covcollab-eve-detection / manifest.json
ashen-navigator's picture
Add covert-collaboration Eve-detection dataset (data + minimal generator source)
b96185c verified
Raw
History Blame
10.9 kB
{
"artifact": "covcollab_eve_detection",
"schema_version": "1.0",
"title": "Covert-Collaboration Eve-Detection Dataset (controlled factorial)",
"created_utc": "2026-08-11T02:24:07Z",
"reproducibility": "bit-identical within the pinned environment below (float reductions are BLAS-dependent, so cross-machine bit-identity is NOT claimed); run `--verify` to regenerate from the seed and assert the per-split Y SHA-256 matches",
"environment": {
"git_sha": "239fac6f654aaa817384556930c2776b08e3e43b",
"numpy": "2.5.1",
"torch": "2.13.0",
"python": "3.14.2",
"platform": "macOS-26.3-arm64-arm-64bit-Mach-O"
},
"generation": {
"base_seed": 7000,
"per_group": 16,
"n_groups_core": 10,
"n_groups_ood": 3,
"sim_device": "cpu",
"sim_dtype": "torch.complex128",
"n_cells": 864,
"n_optimized_policies_solved": 54,
"optimized_psgd": {
"surrogate": "sphericity",
"steps": 40,
"batch": 96,
"n_eval": 256,
"gamma_bob_sinr_floor_db": 0.0,
"design_eve_snr_db": -10.0,
"design_delta_db": -12.0,
"design_bob_snr_db": 2.0
},
"wall_s": 76.49
},
"grid": {
"formats": [
"sc",
"ofdm",
"dfts_ofdm",
"otfs",
"afdm",
"ofdm_comb"
],
"arms": {
"none": "no collaborative matrix (diagonal identity assignment, no cross-user mixing)",
"random": "random feasible collaboration matrix W (fresh per group)",
"optimized": "PSGD sphericity-covert W (designed once per cell at the reference point)"
},
"power_note": "all arms share ||W||_F^2 = P_W = M, so arms differ in structure not received energy",
"M": {
"multicarrier": [
8,
16,
25
],
"single_carrier": [
8,
12,
16
],
"note": "SC ZC pilots cap orthogonal M at 16; both grids are length-3 so per-format counts are equal"
},
"Kd_grid_K_d": [
[
2,
1
],
[
2,
2
],
[
4,
1
],
[
2,
4
]
],
"channels_core": [
{
"family": "flat",
"n_taps": 1,
"eps": 0.0,
"nu_max": 0.0
},
{
"family": "multipath",
"n_taps": 4,
"eps": 0.0,
"nu_max": 0.0
},
{
"family": "doppler",
"n_taps": 4,
"eps": 0.1,
"nu_max": 0.0015625
}
],
"channels_ood": [
{
"family": "doppler",
"n_taps": 4,
"eps": 0.25,
"nu_max": 0.00390625
}
],
"n_rx_eve": 4,
"eve_snr_db_grid": [
-16.0,
-12.0,
-8.0,
-4.0
],
"delta_db_grid": [
-20.0,
-12.0,
0.0,
6.0
],
"T_samples": 320
},
"splits": {
"policy": "group-level: core cells split 80/10/10 of groups into train/val/test_iid so a split's policy realizations are disjoint; strong-Doppler cells -> test_ood (a distribution shift held out of the training channel distribution)",
"counts": {
"test_iid": 10368,
"test_ood": 10368,
"train": 82944,
"val": 10368
},
"train": {
"n": 82944,
"n_H1": 41472,
"by_format": {
"afdm": 13824,
"dfts_ofdm": 13824,
"ofdm": 13824,
"ofdm_comb": 13824,
"otfs": 13824,
"sc": 13824
},
"by_policy_arm": {
"none": 27648,
"optimized": 27648,
"random": 27648
},
"by_regime": {
"comparable": 20704,
"covert": 41888,
"detectable": 20352
},
"by_channel_family": {
"doppler": 27648,
"flat": 27648,
"multipath": 27648
}
},
"val": {
"n": 10368,
"n_H1": 5184,
"by_format": {
"afdm": 1728,
"dfts_ofdm": 1728,
"ofdm": 1728,
"ofdm_comb": 1728,
"otfs": 1728,
"sc": 1728
},
"by_policy_arm": {
"none": 3456,
"optimized": 3456,
"random": 3456
},
"by_regime": {
"comparable": 2400,
"covert": 5136,
"detectable": 2832
},
"by_channel_family": {
"doppler": 3456,
"flat": 3456,
"multipath": 3456
}
},
"test_iid": {
"n": 10368,
"n_H1": 5184,
"by_format": {
"afdm": 1728,
"dfts_ofdm": 1728,
"ofdm": 1728,
"ofdm_comb": 1728,
"otfs": 1728,
"sc": 1728
},
"by_policy_arm": {
"none": 3456,
"optimized": 3456,
"random": 3456
},
"by_regime": {
"comparable": 2528,
"covert": 5248,
"detectable": 2592
},
"by_channel_family": {
"doppler": 3456,
"flat": 3456,
"multipath": 3456
}
},
"test_ood": {
"n": 10368,
"n_H1": 5184,
"by_format": {
"afdm": 1728,
"dfts_ofdm": 1728,
"ofdm": 1728,
"ofdm_comb": 1728,
"otfs": 1728,
"sc": 1728
},
"by_policy_arm": {
"none": 3456,
"optimized": 3456,
"random": 3456
},
"by_regime": {
"comparable": 2528,
"covert": 5216,
"detectable": 2624
},
"by_channel_family": {
"doppler": 10368
}
}
},
"totals": {
"n_samples": 114048,
"n_cells": 864,
"n_files": 7,
"total_bytes": 1121412270,
"total_MiB": 1069.462,
"equal_per_format": true,
"by_format": {
"afdm": 19008,
"dfts_ofdm": 19008,
"ofdm": 19008,
"ofdm_comb": 19008,
"otfs": 19008,
"sc": 19008
},
"by_policy_arm": {
"none": 38016,
"optimized": 38016,
"random": 38016
},
"by_regime": {
"comparable": 28160,
"covert": 57488,
"detectable": 28400
}
},
"storage": {
"format": "parquet (zstd), fixed_size_list<float32> Y",
"reshape": "y_real / y_imag are length R*T=1280; reshape to (R,T)=(4,320); Y = y_real + 1j*y_imag",
"features_note": "model input x=[Re,Im,|Y|^2] is recomputed from Y (per-minibatch energy scale); not stored"
},
"schema": [
{
"name": "label",
"dtype": "B",
"meaning": "hypothesis: 1 = H1 (collaboration signal present), 0 = H0 (noise only)"
},
{
"name": "split",
"dtype": "s",
"meaning": "train | val | test_iid | test_ood"
},
{
"name": "format",
"dtype": "s",
"meaning": "waveform family: sc | ofdm | dfts_ofdm | otfs | afdm | ofdm_comb"
},
{
"name": "policy_arm",
"dtype": "s",
"meaning": "none (no collaborative matrix) | random | optimized"
},
{
"name": "policy_kind",
"dtype": "s",
"meaning": "identity | random | optimized:sphericity"
},
{
"name": "n_tx",
"dtype": "i",
"meaning": "M: number of collaborating transmit users"
},
{
"name": "n_msg_users",
"dtype": "i",
"meaning": "K: number of message-bearing users"
},
{
"name": "msg_dim",
"dtype": "i",
"meaning": "d: message length per user (Kd = K*d)"
},
{
"name": "kd",
"dtype": "i",
"meaning": "Kd: total message dimension"
},
{
"name": "n_rx_eve",
"dtype": "i",
"meaning": "N_E = R: Eve's antenna count (fixed = 4)"
},
{
"name": "n_samples_t",
"dtype": "i",
"meaning": "T: samples per block (fixed = 320); reshape y_real/y_imag to (R,T)"
},
{
"name": "eve_snr_db",
"dtype": "f",
"meaning": "Eve's average received SNR (dB) -- sets the observed noise level"
},
{
"name": "bob_snr_db",
"dtype": "f",
"meaning": "Bob's average received SNR (dB) = eve_snr_db - delta_db"
},
{
"name": "delta_db",
"dtype": "f",
"meaning": "Delta = SNR_E - SNR_B (dB); <0 covert (Eve weaker), >0 detectable"
},
{
"name": "regime",
"dtype": "s",
"meaning": "covert | comparable | detectable (from delta_db)"
},
{
"name": "policy_design_snr_db",
"dtype": "f",
"meaning": "optimized W design Eve-SNR (dB); NaN for none/random"
},
{
"name": "channel_family",
"dtype": "s",
"meaning": "flat | multipath | doppler"
},
{
"name": "n_taps",
"dtype": "i",
"meaning": "delay taps (1 = flat)"
},
{
"name": "nu_max",
"dtype": "f",
"meaning": "max normalized Doppler (cycles/sample); 0 for flat/multipath"
},
{
"name": "pdp_decay",
"dtype": "f",
"meaning": "exponential power-delay-profile decay"
},
{
"name": "noise_kind",
"dtype": "s",
"meaning": "receiver noise model (white for this artifact)"
},
{
"name": "nuisance",
"dtype": "b",
"meaning": "per-block RF nuisances (CFO/timing/phase/gain) applied"
},
{
"name": "constellation",
"dtype": "s",
"meaning": "data constellation (qpsk)"
},
{
"name": "cell_id",
"dtype": "I",
"meaning": "factorial cell index (0..n_cells-1)"
},
{
"name": "group_id",
"dtype": "i",
"meaning": "group index within the cell"
},
{
"name": "base_seed",
"dtype": "L",
"meaning": "master seed for the whole build"
}
],
"y_sha256_per_split": {
"train": "ce98056ef5f92c160958b1b7d1c9ff042b2bdeed16b0d47f1d37f754f22b970f",
"val": "b4b10bc37fdae0c73acf8a84bd32809a1a18092cef1fe9ee67cb83c68328c8d6",
"test_iid": "dbb5768bf3397992111d69970862ecae9e9c65fe49a9a2c08894c179394d4375",
"test_ood": "1ef71a5a69a9f4b7175a6e03c0e6e576e1c9945bfb8a0fcf6276b006a582011f"
},
"regeneration": {
"generate": "uv run --extra hf covcollab-eve-controlled --out huggingface/covcollab-eve-detection --base-seed 7000 --per 16 --groups-core 10 --groups-ood 3 --opt-steps 40 --opt-batch 96",
"verify": "uv run --extra hf covcollab-eve-controlled --out huggingface/covcollab-eve-detection --base-seed 7000 --per 16 --groups-core 10 --groups-ood 3 --opt-steps 40 --opt-batch 96 --verify"
},
"files": [
{
"file": "data/train-00000.parquet",
"split": "train",
"rows": 24576
},
{
"file": "data/train-00001.parquet",
"split": "train",
"rows": 24576
},
{
"file": "data/train-00002.parquet",
"split": "train",
"rows": 24576
},
{
"file": "data/train-00003.parquet",
"split": "train",
"rows": 9216
},
{
"file": "data/val-00000.parquet",
"split": "val",
"rows": 10368
},
{
"file": "data/test_iid-00000.parquet",
"split": "test_iid",
"rows": 10368
},
{
"file": "data/test_ood-00000.parquet",
"split": "test_ood",
"rows": 10368
}
]
}