采用深度图像推断的认知无线电频谱预测算法

Cognitive radio spectrum prediction algorithmbased on depth image prediction

摘要:

针对认知无线电频谱预测效率不高的问题,提出了一种采用深度图像推断的频谱预测算法。该算法将序列预测问题转化为图像推断问题,构建深度图像推断网络实现无线电频谱预测。首先,对历史频谱数据进行预处理,提取频谱数据的变化特征;其次,使用双支路多层并联卷积神经网络提取数据的深度特征,经池化、合并操作输出多层次特征信息;最后,融合不同层次提取的特征信息,实现频谱数据的预测和生成。在真实频谱数据的多个频段对算法性能进行验证,实验结果表明算法能够有效地实现电磁频谱数据的预测,具有预测精度高的特点。

Abstract:

To solve the problem of low efficiency in spectrum prediction of cognitive radio, anefficient spectrum prediction algorithm based on depth image inference was proposed. In thispaper, the problem of sequence prediction was transformed into that of the image inference,and the spectrum prediction of image processing was realized by using deep neural network.Firstly, the historical spectrum data were preprocessed to extract the variation characteristics;Secondly, the multi-level parallel convolution layer was used to extract the features ofthe data, and the feature was output through pooling and merging operations; Finally, thefeature information extracted from different layers was fused to achieve the prediction andgeneration of spectrum data. The network could extract high-dimensional features of depthdata and realize the prediction and generation of spectrum data. The proposed algorithm wasverified in multiple frequency bands, and the results have shown that the proposed algorithmcan effectively predict spectrum data with high prediction accuracy.