Vous êtes sur la page 1sur 5

Novel mobile-based location techniques for UMTS

Jakub Borkowski and Jukka Lempiinen Institute of Communications Engineering, Tampere University of Technology P.O. Box 553, FI-33101 TAMPERE FINLAND
Email: {jakub.borkowski, jukka.lempiainen}@tut.fi Website: http://www.cs.tut.fi/tlt/RNG

Abstract The aim of this paper is to propose two new hybrid cellular techniques for mobile positioning in UMTS. Developed methods are entirely mobile-based since no modifications are required on the network side. Proposed techniques exploit signal strength and delay spread measurements from at least two base stations. The estimation accuracy of the presented location techniques has been assessed by measurement campaigns performed in various network configurations and in diverse propagation environments. Measurement results have indicated that the accuracy of positioning based on signal strength measurements can be improved in majority of propagation environments if one of the proposed methods is used. Namely, in a measured area in an urban micro cellular environment the mean error in the range estimation has been reduced from 58 m to 30 m when introduced method has been used. Slightly smaller improvement (30%) has been observed in macro cellular urban and no improvement - in macro cellular suburban/rural environment. In addition, obtained measurement results have illustrated that the range estimation error resulting from signal strength - based positioning is correlated with the distance from the measured base station. Observations of the range estimation error have been concluded in a distance-dependent model.

Key words: mobile-based, positioning, UMTS 1. Introduction

Numerous developed location techniques provide sufficient accuracy, however very often required hardware and software modifications prevent from fast deployment in the existing networks. Ideally, deployment of the location techniques should not involve any changes in the network infrastructure and in the UE (user equipment). However, practically location methods always interfere with the network implementation. Depending on the strategy, required modifications can be concentrated either on the network or the mobile side. Typically, network-based positioning methods are preferred by network operators in order to enable location-sensitive services without requesting users to change their terminals. Examples of these methods include Cell ID+RTT (cell ID + round trip time) [1] or PCM (pilot correlation method) [2]. In turn, modifications implemented in the UE should enable positioning in a wide range of networks. Hence, mobilebased approaches seem to be preferred by mobile phone

manufacturers. In majority, these techniques estimate the position of the UE based on air interface measurements, for instance, time-biased OTDOA (observed time difference of arrival) [1][3] or signal strength-based methods [4]. The major sources of error in the position estimation constitute multipath and NLOS (non line-of-sight) propagation. A failure to detect the direct signal component degrades the accuracy of time-biased positioning algorithms as well as techniques relying on signal strength measurements. Behavior of the resulting estimation error has been widely studied in order to enable NLOS mitigation methods improving overall positioning accuracy [5][6][7]. In addition, awareness of expected estimation errors facilitates implementation of simulators for reliable performance evaluation of positioning. For instance, in [8] NLOS error in TOA (Time of Arrival)-measurements is approximated by mean excess delay of the channel, which in turn is assumed to be proportional to the observable root mean squared delay spread. Moreover, in [9] time-varying NLOS error is modeled as a distance-dependent random variable proportional to the median excess delay of the channel. In this paper, two entirely mobile-based positioning techniques will be proposed and assessed with comparison to the basic signal strength based estimation. In addition, the analysis of an error in range estimation based on signal strength measurements will be illustrated and concluded in a distance-dependent model. 2. Mobile-based hybrid positioning

In the first proposed hybrid algorithm (Hybrid 1), the distances between the UE and the hearable NodeBs are estimated independently based on the RSCP (received signal code power) and DS (delay spread) measurements conducted on CPICH (common pilot channel). The technique requires hearability of minimum two pilots from different sites. However, high overlapping of cells in typical UMTS (Universal Mobile Telecommunication System) configurations ensures good pilot availability, thus this requirement does not constitute the performance bottleneck. Pairs of distances to each hearable NodeB are estimated from the measurements performed by the UE. Namely,

utilization of a simple path loss prediction model (COST231-Hata [10]) allows for a rough derivation of a propagation distance from the RSCP measurements, as the UE is aware of the CPICH transmit power. Similarly, application of an empirical DS distance-dependent model provides an estimate of a distance (di,DS) to the i-th NodeB based on the measured DS [11], (1):
di , DS RMSi . = T1
2

Table 2: Defined RSCP correction in a function of a measured RMS DS. Measured RMS DS [s] 0.03 < 0.3 < 0.45 < 0.5 < 0.7 0.7 Introduced RSCP correction [dB] -15 -10 -8 0 +5 +10

(1)

In (1), RMSi stands for the RMS (root mean square) of the DS of the signal transmitted by the i-th NodeB and T1 is the mean value of the RMS at 1 km distance from the transmitter. For each measured NodeB, ranges estimated from the RSCP and DS measurements are weighted and averaged. The aim of such combining is to achieve better accuracy of the range estimation than in case of a distance estimated from the individual measurements. The definition of weights for averaging is based on the observation that due to local scattering a median of the received power has a tendency to decrease more sharply than the RMS of the DS at small UE-NodeB distances. Consequently, at larger distances from the NodeB, the propagating rays are exposed for more significant influence of various obstacles causing variations of the DS, while RSCP measurements are considered to be more reliable. According to these assumptions, values of the weights have been empirically defined, Table 1. In table, Wi,RSCP and Wi,DS represent the weights of the ranges di,RSCP and di,DS estimated from the RSCP and DS measurements, correspondingly. Finally, the position of the UE is calculated by an unconstrained LMS (least mean square) optimization algorithm based on the averaged estimated UE-NodeB distances assuming that coordinates of the NodeBs are forwarded to the UE. The second proposed mobile-based positioning method (Hybrid 2) exploits correlation properties between the RSCP and the RMS DS. In the LOS situation, COST-231Hata model does not provide reliable distance estimates from the RSCP measurements. Similarly, in a heavy shadowed area, the estimated range based on the utilized pathloss prediction model is also erroneous. In the Hybrid 2, DS measurements are utilized to provide information about the multipath propagation in the measured radio link. Hence, proper exploitation of the DS measurements can
Table 1: Assignment of weights with respect to the relationship between estimated ranges. Condition between estimated ranges di,RSCP > di,DS di,RSCP < di,DS di,RSCP di,DS; di,RSCP > 60% of the cell range di,RSCP di,DS; di,RSCP < 60% of the cell range Wi,RSCP 0 2 0.4 1.6 Wi,DS 2 0 1.6 0.4

improve the accuracy of the RSCP-based positioning. In the first step of the Hybrid 2 algorithm, the distances between the UE and at least two hearable NodeBs are estimated from the RSCP measurements based on the path loss prediction model. Consecutively, the DS is measured on the CPICH transmitted by the same NodeBs. According to the observed RMS DS, the measured RSCP values are modified in order to improve the accuracy of the range estimation. Namely, the measured RSCP is lowered in locations that are estimated to have LOS connection with the respective NodeB, i.e., when the observed RMS DS is very small. Correspondingly, in locations considered as NLOS, the measured RSCP is enlarged. The degree of introduced corrections depends on the measured RMS DS, as illustrated in Table 2. Presented values of corrections have been defined empirically from measurements conducted in an urban UMTS network. Corrected RSCP values are mapped to the distance based on COST-231-Hata model. At the last stage of the positioning, the obtained ranges are processed by the unconstrained LMS optimization algorithm implemented in the UE that provides the final position. Naturally, coordinates of the adequate NodeBs need to be transported to the UE. 3. Measurement environment

The performance of the proposed methods was assessed by field measurements performed in a commercial UMTS network. In order to evaluate practical applicability of the hybrid positioning techniques, the measurements were carried out in diverse propagation environments: in urban micro cellular, urban macro cellular network, as well as in suburban/rural macro cellular environment. Measurement area in a micro cellular topology mainly consisted of 3sectored sites deployed with a density that on average 400 m site spacing was maintained. In turn, analyzed macro cellular urban network was formed by sites with similar sector configuration deployed with 600 m mean site spacing distances. The average base station antenna height in this scenario exceeded the average rooftop level (15 m) by 5-10 m on average. In turn, sites of the suburban/rural macro cellular network were deployed in a less dense manner, with on average 1.2 km site spacing distances and 30 m mean base station antenna height.

(a)

(b)

(c)

(d)

(e)

Figure 1: Comparison of cdf (cumulative distribution function) of error in range estimation performed by proposed hybrid mobile-based algorithms and pure RSCP-based method in different propagation environments; (a) small cell area in micro cellular urban, (b) large cell area in micro cellular urban, (c),(d) macro cellular urban, (e) macro cellular suburban/rural.

Measurement equipment consisted of a laptop PC with radio interface measurement software connected to the test UE, WCDMA (wideband code division multiple access) scanner, and the GPS (Global Positioning System) receiver. Evaluation of the range estimation accuracy was performed by collecting signal strength and delay profile samples from the serving cell over multiple routes. Measurement routes were defined in such a manner that distribution of samples in respect to the distance to the measured NodeB was reasonably balanced. Over every route, at least 200 samples from each measured cell were collected. In every case, the reported accuracy constitutes a deviation between the estimated range from the measured samples and the indication of the GPS receiver. 4. Measurement results

4.1. Hybrid positioning accuracy Executed measurements illustrate that DS is not distance dependent, as practically range estimation performed by the Hybrid 1 algorithm does not improve the accuracy in the considered environments. Exceptionally, when distances from the serving base station are small, e.g., in the analyzed micro cellular environment with mean UE-NodeB distance 144 m, the Hybrid 1 minimizes the range error by 50 %,

down to 30 m, Fig. 1a. At the same time, in the analyzed micro cellular scenario, the second proposed positioning approach, Hybrid 2, improves the accuracy by 20 m in comparison to the basic RSCP-based estimation, which returns 58 m mean range error. However, in the scenario with larger distances to the serving micro cell (mean UENodeB distance 320 m), basically the proposed methods do not improve the range accuracy of pure RSCP-based estimation, Fig. 1b. Results obtained from performance assessment conducted in the urban macro cellular environment indicate that DS is correlated with RSCP to some extent, Figs 1c and 1d. In locations that are characterized by significant drops in a measured RSCP, the DS is correspondingly higher, indicating impact of multipath. Hence, the Hybrid 2 improves the accuracy of the UE-NodeB distance estimation in comparison to the pure RSCP-based positioning method, Figs. 1c and 1d. Precisely, in the macro cellular urban environment with 290 m mean UE-NodeB distance, the mean error in range estimation is reduced from 108 m in the RSCP-based estimation to 94 m when the Hybrid 2 algorithm is used, Fig. 1c. However, at the same time, the analysis reveals that DS is not essentially distancedependent, thus the Hybrid 1 method does not improve the accuracy. Similar results are obtained from measurements performed in the same propagation environment but with

(a)

(b)

(c)

(d)

Figure 2: Range estimation error in a function of distance for RSCP-based estimation; (a) modeled range estimation error for urban and suburban / rural environments; (b)-(d) comparison of the modeled and measured range estimation error illustrated for different propagation environments; (b) macro cellular urban, (c) macro cellular urban, (d) macro cellular suburban/rural.

slightly larger average UE-NodeB distance (330 m). As illustrated in Fig. 1d, the range estimation accuracy is improved from 124 m (RSCP-based) to 107 m (Hybrid 2). Expectedly, when distances to the serving NodeB are smaller in the macro cellular topology configurations, the range estimation accuracy is at the higher level. However, in these scenarios, the proposed algorithms do not significantly enhance the accuracy of the standard RSCPbased estimation. For instance, in the macro cellular urban environment with 150 m mean UE-NodeB distance, the mean range error is at the level of 73 m and 57 m for Hybrid 1 and Hybrid 2, correspondingly. In the same scenario, RSCP-based estimation returns 61 m mean error (not illustrated in figures). Fig. 1e illustrates the positioning performance based on samples collected in the macro cellular suburban/rural propagation environment with average UE-NodeB distance 670 m. As expected, with larger distances from the serving NodeB, the range estimation error is significantly larger. Moreover, proposed positioning methods do not improve the accuracy.

4.2. Distance-dependent model of the range error Estimated range based on RSCP measurements contains an error due to fading propagation environment and non ideality of the utilized path loss to distance mapping model. Presented measurement results clearly illustrate a tendency of increase of the range estimation error with the distance from the measured NodeB. Naturally, this phenomenon is inline with expectations. However, more detailed analyses of the accuracy of pure RSCP-based estimation indicate that the range error is not essentially correlated with the distance in locations near the measured NodeB. Precisely, it was observed that for UE-NodeB distances below 250 m, range estimation error is not distance-dependent. In turn, for larger distances, range estimation error (E) observed in a macro cellular environment is directly proportional to the UE-NodeB distance (d) and can be described by a simple function:

E (d ) = ( A d ) y .

(2)

In (2), A is a proportional coefficient for a median E(d) and y is a lognormal variate defined as Y=log(y) is a Gaussian random variable, having zero mean and a standard deviation (y). Based on the executed measurement trials, the defined model is the most accurate for A at a level of 0.35 for urban and 0.47 for suburban/rural macro cellular propagation environments. In turn, range error deviations represented by variate y correspond to observations with y defined as 2.2 dB and 2.5 dB for urban and suburban/rural environments, correspondingly. Fig. 2a illustrates an example of the generated range error in a function of the distance for two considered environments. The correspondence of the modeled error to the observations from performed measurements is presented in Figs. 2b and 2c (urban) and Fig. 2d (suburban/rural). In illustrations, the UE-NodeB distance in urban environment is limited to 450 m due to cell sizes in the measured area. In turn, range errors with modeled representations for suburban/rural scenario are illustrated for distances up to 1 km. 5. Conclusions

hand, in the suburban/rural environment proposed techniques do not improve the accuracy of the signal strength based positioning. In the measured scenarios, range error in RSCP-based estimation has been observed to increase with the distance for the locations above 250 m from the measured NodeB. Slightly different behavior of the range estimation error in urban and suburban/rural environments has been described in the distance-dependent model with appropriate parameterization. Acknowledgment Authors would like to thank Elisa Networks Oyj for enabling the measurement campaign, Nemo Technologies Ltd for providing a measurement tool, and European Communications Engineering (ECE) Ltd for helpful comments. References
[1]

Two entirely mobile-based positioning techniques for UMTS have been proposed. Introduced location methods are based on RSCP and DS measurements and do not involve any changes in the network. Therefore, the technique once implemented in a user terminal can be exploited in a wide range of networks. Performance assessment has been carried out by extensive field measurements in various environments. Reported results illustrate the accuracy of the range estimation, which naturally influences the final accuracy of the position estimation. Performed analysis indicates that the RSCP is correlated to some extent with the DS. Thus, the proposed Hybrid 2 method improves the accuracy of the pure RSCP-based positioning in the most of the measured environments. However, the obtained results yields that the observable DS is not essentially dependent on the distance from the serving NodeB. Therefore, the Hybrid 1 method is not applicable. Practically, only in the measured micro cellular urban environment in locations relatively near the serving Node B (mean UE-NodeB distance 144 m), the Hybrid 1 improves the accuracy by 50%. In other analyzed scenarios the Hybrid 1 does not improve the accuracy. In turn, conducted measurements illustrate that in urban macro cellular environments, the range error is reduced from 108 m for RSCP-based positioning to 94 m when the Hybrid 2 algorithm is exploited. The improvement of the Hybrid 2 is expected to be more significant in urban macro cells if the average distance to the measured NodeB is larger. Namely in the analyzed scenario with 330 m mean UE-NodeB distance, the average range estimation error for the Hybrid 2 is 107 m, which is 20% lower in comparison to the positioning based on RSCP measurements. From the other

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

3GPP TS 25.305, UMTS; UE positioning in Universal Terrestrial Radio Access Network (UTRAN); Stage 2, ver. 7.1.0, Rel. 7. J. Borkowski and J. Lempiinen, Pilot correlation method for urban UMTS network, in Proc. European Wireless Conf., 2005, Vol. 2, pp. 465-469. Y. Zhao, Standardization of mobile phone positioning for 3G systems, IEEE Comm. Magazine, 7(50):108-116, July 2002. Liu Bio-Chieh, K. H. Lin, Jieh-Chian Wu, Analysis of hyperbolic and circular positioning algorithms using stationary signal-strength-difference measurements in wireless communications, IEEE Trans. on Vehicular Tech., 2(55):499-509, March 2006. E. S. Lohan, R. Hamila, A. Lakhzouri, M. Renfors, Highly efficient techniques for mitigating the effects of multipath propagation in DS-CDMA delay estimation, IEEE Trans. on Wireless Comm., 1(4):149-162, January 2005. E. S. Lohan, A. Lakhzouri, M, Renfors, LOS estimation in overlapped WCDMA scenarios via adaptive threshold, in Proc. 4th IEEE Workshop on Signal Processing Advances in Wireless Comm., 2003, pp. 353-357. M. P. Wylie, J. Holtzman, The non-line of sight problem in mobile location estimation, in Proc. 5th IEEE Intl. Conf. on Universal Personal Comm., 1996, Vol. 2, pp. 827-831. Jeong Yangseok, You Heungryeol, Lee Chungyoung, Calibration of NLOS error for positioning systems, in Proc. 53rd IEEE Vehicular Tech. Conf., 2001, Vol. 4, pp. 2605-2608. M. P. Wylie-Green, S. S. Wang, Robust range estimation in the presence of the non-line-of-sight error, in Proc. 54th IEEE Vehicular Tech. Conf., 2001, Vol. 1, pp. 101-105. Digital Mobile Radio Towards Future Generation Systems, COST-231 Final Report, Available: http://www.lx.it.pt/cost231/final_report.htm. L. J. Greenstein, V. Erceg, Y. S. Yeh, M. V. Clark, A new path-gain/delay propagation model fo digital cellular channels, IEEE Trans. on Vehicular Tech., 4(43):837-847, November 1997.

Vous aimerez peut-être aussi