Partial Discharge Signals in Insulated Power Cables with Time-of-Arrival Annotations

Summary
Dataset of univariate voltage time series from insulated power cables with manual time-of-arrival (ToA) annotations for partial discharge (PD) pulses. Built for research and benchmarking of end-to-end ToA estimators and time-domain reflectometry (TDR) localization. Records are 10 microsecond windows sampled at 100 MS/s with 10-bit resolution; each window may contain zero, one, or multiple PD pulses. Data cover controlled laboratory tests and real-world field measurements on energized cables with strong EMI.

Scope and sources
• Supervised ToA (SUP-TOA): labeled set for training/validation/testing; includes lab and field signals.
• Operational Evaluation — Laboratory (OE-LAB): shielded lab per IEC 60270.
• Operational Evaluation — Field (OE-FIELD): industrial environment on energized cables with realistic EMI.

Quantitative summary
SUP-TOA: 1,660 signals (all with PD), 5,145 annotated pulses.
OE-LAB: 240 signals (all with PD), 1,440 annotated pulses.
OE-FIELD: 8,135 signals, 1,251 with PD, 2,502 annotated pulses total.
Each signal has 1,000 samples (10 microseconds at 100 MS/s).

Acquisition
• Laboratory: shielded setup, dedicated grounding, coupling capacitor 1 nF, measuring impedance 50 ohms (typical bandwidth 100 kHz–30 MHz), 60 Hz source; high SNR.
• Field: in situ with HFCT sensors (nominal 1–80 MHz, approx. 19 mV/mA at 5 MHz into 50 ohms); lower SNR due to EMI.
• Acquisition system: Techimp Aquila, 100 MS/s, analog UWB filter \~10–30 MHz, 10-bit ADC.

Annotation protocol
For each pulse, define a non-overlapping support interval; find the absolute-amplitude peak in that interval; set ToA as the earliest sample where the absolute amplitude reaches 50 percent of that peak. This is polarity-invariant and robust to gain variation. Windows may include incident and reflected pulses.

File format
HDF5 files per subset (SUP-TOA, OE-LAB, OE-FIELD). Root attributes include: schema_version, created_at, dataset_name, dataset_id, license, doi, organization, time_unit = “s”, signal_unit = “V”, fs_Hz = 100000000.0, adc_bit_depth = 10, description, sha256_dataset.
Groups:
• /signals/data: float32, N x 1000 (each row is one window).
• /signals/experiment_id: text per row.
• /signals/environment: “laboratory” or “field”.
• /signals/phase: electrical phase per row.
• /annotations/pulses/indices: variable-length int64 with ToA sample indices; convert to time by dividing by fs_Hz.
Amplitudes are in volts at the acquisition point; no per-window normalization in the files.

Quality and use
Each file includes a SHA-256 hash for integrity. A single, documented labeling protocol ensures consistency. Recommended uses: training and validation of ToA estimators; evaluation with tolerance-aware metrics (precision, recall, F1, MAE in samples); single-ended TDR localization (incident and first reflection); robustness studies versus SNR, dispersion, and impulsive interference.