The first edition of AmbiLoc contains RSS fingerprints of ambient FM, TV and GSM stations. For wider compatibility, the dataset is published in text-based CSV format.

Testbeds

TestbedDimensionsFloorsTest pointsAvg. spacingDurationSessions
Offices 100×50 m 1, 0, -2 33+36+16 4.6 m 12 months 23
Campus 80×80 m 1, 0 13+13 12.5 m 12 months 20
Apartment 14×7 m 3 37 1.4 m 3 months 6

Floorplans and sampling locations

Radio sampling

The fingerprints are based on a set of raw (baseband, IQ) radio samples collected by a USRP B-210. Each radio band was sampled for 2 s as described below:

BandCenter frequencySampling rate (complex)Antenna
FM 98 MHz 21 MHz VERT400
GSM-900 935 MHz, 950 MHz 20 MHz VERT900
DVB-T 498, 522, 554, 578, 594, 602 9.14 MHz VERT400

(For more details on experimental setup, please check this paper)

The raw radio samples allow extraction of arbitrary fingerprinting features, even those not foreseen at the time of data collection. If you would like to extend AmbiLoc with advanced non-RSS fingerprints, please let me know: Email address is protected from spam. Please enable JavaScript to reveal the email..

Structure

Dataset files are organized in a hierarchy of folders:
Testbed Name / Testbed Floor / Session ID / Coordinates / CSV files
Here, Session ID is the UTC time at the start of the measurement session (YYYYMMDD-hhmmss). Coordinates is the WGS-84 latitude/longitude of the test point (measurement location). The test points are predefined and fixed for the given testbed floor. (Note: the Apartment coordinates have been moved within 100m from the original position for privacy reasons).

RSS data

RSS measurements were calculated from the raw radio samples using GNU Radio toolkit with a GSM-specific module (gr-gsm). The RSS values are sampled at 10 Hz, resulting in 20 fingerprints per test point and per radio type.

BandChannel widthFingerprint size
FM 87.5–108.5 MHz 100 kHz 210
GSM-900 925–960 MHz 200 kHz 175
TV (DVB-T) 498 to 602 MHz 8 MHz 6

The fingerprints are stored in a matrix form in human-readable CSV (comma-separated values) files, one file for each radio type. Each row represents one full-width RSS fingerprint, where each column corresponds to one radio channel.

FM

For FM, the following data are available:

fm.rss
RSS fingerprints (20×210 matrix)
fm.freq
Center frequency of each channel (210-element vector)
fm.fm-pilot-snr
FM stereo pilot SNR (the level of 19-kHz band above its neighboring 17 and 21 kHz bands in the FM-demodulated spectrum). This can be used for distinguishing unused radio channels from those with active FM stereo broadcasts.

GSM-900

Due to the wide width of the GSM-900 band, full-band GSM fingerprints are composed from two raw radio samples. Cell ID parameters were extracted using the gr-gsm module of GNU Radio.

gsm900.rss
RSS fingerprints (20×175 matrix)
gsm900.freq
Center frequency of each channel (175-element vector)
gsm900.(mnc, mcc, lac, cell_id)
Cell ID parameters (175-element vectors). For empty/unrecognized channels, default to zero.

TV (DVB-T)

The active TV channels have been identified manually in the first days of data collection. The initial set of five channels has later been extended with another one. The missing RSS data of that channel in the early sessions is replaced by "nan" string.

dvb.rss
RSS fingeprints (20×6 matrix)
dvb.freq
Center frequency of each channel (6-element vector)

Metadata

Each session includes a JSON-formatted file (session-info.json) that describes the testbed, its state (populated or empty) and ground truth precision. In the early sessions, the ground truth was established manually and should be consistent within few decimeters. Later on, the receiver position was controlled using laser rangefinders, ensuring that the ground truth across sessions is consistent within few centimeters.

The metadata also includes weather reports published by the nearby Luxembourg Findel airport. Weather information includes temperature, humidity, precipitation and several other parameters that can be useful for analyzing weather impact on localization performance.

TestbedGround truthHuman presenceDistance to meteo station
Offices manual (coarse), laser (precise) empty / populated 3 km
Campus manual (coarse), laser (precise) populated only 4 km
Apartment manual (coarse) only empty / populated 7 km

Further details

This documentation is work in progress. If something seems to be missing, please email Email address is protected from spam. Please enable JavaScript to reveal the email..