Deep Learning-based Terahertz Channel Estimation

Terahertz (THz) wireless communication involves highly abrupt channel changes that are vulnerable to profound degradation. This paper proposes a deep learning channel estimate approach for THz wireless communication. We use a Markov process to represent the THz fast-fading communication channel. A Residual structure-based E-ResNet network is also suggested at the same time. To estimate channel state information, this approach interpolates and denoises the pilot grid. This makes constellation diagrams to be recovered in deep fading caused by drastically changing channels. This network's realization is described in the study. Experiments reveal that, when the channels are thoroughly learned, this technique beats both Minimum Mean Squared Error (MMSE) and Convolutional Neural Network (CNN) schemes. The results show the effectiveness of the technique for channel estimation.