• CADDTO-PPO introduces carbon‑aware decentralized task offloading for MIMO‑MEC networks in. • Uses multi‑agent proximal policy optimization to jointly minimize carbon emissions, latency, and energy waste. • Models system as DEC‑POMDP enabling autonomous IoT agents to make fine‑grained power control decisions. • Parameter sharing architecture (DEPS) achieves O(1) inference complexity, scalable for dense networks. • Carbon‑first reward prioritizes green time slots, decoupling throughput from grid carbon footprints. • Experiments show CADDTO‑PPO outperforms DDPG and Lyapunov baselines, achieving lowest carbon intensity. • Maintains near‑zero packet overflow rates even under extreme traffic loads.
Article Summaries:
- A new study introduces CADDTO‑PPO, a carbon‑aware, decentralized task‑offloading framework for multi‑antenna mobile edge computing (MIMO‑MEC) networks. The authors model the system as a Decentralized Partially Observable Markov Decision Process and employ multi‑agent proximal policy optimization (PPO) with parameter sharing to let individual IoT devices make local power‑control and offloading decisions. A carbon‑first reward structure prioritizes green‑energy time slots, decoupling throughput from grid carbon footprints. Experiments show CADDTO‑PPO outperforms DDPG and Lyapunov‑based baselines, achieving the lowest carbon intensity and near‑zero packet overflow even under heavy traffic, while maintaining constant O(1) inference complexity.
Sources: