PaperlayerView exact revision on arXiv

Source-linked AI summary

ARFT: A Synchronized Multimodal RF-Acoustic Dataset for Positioning in Distributed Environments

Daan Delabie, Jarne Van Mulders, Bert Pyck, Gustav Nilsson Gisleskog, Gilles Callebaut

arXiv:2609.02657v1eess.SPeess.AS

TL;DR

ARFT addresses the need for synchronized, aligned RF and acoustic measurements for distributed indoor positioning. It constructs and releases a Techtile campaign dataset with shared acquisition identifiers, multimodal data products, and processing workflows, including a baseline acoustic localization pipeline. The release contains 5011 tri-modal spatial samples, while documented acoustic baselines achieve mean 2D and 3D errors of 0.110 m and 0.130 m, respectively.

  • Problem

    ARFT addresses limited availability of synchronized RF-acoustic-position measurements for evaluating RF-only, acoustic-only, and joint indoor positioning workflows.

  • Method

    The paper constructs an open Techtile campaign release with co-located rover sensing, shared cycle identifiers, RF CSI products, acoustic waveform tensors, parsing code, and notebook-based workflows.

  • Results

    5011 tri-modal spatial samples form the aligned subset, while the baseline acoustic pipeline reports mean 2D error 0.110 m and mean 3D error 0.130 m.

  • Takeaways & Limitations

    ARFT supports spatial analysis, tile-level RF inspection, multimodal cycle alignment, and baseline acoustic localization when the appropriate completeness subset is used.

Abstract

from arXiv · show

This paper documents the acoustic-radio fusion in Techtile (ARFT) dataset, a synchronized measurement campaign for distributed wireless sensing and positioning in the Techtile testbed. Ultrasonic and radio frequency (RF) signals are simultaneously transmitted and captured at multiple positions in a 2D spatial grid inside the Techtile testbed. Each acquisition cycle corresponds to one rover stop, one position sample, one acoustic recording and one RF snapshot. The RF modality is recorded as per-host pilot measurements and released as channel state information (CSI) tensors for 42 antennas mounted at the ceiling of the room. The acoustic chirp is recorded with 91 synchronized microphones and transmitted by a synchronized quasi omni-directional speaker. The campaign spans 5011 spatial position samples spanning a 5.57 m by 2.89 m area with complete RF and acoustic data. We detail how the measurements are recorded, quantify per-experiment coverage, describe the RF and acoustic data contents, and explain how the aligned modalities support RF-only, acoustic-only, and joint positioning workflows. Furthermore, a static acoustic positioning pipeline that performs anchor selection, pulse-compression ranging, and least squares (LS) localization is elaborated.

I. INTRODUCTION

ARFT introduces a synchronized, room-scale RF-acoustic dataset for indoor positioning in Techtile. It combines co-located modalities, shared identifiers, and open processing materials to support reproducible multimodal research.

  • Motivation and contribution: Large-scale experimental datasets expose realistic synchronization errors, coverage gaps, and multimodal alignment constraints for communication, sensing, and positioning research.They also decouple algorithm development from scarce hardware testbed time.
  • Motivation and contribution: Techtile provides the room-scale distributed infrastructure in which ARFT measurements are collected.The supplied passages identify Techtile as an environment for communication, sensing, positioning, and acoustic research.
  • Motivation and contribution: ARFT co-locates an ultrasonic source and RF user antenna on one rover, aligning RF, acoustic, and ground-truth position measurements per acquisition cycle.Common experiment and cycle identifiers remove the need for geometric matching between independent campaigns.
  • Motivation and contribution: The dataset targets room-scale indoor positioning and enables RF-only, acoustic-only, and joint positioning benchmarks.Its RF measurements span 42 ceiling antennas, while acoustic recordings preserve responses from 91 synchronized microphones.
  • Motivation and contribution: ARFT packages acquisition software, rover control, signalling, measurement scripts, parsing code, processed data, and notebook-based analysis.The release therefore includes data and processing material rather than only final benchmark numbers.

II. PLATFORM AND ACQUISITION WORKFLOW

The ARFT platform combines a rover-mounted RF-acoustic payload with distributed Techtile infrastructure and coordinated positioning and measurement services.

  • Platform: Techtile uses 140 detachable tiles, Qualisys subcentimeter tracking, a 2D sampling robot, and 42 phase-coherent antennas operating at 920 MHz.The campaign rig combines an ultrasonic speaker and RF antenna on the rover.
  • Acquisition workflow: A central orchestrator coordinates rover movement, position acquisition, and RF and acoustic sampling.This coordination links the physical rover pose with the sensing operations.

1) Orchestator:

The orchestration layer executes each measurement cycle by sequencing rover motion, ground-truth position logging, acoustic capture, and RF synchronization.

  • Orchestator: The deployment layer distributes code and settings and starts long-lived services before orchestration begins the measurement cycle.The two control layers separate system setup from end-to-end acquisition.
  • Orchestator: Each cycle starts with movement to a commanded waypoint and a fresh Qualisys position sample at the rover stop.The passage then describes subsequent acoustic and RF acquisition steps.

2) Rover:

The rover carries the co-located sensing payload and follows a dense, reproducible multi-resolution scan while RF phase is interpreted relatively rather than absolutely.

  • Rover: Both the RF antenna and ultrasonic speaker are mounted on the rover’s xy-plotter.This co-location supports measurements from the same rover platform.
  • Rover: RF phase information is interpreted relative to other received phases, not as absolute phase.The transmit antenna is not phase-coherent with the receiving antennas, limiting absolute phase interpretation.
  • Rover: Fig. 2 depicts the rover trajectory and sampled measurement locations.The supplied caption identifies the figure’s content without specifying an outcome to compare.
  • Rover: The rover performs a square-spiral seven-sweep scan whose spacing decreases from 120 mm to approximately 50 mm.Progressively denser sweeps support spatial visualization while keeping the campaign operationally tractable.

3) Positioning Ground Truth:

Ground-truth rover positioning is supplied by a Qualisys tracking system, with one fresh 3D position recorded after every completed move.

  • 1.25 mm 3D standard deviation is reported for the Qualisys tracking accuracy.
  • Each rover stop receives one fresh (x, y, z) position update from the orchestrator.The rover moves only in the (x, y) plane, although z-coordinates are retained.

B. RF Orchestrator and Measurements

The RF orchestrator synchronizes measurements across 42 ceiling receiver tiles and converts each pilot response into phase-coherent complex CSI per rover stop.

  • 42 receiver tiles form the active Techtile ceiling RF aperture for the reciprocity workflow.
  • 920 MHz sampling uses a 250 kS/s rate and receiver gain of 80.
  • Each RF cycle waits for all tiles, publishes SYNC, and awaits the corresponding DONE quorum before completion.
  • Pilot phase and RMS amplitude are post-processed with cable-phase correction into one complex CSI value per ceiling antenna.A single physical stop therefore yields up to 42 phase-coherent complex CSI observations.

C. Acoustic Measurements

The acoustic system uses a rover-mounted ultrasonic source and 91 fixed microphones, with synchronized chirp captures organized into processed waveform tensors alongside RF and position records.

  • 91 distributed microphones receive a 30 ms, 20–40 kHz linear chirp from a rover-mounted quasi-omnidirectional ultrasonic speaker.The microphone acquisition window is twice the excitation length to preserve direct and multipath arrivals.
  • The 0.4 s reverberation time constrains acoustic measurements to a low update rate to exclude interference.
  • Nanosecond synchronization is provided by a shared DAQ reference clock and hardware start triggering.
  • Acoustic captures store chirp settings, microphone metadata, coordinates, and waveform samples before parsing into experiment-cycle-microphone waveform tensors.
  • RF parsing applies cable correction, joins measurements with rover positions, removes consecutive duplicate positions, and writes a merged NetCDF.The acoustic parser writes a separate per-experiment waveform NetCDF.

III. EXPERIMENTAL CAMPAIGN AND COVERAGE

The campaign comprises 12 named experiments and 5011 tri-modal spatial samples collected across partially overlapping spatial subregions with differing sample counts and densities.

  • 12 named experiments, EXP001–EXP012, make up the ARFT campaign.
  • 5011 tri-modal spatial samples are included in the dataset.
  • Recorded coordinates span approximately 0.81–6.87 m on the Qualisys x-axis and 0.71–3.60 m on the y-axis.
  • Experiments cover different spatial subregions and do not contain identical sample counts or densities.

IV. RECORDED DATA PRODUCTS AND POSITIONING CAPABILITIES

ARFT provides layered RF and acoustic data products linked by experiment and cycle identifiers, enabling inspection, model-based acoustic localization, and RF, acoustic, or joint positioning workflows. The release includes analysis-ready tensors, trajectory artifacts, and notebooks that demonstrate these capabilities.

  • Data products: Each acquisition cycle is represented in analysis-ready products with operational records converted into hierarchical NetCDF tensors.RF data use experiment, cycle, and hostname dimensions, while acoustic data preserve waveform samples by experiment, cycle, microphone, and sample index.
  • RF data: 5011 RF cycles yield 210068 tile-level complex CSI samples, with 4617 cycles containing all 42 ceiling-tile entries and 394 containing 41.The RF tensor reconstructs each observation as csi_real +j csi_imag and includes an explicit csi_available validity mask.
  • Acoustic data: The acoustic product preserves synchronized chirp waveforms whose direct-path timing, reverberation, and multipath signatures support fingerprinting, feature extraction, and model-based localization.Each trace corresponds to the same rover stop as its paired RF snapshot.
  • Multimodal alignment: Shared experiment_id and cycle_id keys align rover pose, tile-level RF channels, and raw microphone waveforms for RF-only, acoustic-only, or joint positioning.Figures 3 and 4 illustrate cycle-level acoustic waveform inspection alongside RF-position selection and phase/amplitude heatmaps across cycles.
  • Positioning pipeline: The acoustic pipeline filters signals, selects anchors, applies chirp pulse compression, estimates ranges using peak prominence, and computes least-squares rover positions.The current configuration uses up to 50 anchors per stop and a peak-prominence factor of 0.24; documented results cover 50 held-out paths and 550 rover stops.
  • Positioning evidence: 0.110 m mean 2D positioning error and 0.195 m P90, alongside 0.130 m mean 3D error and 0.217 m P90, summarize the calibrated acoustic baseline.The ranging records contain 27500 anchorwise distance estimates with mean absolute error 0.311 m; RF positioning remains exploratory because transmitter-receiver phase synchronization is absent.
  • Analysis workflow: The workflow includes generated training and held-out test walks, with scripts for trajectory inspection, CSI heatmaps, 42-receiver snapshots, joint inspection, and spatial phase-power movies.The release contains 2500 generated training walks and 50 held-out test walks, while Figures 5 and the notebook outputs expose trajectory and channel variation.

V. DISCUSSION AND INTERPRETATION CAVEATS

ARFT’s interpretation depends on distinguishing campaign-wide logs from the smaller analysis-ready merged subset, validating data through availability masks, and respecting the static, environment-specific measurement conditions.

  • The merged analysis set currently contains 9 of 12 experiments, so three logged experiments remain relevant to campaign coverage but lack processed RF and acoustic data.EXP001, EXP002, and EXP004 are excluded from the current merged set despite documenting physically traversed campaign areas.
  • The merged RF cycle_id axis is a shared union across experiments, and availability masks—not coordinate presence—identify valid measurements.Ignoring csi_available or position_available can treat absent data as zeros or mistake coordinate existence for measurement existence.
  • The campaign used static measurements in Techtile, requiring reverberant acoustic signals to be handled as static and limiting direct transfer to other rooms without hardware redesign.The stated environmental scope and hardware availability constrain scaling beyond the Techtile setting.

VI. CONCLUSION

ARFT is presented as a synchronized measurement campaign and open data release rather than a benchmark. Its documented coverage, processing, alignment, and baseline localization workflow support reproducible spatial and multimodal analysis when the appropriate completeness subset is used.

  • ARFT integrates rover motion, Qualisys positioning, acoustic sensing, and distributed ceiling-tile RF measurements in Techtile rather than presenting a benchmark.The release includes 12 experiments and a fully aligned tri-modal subset of 5011 cycle pairs.
  • The release contains 6735 rover stops, 5011 cycles in a merged 9-experiment RF dataset, 210068 CSI values, 5855 acoustic cycles, and 5011 aligned tri-modal cycle pairs.
  • ARFT supports spatial analysis, tile-level RF inspection, multimodal cycle alignment, and baseline acoustic localization when the appropriate completeness subset is used.The paper emphasizes documented coverage, processing, example scripts, and limitations as part of the dataset contribution.
Loading 2609.02657v1…