Data augmentation aided complex-valued network for channel estimation in underwater acoustic orthogonal frequency...

Area of Science:

• Electrical Engineering
• Signal Processing
• Underwater Communications

Background:

• Underwater acoustic orthogonal frequency division multiplexing (UWA-OFDM) systems face challenges with data scarcity and sampling difficulties for deep learning applications.
• Accurate channel estimation is crucial for reliable UWA-OFDM communication.

Purpose of the Study:

• To propose a novel data augmentation aided complex-valued network for UWA-OFDM channel estimation.
• To address data scarcity and sampling issues in real-world UWA communication scenarios.
• To investigate the impact of high-frequency component augmentation on UWA channel estimation.

Main Methods:

• Empirical mode decomposition (EMD) based data augmentation is employed to generate synthetic data.
• A complex-valued neural network is designed to process complex-formatted UWA-OFDM signals efficiently.
• Experiments are conducted using the at-sea-measured WATERMARK dataset.

Main Results:

• The proposed method achieves near-optimal channel estimation performance.
• Augmenting high-frequency components positively influences model training and performance.
• The complex-valued network utilizes real and imaginary parts effectively, requiring fewer resources than real-valued networks.

Conclusions:

• The data augmentation aided complex-valued network offers a viable solution for UWA-OFDM channel estimation.
• The method's low resource requirements enhance its adaptability for practical UWA applications.
• This approach effectively mitigates data scarcity challenges in deep learning for UWA communications.