Ashish Sharma supervised by Dr. Deepak Gangadharan received his Master of Science by Research in Computer Science Engineering (CSE). Here’s a summary of his research work on Securing Low-Cost UAVs: From DDoS Detection to Post-Quantum Cryptography

Low-cost Unmanned Aerial Vehicles (UAVs) are used in agriculture, logistics, inspection, disaster response, and defence. In many of these systems, the UAV and the Ground Control Station (GCS) exchange commands and telemetry over Micro Air Vehicle Link (MAVLink) protocol. MAVLink is lightweight and practical, but it lacks strong built-in security mechanisms. As a result, a Wi-Fi-based UAV link can be exposed to Distributed Denial of Service (DDoS) floods, eavesdropping, spoofing, replay, and command injection. This thesis addresses this problem through two complementary directions. The first part focuses on DDoS detection on a low-cost UAV platform. Seven machine learning (ML) and deep learning (DL) models, Extreme Gradient Boosting (XGBoost), Isolation Forest (IF), Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM), LSTM with attention, Bi-LSTM with attention, and a Time Series Transformer (TST), are evaluated against Transmission Control Protocol (TCP), Internet Control Message Protocol (ICMP), and combined TCP+ICMP flooding on a live Pixhawk/Raspberry Pi testbed. The results show a clear pattern: models with attention handle this task better. The TST gives the best overall performance, with F1 scores of 0.999, 0.997, and 0.943 for the three attack cases. The second part focuses on post-quantum protection for MAVLink communication. It presents PQTUNNEL, a transparent application-layer tunnel that adds post-quantum session establishment and authenticated packet protection without changing the flight controller or the GCS software. Session establishment uses a three-message authenticated one-round-trip (1-RTT) handshake, the benchmark study covers 24 level-matched Key Encapsulation Mechanism (KEM)×signature suites, and the data plane evaluates eight Authenticated Encryption with Associated Data (AEAD) profiles across National Institute of Standards and Technology (NIST) security levels 1, 3, and 5. A lightweight runtime policy is included only as a support layer for choosing among practical configurations under load. The live evaluation covers 198 tunnel sessions on a Raspberry Pi 4 companion computer with AVHzY CT-3 power measurements at 1 kHz. Module-Lattice Key Encapsulation Mechanism (ML-KEM) keeps key establishment in the low-millisecond range, ML-KEM paired with Module-Lattice Digital Signature Algorithm (ML-DSA) gives the most practical handshake behaviour on the drone side, and ChaCha20-Poly1305 remains the fastest software AEAD for the 256-byte MAVLink payload regime studied here. Overall, the thesis shows that both attention-based DDoS detection and post-quantum secure communication are practical on low-cost UAV hardware when the design stays close to the limits of the platform. 

July 2026