Deep Learning Based Cross Frequency Channel Reconstruction and Modeling

Wireless channel modeling is widely considered as foundation of wireless communication system design. Sufficient and diverse channel data provides strong support for wireless channel characterization and modeling. However, channel data from real measurement is usually limited considering complexity of channel measurements for different scenarios and frequency bands. In this work, a deep learning-based cross-frequency channel generation and modeling framework is proposed. Without requiring a traditional parametric channel model, the proposed framework can generate realistic cross-frequency channels by employing generative adversarial networks. Based on vehicular channel measurement data, cross-frequency reconstruction performance of the proposed framework is validated by comparing characteristics of measured and reconstructed channels. It is also found that channel non-stationary characteristics can be well embodied in the reconstructed channels.