BG:

Resource allocation is a key challenge in Integrated Sensing and Communication (ISAC) systems:

• Traditional methods focus on balancing sensing and communication without reducing redundant sensing data transmission, leading to inefficiencies.
• Existing approaches allocate resources uniformly, failing to adapt dynamically to changing sensing requirements and communication needs.
• Moreover, sensor failures or missing data degrade system performance due to the lack of an effective data-recovery mechanism.

Proposed Method:

To address these issues, we propose a Semantic Inpainting-Based Dynamic Resource Allocation mechanism to optimize ISAC performance. Our approach leverages:

• Semantic Inpainting: Inspired by JEPA (Joint Embedding Predictive Architectures) and self-supervised learning, we reconstruct missing or redundant sensor data at the semantic level, reducing the need for raw data transmission.
• Self-Enhanced RF-Sensing: Exploiting the natural transmission of sensing data through wireless channels, further reducing redundant sensor data transmission by enhancing naturally-existing RF-based sensing (e.g., WiFi CSI).
• Dynamic Resource Allocation: We aim to develop a structured ISAC system model that quantitatively analyzes the trade-off between sensing and communication. Unlike our previous work, which only formulated the problem conceptually, this work rigorously models the system and validates our approach through simulation.

In a word, we aim to develop a more efficient, adaptive and robust ISAC system that minimizes sensing resource allocation while maximizing communication capability.

Contributions:

Compared to our previous conf. paper, this journal work introduces the following major advancements:

• From Theoretical Formulation to ISAC System Modeling
  • Transitions from a conceptual trade-off formulation to a complete ISAC system model, enabling real-world trade-off analysis and simulation-based validation.
• Dynamic Resource Allocation with Fine-Tuning for Long-Term Adaptation
  • Introducing a state-based adaptive resource allocation framework which considers long-term performance degradation due to domain shift.
• Simulation-Based Trade-Off Validation
  • Moves beyond proof-of-concept inpainting experiments by validating semantic inpainting’s impact through simulations, enabling deeper quantitative insights.
  • Extends the study of Self-Enhanced RF-Sensing, which was previously unexplored.
• V-JEPA for SpatioTemporal Semantic Inpainting
  • Advances from I-JEPA’s single-time-step spatial inpainting to video-based spatio-temporal inpainting through experiments.
  • Introduces Local-to-Global Semantic Inpainting, integrating multiple local models into a global representation for a holistic ISAC perspective.