Data Associted With the Paper Titled "Online biomimetic electrolocation using causal multi-electrode voltage histories"

This dataset contains ten paired experimental trials acquired on a controlled biomimetic electrolocation platform. In each approximately 30 s trial, a moving signal-generator surrogate was observed simultaneously by a distributed nine-electrode measurement array and an overhead camera while moving in an 86 cm x 80 cm water-tank workspace. The electrode files contain recorded timestamps, ADC counts, converted voltages, and peak-to-peak voltage measurements. The camera files contain the two-dimensional image-plane trajectory of the moving surrogate at 50 frames/s. The paired streams support supervised evaluation of two-dimensional electrolocation, including causal online estimators and offline trajectory-reconstruction methods.

The archive contains 93,627 electrode records and 15,000 camera frames in total. Trial pairs are identified by matching file numbers: M_001.csv corresponds to trial01_tracks.csv, through M_010.csv and trial10_tracks.csv. Camera positions are provided in pixels and can be mapped to physical tank coordinates and interpolated to the electrode time axis using the calibration described below. A derived MATLAB file containing camera-derived reference positions and saved leave-one-trial-out predictions from the original benchmark models is also included. No human participants or live animals were used; the moving electric emitter was an artificial signal-generator surrogate.

Associated manuscript: "Online biomimetic electrolocation using causal multi-electrode voltage histories."

SUGGESTED SUBJECTS

Engineering; Electrical Engineering; Robotics; Machine Learning; Bio-inspired Sensing; Underwater Sensing

SUGGESTED KEYWORDS

biomimetic electrolocation; active electric sensing; multi-electrode sensing; voltage time series; camera tracking; sensor fusion; online position estimation; causal temporal modeling; underwater sensing; machine learning; trajectory data

FULL DATASET README

1. Overview

This archive provides the empirical data used to study two-dimensional online biomimetic electrolocation from causal multi-electrode voltage histories. The experimental platform comprised a water tank, a distributed electrode array, a moving signal-generator surrogate, and an overhead camera used to obtain an independent reference trajectory.

The dataset was designed as a controlled engineering benchmark. It measures the location-dependent voltage pattern generated by an artificial moving emitter and does not represent natural animal behavior. No human participants or live animals were used.

The ten trials may be used to develop and compare regression, sequence, filtering, inverse-search, and other tracking methods. The associated study evaluates models with leave-one-trial-out validation, so each complete trajectory is held out in turn rather than splitting adjacent frames from the same trial between training and testing.

2. Archive Contents

Raw electrode recordings:

M_001.csv
M_002.csv
M_003.csv
M_004.csv
M_005.csv
M_006.csv
M_007.csv
M_008.csv
M_009.csv
M_010.csv

Raw camera tracks:

trial01_tracks.csv
trial02_tracks.csv
trial03_tracks.csv
trial04_tracks.csv
trial05_tracks.csv
trial06_tracks.csv
trial07_tracks.csv
trial08_tracks.csv
trial09_tracks.csv
trial10_tracks.csv

Trial pairing is one-to-one by numerical identifier:

M_001.csv <-> trial01_tracks.csv
M_002.csv <-> trial02_tracks.csv
...
M_010.csv <-> trial10_tracks.csv

3. Trial Inventory

Trial  Electrode file  Electrode records  Camera file          Camera frames  Electrode duration (s)
1      M_001.csv       9,361              trial01_tracks.csv   1,500          29.973027
2      M_002.csv       9,366              trial02_tracks.csv   1,500          29.993882
3      M_003.csv       9,356              trial03_tracks.csv   1,500          29.959727
4      M_004.csv       9,355              trial04_tracks.csv   1,500          29.958234
5      M_005.csv       9,369              trial05_tracks.csv   1,500          29.996883
6      M_006.csv       9,365              trial06_tracks.csv   1,500          29.992062
7      M_007.csv       9,359              trial07_tracks.csv   1,500          29.974798
8      M_008.csv       9,367              trial08_tracks.csv   1,500          29.996128
9      M_009.csv       9,367              trial09_tracks.csv   1,500          29.994830
10     M_010.csv       9,362              trial10_tracks.csv   1,500          29.964581

Total electrode records: 93,627
Total camera frames: 15,000
Missing numeric values detected in the deposited CSV files: none

4. Electrode CSV Data Dictionary

Each M_###.csv file contains a header row and 28 columns: one recorded time column followed by three quantities for each of nine electrode channels.

t_s
Recorded time in seconds for the electrode stream.

<channel>_counts
Raw analog-to-digital converter count for the named channel. Counts are dimensionless integer-valued measurements as recorded by the acquisition system.

<channel>_V
Voltage corresponding to the ADC measurement, in volts.

<channel>_Vpp
Recorded peak-to-peak voltage quantity for the channel, in volts. The associated electrolocation analyses use the nine _Vpp columns as the model input.

The channel groups, in file order, are:

ADCA2
ADCA3
ADCA4
ADCA5
ADCB2
ADCB3
ADCB4
ADCB5
ADCC2

The resulting nine-channel analysis vector is ordered as:

ADCA2_Vpp, ADCA3_Vpp, ADCA4_Vpp, ADCA5_Vpp,
ADCB2_Vpp, ADCB3_Vpp, ADCB4_Vpp, ADCB5_Vpp, ADCC2_Vpp

The electrode files contain approximately 9,360 records over 30 s, corresponding to an average row rate of approximately 312.2 Hz. The recorded t_s field contains repeated timestamp values in the raw files. Users should preserve row order and should not estimate the acquisition interval from the median of all timestamp differences, because many consecutive differences are zero. The original analysis uses the recorded time values for camera interpolation and the ordered samples for causal histories.

5. Camera-Track CSV Data Dictionary

Each trial##_tracks.csv file has no header and contains exactly 1,500 rows and two columns:

Column 1: horizontal image coordinate p_x, in pixels
Column 2: vertical image coordinate p_y, in pixels

The files contain the tracked image-plane location of the moving signal-generator surrogate. Camera coordinates increase to the right and downward. The camera rate is 50 frames/s, and frame timestamps are defined as:

t_camera[m] = m / 50 seconds, for m = 0, 1, ..., 1499.

Thus, each camera trajectory spans nominal times from 0 to 29.98 s. The camera rows are not one-to-one with the electrode rows; the two streams must be aligned by time.

6. Camera-to-Tank Coordinate Mapping

The physical tank workspace used in the analysis is:

Width W = 86 cm
Height H = 80 cm

The image crop/calibration bounds are:

x pixels: 290 to 1190
y pixels: 1 to 720

For a camera point (p_x, p_y), physical coordinates are computed as:

x_cm = (p_x - 290) * 86 / (1190 - 290)
y_cm = (720 - p_y) * 80 / (720 - 1)

The y coordinate is inverted because image y increases downward, while the tank coordinate system uses y increasing upward. After conversion, coordinates are clipped to the tank bounds:

0 <= x_cm <= 86
0 <= y_cm <= 80

These values are camera-derived reference positions. Their accuracy is limited by the camera tracking and spatial calibration; the archive does not provide a separate measurement of camera-reference uncertainty.

7. Synchronizing Camera and Electrode Streams

For trial i:

1. Load the corresponding M_###.csv file and retain its t_s vector.
2. Load the camera file with the matching trial number as the pixel trajectory.
3. Convert the camera pixels to tank coordinates using the equations above.
4. Assign camera times at 50 frames/s, starting at t = 0 s.
5. Linearly interpolate x_cm and y_cm onto the electrode t_s values.
6. Use linear extrapolation only when an electrode timestamp lies slightly outside the camera time range.

This procedure produces one two-dimensional camera-derived reference position for each electrode record. The deposited Python loader implements the same mapping and was checked against the reference positions stored in the MATLAB result file; differences were at numerical floating-point precision.

8. Licensing and Attribution

Before publication, select and state an explicit reuse license in Dataverse. Replace the bracketed text below with the final license and publication information.

If these data or the accompanying code are used in research, please cite both the Dataverse dataset DOI and the associated manuscript.

Contact

Ismail Uyanik
Department of Electrical and Electronics Engineering
Hacettepe University
Ankara, Turkiye

ismailuyanik@hacettepe.edu.tr