ZTSec-FedSDN: A Privacy-Preserving Federated Framework for SDN Attack Detection Using Zero-Trust Blockchain and 6G Terahertz Networks

Abstract

This research presents ZTSec-FedSDN, a privacy-preserving federated framework for SDN attack detection that integrates zero-trust blockchain architecture with 6G terahertz networks. The framework enables collaborative training of Deep Neural Network models across distributed clients while maintaining data privacy and leveraging the high-speed capabilities of 6G terahertz communication. Using the SDNFlow dataset containing diverse network traffic patterns and attack types, we implement four federated optimization strategies: Federated Averaging (FedAvg), Federated Proximal (FedProx), Federated Adam (FedAdam), and Federated Adagrad (FedAdagrad). The zero-trust blockchain layer provides an immutable ledger for recording and verifying model updates, ensuring transparency and trustworthiness in the federated learning process. The integration with 6G terahertz networks enables ultra-low latency communication between federated clients, crucial for real-time intrusion detection in SDN environments. Multiple clients collaboratively train the shared DNN model on local traffic data without exposing sensitive information, preserving data privacy while benefiting from collective intelligence. This work systematically evaluates the performance of different federated learning algorithms by comparing model accuracy, convergence speed, robustness, and efficacy in multi-class network attack classification. The results provide comprehensive insights into each optimizer’s suitability for secure, transparent, and trustworthy intrusion detection systems in next-generation SDN environments powered by 6G terahertz networks.