Application of deep learning for joint channel estimation and signal detection in underwater acoustic WSN
摘要
More recent Underwater Wireless Sensor Networks (UWSNs) systems uncover incredibly complex features that lead to interference and scattering effects of acoustic waves, resulting in multipath signals. These days, UWSNs systems are widely employed in various applications, including seabed exploration, and early acoustic warning systems use acoustic waves as the communication medium to learn the channel parameters. As such, high latency, signal distortions, and other internal disruptions affect the system's overall performance. Efficient approaches for Signal Detection (SD) and Channel Estimate (CE) are required to tackle this issue and enhance communication in the UW Acoustic WSN (UAWSN). This work presents a novel hybridized Deep Learning (DL) based Semantic Guided Attentive Auto-encoder with DenseNet201 (SGAttA-DNet201) technique to efficiently detect the signal at the receiver end and to jointly evaluate the channel parameters. After the DenseNet201 (DNet201) model is introduced for SD at the end of the UAWSN receiver, the Semantic Guided Attentive Auto-encoder (SGAttA) model is proposed to perform CE. Key evaluation measures, such as accuracy rate, Bit Error Rate (BER), Root Mean Square Error (RMSE), Normalized Mean Square Error with Power Delay Profile (NMSE-PDP), and computational complexity, are acquired in order to analyze the performance of the suggested technique. The suggested method yields 100% accuracy, BER of