TinyGS QoS

A machine learning tool for predicting satellite transmission quality of service (QoS) probabilities in the TinyGS network. This project provides a Streamlit web application and utilities for analyzing LoRa satellite transmission success rates based on orbital parameters and radio configuration. https://tinygsqos.streamlit.app/

Overview

TinyGS QoS uses machine learning (Positive-Unlabeled learning) to predict the probability of successful satellite transmissions based on:

• Satellite position (latitude, longitude, altitude)
• LoRa radio parameters (spreading factor, bandwidth)
• Antenna configuration (gain)
• Geometric factors (elevation angle, distance to the nearest ground station)

Features

• Interactive Streamlit web interface for transmission probability visualization
• Global coverage heatmaps showing transmission success probability
• Support for custom satellite orbital parameters
• Batch prediction capabilities
• Sun-Synchronous Orbit (SSO) simulation tools
• Pre-trained ML model using scikit-learn

Installation

Requirements

• Python 3.12 or higher
• Poetry (recommended) or pip

Using Poetry

poetry add tinygs-qos

Using pip

pip install tinygs-qos

Usage

Running the Web Application

Launch the Streamlit interface:

streamlit run main.py

The application provides:

• Interactive parameter controls for SF, BW, antenna gain, and altitude
• Validation for valid LoRa parameter combinations
• Global transmission probability heatmaps
• Overlay of actual transmission data

Using the TransmissionPredictor Class

from tinygs_qos import TransmissionPredictor

# Initialize predictor
predictor = TransmissionPredictor()

# Single prediction
prob = predictor.predict(
    sat_alt=600.0,      # Satellite altitude in km
    sf=10,              # Spreading factor
    bw=125.0,           # Bandwidth in kHz
    min_gain=5.0,       # Antenna gain in dB
    el=45.0,            # Elevation angle in degrees
    distance_to_station=500.0  # Distance to nearest station in km
)

# Batch prediction
import pandas as pd
predictions = predictor.predict_batch(dataframe)

Generating Test Samples

from tinygs_qos.utils.tiny_utils import TestSample, SSOOrbitTestSample

# Random global samples
X_test = TestSample(
    n_samples=10000,
    rand_lat=True,
    sf=[10],
    bw=[125.0],
    gain=[5.0],
    alt=600.0
)

# Sun-Synchronous Orbit samples
X_orbit = SSOOrbitTestSample(
    altitude=600.0,    # km
    sf=10,
    bw=125.0,
    gain=5.0,
    raan=0.0,         # Right Ascension of Ascending Node
    time_step=1.0     # seconds
)

Data Files

The project requires the following data files in the data/ directory:

• PU_optuna_SDG_log_loss_12_7_v2.joblib - Pre-trained ML model
• packet_features.parquet - Historical packet transmission data
• station_locations.parquet - TinyGS ground station locations
• kdtree_stations.joblib - KDTree for fast nearest station lookup
• satellite_min_gain.parquet - Satellite-specific minimum gain data
• kde_satPosAlt.joblib - Kernel Density Estimator for altitude distribution

Valid LoRa Parameter Combinations

The model supports the following (Bandwidth, Spreading Factor) pairs based on actual TinyGS data:

Bandwidth (kHz)	Spreading Factor
125.0	7
62.5	8
125.0	8
125.0	9
500.0	9
125.0	10
250.0	10
125.0	11

Model Details

The prediction model uses:

• Positive-Unlabeled (PU) learning framework
• Modified F1 score metric optimized for PU learning
• Features: satellite altitude, spreading factor, bandwidth, elevation angle, distance to station, antenna gain
• Training via Optuna hyperparameter optimization