- 期刊
The Channel Prediction Combining Deep Convolutional Autoencoder and CNN-BiLSTM
摘要
Aiming at the problems of noise interference, rapid change of channel quality and outdated channel parameters in wireless communication transmission, a channel prediction model combining deep convolutional autoencoder and CNN‐BiLSTM (Convolutional Neural Network ‐ Bidirectional Long Short ‐ Term Memory) is proposed. First, a deep convolutional autoencoder network is constructed as a preprocessing of channel prediction. Secondly, a hybrid model integrating CNN and BiLSTM is constructed, the local features of time series signal along the positive direction of the time axis are extracted by 1D CNN (one‐dimensional convolution), and the BiLSTM extracts the bidirectional global feature, and then fuse the extracted features, which is conducive to the prediction. Finally, transfer learning is adopted for small dataset to fine tune a new model. By comparing the proposed method and its variants on the dataset, the results show that the model proposed in this paper have certain advantages.