Introduction
As the industry pivots toward 6G, the demand for immersive, high-bandwidth data like panoramic (360-degree) video is skyrocketing. Traditional communication systems, which prioritize bit-level accuracy, often suffer from the "cliff effect"—a sudden collapse in video quality when the signal-to-noise ratio (SNR) drops.
Semantic Communication (SemCom) offers a solution by transmitting the underlying meaning or features of data rather than raw bits. However, integrating SemCom into existing networking stacks poses challenges regarding compatibility, security, and the massive data volume inherent in panoramic video projections. To address these, this paper introduces CLESC, a cross-layer encrypted semantic communication framework.
The CLESC Framework
The authors propose a shift in how semantic components are integrated into the traditional networking stack. Unlike naive implementations that treat semantic extraction as a separate pre-processing step, CLESC optimizes the placement of encryption and encoding to maintain data correlations while ensuring security.
As shown in the comparison above, the proposed approach (Scheme 3) integrates the Semantic Extractor, Deep JSCC Encoder, and Encryption within the Application Layer. This allows the system to calculate "semantic importance" before the data hits the lower transport and physical layers, enabling more efficient resource allocation.
Adaptive Panoramic Video Semantic Transmission (APVST)
At the heart of CLESC is the APVST network, which utilizes Deep Joint Source-Channel Coding (Deep JSCC) powered by Swin Transformer v2 blocks. The architecture is specifically designed to handle the unique geometry of 360-degree video.
Key Technical Components:
- Latitude Adaptive Module: In Equirectangular Projection (ERP), pixels near the poles are stretched and contain redundant information. This module dynamically adjusts encoding length based on latitude, significantly boosting efficiency.
- Weight Attention (WA) Module: This prioritizes regions of the frame that contribute most to human-perceived quality, optimizing for metrics like WS-PSNR and WS-SSIM.
- Importance-Driven Reliability: The framework uses an entropy model to identify critical visual features. High-importance packets are assigned more robust error-checking (CRC) and more frequent retransmissions (ARQ) across the layers.
- Secure Integration: Encryption is performed after semantic extraction but before channel coding. This sequence ensures data security without disrupting the model's ability to learn and reconstruct source correlations.
Key Results and Insights
The CLESC framework demonstrates significant improvements over traditional H.264 and H.265 codecs, particularly in challenging environments:
- Bandwidth Efficiency: The system achieved up to an 85% reduction in bandwidth compared to H.264 and a 33% reduction compared to H.265 while maintaining equivalent visual quality.
- Resilience to Noise: By leveraging learned context, the model degrades gracefully at low SNR. Where traditional systems would fail entirely (the cliff effect), CLESC maintains a watchable image by "guessing" missing details based on semantic features.
- Reduced Latency: The system reached near-optimal performance with an average of only 3 retransmissions, compared to the 5 required by traditional H.265 schemes to reach even inferior quality levels.
Conclusion
CLESC bridges the gap between next-generation semantic AI and traditional networking protocols. By accounting for the geometric properties of panoramic video and implementing an importance-aware cross-layer mechanism, the framework provides a robust and efficient path for immersive 6G applications. Future work will explore expanding this framework to multimodal data, combining video with audio and sensor inputs for a more comprehensive communication experience.
Summary updated