DenseNet-Based Robust Channel Estimation in OFDM for Improving Underwater Acoustic Communication

Underwater acoustic (UWA) communication presents unique challenges due to the unpredictable and dynamic nature of acoustic channels, influenced by Doppler spread, low signal-to-noise ratios (SNRs), and the general need for complex channel characteristics, coupled with a scarcity of real-world data. Accurate orthogonal frequency division multiplexing (OFDM) channel estimation is pivotal for ensuring reliable data transmission in such challenging environments. In this study, we introduce the DenseNet estimator, which is specifically used for OFDM channel estimation in UWA communication. The use of dense connectivity within the DenseNet structure proves to be advantageous in capturing the intricacies of the complex and dynamic UWA channels. This architecture, showcasing robustness even when there's a limited number of pilots, sets it apart from conventional methods. The DenseNet estimator is trained on the WATERMARK data set, leveraging the richness of real-time varying channel impulse responses to provide the necessary diversity for accurate channel estimation. Uniquely, once trained, our DenseNet estimator operates without necessitating additional channel statistics like SNR, relying solely on the received signal as its primary input. This approach offers a simplified and more direct application in real-world scenarios. Our numerical results underscore the DenseNet estimator's efficacy: It consistently outperforms traditional methods such as least square, minimum mean square error, and fully connected neural network, recording improvements of up to 96.3%, 94.2%, and 40.0% in bit error rate. Performance assessments across various watermark underwater channels demonstrate the DenseNet estimator's adaptability and robustness in both stable and challenging environments.