CARSA: context-aware resilient semantic architecture: semantic communication for sustainable next-generation networks

Semantic communication is developing into a very promising paradigm for sixth-generation (6G) wireless systems. However, most existing approaches suffer from equal treatment of all semantic data, predominantly neural recovery methods, and catastrophic forgetting of previously learned information when adapting to new domains. This paper presents a context aware resilient semantic architecture (CARSA), which utilizes three main innovations: hierarchical semantic importance attention (HSIA); The use of common sense reasoning in conjunction with neural denoising for robust recovery in neuro-symbolic semantic recovery (NSSR); and elastic continual semantic adaptation (ECSA) utilizing Fisher Information weighted continual learning for near-zero forgetting of previously learned information when adapting to a new domain. CARSA achieved a 14.5% increase in semantic similarity at an SNR of − 5 dB and 122.9% increase in semantic similarity under Rayleigh fading compared to other baselines. Ablation tests confirmed that HSIA achieved a 6.6% increase in semantic similarity, while NSSR maintained recovery accuracy greater than or equal to 0.73 at 60% corruption. ECSA demonstrated less than 5% loss of previously learned information across five domains as opposed to the 60% loss associated with fine tuning. CARSA-Lite also achieved a 53% reduction in carbon footprint. CARSA established a new paradigm for the development of resilient and sustainable semantic communication for future generations of wireless networks, and demonstrates significant improvements in performance in low-SNR conditions, fading channels and multiple domain deployments.