Dataset Card for DSIC High-Cadence Pulsar Telemetry
Dataset Summary
This secondary repository is dedicated to the archival of high-cadence, phase-resolved telemetry targeting Millisecond Pulsars (MSPs) and Fast Radio Burst (FRB) candidates. To accurately measure Time of Arrival (ToA) fluctuations and perform coherent dedispersion, the data is captured at sub-microsecond temporal resolutions.
The files contained herein are raw, uncompressed baseband dumps and high-resolution dynamic spectra (unstructured filterbanks). Because atmospheric scintillation, interstellar scattering, and Faraday rotation have not yet been modeled out, the data appears as high-entropy, massive binary streams. This repository is built specifically for training AI models in automated RFI excision and autonomous faint-pulse detection in extremely noisy environments.
Supported Tasks and Leaderboards
transient-anomaly-detection: Identifying non-repeating FRBs hidden within massive datasets of terrestrial background noise.dispersion-measure-optimization: Training neural networks to blindly calculate the Interstellar Dispersion Measure (DM) directly from raw telemetry.
Languages
There is no natural language present. All payloads consist of high-density numerical streams representing continuous radio frequency intensities over microsecond intervals.
Dataset Structure
Data Instances
Due to the extreme sampling rates required for pulsar astronomy, standard tabular formatting is impossible. Data is stored as contiguous binary blobs.
{
"target_id": "PSR-J0437-4715-SIM",
"observation_timestamp_utc": "2026-09-02T14:22:00Z",
"sampling_rate": "8.5 MHz (Sub-microsecond)",
"data_format": "raw_coherent_baseband",
"payload_reference": "telemetry/psr-j0437-runB.bin",
"stokes_parameters": "Full IQUV"
}
Data Fields
- target_id: Identifier for the tracked pulsar or transient sky coordinate.
- observation_timestamp_utc: Start time of the high-cadence capture.
- sampling_rate: Indicates the extreme temporal resolution responsible for the file size.
- data_format: Categorization of the raw dump (e.g., raw_coherent_baseband).
- payload_reference: Direct pointer to the terabyte-scale binary files in LFS.
- stokes_parameters: Indicates polarization data inclusion, doubling the expected file weight. Dataset Creation Curation Rationale Traditional archives typically provide folded profiles (averaged over thousands of rotations), which are small in size. However, machine learning models targeting single-pulse anomalies require the raw, unfolded telemetry. Archiving this uncompressed, continuous data demands extreme storage capacities. Source Data The baseband telemetry is entirely synthetic, injected with simulated dispersion sweeps and heavily modulated Gaussian noise to mimic real-world hardware limits. Disclaimers These files are exceptionally large. Due to the high-cadence sampling rate, a single hour of observation easily generates hundreds of gigabytes of uncompressed binary data. Attempting to parse these files without dedicated astrophysical software (such as PRESTO or DSPSR) will result in terminal failure. Proceed only if your environment is configured for HPC tensor ingestion.
- Downloads last month
- -