Post-Quantum Authenticated Encryption on Networked Raspberry Pi Hardware: Security Analysis and Masking-Mechanism Ablation

Post-quantum migration for constrained IoT devices is usually assessed one primitive at a time, so little is known about how a complete protocol behaves once deployed. This study examines a post-quantum authenticated-encryption construction combining ML-KEM key establishment, curve-based masking, and ML-DSA or Falcon signatures. Security is analysed in the classical random oracle model, and sender unforgeability and replay resistance are checked with Tamarin. A GMP-accelerated implementation is benchmarked on desktop hardware and deployed unchanged on two networked Raspberry Pi 4 boards, an embedded-class test platform and not a microcontroller. An ablation study separates the masking layer from a conventional AEAD layer and a hash-derived mask. Tamarin finds forgery and replay attacks against a plain ML-KEM+HKDF+AEAD composition that the construction resists. An authenticated baseline with the same signature and sequence-number logic is faster at every message size, by 1.9 to 37.0 times on desktop and 1.45 to 28.58 times on the Raspberry Pi (ratios of mean encryption times). At 1024 bytes the curve-based construction uses 21 to 24 times the raw energy per operation of that baseline, whereas a hash-masked variant uses 1.4 to 1.5 times. The hash-derived mask matches the curve-based mask on the measured properties at much lower cost.