🌌 EPHEMRA — AI Prediction of GNSS Satellite Clock & Ephemeris Errors

AI/ML-based system aligned with ISRO Problem Statement 25176
Predicting time-varying GNSS satellite clock & orbit error build-up using Transformer models.

---

🛰️ Problem Statement (ISRO)

Goal:
Predict the time-varying error build-up between uploaded and modeled values of:

• Satellite Clock Bias
• Satellite Ephemeris (Orbit)

These errors directly affect navigation accuracy and timing precision.

---

🌠 Application Preview

🏠 Landing Experience

EPHEMRA provides an interactive research dashboard to analyze and forecast satellite error behavior.

---

🛰️ Orbit Selection Interface

Users can choose between:

• MEO Constellation – GNSS navigation satellites
• GEO Constellation – Geostationary satellites

---

📊 MEO Orbital Analysis Dashboard

Features:

• Positional deviation plots (Radial / Along-Track / Cross-Track)
• Satellite clock bias prediction
• Residual diagnostics
• Ephemeris log export

---

🌍 GEO Orbital Analysis Dashboard

Includes:

• Drift monitoring
• Error component tracking
• Clock error analysis
• Statistical error logging

---

🎯 Project Objective

EPHEMRA predicts the error growth between:

📡 Uploaded broadcast parameters
📐 Modeled ICD parameters

for satellites operating in:

• MEO (Medium Earth Orbit)
• GEO / GSO (Geostationary Orbit)

Predictions generated every 15 minutes up to 24 hours ahead.

---

🧠 AI/ML Approach

Models Implemented

• 🔁 Transformer-based time series forecasting
• 📊 Deep learning sequential models
• ⏱️ Long-horizon prediction (15min → 24hr)

Why Transformers?

Satellite error propagation has long-range temporal dependencies, making Transformers ideal.

---

📊 Prediction Horizons

Horizon	Supported
15 minutes	✅
30 minutes	✅
1 hour	✅
2 hours	✅
24 hours	✅

Training: 7 days data
Prediction target: Unseen 8th day

---

🏗️ Architecture

Frontend (Next.js Dashboard)
        ↓
FastAPI Inference API
        ↓
Transformer Models (TensorFlow)
        ↓
GNSS Ephemeris Dataset

---

⚙️ Tech Stack

Backend (ML & API)

• Python 3.9+
• FastAPI
• NumPy & Pandas
• TensorFlow / Keras
• Custom Transformer Models

Frontend (Dashboard)

• Next.js (App Router)
• TypeScript
• Tailwind CSS
• Research-style visualization UI

---

📈 Evaluation (ISRO Criteria)

Model performance measured by:

• Prediction accuracy across horizons
• Gaussian distribution of residuals
• Long-term prediction stability
• Statistical consistency

---

✨ Key Features

• GEO & MEO separate pipelines
• Transformer long-horizon forecasting
• API-driven inference
• Research visualization dashboard
• Expandable to probabilistic forecasting

---

🔮 Future Work

• Probabilistic uncertainty estimation
• GAN-based error synthesis
• Cloud inference pipeline
• Real-time GNSS ingestion
• Research paper publication

---

👩‍🚀 Author

Sejal Mukane
AI/ML Engineer — Space Technology & Scientific ML

Focus: GNSS • Time-Series Forecasting • Research Systems

---

⭐ Conclusion

EPHEMRA demonstrates how modern AI and Transformer models can improve GNSS navigation reliability and directly support ISRO’s real-world satellite navigation challenges.