V2X systems operate under strict real-time constraints and utilize diverse hardware configurations. As a result, a single, fixed post-quantum cryptography (PQC) solution is often insufficient to address all scenarios. With the advent of quantum-safe schemes requiring larger keys and greater computational power, there is a pressing need for adaptive defenses that match cryptographic strength to current conditions without compromising security.
This thesis presents a practical and measurable approach, involving the development of a policy that selects the most appropriate PQC primitive based on specific context, quantification of its costs and benefits using realistic data, and demonstration of how to transition between algorithms as conditions evolve safely.
The goal is to design and benchmark a context-aware, adaptive PQC stack for connected and autonomous vehicles. You will create a selector that dynamically switches among various PQC options (lattice, code-based, or hash-based) depending on runtime conditions such as latency, CPU load, link quality, and mobility, all while maintaining a minimum level of security. The work will focus on handling real data, performing rigorous benchmarking, and ensuring a clean, reproducible engineering process.
Core questions
- How do different PQC families trade off latency/throughput/CPU/memory under realistic V2X loads?
- Can a learned/contextual policy maintain a security floor while minimizing end-to-end delay?
- What is the handover overhead (keys, state, session continuity) when switching algorithms on the fly?
- How stable is the selector under rapid context shifts (e.g., handovers, congestion, bad weather)?
Tasks
- Implement a context profiler (compute/network/mobility) and an adaptive selector for PQC algorithms.
- Benchmarking with real data handling: ingest and engineer vehicular/network traces; run head-to-head tests on latency, throughput, CPU/memory, key/ciphertext sizes, and handover cost.
- Build a decision engine (rules or ML) with a tunable security constraint; evaluate policy quality vs. fixed baselines.
- Engineering for reproducibility: Dockerized experiments, config files, unit tests, and clear plotting scripts.
- Implement a bandit/RL selector for algorithm choice; compare against rule-based and supervised baselines using regret and latency/security cost metrics.
Expected outcomes
Reproducible code & docs; benchmark tables/plots; concise thesis & 10-slide defense deck.
Student profile / prerequisites
- Strong coding for benchmarking & data pipelines (Python; C/C++ a plus), solid Git/Docker habits.
- Basics of PQC and math (linear algebra, probability, asymptotics); able to read specs and adapt them.
- Comfortable with real data handling (cleaning, feature engineering, repeatable experiments).
Nice-to-have
- Experience with networking/perf testing, crypto APIs, or simple ML (classification/policy learning).
- Familiarity with Linux tooling, CI, and profiling.
Datasets
Public vehicular/V2X and network traces where available; augmented with controlled synthetic scenarios (e.g., mobility/traffic simulators) to cover edge cases.
Tools/Stack
Python, NumPy/Pandas, matplotlib; Docker; Bash; optional C/C++ bindings; optional PQC libraries (e.g., Open Quantum Safe/liboqs or similar); simple DB for logs/metrics (SQLite/Postgres).
Deliverables
Thesis PDF, code repository, dataset cards, benchmarking report with plots, experiment configs.