基于注意力机制的混合CNN-BiLSTM低轨卫星信道预测算法

针对低轨道卫星信道质量变化迅速、信道参数“过时”的问题, 提出了一种基于注意力机制的卷积神经和双向长短时记忆神经网络(attention-convolutional neural network and bi-directional long-short term memory neural network, AT-CNN-BiLSTM)融合的信道预测方法。该方法由信号预处理、网络训练和信号预测3部分组成。首先在高斯白噪声条件下模拟室外卫星信号, 得到卫星信号的训练集和测试集; 然后将训练集输入构建的训练网络进行特征提取; 最后将测试数据输入网络进行预测分析。仿真结果表明, 在与其他4种人工智能方法的对比中, 所提出的混合神经网络能够在较快的收敛速度下达到较高的准确率(91.8%), 有效地缓解了低轨道卫星信道参数“过时”的现状, 对提升卫星通信质量和节省卫星信道资源有良好的改善作用。

Low-orbit satellite channel prediction algorithm based on the hybrid CNN-BiLSTM using attention mechanism

To deal with the problem of the low earth orbit (LEO) satellite rapidly changing channels' qualities and outdated parameters, a channels prediction algorithm attention-convolutional neural network and bi-directional long-short term memory neural network(AT-CNN-BiLSTM) is proposed. The proposed method consists of the signals preprocessing, the network training and the signals prediction components. Simulating the outdoor satellite signals under the Gaussian white noise, a training set and a testing set of the satellite signals will be obtained. Then we input the training set into the training unit for features extraction, and finally input the testing set into the network for prediction analysis. The simulation results show that in the comparison of other four artificial intelligence methods, the hybrid neural network can achieve a high accuracy (91.8%) at a fast convergent speed, which effectively solves the difficulty of the LEO satellite outdated channel parameters. The proposed method can improve the satellite communication qualities and save the satellite channels' resources.