Performance Analysis of UMTS/HSPA at the Cellular Level

Joo Lopes, Lus M. Correia

Instituto Superior Tcnico / Instituto de Telecomunicaes Technical University of Lisbon, Lisbon, Portugal jmclo@ist.utl.pt, luis.correia@lx.it.pt

Lus Santo
Optimus Lisbon, Portugal luis.santo@sonae.com

Abstract This paper analyses and compares HSDPA and HSUPA capacity and coverage. For this purpose, a single user model is developed for the estimation of the maximum cell radius, which is then adapted to a multiple users and services scenario. The results from the single user model show that, in an indoor scenario, HSUPA can serve 1.22 Mbps up to 0.19 km, while HSDPA with 10 HS-PDSCH codes can serve 6.0 Mbps up to 0.17 km. In a pedestrian environment, HSUPA can offer 1 Mbps up to 0.7 km, while at the same distance HSDPA can offer almost 3.0 Mbps. The results from the multiple users simulator show an average cell radius of 0.28 km and 0.25 km for HSDPA and HSUPA, respectively. The HSDPA BS throughput is 2.39 Mbps, with an average throughput per user of 0.55 Mbps, while for HSUPA, the BS throughput is 0.62 Mbps with an average user throughput of 0.21 Mbps. Keywords: UMTS, Multiservice Traffic. HSDPA, HSUPA, Coverage, Capacity,

I.

INTRODUCTION

niversal Mobile Telecommunications System (UMTS) was introduced to address the growth and interest achieved by data communications and to increase throughput and capacity, [1]. The growth of data communications forced a further development of systems that would be capable of offering higher capacity, throughput and enhanced multimedia services, available to consumers anywhere, anytime, [2]. High Speed Downlink Packet Access (HSDPA) was set with an initial peak data rate of 1.8 Mbps, increased to 3.6 Mbps and later to 7.2 Mbps, starting the mobile Internet Protocol (IP) revolution, [3]. For the Uplink (UL), High Speed Uplink Packet Access (HSUPA) was launched, with expected peak data rates up to 1.45 Mbps and 5.7 Mbps with a Time Transmission Interval (TTI) of 2 ms in later releases, [3]. HSDPA and HSUPA, when deployed together, are referred to as High Speed Packet Access (HSPA), deployed on top of UMTS networks, minimising equipment upgrade, with practical data rates beyond 2 Mbps in Downlink (DL) and up to 1 Mbps in UL, keeping latency under 100 ms, [3]. For HSDPA, several papers on performance and capacity analysis can be found, although they refer mainly to the improvement over Release 99, and not the influence of the number of High Speed Physical Downlink Shared Channel (HS-PDSCH) codes on network throughput and capacity.

These analyses are mainly based on the influence of specific topics, as scheduling or admission control algorithms, not giving a global perspective of the system as a whole. Few papers are available on HSDPA, and even fewer on HSUPA. An analytical model for HSDPA cell radius calculation as a function of outage probability is presented in [4]. Constraints on HSDPA throughput and code multiplexing, as well as the impact of shared carriers in the cell radius, are not considered, and there is no multiservice differentiation. An analytical approach to HSDPA radio interface dimensioning is presented in [5], providing a link budget estimation; only 5 HS-PDSCH codes are considered, and the impact of deep indoor is not addressed. The influence in capacity due to the introduction of HSUPA is analysed in [6], but Multimedia Message Service (MMS) and E-mail are the only services considered. HSUPA end-to-end performance is studied in [7], showing low latency and enhancements in file upload, 5 and 15 being the only set of HS-PDSCH codes analysed, and no radio environment variations are considered. The main purpose of this work is to study HSDPA and HSUPA at the cellular level in a multiple users and services scenario, in particular, addressing coverage and capacity aspects. Recent studies only analyse one of the systems, being difficult to compare both, due to the use of different models and traffic considerations. These goals are accomplished thought the development and implementation of a multiple users and multiple services scenario. Cell coverage, capacity and user throughput, among others, are compared. A simple model to estimate the HSDPA and HSUPA single user throughput as a function of distance is implemented. These models are adapted to a multiple users simulator, allowing the analysis of HSDPA and HSUPA behaviour in real networks. Additionally, expressions to calculate HSDPA and HSUPA user throughput in different scenarios are given. The influence of the variation of the systems parameters in network performance, such as the number of HS-PDSCH codes and BS transmission power, are also assessed. Section II describes the models for the theoretical calculations. In Section III, the scenarios used, parameters and results are analysed. In Section IV the main conclusions are drawn.

978-1-4244-2517-4/09/$20.00 2009 IEEE

II.

MODELS

To assess HSPA capacity and coverage, two models were developed: the single user, and the multiple users. The former is intended to assess the maximum cell radius in a single user scenario, which can be used in the first phase of network planning, to estimate the maximum cell radius. The latter is intended to study HSPA performance with the objective of analysing a more realistic traffic scenario, with multiple users performing multiple services, non-uniformly spread. A. Single User Model The HSDPA single user model computes the maximum cell radius according to the desired throughput, the number of HS-PDSCH codes, type of scenarios and Base Station (BS) transmission power, among others. For HSUPA, the same approach is followed, but the parameter having higher influence on the cell radius is the desired throughput, since the inter- to intra-cells interference, as well as the activity factor, have a major influence only in a multiple users scenario. In these models, all network resources are available to the user. The HSPA single user model can be reduced to a unique expression for the cell radius calculation, the difference being in the estimation of the receivers sensitivity. The path loss is calculated using the link budget with the COST-231 WalfischIkegami model [8]. The cell radius can be given by:
EIRP dBm ] Pr [ dBm ] + Gr[ dBi ] M [ dB ] L'tt [dB] Ltm [dB] L'0 [dB] [

R[km] = 10
where:

20 + kd

(1)

EIRP: Equivalent Isotropic Radiated Power; Pr: available receiving power at antenna port; Gr: receiving antenna gain; M: total margin due to additional losses, as propagation and environment; L'tt = Ltt kd log(d[km] ) ;

Ltt: rooftop-to-street diffraction losses; kd: dependence of the multiscreen diffraction loss versus distance; L'0 = L0 20 log(d[ km] ) ;

L0: free space loss. Using polynomial expressions, HSDPA available throughput is calculated as function of the average High-Speed Downlink Shared Channel (HS-DSCH) Signal to Interference plus Noise Ratio (SINR), whereas for HSUPA, the available throughput is calculated as function of the received Ec/N0 (energy per chip to noise spectral density ratio) per BS antenna, [3]. B. Multiple Users Simulator The multiple users simulator consists of 3 main modules, based on [9]. In the first one, users are generated and spread non-uniformly, according to population density areas. Each user is assigned a service according to a list of services penetration percentage, with different throughputs associated to each user, according to the service requested. The second module places users, as well as BSs, according to theirs geographical information, and performs a simple network analysis regarding single user radius. HSDPA and HSUPA

modules, developed in C++, analyse networks capacity and coverage, through a snapshot approach, calculating instantaneous network results as average radius, both user and BS throughput, number of user per BS, satisfaction grade (defined as the ratio between users requested and served throughputs), as well as carried traffic. Results for the busy hour analysis, as the number of users served per hour, are also computed. HSDPA and HSUPA modules are independent of each other, and are simulated separately. Three rate reduction strategies, with different QoS requirements, are developed for the cases where the offered traffic exceeds BSs capacity, which was taken as 3.0, 6.0 and 8.46 Mbps for 5, 10 and 15 HS-PDSCH codes and 1.22 Mbps for HSUPA. These values are calculated at the application level, obtained from the maximum physical throughput by reducing the Medium Access Control (MAC), Radio Link Control (RLC), Block Error Rate (BLER) and application layers overhead. The proposed rate reduction strategies are: Throughput Reduction, where the throughput of all users is reduced by a pre-defined percentage; QoS Class Reduction, where all users throughput of the same service is reduced by a given percentage, according to a list containing services priorities; QoS One by One Reduction, where, for a specific service, each user throughput is reduced in a one by one strategy, also according to a priority list (based on application type). In the multiple users model, an interference margin comprising both intra- and inter-cell interferences and the influence of the number of user in the cell is introduced. The interference margin emulates cell load, varying from BS to BS according to the number of users in the coverage area. The interference margin reduces path loss, leading to a lower cell radius and user throughput, simulating cell breath. The same approach was taken for both HSDPA and HSUPA. The user is connected to the closest BS. Next, the throughput associated to the user distance is calculated, i.e., the maximum throughput that the BS can offer to the user, considering path loss. Cell radius is defined as the distance of the user served further away from the BS. Capacity also limits cell radius, as when the reduction strategies are executed, users further away of the BS have higher probability of being delayed, hence, decreasing cell radius. III. RESULTS ANALYSIS

A. Single User Scenarios The considered environments are: pedestrian, vehicular, and indoor with low and high penetration losses. These environments are distinguished by the different values of the slow and fast fading margins, as well as the extra penetration attenuation that differentiates the indoor low loss from the high loss one, 11 and 21 dB, respectively. The default scenario in this model considers a low loss environment, UL frequency of 1922.5 MHz and DL frequency of 2122.5 MHz, BS and user transmission powers of 44.7 and 24 dBm, respectively, and 10 HS-PDSCH codes for HSDPA and a diversity and Soft Handover (SHO) gain of 2 dB.

B. Multiple Users Scenarios In these scenarios, indoor environments represent the largest percentage of the overall users, as the services considerer in this analysis are, at present, mainly used with laptops in indoor environments. Six services with different QoS classes are considered. The penetration percentages, the QoS priority according to which the services are reduced, as well as the target throughput values for the services considered in the default multiple users scenario in UL and DL are presented in Table I [10]. The first services to be reduced are the ones with higher QoS priority value.
TABLE I. PENETRATION PERCENTAGE, QOS PRIORITY AND MAXIMUM THROUGHPUT. QoS Service Web P2P Streaming Chat E-mail FTP Penetration Percentage [%] 46.4 42.3 6.2 3.1 1.0 1.0 Priority 1 6 2 5 3 4 Maximum Throughput [Mbps] HSDPA 1.536 1.024 1.024 0.384 1.536 2.048 HSUPA 0.512 0.384 0.064 0.384 0.512 0.512

parameters. These models allow calculating results for the busy hour analysis. For HSUPA, the same traffic models cannot be used, due to the asymmetry in the UL services. For this reason, some of the traffic models characteristics must be redefined, namely the real asymmetric ones, like Web Browsing and Streaming, which use the UL channel mainly for signalling purposes. For Web Browsing, the same profile is used, but only 20 kB are considered for the UL data volume. The same data volume was used for Streaming. For File Transfer Protocol (FTP), instead of 10, a 2 MB file is used. Chat and Peer-to-Peer (P2P) are typically symmetric services, and for this reason, the same traffic model is used for UL and DL. In these simulations, the city of Lisbon was used, with 228 BSs and 1600 users over the service area. Ten simulations were done for each case, and average and standard deviation werere calculated for each parameter. C. Single User Model Results The highest single user throughputs for different cell radius for the indoor low loss environment are shown in Fig. 1.
HSUPA 9 8 7 6 5 4 3 2 1 0 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 Cell Radius [km] 0.4 0.45 0.5 0.55 Throughput [Mbps] ghj HSDPA 5 codes HSDPA 10 codes HSDPA 15 codes

The parameters for the link budget evaluation and the default values are listed in Table II, based on [9] and [10].
TABLE II. Parameters BS Transmission Power [dBm] User Transmission Power [dBm] Frequency [MHz] Number of HS-PDSCH codes User Antenna Gain [dBi] Maximum BS Antenna Gain [dBi] Half Power Beam Width [] User Losses [dB] Cable Losses between Transmitter and Antenna [dB] Noise Figure [dB] Diversity Gain [dB] SHO Gain [dB] Interference Margin [dB] Power for signalling and control [%] Reduction Strategy (10% rate reduction) DEFAULT LINK BUDGET VALUES. HSDPA 44.7 --2142.5 10 0 17 65 1 3 9 ----6 5 2 2 6 HSUPA --24 1952.5 ---

Figure 1. HSDPA and HSUPA throughput for different maximum cell radii for the indoor low loss environment.

R99: 25 --HSDPA: 10 QoS Class Reduction

Other scenarios with 5 and 15 HS-PDSCH codes, other reduction rate strategies and different services penetration percentages were also simulated. A simple empiric HSDPA and HSUPA traffic model is defined for each service, based on volume, duration, number of messages exchanged, reading time, among other

For HSDPA, the maximum throughput can be served to distances up to 0.15 km, without reducing the nominal bit rate, while for HSUPA that distance is around 0.30 km. For higher distances, HSDPA with 15 HS-PDSCH codes has the fastest user throughput decay. For a cell radius higher than 0.35 km, HSDPA with 10 or 15 codes has similar performance, due to the similar values of the SINR curves. For cell radius of 0.50 km, HSDPA can serve around 1 Mbps, regardless of the number of codes used. By using 15 instead of 5 codes, the cell radius increases by 0.15 km for a throughput of 3 Mbps. HSUPA can deliver 1 Mbps up to 0.35 km, with a small users throughput decay, which depends not only of the users distance, but also of the Spreading Factor (SF) of the Enhanced Dedicated Channel (E-DCH) channel. This fact together with the lower sensitivity of the Ec/N0 curves with distance, and the lower throughput range, explain the fact that HSUPA can deliver the maximum throughput at larger distances, showing lower user distance sensitivity. For HSDPA, and considering a user throughput of 3 Mbps and 10 HS-PDSCH codes, the maximum cell radii are 0.68, 0.28, 0.30 and 0.16 km for the pedestrian, vehicular, low loss and high loss environments. Regarding HSUPA, and for a user throughput of 1 Mbps, the maximum cell radii are 0.71, 0.29, 0.31 and 0.17 km for the same environments.

D. Multiple Users Capacity Results In the comparison between HSDPA and HSUPA total number of users served per hour, the influence of the HSDPA higher throughput per user can be noticed. HSDPA has the capacity to serve up to 3 times more users than HSUPA, which serves approximately 8 900 users in the busy hour, as the HSDPA average instantaneous throughput per user, near 0.56 Mbps, is almost 3 times higher than the HSUPA one. Regarding the HSDPA BS throughput, an average of 2.39 Mbps is obtained, compared to 0.62 Mbps for HSUPA. The global carried traffic is 85 GB/h for HSDPA and 18 GB/h for HSUPA. In this analysis, the P2P influence is also important, since this service is the only one that is volume symmetric, with significant different UL and DL throughputs. P2P has a high influence on the number of users served per hour, as it is the service with higher volume involved. The three rate reduction strategies have similar results when a macroscopic analysis considering all BSs in the network is performed, due to an average effect, as these strategies are not used in all BSs, but only in the ones that are overloaded. Only on a microscopic level, the improvement in the users instantaneous throughput can be noticed when the QoS One by One Reduction strategy is used instead of the QoS Class Reduction or the Throughput Reduction ones: the first one has the best results, as users are reduced individually. E. Multiple Users Radius Results The difference in the multiple users cell radius for the default scenario is around 15 %, with HSDPA achieving an average cell radius of 0.28 km, while HSUPAs is 0.25 km, this difference not being relevant for coverage purposes. Capacity also influences cell radius, as in systems with reduced capacity, more reduction rate strategies are executed, leading to a radius reduction. These are average results, as both systems can serve users at higher distances. Cell radii are also affected by the average distance between BSs, by users distribution, and by the services profiles distribution. For users performing both HSDPA and HSUPA, the latter system limits coverage, mainly due to the user terminal limitation transmission power. The cell radius is almost constant for the several scenarios simulated, except for the one with higher number of users, where HSDPA cell radius increases to near 0.32 km and to 0.28 km for HSUPA. Cell radius increases in this scenario as more users are placed away from the BS. F. Multiple Users Satisfaction Grade Results The average satisfaction grade (the ratio between users requested and served throughputs) for HSDPA and HSUPA, detailed by service is shown in Fig. 2. For Web, a satisfaction grade of 0.93 and 0.92 was achieved for HSDPA and HSUPA, respectively. The global satisfaction grade is 0.87 for HSDPA and nearly 0.91 for HSUPA. In the majority of the services, HSUPA has a higher satisfaction grade, due to the lower throughput requested by users. These results must be analysed together with the average number of served users; HSDPA can always serve more users than HSUPA. This fact is important for P2P,

Streaming and Chat, where HSUPA has a higher satisfaction grade, but a reduced number of served users, compared to HSDPA. For E-mail and FTP, the higher HSUPA satisfaction grade is mainly due to the lower user throughput requirements, as the difference in the number of served users is minimal.
Average Satisfaction Grade 1.00 0.95 0.90 0.85 0.80 0.75 0.70 0.65 0.60 Web P2P HSUPA HSDPA

Streaming Chat E-mail FTP Type of Service Figure 2. HSDPA and HSUPA satisfaction grade by service.

FTP and E-mail are the services with the highest difference, excluding Streaming - in UL, this service has a fixed throughput of 64 kbps, and so a satisfaction grade of 1 because these services have the maximum gap between UL and DL throughputs. Even though Web is essentially a DL service with few UL requirements, it has a higher DL satisfaction grade. Chat, which is the most symmetric service, has the same volume and maximum throughput in both links, a higher satisfaction grade, and lower standard deviation in DL than in UL, mainly due to the higher HSDPA capacity. HSUPA has a higher satisfaction grade for all the simulated scenarios, except for Chat and Web. G. Multiple User Throughput Expression HSDPA and HSUPA expressions and results presented in this subsection are obtained by limiting the distance to 0.4 km, to eliminate the distances where there is a higher dispersion in the user throughput, hence, having low statistical relevance. The comparison between HSDPA and HSUPA average instantaneous throughput per user, based on (2) and (3), is presented in Fig. 3, where the users throughput as a function of the distance to the BS can be obtained for a loaded network.
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 HSDPA HSUPA Average Instant. User Throughput [Mbps]

0 0.1 0.2 0.3 0.4 Distance [km] Figure 3. HSDPA and HSUPA average instantaneous throughput per user comparison for the scenario presented in Table II.

The HSDPA network trend line regarding distance and the average user throughput, (2), was computed for the multiple users default scenario: [Mbps] = 0.521 d[km] + 0.672 , (2) where: d: distance between BS and user.

Higher interpolation orders were studied, resulting in higher correlation values. A linear approach is taken, with a correlation of 0.90 and a mean relative error of 5.5 %. A correlation of 0.95 is obtained for a 6th order interpolation, but the gain is not worth given the increase in interpolation complexity, hence, a linear approach is taken. Unlike HSDPA, in HSUPA the average instantaneous user throughput is approximately constant for distances up to 0.4 km. This is due to the fact that, for the range of throughputs in this analysis, between 0 and 0.512 Mbps, the Ec/N0 value is always above the threshold, and the SF of the used E-DCH allows the maximum throughput to be achieved, being independent of the user distance. As seen for the single user analysis, and for the indoor low loss environment, HSUPA can deliver 0.7 Mbps to distances up to 0.4 km, Fig. 1. Therefore, even when introducing the interference margin due to the multiple users scenario, the Ec/N0 value does not cross the threshold, hence, there is no throughput reduction. The use of SHO also justifies the constant throughput: when the distance increases, one would expect a reduction of the user throughput, but the probability of the user being in SHO also increases, as the user is more likely to be near the cell edge, entering another BS coverage area, hence, there is an extra gain in the pass loss calculation. It can be easily recognised that, for HSUPA, the curve that best fits the users throughput in Fig. 3 for distances up to 0.4 km, is a horizontal straight line, and so, a constant equation is used: [Mbps] = 0.204 . (3) The correlation for (3) is only 0.12. This low value is due to the derivative of the linear interpolation being 0, with (3) having a mean relative error of 7.4 %. Considering the cell radius calculated for HSDPA and HSUPA, and using (2) and (3), the average user throughput at the cell edge is 0.53 Mbps for HSDPA and 0.20 Mbps for HSUPA. Expressions similar to (2) and (3) were computed for all scenarios, the HSDPA maximum throughput scenario being the one with the most significant increase. HSDPA can always serve a higher throughput, since HSUPA is limited by the user transmission power, as well as the BSs noise figure. For the distances considered in Fig. 3, HSUPA presents an almost constant user throughput, but for higher distances, the HSUPA throughput decreases due to the use of a lower SF. The decrease of the average user throughput in HSDPA can be explained due to the use of adaptive modulation, since when the user is closer to the BS a higher SINR is obtained, and the user can be served with higher throughput. This fact does not occur in HSUPA, as Binary Phase Shift Keying (BPSK) modulation is always used, and even though the user might obtain a higher Eb/N0 (energy per bit to noise spectral density ratio), the throughput is limited by the used SF. IV. CONCLUSIONS This paper addresses the comparison of HSDPA and HSUPA when deployed together, focusing on capacity and coverage aspects, considering equal conditions, as the number

of user offering traffic to the network and the same services penetration percentage. For the single user default scenario, the HSUPA cell radius is 0.19 km for the maximum throughput, 1.22 Mbps. For HSDPA with 5, 10 and 15 HS-PDSCH codes, the maximum cell radius, considering the maximum throughput for each number of codes is 0.21 km for 5, and 0.17 km for both 10 and 15 HS-PDSCH codes. The results from the single user model show that HSUPA can deliver the maximum throughput for larger distances than HSDPA, but the latter allows higher throughputs. The advantages of having a higher number of HS-PDSCH codes available fade away rapidly in indoor scenarios for distances further than 0.45 km. The comparison between HSDPA and HSUPA in a multiple users scenario shows that, for the default scenario simulated, HSDPA has a bigger cell radius than HSUPA, near 15 %, even though this difference is not significant regarding coverage aspects. HSDPA also has higher capacity, as can be seen by the number of users served in the busy hour analysis, since HSDPA can serve up to 32 000 users, while HSUPA can only serve 8 900. A mean satisfaction grade of 87 % and 91 % is obtained for HSDPA and HSUPA. Regarding the traffic carried by the network, HSDPA carries 85 GB/h, while HSUPA carries 18 GB/h. This difference is due not only to the different capacity of the 2 systems, but also due to the different characterisation of the traffic models used for DL and UL directions. Regarding QoS rate reduction strategies, the advantages of using the QoS One by One Reduction instead of the QoS Class Reduction or the Throughput Reduction strategies at the user level can only be noticed on a per BS analysis, with the first strategy having the best results. REFERENCES
Holma,H. and Toskala,A., WCDMA for UMTS, John Wiley & Sons, Chichester, UK, 2004. [2] Laiho,J., Wacker,J. and Novosad,T., Radio Network Planning and Optimisation for UMTS, John Wiley & Sons, Chichester , UK, 2006. [3] Holma,H. and Toskala,A., HSDPA/HSUPA for UMTS, John Wiley & Sons, Chichester, UK, 2006. [4] Avidor,D. and Mukherjee,S., Downlink Dimensioning for HSDPA Standard, Wireless Personal Communications, Vol. 32, No. 2, Jan. 2005, pp. 139-152. [5] Zanier,P. and Soldani,D., A Simple Approach to HSDPA Dimensioning in Proc. of PIMRC05 16th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, Berlin, Germany, Sep. 2005. [6] Helmersson,K., Englund,E., Edvardsson,M., Edholm,C., Parkvall,S., Samuelsson,M., Wang,Y. and Cheng,J., System Performance of WCDMA Enhanced Uplink, in Proc. of VTC05 Spring 61st IEEE Vehicular Technology Conference, Stockholm, Sweden, June 2005. [7] Peisa,J., Ekstrm,H., Hannu,H. and Parkvall,S., End-to-End Performance of WCDMA Enhanced Uplink, in Proc. of VTC05 Spring 61st IEEE Vehicular Technology Conference, Stockholm, Sweden, June 2005. [8] Damasso,E. and Correia,L.M., Digital Mobile Radio Towards Future Generation, COST 231 Final Report, 1999 (http://www.lx.it.pt/cost231). [9] Costa,P., Ladeira,D., Correia,L.M. and Santo,L., Planning of UMTS Cellular Networks for Data Services Based on HSDPA in Proc. of EW2007 13st European Wireless Conference, Paris, France, Apr. 2007. [10] Lopes,J., Performance Analysis of UMTS/HSDPA/HSUPA at the Cellular Level, M.Sc. Thesis, IST-UTL, Lisbon, Portugal, Mar. 2008. [1]