enhanced and attention mechanisms across receptive field,channel,and spatial dimensions were introduced.By integrating residual and dense connections,a powerful deep feature extraction framework was formed,which was termed residual in dense(RID),its feature extraction and sensing performance had been significantly boosted.Experimental results show that under low signal-to-noise ratio(SNR)conditions,the RADN algorithm outperforms traditional deep learning methods,adapting well to various modulation schemes and achieving high detection probability and robustness.

关 键 词：协作频谱感知 卷积神经网络 注意力机制 密集连接 残差连接 

分 类 号：TN911[电子电信—通信与信息系统]