Minimal dataset for the manuscript
"Real-Time Reconstruction of the Eddy-Current Magnetic Field in a Local Region of a Maneuvering UUV Platform"
(Sensors, MDPI).

The dataset contains (1) the finite element solutions exported from the UUV
models described in Section 4.1 of the manuscript, and (2) the numerical results
from which every number in Tables 1, 3, 4 and 5 and Figures 2, 4, 5, 6 and 7 is
taken. The finite element model files themselves are several gigabytes each and
are not included; they are available on request from the corresponding author.

----------------------------------------------------------------------
1. Coordinate frame and units
----------------------------------------------------------------------
Body frame O-xyz as defined in Section 2.1: x along the longitudinal axis of the
hull, pointing aft; y vertically upward when the vehicle is level; z by the
right-hand rule. The origin lies on the longitudinal axis. Positions are in meters,
time in seconds, magnetic flux density in nanotesla (nT) unless the column name
ends in _T (tesla). For the transient cases (sections 2.1 to 2.5) all field values
are the total field in the body frame, that is, the background (geomagnetic) field
plus the eddy-current field; the eddy-current field is obtained by subtracting the
background vector given in the *_bg.csv file of the same folder (after converting
T to nT). For the static anomaly case (section 2.6) the files contain the anomaly
field itself (no background field is applied).

----------------------------------------------------------------------
2. fem_exports/  (finite element solutions)
----------------------------------------------------------------------
Observation sets used by all files:
  axis line   : 501 points on the longitudinal axis, x = -3.5 ... 2.5 m, y = z = 0
  offset line : 501 points on the line y = 0.1 m, z = 0, same x range
  volume grid : 3321 points covering the sensor candidate region,
                x = -2.70 ... -0.70 m, y = -0.20 ... 0.20 m, z = -0.20 ... 0.20 m,
                spacing 0.05 m in all three directions

File types (PREFIX denotes the folder prefix listed below):
  PREFIX_axis_Bx.csv, _By.csv, _Bz.csv   columns time_s, x_m, y_m, z_m, B?_nT   (axis line)
  PREFIX_y01_Bx.csv,  _By.csv, _Bz.csv   same columns                           (offset line)
  PREFIX_volume.csv                      columns time_s, x_m, y_m, z_m, Bx_nT, By_nT, Bz_nT (volume grid)
  PREFIX_volume_static.csv               columns x_m, y_m, z_m, dom, sigma_S_m
                                         dom: finite element domain index at the grid point;
                                         sigma_S_m: electrical conductivity at the grid point, S/m
                                         (3.03e7 S/m inside the aluminum structure of the nominal
                                         model, 2.727e7 S/m in the perturbed model, 1 S/m in the
                                         surrounding domain, which is treated as air)
  PREFIX_bg.csv                          columns time_s, gbox_T, gboy_T, gboz_T
                                         background (geomagnetic) vector in the body frame, tesla

2.1 Unit-excitation set (Section 4.1, mode extraction and step-response fitting)
  A_x/  prefix Ax    ramp-and-hold background field along the body x axis
  A_y/  prefix Ay    ramp-and-hold background field along the body y axis
  A_z/  prefix Az    ramp-and-hold background field along the body z axis
  The background field rises linearly from 0 to 50 microtesla during 0-1 s and is
  held constant during 1-2 s. Stored times (s): 0, 0.01, 0.02, 0.04, 0.06, 0.08,
  0.1, 0.2, 0.5, 0.9, 0.99, 1, 1.01, 1.02, 1.04, 1.06, 1.08, 1.1, 1.2, 2.
  Only the volume-grid files are provided for this set (no line or bg files; the
  background is the ramp described above, applied along the axis named by the folder).

2.2 Maneuver set at pi/9 rad/s (main verification, Tables 3 and 4, Figures 4, 6 and 7)
  B_yaw_pi9/    prefix Byaw20     yaw maneuver
  B_pitch_pi9/  prefix Bpitch20   pitch maneuver
  B_roll_pi9/   prefix Broll20    roll maneuver
  Background-field magnitude 50 microtesla. The angular rate rises linearly from 0
  to pi/9 rad/s during the first 0.1 s and is then held constant; the platform
  rotates through 19 degrees within 1 s. 20 stored times, 0.05 to 1.00 s in steps
  of 0.05 s. Each folder contains the 9 files listed above.

2.3 Maneuver set at pi/3 rad/s (angular-rate comparison and T-L calibration)
  B_turn/   prefix Bturn    yaw maneuver at pi/3 rad/s
  B_dive/   prefix Bdive    pitch maneuver at pi/3 rad/s
  B_roll/   prefix Broll    roll maneuver at pi/3 rad/s
  Background-field magnitude 50 microtesla; 10 stored times, 0.1 to 1.0 s in steps
  of 0.1 s. Each folder contains the 9 files listed above.

2.4 Combined reversing maneuver (Section 4.1, Table 3, Figures 2c and 5)
  R1_combined/  prefix Bcomb15
  Nominal model. The platform rotates about the fixed body-frame axis proportional
  to (0.40, 0.80, 0.45); the background field of 50 microtesla is directed along
  (-0.81, -0.30, 0.51). The angular rate rises linearly to pi/9 rad/s within 0.1 s,
  is held until 0.7 s, reverses linearly to -pi/9 rad/s between 0.7 and 0.9 s, and
  is held until 1.5 s. 30 stored times, 0.05 to 1.50 s in steps of 0.05 s. The
  folder contains the 9 files listed above; the bg file gives the background
  vector at every stored time.

2.5 Independent perturbed model (Sections 4.1 and 4.4)
  R1_perturbed/  prefix Bpert20
  Yaw maneuver identical to B_yaw_pi9 (same background field and stored times),
  solved on a model that differs from the nominal one in three respects: hull
  conductivity reduced by 10 percent to 2.727e7 S/m; a solid aluminum cylinder of
  radius 0.12 m and length 0.30 m added on the axis at x = -1.575 ... -1.275 m;
  new mesh (2,924,263 tetrahedra). The 126 volume-grid points inside the cylinder
  carry dom = 13 and sigma_S_m = 2.727e7 in Bpert20_volume_static.csv. The folder
  contains the 9 files listed above.

2.6 Static anomaly source (Section 4.4)
  R1_anomaly/  prefix Banom
  Stationary magnetostatic solution (scalar potential formulation) of the nominal
  hull geometry with a uniformly magnetized sphere of radius 0.10 m centered at
  (-1.80, -3.00, 0) m, remanent flux density (0, -0.05, 0) T, recoil permeability
  1 (equivalent dipole moment 166.7 A m^2 along -y); no background field; magnetic
  insulation on the outer boundary of the 20 m cubic air domain. The files give
  the anomaly field itself in nT (the time_s column is 0 for all rows):
    Banom_axis_Bx.csv, _By.csv, _Bz.csv     axis line
    Banom_y01_Bx.csv,  _By.csv, _Bz.csv     offset line
    Banom_volume.csv                        volume grid
    Banom_stations.csv                      columns station, x_m, y_m, z_m, Bx_nT, By_nT, Bz_nT
                                            (primary point and auxiliary location)
    Banom_meta.txt                          setup of the static anomaly case

----------------------------------------------------------------------
3. results/  (processed results, JSON)
----------------------------------------------------------------------
Each file is a nested JSON object of plain numbers and arrays produced by the
Python processing scripts. Maneuver keys: yaw20 / pitch20 / roll20 are the pi/9
maneuvers (19 degrees in 1 s); yaw60 / pitch60 / roll60 are the pi/3 maneuvers
(60 degrees in 1 s); comb15 is the combined reversing maneuver (1.5 s); pert20 is
the yaw maneuver on the perturbed model. Case keys: ideal / good / typical /
typical_cal are the ideal case, the favorable installation, the typical
installation, and the typical installation with calibrated scale factor and
misalignment of Table 3. Method keys: hyb = proposed method; ff = open-loop
feed-forward; array or array_m4 = full-coefficient estimation with the M=4
configuration; tl = Tolles-Lawson model (trunc = eddy-current term only, full6 =
complete model with symmetric induced term, full9 = complete model with full
induced term); hyb_cum = proposed method with the cumulative (not sample-by-sample)
estimate of the amplitude factor; tot_hyb / tot_unc = total-field residual after
compensation by the proposed method / without compensation. Error metrics are
defined in Section 4.1: "region" values are the region reconstruction error on
the volume grid, the other residuals are RMS residuals at the primary point in nT.

Files:
  v15_results.json   mode library (Table 1, Figure 2), sensor placement (Figure 3),
                     single-factor magnetometer error sweeps and offline-model
                     error sweeps (Table 4, Figure 7)
  v17_compare.json   three main comparison cases (ideal, favorable installation,
                     typical installation) for yaw and pitch, relocation case,
                     Monte Carlo samples (Table 3, Figures 4 and 6)
  v17_roll.json      the same cases for the roll maneuver (Table 3, roll column)
  v17_neg.json       amplitude error eps_s = -10 percent (Section 4.3)
  v18_extra.json     single-source injection for the T-L model and the proposed
                     method, Monte Carlo standard deviations, start-up-stage
                     diagnostics (Sections 4.3 and 4.4)
  v20_extra.json     typical installation with calibrated scale factor and
                     misalignment (Table 3), start-up residual energy fraction,
                     threshold-switching test (Section 4.3)
  r1_kappa.json      identifiability and variance of the amplitude factor: differential
                     signal norm, Monte Carlo and theoretical standard deviation of the
                     amplitude factor, settling times, noise sweep, threshold rule,
                     condition number and noise gain of full-coefficient estimation
                     (Sections 3.2, 4.2 and 4.3, Figure 4g,h)
  r1_adaptive.json   adaptive interference cancelation with a pair of magnetometers:
                     implementation A (as published) and B (geomagnetic change removed),
                     whole-maneuver and five-sample windows (Section 4.3, Table 3)
  r1_tl_full.json    complete T-L model calibrated on two and on five constant-rate
                     maneuvers, with condition numbers and relocation results
                     (Sections 4.2 and 4.3)
  r1_tl_full_s1.json complete T-L model with the leave-one-out calibration set that
                     includes the combined reversing maneuver: eddy-current and total
                     residuals, condition numbers, relocation (Table 3, Sections 4.2,
                     4.3 and 5)
  r1_combined.json   combined reversing maneuver: background-field check, excitation
                     statistics, projection errors onto the mode basis, online
                     reconstruction in the four cases, amplitude factor statistics,
                     residual energy of the transients (Sections 4.1 and 4.3, Table 3,
                     Figures 2c and 5)
  r1_perturbed.json  independent perturbed model as ground truth: grid statistics,
                     ratio to the nominal field, projection errors, scalar fits,
                     online reconstruction in the four installation cases with and
                     without injected model error, amplitude factor (Section 4.4)
  r1_anomaly.json    external anomaly by superposition: dipole scenarios (depth,
                     amplitude, speed, orientation), residuals, unexplained part of
                     the differential measurement, joint drift estimation
                     (Section 4.4, Table 5)
  r1_s3_check.json   stationary finite element solution with the inserted sphere
                     compared with the free-space dipole field and with the dipole
                     field including the images of the outer boundary, at the two
                     magnetometer positions, on the two lines and on the volume grid
                     (Section 4.4)

The processing scripts are available on request from the corresponding author.
