PQC-Enhanced Privacy-Preserving Federated Learning for Edge-Based IoT Ecosystems

With the emergence of quantum computing, traditional cryptographic methods face significant threats, making it increasingly challenging to ensure data privacy and integrity. How to design a secure and efficient federated learning (FL) algorithm against quantum attacks for edge-based IoT ecosystems is a valuable problem. To address this challenge, this paper proposes a novel FL privacy preservation scheme that fuses trusted execution environment (TEE) and lattice cipher. The scheme generates a joint seed and clients seeds through TEE, using a lattice-based key encapsulation protocol, and symmetric authentication encryption, to realize a secure channel establishment and seed distribution mechanism between client and server, by which the seeds are distributed to the participants securely, and then fed into a lattice-based homomorphic seudorandom sequence generator to generate masks for encrypting the model gradients. This combination ensures strong privacy preservation and efficient aggregation of model updates under semi-honest adversary models. Experimental results show that this scheme outperforms existing methods in terms of resistance to quantum attacks, communication efficiency and computational overhead, while enhancing the dropout robustness, making it suitable for resource-constrained IoT environments.