A Transformer-Based OFDM Receiver for Underwater Acoustic Communication

In recent years, the applications of deep learning have aroused growing interests in wireless communication. However, existing deep neural networks often fails to consider the scenarios invoving long transmitted sequences, whose complex correlations are essential for symbol detection. This paper proposes a transformer-based OFDM receiver, which completely replaces the channel estimation, equalization, and demodulation modules in traditional methods. The network consists of an encoder with long sequence distance perception and a multi-layer perceptron with a hidden layer. Among them, the encoder utilizes attention mechanisms to capture the correlation between the receiving signals and discovers the change in channel characteristics, while the multi-layer perceptron(MLP) is employed for detection and recovery of all original signals. Simulation and experiments show that our model outperforms fully connected deep neural networks (DNN) and skip-connected convolutional neural networks (CNN) in terms of bit error rates, and exhibits greater adaptability in various scenarios.