Blockchain-based Secure and Truthful Entity Participation Scheme in Heterogeneous Federated Learning

Federated learning (FL) preserves privacy yet struggles to secure truthful participation and adequate effort. Because participant types are private and effort is unobservable, adverse selection and moral hazard may arise. In this paper, we propose a blockchain-supported, contract-theoretic compensation and settlement framework to address asymmetric information and energy constraints. Utilities are formulated for coupled sensing and training energy under a global budget, and a contract menu specifies the minimum effective contribution and payment. A closed-form optimal payment is derived under binding lowest-type individual rationality and adjacent-upward incentive compatibility constraints. When budget truncation violates monotonicity, we restore implement ability via bunching. A minimal on-chain metadata design is adopted to reduce trust and compliance overhead. Only essential metadata is recorded, and a reputation-driven witness committee performs consensus, settlement, and auditing. Simulations indicate that truthful type disclosure and positive effort constitute best responses. The sensitivity analyses show robustness to parameter perturbations. The framework provides an end-to-end path to verifiable contribution accounting and trustworthy settlement in resource-constrained FL.