Impact of Intra- and Inter-Cell Interferences on UMTS-FDD

Transcription

1 Impact of Intra- and Inter-Cell Interferences on UMTS-FDD Hugo Esteves (1), Mário Pereira (1), Luis M. Correia (1), Carlos Caseiro (2) (1) Instituto Superior Técnico/Instituto de Telecomunicações, Tech. Univ. Lisbon, Lisbon, Portugal (2) Vodafone, Lisbon, Portugal Abstract The goal of this work is to study the interference impact on a UMTS cellular network, in FDD mode. A UMTS network simulator was updated, which now includes the interference calculation. The results show that the inter-to-intra cell interferences ratio is always less than.65, increasing when slow fading is added and decreasing when the number of available carriers in the system increases. Different values for the interference margin can be applied according to the zone, being less than 3 db. A cell radius optimisation can reach a 1 m increase in a low user density zone and, approximately, 13 % in an indoor scenario. I. INTRODUCTION Interference is part of every mobile cellular communications system developed until today, and since it constitutes a limitation to both radio network capacity and quality of service provided to users, its study remains important. Although UMTS is already implemented in many parts of the world, there are still not many studies and conclusions about the impact of interference on radio network performance, namely in a perspective of users being non-uniformly distributed in the service area, and offering multi-services traffic; this work intends to be a step forward towards a better understanding of this problem. In the FDD mode of UMTS, due to its nature, interference can be caused by Mobile Terminals (MTs) on Base Stations (BSs), i.e., the uplink () case, or on the reverse condition, i.e., the downlink () one. However, there is no interference among MTs, or among BSs (as it would be possible if, e.g., the TDD mode would be under analysis). Therefore, only the and situations are considered. In the case, one needs to distinguish interference caused by MTs within the cell, intra-cell, or by MTs from adjacent cells, inter-cell. In the case, the BS can cause interference on an MT due to partial loss of orthogonality between the different codes used for all users in the same cell, intra-cell, but the power transmitted from BSs in adjacent cells can also cause interference on an MT, intercell. In [1], a model is presented for the calculation of intra-cell interference in. The model considers the total power transmitted by the BS, the orthogonality factor and the propagation losses between BS and MT. It uses a Gaussian distribution to describe slow fading, but it does not take multi-service users into account. In [2], intra-cell interference is calculated, considering the transmitted power from MTs, as well as the number of users. Perfect power control is assumed, and inter-cell interference is taken as known a priori. Intra-cell interference is presented as depending on inter-cell one, the number of users, the service used, and the target equivalent Signal-to-Noise Ratio (SNR), E b /N. In [3], the capacity is analysed as a function of the inter-to-intra cell interferences ratio. A model for intra-cell interference is presented, considering the received power on the BSs, the number of users, their distance to the BS, and the activity factor for voice. However, it does not consider multiservices. Many times, inter-cell interference calculation is done by multiplying the number of users in a cell by the average interference offered in this cell, which is the same to say that any user inside this cell contributes with the same value of interference, regardless of his/her location, e.g., [4] for. However, this kind of calculation, being suitable for some types of analyses, e.g., real-time interference simulations, is really not adequate if one intends to analyse the effect of users dispersion in the service area. A model based on the number of users, their path loss, slow fading, and the cell area, is presented in [5], but users are uniformly distributed, and only one service is considered. In [6], an inter-cell interference density analysis is performed, assuming perfect power control; the number of users is taken into account, as well as the received signal power and the activity factor, according to the user s service, and the model calculates the average inter-cell interference per cell, being necessary to use a user distribution in the cell area. Multi-service scenario is also supported in this model, developed from [4]. In [7], one can find a model for inter-cell interference calculation as well, taking BS total transmitted power, path loss, and slow fading into account, but it does not assume a multi-service scenario. Usually, in radio network planning [8], a fixed value is taken for the inter-to-intra cell interferences ratio (i factor), i.e.,.65, as well as for the interference margin in the power budget, i.e., 3 db. This work addresses the validity of these values under several multi-service traffic scenarios, for nonuniform realistic distribution of users, using a radio network simulator perspective. In Section II, the interference problem is presented, together with the interference models used in this work. Section III contains a brief description of the simulator, while

2 the analysis of results is done in Section IV, where several cases and scenarios are presented. Finally, conclusions are drawn in Section V. II. INTERFERENCE MODELS A. Intra-Cell The model used for the calculation of the intra-cell interference in, on MT i is given by [7]: ( ) α [ ] I = P P L W (1) Intra, i Total, BS BS MTi BS MTi where P Total,BS is the total power transmitted by the BS, P BS MTi is the power transmitted by the BS to the MT in which interference is being calculated, and L BS MTi is the propagation loss between BS and MT, which in this work is calculated by the COST 231-Walfish-Ikegami [9]. The orthogonality factor α can take values between (for complete orthogonal codes) and 1 (absence of orthogonality). In, interference is given by [6]: G Intra, j = BS j MT ηg j, g g= 1 I P N [ W] where P BSj MT is the power received at BS j from an MT (as a result of perfect power control assumed at the BS, the received power is equal for all MTs using the same service in the cell), η g is the activity factor of service g, N j,g is the number of MTs using service g on the cell of BS j, and G is the total number of services used. B. Inter-Cell In, the model used for inter-cell interference, in an MT i using a service g is [1]: NBS a Inter, i Total, BS α j j j= 2 ΔLj 1 1 [ ] I = P r W (3) where P Total, BSj is BS j total transmitted power, including antennas gain, N BS is the number of interfering BSs, r j -a is the path loss, a representing path loss exponent, ΔL j is associated to slow fading, following a statistical distribution with zero mean and a certain standard deviation, and r j represents the distance between the interfering BS j and the MT. Factor α was introduced due to the lack of the orthogonality factor in the original model. The path loss also follows the COST 231-Walfish-Ikegami propagation model. Concerning slow fading, it is treated as a random component of the propagation loss, for each user, being given by a log-normal distribution, according to the statistical distribution defined by [9]. As far as is concerned, the following expression [6] is used: (2) NBS G Inter, j BS j TM η k g k, g k= 1, k j g= 1 I = P N A [ W] (4) where 2 ( γδl) r A= e (5) N kg, a kn, a n= 1 rjn, and P BSj TMk is BS j power received from the MT that is being covered by an adjacent cell k, N k,g is the number of users using service g in interfering cell k, r k,n and r j,n are, respectively, the distance from MT n using service g to BSs k and j, γ=ln(1)/1 is a constant, and ΔL is associated to slow fading. In the original model [6], a non-uniform user distribution statistical function in the cell area is used, with a certain standard deviation and zero mean. In this work, that approach is not appropriate, because users are statistical generated in the network, after which interference is calculation. Hence, instead of integrating the user distribution function in the cell area, a sum of all users distance to theirs serving BSs is carried out, as presented in (5). Beyond this, it was necessary to perform a change, modifying N k,g for a users sum, because one can not consider that all users covered by a certain sector, using the same service throughput, have the same slow fading attenuation value, as it was assumed in the original model. This attenuation affects the path loss, as presented in (4), but the impact is calculated on a per user basis, instead of the factor ( L) e γδ 2 being calculated for each cell. Again, as in, a lognormal distribution with zero mean and a given standard deviation is used to simulate slow fading. III. SIMATOR The simulator used in this work is a spatial one, being composed of three main modules [1]: user generator (SIM), network creation (UMTS_Simul), and network dimensioning (Net_opt). The SIM module is responsible for creating users on a statistical approach, through previous defined parameters. Based on traffic information, users distribution, multiple services profiles, and clutter regions in the study area, the SIM module creates users, assigning them specific characteristics. A service, out of a set of 8 (voice and videotelephony over circuit switch (CS), and video streaming, ftp, , Internet browsing, location based, and multimedia messaging over packet switch (PS)) is statistically allocated to users, which are non-uniformly distributed in the service area. The information resulting from SIM is received by UMTS_Simul, in which, together with network parameters information, it creates and analyses the radio network in the studied area. The Net_opt block performs the liaison between BSs and MTs, by calculating BSs coverage area, BS load, among many other radio network parameters.

3 Interference calculation is done in the Net_opt module, being responsible for calculating intra- and inter-cell interferences, and comparing the new values obtained for the interference margin with the ones obtained by using.65 for the inter-to-intra cell interferences ratio, the i factor. IV. ANALYSIS OF RESTS A. Reference Scenario It is important to have a reference scenario, to which comparisons are performed when some parameters are modified. This reference scenario should be chosen in such a way that it represents a possible real network. The used reference service is based on 128 kb/s for data PS traffic, users being in a pedestrian scenario. The percentages used for services penetration are the ones defined in the MOMENTUM project [11]. The city of Lisbon is taken as example, and number of users generated in the service area is 7 145, on average, leading to a blocking probability for voice users less then 1 %. A radio network of UMTS BSs colocated with GSM ones is taken, in the sense that no radio network optimisation is at stake, but rather comparing interference evaluation in a given network. At least, 1 simulation runs were performed for each situation, in order to get statistical relevance. Calculations were done for interference, capacity, interference margin, inter-to-intra cell interferences ratio, as well as other performance indicators. In what follows, several situations are analysed. Detailed information can be found in [1]. B. Variation on the Reference Service The reference service has impact on two main aspects: the nominal cell radius, since the higher the throughput is the lower the cell area covered by sectors will be, and the allocation of a throughput associated to a certain service by a user, since he/she can see his throughput successively reduced, until it reaches the reference value, in the worst case, due to cell coverage restrictions. It is likely that with a high throughput reference service, the number of users using the service they asked for will increase, and, as a consequence, a higher load in the network will exist, which will lead to a decrease on the cell radius. There is a balance between the network load and the cell radius, resulting in a negligible difference in the interference parameters in this case study. Intra- and inter-cell interferences, in both and, remain more or less constant when the reference service is changed, therefore, the i factor does not suffer significant changes, as it can be seen in Fig. 1. Since MTs need a higher power level to communicate with the BS when the service throughput is higher, generating more interference, cell radius decreases, thus, yielding a reduction in the overall network interference. Inter-cell interference is higher than intra-cell one, in, because, on average, the number of interfering MTs taken into account in the former is higher. With the low variation of the i factor previously observed (Fig. 1), other parameters, like the load factor and the interference margin, do not have a significant variation with the reference service. However, one should emphasise that, mainly in, the new load factor value, calculated with the new i, is much higher than the one calculated with i =.65 (almost the double). In, the new load factor value is close to the one calculated with i =.65, because the new i is also close to this value. These new values for the load factor result in a very different interference margin, compared with the one calculated with i =.65, mainly in, the difference between them reaching.9 db.,6,5,4 i,3,2,1 Fig. 1. i vs. reference service. 64 kb/s 128 kb/s 384 kb/s C. Variation on the Reference Environment The scenario can vary from pedestrian (reference environment), to vehicular and indoor. The main difference resulting from the simulations in the various environments is on the values of the effective cell radius, which decrease from pedestrian to indoor. Comparing the i factor, Fig. 2, a decrease is noticeable in, from pedestrian to indoor, resulting from a decrease in the inter-cell interference, and leading to a decrease in the load factor as well i.3.1 Fig. 2. i vs. reference environment. Pedestrian Vehicular Indoor On the other hand, due to the use of perfect power control, the intra-cell interference does not vary when the environment is changed. In, the i factor does not have significant changes when the environment varies, because the power transmitted by the BSs is similar, resulting in negligible changes on both intra- and inter-cell interferences.

4 The difference ΔM between an interference margin calculated with i =.65 and the one calculated with i obtained as a result of the interference calculations from the simulator is illustrated in Fig. 3. It can be seen that, when changing the environment, ΔM increases in, but in, as expected, there are no changes. One can also observe that, in, the differences between the margins are higher than in, which is due to the lower value of i. ΔM [db] Pedestrian Vehicular Indoor Fig. 3. ΔM vs. reference environment. D. Variation on the Maximum Number of Carriers In the reference scenario, 4 carriers are used, corresponding to a number of users of 7 145; when reducing the number of carriers available in the network (i.e., the maximum one available for a BS), the number of users is also reduced in order to maintain a blocking probability less than 1 %. Comparing the values obtained for the i factor, Fig. 4, there is a decrease in both and decreases when the number of carriers increases, as a result of the intra-cell interference being higher than the inter-cell one i.3.1 Fig. 4. i vs. number of carriers. 1 carrier 2 carriers 3 carriers 4 carriers there is a decreasing number of BSs with an increasing number of carriers. i Fig. 5. i vs. four carriers. 1st carrier 2nd carrier 3rd carrier 4th carrier A decrease is observed from the first to the fourth carrier, as a consequence of previous analysis on the number of carriers per BS. When taking only the first carrier into account, the value obtained for the case is near to the usual one of.65. E. Variation on the Users Service Penetration As previously mentioned, the reference scenario uses the service penetration from the MOMENTUM project. By changing the relative percentages among services, different traffic profiles are obtained, which are expected to have impact on the interference levels. Two new services penetration profiles were created (relative to the MOMEMTUM definition): in the Busy Hour Voice (BHV) one, voice becomes predominant relative to all data services; in the Busy Hour Data (BHD) one, the penetration of data services is similar (to MOMENTUM), but in BHD the main data services require higher data rates (e.g., ftp). As a consequence, in terms of load being put to the BS, BHV is less demanding than MOMENTUM, but BHD is the opposite. The values obtained for i are not that much different among the three situations. BHV presents smaller values, as expected, due to the lower value of the load. A comparison of the load factors obtained from taking i =.65 and the ones for simulations, in, Fig. 6, shows that in fact the different services penetrations, and the use of an appropriate model for the estimation of i, can lead to differences that impact on the interference margin, hence, on link budget, and ultimately on radio network design, and the required number of BSs. With only one carrier, the values obtained in are very close to the usual one of.65. As a consequence, the difference between margins will increase when the number of carriers increases. An analysis among carriers, in the 4 carriers situation, was also done. Fig. 5 shows results for i for each of the carriers, when using a maximum of 4 carriers in the network (the reference scenario), in order to show the interference behaviour in each carrier individually; one should note that not all BSs will have the maximum number of carriers, due to the non-uniformity of the spatial traffic distribution, and that Load factor [%] i obtained i=.65 MOMENTUM BHV BHD Service percentage Fig. 6. Load factor in vs. service percentage.

5 F. Interference in Different Zones of Lisbon An analysis of the interference was performed in the different zones of Lisbon, which have different traffic profiles (coming from the non-uniform distribution of users, as well as from the different characteristics they have, i.e., predominance of either business or residential markets). The map of Lisbon was divided into several small areas, Fig. 7, which enabled the calculation of users density, the overall average being 172 MT/km 2. With this approach, one can distinguish different zones according to the average users density in each one, and establish upper and lower thresholds. Zones that present more than 2 MT/km 2 were designated by High Density Zones (HDZs), whereas those that present less than 75 MT/km 2 are Low Density Zones (LDZs); the result is presented in Fig. 8. Average values for the BS density were calculated as well; as expected, there is a strong correlation between the two types of densities, as more BSs are located in areas with more users. LDZ), HDZ presents a higher load factor, hence, a higher interference margin. G. Adding Slow Fading The goal of adding slow fading in the calculations is, of course, to have a perspective closer to reality. The impact of slow fading was analysed by changing the standard deviation from to 1 db. By assuming perfect power control in the BS, slow fading has no impact on intra-cell interference, thus, only the inter-cell one suffers from the changes in path loss. Some changes are noticeable, namely on i, which presents higher values for an increase in the standard deviation; also, as expected, the variation of i around the mean value increases as well. Along with the increase of i, there is also an increase of the of the load factor, in both and. The load factor in is still lower than in, since the value of i is lower as well. As a consequence, there is impact on the interference margin. Fig. 1 represents the difference between the interference margin calculated with i =.65 and the one calculated using the new value of i. A degreasing trend is observed when a higher slow fading standard deviation is considered, which is due to the inter-cell interference increase, which leads to the increase of i, and consequently of the load factor and of the interference margin. Fig. 7. Lisbon with users. High Density Zone. Intermediate Zone. Low Density Zone. Fig. 8. The zones with different densities. In Fig. 9, one can see the results for the interference margin in both types of zones, for and. It is clear that different margins should be used in each type of zone, and even differentiated between and. MI [db] HDZ.8 Fig. 9. Interference margin vs. zone type. LDZ On the other hand, all these values differ quite a bit from the usual value of 3 db. This difference is a result from the strong relationship between the load factor and the number of users within each sector. By having a higher density of users (the average value in HDZ is around 8 times higher than in.5 ΔM [db] σ = db σ = 5 db σ = 1 db Fig. 1. ΔM vs. slow fading. As a conclusion, one can say that by adding slow fading, i increases, although not reaching the usual value of.65. H. Capacity results Some calculations were performed in order to evaluate the impact on system capacity at large, namely, on the nominal cell radius, from using lower values for i. The decrease obtained in the interference margin, coming from the lower values of i. lead to a decrease in the overall attenuation in the link budget, therefore, enabling the increase in the cell radius. Fig. 11 shows the values for the nominal cell radius, coming for different interference margin values, for different reference scenarios. Comparing the cases where the interference margin is 3 db (the usual value) or 1.2 db (obtained with the calculated values of i), a difference of 8 m (around 1 %) is obtained for a 128 kb/s service, and of 9 m for a 384 kb/s one, in the pedestrian reference scenario. This difference is even higher, around 1 m (more than 1 %), when the margin obtained

6 for LDZs is considered (1 db), and the new covered area is approximately exceeded in.52 km 2. Radius [km] M I [db] Pedestrian Vehicular Indoor Fig. 11. Nominal radius vs. interference margin for a 128 kbps service. In terms of percentage, Fig. 12, the other environments also present a somehow significant enhancement in terms of coverage area, when a lower interference margin is considered (instead of 3 db). When 1 db is taken, one gets the highest increase for the vehicular environment. On the other hand, it is noticeable that for indoors the increase, although being high, is almost independent of the differences in the interference margins (there is a small increase in absolute values); this is due to the high value of indoor penetration attenuation, and at a small scale to the E b /N one as well. Cell radius raise [%] pedestrian vehicular indoor M = 1.6 db M = 1.2 db M = 1 db Fig. 12. Nominal cell radius increase in percentage. V. CONCLUSIONS This work deals with the problem of interference in UMTS radio networks. It presents a perspective for interference estimation under several multi-service traffic scenarios, for non-uniform realistic distribution of users, using a radio network simulator perspective. Usually, in radio network planning, a fixed value is taken for the inter-to-intra cell interferences ratio, i.e.,.65, as well as for the interference margin in the link budget, i.e., 3 db. This work presents a study on these values, and their dependence on various parameters. A simulator with a statistical approach has been used to implement the models. The city of Lisbon, with a radio network co-located with the GSM one, was taken for simulations, with real traffic associated to it. The main conclusion is that the inter-to-intra cell interferences ratio is always less than.65. This parameter is higher in than in, on average being.39 and 2, respectively, in the reference scenario, without slow fading. By introducing slow fading, it increases to.46 in. It shows also a strong dependence on the number of carriers present at each base station. Regarding the interference margin, it stays always below the value of 3 db. Different margins should be used for different parts of the service area, since average values as low.5 db can be found in for zones with a low density of users. As a consequence, there is a non-negligible impact on system capacity, in this case expressed by an increase of the nominal cell radius, which can increase more than 1 % with the new values taken for the interference margin. The increase of the nominal cell radius can reduce the number of BSs used for the same quality of service, therefore, reducing the costs in infrastructures. REFERENCES [1] Chen,Y., Soft Handover Issues in Radio Resource Management for 3G WCDMA Networks, Ph.D. Thesis, Queen Mary College, University of London, London, UK, Sep [2] Mäder,A. and Staehle,D., An Analytic Approximation of the Uplink Capacity in a UMTS Network with Heterogeneous Traffic, Research Report No. 31, Institute of Computer Science, University of Würzburg, Würzburg, Germany, May TR/tr31.pdf [3] Owen,R., Jones,P., Dehgan,S. and Lister,D., Uplink WCDMA Capacity and Range as a Function of Inter-to-Intra Interference: theory and practice, in Proc. of VTC 2 Spring - IEEE Vehicular Technology Conference, Tokyo, Japan, 2. [4] Parvez,A., Impact of Actual Interference on Capacity and Call Admission Control in a CDMA Network, Master Thesis, University of North Texas, TX, USA, May [5] Akl,R., Hegde,M., Naraghi-Pour,M. and Min,P., Multicell CDMA Network Design, IEEE Transactions on Vehicular Technology, Vol. 5 No. 3, May 21, pp [6] Nguyen,S., Capacity and Throughput Optimization in Multi- Cell 3G WCDMA Networks, Master Thesis, University of North Texas, TX, USA, Aug [7] Sugano,M., Kou,L., Yamamoto,T. and Murata,M., Impact of Soft Handoff on TCP Throughput over CDMA Wireless Cellular Networks, in Proc. of VTC 23 Fall - IEEE Vehicular Technology Conference, Orlando, FL, USA, Oct sugano3vtccdmasofthandoff.pdf [8] Holma,H. and Toskala,A., WCDMA for UMTS, John Wiley & Sons, Chichester, UK, 24. [9] Damasso,E. and Correia,L.M. (eds.), Digital Mobile Radio Towards Future Generation - COST 231 Final Report, COST Office, Brussels, Belgium, [1] Esteves,H. and Pereira,M., Impact of Intra- and Inter-Cell Interferences on UMTS-FDD (in Portuguese), Graduation Thesis, Instituto Superior Técnico, Lisbon, Portugal, July 26. [11] IST-MOMENTUM (Models and Simulations for Network Planning and Control of UMTS), European Project, European Commission, Brussels, Belgium,