PaperlayerView exact revision on arXiv

Source-linked AI summary

Memoryless Techniques and Wireless Technologies for Indoor Localization with the Internet of Things

Sebastian Sadowski, Petros Spachos, Konstantinos Plataniotis

arXiv:2005.01877v2eess.SPcs.NI

TL;DR

Indoor localization faces large errors from indoor multipath and complex environments, and there is no IPS standard. The paper compares localization techniques across wireless technologies and experimental settings; KNN is reported as the most accurate and precise overall.

  • Problem

    Indoor multipath and complex environments can cause large position errors, and no standards are in place for an IPS.

    Source: §I · p. 1

  • Method

    The study compares localization techniques using commercially available hardware and multiple wireless technologies.

    Source: §I · p. 1, §VIII · p. 9

  • Results

    KNN with k = 4 was the most accurate and precise technique overall, followed by Naive Bayes.

    Source: §VIII · p. 9

  • Takeaways & Limitations

    KNN and Naive Bayes require database computations with complexity O(mn), while trilateration has complexity O(1) and is described as the fastest.

    Source: §VIII · p. 9, B. Discussion · p. 9

Abstract

from arXiv · show

In recent years, the Internet of Things (IoT) has grown to include the tracking of devices through the use of Indoor Positioning Systems (IPS) and Location Based Services (LBS). When designing an IPS, a popular approach involves using wireless networks to calculate the approximate location of the target from devices with predetermined positions. In many smart building applications, LBS are necessary for efficient workspaces to be developed. In this paper, we examine two memoryless positioning techniques, K-Nearest Neighbor (KNN), and Naive Bayes, and compare them with simple trilateration, in terms of accuracy, precision, and complexity. We present a comprehensive analysis between the techniques through the use of three popular IoT wireless technologies: Zigbee, Bluetooth Low Energy (BLE), and WiFi (2.4 GHz band), along with three experimental scenarios to verify results across multiple environments. According to experimental results, KNN is the most accurate localization technique as well as the most precise. The RSSI dataset of all the experiments is available online.

I. INTRODUCTION

Indoor localization must contend with complex environments and the absence of IPS standards. This work compares localization techniques and wireless technologies across three rooms, identifying KNN fingerprinting as highly accurate.

I. INTRODUCTION

  • Indoor multipath and complex environments can cause large position errors, and no IPS standards are in place.Room layout, size, and obstacles can make a system designed for one location unsuitable elsewhere.Source: §I · p. 1
  • The study compares KNN, Naive Bayes, and trilateration using Zigbee, BLE, and 2.4 GHz WiFi across three environments.The environments represent low, average, and high interference levels.Source: §I · p. 1
  • The experiments produced an RSSI dataset available online.The scenarios span rooms with different dimensions and interference.Source: §I · p. 1
  • KNN fingerprinting is identified as a highly accurate localization technique.The introduction does not quantify this result or compare precision across techniques.Source: §I · p. 1

II. RELATED WORK

Prior localization comparisons report accuracy and computation trade-offs, while the relative merits of wireless technologies remain insufficiently compared. The paper frames localization methods as model-based or survey-based and describes trilateration as a distance-based model.

II. RELATED WORK

  • Prior comparisons found KNN highly accurate but slow to compute, while other fingerprinting comparisons favored Isoline.A separate WiFi study reported higher accuracy for Nearest-Neighbor than multi-trilateration, alongside greater computational demand.Source: §II · p. 2
  • Earlier work examined fingerprinting complexity and WiFi, BLE, and Zigbee use, but the paper identifies a need to compare wireless technologies directly,.The authors compare KNN and Naive Bayes with trilateration across three scenarios and the three technologies.Source: §II · p. 2

III. LOCALIZATION TECHNIQUES

  • Model-based localization calculates positions mathematically, whereas survey-based systems build an environment database before deployment.The passage classifies lateration as model-based and fingerprinting as survey-based,.Source: §III · p. 2

A. Trilateration

  • Trilateration estimates receiver location from distances to transmitters, using RSSI as a low-cost way to estimate distance.Three transmitters are required for a 2D position and four for a 3D position; unreliable RSSI from interference, obstacles, and reflections can impair accuracy.Source: A. Trilateration · pp. 2–3

B. Fingerprinting

Fingerprinting stores RSSI measurements at known locations, then matches measurements from an unknown location against that database. KNN compares RSSI values using Euclidean distance and averages the positions of the nearest matches.

B. Fingerprinting

  • Fingerprinting has a sampling phase that stores RSSI readings and locations, followed by matching unknown-location readings against the database.The algorithm returns the stored position that best matches the unknown reading.Source: B. Fingerprinting · p. 3
  • Fingerprinting can be time-consuming and labor-intensive to map, and environmental changes may require recreating its database.RSSI values are influenced by obstacles, reflections, and multipath effects.Source: B. Fingerprinting · p. 3

1) K-Nearest Neighbor:

  • KNN compares unknown-location RSSI readings with stored fingerprints using Euclidean distance, then selects the k nearest matches.The final location estimate averages the (x,y) positions of those matches.Source: 1) K-Nearest Neighbor · p. 3

2) Naive Bayesian Estimation:

Naive Bayes compares an unknown location's RSSI measurements with stored values and estimates position using probabilities. It chooses the stored position with the highest likelihood, assuming RSSI values follow Gaussian distributions based on training measurements.

2) Naive Bayesian Estimation:

  • Naive Bayes estimates an unknown device's position by comparing RSSI measurements with stored values and selecting the most likely position.It uses probabilities of stored RSSI values to determine location likelihood.Source: 2) Naive Bayesian Estimation · pp. 3–4
  • The posterior probability for a stored position is calculated from the likelihood and prior, normalized by the marginal likelihood.The procedure sums location-specific probabilities to calculate the marginal likelihood for an observed RSSI value.Source: 2) Naive Bayesian Estimation · pp. 3–4
  • The method assumes possible RSSI values follow Gaussian distributions estimated from training-phase measurements.The database contains n stored locations.Source: 2) Naive Bayesian Estimation · p. 4

C. Error Analysis

The error analysis compares calculated positions with real positions using Mean Squared Error, then averages errors across tests to compare localization techniques.

C. Error Analysis

  • Mean Squared Error compares each calculated position (xcalc, ycalc) with its real position (xreal, yreal).The paper uses this measure to assess localization accuracy.Source: C. Error Analysis · p. 4
  • The authors average the errors across all tests and compare techniques by how close their results are to the actual location.Source: C. Error Analysis · p. 4

D. Computational Complexity

The paper contrasts the computational cost of trilateration with database-search methods and describes the wireless technologies used in the experiments. Technology selection depends on application requirements, and the implementation details vary across Zigbee and BLE.

D. Computational Complexity

  • Trilateration has complexity O(1), while KNN and Naive Bayes have complexity O(mn) because they search stored RSSI points across transmitters.A larger database can increase computations and runtime, even as it provides more candidate points for matching.Source: D. Computational Complexity · p. 4

IV. WIRELESS TECHNOLOGIES

  • No standard wireless technology is prescribed; designers select technology according to application requirements.The paper notes that energy requirements differ among IoT technologies.Source: §IV · p. 4

A. Zigbee - IEEE 802.15.4

  • The Zigbee setup used Series 2 XBee antennas and an Arduino Uno to control information flow because the XBees had limited processing power.Source: A. Zigbee - IEEE 802.15.4 · p. 4

B. Bluetooth Low Energy (BLE)

  • The BLE experiments used iBeacon-configured Gimbal beacons with distinct Minor values, while a Raspberry Pi received signals and measured RSSI.One-way beacon transmission allows the receiver, rather than the transmitters, to be used for tracking.Source: B. Bluetooth Low Energy (BLE) · pp. 4–5

C. WiFi (2.4 GHz) - IEEE 802.11.N

WiFi RSSI was collected with Raspberry Pi 3 devices, and the experiments compared localization across environments with low, high, and average interference. The first two scenarios used small meeting rooms with different fingerprint-map densities.

C. WiFi (2.4 GHz) - IEEE 802.11.N

  • Raspberry Pi 3 devices used their onboard WiFi antennas to create a WLAN and measure RSSI from signals of interest.RSSI was collected by polling for available signals.Source: C. WiFi (2.4 GHz) - IEEE 802.11.N · p. 5

V. EXPERIMENTAL SCENARIOS

  • Three scenarios provided low, high, and average interference environments for comparing techniques and wireless technologies.Source: §V · p. 5

A. First Scenario - Small room with low interference.

B. Second Scenario - Small room with high interference.

C. Third Scenario - Large room with average interference.

The third scenario tested a large computer lab with average interference, while the procedure maintained line of sight and averaged repeated RSSI scans. The test area was limited to preserve a comparison with trilateration.

C. Third Scenario - Large room with average interference.

  • The large computer lab contained people, wireless equipment, BLE devices, and WiFi-enabled computers, creating obstructions, interference, and reflections.The room measured 10.8 × 7.3 m, approximately 79 m2.Source: C. Third Scenario - Large room with… · p. 6
  • Only part of the lab was fingerprinted because line of sight was needed to compare trilateration with fingerprinting techniques.The scenario used 40 reference points and 16 randomly selected test points.Source: C. Third Scenario - Large room with… · p. 6

VI. EXPERIMENTAL PROCEDURE

  • The experiment was designed to determine which localization technique produced the highest accuracy.Source: §VI · p. 6

A. Process

  • Transmitters and receivers were kept in line of sight and placed on tables to approximate a device carried in a pocket or bag.Each reference location used 300 scans, followed by a moving average of RSSI values; the same measurement procedure was used at test locations to reduce interference effects.Source: A. Process · p. 6

B. Device configuration

The experiments used shared transmission settings and calibrated a path-loss model for each technology and scenario. Device orientation and placement were kept consistent to limit RSSI variation.

B. Device configuration

  • Transmission power and interval affect accuracy, response time, and device power consumption, so the settings balance localization updates with device lifetime.Source: B. Device configuration · p. 6
  • Devices broadcast at −10 dBm and 2 Hz, corresponding to a 0.5 s interval.The power setting was the common preconfigured value for the XBee and Gimbal devices; the interval was chosen for quick location updates.Source: B. Device configuration · p. 6

C. Path-loss model

  • Trilateration converted RSSI to distance using path-loss parameters calibrated for each wireless technology and scenario.RSSI was measured at 18 distances, then curve fitting produced models and parameters summarized in Table II.Source: C. Path-loss model · p. 6
  • Keeping device orientation constant and centering devices at collection points was intended to reduce RSSI error.Device direction can significantly affect RSSI readings.Source: C. Path-loss model · p. 6

D. Techniques configuration

The study configured trilateration with Kalman-filtered RSSI and used k = 4 for KNN. Across the tested scenarios and technologies, KNN had the greatest overall accuracy and precision, while trilateration performed worst.

D. Techniques configuration

VII. RESULTS AND DISCUSSION

A. Experimental Results

  • KNN performed better than Naive Bayes and trilateration in all three scenarios; BLE led in Scenarios 1 and 2, while WiFi led in Scenario 3.Technology precision varied with scenario: WiFi had lower variance than BLE in Scenario 1, and BLE had lower variance than Zigbee in Scenario 2.Source: A. Experimental Results · pp. 7–8, Figure 7 · p. 8
  • KNN with k = 4 produced the greatest overall accuracy and precision, with 1.602 m deviation and 0.6662 m variance; trilateration had the worst overall results.Source: A. Experimental Results · pp. 7–8, Table III · p. 8

B. Discussion

Across the tested environments, KNN was the most accurate and precise technique, followed by Naive Bayes, while trilateration was least accurate and inconsistent. The discussion also weighs these outcomes against runtime, database requirements, and how fingerprints and interference shape performance.

B. Discussion

  • KNN was most accurate and precise, followed by Naive Bayes, while trilateration was least accurate and produced large variances.For KNN and Naive Bayes, estimates were limited to database locations; trilateration could estimate any location using transmitter positions and the environment’s path-loss.Source: B. Discussion · pp. 8–9
  • KNN was simpler to use than Naive Bayes, which must recompute stored RSSI probabilities after database changes.KNN estimates location by averaging the k database points with the smallest Euclidean distances; new access points can be integrated after creating a new database.Source: B. Discussion · pp. 8–9
  • KNN requires a database, and its runtime grows with testing points, access points, and database size; closer fingerprints may improve accuracy but can create redundant RSSI entries.If fingerprints are too close together, similar RSSI readings may hinder overall performance.Source: B. Discussion · p. 9
  • Trilateration executed fastest but was inaccurate and inconsistent; its simple RSSI-to-distance calculation needs no database and is not limited to stored reference points.Source: B. Discussion · p. 9
  • Accuracy and precision generally aligned across algorithms, but Scenario 1’s Zigbee KNN had the worst accuracy and highest precision among its configurations.Source: B. Discussion · p. 9
  • Trilateration errors were larger in the research lab than in the small meeting rooms, and the authors linked poorer performance to environmental interference.The small-room scenarios differed in variance, while the lab’s larger space changed the effect of wall reflections.Source: B. Discussion · p. 9
  • KNN errors were similar across scenarios, but the 16-fingerprint Scenario 2 had the lowest error, while the 49-fingerprint Scenario 1 had the largest.The authors relate accuracy to fingerprint number and placement; Scenario 1 nevertheless had the lowest variance.Source: B. Discussion · p. 9

VIII. CONCLUSION

The conclusion compares KNN, Naive Bayes, and trilateration across three rooms and three wireless technologies. KNN with k = 4 was the most accurate and precise overall, while trilateration had the best complexity; the experimental RSSI dataset is open-source.

VIII. CONCLUSION

  • KNN with k = 4 was most accurate and precise overall, followed by Naive Bayes; trilateration was worst overall.Source: §VIII · p. 9
  • KNN and Naive Bayes had complexity O(mn), while trilateration had O(1) and required little running time.Source: §VIII · p. 9
  • The experiments compared the techniques across three rooms and three wireless technologies: Zigbee, BLE, and WiFi.The comparison covered accuracy, precision, and complexity.Source: §VIII · p. 9
  • The experimental RSSI dataset was made open-source and is presented as an indicator for selecting indoor localization techniques in smart buildings.Source: §VIII · p. 9

Related papers

Loading 2005.01877v2…