基于残差神经网络和注意力机制的频谱感知方法

Spectrum Sensing Method Based on Residual Neural Network and Attention Mechanism

随着通信技术的发展,频谱感知技术已经成为解决频谱资源稀缺的重要解决手段之一。针对传统的频谱感知方法在低信噪比(Signal to Noise Ratio,SNR)下准确率较低的问题,提出一种基于残差神经网络和注意力机制相结合的正交频分复用(Orthogonal Frequency Division Multiplexing,OFDM)频谱感知方法。将频谱感知问题转化为图像二分类任务。通过分析OFDM信号的循环自相关特征,将其灰度处理以生成循环自相关灰度图像。利用改进后的残差神经网络进行训练,提取这些灰度图像的深层特征,使用测试数据验证所得到的频谱感知模型。仿真实验结果表明,在低SNR条件下,所提方法表现出更出色的频谱感知性能,优于传统频谱感知技术。

With the development of communication technology,spectrum sensing technology has become one of the important solutions to solve the scarcity of spectrum resources.For the low accuracy of traditional spectrum sensing methods under low Signal to Noise Ratio(SNR),an Orthogonal Frequency Division Multiplexing(OFDM)spectrum sensing method based on the combination of residual neural network and attention mechanism is proposed.The spectrum sensing problem is transformed into a binary image classification task.The cyclic autocorrelation grayscale images are produced by analyzing the cyclic autocorrelation characteristics of OFDM signals to perform grayscale processing.Subsequently,deep features from the grayscale images are extracted through training an improved residual neural network,and the resulting spectrum sensing model is validated using a test dataset.The simulation experiments show that the proposed method exhibits superior spectrum sensing performance under low SNR conditions,surpassing conventional spectrum sensing techniques.