CNN-Based Underwater Acoustic OFDM Communications over Doubly-Selective Channels

Due to the serious frequency selective fading and time selective fading in underwater acoustic (UWA) channel, the design of receiver becomes more difficult. We present a convolutional neural network (CNN) based orthogonal frequency division multiplexing (OFDM) receiver that fully considers the banded channel matrix features of doubly-selective channels in the UWA communication to achieve the integrated design of channel estimation and equalization. We propose a novel architecture using decoder and encoder convolutional neural network, named DECCN, which uses twenty-one convolution layers to compose an encoder and a decoder based on the dilation convolution and the feature reuse. DECCN focuses on reducing the complexity without considering the full connection (FC) layer. DECCN performs well for various length signals without changing the structure of system, such as 1024 bits and 2048 bits. Simulation results show that the proposed scheme reduces the bit error rate compared with the traditional algorithms and other DL-based schemes, especially with the pilot only 1/8 length of the signal or without cyclic prefix.

Automated summary

This paper proposes a convolutional neural network-based orthogonal frequency division multiplexing receiver for underwater acoustic communication, which integrates channel estimation and equalization for doubly-selective channels. The architecture named DECCN reduces complexity by using an encoder and a decoder based on convolutional neural networks, achieving good performance for various signal lengths without changing the system structure. The proposed scheme effectively reduces bit error rate, especially with pilot length only 1/8 of the signal or without cyclic prefix.