Results of an indoor deployment Yann-A el Le Borgne, Mathieu Van - - PowerPoint PPT Presentation

Range based localization with CC2420

Results of an indoor deployment

Yann-A¨ el Le Borgne, Mathieu Van Der Haegen, Gianluca Bontempi

Machine Learning Group, Universit´ e Libre de Bruxelles Brussels - Belgium

June 4, 2007

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 1 / 13

Machine Learning Group

Wireless sensor networks

Main research axes : Use of prediction/compression techniques to reduce the amount of data generated by the network More particularly :

1

Time series prediction at the sensor level

2

Spatial/temporal in-network compression with PCA (Principal component analysis)

3

In-network feature extraction with PCA for network-wide prediction task

4

Localization : Multidimensional Scaling (close to PCA)

Experimental platforms :

◮ 22 TMote Sky (MoteIV) ◮ 8 TMote Invent

8 MHz microcontroller 48KB + 10KB RAM + 1MB memory CC2420 256kbps radio Strong energy constraints

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 2 / 13

Agenda

1

Range based localization

2

CC2420 distance indicators

3

Comparison RSSI/LQI

4

Impact of the power level

5

Conclusion

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 3 / 13

Range based localization

Let consider a set of N sensors in a 2D space with :

◮ pi ∈ R2, 1 ≤ i ≤ N, be the coordinates of sensor i ◮ dij = ||pi − pj|| be the euclidean distance between sensors i and j ◮ There exists geometrical relations betwen pi and dij NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 4 / 13

Range based localization

Range based localization is a two stage process :

1 Internode distance estimation : ˆ

dij = f (Iij)

◮ Iij is an indicator of distance ◮ f is a the function transforming the indicator into a distance ◮ The set of all distance estimates is stored in a ˆ

Dij = [ˆ dij] matrix

2 Position derivation : {ˆ

pi} = argmin{φi}L(ˆ Dij, {φi})

◮ For example L(ˆ

Dij, {φi}) =

i,j(ˆ

dij − ||φi − φj||)2)

Accuracy of the localization is given by 1

N

N

i=1(ˆ

pi − pi)2

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 5 / 13

CC2420

CC2420 : Chipcon radio, 802.15.4 IEEE compliant, featured by TMote (MoteIV) MICAz (Crossbow) TelosB (Crossbow) IMote 2 (Intel) Operates in 2.4GHz GSM band, same as Wifi 16 channels, data rate of 256 kbps Two indicators of signal quality :

◮ RSSI : Received Signal Strength ◮ LQI : Link quality indicator NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 6 / 13

Radio link quality indicators

RSSI : Received Signal Strength Indicator

◮ Benefits from a theoretical basis : P(d) = βd−α ◮ However, it is depreciated in practice because of a high variability ◮ In CC2420 : computed over eight 4 bits symbol periods :

P=RSSI VAL-45 (in dBm)

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 7 / 13

Radio link quality indicators

RSSI : Received Signal Strength Indicator

◮ Benefits from a theoretical basis : P(d) = βd−α ◮ However, it is depreciated in practice because of a high variability ◮ In CC2420 : computed over eight 4 bits symbol periods :

P=RSSI VAL-45 (in dBm)

LQI : Link Quality Indicator

◮ Chip error rate, i.e. number of bits flipped, computed over two 4 bits

symbol periods

◮ Featured by CC2420 radios (TMote, TelosB, Micaz), and assumed to

be more reliable to estimate link quality

◮ Ranging from 50 to 110 (Minimum and maximum quality frames,

respectively)

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 7 / 13

Radio link quality indicators

RSSI : Received Signal Strength Indicator

◮ Benefits from a theoretical basis : P(d) = βd−α ◮ However, it is depreciated in practice because of a high variability ◮ In CC2420 : computed over eight 4 bits symbol periods :

P=RSSI VAL-45 (in dBm)

LQI : Link Quality Indicator

◮ Chip error rate, i.e. number of bits flipped, computed over two 4 bits

symbol periods

◮ Featured by CC2420 radios (TMote, TelosB, Micaz), and assumed to

be more reliable to estimate link quality

◮ Ranging from 50 to 110 (Minimum and maximum quality frames,

respectively)

Srinastava and Levis (IPSN-06) stressed :

◮ The lack of public experiments comparing RSSI and LQI with CC2420 ◮ Showed that RSSI was better for determining link quality NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 7 / 13

Experimental results 1

RSSI, LQI and distance

N = 20 TMotes arranged in a 5*4 grid Inter node distance was about 1.5 meter RSSI and LQI were sampled every minute

Distance vs LQI

Raw LQI measurement Distance Between nodes (m)

Distance vs RSSI

Distance Between nodes (m) RSSI measurement (dBm)

+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + −11 −10 −9 −8 −7 −6 −5 2 4 6 8 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + ++ ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + + + + + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + 60 70 80 90 100 110 2 4 6 8

RSSI not correlated with distance

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 8 / 13

LQI - Detailed results

• ●
• 70

80 90 100 110 1 2 3 4 5 6 7

Distance vs LQI (RMSE=1.21)

LQI Distance (m)

Distribution of error residuals

Error Frequency −4 −2 2 4 10 20 30 40 50

Good correlation Linear model seems adequate Gaussian-like error distribution

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 9 / 13

Localization accuracy at maximum power level

Multidimensional scaling (MDS) was used (Whitehouse et al., IPSN06 survey) It solves {ˆ pi} = argmin{φi}

• i,j(ˆ

dij, ||φi − φj||)2) Error is estimated with RMSE =

• 1

N

N

i=1(ˆ

pi − pi)2 after procrustes analysis (translation, rotation, scaling)

−3 −2 −1 1 2 3 −3 −2 −1 1 2 3

MDS Localization with LQI

Dimension 1 Dimension 2

• End of arrows are true positions pi

Start of arrows are estimated positions ˆ pi RMSE = 1.49

Distribution of errors Position error (m) Frequency 0.0 0.5 1.0 1.5 2.0 2.5 2 4 6 8

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 10 / 13

Experimental results 2

Varying power level

Power is increased from -25dB to 0dB

−25dBm −15dBm −10dBm −7dBm −5dBm −3dBm −1dBm −0dBm 0.5 1.0 1.5 2.0 2.5 3.0

Position error for different transmission power

Position error (m) −3 −2 −1 1 2 3 −3 −2 −1 1 2 3

MDS localization at optimal power transmission− RMSE=1.21

Dimension 1 Dimension 2

• At low transmission power, long distances are poorly estimated

At high power transmission, short distances are poorly estimated

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 11 / 13

Conclusion

Summary of the results

◮ LQI was well correlated with distance ◮ RSSI turned out to be uncorrelated with distance ◮ Radio transmission power may improve the localization

RSSI is however good at characterizing connectivity Future work :

◮ Experimental work is not complete : needs comparison with outdoor

and other indoor environments

◮ Determine how the distance model can be identified before the

deployment

◮ Combine the maps obtained for different transmission powers NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 12 / 13

Thank you for your attention Questions ?

Site web : http ://www.ulb.ac.be/di/mlg http ://www.ulb.ac.be/di/labo

NOTERE - WSN workshop (Marrakech) Localization with CC2420 radio June 4, 2007 13 / 13