An experimental evaluation of network-based methods for mobile ...
AN EXPERIMENTAL EVALUATION OF NETWORK-BASED METHODS FOR MOBILE STATION POSITIONING Magne Pettersen1, Ragnar Eckhoff2, Per H. Lehne1, Tore A. Worren1, Elin Melby1 1
Telenor R&D, Snarøyveien 30, N-1331 Fornebu, Norway, {magne.pettersen, per-hjalmar.lehne, tore-arthur.worren, elin.melby}@telenor.com 2 Kongsberg Simrad, Strandpromenaden 50, PO. Box 111, N-3191 Horten, Norway, [email protected] Abstract - Different methods for network-based mobile positioning were compared experimentally. The Cell-id and TA method provided limited accuracy, the uncertainty ranging from 500 m to 3000 m for the median, dependant on area type. This could be reduced up to 25% by using alternative radii based on observations in the “doughnut” shaped prediction area. The Planning tool method improved the accuracy by approximately 30-50 % compared to this in urban area. The Forced handover method provided the best overall performance. In rural area the improvement was approximately 40-55 % compared to Cell-id and TA.
A. Cell-id and TA
I. INTRODUCTION
This is a relatively simple approach using knowledge about the base station (BS) position (from the cell-id) and the GSM parameter Timing Advance (TA), which provides a measure of the MS distance from the serving cell BS. This measure will not however have a higher resolution than that provided by the GSM bit rate, which is in the order of 500 m. The prediction using this method is in the shape of a “cut-off doughnut”, as illustrated in Figure 1, with an inner and outer radius relative to the BS position, and a left and right angle, relative to north. The radii are found from the current TA value. The angles are fixed for each cell and are dependent on the BS antenna pointing direction and the cell type, sector cell or omnidirectional cell.
Mobile location services are believed to become very important in the near future, both for existing systems, like GSM, as well as for future systems, like UMTS.
In the implementation a margin was added to the outer radius and subtracted from the inner radius, so that inner_rad-outer_rad was typically 1000 m.
Keywords – Location technology, positioning, propagation
There are a large number of possible applications using mobile location, such as navigation, mobile yellow pages or emergency services. Other potential useful aspects of accurate location estimates are e.g. criminal observation and prevention. The US Federal Communications Commission (FCC) has issued requirements for the necessary positioning accuracy for 3G mobile systems for reasons of emergency calls [1]. Since the accuracy of the localization of user equipment is the fundament for applying location-based services and because the different location based services will require a different degree of accuracy, it is essential to have knowledge about the obtainable accuracy. Therefore, an evaluation of different approaches to network-based location estimation was undertaken. The experiments and simulations described in this paper are for GSM, but the same principles can also be applied to other systems, such as UMTS. II. LOCATION METHODS A large number of methods to perform mobile station (MS) positioning exist, ranging from simple methods using cell-id only to more advanced approaches, such as Enhanced Observed Time Difference (E-OTD) [2] [3]. In this paper three methods were tested, having in common that they do not require severe changes into the GSM network:
0-7803-7589-0/02/$17.00 ©2002 IEEE
Figure 1. Illustration of the shape of the location prediction area using Cell id and TA In the implementation used the inner and outer radii for each TA was based on free space theoretical propagation values. However, it was observed that due to the propagation channel not always being line-of-sight, this approach tended to overestimate the MS-BS distance. Therefore, also a different approach was used where the radii were optimised based on observations rather than theoretical values. These radii were found based on an optimisation where the goal was that the probability of the MS being outside the outer
PIMRC 2002
radius or inside the inner radius was below 10% averaged over all area types. The cell-id and TA method was used as a reference method with which the other methods were compared. The results presented for the other two methods described below are from simulations using the alternative radii. B. Planning tool In this approach the predicted area using the Cell-id and TA method is reduced using signal level predictions from a planning tool, and comparing this to the actual signal level observed (RxLev) from various BSs. In GSM the RxLev parameter is available on the MS for serving cell as well as for up to 15 neighbouring cells. The method first defines a coverage area for the serving cell by finding the area with predicted level equal to or greater than RxLev minus an error margin. This is then repeated for the strongest neighbour, the second strongest neighbour and so forth. In the analysis up to three neighbours were used. The total predicted area is the overlay of these resulting coverage areas. The error margin was 10 dB, chosen because it is in the same order of magnitude as the uncertainty reported for typical high-performance propagation prediction tools [4]. Figure 2 illustrates the principle by showing how the original prediction area (circular) is reduced first by overlaying the serving cell coverage (a) and then the strongest neighbour coverage (b).
Figure 2. Illustration of principle using the Planning tool method. a) Reduced prediction area using serving cell coverage. b) Reduced prediction area using strongest neighbour coverage C. Forced handover This method assumes that TA is available from more than one BS simultaneously. This can be achieved by forcing the MS to perform one or more handovers. The total prediction area is found as an overlay of the area from the different BSs, using the Cell-id and TA method in each case. This is illustrated in Figure 3. In the simulations one or two neighbours were used.
a)
a)
b)
b)
c) Figure 3. Illustration of principle using the Forced handover method. A prediction area is found using Cell-id and TA from both the serving cell (a) and the strongest neighbour (b) resulting in an overlay total area (c) The drawback of this method is that the handovers will introduce extra traffic into the network. Also, the technical implementation is not straight-forward. III. EXPERIMENTAL APPROACH AND DEFINITION OF PARAMETERS A. Experimental approach The experiments were performed by driving a van and by simultaneous logging of information from a mobile positioning server, GSM test equipment and a D-GPS receiver. The positioning server provided the (serving cell) BS position, the inner and outer radius, and the left and right angles as required for the Cell-id and TA method described in the previous chapter.
•
Urban area: Generally a quadratic street structure with dense building blocks. Typical cell diameters up to 500 m.
•
Suburban area: Mainly wooden residential houses. Typical cell diameters 0.5 – 2 km.
•
Rural flat: Mainly farmland and woods, some smaller hills. Cell diameters typically 5-20 km.
•
Rural valley: Steep hillsides, mountains. Cell diameters typically 5-20 km.
The analysis for the Planning tool and Forced handover methods were only carried out for the urban and rural flat areas. B. Definition of parameters In all the methods the location prediction is in the form of an area. In the analysis two parameters were therefore used to characterize the prediction; the position uncertainty and the miss probability. These parameters were most often competing. 1) Position uncertainty For each measurement the distribution of the distance from the actual position (from the D-GPS) to the prediction area was found. For each area type the overall distribution for all the measurements were then found as well as the cumulative distribution function (cdf) resulting from it. These statistics were also found for subsets of the measurements, for instance for each TA-value. It is obviously desirable that the location uncertainty is as low as possible. This parameter is the most important one for services requiring accurate location knowledge, e.g. navigation services.
The GSM test equipment provided GSM parameters, such as RxLev for the strongest neighbours as required by the Planning tool and Forced handover methods. The TA values were not available from the neighbouring cells, however, as required for Forced handover. These values were therefore estimated during post-processing. This was done by taking the actual MS-BS distance and “drawing” a TA based on the distance probabilities for the different TAs earlier estimated from measurements.
2) Miss probability
The D-GPS position was used as a position reference for each measurement.
A. Cell-id and TA
Measurements were performed in four area types, and an extensive measurement series was undertaken for each of them. In each area type between 5000 and 10000 measurements was performed, representing at least 30 serving BSs. In all areas the cells were a mixture of sector cells and omnidirectional cells. The network consisted of both GSM900 and GSM1800 BSs, but outside the urban area few GSM1800 BSs were used. The area types were:
This is the probability, estimated from the measurements, of the actual position being outside the predicted area. This probability should be as low as possible, which is particularly important for e.g. emergency services. IV. RESULTS
Figure 4 shows the cdf of the uncertainty for urban, suburban, rural flat and rural valley. Table 1 shows the 50 percentiles for each area as well as the miss probability. The uncertainty is lowest in the urban area and largest in the two rural areas. This is mainly due to the most urban areas having the smallest cells, leading the values of TA to be lower, and the prediction areas to be smaller. Figure 5 shows the mean position uncertainty as a function of TA for rural flat area. The dependence between position uncertainty and
TA is close to linear. Similar results were seen for the other area types.
The main reason for the misses was errors in the radial direction, less often in angle. The two rural areas show similar performance in terms of uncertainty, but the rural valley area has higher miss probability. This is due to the complicated radio channel in the hilly area, amplifying the tendency to overestimate the MS-BS distance. Table 2 shows the 50 percentiles for each area as well as the miss probability using the alternative radii described in Chapter 2. The uncertainty is somewhat reduced compared to the original case, up to 25% for the median. The miss probability is reduced significantly in the rural areas.
Figure 4. cdf of position uncertainty in urban (dashdot), suburban (dashed), rural flat (solid) and rural valley (dotted)
Table 2. 50 percentile (m) of uncertainty and miss probability for the different area types using cell-id and alternative radii
Table 1. 50 percentile (m) of uncertainty and miss probability for the different area types using cell-id and TA
50 percentile uncertainty (m) Miss probability (%)
Urban
Suburban
450
670
12
10
Rural flat 2625
21
Rural valley 2825
30
50 percentile uncertainty (m) Miss probability (%)
Urban
Suburban
Rural flat
Rural valley
335
735
2125
2125
9
12
15
18
B. Planning tool Table 3 shows the 50 percentiles of the position uncertainty using the Planning tool method for urban and rural flat area, for serving cell only, and for 1, 2, and 3 neighbours, respectively. Comparing with Table 2 it can be seen that this method provides a significant improvement compared to the Cell-id and TA method in the urban area, approximately 3050 %. The improvement in rural area is limited. The lowest uncertainty was seen in the 3 neighbour case. Table 4 shows the miss probability in the same cases as in Table 3. The miss probability is increasing with the number of neighbours, as expected, but did not exceed 15%. Reducing the error margin (Section 2) from 10 dB to 5 or 0 dB would reduce the position uncertainty, but would significantly worsen the miss probability. Table 3. 50 percentile (m) of position uncertainty using the Planning tool method
Figure 5. Mean position uncertainty as a function of TA for rural flat area
Serving cell Serving cell + 1 neighbour Serving cell + 2 neighbour Serving cell + 3 neighbour
Urban 226 193
Rural flat 2527 2247
180
2133
172
2068
V. CONCLUSIONS Table 4. Miss probability (%) using the Planning tool method
Serving cell Serving cell + 1 neighbour Serving cell + 2 neighbour Serving cell + 3 neighbour
Urban 5 8
Rural flat 10 12
11
14
13
14
C. Forced handover Table 5 shows the 50 percentiles of the position uncertainty using the Forced handover method for urban and rural flat area, for one and two neighbours, respectively. The position uncertainty is lowest for the 2 neighbour case. The main reason for this is the tendency in the 1 neighbour case of the resulting area to consist of two unconnected areas, as seen in Figure 3 c). Comparing with Table 2 it can be seen that this method provides a large improvement compared to the Cellid and TA method, especially for rural flat area. The improvement in this case was typically in the range 40-55 %. For urban area the improvement was slightly poorer that for the Planning tool method. Table 6 shows the miss probability in the same cases as in Table 5. The miss probability is higher for the 2 neighbour case than for the 1 neighbour case, and it is higher for the rural flat area than the urban area. The miss probability could be reduced by using more conservative radii.
Table 5. 50 percentile (m) of position uncertainty using the Forced handover method
1 neighbour 2 neighbours
Urban 263 229
Rural flat 1423 905
Table 6. Miss probability (%) using the Forced handover method
1 neighbour 2 neighbours
Urban 17 19
Rural flat 24 31
Three methods for network-based MS positioning have been tested using results from an extensive measurement campaign. The Cell-id and TA method shows limited accuracy, the 50 percentile of position uncertainty being in the range 5003000 m, depending on area type. The most important parameter for the accuracy is the cell size, therefore the best results were found in an urban area. The miss probability was in the range 10-30 %, with the highest values seen in the rural valley area, due to the complicated radio channel in this case. Using alternative radii in the cell-id and TA method would improve the position uncertainty somewhat in most area types, up to 25% for the median. It would reduce the miss probability significantly in rural areas. The Planning tool method provided an improvement with respect to the Cell-id and TA method in the range 30-50 % for the position uncertainty in urban area, depending on the number of neigbouring cells taken into account in the prediction. In rural area this method showed minimal improvement. The miss probability would increase with the number of neighbours, but did not exceed 15 %. The Forced handover method showed the best overall performance. The improvement for positioning uncertainty compared to Cell-id and TA being in the range 40-55 % for rural area, dependant on the number of neighbours. The miss probability was relatively high however, up to 30 %. This could be improved by using more conservative radius values. All the measurements and simulations described in this paper were performed for GSM, but the principles are equally applicable to other systems, such as UMTS. ACKNOWLEDGEMENTS The work described in this paper was performed in cooperation with Telenor Mobile Communications. REFERENCES [1] The Federal Communications Commission. Enhanced Wireless 911 Requirements (E911), 1996 [2] Lindsey, M, et.al. GSM Mobile Location Systems. Onipoint Tech inc. July 1999 [3] Schneider, C. Mobile Location systems. Kjeller, Norway, Telenor R&D. (Telenor R&D Scientific Document N 23/2000), 2000 [4] Correia, L (editor). Wireless Flexible Personalised Communications. John Wiley & Sons Ltd. 2001
Telenor R&D, Snarøyveien 30, N-1331 Fornebu, Norway, {magne.pettersen, per-hjalmar.lehne, tore-arthur.worren, elin.melby}@telenor.com 2 Kongsberg Simrad, Strandpromenaden 50, PO. Box 111, N-3191 Horten, Norway, [email protected] Abstract - Different methods for network-based mobile positioning were compared experimentally. The Cell-id and TA method provided limited accuracy, the uncertainty ranging from 500 m to 3000 m for the median, dependant on area type. This could be reduced up to 25% by using alternative radii based on observations in the “doughnut” shaped prediction area. The Planning tool method improved the accuracy by approximately 30-50 % compared to this in urban area. The Forced handover method provided the best overall performance. In rural area the improvement was approximately 40-55 % compared to Cell-id and TA.
A. Cell-id and TA
I. INTRODUCTION
This is a relatively simple approach using knowledge about the base station (BS) position (from the cell-id) and the GSM parameter Timing Advance (TA), which provides a measure of the MS distance from the serving cell BS. This measure will not however have a higher resolution than that provided by the GSM bit rate, which is in the order of 500 m. The prediction using this method is in the shape of a “cut-off doughnut”, as illustrated in Figure 1, with an inner and outer radius relative to the BS position, and a left and right angle, relative to north. The radii are found from the current TA value. The angles are fixed for each cell and are dependent on the BS antenna pointing direction and the cell type, sector cell or omnidirectional cell.
Mobile location services are believed to become very important in the near future, both for existing systems, like GSM, as well as for future systems, like UMTS.
In the implementation a margin was added to the outer radius and subtracted from the inner radius, so that inner_rad-outer_rad was typically 1000 m.
Keywords – Location technology, positioning, propagation
There are a large number of possible applications using mobile location, such as navigation, mobile yellow pages or emergency services. Other potential useful aspects of accurate location estimates are e.g. criminal observation and prevention. The US Federal Communications Commission (FCC) has issued requirements for the necessary positioning accuracy for 3G mobile systems for reasons of emergency calls [1]. Since the accuracy of the localization of user equipment is the fundament for applying location-based services and because the different location based services will require a different degree of accuracy, it is essential to have knowledge about the obtainable accuracy. Therefore, an evaluation of different approaches to network-based location estimation was undertaken. The experiments and simulations described in this paper are for GSM, but the same principles can also be applied to other systems, such as UMTS. II. LOCATION METHODS A large number of methods to perform mobile station (MS) positioning exist, ranging from simple methods using cell-id only to more advanced approaches, such as Enhanced Observed Time Difference (E-OTD) [2] [3]. In this paper three methods were tested, having in common that they do not require severe changes into the GSM network:
0-7803-7589-0/02/$17.00 ©2002 IEEE
Figure 1. Illustration of the shape of the location prediction area using Cell id and TA In the implementation used the inner and outer radii for each TA was based on free space theoretical propagation values. However, it was observed that due to the propagation channel not always being line-of-sight, this approach tended to overestimate the MS-BS distance. Therefore, also a different approach was used where the radii were optimised based on observations rather than theoretical values. These radii were found based on an optimisation where the goal was that the probability of the MS being outside the outer
PIMRC 2002
radius or inside the inner radius was below 10% averaged over all area types. The cell-id and TA method was used as a reference method with which the other methods were compared. The results presented for the other two methods described below are from simulations using the alternative radii. B. Planning tool In this approach the predicted area using the Cell-id and TA method is reduced using signal level predictions from a planning tool, and comparing this to the actual signal level observed (RxLev) from various BSs. In GSM the RxLev parameter is available on the MS for serving cell as well as for up to 15 neighbouring cells. The method first defines a coverage area for the serving cell by finding the area with predicted level equal to or greater than RxLev minus an error margin. This is then repeated for the strongest neighbour, the second strongest neighbour and so forth. In the analysis up to three neighbours were used. The total predicted area is the overlay of these resulting coverage areas. The error margin was 10 dB, chosen because it is in the same order of magnitude as the uncertainty reported for typical high-performance propagation prediction tools [4]. Figure 2 illustrates the principle by showing how the original prediction area (circular) is reduced first by overlaying the serving cell coverage (a) and then the strongest neighbour coverage (b).
Figure 2. Illustration of principle using the Planning tool method. a) Reduced prediction area using serving cell coverage. b) Reduced prediction area using strongest neighbour coverage C. Forced handover This method assumes that TA is available from more than one BS simultaneously. This can be achieved by forcing the MS to perform one or more handovers. The total prediction area is found as an overlay of the area from the different BSs, using the Cell-id and TA method in each case. This is illustrated in Figure 3. In the simulations one or two neighbours were used.
a)
a)
b)
b)
c) Figure 3. Illustration of principle using the Forced handover method. A prediction area is found using Cell-id and TA from both the serving cell (a) and the strongest neighbour (b) resulting in an overlay total area (c) The drawback of this method is that the handovers will introduce extra traffic into the network. Also, the technical implementation is not straight-forward. III. EXPERIMENTAL APPROACH AND DEFINITION OF PARAMETERS A. Experimental approach The experiments were performed by driving a van and by simultaneous logging of information from a mobile positioning server, GSM test equipment and a D-GPS receiver. The positioning server provided the (serving cell) BS position, the inner and outer radius, and the left and right angles as required for the Cell-id and TA method described in the previous chapter.
•
Urban area: Generally a quadratic street structure with dense building blocks. Typical cell diameters up to 500 m.
•
Suburban area: Mainly wooden residential houses. Typical cell diameters 0.5 – 2 km.
•
Rural flat: Mainly farmland and woods, some smaller hills. Cell diameters typically 5-20 km.
•
Rural valley: Steep hillsides, mountains. Cell diameters typically 5-20 km.
The analysis for the Planning tool and Forced handover methods were only carried out for the urban and rural flat areas. B. Definition of parameters In all the methods the location prediction is in the form of an area. In the analysis two parameters were therefore used to characterize the prediction; the position uncertainty and the miss probability. These parameters were most often competing. 1) Position uncertainty For each measurement the distribution of the distance from the actual position (from the D-GPS) to the prediction area was found. For each area type the overall distribution for all the measurements were then found as well as the cumulative distribution function (cdf) resulting from it. These statistics were also found for subsets of the measurements, for instance for each TA-value. It is obviously desirable that the location uncertainty is as low as possible. This parameter is the most important one for services requiring accurate location knowledge, e.g. navigation services.
The GSM test equipment provided GSM parameters, such as RxLev for the strongest neighbours as required by the Planning tool and Forced handover methods. The TA values were not available from the neighbouring cells, however, as required for Forced handover. These values were therefore estimated during post-processing. This was done by taking the actual MS-BS distance and “drawing” a TA based on the distance probabilities for the different TAs earlier estimated from measurements.
2) Miss probability
The D-GPS position was used as a position reference for each measurement.
A. Cell-id and TA
Measurements were performed in four area types, and an extensive measurement series was undertaken for each of them. In each area type between 5000 and 10000 measurements was performed, representing at least 30 serving BSs. In all areas the cells were a mixture of sector cells and omnidirectional cells. The network consisted of both GSM900 and GSM1800 BSs, but outside the urban area few GSM1800 BSs were used. The area types were:
This is the probability, estimated from the measurements, of the actual position being outside the predicted area. This probability should be as low as possible, which is particularly important for e.g. emergency services. IV. RESULTS
Figure 4 shows the cdf of the uncertainty for urban, suburban, rural flat and rural valley. Table 1 shows the 50 percentiles for each area as well as the miss probability. The uncertainty is lowest in the urban area and largest in the two rural areas. This is mainly due to the most urban areas having the smallest cells, leading the values of TA to be lower, and the prediction areas to be smaller. Figure 5 shows the mean position uncertainty as a function of TA for rural flat area. The dependence between position uncertainty and
TA is close to linear. Similar results were seen for the other area types.
The main reason for the misses was errors in the radial direction, less often in angle. The two rural areas show similar performance in terms of uncertainty, but the rural valley area has higher miss probability. This is due to the complicated radio channel in the hilly area, amplifying the tendency to overestimate the MS-BS distance. Table 2 shows the 50 percentiles for each area as well as the miss probability using the alternative radii described in Chapter 2. The uncertainty is somewhat reduced compared to the original case, up to 25% for the median. The miss probability is reduced significantly in the rural areas.
Figure 4. cdf of position uncertainty in urban (dashdot), suburban (dashed), rural flat (solid) and rural valley (dotted)
Table 2. 50 percentile (m) of uncertainty and miss probability for the different area types using cell-id and alternative radii
Table 1. 50 percentile (m) of uncertainty and miss probability for the different area types using cell-id and TA
50 percentile uncertainty (m) Miss probability (%)
Urban
Suburban
450
670
12
10
Rural flat 2625
21
Rural valley 2825
30
50 percentile uncertainty (m) Miss probability (%)
Urban
Suburban
Rural flat
Rural valley
335
735
2125
2125
9
12
15
18
B. Planning tool Table 3 shows the 50 percentiles of the position uncertainty using the Planning tool method for urban and rural flat area, for serving cell only, and for 1, 2, and 3 neighbours, respectively. Comparing with Table 2 it can be seen that this method provides a significant improvement compared to the Cell-id and TA method in the urban area, approximately 3050 %. The improvement in rural area is limited. The lowest uncertainty was seen in the 3 neighbour case. Table 4 shows the miss probability in the same cases as in Table 3. The miss probability is increasing with the number of neighbours, as expected, but did not exceed 15%. Reducing the error margin (Section 2) from 10 dB to 5 or 0 dB would reduce the position uncertainty, but would significantly worsen the miss probability. Table 3. 50 percentile (m) of position uncertainty using the Planning tool method
Figure 5. Mean position uncertainty as a function of TA for rural flat area
Serving cell Serving cell + 1 neighbour Serving cell + 2 neighbour Serving cell + 3 neighbour
Urban 226 193
Rural flat 2527 2247
180
2133
172
2068
V. CONCLUSIONS Table 4. Miss probability (%) using the Planning tool method
Serving cell Serving cell + 1 neighbour Serving cell + 2 neighbour Serving cell + 3 neighbour
Urban 5 8
Rural flat 10 12
11
14
13
14
C. Forced handover Table 5 shows the 50 percentiles of the position uncertainty using the Forced handover method for urban and rural flat area, for one and two neighbours, respectively. The position uncertainty is lowest for the 2 neighbour case. The main reason for this is the tendency in the 1 neighbour case of the resulting area to consist of two unconnected areas, as seen in Figure 3 c). Comparing with Table 2 it can be seen that this method provides a large improvement compared to the Cellid and TA method, especially for rural flat area. The improvement in this case was typically in the range 40-55 %. For urban area the improvement was slightly poorer that for the Planning tool method. Table 6 shows the miss probability in the same cases as in Table 5. The miss probability is higher for the 2 neighbour case than for the 1 neighbour case, and it is higher for the rural flat area than the urban area. The miss probability could be reduced by using more conservative radii.
Table 5. 50 percentile (m) of position uncertainty using the Forced handover method
1 neighbour 2 neighbours
Urban 263 229
Rural flat 1423 905
Table 6. Miss probability (%) using the Forced handover method
1 neighbour 2 neighbours
Urban 17 19
Rural flat 24 31
Three methods for network-based MS positioning have been tested using results from an extensive measurement campaign. The Cell-id and TA method shows limited accuracy, the 50 percentile of position uncertainty being in the range 5003000 m, depending on area type. The most important parameter for the accuracy is the cell size, therefore the best results were found in an urban area. The miss probability was in the range 10-30 %, with the highest values seen in the rural valley area, due to the complicated radio channel in this case. Using alternative radii in the cell-id and TA method would improve the position uncertainty somewhat in most area types, up to 25% for the median. It would reduce the miss probability significantly in rural areas. The Planning tool method provided an improvement with respect to the Cell-id and TA method in the range 30-50 % for the position uncertainty in urban area, depending on the number of neigbouring cells taken into account in the prediction. In rural area this method showed minimal improvement. The miss probability would increase with the number of neighbours, but did not exceed 15 %. The Forced handover method showed the best overall performance. The improvement for positioning uncertainty compared to Cell-id and TA being in the range 40-55 % for rural area, dependant on the number of neighbours. The miss probability was relatively high however, up to 30 %. This could be improved by using more conservative radius values. All the measurements and simulations described in this paper were performed for GSM, but the principles are equally applicable to other systems, such as UMTS. ACKNOWLEDGEMENTS The work described in this paper was performed in cooperation with Telenor Mobile Communications. REFERENCES [1] The Federal Communications Commission. Enhanced Wireless 911 Requirements (E911), 1996 [2] Lindsey, M, et.al. GSM Mobile Location Systems. Onipoint Tech inc. July 1999 [3] Schneider, C. Mobile Location systems. Kjeller, Norway, Telenor R&D. (Telenor R&D Scientific Document N 23/2000), 2000 [4] Correia, L (editor). Wireless Flexible Personalised Communications. John Wiley & Sons Ltd. 2001
