Deep Signal Lab – SETI-Inspired Signal Anomaly Explorer
An anomaly detection system for monitoring radio telescope signals, designed to identify unusual patterns in time-frequency data that might indicate signals of interest.
Overview
An anomaly detection system for monitoring radio telescope signals, designed to identify unusual patterns in time-frequency data that might indicate signals of interest.
Screenshots1/4
Spectrogram overview
This plot shows the simulated radio spectrogram: time left to right, frequency bottom to top, and brightness as signal power. You can see the noisy background, steady narrowband lines, and injected bursts and drifting tracks that the detector is meant to find.
Tech Stack
Key Features
- ▸Synthetic SETI-inspired time–frequency dataset with labeled anomalies
- ▸Spectrograms with highlighted candidate signals and bursts
- ▸Time-series plots with anomaly overlays and threshold explanations
- ▸Rule-based and ML-assisted anomaly detection approaches
- ▸Plain-language insights that connect to real-world analytics problems
Design Philosophy
Focus on clarity and interpretability: simple color palettes, clear annotations, and layouts that make it easy to understand what the system is detecting and why.
Design
Focused on clarity and speed: typography choices to improve scan-ability, high-contrast UI for readability, and a layout that highlights primary actions and content without distraction.
Development
Built with Python, Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, Jupyter Notebook. Emphasis on modular components, predictable state handling, and accessible interactions. Performance budgets guided media usage and bundle size.
Target Audience
Technical teams and stakeholders who need to understand signal monitoring systems and anomaly detection workflows.
Deployment
Analysis pipeline built with Jupyter notebooks and reproducible data generation. Results are documented with visualizations and exported for stakeholder review.
How to read these plots
The main spectrogram (time × frequency) shows what the \"radio sky\" looks like over time: most of the frame is background noise, while bright streaks and bursts are unusually strong activity at specific frequencies.
The time-series charts collapse that into a single signal so you can see when the overall system is quiet versus spiky; highlighted points are where a robust z-score detector believes something stands out enough to raise an alert.
Finally, the distribution views show that most flags sit in the extreme right tail of the power values rather than in the bulk of the noise, making the detector easy to explain to non-technical stakeholders as \"we only alert on the rarest, strongest events\".
Challenges
Designing realistic synthetic signals for testing, choosing interpretable detection rules that balance sensitivity and false positives, and presenting findings in a way that technical and non-technical stakeholders can understand.
Outcomes
A working anomaly detection pipeline that can process time-frequency data, flag unusual events, and provide clear visualizations for follow-up investigation. The system uses robust statistical methods (median + MAD) that handle noisy backgrounds better than simple thresholds.