Aegis-RF

Development of a modular anomaly detection system for complex RF (I/Q) signals, oriented toward spectrum surveillance and electronic warfare scenarios. The project includes advanced synthetic signal generation, interference and jamming modeling, spectral feature extraction, and unsupervised machine learning models for real-time anomaly detection.

Overview

This project implements a complete simulation and anomaly detection framework for radio-frequency (RF) signals, with a focus on defense-oriented applications such as:

Spectrum surveillance Intentional interference (jamming) detection Tactical communication link monitoring RF intrusion detection

The system operates on complex baseband I/Q signals generated under realistic mathematical models of digital modulation and channel effects.

Technical Objectives

1. Design a synthetic RF signal generator supporting digital modulations (BPSK, QPSK, QAM).

2. Simulate realistic channel effects including AWGN and controlled degradations.

3. Model representative RF anomalies:

Broadband jamming Narrowband interference bursts Unexpected carrier injection Abrupt modulation changes Frequency drift (oscillator instability simulation)

4. Build a signal processing pipeline:

Windowing and segmentation Short-Time Fourier Transform (STFT) Spectral and statistical feature extraction

5. Implement unsupervised anomaly detection models:

Isolation Forest One-Class SVM Autoencoders (dense and convolutional variants)

6. Evaluate performance using defense-relevant metrics:

False Alarm Rate (FAR) ROC-AUC Detection delay Sensitivity to SNR variations

System Architecture

The system is structured into modular layers:

1. Signal Generator

High-fidelity simulation of digital modulations (BPSK, QPSK, QAM) and controlled channel degradation (AWGN, fading).

2. Signal Processor

Real-time feature extraction using Short-Time Fourier Transform (STFT) to convert complex I/Q data into time-frequency representations.

3. Anomaly Detector

Deep learning-based unsupervised engine (Autoencoder) that learns the "normal" spectral footprint and identifies deviations based on reconstruction error (MSE).

4. Evaluation Engine

Metrics-focused suite measuring Probability of Detection (Pd) vs. False Alarm Rate (Pfa) under varying SNR scenarios.

Key Differentiators

Defense-Oriented Metrics: Focus on robust performance under low-SNR conditions. Modular Pipeline: Easily swappable models and signal generators. Edge-Ready Design: Architected for potential deployment on embedded hardware (FPGA/SoC) for real-time tactical surveillance.

Performance Analysis

he system evaluates performance using:

1. ROC-AUC Curves

Visualizing detection capability across different noise floors.

2. Reconstruction MSE

Using the Mean Squared Error of the autoencoder to define the anomaly threshold.

Technologies

Python (NumPy, SciPy, scikit-learn, PyTorch) Digital Signal Processing (DSP) Mathematical modeling of digital modulation schemes Modular and reproducible ML architecture design