Enhancing Underwater Acoustic Orthogonal Frequency Division Multiplexing Based Channel Estimation: A Robust Convolution-Recurrent Neural Network Framework with Dynamic Signal Decomposition

MANSOOR JAN^1*Muhammad Aman²Syed Agha Hassnain Mohsan³Samih Mostafa⁴Faten Khalid Karim^4*

• ¹Hohai University, Nanjing, China
• ²Shanxi Agricultural University, Taiyuan, China
• ³Peking University Shenzhen Graduate School, Shenzhen, China
• ⁴Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia

Underwater acoustic (UWA) communication presents significant challenges owing to the unique, unexpected, and dynamic nature of the acoustic channel, which is affected by low signal-to-noise ratio (SNR), severe multipath propagation, latency, and Doppler spread, coupled with a shortage of real-world data. Accurate orthogonal frequency division multiplexing (OFDM) is essential for establishing resilient and reliable data transmission in challenging environments. This work introduces a CRNet estimator, including dynamic signal decomposition (DSD) techniques to properly estimate UWA-OFDM channel characteristics and mitigate noise-induced effects in received signals. The suggested CRNet model effectively captures both spatial and temporal aspects in the complex and dynamic UWA channel, demonstrating robustness despite a small number of pilots. The proposed CRNet model is trained using received pilot symbols combined with transmitted pilots and accurate channel impulse responses; hence, the CRNet model provides estimated channel impulse responses during the operational phase. Upon completion of training, our CRNet model does not need supplementary channel characteristics such as SNR, relying only on the received signal as its input. The numerical findings indicate that the suggested model regularly surpasses benchmarks, including least squares, minimal mean square error, and backpropagation neural network techniques, in terms of bit error rate and amplitude and phase error. Our suggested CRNet model demonstrates superior performance with QPSK than QAM. Moreover, performance evaluation on both training and unseen data sets illustrates the resilience and flexibility of the CRNet estimator in demanding underwater acoustic environments.