Presentation is loading. Please wait.

Presentation is loading. Please wait.

Location Based Services by: Imar Sanmartí Germà Jordi Anguela Rosell.

Similar presentations


Presentation on theme: "Location Based Services by: Imar Sanmartí Germà Jordi Anguela Rosell."— Presentation transcript:

1 Location Based Services by: Imar Sanmartí Germà Jordi Anguela Rosell

2 Index ● General introduction ● Objectives ● Workspace ● Methods ● Conclusions ● Future improvements

3 General introduction ● What is a 'Location Based Service'? ● LBS have two major actions: – Obtaining the location of a user – Utilizing this information to provide a service ● Where am I?, Where is the nearest …?, Where is my …?, How do I get there? ● 3 types of LBS: pull, push, and tracking services.

4 Objectives Estimate the position of a cell phone Restrictions: Average error < 100m. Time to process the request < 1sec.

5 Workspace ● Mobile positioning techniques – Rx-Level Based ● Propagation models – Planet DMS Macrocell prediction ● How to manage all this data – Spatial Data Base Introduction ● Features and limitations of a Spatial DB ● Saving and access to the data

6 Mobile positioning techniques Radiolocation techniques Angle-of-Arrival (AoA) Time-Based methods Cell ID methods Rx Level-Based Positioning Satellite-Based techniques Assisted GPS ● Techniques used in Proximity Systems ● Dead Reckoning techniques

7 Propagation models (I) Macrocell prediction models – Okumura-Hata model – Parabolic Equation Method – Empirical two ray model (DTR) – Dominant path prediction model Microcell prediction models – Empirical COST-Walfisch-Ikegami Model – Ray tracing

8 Propagation models (II) Propagation scenario in a typical urban environment

9 Planet DMS Macrocell prediction(I) ● File of path loss values over a certain area for a base station. ● Each value is assigned to a slot of terrain of 25m x 25m. ● 1601 x 1601 slots in a base station map (cell). ● BS centred in the map. 25m -70dB Base Station 1601 x 1601 path loss values

10 Planet DMS Macrocell prediction(II) 120º 200º 456241_5550700_0_25_25m_1800_global_25m_EiRP_fast_739496_120_1 455997_5549964_0_22_25m_1800_global_25m_EiRP_fast_739496_200_-5 (456241,5550700) 456241_5550700_0_25_25m_1800_global_25m_EiRP_fast_739496_120_1 455997_5549964_0_22_25m_1800_global_25m_EiRP_fast_739496_200_-5 (455997,5549964)

11 Introduction Spatial Database ● PostGIS is an extension to the PostgreSQL object-relational database system which allows GIS (Geographic Information Systems) objects to be stored in the database ● Why PostgreSQL+PostGIS? – Because it's Open-source (It’s free) – Doesn’t need an expensive license like Oracle

12 Spatial DB Features ● Features: – Three types of indexes: B-Tree, R-Tree, GiST – Geometry Relationship Functions like: ● Intersects(geometry,geometry) – Geometry Processing Functions ● Centroid(geometry) ● Area(geometry) ● GeomUnion(geometry,geometry)

13 Saving data into spatial DB cell_1001 ● One table per each cell attribute fields: level cell_geometry indexs: level_index - btree geometry_index - gist

14 Reducing amount of data ● Our maps have: 1601x1601slots≈ 2.500.000slots/map ● To reduce this we used a 'threshold' Rx level < -120dB are not stored in DB 2.500.000slots => 2.500slots (1000 times less)

15 Methods ● Minimization of an error function ● Map intersection without division ● Intersection with map division ● Direct intersection using indexes – Basic mode – Cutting the maps under threshold x dB – Powering the received levels with y dB – Hybrid mode

16 Minimization of an error function (I) Uses measured signal strength at the MS. Example: – P rx1 = -70dB – P rx2 = -89dB – P rx3 = -64dB

17 Minimization of an error function (II) (x 1,y 1 ) (x 2,y 2 ) (x 3,y 3 ) (x,y) Requirements: Minimum 3 Rx potencies. Position of the BSs. Selection of the better propagation model.

18 Minimization of an error function (III) Circle of radium d i from with centre in (x n,y n ). If the circles of the BSs don’t intersect, we define The point (x,y) OPT will be:

19 Minimization of an error function Results Results obtained with Extended Okumura-Hata propagation model. The method works better with 3 BSs: – The BSs that is serving – The next two BSs with higher rx-level reception Time of computation fast. – Average time:59 ms High error: – Average error:258 m No need to use the maps.

20 Map intersection without division Select the polygons between the range ±x/2(dB) for each cell. P rx1 =-64dBP rx2 =-73dBP rx3 =-87dB -x/2+P rx1 < Level < x/2+P rx1 -x/2+P rx2 < Level < x/2+P rx2 -x/2+P rx3 < Level < x/2+P rx3

21 POLYGON ((x11,y11),(X12,y12),(x13,y13),(x14,y14),(x11,y11)) POLYGON ((x21,y21),(x22,y22),(x23,y23),(x24,y24),(x21,y21)) POLYGON ((x31,y31),(x32,y32),(x33,y33),(x34,y34),(x31,y31)).................................................................................................. ……………………………………………………………………….. POLYGON ((xn1,yn1),(xn2,yn2),(xn3,yn3),(xn4,yn4),(xn1,yn1)) Map intersection without division Geomunion of the resultant query. MULTIPOLYGON ((x11,y11),(X12,y12),(x13,y13),(x14,y14),(x21,y21),(x22,y22),(x23, y23),(x24,y24),(x21,y21),(x31,y31),(x32,y32),(x33,y33),(x34,y34), (x31,y31).......................................................................................... ……………………………………………………………………………. (xn1,yn1),(xn2,yn2),(xn3,yn3),(xn4,yn4),(xn1,yn1))

22 Map intersection without division Centroid of the resultant intersection Intersection of the resultant GeomunionsCalculate centroid of the resultant intersection

23 Map intersection without division Results The computation time is an exponential function of the number of polygons to join. Error approximation (140m approx.). High computation time: – Need to split the multipolygons.

24 Intersection with map division(I) ● Methodology – Split the maps to work faster with them

25 Intersection with map division(II) ● This division is done recursively if one of the partitions has more than a slot_threshold=700slots 2500 slots SPLIT!!!! > threshold 100 slots400 slots 300 slots 1700 slots 100 slots400 slots 300 slots 600 slots 400 slots 300 slots

26 Intersection with map division(III) ● Methodology – Select the polygons between the range ±x/2(dB) for each map. – Geomunion of the resultant query – Geomunion of the divided maps – Intersection of the resultant Geomunions of the divided maps – Calculate centroid of the resultant intersection

27 Intersection with map division Results ● Error approximation: 147m ● Computation time: 109 s ● Conclusion: – It still spends a lot of time and the results obtained aren’t so good as we expected ● Solution: don’t use the functions GeomUnion() and Intersection() to estimate the position

28 Direct intersection (I) ● How it works: – We use a prepared SQL query to estimate the mobile position directly from the spatial DB. ● What do we expect? – At least, faster computation time.

29 Direct intersection (II) ● How the SQL query chooses the data? – It takes profit of the indexes: ● level_index ● geometry_index: with operator (&&) much much faster Ex: SELECT … FROM… WHERE cell1.geom && cell2.geom …

30 Direct intersection (III) ● How the mobile position is estimated? – The final position is estimated doing the average of the centers of all slots that have been selected. SELECT AVG(Y(Centroid(cell1.geom))) AS North AVG(X(Centroid(cell1.geom))) AS East FROM …

31 Direct intersection (IV) RESULTS ● Features: – We have the same error – Average time is reduced to 2,5 s. – Method very simple ● Next step: – Reduce the time -> use threshold in Rx levels – Improve the error -> power Rx levels

32 Direct intersection (V) ● Tested solutions – Cutting the maps under threshold -85dB, -80dB – Powering the received levels with 3dB, 5dB, 7dB – Hybrid mode:combination of both methods

33

34

35 Direct intersection (VII) Results ● The best method used: – Cutting the maps under -80dB. – Powering 5dB the received signals – Direct intersection using indexes with 30dB of range ● The obtained results: – Average error:105,3 m – Average time:0,92 s

36 Direct intersection (VI) Results ● Cutting the maps under some threshold increases the precision. ● Cutting over -80dB is the limit where some tests start not to intersect. ● Powering the received signals also increases the precision, but increases the computation time. ● Powering with 7dB is the limit where some tests start no to intersect.

37 Direct intersection (VIII) Our proposal ● Requirements: – Map of the service station is needed – Discard maps with Rx received level under -80dB – Minimum 2 maps over threshold -80dB are needed. – Power the resultant Rx-levels with 5dB.

38 Conclusions We have accomplished the objectives: – Error ≈ 100 m – Time < 1 s Large dispersion in the results (high variance) Methods were only tested in Dejvice area: – Need to test in other parts of Prague. An error of 100 m means a distance of 4 resolution units of the DB (res. 25m). The use of indexes is the optimal method.

39 Future improvements (I) ● Use of improved maps. How? – Using other propagation models considering more factors (rooftop, diffraction on corners, street canyon effect...). – Change of macrocell to microcell maps: ● Shorter maps with higher resolution.

40 Future improvements (II) Average of multiple measurements. Why? – Avoid the fast fading – Change of LOS to NLOS – Have more accuracy in the rx-level values

41 Future improvements (III) Use of vector files with information of the city. Use the map as a mask over the pathloss files, to reduce the amount of data to store.

42 The End Thanks for the assistance Any question?


Download ppt "Location Based Services by: Imar Sanmartí Germà Jordi Anguela Rosell."

Similar presentations


Ads by Google