LAUSR

In the letter, we propose to estimate uplink channel in urban 3-D MIMO systems utilizing denoising convolutional neuron networks (DnCNNs), which is recently proposed and has succeeded in tackling with image denoising tasks in computer vision. To process the complex numbers in channel estimation (CE), based on complex DnCNN (CDnCNN), a variant of DnCNN, we propose the simplified DnCNN (S-DnCNN) CE. S-DnCNN CE includes two major parts. In the first part, the channel frequency impulses are rearranged to form feature maps. In the second part, considering the practical latency constraints, we provide the S-DnCNN structure and fast training/testing (FT) methods. The practical 3-D MIMO channel model is used in the simulation. The results show that S-DnCNN CE outperforms traditional methods and CDnCNN-based ChannelNet. Besides, our FT S-DnCNN CE greatly reduces the time complexity compared to ChannelNet.