wavecall The Reliable Wireless Connection The impact of radio propagation prediction on urban UMTS planning

wavecall The Reliable Wireless Connection The impact of radio propagation prediction on urban UMTS planning Mathias Coinchon 27.9.2001 WaveCall SA

Executive Summary This case study outlines the importance of an accurate radio propagation planning for UMTS networks. Due to the specific UMTS technology 3G users may experience much lower service levels then expected or even failures if the network is planned inaccurately. But dissatisfied customers and vendors of Value Added Services can be avoided if the network is planned accurately. Based on an urban area planning in the city of Paris this case study demonstrates the capabilities of sophisticated ray tracing model such as WaveSight. As a result, the study finds that WaveSight is able to accurately predict network coverage and transmission rates for a given user pattern. It remarkably outperforms the classical prediction models. Hence, optimal levels for hardware configuration and customer satisfaction can be achieved from the beginning. 2

Table of content 1 INTRODUCTION... 4 1.1 CONTEXT... 4 2 GENERAL OVERVIEW... 5 2.1 W-CDMA... 5 2.1.1 PROCESSING GAIN... 5 2.1.2 POWER CONTROL... 6 2.1.3 SOFT HANDOVER... 7 2.2 RADIO NETWORK PLANNING... 7 2.2.1 MAIN ISSUES... 7 2.2.2 OTHER ASPECTS... 8 3 CASE STUDY... 9 3.1 COVERAGE MAPS... 9 3.1.1 COVERAGE BY SIGNAL LEVEL... 9 3.1.2 BEST SERVER (-102 DBM)... 10 3.1.3 OVERLAPPING (-80 DBM)... 10 3.1.4 SOFT HANDOVER NO HANDOFF AREA... 11 3.1.5 64 KBPS SERVICE AREA DIFFERENT EB/N0 THRESHOLDS... 11 3.2 NETWORK SIMULATION... 12 3.2.1 QUALITATIVE RESULTS... 12 3.2.2 QUANTITATIVE RESULTS... 14 4 CONCLUSION... 17 ANNEXES... 19 UMTS... 19 ALLOCATED FREQUENCY SPECTRUM... 19 SCENARIO OF THE PARIS CASE STUDY... 20 Page 3

1 Introduction This case study aims to pick up the difficulties faced by the planner of UMTS networks. It will emphasise the crucial importance of radio propagation prediction in urban UMTS network planning. Hence, a comparison will be made between the common empirical propagation model that is based on statistical data and a deterministic (or physical) model that tries to be as accurate as possible by taking as much geographical information as possible into account (terrain, buildings, heights, ). We will first give an overview of the physical layer of UMTS. Afterwards, we will demonstrate the importance of using an adequate propagation model. A nominal network planning will be conducted in the city of Paris, comparing an empirical propagation model and the WaveSight ray-tracing model 1.1 Context During the last decade the world has experienced an explosion in mobile communication. The rise of the GSM (Global System for Mobile communications) technology was a major reason for this stunning development. GSM has been the first civilian digital system and is commonly termed as a 2 nd Generation system. However, GSM designers never had expected to achieve such results, and the system is now a victim of its own success story. Overcrowded networks and its weak performance in data transmission have severely outdated this system in a new decade that is characterized by the continuous rise of data transmission. The aim of 3 rd generation systems commonly known as UMTS- is to overcome these limitations by: Offering a flexible range of mobile services, from the voice services to the very high bit rate data transmission services Improving the efficiency in the use of the radio resource (spectrum, frequencies) Operators are now in the early stage of deployment of such network and face the problem of network planning. In UMTS, the capacity and functionality of the network depends as well on the kinds services used as on the distribution of the users themselves. Hence, the planning process of an UMTS network is completely different than with 2G systems (e.g. GSM). Consequently, the system must be simulated using statistical distribution of users among the environments, compiled with different services in use. Page 4

2 General overview General overview of the physical layer aspects of UMTS. 2.1 W-CDMA Second Generation networks use a completely different transmitting technology than 3G do. In the former users share the radio channel by utilizing different frequencies (FDMA, Frequency Division Multiple Access) during allocated slices of time (TDMA, Time Division Multiple Access). 3 rd generation systems will mainly use Wideband Code Division Multiple Access (W-CDMA) for radio channel sharing. Code Division Multiple Access (CDMA) has a completely different approach. The users spread the signal they have to transmit on a large bandwidth according to a code sequence. The longer the code is, the more the signal is spread on the frequency band. At the receiver end -in order to be able to catch the signal- the receiver must use the same code that has been used for transmission. Then the principle of CDMA is to allocate an independent code to each user in a cell. At the receiver the transmissions from other users using different codes will be seen as interfering noise. In general, to characterize the useful signal from the rest (noise+interference), we use the ratio useful signal power / total interference power (C/I = Carrier/Interference). All users are on the same frequency. Differentiation is done by codes. Other users are seen like interference. CDMA systems have in general a greater capacity than FDMA, TDMA systems for the same bandwidth, but there are some unique condition to meet. 2.1.1 Processing gain On the UMTS, the signal is always taking the same bandwidth that is 5 Mhz. However, a question arises. A voice signal that has a lower bit rate (12,2 kbps) than a video conference stream (128kbps) should require less bandwidth. So what is the difference if both must take at the end 5 Mhz bandwidth? Page 5

The answer is that the energy of the voice signal will be spread more than the one of the video conference stream. Voice Energy W-CDMA modulation Video Bandwidth 5 Mhz The consequence of spreading is that the more the signal is spread from its original bandwidth, the more it can be buried in the noise. The spreading is expressed by the processing gain, which is the ratio CDMA code bit rate/original bit rate. If we take the example of the voice signal, we have a processing gain of 25 db It means that the useful signal may have 100 less power than the noise and interference power! In comparison: A videoconference stream will have a processing gain of about 15 db and therefore requires a noise / power ratio lower than 10 time the useful signal power (taking an initial signal-to-noise ratio requirement of 5dB). As we have seen, the principle of W-CDMA is to add many users on the same carrier. As all other users are seen as noise to the single signal, the capacity of such a system is directly dependent on the interference coming from the power that everybody send on this carrier. Consequently, the more users are active; the lower is the transmission rate. Imprecise planned networks may result in voice-only communication in densely populated areas, reducing the core advantage of 3G to zero! 2.1.2 Power control As pointed out above, the capacity limit of CDMA systems occurs when the interference from other users become so high that the signal quality (C/I) becomes bad. That s why in order to increase the capacity, each user has to transmit with the minimum power to ensure that he causes the least possible interference to the others. In UMTS there are strong power control algorithms that update the power of the mobile and base station 1500 times per second. Power control is very important for WCDMA, a mobile using too much power may harm all the users on the same cell. Page 6

2.1.3 Soft handover To reduce the effect of shadowing on CDMA network, there s a feature called soft handover. Soft handover is the ability for one user to be in communication simultaneously with two or more base stations. This mechanism offers macro diversity in the downlink (gain of 2-3 db). Soft handover. One mobile can be in communication with many base station at the same time. So the risk of dropped call due to shadowing is reduced and the user can use less power to reach the base station. The ratio area of handoff region / area of a cell can be adjusted for example by adjusting the distance between base stations. Among the disadvantages of soft handover we have: Additional resources are used that are not available for other users. Signalling load increased on the fixed network The interference caused to other users by downlink increase because many BS transmits instead of one. 2.2 Radio network planning 2.2.1 Main Issues We ll see here the main aspects of UMTS radio network planning as far as of importance for propagation models. We have just seen that the behaviour of a network depends on many different aspects like power, interference and services used (through processing gain). These aspects depend directly on the user location and on the services used. In order to study and plan a network the users have to be taken into account. That s why UMTS radio network planning is performed through simulations trying to simulate a real-life network. Page 7

UMTS planning tools use Monte Carlo simulations. They spread the users on the environment and then compute the necessary power for each user to fulfil the requirements. The power is calculated through steps adjusting the power according to the power control algorithms, the simulation stops when the power converge. The result is a snapshot of the network for a typical user configuration. Among the well known planning tools there are Asset from Aircom, Odyssey from Logica, Atoll from Forsk, CellPlan, Planet,... The simulation in this case study are performed using Atoll, but they could have been done with the other tools as well. 2.2.2 Other Aspects The link budget is a summary taking into account all elements of the end-to-end transmission from the mobile to the base station: The link budget is different for each service, type of mobility (pedestrian, 50km/h, 120km/h in car, ) and terminal (phone, PDA,..). The result of the link budget is a maximum allowed propagation loss for the cell. In the simulations, the link budget is computed for each user. The propagation loss is given by the propagation model, if the loss is higher then the maximum allowed path loss the link can t be established. The maximum allowed propagation path loss is mapped to the coverage zone. Errors of the propagation prediction model leads to wrong coverage computation. To overcome this error and so increase the coverage probability a log normal fading margin is taken depending on the error standard deviation of the propagation model. This margin is a penalty on the link budget and so lead at the end of day to plan a higher base station density then needed (costly!). The more accurate the propagation model, the lower the penalty on the link budget is. For further details please see Annex A. Page 8

3 Case study In the following we will compare WaveSight s ray tracing capabilities with the traditional empirical model. An area with a site-to-site distance of 700 meters was selected. It is located in the city of Paris. At first the radio propagation model is computed, using WaveSight as well as the empirical Cost Hata model. Afterwards we will run UMTS simulations on both network predictions that point out the major differences that would be experienced by the users of a UMTS phone. For further technical details of the scenario please refer to Annex B. 3.1 Coverage maps 3.1.1 Coverage by signal level Here are screenshots showing different maps with on the left computation using Cost-Hata predictions and on the right WaveSight predictions >=-75 dbm >=-81 dbm >=-85 dbm >=-93 dbm >=-96 dbm >=-102 dbm It is obvious that the empirical model does not take into account the buildings. WaveSight on the other hand visualises the canyon effect of streets and the impact of physical structures (buildings) on the propagation. Page 9

3.1.2 Best server (-102 dbm) Here is the best server map taken with a minimum threshold of 102 dbm. Again on the left the empirical model and WaveSight on the right Again the differences are obvious. WaveSight accurately predicts the impact of buildings and gives a more precise picture. 3.1.3 Overlapping (-80 dbm) Cells are considered to overlap here if the level of the second is at least 80 dbm. Overlapping raises the noise -level. Hence, large areas over overlapping reduce the performance. Again, WaveSight results are pictured on the right. Page 10

The overlapping is underestimated in the empirical prediction case. The consequence will be more intercell interference than expected. The noise rise will be reached quicker. This would then lead to a loss of offered capacity. 3.1.4 Soft handover no handoff area This maps shows area where no handoff occurs in red. Smaller handoff areas are preferable as they reduce the burden for the infrastructure. 3.1.5 64 kbps service area different Eb/N0 thresholds These maps shows the service area for 64kbps data connection with different Eb/N0 minimum requirements. As usual, WaveSight predictions are on the right. Page 11

Eb/N0 > 10 db : Eb/N0> 8 db Eb/N0 > 5 db The Eb/N0 is directly related to the bit error rate of the transmission. A higher ratio indicates a better quality of the connection. But there s also something interesting with these maps: as the Eb/N0 is part of the link budget, this map can show the impact of difference in db on any element of the link budget. Again, the more accurate predictions of WaveSight enables the planner to build a network that features the quality level that is demanded by the operator and users. 3.2 Network simulation When considering the results from the above computations, what does this mean for the UMTS user? To answer these questions a number of users are spread over the area and their impact is simulated. 3.2.1 Qualitative results Here is a snapshot of many simulations. Each point represents a mobile user. Again, the right picture uses WaveSight. Simulations snapshots with classical empirical prediction and with WaveSight OK inactive base station power outage overload (load factor limit reached) mobile power outage Page 12

The simulation using empirical predictions is far more optimistic then the prediction with the physical model. However, the fact is that the empirical model does not take into account buildings while WaveSight does. So in order to be fairer in the comparison, the simulation has been recomputed with a propagation penetration loss factor for buildings of 0dB (instead of 20 db before). Simulation snapshots with WaveSight (no building penetration loss) before and after optimisation The major problem is that the load factor limit is reached and so the admission control does not let more users come in. This is a consequence of too much interference. So these users won t be allowed to access the service they ask. If the base station is out of power it means that the 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. In the next step, we have tried to increase the performance of the network. By investigating the problem of noise rise, it appears that the origin is coverage overlapping. Some transmitters are redundant and generate interference. By removing these transmitters (5 transmitters) the performance have increased. (see right picture). This is a rough way to optimise. The traffic of suppressed cells will be transposed on other one. If we suppress the same transmitters with the classical predictions simulations, problems rise because of low predicted signal level. We have here a demonstration of the importance of propagation predictions in UMTS radio planning with simulations. The classical model predictions are far too Page 13

optimistic and so hide the problem that will occur in reality. It is not possible to pinpoint location where troubles will appear because everything appears to be good. With WaveSight we are able to see the problems and act on the network until things get better. 3.2.2 Quantitative results For every case we have an average of 346 users spread on the environment (average of 181 on voice services, 99 on 64kbps connections and 64 on 144kbps connections). With the classical model we have an average of 3 rejections (0.9%) while with WaveSight we have an average of 68 rejections (20%) with a building penetration loss of 20 db and 42 rejections (12%) without taking into account buildings penetration loss. The classical predictions lead to 14 times less predicted rejections than with WaveSight! After suppressing 5 redundant transmitters and running again simulations with WaveSight predictions we go down to an average of 22 rejections (6.4%). If we look at the data transmission aspect (64kbps and 144kbps service) the total requested downlink data rate by users is 647 Kbyte/s for 64 kbit/s connections and 947 Kbyte/s for 144 kbit/s connections. The total is 1594Kbyte/s. The network will be able to serve only a fraction of this request that we call offered traffic. The offered traffic in the classical predictions simulations is 640Kbyte/s (64kbps service) and 921Kbyte/s (144 kbps) with a total of 1561Kbyte/s. The offered traffic in the simulation with WaveSight predictions (not taking into account building penetration loss) is 559Kbyte/s (64kbps service) and 794Kbyte/s (144 kbps) with a total of 1353Kbyte/s. The offered traffic of the optimised network (using WaveSight predictions) is 597 Kbyte/s (64kbps service) and 831 Kbyte/s (144 kbps) with a total of 1428Kbyte/s. The difference between simulations with classical predictions and WaveSight predictions is 1561-1353= 208 Kbyte/s. If we take today billing schemes, each MByte is billed about 9 (ref. Deutsche Telekom, D2-Vodafone). So if we assume a network planned with a classical model, we can give here an estimation of the money lost due the difference of what is expected and what will really happen. This give for this area (5km2) 9x0.208=1.9 lost per second, which correspond to about 6 840 lost per hour and about 54 720 lost for 8 hours! The simple optimisation that we have performed by removing redundant transmitters (less transmitters equal also less infrastructure costs) lead to a gain of 0.075*9*3600*8=19 440 for 8 hours. That s why the network has to be carefully planned. Experience with GSM shows also that users who have difficulty to obtain a service will finally give up and not using it anymore (e.g. people will give up making phone call in trains if the they get always dropped). There are 14 times more rejections in the physical propagation prediction case than in the empirical one. The fact is that empirical models are far less precise then physical one, resulting in far worse results than expected. Page 14

By using WaveSight these pitfalls can be avoided. It allows to plan a network in a way that the desired service level can be maintained at minimum hardware cost. Page 15

4 Conclusion We have seen that UMTS simulations are completely based on path loss prediction. So from this point it is obvious that if predictions are wrong, the results are nonsense. Particularly, empirical models are known to be inaccurate in the first hundred meters near the antenna and become more accurate further away. So, as with UMTS the site density will be higher, the accuracy of propagation in the first hundreds of meters from the base station is more important. Taking into account propagation paths such as canyon effects of street is crucial. The danger here is to underestimate the overlapping of cells and so meet more interference than expected. More interference in UMTS means less capacity and so less offered service to customers. On the other hand, over estimating the signal on some area would mean that users may need low power to reach services. The danger here is that planned capacity for this zone will be again higher then in reality because users will need more power than expected and so will make that the soft capacity limit is reached more rapidly. UMTS planning is far more complicated than GSM voice planning. It is not possible anymore to think of one minimum received signal level to achieve and one maximum interference threshold. Each service requires specific threshold values and network behaviour changes with traffic. Data transmission cannot be planned like voice service by saying good or not good. 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 before.by using a physical ray tracing model like WaveSight the network operator can optimise the service level for his customers and hence their satisfaction level. Page 17

Annexes Annex A UMTS The air interface of the UMTS Terrestrial Radio Access (UTRA) uses wideband CDMA (W-CDMA). Two modes are available: WCDMA/FDD (Frequency Division Duplex) for the paired frequency bands. It is suitable for large coverage areas and permits bit rates up to 2Mb/s in small cells. WCDMA/TDD (Time Division Duplex) for the unpaired frequency bands. It is particularly well suited for small cells and for asymmetrical traffic. It should be used for indoor. The chip rate is 3.84 Mcps, leading to a bandwidth of 5 MHz. The power control rate is 1500 Hz (1500 power updates per second) on both downlink and uplink. Allocated frequency spectrum The allocated spectrum for UMTS is 2x60MHz (1920-1980MHz for uplink, 2110-2117 MHz for downlink) for the paired bands and 15+20 Mhz (1900-1920MHz, 2010-2025MHz) for the unpaired bands. Example of UMTS frequency allocation in UK: Licence A (Reserved for a new entrant): 2x15 MHz paired spectrum plus 5 MHz unpaired spectrum. Licence B: 2x15 MHz paired spectrum. Licence C: 2x10 MHz paired spectrum plus 5 MHz unpaired spectrum. Licence D: 2x10 MHz paired spectrum plus 5 MHz unpaired spectrum. Licence E: 2x10 MHz paired spectrum plus 5 MHz unpaired spectrum Page 19

Annex B Scenario of the Paris case study Hexagonal planning on the city of Paris: 3-sectored sites, 65 degrees beamwidth antennas, ~2 meters over 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 : 50 % speech, 30% 64 kbps, 20% 144kbps Required Eb/N0 : 64kbps and 144kbps: +3dB UL +4dB DL; Speech: +6dB UL +8dB DL. Maximum load factor: 75% (6dB noise rise, interference margin) Maximum BS power per service: +30 dbm speech, +36dBm 64kbps, +36 dbm 144 kbps Working zone 5.4 km2 (but taking into account Base Station on the whole city) The prediction have been computed on a distance extending up to 1.5 kilometers from base station. The Cost-Hata model has been tuned with measurement routes in the same area of Paris. Page 20

Glossary BS: Base Station References CDMA: Code Division Multiple Access. Method that differentiates users by channel codes when using a common radio resource. FDD: Frequency Division Duplex. In a two-ways communication (duplex), the two channels are separated by using different frequencies. FDMA: Frequency Division Multiple Access. GSM: Global System for Mobile communications Handover: Switching of a communication between from one base station to another base station. Shadowing: Effect of an obstacle between a transmitter and a receiver in a radio link. Shadowing effect is a decrease of signal strength. TDD: Time Division Duplex. In a two-ways communication (duplex), the two channels are on the same frequency but are separated by the time where they can use it. W-CDMA: Wideband CDMA 1. D.Wong, T.J. Lim: Soft Handoffs in CDMA mobile systems; IEEE Personal Communications 12/1997. 2. B.Melis, G.Romano: UMTS W-CDMA, Evaluation of radio performance by means of link level simulations, IEEE Personal Communications 06/2000. 3. Proakis: Digital communications; McGrawHill 4. H.Holma, A.Toskala: WCDMA for UMTS; Wiley 5. 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 Acknowledgments Special thanks to Ari-Pekka Salovaara from Sonera who gave important guidelines for this case study. Thanks to company ISTAR (http://www.istar.com) in France who provided the building and terrain data necessary for WaveSight predictions. Thanks to company Forsk (http://www.forsk.com) in France for providing licenses and training for their UMTS planning tool Atoll. Page 21