SemCom-OPTIMA: Empirically-Driven Optimization of Semantic Image Transmission Across Heterogeneous Edge–Cloud Systems

Semantic communication (SemCom) leverages artificial intelligence (AI) to prioritize meaningful information exchange, significantly enhancing efficiency and resource utilization in next-generation wireless networks. Despite its potential, practical deployment within multi-tier, heterogeneous edge-cloud systems presents substantial challenges, including diverse device computational capabilities, varying communication resources, and semantic mismatches between user encoders and edge-server decoders. To overcome these challenges, we propose SemCom-OPTIMA, a comprehensive optimization framework specifically designed for maximizing semantic image reconstruction accuracy in multi-user, multi-edge environments under strict latency and energy constraints. The problem is formulated as a joint mixed-integer nonlinear programming (MINLP) model, incorporating user-edge associations, semantic masking ratios, CPU frequency scaling, and joint power-bandwidth allocation. Given its NP-hard complexity, we present a multi-stage decomposition algorithm, solving iteratively tractable subproblems for semantic-computational allocation, channel resource management, and adaptive user-edge association, while ensuring convergence and feasibility. Distinctively, this work is grounded in extensive empirical analysis involving a rigorous 600,000-sample design-space exploration spanning masking ratios (0–0.99), signal-to-noise ratios (1–16 dB), and diverse semantic image classes (1000 ImageNet labels). Experiments conducted on a realistic 10-user, 3-edge testbed demonstrate that SemCom-OPTIMA consistently achieves a 10–12% higher peak signal-to-noise ratio (PSNR) and robustly meets energy and latency constraints compared to state-of-the-art baseline methods.