Cross-Band Spectrum Prediction Algorithm Based on Transfer Learning and Meta Learning
Chuang Peng, Mengbo Zhang, Weilin Hu, Lunwen Wang · 2021 7th International Conference on Computer and Communications (ICCC) · 2021
Spectrum prediction is an important research task in cognitive radio. It can predict the state change of spectrum and play an important role in improving spectrum sensing performance. Existing spectrum prediction algorithms based on deep learning use a large amount of training data, and due to the difference of frequency band data, the prediction model cannot be directly used across bands. In order to solve this problem, spectrum prediction methods based on meta-learning and transfer learning are studied in this paper. Firstly, the VGG16 pre-training model is used to analyze the data in different frequency bands, to clarify the differences and similarities between different frequency bands, and to give the theoretical analysis that the prediction model cannot be used across frequency bands. Secondly, a meta-learning data set is constructed and the model is trained by meta-learning to find the optimal distribution of model parameters. Finally, the method based on meta-learning and transfer learning can effectively predict the target frequency band with a small amount of spectral data. Experimental results show that the proposed method can efficiently and rapidly realize the cross-band prediction of the model using a small amount of target band data, and has better stability than only using transfer learning.