The comparison of methods for constructing the radio frequency layer of radio environment map using participatory measurements

Transcription

1 4th Workshop of COST Action IC0902 Rome, Italy, October 9-11, 2013 The comparison of methods for constructing the radio frequency layer of radio environment map using participatory measurements Marko Pesko 1, Tomaž Javornik 2, Mitja Štular 1, Mihael Mohorčič 2 1 Telekom Slovenije, d.d., Ljubljana, Slovenia 2 Jozef Stefan Institute, Ljubljana, Slovenia

2 Outline Motivations Related work Improving RF-REM construction Self-tuning construction method RF-REM area of interest Participatory measurement sets The performance metric Results Conclusions 2

3 Motivations 3 RF-REM construction is so far mostly performed by direct interpolation-based methods Spectrum measurements are nowadays obtained from dedicated sensing deployments or measurement campaigns The future radio environment is expected to be more dynamic and participatory sensing could replace dedicated sensing The performance of current and to participatory measurements adapted RF-REM construction methods must be evaluated

4 Related work Recently proposed indirect method LIvE shows great potential of indirect methods: H. B. Yilmaz and T. Tugcu, Location Estimation-Based Radio Environment Map Construction in Fading Channels, Wireless communications and mobile computing, Test scenario: 4 Grid-based sampling, omnidirectional transmitter, assumed known information of channel parameters, log-normal shadowing and Rayleigh fading The simulation results suggest that the location estimation based REM construction outperforms the compared methods Consideration of operating environment and Tx characteristics could further improve such indirect RF-REM construction

5 Improving RF REM construction Making a step further, we developed a new indirect RF REM construction method to consider: the operating environment, transmitter parameters, including the antenna pattern, calibration of the selected propagation model 5 An indirect self-tuning method construction: (STM) for RF-REM M. Pesko, L. Benedicic, T. Javornik, A. Kosir, M. Stular, M. Mohorcic An indirect self-tuning method for constructing the radio frequency layer of radio environment map, submitted to IET Electronics Letters in August 2013, under review and subject to Institution of Engineering and Technology Copyright.

6 Self-tuning construction method Estimted signal strength on i-th RF-REM location: where: RF RE M i = P Tx L RE M i = P Tx ( L i + L dif f i + L clu t i G an t i ) G an t i = G m FBR+FBR cos n ( Θ 0 Θ i /2 ) 6 L i = A 0 + A 1 log 10 (d)+ A 2 log 10 ( H eff ) + A 3 log 10 (d) log 10 ( H eff ) 3.2 [ log 10 (11.75 H m ) ] log 10 (f) 4.78 [ log 10 (f) ] 2 F( β,j)= 1/N i=1 N ( P i ( P Tx L REM i ( β ( T j )))) 2 β =[ A 0, A 1, A 2, A 3, G m,fbr, Θ 0, n] ( β, j )=argmin F( β,j)

7 RF-REM area of interest The size of area: 5.15 km by 6.75 km with resolution of 25 m 9 different clutter categories Dense urban area Urban area Urban area without buildings, mostly roads Suburban area Dry open land without special vegetation Agricultural area Forestall area Swamp area Water area 7

8 Participatory measurement sets 8 + Rayleigh fading GRASS-RaPlaT + information from mobile operator s disposal real BS transmitter information (location, antenna pattern, Tx power) digital elevation model (DEM) clutter map Matlab: Sampling of such RF REM and formation of measurement sets of different sizes Multiple measurement sets of the same size (location dependency)

9 Methods performance metric RF REM performance metrics from literature: mean squared error (MSE) mean absolute error (MAE) relative Mean Absolute Error (RMAE) root mean square error (RMSE) FAZR and CDZR (introduced by Yilmaz) etc. The selected performance metric is: RMSE 9 i.e. average of RMSE values calculated between TRUE and reconstructed RF REMs for different measurement sets of the same size

10 Results 10 The considered RF REM construction methods: IDW IDW2 Kriging LIvE STM The same methods as in LIvE test scenario Preliminary results of the considered construction methods comparison R M S E Measurements set size IDW LIvE IDW2 Kriging STM

11 Preliminary results MCDs distribution: random IDW IDW2 Kriging LIvE STM:NoAntenna STM:Antenna MCDs distribution: all in main lobe of Tx antenna IDW IDW2 Kriging LIvE STM:NoAntenna STM:Antenna MCDs distribution: 95% in main lobe of Tx antenna IDW IDW2 Kriging LIvE STM:NoAntenna STM:Antenna R M S E R M S E R M S E Number of measurements N Number of measurements N Number of measurements N MCDs distribution: 95% outside main lobe of Tx antenna IDW IDW2 Kriging LIvE STM:NoAntenna STM:Antenna MCDs distribution: clusters (clutter) IDW IDW2 Kriging LIvE STM:NoAntenna STM:Antenna MCDs distribution: real traffic (distances from Tx) IDW IDW2 Kriging LIvE STM:NoAntenna STM:Antenna R M S E R M S E R M S E Number of measurements N Number of measurements N Number of measurements N

12 Conclusions 12 RF REM performance evaluation results were presented for several existing methods in parallel with the performance of the STM method All methods were tested on the same sets of spatially distributed participatory measurements and compared in terms of RMSE The results confirm that the accuracy of the RF REM can be notably enhanced by construction methods considering the operating environment, propagation model tuning and transmitter characteristics

13 Acknowledgements 13 The research leading to these results has been partially funded by the European Union, European Social Fund and the FP7 projects ABSOLUTE (FP7 ICT ) and CREW (FP7-ICT ).

14 4th Workshop of COST Action IC0902 Rome, Italy, October 9-11, 2013 Thank you! Marko Pesko

15 4th Workshop of COST Action IC0902 Rome, Italy, October 9-11, 2013

16 GRASS-RaPlaT Developed by JSI ( Open-source radio coverage simulation tool based on GRASS with user extendible set of radio propagation models For research and professional communication network planning modules for a number of channel models module for sectorisation according to given antenna radiation patterns module for calculating and storing the complete radio network coverage data supporting modules (e.g. for adapting input data and analyzing simulation results) 16 Python The accuracy has been validated on existing real GSM network data Cell list Antenna diagram r.fspl r.hata r.cost231 r.hatadem r.waik r.sector Loop db.generatetable r.maxpower DEM Attenuation Cell attenuation Power table (dbf) Maximal signal Land usage [db] r.clutconvert Land usage table - db db.compareresults r.comparemobitel r.compare Land usage Results comparison Measurements Simulations TEMs Results comparison Comparison