Update dataset card to v2 (7 classes, 800 samples, 96.9% accuracy)

README.md

@@ -13,44 +13,68 @@ tags:
 - machine-learning
 - telecommunications
 - software-defined-radio
-pretty_name: RTL-ML RF Signal Classification Dataset
+pretty_name: RTL-ML RF Signal Classification Dataset v2
 size_categories:
 - 100K<n<1M
 ---
 
-# RTL-ML Dataset
-
-> ⚠️ **v1.0 Dataset Notice**: Community feedback identified that several signal classes have suboptimal SNR due to antenna/gain configuration during capture. The classifier achieves stated accuracy but is partially learning frequency-specific noise characteristics rather than pure signal features. v2.0 with improved captures is in progress.
+# RTL-ML Dataset v2
 
 ## Dataset Summary
 
-This dataset contains 240 validated RF signal samples captured using an RTL-SDR Blog V4 dongle. It's designed for training machine learning models to classify common RF signals.
+This dataset contains **800 validated RF signal samples** captured using an RTL-SDR Blog V4 dongle on an Indiedroid Nova (RK3588S). Designed for training machine learning models to classify common RF signals.
 
-**Total Size:** 1.9 GB  
-**Samples:** 240 (30 samples × 8 classes)  
-**Format:** NumPy arrays (.npy files)  
-**Sample Rate:** 1.024 MSPS  
-**Sample Duration:** 1 second per capture  
+**Samples:** 800 (7 classes)
+**Format:** NumPy arrays (.npy files) — each file is a dict with IQ data + metadata
+**Sample Rate:** 1.024 MSPS
+**Sample Duration:** 0.5 seconds per capture
+**Quality Gates:** DC removal, auto-gain, 6 dB minimum SNR, per-class validation
 
 ## Signal Classes
 
 | Class | Frequency | Count | Description |
 |-------|-----------|-------|-------------|
-| ADS_B | 1090 MHz | 30 | Aircraft transponder signals |
-| APRS | 144.39 MHz | 30 | Amateur radio position reporting |
-| FM_broadcast | 88-108 MHz | 30 | Commercial FM radio stations |
-| ISM_sensors | 433.92 MHz | 30 | Wireless sensors & remote controls |
-| NOAA_APT | 137.5 MHz | 30 | Weather satellite imagery |
-| NOAA_weather | 162.4 MHz | 30 | Weather radio broadcasts |
-| noise | Various | 30 | Background RF noise baseline |
-| pager | 152.84 MHz | 30 | POCSAG pager transmissions |
-
-## Validation Metrics
-
-- **ISM Sensors:** 20.6x burst ratio (strong on/off keying)
-- **NOAA Weather:** 14.4 dB SNR (clear signal)
-- **Pager/APRS:** 12.7 dB SNR (good quality)
-- **Model Accuracy:** 87.5% on test set
+| FM_broadcast | 88.5, 93.3, 98.7, 101.1, 105.7 MHz | 200 | Commercial FM radio (5 stations) |
+| NOAA_weather | 162.4 MHz | 100 | Weather radio broadcasts |
+| APRS | 144.39 MHz | 100 | Amateur radio position reporting |
+| pager | 152.84 MHz | 100 | POCSAG pager transmissions |
+| ISM_sensors | 433.92 MHz | 100 | Wireless sensors & remote controls |
+| FRS_GMRS | 462.5625 MHz | 100 | Family/general mobile radio |
+| noise | 145.0 MHz | 100 | Background RF noise baseline |
+
+## What Changed from v1
+
+- **7 classes** (removed ADS-B — 1090 MHz out of R828D tuner range; removed NOAA APT — decommissioned Aug 2025; added FRS/GMRS)
+- **800 samples** (up from 240) with 100+ per class
+- **DC offset removal** on every capture
+- **Auto-gain calibration** per frequency
+- **6 dB SNR gate** — rejects weak/empty captures
+- **Per-class quality validators** (bandwidth, burst ratio, packet detection)
+- **Temporal train/test split** — first 80% train, last 20% test (no data leakage)
+- **Multi-frequency FM** — trained on 5 stations for frequency-invariant classification
+- **Metadata in every file** — center_freq, sample_rate, timestamp, label, snr_db, version
+
+## Model Performance
+
+- **Random Forest:** 96.9% accuracy (155/160 test samples correct)
+- **Temporal split:** No data leakage between train and test
+- **Cross-frequency FM:** Generalizes to unseen FM stations
+
+## Sample Format
+
+Each .npy file contains a dict:
+```python
+{
+    'samples': np.array([...], dtype=complex64),  # IQ data
+    'center_freq': 98700000.0,
+    'sample_rate': 1024000.0,
+    'timestamp': '2026-01-15T14:23:01',
+    'label': 'FM_broadcast',
+    'duration': 0.5,
+    'snr_db': 17.5,
+    'version': 'v2'
+}
+```
 
 ## Usage
 
@@ -65,8 +89,8 @@ dataset_path = snapshot_download(
 )
 
 # Load a sample
-sample = np.load(f"{dataset_path}/datasets_validated/ADS_B_0.npy")
-print(f"Signal shape: {sample.shape}")  # (1048576,) complex64
+data = np.load(f"{dataset_path}/datasets_validated/FM_broadcast/FM_broadcast_0.npy", allow_pickle=True).item()
+print(f"Signal: {data['label']}, SNR: {data['snr_db']:.1f} dB, Freq: {data['center_freq']/1e6:.1f} MHz")
 ```
 
 ## Dataset Structure
@@ -74,40 +98,27 @@ print(f"Signal shape: {sample.shape}")  # (1048576,) complex64
 ```
 rtl-ml-dataset/
 └── datasets_validated/
-    ├── ADS_B_0.npy ... ADS_B_29.npy       (30 files)
-    ├── APRS_0.npy ... APRS_29.npy         (30 files)
-    ├── FM_broadcast_0.npy ... _29.npy     (30 files)
-    ├── ISM_sensors_0.npy ... _29.npy      (30 files)
-    ├── NOAA_APT_0.npy ... NOAA_APT_29.npy (30 files)
-    ├── NOAA_weather_0.npy ... _29.npy     (30 files)
-    ├── noise_0.npy ... noise_29.npy       (30 files)
-    └─��� pager_0.npy ... pager_29.npy       (30 files)
+    ├── FM_broadcast/          (200 files from 5 frequencies)
+    ├── NOAA_weather/          (100 files)
+    ├── APRS/                  (100 files)
+    ├── pager/                 (100 files)
+    ├── ISM_sensors/           (100 files)
+    ├── FRS_GMRS/              (100 files)
+    └── noise/                 (100 files)
 ```
 
-Each `.npy` file contains:
-- **Shape:** (1048576,) - 1 second @ 1.024 MSPS
-- **Dtype:** `complex64` (I/Q samples)
-- **Size:** ~8.4 MB per file
-
 ## Hardware
 
-- **SDR:** RTL-SDR Blog V4 ($39.95)
-- **Computer:** Indiedroid Nova 16GB or Raspberry Pi 4/5 (8GB+ recommended)
-- **Antenna:** Telescopic dipole (included)
-
-## Model Performance
-
-When trained with Random Forest (100 trees):
-- **Overall Accuracy:** 87.5%
-- **Perfect Classes:** ADS-B, FM, ISM, NOAA APT, Weather, Pager (100%)
-- **Challenging:** APRS ↔ Noise confusion (sparse packets)
+- **SDR:** RTL-SDR Blog V4 ($39.95) — **requires [RTL-SDR Blog driver fork](https://github.com/rtlsdrblog/rtl-sdr-blog)** for R828D tuner support
+- **Computer:** Indiedroid Nova 16GB ($179.95)
+- **Antenna:** Telescopic dipole (included with V4)
 
 ## Citation
 
 ```bibtex
-@misc{rtl-ml-dataset,
+@misc{rtl-ml-dataset-v2,
   author = {TrevTron},
-  title = {RTL-ML Dataset: Validated RF Signal Captures},
+  title = {RTL-ML Dataset v2: Validated RF Signal Captures},
   year = {2026},
   publisher = {Hugging Face},
   howpublished = {\url{https://huggingface.co/datasets/TrevTron/rtl-ml-dataset}}
@@ -116,19 +127,9 @@ When trained with Random Forest (100 trees):
 
 ## License
 
-MIT License - Free for commercial and non-commercial use.
+MIT License — Free for commercial and non-commercial use.
 
 ## Related
 
 - **Code:** [github.com/TrevTron/rtl-ml](https://github.com/TrevTron/rtl-ml)
 - **Blog:** [unland.dev/blog/building-ai-radio-scanner-rtl-sdr-machine-learning](https://unland.dev/blog/building-ai-radio-scanner-rtl-sdr-machine-learning)
-- **Hardware Guide:** [Indiedroid Nova Setup](https://github.com/TrevTron/rtl-ml/blob/main/docs/HARDWARE_SETUP.md)
-
-## Contributions
-
-Captured in Temecula, CA (Southern California) using:
-- Clear line of sight to multiple signal sources
-- Validated with spectral analysis and manual inspection
-- All samples meet minimum SNR requirements (>10 dB for modulated signals)
-
-For questions or improvements, see the [GitHub repository](https://github.com/TrevTron/rtl-ml).