README.md

metadata

license: cc-by-4.0
task_categories:
  - time-series-forecasting
  - tabular-classification
  - other
tags:
  - uav
  - navigation
  - gps-denied
  - sensor-fusion
  - electronic-warfare
  - integrity-monitoring
  - imu
  - gnss
  - anomaly-detection
  - synthetic-data
language:
  - en
size_categories:
  - 10M<n<100M

HiFi-NAV-38M: High-Fidelity Synchronized IMU-GPS Dataset for Navigation Integrity Monitoring under Electronic Warfare Conditions

Dataset Description

HiFi-NAV-38M is a large-scale synthetic dataset of synchronized Inertial Measurement Unit (IMU) and GPS sensor readings for unmanned aerial vehicle (UAV) navigation under nominal and electronic warfare (EW) conditions. The dataset provides time-aligned multi-sensor data with full ground truth and threat labels, enabling research in navigation integrity monitoring, GPS spoofing/jamming detection, and resilient sensor fusion.

This dataset was generated using a purpose-built high-fidelity simulation engine developed by the author for navigation research in contested environments. Custom dataset configurations (scenario mix, threat profiles, sensor parameters, scale) are available on request.

Key Properties

Property	Value
Total Samples	37,890,000
Total Scenarios	2,549
IMU Rate	100 Hz
GPS Rate	10 Hz
Simulation Fidelity	High-fidelity physics-based
Coordinate Frame	NED (North-East-Down)
File Format	JSONL (gzip compressed)
Total Size	6.4 GB (compressed)
License	CC-BY-4.0

What Makes This Dataset Unique

Existing public GPS spoofing datasets (e.g., TEXBAT, OAKBAT) provide GPS-only recordings without synchronized inertial data. This fundamentally limits their use for multi-sensor integrity monitoring research, where cross-validation between GPS and IMU is the primary detection mechanism.

HiFi-NAV-38M provides all of the following simultaneously:

• High-rate IMU (accelerometer + gyroscope) at 100 Hz
• GPS with constellation-level metadata (satellites, HDOP, C/N0) at 10 Hz
• Navigation filter output (estimated position, velocity, errors)
• Ground truth state (position, velocity, attitude)
• Per-sample threat labels (type, active/inactive, GPS denied)
• Flight phase labels

This enables research that requires synchronized multi-sensor data with ground truth under adversarial conditions -- a combination not available in existing public datasets.

Threat Scenarios

Threat Type	Scenarios	Share	Description
none (Nominal)	1,060	41.6%	Clean flight with nominal sensor noise
jamming	735	28.8%	RF interference causing GPS signal degradation or denial
spoofing	495	19.4%	False GPS signals injected to manipulate position solution
drfm	259	10.2%	Digital Radio Frequency Memory -- coherent replay attack

Scenarios include varying flight profiles (waypoint navigation, orbit, hover) with threat onset at different mission phases.

Data Structure

Each line in the JSONL files is a single timestamped sample with the following structure:

Metadata

Field	Type	Description
meta.scenario	string	Scenario identifier
meta.scenario_type	string	Flight profile type
meta.threat_type	string	none, jamming, spoofing, or drfm
meta.sample_index	int	Index within scenario

Ground Truth

Field	Type	Unit	Description
timestamp_s	float	s	Time since scenario start
ground_truth.pos_ned_m	float[3]	m	True position [North, East, Down]
ground_truth.vel_ned_mps	float[3]	m/s	True velocity [North, East, Down]
ground_truth.attitude_euler_rad	float[3]	rad	True attitude [roll, pitch, yaw]

IMU Sensor

Field	Type	Unit	Description
sensors.imu.accel_body_mps2	float[3]	m/s^2	Specific force in body frame [x, y, z]
sensors.imu.gyro_body_rads	float[3]	rad/s	Angular rate in body frame [x, y, z]

IMU readings include realistic noise characteristics (bias, random walk).

GPS Sensor

Field	Type	Unit	Description
sensors.gps.valid	bool	-	GPS fix available
sensors.gps.pos_ned_m	float[3]	m	Measured position [North, East, Down]
sensors.gps.vel_ned_mps	float[3]	m/s	Measured velocity [North, East, Down]
sensors.gps.num_satellites	int	-	Number of visible satellites
sensors.gps.hdop	float	-	Horizontal Dilution of Precision
sensors.gps.cn0_dbhz	float	dB-Hz	Carrier-to-noise density ratio

During jamming, GPS validity degrades. During spoofing and DRFM, GPS reports manipulated positions while appearing valid.

Navigation Filter Output

Field	Type	Unit	Description
navigation.estimated_pos_ned_m	float[3]	m	Estimated position
navigation.estimated_vel_ned_mps	float[3]	m/s	Estimated velocity
navigation.position_error_m	float	m	Euclidean position error vs. ground truth
navigation.velocity_error_mps	float	m/s	Euclidean velocity error vs. ground truth

Labels

Field	Type	Description
labels.threat_active	bool	Whether a threat is currently active
labels.threat_type	string	Active threat type or none
labels.gps_denied	bool	Whether GPS is denied at this timestep
labels.flight_phase	string	Current flight phase (e.g., cruise)

Usage

Loading the Data

import gzip
import json

def load_samples(filepath, max_samples=None):
    samples = []
    with gzip.open(filepath, 'rt') as f:
        for i, line in enumerate(f):
            if max_samples and i >= max_samples:
                break
            samples.append(json.loads(line))
    return samples

# Load first 1000 samples from first file
samples = load_samples("uav_data_00000.jsonl.gz", max_samples=1000)

# Access fields
for s in samples[:5]:
    print(f"t={s['timestamp_s']:.2f}s | "
          f"threat={s['labels']['threat_type']} | "
          f"pos_err={s['navigation']['position_error_m']:.2f}m | "
          f"sats={s['sensors']['gps']['num_satellites']}")

Streaming Large Files

import gzip
import json
from pathlib import Path

data_dir = Path(".")
for gz_file in sorted(data_dir.glob("uav_data_*.jsonl.gz")):
    with gzip.open(gz_file, 'rt') as f:
        for line in f:
            sample = json.loads(line)
            if sample['labels']['threat_active']:
                # Handle threat sample
                pass

Extracting NumPy Arrays

import numpy as np
import gzip, json

imu_data = []
with gzip.open("uav_data_00000.jsonl.gz", 'rt') as f:
    for line in f:
        s = json.loads(line)
        imu_data.append(s['sensors']['imu']['accel_body_mps2'])

accel = np.array(imu_data)  # Shape: (N, 3)
print(f"Mean gravity: {np.mean(np.linalg.norm(accel, axis=1)):.3f} m/s^2")

Generation

This dataset was produced using a high-fidelity UAV navigation simulation engine developed by the author. The engine models multi-sensor dynamics under nominal and adversarial conditions with physics-based fidelity.

Custom datasets with tailored scenario configurations, threat profiles, sensor parameters, or larger scale are available on request. Contact the author for collaboration or licensing inquiries.

Intended Uses

• Navigation integrity monitoring: Training and evaluating detection algorithms for GPS manipulation
• Sensor fusion research: Developing IMU-GPS fusion methods robust to adversarial conditions
• Anomaly detection benchmarks: Binary/multi-class classification of threat conditions
• Resilient PNT research: Studying degradation patterns under electronic warfare

Limitations

• Synthetic data -- real-world sensor artifacts (temperature drift, vibration coupling, antenna patterns) are modeled statistically, not from hardware measurements
• Single-vehicle scenarios (no multi-agent or swarm configurations)
• NED frame relative to scenario start (no absolute geodetic coordinates)

Citation

@dataset{foss2026hifnav,
  author    = {Foss, David Tom},
  title     = {{HiFi-NAV-38M}: High-Fidelity Synchronized {IMU-GPS} Dataset
               for Navigation Integrity Monitoring under Electronic Warfare Conditions},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/chkmie/hifi-nav-38m}
}

Author

David Tom Foss Independent Researcher ORCID: 0009-0004-0289-7154 Contact: david@foss.com.de

Custom dataset generation and research collaboration inquiries welcome.

License

This dataset is released under CC-BY-4.0. You are free to share and adapt the data for any purpose, provided appropriate credit is given.