Device Signal Strength Self-Calibration Using Histograms

Introduction - Motivation Device Calibration - Device Diversity Device Signal Strength Self-Calibration - Manual Calibration Self-Calibration using Histograms - RSS Histograms - Self-Calibration Method

Performance ∗ † ∗ Evaluation Christos Laoudias , Robert Pich´e and Christos Panayiotou - Measurement Setup - Experimental Results ∗KIOS Research Center for Intelligent Systems and Networks, University of Cyprus † Conclusions Tampere University of Technology, Tampere, Finland - Concluding Remarks

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Outline

Introduction - Motivation Introduction Device Calibration - Device Diversity - Manual Calibration Device Calibration Self-Calibration - RSS Histograms - Self-Calibration Method Self-Calibration Performance Evaluation - Measurement Setup Performance Evaluation - Experimental Results Conclusions - Concluding Remarks Conclusions

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 I Diﬀerent devices do not report RSS values in the same way

I The WiFi standard (IEEE 802.11) deﬁnes the RSS Indicator (1 byte integer) for measuring RSS in [0 255] I Each vendor’s implementation is limited up to RSSImax I RSSI is mapped to power values in dBm internally by the device driver (proprietary information) I Even worse: same chipsets may not report the same RSS values due to diﬀerent antennas or packaging

I Using a new device for positioning is feasible, but the RSS values are not compatible with the radiomap

I Best accuracy is guaranteed only if the user carries the same device during positioning, otherwise calibration is required

I Existing calibration methods do not ﬁt well in real-time positioning scenarios

Motivation of our work

I RSS is intended for determining the signal quality and not for positioning purposes Introduction - Motivation Device Calibration - Device Diversity - Manual Calibration Self-Calibration - RSS Histograms - Self-Calibration Method Performance Evaluation - Measurement Setup - Experimental Results Conclusions - Concluding Remarks

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 I Using a new device for positioning is feasible, but the RSS values are not compatible with the radiomap

I Best accuracy is guaranteed only if the user carries the same device during positioning, otherwise calibration is required

I Existing calibration methods do not ﬁt well in real-time positioning scenarios

Motivation of our work

I RSS is intended for determining the signal quality and not for positioning purposes Introduction - Motivation I Different devices do not report RSS values in the same way Device Calibration I The WiFi standard (IEEE 802.11) defines the RSS - Device Diversity - Manual Calibration Indicator (1 byte integer) for measuring RSS in [0 255] Self-Calibration I Each vendor’s implementation is limited up to RSSImax - RSS Histograms - Self-Calibration I RSSI is mapped to power values in dBm internally by Method the device driver (proprietary information) Performance Evaluation I Even worse: same chipsets may not report the same - Measurement Setup - Experimental RSS values due to different antennas or packaging Results Conclusions - Concluding Remarks

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 I Best accuracy is guaranteed only if the user carries the same device during positioning, otherwise calibration is required

I Existing calibration methods do not ﬁt well in real-time positioning scenarios

Motivation of our work

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 I Existing calibration methods do not ﬁt well in real-time positioning scenarios

Motivation of our work

I Best accuracy is guaranteed only if the user carries the same device during positioning, otherwise calibration is required

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Motivation of our work

I Best accuracy is guaranteed only if the user carries the same device during positioning, otherwise calibration is required

I Existing calibration methods do not ﬁt well in real-time positioning scenarios

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Device Diversity

Introduction - Motivation Device Calibration - Device Diversity - Manual Calibration Self-Calibration - RSS Histograms - Self-Calibration Method Performance Evaluation - Measurement Setup - Experimental Results Conclusions - Concluding Remarks

Source: K. Kaemarungsi (2006)

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 I Limited Applicability: (i) User needs to be familiar with the indoor area and (ii) a considerable data collection eﬀort is required

Good News: Linearity between RSS values

−20 −20

−30 −30

−40 −40 Introduction - Motivation −50 −50

Device Calibration −60 −60 - Device Diversity −70 −70 - Manual Calibration

Mean RSS from HP iPAQ [dBm] −80 −80 Self-Calibration Mean RSS from Asus eeePC [dBm]

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Good News: Linearity between RSS values

−20 −20

−30 −30

−40 −40 Introduction - Motivation −50 −50

Device Calibration −60 −60 - Device Diversity −70 −70 - Manual Calibration

Mean RSS from HP iPAQ [dBm] −80 −80 Self-Calibration Mean RSS from Asus eeePC [dBm]

- RSS Histograms −90 RSS pairs −90 RSS pairs Cloud center Cloud center - Self-Calibration Least−squares Fit Least−squares Fit Method −100 −100 −100 −90 −80 −70 −60 −50 −40 −30 −20 −100 −90 −80 −70 −60 −50 −40 −30 −20 Performance Mean RSS from HTC Flyer [dBm] Mean RSS from Samsung Nexus S [dBm] Evaluation - Measurement Setup - Experimental Results Manual Calibration: Collect several colocated RSS pairs at Conclusions I - Concluding Remarks known locations and estimate the linear coefficients through least squares (2) (1) ¯rij = α12¯rij + β12 I Limited Applicability: (i) User needs to be familiar with the indoor area and (ii) a considerable data collection effort is required International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Can we do it more efficiently?

Introduction Objectives - Motivation Device Calibration - Device Diversity I Fully automatic approach with short calibration time - Manual Calibration Self-Calibration I Runs concurrently with positioning while the user walks - RSS Histograms - Self-Calibration around Method Performance I No user intervention or tedious data collection Evaluation - Measurement Setup - Experimental Results Idea Conclusions - Concluding Remarks I Perform device self-calibration on-the-ﬂy using histograms of RSS values observed simultaneously with positioning

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 RSS Histograms

0.05 0.05

0.045 0.045

0.04 0.04

0.035 0.035

Introduction 0.03 0.03

- Motivation 0.025 0.025 Probability Probability Device Calibration 0.02 0.02 0.015 0.015 - Device Diversity - Manual Calibration 0.01 0.01 0.005 0.005

Self-Calibration 0 0 −100 −90 −80 −70 −60 −50 −40 −30 −20 −10 −100 −90 −80 −70 −60 −50 −40 −30 −20 −10 - RSS Histograms Mean RSS Value Mean RSS Value - Self-Calibration Method Figure: HP iPAQ (left) and Asus eeePC (right) Performance Evaluation

- Measurement Setup 0.05 1 - Experimental Results 0.045 0.9 0.04 0.8

Conclusions 0.035 0.7 HP iPAQ Asus eeePC - Concluding Remarks 0.03 0.6 HTC Flyer 0.025 0.5 Probability 0.02 Probability 0.4

0.015 0.3

0.01 0.2

0.005 0.1

0 0 −100 −90 −80 −70 −60 −50 −40 −30 −20 −10 −100 −90 −80 −70 −60 −50 −40 −30 −20 −10 Mean RSS Value Mean RSS Value Figure: HTC Flyer (left) and Empirical cdfs (right)

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 1. Create the ecdf of the reference device from the radiomap 2. Create and update the ecdf of the new device by using s(k) 3. Fit a linear mapping between the reference and new device to obtain (α, β) by using “representative” ecdf values

4. Transform the observed RSS values with ˜sj (k) = αsj (k) + β 5. Estimate location `ˆ(k) with any ﬁngerprint-based algorithm

Self-Calibration Method

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 2. Create and update the ecdf of the new device by using s(k) 3. Fit a linear mapping between the reference and new device to obtain (α, β) by using “representative” ecdf values

4. Transform the observed RSS values with ˜sj (k) = αsj (k) + β 5. Estimate location `ˆ(k) with any ﬁngerprint-based algorithm

Self-Calibration Method

Reference Device User Device Device ecdf Calibration ecdf Introduction (,)αβ - Motivation Device Calibration Transformation sk( ) - Device Diversity - Manual Calibration sk%() Self-Calibration Positioning - RSS Histograms r lˆ()k - Self-Calibration i Algorithm Method Radiomap Performance Evaluation - Measurement Setup - Experimental Results Conclusions 1. Create the ecdf of the reference device from the radiomap - Concluding Remarks

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 3. Fit a linear mapping between the reference and new device to obtain (α, β) by using “representative” ecdf values

4. Transform the observed RSS values with ˜sj (k) = αsj (k) + β 5. Estimate location `ˆ(k) with any ﬁngerprint-based algorithm

Self-Calibration Method

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 4. Transform the observed RSS values with ˜sj (k) = αsj (k) + β 5. Estimate location `ˆ(k) with any ﬁngerprint-based algorithm

Self-Calibration Method

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 5. Estimate location `ˆ(k) with any ﬁngerprint-based algorithm

Self-Calibration Method

4. Transform the observed RSS values with ˜sj (k) = αsj (k) + β

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Self-Calibration Method

4. Transform the observed RSS values with ˜sj (k) = αsj (k) + β 5. Estimate location `ˆ(k) with any ﬁngerprint-based algorithm International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Inverse ecdf Linear ﬁtting

0.9

0.8 HP iPAQ HTC Flyer 0.7 Introduction 0.6 - Motivation 0.5 Device Calibration - Device Diversity Probability 0.4 - Manual Calibration 0.3

Self-Calibration 0.2

- RSS Histograms 0.1 - Self-Calibration Method 0 -100 -90 -80 -70 -60 -50 -40 -30 -20 -10 Performance Mean RSS Value Evaluation - Measurement Setup - Experimental Results I F (x) gives the probability that the RSS value is less than x, Conclusions F −1(y) returns the RSS value that corresponds to the y-th - Concluding Remarks cdf percentile

I Fr (x) and Fu(x) are the ecdfs of the reference and user device −1 −1 I Fr (y) = αFu (y) + β, y ∈ {0.1, 0.2,..., 0.9} I (α, β) are initialized to (1, 0) and updated periodically (e.g. every 10 sec) thereafter, while the user is walking International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Inverse ecdf Linear ﬁtting

0.9

0.8 HP iPAQ HTC Flyer 0.7 Introduction 0.6 - Motivation 0.5 Device Calibration - Device Diversity Probability 0.4 - Manual Calibration 0.3

Self-Calibration 0.2

0.9

0.8 HP iPAQ HTC Flyer 0.7 Introduction 0.6 - Motivation 0.5 Device Calibration - Device Diversity Probability 0.4 - Manual Calibration 0.3 -87 dBm -66 dBm Self-Calibration 0.2

Introduction - Motivation Device Calibration - Device Diversity - Manual Calibration START Self-Calibration - RSS Histograms END - Self-Calibration Method Performance Evaluation - Measurement Setup - Experimental Results 2 I 560 m oﬃce, 9 WiFi APs, 5 devices (1 HP iPAQ PDA, 1 Conclusions - Concluding Remarks Asus eeePC laptop, 1 HTC Flyer Android tablet, 2 Android smartphones)

I Training Data: 105 reference locations, 20 ﬁngerprints per location (2100 in total) with each device for comparison

I Testing Data: Route with 2 segments, 96 test locations, 1 ﬁngerprint per location, route sampled 10 times International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Experimental Results – 10 Routes

7 Introduction

- Motivation 6

Device Calibration 5 - Device Diversity - Manual Calibration 4

Self-Calibration 3 - RSS Histograms - Self-Calibration 2 Method Mean Positioning Error Per Route [m] 1 Performance Evaluation 0 No Calibration Self−Calibration Manual Calibration Device−Specific - Measurement Setup - Experimental Results Conclusions 1 - Concluding Remarks 0.9

0.8 HP iPAQ HTC Flyer 0.7

0.6

0.5

Probability 0.4

0.3

0.2

0.1

0 -100 -90 -80 -70 -60 -50 -40 -30 -20 -10 Mean RSS Value Figure: HTC Flyer user with HP iPAQ radiomap International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Experimental Results – 10 Routes

3.5

Introduction 3 - Motivation 2.5 Device Calibration - Device Diversity 2 - Manual Calibration 1.5 Self-Calibration - RSS Histograms 1 - Self-Calibration Mean Positioning Error Per Route [m] Method 0.5 Performance 0 Evaluation No Calibration Self−Calibration Manual Calibration Device−Specific - Measurement Setup - Experimental Results

Conclusions 1 0.9 - Concluding Remarks 0.8

0.7

0.6

0.5

Probability 0.4

0.3

0.2

0.1 Asus eeePC HTC Flyer 0 -100 -90 -80 -70 -60 -50 -40 -30 -20 Mean RSS Value Figure: HTC Flyer user with Asus eeePC radiomap International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Experimental Results – Single Route

18 No Calibration Self−Calibration 16 Device−Specific

Introduction 14 - Motivation 12 Device Calibration 10 - Device Diversity - Manual Calibration 8 Self-Calibration Positioning Error [m] - RSS Histograms 6 - Self-Calibration Method 4

Performance 2 Evaluation - Measurement Setup 0 0 10 20 30 40 50 60 70 80 90 100 - Experimental Samples Results Conclusions - Concluding Remarks I iPAQ radiomap with Flyer user-carried device

I For the ﬁrst 10 sec the device is uncalibrated and accuracy is not adequate

I Beyond that point, the device is automatically calibrated and accuracy is greatly improved International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Results with pairwise device combinations

Table: Median of the mean error ¯ [m], with and without calibration. Introduction - Motivation Device Calibration iPAQ eeePC Flyer Desire Nexus S - Device Diversity - Manual Calibration iPAQ 2.7 2.8 (6.6) 3.0 (7.5) 2.9 (8.4) 2.6 (7.7) Self-Calibration eeePC 2.8 (4.4) 2.3 2.3 (2.8) 2.6 (3.5) 2.5 (2.9) - RSS Histograms - Self-Calibration Flyer 3.2 (5.9) 2.6 (3.0) 1.9 2.1 (2.3) 2.6 (2.7) Method Desire 3.4 (6.1) 2.8 (3.2) 2.5 (2.5) 2.4 2.5 (2.6) Performance Evaluation Nexus S 3.0 (6.2) 2.6 (2.8) 2.7 (2.7) 2.4 (2.5) 2.3 - Measurement Setup - Experimental Results Conclusions I All 5 devices used as a reference (row) and test device (column) - Concluding Remarks I Mean positioning error using device self-calibration (results without calibration shown in parentheses) I The diagonal cells report the accuracy when the reference and test devices are the same (i.e. device-speciﬁc radiomap is used) I Self-calibration improves the accuracy for all device pairs

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Concluding Remarks

Device diversity is one of the reasons that hinders the Introduction proliferation of RSS-based positioning systems. - Motivation Device Calibration Our Contributions - Device Diversity - Manual Calibration Self-Calibration I Low-complexity, yet eﬀective method that allows any mobile - RSS Histograms device to be self-calibrated - Self-Calibration Method I Automatic calibration is attained shortly after the user has Performance Evaluation started positioning - Measurement Setup - Experimental Results Future Work Conclusions - Concluding Remarks I Application in larger scale setups featuring non uniform WiFi AP layouts (possible skewness of the RSS histograms) I Integrate with our Airplace indoor positioning platform developed for Android smartphones

http://www2.ucy.ac.cy/~laoudias/pages/platform.html International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Introduction - Motivation Device Calibration Thank you for your attention - Device Diversity - Manual Calibration Self-Calibration - RSS Histograms - Self-Calibration Contact Method Performance Christos Laoudias Evaluation KIOS Research Center for Intelligent Systems and Networks - Measurement Setup - Experimental Results Department of Electrical & Computer Engineering Conclusions University of Cyprus - Concluding Remarks Email: [email protected]

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Introduction - Motivation Device Calibration - Device Diversity - Manual Calibration Self-Calibration - RSS Histograms - Self-Calibration Method Extra Slides Performance Evaluation - Measurement Setup - Experimental Results Conclusions - Concluding Remarks

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 RSS Diﬀerence Approach

Assume that a mobile device resides at a location `, which is Introduction covered by 2 WiFi APs, namely AP1 and AP2. The RSS values - Motivation recorded by the device are given by Device Calibration - Device Diversity - Manual Calibration RSS1 = A − 10γ log10 d1 + X1 Self-Calibration - RSS Histograms RSS2 = A − 10γ log10 d2 + X2 - Self-Calibration Method Performance where di , i = 1, 2 is the distance from the i-th AP, while Evaluation 2 X1, X2 ∼ N (0, σn) are independent Gaussian noise components - Measurement Setup - Experimental disturbing the RSS values. Results Taking the diﬀerence of these RSS values, denoted as RSSD12, Conclusions - Concluding gives Remarks d2 0 RSSD12 = RSS1 − RSS2 = 10γ log10 + X d1 0 2 where X ∼ N (0, 2σn) is the linear combination of X1, X2.

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012 Inverse ecdf Least Squares Fitting

If u is a continuous random variable and y = f (u) with −1 monotonically increasing f then f = Fy ◦ Fu. In particular, the Introduction inverse cdf ordered pairs - Motivation Device Calibration −1 −1 {(ui , yi ) = (F (qi ), F (qi )) : qi ∈ {0.1,..., 0.9}} - Device Diversity u y - Manual Calibration Self-Calibration lie on the curve y = f (u). - RSS Histograms - Self-Calibration Method Proof: Performance We have Evaluation - Measurement Setup - Experimental Fu(u) = P(u ≤ u) = P(f (u) ≤ f (u)) = Results Conclusions = P(y ≤ f (u)) = Fy(f (u)). - Concluding Remarks −1 −1 Applying Fy to both sides gives the identity f = Fy ◦ Fu. Also, the components of the inverse cdf ordered pairs satisfy

−1 −1 yi = Fy (qi ) = Fy (Fu(ui )) = f (ui ).

International Conference on Indoor Positioning and Indoor Navigation, Sydney, Australia 15 November 2012