You are on page 1of 5

1

HSUPA 5,76 Mbps mobiles: Throughputs evaluation on a live network


Jean-Baptiste Landre, Stefan Wendt, Orange Labs
Abstract The E-DCH transport channel uses multi-code transmission together with a special kind of QPSK modulation (BPSK modulation on I and Q axis) to achieve high throughput. On frequency selective channels the orthogonality between the spreading codes is lost and inter-code-interference (ICI) arises. This interference limits performances in case the base stations have a classical receiver (Rake), especially for HSUPA 5,76 Mbps mobiles. Realistic performances of those mobiles are shown in this paper. A new kind of simulation method, based on existing drive test measurements, is used. The measurements are combined with link level simulation results to obtain reliable performance estimation. Simulation results show that the throughput does not exceed 2,5Mbps if the base stations have a Rake receiver. Index Terms HSUPA, drive test measurements, cell planning

ntroduction of HSUPA has been a major evolution of WCDMA networks. HSDPA has provided a way to increase the achievable throughputs in the downlink, but the uplink capabilities did not match the downlink. HSUPA aims at providing higher data rates in the uplink than what is available with the uplink associated dedicated channels (DCH) to HSDPA in 3GPP release 5. To achieve high throughput, a new transport channel called Enhanced Dedicated Channel (EDCH) has been specified in 3GPP/UTRAN Release 6. It uses up to 4 codes of spreading factor 2 or 4 in parallel. Several transport format combination (E-TFC) are defined. Their selection depends on the radio link conditions and the coding rate is periodically adjusted to match the channel quality. A shorter 2ms Transmission Time Interval (TTI) is optionally used. Fast retransmissions are possible thanks to a Hybrid Automatic Repeat Request (HARQ) mechanism managed in the base station. Capabilities of HSUPA mobiles are currently increasing. At the beginning of HSUPA networks operations, 2Mbps mobiles have been available (category 5). Those mobiles can use up to 2 codes of spreading factor 2, and support a 10ms TTI. 3GPP has also specified 5.76Mbps mobiles (category 6), which can use up to 4 codes in parallel, and support a 2ms TTI. Those mobiles are the best performing ones that are specified in 3GPP/UTRAN Release 6 specifications [1]. They are however not available yet in the marketplace. It is important to keep in mind that those throughputs associated to one HSUPA mobile category are the peak ones. This means that such throughputs can only be achieved under optimal network conditions: very close to the base station, during off-hours, and in good fading conditions. It is then critical for operators to know the actual performance of those mobiles on their network. It is also critical to figure out what

I. INTRODUCTION

will be the performance of yet-to-be-released ones. In the case of 5.76 Mbps HSUPA mobiles, it turns out that the existing base stations can not support them: hardware upgrades are required. Before starting expensive upgrades, operators need to have an estimate of the performance improvement that can be expected from those mobiles. Several previous works evaluated the performance of HSUPA 10ms at the link level and extended them to the system level [6] [9] [10]. However the real topography of a network and the sites parameters from an actual cell planning database has never been considered. This paper provides an assessment of the throughputs that can be achieved by those 5,76 Mbps HSUPA 2ms mobiles under real network conditions, and compares them to what can be achieved by 2 Mbps mobiles. Drive test measurements are used to get the signal to noise ratio estimates. Actual network topology and real propagation conditions are therefore taken into account. Signal to noise ratio figures are then combined with link-level simulations results get an estimate of the HSUPA throughputs. This paper is organized as follows. In section II, the link level receiver structures are presented. In section III, the choice of the evaluation method is discussed. This method is then fully described in section IV. Simulation scenarios and results are presented and discussed in section V. II. LINK-LEVEL RECEIVER MODELLING A. System model All dedicated physical channels involved in the E-DCH transmission are spread with a sequence C and multiplied by a gain factor , with the condition that 2 = 1 . In the considered scenario, the E-DPDCH 1 and 3 are mapped on I axis, E-DPDCH 2 and 4 on Q axis. E-DPCCH is always

Figure 1: System model - E-DCH transmission

mapped on I axis and DPCCH is always mapped on Q axis [2]. Indeed, the UMTS UL uses a BPSK on I and Q axis

978-1-4244-2644-7/08/$25.00 2008 IEEE

2 (pseudo QPSK). E-DCH transmission can use 1, 2 or 4 EDPDCHs. We find therefore 3 situations: 1) 1 E-DPDCH: 2 codes on I, 1 code on Q. 2) 2 E-DPDCHs: 2 codes on I, 2 codes on Q. 3) 4 E-DPDCHs: 3 codes on I, 2 codes on Q. The I and Q axes are separated with the complex scrambling code. This allows considering the axis separately as BPSK transmissions. We consider a transmission of a sum of K spreading codes. The transmitted signal is then K (1) x(t ) = x (t ) We recall the results in the following. A transmission that uses K spreading codes in parallel requires a bank of K Rake receivers, each having L fingers. Considering the first Rake corresponding to the code index 0, the lth finger output is denoted by y ( l ) and is equal to
y ( l ) =
+

r (t + )a
l

* 0, i

(t )dt

(7)

The

y ( l ) are weighted by the coefficients wl* and combined


z = wl* y ( l ) = w H y
l =0 L 1

to produce the decision (8)

k =0

k k

where xk(t) is the signal corresponding to the kth code. It is expressed


x k (t ) =
i =

where y = [ y ( 0 ), y ( 1 ),..., y ( L 1 )]T and w = [ w0 , w1 ,..., w L 1 ]T . Finally the SINR is deduced as


SINR = Es , k w H hh H w w H Ru w H

(i ) a k ,i (t iT )

(2)

(9)

where T is the duration of the symbol sk(i) which is spread by the spreading waveform ak,i(t). Each sk(i) is a BPSK symbol and is assumed to have unity average energy: E(|sk(i)|2 )= 1 (3)

where E s , k = N k E c is the transmitted energy for symbol sk(i) on code k, h = [ h0 , h1 ,..., hL 1 ] t are the L channel coefficients, coefficients and

~ ~ w = h the weighting vector, h the estimated channel

The spreading waveform for the kth code and the ith symbol is a real chip sequence, {ck ,i ( j )}N k -1 , where Nk is the spreading
j =0

Ru

is

the

interference

and

noise

factor, convolved with the pulse shape filter, p(t). The amplifier multiplies every element of the chip sequence by

autocorrelation matrix, taking into account ISI, ICI and noise, defined as:

Ec giving
1 N k 1 (4) Ec ck ,i ( j ) p(t jTc ) N k j =0 where Tc is the chip duration. It is supposed that the spreading sequence is the product of the kth spreading code (which is a Walsh-Hadamard code) and the base-station specific pseudonoise (PN) scrambling code. The transmitted signal passes through the multi-path propagation channel characterized by the impulse response a k ,i (t ) =
h( ) = hl ( l )
l =0 L 1

R u = E s R ISI + E I R ICI + N 0 R n '


where E I =
K 1 k =1

(10)

s,k

. The cross and autocorrelation function

(5)

where L is the number of resolvable paths,

hl and l are

respectively the coefficient and the delay of the lth path. The received signal can be hence expressed as

between the waveforms of the different spreading codes is quite difficult to determine analytically for little spreading factors. Furthermore, channel estimation effects are difficult to take into account analytically as well. Also, the use of realistic power control is very difficult to treat theoretically. And finally, the information data is transmitted over all K codes (having not always the same spreading factor). The HSUPA QoS analysis needs a detailed calculation of all these points. Approximations cannot be done. Therefore, an analytical evaluation wouldn't be exact. We opted hence to take into account all these issues in our link level simulation in order to obtain exact values for SINR and QoS. C. Link level layer outputs The E-DCH performance evaluation at the link layer consists into providing by simulations the average QoS and throughput for the average chip signal to noise ratio defined as

r (t ) = hl x0 (t l ) + hl xk (t l ) + n(t )
l =0 k =1 l = 0

L 1

K 1 L 1

(6)

where n(t) is a white Gaussian noise with power spectral density N0. B. Receiver model We consider the classical Rake receiver to analyze the performance. Indeed, first HSUPA base stations supporting class 6 mobiles will have implemented a classical Rake receiver. It is clear, that this receiver cannot cope with the inter-code interference (ICI) introduced due to the loss in orthogonality encountered in multi-path propagation channels. In [5] they deduced analytically the SINR of the classical Rake receiver.

SNRc =

Ec N0

(11)

where Ec is the received total energy of all E-DPDCHs, EDPCCH and DPCCH per PN chip and per antenna. N0 is the total one-sided noise power spectral density due to all noise sources. This definition is conform to 3GPP specifications (cf.[3]). Real channel estimation is used by a Rake receiver. Realistic power control and well as HARQ Chase Combining are also considered. The average is performed on the different

3 multipath channel realizations using a fast link adaptation on the actual channel realization. III. EVALUATING FUTURE HSUPA MOBILES Running network simulations using a cell planning tool is the classical way to assess the performances of future mobiles. For example, this is what had been done in [4] for HSDPA mobiles. However, although system simulations are the only way to get an insight into the performances of future mobiles, it turns out that they have several drawbacks. Indeed, system simulations use digital terrain maps that are not always up-todate, and propagation calculations are not always accurate. All this can lead to unreliable results. On the other hand, drive test measurements are an accurate way to assess user-experienced throughputs. However, this method requires that terminals, or at least prototypes, are available. This is obviously not the case for future mobiles, such as HSUPA 5.76 Mbps mobiles, that are not available at all today. A method to plan HSDPA throughputs based on basic measurement samples from drive tests has been proposed in [8]. In this study, we reuse this idea to use drive test measurements samples, and we adapt it to HSUPA. A software tool has also been developed to run the calculations for each measurement sample, which are then displayed on a digital map using the MapInfo tool. This last point makes the method a real alternative to classical simulations with a cell planning tool, at least for coverage evaluation. IV. THROUGHPUT EVALUATION METHODOLOGY A. Available measurements Drive tests provide the downlink CPICH RSCP. We assume that path loss and shadowing are mainly due to the geographical situation. Therefore, we assume that reciprocally path loss and shadowing are approximately the same in the uplink, as uplink and downlink frequency bands are very close to each other. Of course this is not the case for the fast fading. But fast fading is modelled in the way that the average energy is constant, i.e. E{X} = 1, with X the energy of all paths of the channel model. Hence, no loss of (average) power is due to fast fading. This part shows that it is possible to get a good estimate of HSUPA throughputs with this single figure. B. Interference budget The first step of the calculations consists in calculating the interference budget at the base station receiver. The received power at the base station receiver Io is kept constant by the base station. This power is shared between the thermal noise power, the power received by R99 users and HSUPA users from the same cell (in-cell) and from other cells (out-of-cell). in out in cell out cell (12) I 0 = I R 99cell + I R 99 cell + I HSUPA + I HSUPA + N 0 We assume here that this user is alone in its serving cell.

Figure and R99 power from other cells 1) HSUPA2: Interference budget at the base station receiver We define the ratio of all interference from other cells (Ioc) to the interference from transmissions of the own cell (Ior) as

i=
It follows that:

I oc I or

(13)

in in cell I 0 = (1 + i )( I R 99cell + I HSUPA ) + N 0

(14)

Geometry calculations in [7] showed that i = 0,65 for trisectorial sites. 2) HSUPA max useful received power The received power from the HSUPA user is then :
in cell in out out cell I HSUPA = RoTmax + N 0 I R 99cell I R 99 cell I HSUPA in in cell = RoTmax + N 0 (1 + i ) I R 99cell i * I HSUPA

(15)

in RoTmax + N 0 (1 + i ) I R 99cell 1+ i

3) Link budget The link budget calculates the throughput as a function of path loss from the base station to the cell edge. Close to the serving site, the mobile power is limited by the maximum allowed RoT at the base station. On the other hand, when moving towards the cell edge, the mobile gets limited by its maximum transmission power, the received power gets low because the path loss gets high. 4) E-DCH Signal Interference, Noise Ratio (SNR) The interference perceived at the base station receiver has several components: Noise N0 Interference from neighboring cells (Ioc): R99+other Intra-cell interference (Ior) coming from R99 and other HSUPA users in the same cell Inter-code interference (ICI) due to orthogonality loss of the codes used by the considered HSUPA user. Since different long scrambling codes are used, Ioc and Ior can be approximated by a noise rise. Indeed, the interscrambling code correlation product is small and can be approximated as Gaussian noise perceived by the receiver. Inter-Code Interference (ICI) comes from the loss of orthogonality in the frequency selective channel. The exact amount of interference is very difficult to calculate analytically, especially for the low spreading factors of 2 and 4. We can represent this loss by an environment specific factor, . The total amount of ICI is then (1-)*IICImax where =1 means no loss and =0 complete loss of orthogonality.

Total noise rise=6dB (totalRoTmax)

C: Received power from in-cell HSUPA users Received power from out-of-cell HSUPA connections Received power from DCH connections (in-cell+out-of-cell) Total received power Io

Interference I

4 Note that depends on the base station receiver. Chip SINR can be defined as follows
E DCH E c,RX SINR E-DCH = 10 log c (1 - )I ICI,max + I or + I oc + N o

RSCP samples are between -110 dBm and -50 dBm, which means that the coverage is good in that area.
TABLE I CALCULATION PARAMETERS Parameter Multipath channel RoT (Rise over thermal noise) max Indoor margin Base station noise figure Base station antenna gain Mobile antenna gain Mobile maximum transmit power CPICH transmit power Mast head amplifier Neighboring cells load Setting Vehicular A 3km/h 6 dB 21 dB 2,5 dB 17 dBi 0 dBi 24 dBm 31 dBm Yes 25%

(16)

where

E DCH c, RX

is the received total energy of all dedicated

channels involved in E-DCH transmission (i.e. E-DPDCHs, EDPCCH and DPCCH) per PN chip per antenna. It is important to note that the ICI is taken into account in the link level simulations. 5) Throughput calculation Under the assumption that Ioc and Ior are Gaussian, we use look up tables from link level simulations with the SINR
SINR E-DCH c
E DCH E c,RX = 10 log I +I +N o or oc

(17)

to determine QoS and throughput The correspondence is done by identifying each computed SNR to the link level one and the optimal throughput provided by the link level results is hence selected. 6) Discussion As explained above, the evaluation method used here is as accurate as possible, since it is based on drive test measurements. However, some modeling estimates are still there: first, a multipath model (Vehicular A 3 km/h) is used. This model represents well urban cells, but can be a bit pessimistic for areas that are close to lineof-sight conditions, where it can overestimate intra-cell interference. Nevertheless, the evaluation method that has been described here turns out to be a very accurate way to get an estimate of the performance of future mobiles that will be used in the network that is available today. The use of this method is promising. Assessing the performances of future 3GPP systems, such as HSPA+ and LTE, can be done using this method. V. SIMULATION SCENARIOS AND RESULTS A. Scenarios and Drive test measurements Performances of 5.76 Mbps and 2 Mbps HSUPA terminals are studied using the software tool described in the previous paragraphs. The base stations in the drive test area are all equipped with two receive antennas and a mast-head amplifier. They all have Rake receivers. The 2 Mbps mobiles use up to 2 codes of spreading factor 2 and a BPSK modulation, the TTI length is 10 ms. The 5.76 Mbps mobiles use up to 4 spreading codes in parallel (2 codes of spreading factor 2 and 2 codes of spreading factor 4) and a BPSK modulation; the TTI length is 2ms. Their channel coding rate varies depending on the radio conditions. Simulations have been carried out in the centre of a major European town in which Orange operates a network. The drive test area is around 2 sq. kms. Most of the sites that have been rolled out in that area are tri-sectorial ones, but some are bisectorial ones. Cells are all macro ones. Simulation parameters are given in table 1. Around 4000 drive test samples are available. RSCP has been measured during the drive tests.

B. Simulation results and discussions The parameters that have been used in the link budget calculations are given in Table 1. The link budget calculations give the chart shown in Figure 3. One screenshot of the drive test is available for 2Mbps mobiles, and another one for 5,76 Mbps mobiles. They are shown in Figure 4 and 5. Drive test plots show that achievable throughputs change significantly along the drive test road:
HSUPA throughput vs CPICH RSCP
3000 2500

Throughput per user

2000 1500 1000 500 0 0 -10 -20 -30 -40 -50 -60 -70 -80 -90 -100 -110 -120 -130 2 Mbps mobiles 5 Mbps mobiles

Outdoor CPICH RSCP (dBm)

Figure 3: HSUPA indoor throughput vs. outdoor CPICH RSCP

HSUPA is definitely not a fair technology since users close to the radio site can expect high throughputs in average, while users at the cell edge cannot. In the 2Mbps HSUPA mobile case (Figure 4), more than half of the samples are above 1 Mbps. The number of samples below 1 Mbps corresponds to locations where the throughput is limited by the path loss. Most of the other ones correspond to locations where it is limited by the maximum RoT. The complementary cumulative density functions (CCDF) for 2 Mbps and 5.76 Mbps mobiles are shown in Figure 6. These functions give the percentage of area where this throughput (or a higher one) can be achieved. The comparison of the two charts shows that an overall improvement is achieved by 5.76 Mbps mobiles. However those it turns out that those mobiles cannot do better than 2.5 Mbps in the drive test area, which remains significantly below the expected 5.76 Mbps throughput. Figure 7 shows that the maximum achievable throughput increases by around 65% when using a 5.76 mobile instead of a 2 Mbps one.

5
Achievable throughput (kbps) Trisectorial radio sites

However, the average throughput on the drive test remains roughly the same. The minimum throughput is the same for both mobiles, which means that the 5.76 mobiles do not improve anything in poor coverage areas. It turns out that 5,76 mobiles give relatively poor results. The Number of reason is obviously the use of Rake receiver in the base samples stations. Indeed, the Rake receiver cannot handle inter-codes interference. E-TFC that provide throughputs close to 5,76 mobiles cannot be used. The use of advanced receivers (e.g. LMMSE) in the base stations is the only way to achieve good throughputs with 5,76 Mbps mobiles. Further work will be done to assess the gain of those receivers. VI. CONCLUSION

Drive test road 1,5 kilometer Figure 4: HSUPA indoor throughput for 2 Mbps mobiles Achievable throughput (kbps)

Realistic performance of HSUPA 5,76 Mbps mobiles have been shown in this paper. Throughputs that will be experienced by those mobiles will be in fact significantly lower. As expected, they depend on the user's location in the network. The name "5,76 Mbps mobiles", which is commonly referred to, is therefore misleading. Network operators are encouraged to pay a high attention to the availability of advanced receivers in the Node Bs provided by their vendors, since they should definitely provide better performances.

Minimum, average and maximum throughputs


3000 2500 2456 1927

Throughput (kbps)

2000 1500 1000 500 0

1292 min max average 0 Category 5

1491

0 Category 6

Figure 7: Throughputs comparison between the 2 mobile categories Figure 5: HSUPA indoor throughput for 5,76 Mbps mobiles
Throughputs Complementary cumulative density function

REFERENCES
[1] [2] [3] 3GPP TS 25.214 V5.6.0: Physical layer procedures (FDD).. 3GPP TS 25.213 V7.3.0: Spreading and modulation (FDD). 3GPP TS 25.104 V7.5.0: Base Station radio transmission and reception (FDD). [4] J.B. Landre, A. Saadani, Receive diversity and LMMSE equalization benefits for HSDPA: realistic network throughputs, in IEEE PIMRC, September 2007, Athens, Greece.. [5] G.E. Bottomley, et al. A generalized Rake Receiver for Interference Supression, IEEE journal on selected areas in communications, vol.18, No. 8, August 2000. [6] H. Holma, A. Toskala, HSDPA/HSUPA for UMTS, May 2006. [7] H. Holma, A. Toskala, WCDMA for UMTS, 2004. [8] J.B. Landre, A. Saadani, Hsdpa 14,4 Mbps mobiles-Realistic throughputs evaluation, in IEEE VTC Spring, 2008 [9] J. Wigard, S. Corneliussen, High speed uplink packet access evaluation by dynamic simulations, in IEEE PIMRC, 2006 [10] E. Rodrigues, Third generation WCDMA uplink capacity improvements with3.5G HSUPA for ITU Macro-Cell Channels, in 9th European Conference on Wireless Technology, 2006

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%
0 0 30

Area

Category 6 (5.76Mbps) Category 5 (2 Mbps)

0 60

0 0 0 0 0 0 0 0 90 120 150 180 210 240 270 300


Throughput (kbps)

Figure 6: Throughputs comparison between 2 Mbps and 5.76 Mbps mobiles

You might also like