The impact of radio propagation predictions on urban UMTS planning

Mathias Coinchon*, Ari-Pekka Salovaara#, Jean-Frdric Wagen*+ *Wavecall SA, EPFL- PSE/B, 1015 Lausanne, Switzerland # Sonera, PO BOX 487, 90101 Oulu, Finland +EIA-FR, Univ. of Applied Sciences of Western Switzerland in Fribourg, 1705 Fribourg, Switzerland
This investigation outlines the importance of accurate radio propagation predictions in the planning for UMTS radio networks. The UMTS radio technology based on variable spreading DS-CDMA techniques is flexible but complex. The sensitivity of a UMTS radio network to the radio environment and traffic conditions is well publicized but rarely quantified. Theoretical analysis and commercial implementation of CDMA cellular systems also tend to show that radio network planning, simplified on one hand because of the lack of frequency planning, could be cumbersome due to difficulties linked to the so-called cell breathing effects. Since this effects is related to the traffic conditions and the radio propagation environments, the more accurate the coverage predictions are, the closer the expected performance will be from the measured performance. To quantify the advantage of accurate radio coverage predictions a case study was undertaken for a UMTS radio network planning in a 5km2 area of Paris (France). This study in an urban environment concludes that a conventional propagation model could lead to erroneous planning with less than expected quality of service, unacceptable interference, and more base stations then necessary. An accurate ray-tracing model integrated in an UMTS-capable planning tool allows the radio network designer to reach optimal levels for the base station deployment and configuration while meeting the expected service level requirements. Index TermsRadio propagation model, ray tracing, UMTS, Radio planning, simulation.

combines many parameters and can be approximately fitted to measurements. The ray tracing based model [9] is a deterministic physical model that gains its accuracy by computing radio ray propagation in the environment, taking into account the geographical information: terrain height, buildings coordinates and heights, and trees. The conventional COST231-Hata model and the ray-tracing model have been used within a commercially available planning tool [10]. Predictions and simulations of a UMTS radio network artificially deployed over an urban area have been computed. The availability of geographical data kindly provided by ISTAR [11] lead our investigation to be conducted in a 5.4 km2 portion of the city of Paris. The specific investigations carried in this paper is claimed to be generic enough to lead to general conclusions. The propagation models used are described briefly in the next section. The scenario considered for the UMTS planning case study over a portion of Paris is explained in Section III. Qualitative and quantitative results are then provided in Section IV. Finally, concluding remarks in Section V summarize the advantages of accurate coverage prediction for UMTS radio network planning. II. PROPAGATION MODELS The COST231-Hata model used a curve fitted power-law model derived from several measurements performed on various European cities [7]. Obviously this model usually requires a number of measurements campaigns in order to tune it and thus provides satisfying results. Ideally, a differently fitted model should be used for each cell, but this is unrealistic, as it would require too many measurements. Furthermore, measurements are obviously impossible in a new area where the base station are not already deployed or in the case of prospective studies before network deployment. Test transmitters can be used but this technique is usually cost intensive. The COST231-Hata model used here has been tuned with some measurement routes in the same area of Paris. The results lead to a mean prediction error of 0 dB and a standard deviation of 10 dB. The performance of a propagation model cannot be fairly described by these two numbers especially in urban environment. However, the mean and standard deviation of the prediction error are commonly used as performance metrics [e.g., 8]. Thus, they are provided here for reference purpose.

I. INTRODUCTION This investigation analyses the difficulties faced by the planners of future UMTS radio networks that have been reported in many recent publications such as [1-6]. The crucial importance of radio propagation prediction in urban UMTS network planning is emphasized. The sensitivity of UMTS radio planning with respect to coverage predictions is known to be an issue [4-6] but is rarely detailed. An investigation was performed in [7] but for a GSM network. Since a GSM radio network is expected to be more robust due to its TDMA/FDMA scheme and its relative simplicity, especially in terms of services, it can be easily argued that UMTS planning requires even more accurate radio predictions. The effects however remain to be investigated in some details. Thus, this investigation compares, the performance of radio network planning using on one hand a common empirical propagation model and a more accurate ray tracing based model on the other hand. The empirical propagation model is the well-known COST231-Hata model [8]. This model

The ray-tracing model used here is a physical deterministic model [9,12,13]. An efficient ray tracer computes the diffractions and reflections that occur on the terrain and buildings detailed on geographical databases. Single and multiple edge diffraction theory and Fresnel reflection are used to predict the received field strength transmitted by a base station antenna. Since the ray-tracing model is based on a physical model, the only tuning would be on the electrical parameters of the terrain and buildings. It has been shown [9,12,13] that for all measurements taken so far in various cities, the electrical parameters do not need to be changed. This fact may be explained from the averaging behaviour of the multiple rays. In addition, it is after all not surprising that the electrical parameters of building construction average over a city do not differ significantly from city to city (with the exception of area with a concentration of metallic glass skyscraper). Furthermore, there is a relative insensitivity of the results in the range to be considered for the values of the electrical parameters. Thus, the ray-tracing model used here does not require any tuning. The model applies to any urban and sub-urban environments. The accuracy of the ray-tracing model considered here has been validated by numerous measurements in different cities. Typical values of mean prediction error are less than 3 dB and 7 dB or less for the standard deviation. Ray tracing predictions for the same measurement routes that were used to tune the COST231Hata model showed a typical mean error of 0.1 dB and a standard deviation of 7.5 dB. III. UMTS SCENARIO The UMTS differs significantly from the GSM system [1-6]. The differences lead to the UMTS being a more flexible system but at the cost of being more sensitive to the environment. The capacity and the achieved service quality in a UMTS network depend on many different parameters like power, code, processing gain and service distributions. Unlike in GSM, users in a UMTS network will share the same 5 MHz frequency block with other users in the same cells and also with users in other cells. Frequency planning is not needed. The CDMA radio technology of UMTS provides the required orthogonality between users at the cost of accurate power control, some restriction in code allocation, and limited cell range depending on the interference level. Since the interference level depends on the activity of all other users and on their allocated power, the cell range depends on the traffic and power distribution in the cell. This well-known cell breathing effects lead to an iterative approach to compute the power allocated to each user and the level of interference [6]. These computations depend directly on the user locations and on the services (data rate) used. Thus, UMTS radio network planning is performed through simulations of a user and service distributions. Thus, state-of-the-art planning tool for UMTS radio network uses Monte-Carlos simulations [e.g., 10]. In a network simulation, the users and their service (data rate) requirements are spread over the area under investigation. The required power and the interference are computed

iteratively for each user. The power is calculated through steps adjusting it according to the power control algorithms. The simulation stops when power converges. The result is a snapshot of the network for a typical user configuration. The snapshot shows where the service requirements of each user can be fulfilled or not. Thus, coverage results can be presented as a plot of dots (user locations) of different colour depending on whether or not the expected service quality is achieved. Since the whole simulation steps depend on power computations, the propagation predictions become crucial. To analyse the effect of using a conventional propagation model as compared to a more accurate model, the following procedure was used. First the radio propagation predictions are computed using the ray tracing model and then using the empirical COST231-Hata model. The results are described in the next section (Section IV). Both predictions are then used for the UMTS simulations presented in Section V. Here are the main parameters taken for the study: Hexagonal planning on the city of Paris: 3-sectored sites, 65 degrees beam width antennas, ~2 meters over the roof (~30 meters from ground), 700 meters between sites. Base Station maximum power: +43 dBm, pilot power: +30 dBm Mobile: max. power +21dBm, active set 3, noise factor 8 dB. 800 users/km2; traffic repartition: 50 % speech, 30% 64 kbps service, 20% 144kbps service. (Speech: 0.4 calls/hour & 120 seconds average duration; 64kbps: 0.1 calls/h & 288s; 144kbps: 0.1 calls/h & 192s.) Speech: coding factor 1.75, activity factor 0.7 Uplink (UL)/downlink (DL). Required Eb/N0 6dB UL, 8dB DL. 64kbps and 144kbps: coding factor 2, activity factor 0.4 UL, 0.8 DL. Required Eb/No 3dB UL, 4dB DL. Maximum load factor: 75% (equivalent to a 6dB noise rise or interference margin). Maximum BS power per service: +30 dBm for speech, +36dBm for 64kbps, +36 dBm for 144kbps. Working zone 5.4 km2 (but taking into account base stations outside this zone, over the whole city). The coverage predictions have been computed on a distance extending up to 1.5 km from each base station. The coverage results are discussed below. IV. COVERAGE MAPS Figure 1 shows the received power predicted by the COST231-Hata model. For clarity only a portion of the area under investigation is shown. The pattern obtained in Figure 1 is typical for a power law based empirical model used in an urban environment. The building layout is not taken into account. The 3-sector antenna patterns are clearly seen from the shape of the results. The lobes in the vertical pattern of each antenna explain the alternating colours along a radius away from each antenna.

Thus, Figure 3 and 4 shows only the area where the received power is above 80 dBm for at least two base station antennas. Fig. 3 and 4 provide the overlap results for the COST231Hata and ray-tracing model, respectively. The overlapping is severely underestimated in the empirical prediction case. The regions with overlap are concentrated near the base station when predicted by the empirical model. However, the more accurate ray-racing model shows that the inter-cell interference is wider spread especially in open areas. The users in these regions (where the interference was underestimated) will not only suffer themselves but also will lead to a decrease in quality or even a lack of service for others in the network because of the power control algorithm. This leads to a loss of offered UMTS capacity as shown in the next section. V. UMTS NETWORK SIMULATION Based on the coverage prediction shown above, Monte-Carlo simulations of users in a UMTS network were computed as described in section III. For a given user and traffic distribution, the coverage predictions are used to compute the offered service based on UMTS simulations. The results of one user and traffic distribution are gathered in a so-called snapshot. The results presented here are a superposition of 30 snapshots. In Figures 5 and 6 each dot represents a mobile user. A dark colour (red or purple) means that the user cannot be served using the service that was required. Lack of sufficient power is coded in red. Outage due to interference is shown in purple. Most interfered (purple) dots are mainly in the open areas and the insufficient power (red) dots are the remaining dark dots. The simulations based on the empirical COST231-Hata coverage predictions are far more optimistic then when using the physical ray tracing propagation model: almost no outage is found. This would be ideal if this result was correct. But this would be a severe problem for any network operator desiring to implement such a real UMTS network. Expectation of good service quality would not be met but this fact would discover only after deployment. Corrective actions or the re-design of the network would drain human and financial resources.
Fig. 2 Signal strength with ray tracing propagation predictions [9]

Fig. 1 Signal strength from empirical propagation predictions.

A. Qualitative results The simulations based on the empirical COST231-Hata coverage predictions are far more optimistic then when using the physical ray tracing propagation model: almost no outage is found. This would be ideal if this result was correct. But this would be a severe problem for any network operator desiring to implement such a real UMTS network. Expectation of good service quality would not be met but this fact would be discovered only after deployment. Corrective actions or the re-design of the network would drain human and financial resources.

Figure 2 shows the results for the ray-tracing model [9]. The canyon effect of streets and the impact of the buildings on the propagation are obvious. Comparing Figures 1 and 2, it is observed that the field strength predicted by the empirical COST231-Hata model is underestimated in most streets and in open spaces. This is readily explained by the contribution from diffractions and reflections that cannot be taken into account in the empirical. Since UMTS radio network is sensitive to interference, it is of interest to visualize the area where strong interference may occur due to cell overlap.

results of Figure 7. The users who in Figure 6 had not enough power are now assigned to serving sites. This increases slightly the overload of the network.

Fig. 3 Cell overlapping (-80dBm threshold) with empirical propagation predictions

Fig. 5 Superposition of 30 UMTS simulations snapshots based on classical empirical propagation predictions. Less than 1 % rejections.

Fig. 4 Cell overlapping (-80dBm threshold) with ray tracing propagation predictions

It could be pointed out that since the empirical model does not take into account buildings, and since some simulated users are located in a building (several red dots in buildings), a fairer comparison could be made between the empirical and the ray tracing models. Indeed, the results (in Figure 6) of the ray tracing model include a 20 dB extra loss for inbuilding penetration. Removing this extra loss lead to the

Fig. 6 Superposition of 30 UMTS simulations snapshots based on ray tracing propagation predictions and 20dB building penetration loss. 20 % rejections. 6 % rejections because of too low signal power (~600 red dots). 9 % because of overload (~960 purple dots).

power limit set for a given service has been reached. Some other users have a mobile power outage. It means the necessary power to reach the base station with a given service is exceeding the available power. These users would have to switch to lower data rate services or move to an area where the propagation path loss is lower. Considering the results, it appears that we can increase the performance of the hexagonal layout. By investigating the problem of noise rise; it appears that the origin of interference is coverage overlapping. Some transmitters are not really needed and thus generate mainly interference instead of capacity relief. By removing some transmitters, the 5 crossed sectors in Figure 8, the performance have increased. Half less rejections have been obtained. This removal of base station is only a rough optimisation. The traffic of the suppressed sectors will be handled by surrounding cells. Note that if we suppress the same base stations when using the COST23-Hata coverage predictions, the results show spurious problems because of the insufficient signal level. The importance of propagation predictions in UMTS radio planning with simulations was examined. The classical model predictions are far too optimistic and so it hides the problem that will occur in reality. Furthermore, it is not possible to predict locations where problems will appear. With an accurate ray-tracing tool, a radio network planner is able to detect problem areas and act on the network planning before deployment. B. Quantitative results As mentioned above the results are averaged over 30 MonteCarlo realizations of the user and service distributions described in Section III. In the UMTS simulations performed, an average of 344 users were spread in the area under investigations. On average 181 users (~50%) are using the voice service, 99 (~30%) are using 64kbps connections and 64 (~20%) on 144kbps connections. With the classical COST231-Hata model, only an average of 3 rejections (1%) is obtained while, with the accurate ray-tracing model, an average of 68 rejections (20%) is obtained with a 20 dB building penetration loss. When the 20 dB building penetration loss is not taken into account fewer rejections are obtained but still 43 users are rejected (12%). The classical propagation model leads to predict (43/3=)14 times less rejections than the raytracing model. Thus, UMTS network planning using a simplified coverage model or a not perfectly tuned propagation model could lead to very erroneous results. The resulting network performance would then be far below expectation. After removing 5 transmitters producing overlapping coverage, the resulting network leads to fewer rejections (22 or 6.4%). We did not try to improve the results further. However, these preliminary results shows that a classical

Fig. 7 Superposition of 30 UMTS simulations snapshots based on ray tracing propagation predictions and no (0dB) building penetration loss. 12 % rejections, mainly because of overload saturation (8% or 1100 purple dots). Close to 1% rejections because of too low signal power (red dots).

Fig. 8 As in Fig 7 but with sites removed to avoid interference (red crossed serctors). 6 % rejections, half because of overload saturation (~240 purple dots). Close to 1% rejections because of too low signal power (red dots).

The major problem in Figure 7 is the overload saturation. The noise rise limit is reach and the base station admission control does not let more users come in. This is a consequence of too much uplink interference. So these users are not allowed by the base station to access the service they ask. If the base station is out of power it means that the

propagation model could lead to a poor and costly UMTS network. As a further example of the results, the data services (at 64kbps and 144kbps) are considered. The total data rates requested by users on the downlink are given in Table 1. The offered traffic computed from the UMTS simulations is only close in the case of predictions from the classical COST231Hata model. More accurate propagation predictions lead to an offered traffic that is roughly 10% less than the required traffic. These results are in line with the results obtained above for the rejection ratio. Removing some overlapping cells improve the results but only slightly. Thus, a complete re-engineering of the UMTS radio network is probably required. Again, the results shows the danger of planning UMTS services based on erroneous propagation predictions.
Table 1 Results for data services in [Kbytes/s] (useful downlink rate after decoding)

UMTS planning is more complicated than GSM voice planning. It is very dangerous to think of one minimum received signal level to achieve coverage and one maximum interference threshold to insure quality. Each service requires specific threshold values and the network behaviour and size changes with traffic. Data transmission adds also another layer of complexity. Data throughput and quality of service variations are directly tied to variations of radio propagation. The need of an accurate propagation model is now more important than with GSM. By using a physical ray-tracing model the network operator can optimise the service level to meet customers expectations. It was shown through a rather typical case of planning in an urban environment that a conventional empirical propagation model could lead to accept a set of base stations for a UMTS radio network since, for example, only a 1% rejection/call ratio is predicted while more accurate radio coverage predictions, based on ray tracing for example, enable the planners to detect problems (e.g., a 12% rejection/call ratio) and to take the appropriate solutions (e.g., remove, displace or re-configure some base stations). The same conclusion was reached by analysing the effects on data services.
VII. ACKNOWLEDGEMENTS Thanks to ISTAR (http://www.istar.com), who provided the building and terrain data necessary for the ray tracing predictions. Thanks to Forsk (http://www.forsk.com), for providing licenses and training for their UMTS planning and simulation tool Atoll. Last, but not least, Wavecall management is thanked for the support in producing this paper.
[1] [2]

Service [kbps]
Av. tot. Requested traffic [Kbytes/s] Av. tot. Offered traffic [Kbytes/s] COST231-Hata Ray tracing with 0dB bldg penetration Ray tracing with 5 BS removed VI. CONCLUSION

64
647 640 559 597

144
947 921 794 831

UMTS simulations are heavily based on path loss and traffic prediction. If predictions are inaccurate, the results will be also inaccurate. This paper considers a typical urban planning example to quantify the relationship between the inaccuracy of the propagation predictions and the effects on the planning of the UMTS radio network. Current empirical propagation prediction models are known to be inaccurate in the first hundred meters near the base station antenna although a better accuracy is to be expected further away, usually at the cost of making measurements and then tuning the model. It is anticipated that UMTS sites density will be higher than for GSM to insure higher data rate services. Thus the accuracy of the propagation models in the first few hundreds of meters near the base station is becoming more important. Conventional propagation models are based on a power law distance dependence and cannot easily take into account the propagation paths guided by the street (canyon effect). This inaccuracy tends to lead to an underestimation of the overlapping between cells. Thus less interference than to be expected is predicted. For a UMTS radio network, more interference means less capacity and so less offered service and/or at a lower quality. Furthermore, under estimating the path loss over a given area means that users only need a low power. Thus, the predicted capacity for this zone will be higher then in reality because users will, in reality, need more power than it was predicted and so will make that the soft capacity limit to be reached more rapidly.

[3] [4] [5]

[6]

[7]

[8]

[9] [10] [11] [12]

[13]

VIII. REFERENCES D.Wong, T.J. Lim: Soft Handoffs in CDMA mobile systems; IEEE Personal Communications 12/1997. B.Melis, G.Romano: UMTS W-CDMA, Evaluation of radio performance by means of link level simulations, IEEE Personal Communications 06/2000. H.Holma, A.Toskala: WCDMA for UMTS; Wiley, ISBN 0-47172051-8. J.P. Castro: The UMTS network and radio access technology. Wiley, ISBN 0-471-81375-3 K.S Gilhousen, I.M Jacobs, R. Padovani, A.J. Viterbi, L.A. Weaver, C.E. Wheatley: On the capacity of a cellular CDMA system; IEEE Trans. on vehicular technology vol.40 no2 , may 1991 L. Correia, editor: Wireless Flexible Personalised Communications: COST 259, European Co-operation in Mobile Radio Research, Wiley, 2001, ISBN 0-471-49836 T. Krner, R. Fauss. Impact of digital terrain databases on the prediction accuracy in urban areas-comparative study, Proc. EPMCC97, Bonn, Germany, Oct. 1997. COST 231, Digital mobile radio towards future generation systems, Final Report, COST Telecom Secretariat, European Commission, Brussels, Belgium, 1999. WaveSight is the ray tracing model for urban areas developed by Wavecall. SA (http://www.wavecall.com). Atoll v.1.9 User Manual, 2001, Forsk SA, France. Terrain height 5mx5m raster, 3-D building vectors, ISTAR, France. Karim Rizk: Propagation in microcellular and small cell urban environment , Thesis #1710 (1997), Swiss Federal Institute of Technology of Lausanne. (downloadable on www.wavecall.com). K. Rizk, J.-F. Wagen, F. Gardiol, Two Dimensional Ray Tracing Modeling For Propagation Prediction In Microcellular Environments, IEEE Trans, Veh. Technol., vol 46, no 2, pp. 508-517, May 1997.