Implicit codebook-based noise resilient underwater digital semantic communication system.

Area of Science:

• Underwater optical semantic communication (UWOSC)
• Digital communications
• Image processing

Background:

• State-of-the-art UWOSC systems use vector quantization (VQ) with explicit codebooks for digital compatibility.
• VQ methods face challenges like structural mismatches and codebook collapse, hindering performance in underwater wireless optical communication (UWOC).

Purpose of the Study:

• To develop a novel, robust implicit codebook-based UWOSC system for enhanced image transmission.
• To address the limitations of explicit codebook methods in UWOSC systems.

Main Methods:

• Utilized a CNN-based semantic feature transceiver for image feature extraction.
• Employed a finite scalar quantization (FSQ)-based implicit codebook construction with dimension-wise bit mapping.
• Incorporated adversarial training with a patch discriminator to improve visual fidelity.
• Implemented a dual-phase training strategy to enhance model robustness.

Main Results:

• The FSQ-based implicit codebook approach demonstrated superior performance compared to VQ-based explicit codebook methods.
• Achieved reduced communication overhead in image transmission.
• Showcased enhanced robustness under challenging UWOC channel conditions.

Conclusions:

• The proposed implicit codebook method offers a robust and efficient solution for UWOSC image transmission.
• The dimension-wise bit mapping and predefined codebook structure prevent codebook collapse and stabilize training.
• This approach significantly improves UWOSC system performance, especially in adverse underwater environments.