Symbiotic Value Capacity Optimization in Satellite-Terrestrial Integrated Communication and Computing Networks

Abstract

Satellite-terrestrial integrated communication and computing network (STICCN) faces the core challenge of supporting highly heterogeneous tasks with differentiated requirements, under stringent dual constraints of communication and computing resources. Most existing scheduling schemes focus on macroscopic system performance, while overlooking two key limitations: value differentiation across heterogeneous tasks and the hard minimum startup computing power constraint for AI-driven tasks. This results in low service utility of scheduling policies, or even infeasibility for practical deployment. To address these gaps, we first introduce the minimum startup computing constraint into task modeling, and construct a symbiotic value capacity metric to quantify the service contribution of resource allocation. The joint offloading and resource allocation problem is then formulated as a Markov Decision Process (MDP). We further propose a Value-Aware Actor-Critic (VAAC) algorithm, which leverages a gated attention mechanism to coordinate hybrid discretecontinuous actions, and optimizes policies via a tailored valueaware advantage function and dedicated resource constraint loss function. Simulation results demonstrate the proposed algorithm consistently outperforms mainstream baselines in task completion rate, symbiotic value capacity, and user satisfaction factor.