Vous êtes sur la page 1sur 2

Poster Abstract: Bayesian Localization in Wireless Networks Using Angle of Arrival

Eiman Elnahrawy, John-Austen Francisco, Richard P. Martin


{eiman,deymious,rmartin}@cs.rutgers.edu

Department of Computer Science, Rutgers University 110 Frelinghuysen Rd, Piscataway, NJ 08854 ABSTRACT
Using existing wireless communication networks as a localization infrastructure promises enormous cost and deployment savings over specic localization infrastructures. In this work we investigate a Bayesian network approach that uses a combination of radio signal strength (RSS) to distance estimation along with angle-of-arrival (AoA) information. We characterize the resulting localization accuracy using data collected outdoors using different radios, indoor data, and simulated data. We show how the localization performance degrades in indoor environments and analyze the different sources of errors that cause this performance degradation as compared to outdoor settings. We found our network is quite sensitive to variations in the distance to signal strength, and the additional angle information had only a small impact on localization accuracy. antenna. This function of RSS to angle, or the, AoA curve, provides the base radio property of our approach. We use both our own data, as well as that in [3]. We prove that our Bayesian technique is feasible by rst showing that a simple propagation model ts the observed AoA as a function of the angle (the AoA curve), in outdoor settings. Figure 1 shows a sample raw and smoothed sample AoA curve for the Telos Mote in an outdoor setting. We also found our Bayesian approach has equivalent accuracy to [3], but uses less base-stations (4 vs. 7). We also demonstrate that our approach is applicable to a variety of technologies by showing that the propagation models work well for both 802.11 as well as 802.15.4 networks. Using our derived propagation models, we show that our proposed Bayesian network can give highly accurate results and it is robust to signicant errors in the angle estimation. With minimal errors in the AoA curve, the network can provide a mean error of less than 1 ft and a maximum error of 3 ft. Even with random errors of up to 45 degrees, the maximum error for a building oor is still only 7ft.

Categories and Subject Descriptors


C.2.5 [Local and Wide-Area Networks]

General Terms
Telos Mote AoA Curve at 60 ft

Algorithms, Measurement, Performance, Design, Experimentation


-65

Original Data

Smoothed Data

Cosine Fit

Signal Strength (dBm)

Keywords
Localization, Angle of Arrival, Wireless Local Area Networks, Bayesian Statistics

-70 -75 -80 -85 -90 -95 -100 -105 180

1.

INTRODUCTION

Recent years have seen intense research investigating using wireless networks as a localization infrastructure. If successful, using the same infrastructure for both communication and positioning would provide a tremendous cost and deployment savings over a specic localization infrastructure, such as ceiling-based ultrasound. In this research we investigate using a machine learning technique that uses received signal strength (RSS) in combination with both Angle of Arrival (AoA) and distance estimation to provide localization in wireless networks. More specically, we use a Bayesian network that incorporates both angle and signal to distance in a single network in order to estimate the propagation parameters as well as estimate a radios position. We build on previous work using Bayesian networks using only RSS for distance estimation [2, 1]. We rst constructed a simple base station incorporating a rotating directional parabolic antenna. The base station can then construct a curve of the measured RSS as a function of the angle of the
This research was supported by NSF grant #CNS-0448062 and NSF grant #CCR 03-14161.

230

280

330

20

70

120

170

Antenna Angle (deg)

Figure 1: A sample AoA curve, this is for the Telos Mote (using 802.15.4) at 60ft taken outdoors.

2. THE BAYESIAN NETWORK


Figure 2 shows our graphical model for the location estimation. This model incorporates both the knowledge of AoA and the knowledge of RSS from the n different base stations in the localization system in order to localize an object. The vertices X and Y represent location. The vertex D1 (respectively D2 , . . . , and Dn ) represents the Euclidean distance between the location specied by X and Y and the rst (respectively second, . . . , and nth) base station. We assume the locations of the access points are known and hence the Di s are deterministic functions of X and Y . The vertex a1 (respectively a2 , . . . , and an ) represents the angle-of-arrival (AoA) of the signal received from the rst (respectively second, . . . , and nth) base station.

Copyright is held by the author/owner. SenSys05, November 24, 2005, San Diego, California, USA. ACM 1-59593-054-X/05/0011.

Error CDF:AoA and SS vs SS Only 4 Basestations


1 0.9

Error CDF:Effect of Random Angle Ahift, Lobes Noise and Attenuation on AngleModel
1
1

Error CDF:Effect of Different Corrections on Indoor Data


0.9

0.9

0.8

0.8

0.8

0.7

0.7

0.7

Probability

Probability

0.5

0.5 lobes0,shift0,atten0 lobes0,shift0,atten5 lobes0,shift45,atten5 lobes0.8,shift0,atten5 lobes0.8,shift45,atten5 lobes0.8,shift45,atten15

Probability

0.6

0.6

0.6

0.5

0.4

0.4

0.4 A=0, S=0, W=1, L=120 A=0, S=0, W=181, L=0 A=0, S=0, W=241, L=0 A=0, S=100, W=1, L=0 A=100, S=0, W=1, L=0 A=100, S=100, W=61, L=60 A=100, S=100, W=181, L=120 A=100, S=100, W=241, L=60 A=100, S=100, W=61, L=0 A=100, S=100, W=241, L=0 0 5 10 15 20 25 30 35 40 45 50

0.3

0.3

0.3

0.2

0.2

0.2

0.1 AoA and SS M1SS Only 0 0 5 10 15 20 25 30 35 40 45 50

0.1

0.1

10

20

30

40

50

60

70

Error in Feet

Error in Feet

Error in Feet

(a)

(b)

(c)

Figure 3: The gures show CDFs of localization accuracy. Figure (a) compares our Bayesian network to one that does not incorporate angle information, while (b) presents error CDFs of the synthetic data with varying degrees of distortion, and (c) presents accuracy results for real indoor data with varying amounts of correction applied. are signicant deviations from the outdoor models in terms of distance estimation. Finally, we found that the shapes of indoor AoA curves are quite distorted, with many peaks in the data that are close to the true peak. In order to explore the impact of these effects on the AoA curve on our localization approach we both perturbed a set of synthetic data as well as correct a set of real data. We systematically altered the (1) angle of the peak (shift), (2) mean RSS as a function of distance (attenuation), and (3) sizes of the side lobes, as percentage of the main lobe. Our work shows that accuracy degrades as a function of all these error types. Figure 3(b) shows the results of applying increasing distortions to a set of synthetic data. The results show that although our model is robust to a single type of error, the combination of errors in both the angle and the distance to signal measurements make indoor localization very challenging. Figure 3(c) shows the results of xing the 3 error types in measured data in indoor environments. The gure shows signicant improvements in localization accuracy as we adjust the AOA curves towards those predicted by outdoor measurements. Our results show that the resulting AoA curves are quite distorted in terms of distance, shift and lobe errors. Both of these results also show that our network is quite sensitive to RSS to distance distortions, and we found that these are the most common and severe ones observed. Also, comparing the distorted synthetic and correct real data shows that the performance of our network is consistent with the classes and magnitudes of these errors on the AoA curves. A key open question is thus how to resolve the better performance of many other algorithms as compared to our network, despite our addition of angle information.

a1

D1

a2

D2

an

Dn

S1

S2

Sn

b10

b11 b13

t1

b20

b21 b23

t2

b n0

bn1 bn2 b n3

tn

b12

b22

Figure 2: A Bayesian network for location estimation that combines both AoA and signal to distance in one graphical model.

The vertex Si represents the combined function of both the signal strength measured at (X, Y ) with respect to the ith base station, i = 1, . . . , n and the angle information at this base station. The model assumes that X and Y are marginally independent and are independent of the angle measurement. Note that we must quantize the AoA curve for Si in 10o increments and t it to a cosine curve in order to make the model tractable. Figure 1 shows a tted cosine curve.

4. REFERENCES
[1] E IMAN E LNAHRAWY, X IAOYAN L I , AND R ICHARD P. M ARTIN The limits of Localization Using Signal Strength: A Comparative Study. In IEEE SECON (October 2004). [2] DAVID M ADIGAN , E IMAN E LNAHRAWY, R ICHARD P. M ARTIN , W EN -H UA J U , P. K RISHNAN , AND A.S. K RISHNAKUMAR. Bayesian Indoor Positioning Systems. IEEE Infocom (March 2005). [3] D RAGOS N ICULESCU AND BADRI NATH. VOR Base Stations for Indoor 802.11 Positioning. ACM MobiCom (September 2004).

3.

LOCALIZATION PERFORMANCE

Using primarily 802.11, we found that indoor environments have a high degree of error in both the AoA measurement as well as the distance measurement results in performance that is only marginally better than not using angle information at all, as is shown in Figure 3(a). In order to understand the differences between the indoor and outdoor environments, we rst characterize the resulting AoA curves. We found that the curves are quite noisy, with angle estimation errors of up to 60 degrees not uncommon. We also found that there

Vous aimerez peut-être aussi