Impulsive Noise Mitigation for Underwater Acoustic OFDM Systems Based on 1DCNN With Multiattention Mechanism and Transfer Learning

Underwater acoustic (UWA) communication is until now the only effective means for long distance underwater wireless communication, and hence it is the key foundation for Internet of Underwater Things (IoUT). However, in ocean environment, impulsive noise (IN) generated by natural and human factors usually seriously affects the performance of UWA communication. In this article, utilizing the powerful capability of deep learning, a 1-D convolutional neural network based on multiattention mechanism (1DCNN-MAM) for IN mitigation in UWA orthogonal frequency division multiplexing (OFDM) systems is proposed. To enhance the generalization performance of the network, it utilizes minimization of the energy on null subcarriers as an auxiliary task for network training. Furthermore, to adapt to specific environment quickly and reduce the amount of real data required for training, it adopts a network-based deep transfer learning approach for fine-tuning. To verify the performance of the proposed scheme, a sea trial has been carried out along with simulations, and both demonstrate that the proposed scheme can effectively suppress the IN in UWA OFDM systems.