Model-Driven Based Deep Unfolding Equalizer for Underwater Acoustic OFDM Communications

It is a challenge to design a equalizer for complex time-frequency doubly-spread channels. In this paper, we employ the deep learning (DL) architecture by that unfolding an existing iterative algorithm to build an equalizer named underwater deep network (UDNet) for underwater acoustic (UWA) orthogonal frequency division multiplexing (OFDM) signal. Considering constellation recognition is a classification issue, the one-hot coding and softmax layer are adopted in the proposed network to achieve the minimum Kullback-Leibler (KL) criterion. Simultaneously, we introduce a sliding structure based on the banded approximation of the channel matrix to reduce computational complexity and aid UDNet performs well for different length signals without changing the network structure. Furthermore, we apply the environment of the true UWA channel as much as possible, including utilize measured doubly-spread UWA channel and offshore background noise to evaluate the UDNet. Experimental results show that in the case of 10-35dB SNR, UDNet achieves better performance with low computational complexity.