Split-Plane Decentralized Middleware For Secure Multi-Protocol Iot Communication: Architecture, Formal Verification, And Experimental Validation

Abstract

The rapid expansion of Internet of Things (IoT) deployments across smart cities, industrial automation, and critical infrastructure has exposed fundamental limitations in prevailing IoT security architectures. Centralized cloud-based solutions introduce single points of failure and incur latency incompatible with real-time operation, while fully decentralized blockchain-based approaches suffer from scalability and throughput constraints. This paper proposes a decentralized middleware for secure multi-protocol IoT communication and analytics based on a split-plane architecture that explicitly decouples real-time security enforcement from auditability and analyt-ics. The middleware decomposes into four independently optimized planes: a Security Plane with distributed Policy Decision Points synchronized via gossip protocols for sub-40 ms admission decisions; a Data Plane using RabbitMQ Quorum Queues and Apache NiFi for fault-tolerant, high-throughput protocol mediation; an Analytics Plane employing a two-stage Random Forest and Deep Autoen-coder intrusion detection pipeline with ADWIN-based concept drift detection and Krum-robust federated aggregation; and an Audit Plane backed by Hyperledger Fabric with periodic Merkle-root anchoring to Polygon PoS at a cost of approximately $2.16 USD per month. Formal verification using TLA+ exhaustively explores over 1.84 million states to confirm that the audit-layer consensus pre-serves agreement and validity under crash-fault assumptions, with zero safety violations detected. Extensive experimental evaluation across five public benchmark datasets on both cloud-tier (Intel i7) and edge-tier (Raspberry Pi 4) hardware demonstrates that the system maintains peak throughputs exceeding 10,000 transactions per second, incurs less than 15% security overhead, achieves aver-age enforcement latencies below 40 ms, and sustains detection F1-scores up to 0.97. A systematic ablation study confirms that each plane contributes necessarily: disabling audit-plane decoupling increases latency by 379%, while removing adaptive trust decay raises the false positive rate by 158%. These results confirm that decentralized trust and real-time IoT performance can be reconciled through explicit architectural separation