electricsheepafrica/radar-phased-array-signatures

---
license: apache-2.0
task_categories:
- tabular-classification
language:
- en
tags:
- radar
- phased-array
- multi-band
- long-range
- drone-detection
- micro-doppler
- synthetic-data
- hypersonic
- swarm-detection
- electronic-warfare
pretty_name: "Long-Range Phased-Array Radar Signatures"
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: data/features.parquet
---

# 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](https://huggingface.co/datasets/electricsheepafrica/radar-cuas-signatures) for full feature descriptions.

## Usage

```python
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

```bibtex
@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}
}
```