Important
This repository contains the project implementation and final report for an adaptive machine-learning approach to sub-threshold eavesdropping detection in Quantum Key Distribution (QKD).
The finalized project report, detailing the statistical methodology, ML architecture, and simulation results: Download Project Report (PDF)
The simulation framework evaluates the Adversarial Error-Classification Engine (AECE) across high-volume, 1,000-qubit distribution streams.
The AECE analyzes the error rates of non-cryptographic sentinel qubits to detect malicious interference that remains below the 11% QBER threshold.
Figure 1: Distribution of QBER Delta. The gray area represents the 95% confidence interval for stochastic noise ($H_0$).
By recovering noisy blocks that are statistically uncompromised, the project demonstrates a significant improvement in effective key throughput.
Figure 2: Performance comparison between static 11% abort strategies and the AECE filtering suite.
Analysis of the Random Forest model shows which temporal correlations are most effective at identifying adaptive adversaries.
Figure 3: Feature importance weights for the AECE classification model.
The bb84.ipynb serves as the primary artifact for the project. It features a high-performance vectorized numerical engine for simulating complex quantum channels.
Prerequisites:
numpy,scikit-learn,matplotlib
To run the project locally:
- Ensure
python 3.9+is installed. - Launch the Jupyter Notebook in the
notebook/directory. - Figures are automatically generated and saved to the
plots/folder.
Course/Institution: IIIT - Andhra Pradesh
Author: Sree Charan Desu (sreecharan309@gmail.com)