Signal Detection in Uplink Time-Varying OFDM Systems Using RNN With Bidirectional LSTM
Shengyao Wang, Rugui Yao, Theodoros A. Tsiftsis, Nikolaos I. Miridakis, Nan Qi
Abstract
In this letter, we propose a deep learning-assisted approach for signal detection in uplink orthogonal frequency-division multiplexing (OFDM) systems over time-varying channels. In particular, we utilize a recurrent neural network (RNN) with bidirectional long short-term memory (LSTM) architecture to achieve signal detection. In addition, with the help of convolutional neural network (CNN) and batch normalization (BN), a new network structure CNN-BN-RNN Network (CBR-Net) is proposed to obtain better performance. The sequence feature information of the OFDM received signal is extracted from big data to successfully train a RNN-based signal detection model, which simplifies the architecture of OFDM systems and can adapt to the change of channel paths. Simulation results also demonstrate that the trained RNN model has the ability to recall the characteristics of wireless time-varying channels and provide accurate and robust signal recovery performance.