electricsheepafrica/radar-phased-array-signatures

⚠️ Synthetic dataset — Parameterized from published SSA literature, not real observations. Not suitable for empirical analysis or policy inference.

Long-Range Phased-Array Radar Signatures (X/C/Ku-band)

Dataset Description

Physics-based synthetic radar signature dataset for long-range airspace threat detection using multi-band phased-array radar. Contains 6,600 labeled samples across 11 target classes including stealth drones, swarm formations, loitering munitions, hypersonic glide vehicles, and environmental clutter. Each class is observed across three radar bands (X-band 9.5 GHz, C-band 5.5 GHz, Ku-band 15 GHz), enabling multi-band fusion research.

Dataset Summary

Property	Value
Total samples	6,600 (200 per class x 3 bands)
Classes	11
Features	35 columns (31 numeric, 4 categorical/ID)
Radar bands	X-band (9.5 GHz), C-band (5.5 GHz), Ku-band (15 GHz)
Range coverage	1 km to 15 km
SNR range	6 to 55 dB
CPI	256 pulses
Environments	Clear, light rain, heavy rain, urban multipath, coastal
Jamming	None (60%), barrage noise (20%), DRFM (20%)
Format	Apache Parquet

Radar Band Configurations

Parameter	X-band	C-band	Ku-band
Frequency	9.5 GHz	5.5 GHz	15 GHz
Bandwidth	50 MHz	40 MHz	100 MHz
PRF	20 kHz	15 kHz	25 kHz
Antenna gain	35 dBi	30 dBi	38 dBi
Tx power	45 dBm	43 dBm	42 dBm
Range resolution	3.0 m	3.75 m	1.5 m
Velocity resolution	0.24 m/s	0.35 m/s	0.15 m/s
Noise figure	3.5 dB	4.0 dB	4.5 dB
Beamwidth (az)	3.0 deg	5.0 deg	2.0 deg

Target Classes

Class	Category	RCS (m^2)	Max Speed (m/s)	Micro-Doppler	Range Profile
loitering_munition_lr	Threat	0.02	55	Pusher prop, 267 Hz blade rate	5 +/- 3 km
stealth_drone	Threat	0.005	40	None (low observability)	8 +/- 4 km
small_uav_lr	Threat	0.3	50	2-rotor, 167 Hz blade rate	6 +/- 3 km
swarm_small	Threat	0.008/target	30	Swarm (3-8 members)	4 +/- 2 km
swarm_large	Threat	0.005/target	25	Swarm (20-60 members)	5 +/- 2.5 km
fast_munition	Threat	0.01	200	None (ballistic)	10 +/- 4 km
hypersonic_glide	Threat	0.1	2,000	None (Mach 6)	12 +/- 5 km
large_aircraft	Clutter	100	250	None	12 +/- 4 km
flock_lr	Clutter	0.0004/bird	20	Multi-wing, 4 Hz	3 +/- 1.5 km
weather_clutter	Clutter	distributed	15	Distributed, 500 Hz spread	4 +/- 2 km
ground_clutter	Clutter	50	5	Vibrating, 20 Hz	2 +/- 1 km

Signal Generation

The generation pipeline follows the same physics as the C-UAS dataset but with band-specific parameters:

1. Per-band radar equation: Received power computed using band-specific frequency, gain, and system losses. The RCS model applies aspect-dependent scintillation for ellipsoidal and cylindrical targets.

2. Band-specific micro-Doppler: Templates define per-class sideband injection parameters (blade rate, Doppler spread, amplitude fraction, modulation depth). Swarm targets inject multiple member returns with individual Doppler offsets.

3. Long-range propagation: At 15 km, atmospheric attenuation and rain absorption become significant. Two-way path loss is modeled as attenuation_dB = (0.01 + rain_rate) * range_km.

4. Multi-band RCS variation: The same target observed at different bands exhibits different RCS values due to wavelength-dependent scattering. This is captured through band-specific radar constants in the feature extraction.

Features

Identical 35-feature schema to the C-UAS dataset. The band information is encoded in the sample_id field (format: {class}_{band}_{index}_{uuid}). Key differences from D1:

• Range: 1-15 km (vs 100 m - 10 km)
• Velocity: Up to +/- 2,000 m/s (hypersonic class)
• RCS range: 0.005 m^2 (stealth drone) to 100 m^2 (large aircraft)

See the C-UAS dataset for full feature descriptions.

Usage

from datasets import load_dataset

ds = load_dataset("electricsheepafrica/radar-phased-array-signatures")
df = ds['train'].to_pandas()

# Extract band from sample_id
df['band'] = df['sample_id'].str.extract(r'_(x_band|c_band|ku_band)_')

# Band-specific analysis
for band in ['x_band', 'c_band', 'ku_band']:
    band_df = df[df['band'] == band]
    print(f"{band}: {len(band_df)} samples, {band_df['class_label'].nunique()} classes")

Data Quality

• Balanced: 200 samples per class per band (600 per class total)
• No NaN or Inf values
• RCS estimation correlates with ground truth (r ~ 0.81)
• Velocity ranges match class kinematics (up to +/- 2000 m/s for hypersonic)
• Per-band radar constants ensure physically consistent feature extraction

Citation

@misc{electricsheep2025phased,
  title={Long-Range Phased-Array Radar Signatures: Multi-Band Synthetic Features for Airspace Threat Detection},
  author={Electric Sheep Africa},
  year={2025},
  publisher={HuggingFace},
  url={https://huggingface.co/datasets/electricsheepafrica/radar-phased-array-signatures}
}