Maritime Channel Estimation with Enhanced Super Resolution Based Convolutional Neural Network

Maritime communication is one of the key enabling technologies for human's maritime activities in modern days. To overcome the challenges imposed by the complicated sea surface conditions on channel estimation (CE), we propose a sparse feature extraction (SFE) and endmost upsampling (EU) aided channel estimation network (SE-CENet) for the island-to-sea (I2S) scenario. In the SFE module, a sparse transform unit (STU) is devised for extracting the sparsity characteristics of the I2S channel from the estimated channel frequency-domain response (CFR) at pilot positions. In the EU module, a learnable sub-pixel convolution layer is adopted, followed by a cropping layer that enables the network to adapt to different pilot patterns. Simulation results demonstrate the superiority of SE-CENet in both estimation performance and computation complexity in comparison to existing schemes.