Impact of Angular Spread on Higher Order Sectorization in WCDMA Systems Afif Osseiran, Andrew Logothetis Ericsson Research, SE-164 80 Stockholm, Sweden {Afif.Osseiran,Andrew.Logothetis}@ericsson.com

Abstract— This paper analyzes higher order sectorization (HOS) in WCDMA. Sectorization is achieved by splitting the sites into smaller sectors using highly directional antennas. The impact of angular spread on the system throughput is evaluated using a dynamic radio network simulator. Increasing the number of sectors per site from 3 to 6 and 12 in a typical urban radio channel, yields a downlink system capacity gain of 80% and 200% respectively. Simulations have shown that the increase in the ratio of users in soft and softer handover per sector is negligible and the impact of angular spread on the system performance of a 6 and 12 sector sites is minor and negligible for the 3 sector sites.

I. I NTRODUCTION A well known method that increases the capacity of a cellular system is sectorization. This is done by splitting the sites into smaller sectors. The site splitting is achieved using highly directional antennas, that provide higher antenna gains for the served users in the cell and ensure reduced interference to adjacent cells. When signals transverse a radio channel the signals become subject to spatial and temporal dispersions. The spatial distribution of the signal power is known as the Power Azimuth Spread (PAS). The standard deviation of the PAS is commonly referred to as the Angular Spread (AS). The degree of AS directly impacts the signal strength at the mobile and correlates the signal power from adjacent antennas. As the angular spread increases the effective antenna gain decreases. On the other hand, as the angular spread decreases the variation of the signal power from adjacent Base Station (BS) antennas becomes increasingly correlated. The purpose of this paper is twofold: firstly, to investigate the downlink (DL) system throughput gain going from a 3 sector to 6 or 12 sector sites in a WCDMA system, and secondly to analyze the system degradation due to spatial and temporal dispersions of the radio propagation channel. Very few studies have evaluated the impact of AS on higher order of sectorization. For instance, [1] and [2] assumed a simplified AS model and evaluated sectorization using a static simulator. Similarly, [3] and [4] used a static system simulation with no soft handover or power control and simplified angular spread modelling to evaluate sectorization. Recently [5] and [6], evaluated sectorization using a Uniform Linear Array (ULA) [5] in a dynamic system simulator but either did not investigate the impact of AS [5] or did not even considered it [6]. Finally, [7] reported only the gains that 12 sector sites offers in terms of SIR using ULAs or circular arrays in a static system simulator. In this paper, sectorization is evaluated in a dynamic system simulation taking into account the impact of AS in the system performance using an accurate intra- and

inter-cell interference modelling involving instantaneous SINR calculations. II. S YSTEM S ETUP The simulated area consists of a central site and two surrounding tiers of sites. The total number of sites is 19. Each site comprises of 3, 6 or 12 sectors (i.e. cells). Users are dynamically generated in the central site and the first tier (which consists of 6 sites). The second tier consists of 12 sites where no users are generated. Instead the Base Stations (BS) power of the second tier is time varying and is modelled as a random walk with upper and lower bounds determined by the 90th and 10th percentile of the BSs power located in the central site and first tier. An illustration of the simulated area is drawn in Fig. 1. Each star represents a site where the center of each star represents the site’s coordinates. The arrows indicate the orientation of the cells. In Fig. 1, sites with 6 sectors are assumed. The site-to-site distance is 3 km. Note that the simulation tool is similar to the one used in [8]. On each iteration of the main loop, the simulator time is increased by the duration of one frame and all radio network algorithms are executed, except for the power control which is executed on a slot level. The most relevant system parameters are summarized in Table I. System parameter Number of sites Site type: HOS Site-to-site distance [m] Max TX power [W] Channel model Number of RAKE fingers SF of the 64 kbps user

Value 19 3, 6, 12 sectors per site 3000 20 per sector COST259, Typical Urban 10 32

TABLE I S YSTEM PARAMETERS .

A. Propagation Environment The propagation model used is the COST 259 channel model [9], which is a spatial temporal radio propagation model that includes the effect of fast and slow fading. The COST 259 version used in the current system simulator yields an instantaneous Power Delay Profile (PDP) pm , the Rice factor κm , and the angular spread σm for the mth link in the system. The COST 259 can model several radio environments. Here we investigate the Typical Urban (TU) channel model which has a median AS of 8 degrees. In an urban environment

where Nm , Gm , Pm and N0 denote the spreading factor, the path gain, the transmitted power to the mth user, and the thermal noise respectively. Po is the total BS power allocated to signals using the same scrambling code as m. Im is the interference from the non-orthogonal signals originating from the own cell and other cells. Finally αm is the downlink orthogonality factor, which represents the fraction of the wide band received power of the orthogonal signals causing interference to user m. It can be shown (see [11]) that the orthogonality factor may be written as follows:

5000 4000 3000

Distance[m]

2000 1000 0 −1000 −2000 −3000 −4000 −5000 −6000

−4000

−2000

0 Distance[m]

2000

4000

6000

αm [k] =

nX F −1

l=−nH +1,l6=0

Fig. 1.

|rm,l [k]|2 /|rm,0 [k]|2

(4)

The simulated cell plan for 6 sector sites.

the power azimuth spectrum (PAS) is accurately described by a Laplacian pdf [10], that is:   √ |θ − θm | 1 f (θ|θm , σm ) = √ (1) exp − 2 σm 2σm where θm denotes the nominal direction to the mobile. Let a(θ) denote the spatial signature, i.e. the response of the BS antennas when a planar wave is impinging from an angle θ. Given the PAS the user dependent channel correlation matrix is given by Z ∞ R(θm , σm ) = a(θ)aH (θ)f (θ|θm , σm )dθ (2)

where {rm,l [k] : l = −nH + 1, . . . , nF − 1} is the impulse response of the combined effect of the transmit weights, radio channel and the receiver filter at time instance k. nH and nF are the channel length and number of receiver filter taps respectively.

D. Antenna Configuration: The antenna configuration employed in this study consisted of placing 3, 6 or 12 antennas equally spaced on a circle of radius r. The antenna patterns of the three, six and twelve sector antennas are shown in Fig. 2. The 3 dB beam-width of the 3, 6 and 12 sector beams are 63◦ , 35◦ and 20◦ , respectively.

−∞

The correlated channel impulse responses for the mth link can straightforwardly be derived from Eq. (2).

25 3 Sector Ant 6 Sector Ant 12 Sector Ant

20 15

B. Receiver Structure

C. Orthogonality factor As shown in [11], the SINR is a function of the orthogonality factor. The expected SINR for the mth user after despreading is generally modelled as follows SINRm =

Nm Gm Pm αm Gm Po + Im + No

(3)

10

Antenna Gain (dBi)

Each mobile is assumed to have a single receive antenna. Furthermore, perfect channel estimation is assumed in the terminals. The terminals employ a conventional Maximum Ratio Combining (MRC) receiver, i.e. a RAKE receiver with 10 fingers for the TU channel model. Power Control (PC) is also implemented and consists of inner and outer loop. The inner loop power control and the fast fading act on slot level. The inner loop PC assumes ideal Signal to Interference plus Noise Ratio (SINR) estimation (i.e. no measurement error is considered). After the slot loop, the instantaneous SINR are averaged and mapped to a BLock Error Probability (BLEP). Each block is then classified as erroneous or not, which gives the block error rate (BLER) estimates. The BLER estimates are used by the outer loop algorithm in order to decide if the SINR target should be increased or decreased.

5 0 −5 −10 −15 −20 −25 −30

Fig. 2.

−150

−100

−50

0 Angle (degrees)

50

100

150

Gains for 3, 6, and 12 sector antennas.

Note that the COST259 channel model yields a position dependent AS. In the TU case the 90th percentile of the AS is 19 degrees, thus more severe losses in the effective antenna gains are anticipated. As an example, Fig. 3 shows the effective antenna gains for a 12 sector site when AS is 8 degrees, which corresponds to the median AS in the TU case. It is clear that AS results in reducing the maximum effective antenna gain and widening of the main lobe.

Antenna Gain 90

0 deg. 8 deg.

25 120

60 20

12 sectors, respectively. Similar results for Q2 are obtained. The results are summarized in Table III. Obviously, a narrower beam is more susceptible to the effects of the angular spread.

15 150

30 Stream64k, PCPICH Tune=Off

10

65

5 60 180

0

330

210

240

300 270

Average user bitrate (kbps)

55

50

3Sec, AS0 6Sec, AS0 12Sec, AS0 3Sec, TU 6Sec, TU 12Sec, TU

45

40

35

Fig. 3.

Antenna gains of a 12 sector site with and without AS. 30

0

0.5

1

1.5 2 2.5 Normalized system throughput

3

3.5

4

E. Mobility & Traffic Models The mobile users are uniformly distributed in the cells. The average user speed is 3 km/h with small variations around the mean value. A poisson distribution time of arrival is assumed for the users. Furthermore, the user session time is exponentially distributed with mean holding time of 5s. The data are transmitted in a continuous stream (no TCP) using a 64kbps RAB (Radio Access Bearer) with retransmissions. F. Performance Measure The total system throughput is defined by the sum of correctly delivered bits to all users connected to the central site divided by the simulation period and the number of simulated cells in the central site. The user bit rate is given by the ratio of the total received bits over the length of the user’s session time. The Quality of Service (QoS) depends on the user bit rate. Two QoS are defined, Q1: when 90 percent of the users are guaranteed on the average 50kbps or better, and Q2: when the average bit rate of all users is greater than 50kbps. The system capacity is defined as the total system throughput when the QoS is met1 . III. R ESULTS

Fig. 4. Average user bit rate versus the relative system throughput in the TU and AS0 radio channels.

Sectors 3 6 12 3 6 12

Criterion Q1

Q2

Relative Gain 1.0 1.77 3.11 1.0 1.65 3.13

TABLE II S YSTEM THROUGHPUT IN TU.

Sectors

Channel

3

AS0 TU AS0 TU AS0 TU

6 12

Relative Loss Q1 Q2 1.00 1.00 0.98 0.99 1.00 1.00 0.90 0.95 1.00 1.00 0.84 0.91

TABLE III R ELATIVE S YSTEM THROUGHPUT IN AS0 AND TU.

A. Impact of Angular Spread The system performance is evaluated with and without angular spread modelling. Fig. 4 shows the average user bit rate versus the system throughput for 3, 6 and 12 sector sites. The system throughput results are normalized to the system throughput of the 3 sector TU case and is summarized in Table II. The performance when the standard deviation σm of the AS is set to zero, which is labelled as ”AS0” in Fig. 4, is also shown for 3, 6 and 12 sector sites. It is clear, based on Q1 that the AS induces a 2%, 10% and 16% system throughput loss in a TU channel compared to the AS0 case for 3, 6, and 1 Absolute numbers depend on many parameters. The focus should be on relative performance, instead.

B. Power consumption The cumulative density function (cdf) of the link and the total BS power for two Offered Traffic (OT) loads (OT = 5 and OT =10) are shown in Fig. 5. It can be seen that the link and BS power increased as the number of sectors per site increases. It has been observed (but not shown here) that the increase is mainly due to the inter-sector interference. In fact the interference increases more rapidly due to the substantial increase of the number of users per site when going from 3 to 6 and to 12 sectors.

Stream64k, TU, PCPICH Tune=Off 100 80 3Sec, OT=5 3Sec, OT=10 6Sec, OT=5 6Sec, OT=10 12Sec, OT=5 12Sec, OT=10

cdf

60 40 20 0

0

2

4

6

0

0.2

0.4

0.6

8 10 12 Total BS power (W)

14

16

18

20

1.4

1.6

1.8

2

100 80

cdf

60 40 20 0

Fig. 5.

0.8

1 1.2 Link power (W)

cdf of the link and the total BS power in the TU channel.

mobiles periodically report this measure. It is a common rule to allocate 10% of the BS power to the P-CPICH, but from analyzing Fig. 6, it can be seen that for low loads (OT=5), 90% of the users have a CIR around -15 dB and -14.5 dB for 3 and 6 sectors per site, respectively. Hence less power can be allocated to the P-CPICH without sacrificing the cell coverage. Since the P-CPICH quality is load dependent, for high loads and certain antenna configurations, the P-CPICH power needs to be increased. The proposed tuning method is compared to the case when the power of the P-CPICH is fixed at 33 dBm in Fig. 7. At high loads (see Fig 7), it is clear that tuning the P-CPICH power ensured that 90% of the users met their P-CPICH quality regardless of the number of sectors, whereas for the fixed PCPICH power case, slightly less than 85% of the users were able to meet the required P-CPICH quality for the 6 and 12 sector sites. Stream64k, TU, OT=15 100

C. Primary CPICH: Coverage and Tuning

3Sec, PCPICH Tune=On 6Sec, PCPICH Tune=On 12Sec, PCPICH Tune=On 3Sec, PCPICH Tune=Off 6Sec, PCPICH Tune=Off 12Sec, PCPICH Tune=Off

90

80

70

60 cdf

The cumulative density function of the CIR of the primary control physical channel (P-CPICH) for various offered traffic loads is shown in Fig. 6. A target of -16.5 dB is considered more than adequate to detect the cell and perform measurements on the P-CPICH [12].

50

40 Stream64k, TU, PCPICH Tune=Off 100

90

80

30

3Sec, OT=5 3Sec, OT=10 6Sec, OT=5 6Sec, OT=10 12Sec, OT=5 12Sec, OT=10

20

10 70

0 −24

cdf

60

−22

−20

−18

−16 −14 −12 CIR of the P−CPICH (dB)

−10

−8

−6

50

40

Fig. 7. cdf of the CIR of the P-CPICH with and without P-CPICH power tuning. High load (OT = 15).

30

20

10

0 −24

Fig. 6.

−22

−20

−18

−16 −14 −12 CIR of the P−CPICH (dB)

−10

−8

−6

cdf of the CIR of the P-CPICH for various OT in the TU channel.

From Fig. 6, it can be seen there are occasions (e.g. 12 sectors with OT=10) where more that 10% of the users fail to meet the P-CPICH target. If the number of users in the system further increases, then more users will not be able to decode the P-CPICH. Thus the CIR of the P-CPICH is load dependent. In order to compensate those users with unsatisfactory PCPICH quality, and to ensure a fair comparison between the 3, 6 and 12 sectors, an adaptive P-CPICH tuning algorithm was implemented. The algorithm proposed applies power control to the transmitted P-CPICH signal from all the BSs such that 90% of the users have their P-CPICH CIR greater than −16.5 dB. Feedback of the P-CPICH quality to the BSs is possible, since according to the WCDMA standard, the

The impact of the P-CPICH tuning on the system capacity is shown in Table IV. There is no system capacity gain for the 12 sector case when adaptively tuning the P-CPICH power. However, we must point out that, it is necessary to adapt the P-CPICH power (especially in high traffic loads) in order to ensure the desired P-CPICH quality for all the antenna systems investigated here. In doing so, the relative system capacity gain going from 3 sectors to 6 and 12 sectors is 1.65 and 2.74, respectively. Sectors

Relative Gain

3 6 12

12% 6% 0% TABLE IV

R ELATIVE S YSTEM THROUGHPUT GAIN WHEN P-CPICH IS TUNED ON VERSUS TUNED OFF .

D. Handover ratio The expected number of soft and softer links per sector for the 3, 6 and 12 sector sites are shown in Table V. Approximately the soft/softer handover overhead per sector is between 30% to 40% in the non P-CPICH tuning case, and 40% to 50% in the P-CPICH tuning case. This implies that on the average we expect up to 1 and a half links are utilized per user. It is interesting to notice that the overhead is almost traffic independent. In fact the overhead is tightly connected to the log-normal fading and the antenna beam-width (which impacts the overlapping region between the sectors, see Fig. 3 for the 12 sector site case). Sectors 3 6 12

Soft Tuning off Tuning on 0.28 0.35 0.29 0.36 0.29 0.36

Softer Tuning off Tuning on 0.04 0.04 0.05 0.06 0.11 0.13

an accurate intra- and inter-cell interference modelling. The relative gain of 6 and 12 sector sites compared to a 3 sectors is approximately 1.8 and 3 respectively in a Typical Urban radio channel. The system degradation due to spatial dispersion of the channel is minor in terms of system throughput for 3, 6 and 12 sectors. Finally some interesting observations can be mentioned: • The ratio of users in soft and softer handover per cell increased slightly which implies that the handover signalling over the Iub interface slightly more than doubled or quadrupled when going from 3 to 6 or from 3 to 12 sectors, respectively. • Besides ensuring a good quality for 90% of the users, the tuning of the P-CPICH power, offered a 12% capacity gain for 3 sector sites but little or no gain for 6 and 12 sector sites.

TABLE V

R EFERENCES

E XPECTED NUMBER OF SOFT AND SOFTER HANDOVERS PER USER .

[1] T. Baumgartner, T. Neubauer, and E. Bonek, “Performance of Downlink Beam Switching for UMTS FDD in the Presence of Angular Spread,” in IEEE International Conference on Communications. NY, USA: IEEE, May 2002. [2] T. Baumgartner, “Smart Antenna Strategies for the UMTS FDD Downlink,” Ph.D. dissertation, Technische Universitat Wien, Austria, Aug. 2003. [3] A. Wacker, J. Laiho-Steffens, K. Sipilae, and K. Heiska, “The impact of the base station sectorisation on WCDMA radio network performance,” in Proceedings IEEE Vehicular Technology Conference, Fall, vol. 5, no. 50, Amsterdam, The Netherlands, Sep. 1999, pp. 2611–2615. [4] M. Schacht, A. Dekorsy, and P. Jung, “System Capacity from UMTS Smart Antenna Concepts,” in Proceedings IEEE Vehicular Technology Conference, Fall. Orlando, USA: IEEE, October 2003. [5] K. I. Pedersen, P. E. Mogensen, and J. R. Moreno, “Application and Performance of Downlink Beamforming Techniques in UMTS,” IEEE Communications Magazine, pp. 134–143, October 2003. [6] H. Zhu, T. Buot, N. Rin, and H. Schreuder, “Performance Study of Space-Time Transmit Diversity and Sectorisation in WCDMA Radio Networks,” in Proceedings IEEE Vehicular Technology Conference, Spring, Milan, Italy, May 2004. [7] A. Czylwik and A. Dekorsy, “System Level Simulations for Downlink Beamforming with Different Array Topologies,” in Global Conference on Communications. San Antonio, USA: IEEE, November 2001. [8] A. Osseiran et al., “Downlink Capacity Comparison between Different Smart Antenna Concepts in a Mixed Service WCDMA System,” in Proceedings IEEE Vehicular Technology Conference, Fall, vol. 3, Atlantic City, USA, 2001, pp. 1528–1532. [9] L. Correia, Ed., Wireless Flexible Personalized Communications - COST 259 Final Report. John Wiley & Sons, 2001. [10] K. Pedersen, P. Mogensen, and B. Fleury, “A stochastic model of the temporal and azimuthal dispersion seen at the base station in outdoor propagation environments,” IEEE Transactions on Vehicular Technology, vol. 49, no. 2, pp. 437–447, March 2000. [11] A. Logothetis and A. Osseiran, “SINR Estimation and Orthogonality Factor Calculation of DS-CDMA Signals in MIMO Channels Employing Linear Transceiver Filters,” Wiley, Journal of Wireless Comunication and Mobile Computing, 2005, to appear. [12] Y. Sun, F. Gunnarsson, and K. Hiltunen, “CPICH Power Settings in Irregular WCDMA Macro Cellular Networks,” in IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Sept. 2003. [13] P. Zetterberg, “Performance of Three, Six, Nine and Twelve Sector Sites in CDMA - Based on Measurements,” in IEEE International Symposium on Spread Spectrum Techniques and Applications, Aug. 2004. [14] J.-H. Yea, “Smart antennas for multiple sectorization in CDMA cell sites,” in RF Design magazine, http://rfdesign.com/, 2001, pp. 28–38.

It is worth mentioning that although the soft handover ratio remains similar for the 3, 6 and 12 sector sites, the signalling overhead over the Iub interface between the BS and the RNC will more than double (when going from 3 to 6 sectors) or quadruple (when going from 3 to 12 sectors). The expected number of soft/softer links is also shown in Table V when the P-CPICH power is adaptively tuned to the radio conditions. A relatively small increase is observed. This is not surprising since the quality of the pilot channels is adjusted thus increasing the probability that more BSs will be included in the active set of the mobiles. E. Comparison to Prior Work on System Level Results Although some assumptions of the system model (e.g. antenna patterns and configuration, channel models, mobile velocity, angular spread) may have been simplified in other published work, the system capacity gain of 6 sector over 3 sector sites in WCDMA obtained here is in line with other work. A system capacity gain of 1.6 to 1.7 were shown in [3], [4] and [6]. Further [1] and [5] obtained a lower system capacity gain (1.4 and 1.5 respectively) due to the total site power limitation of 60W. A mere 20% gain was found in [13] where the sector-to-sector isolation was rather poor. A live field trial in a CDMA system using 6 sectors was conduced by [14] and led to a gain of 1.73 compared to a 3 sector site. Higher order sectorization was implemented using 3 ULAs per site, each comprising of two antenna elements that formed 2 sectors. In [7], it was shown that 12 sector sites offer 3dB gains in terms of SIR when implementing sectorization. Unfortunately, it is not clear how much of the reported SIR gain translates into system capacity gain. Finally in [5], a gain of 1.8 was obtained using 18 sector sites, but the total BS power was limited to 60W. IV. C ONCLUSIONS The system throughput gain going from 3 sector to 6 or 12 sector sites is evaluated in a dynamic system simulation with