A Deep Learning Method Based Receiver Design
Lei Liu, Tian Ran Lin, Yu Zhou · 2020
The design of the communication receiver for recovering the transmitted signals in conventional works is usually divided into several independent processing blocks, respectively corresponding to specific sub-tasks. In contrast, we propose an end-to-end receiver design based on deep learning (DL) technique in this paper. The proposed DL-based scheme feeds the received signals (including the pilots and transmitted data) at the receive antenna into the neural network as the input, the DL model is trained as a communication receiver to directly output the originally transmitted bits. The proposed DL-based scheme jointly solves the different tasks such as the carrier frequency offset estimation, channel state information estimation, and signal detection using a single neural network, which achieves performance gains. To reveal the performance improvements achieved by the proposed novel scheme, we provided numerical simulations compared with benchmarks. We further verified its feasibility in practical implementation via software-defined radio and illustrated the strong robustness.