融合多模态数据的移动物联网频谱智能感知算法OA

For the problem of low spectrum utilization in the complex and changeable mobile Internet of Things(MIoT)environment,an intelligent spectrum sensing algorithm integrating multimodal data is proposed.Firstly,multiple secondary user signals obtained by the system are used to construct time series data through collaboration,and the sample covariance matrix is calculated to obtain the corresponding image data.Time series data and image data constitute multimodal data,which not only provides a more comprehensive temporal and spatial perspective,but also improves the diversity of features.Then,a parallel dual-branch Time-ImageNet network is constructed based on multi-modal data.The TimeNet branch network consists of multi-layer Conv1D and multi-layer Transformer to extract key time features from time series data.The ImageNet branch is based on the FasterNet network.On the one hand,it reduces the complexity by optimizing the network structure.On the other hand,it uses the channel shuffle operation to perform cross-group channel information interaction on the feature map to break the grouping boundary.Finally,the output time series features are fused with the corresponding image features,and input into the modal fusion network for spectrum sensing to determine whether the spectrum is available.Based on the N-Nakagami channel conditions,in the low signal-to-noise ratio range of-21 dB to-9 dB,compared with other algorithms,when using QPSK modulation,the Time-ImageNet's detection probability is increased by 35%,and the false alarm probability is reduced by 38%.