Zsolt T. Kardkovács Budapest University of Technology and Economics Dept. Telecommunications and Mediainformatics MŰEGYETEM 1782 High Precision Indoor.

Zsolt T. KardkovácsBudapest University of Technology and EconomicsDept. Telecommunications and Mediainformatics

MŰEGYETEM 1782

High Precision Indoor Localisation by Non-Linear Noise Reduction

Problem statement

Is it an important task?

No, ○ people usually know where to go○ RFID solutions are better suite for this problem

Yes, ○ location based services are needed in-door○ i-home solutions require in-door positioning○ if one of the following applies

in-door traffic logging is important in-door tracking is important interactivity (or communication) is a requirement fast & easy set-up is necessaryyou need cheap & general solution in a single device

04/10/23 FuturICT 2009, High-precision in-door localization 2

Problem statement

This is a solved problem, isn’t it?

Yes, there are great many in-door positioning applications, e.g.○ map based solutions○ if background knowledge is available○ great many training data is available○ needs access point (AP) programming

No,○ they are either need lot of background work○ or they have a low precision rate (RME > 2m)

04/10/23 FuturICT 2009, High-precision in-door localization 3

Focus on the resources

Ideal in-door positioning

Needs○ No map information○ No information on AP locations○ No AP programming○ Low amount of training data○ Low computational resources

It can be placed either on client or AP side Fast Adaptive to small changes Easy re-calibration if necessary

04/10/23 FuturICT 2009, High-precision in-door localization 4

Foundations

Received Signal Strength Indication (RSSI)

Close (visible) space Relaxation: quadratic to distance Simple to calculate Precise, although there is a sphere with iso-RSSI points …if possible priories close space information

Far (non-visible) space Signals suffer from diffraction, reflection, and multipath propagation effect Vague, there are a limited number of iso-RSSI point …approximation is required …the only fix point is the shortest route from AP to target

04/10/23 FuturICT 2009, High-precision in-door localization 5

RSSI

A kind of distance function log-linear to RF path length:

where ○ d[m] and d[dBm] – distances in meters and dBm respectively○ Φ is a frequency dependent parameter (for Ch. 7 it is 0.009)○ β is a relaxation coefficient (in ideal environment it is 20)○ α is a signal to noise ratio coefficient (AP dependent)

Channel is noisy - it varies○ in distance○ in time○ in motion

04/10/23 FuturICT 2009, High-precision in-door localization 6

1/][ *100010*][ dBmdmd

Our solution

Observation only the first RSSI is important others can be used for noise reduction(!)

Visible space is preferred within 1.5 d[m] range use AP coordinates farther use other APs to determine the exact location

Far space A range of RSSI is valid for any location in d[m] space if for any training data point calculated distance is 0 if

and d[m] otherwise K-nearest neighbor algorithm

04/10/23 FuturICT 2009, High-precision in-door localization 7

][2|)()}(max),({min| mdtRSSItVtV

Evaluation

If at least 2 APs are available for each location RMSE is 3,8m (RME < 2m) 63% precision, i.e. within 1.5m (31% within 0.5m) The target location is within the top 3 places (85%) IEEE ICDM challenge 5th prize award (the best European competitor)

Other features Performance not depends on the size of training data Needs no additional configuration time Needs no information on APs location or Ch. Information Needs no map information Easy deployment

Extensions GSM based out-door positioning (<15m in 83%, city) If tracking information is available precision is higher (RMSE 2,8m)

04/10/23 FuturICT 2009, High-precision in-door localization 8

Zsolt T. KardkovácsBudapest University of Technology and EconomicsDept. Telecommunications and Mediainformatics

MŰEGYETEM 1782

High Precision Indoor Localisation by Non-Linear Noise Reduction

Thank you for your attention and time