CISR24 storage schema
HDF5
Each split file contains:
| Dataset | Dtype | Shape | Meaning |
|---|---|---|---|
/iq |
float32 | [N, 2, 1024] |
I and Q channels |
/class-id |
uint16 | [1, N] |
IDs 0 through 23 |
/snr-db |
float32 | [1, N] |
SNR in dB |
/sample-index |
uint32 | [1, N] |
global 1-based sample index |
Root attributes include release_name, split, fs_hz, record_len,
total_count, and effective_snr_basis. The public loader also recognizes the
legacy underscore aliases for provenance files, but canonical releases use the
hyphenated names above to match the paper.
No normalization, random crop, or test-time augmentation is applied when reading the fixed 1024-sample records.
Manifest
Each manifest row maps one HDF5 row to its taxonomy, generation condition, and waveform parameters. Canonical field names use lowercase kebab case, including:
sample-id, split, h5-rel-path, h5-dataset, h5-index,
class-id, super-class, class-name, waveform-family,
snr-db, fs-hz, n-samples, seed, active-len, start-offset,
channel-profile, release-name, iid-or-ood, effective-snr-basis
Waveform-specific fields include symbol-rate-hz, sps, nfft, cp-len,
n-active, chirp-bw-hz, chirp-dir, pulse-width-us, barker-len,
costas-len, frank-order, p4-len, rrc-beta, ofdm-num-symbols,
costas-df-hz, costas-perm-id, chip-samples, gmsk-bt,
cpfsk-modulation-index, and cpfsk-tone-spacing-hz. Barker and Costas may distribute a target active length
across chips by one-sample differences; for those classes, chip-samples is
NaN and chip-samples-min/chip-samples-max record the actual bounds.
Non-applicable values are encoded as NaN.