无线电工程

基于自回归模型(Autoregressive Model, AR)的传统信道预测方法在高速移动中，信道具有较大的时变性，导致信道发生了非线性的改变。基于反向传播(Back Propagation, BP)网络的信道预测方法通过适当地调整权重可使模型更加稳健，但是算法效率较低。提出一种新的模型，设计长短期记忆(Long Short-Term Memory, LSTM)网络和增量学习相结合的在线信道预测模型，实现时变信道的在线预测。模型应用LSTM神经网络学习长时间序列的特性来处理时间相关通信系统中的信道状态信息，增量学习(Incremental Learning, IL)在运行期间不断预测系统状态，同时更新LSTM神经网络的现有权重，交替执行训练和预测过程，模型可以很好地适应无线信道的动态变化。实验结果表明，提出的模型能有效地改善时变信道的预测准确率。

The traditional channel prediction method based on Autoregressive Model(AR)has a large time-varying channel in high-speed movement, resulting in nonlinear changes in the channel. The channel prediction method based on Back Propagation(BP)networks can make the model more robust by adjusting weights appropriately, but this algorithm is less efficient. Therefore, a new model which designs an online channel prediction model combining Long Short-Term Memory(LSTM)network and incremental learning to achieve online prediction of time-varying channels is proposed. The model applies the characteristics of learning long-time series of LSTM neural network to process channel state information in time-related communication systems, and Incremental Learning(IL) continuously predicts the system state during operation, while updating the existing weights of the LSTM neural network. By alternately performing training and prediction processes, the model can be well adapted to dynamic changes of wireless channel. Experimental results show that the proposed model can effectively improve the prediction accuracy of time-varying channels.