Smart cell sectorization for third generation CDMA systems

Transcription

1 WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wirel. Commun. Mob. Comput. 2002; 2: (DOI: /wcm.56) Smart cell sectorization for third generation CDMA systems Romeo Giuliano, Franco Mazzenga*, and Francesco Vatalaro Dipartimento di Ingegneria Elettronica Università diromatorvergata Via del Politecnico Roma, Italy Summary Smart antennas are a powerful means to provide the increased bandwidth efficiency needed to deliver advanced third generation (3G) interactive multimedia services anywhere and anytime. In this paper we focus on the development of smart antennas located at the base stations of a code division multiple access (CDMA) cellular system. First, we review the main characteristics of antenna array systems and highlight some issues related to the inclusion of smart antennas in a 3G CDMA system. A classification of the proposed beamforming techniques and a discussion on the advantages and drawbacks of smart antennas are provided. Then, we introduce an original smart antenna technique based on rotation and resizing of the available sectors much simpler than common smart antennas based on beamforming techniques addressing a specific user. We discuss on the impact of the proposed technique in the base station architecture. Finally, we provide simulation results in order to show the effectiveness of the proposed approach and to relate transmission performance to system parameters. It turns out that the smart sectorization technique provides significant capacity improvements under statistically uniform user distribution (up to about 80 per cent for data traffic). Copyright 2002 John Wiley & Sons, Ltd. KEY WORDS cellular systems smart antennas interference mitigation techniques base station architecture Ł Correspondence to: Franco Mazzenga, Dipartimento di Ingegneria Elettronica, Università di Roma Tor Vergata, Via del Politecnico 1, Roma, Italy. mazzenga@ing.uniroma2.it Copyright 2002 John Wiley & Sons, Ltd.

2 254 R. GIULIANO, F. MAZZENGA AND F. VATALARO 1. Introduction Today, one of the main goals of the telecommunications community is to provide customers with a large variety of interactive multimedia services, anywhere and anytime. To this end the convergence of the Internet and of mobile communications is needed, and this asks for a significantly increased bandwidth, and/or spectrum efficiency in the radio access network. A very large spectrum efficiency cannot be provided with the present generation radio interfaces, such as the GSM General Packet Radio Service (GPRS). Therefore, Third Generation (3G) radio interfaces, such as the Universal Mobile Telecommunications System (UMTS), are being introduced to provide high-speed wireless access to the Internet and to the forthcoming mobile multimedia contents. All radio access systems of the future will not only include cellular systems but also satellite systems, Wireless Local Area Networks (WLANs), radio local loop systems, and very short-range ad-hoc networks. As the traffic grows, these systems must better manage frequency resources to increasingly improve spectrum efficiency. Therefore, several physical layer techniques are the subject of continuous investigation, such as: 1. multiple access techniques in the radio interface, to increase system capacity; 2. efficient compression techniques, to reduce the user bit rate without penalizing the perceived QoS; 3. improved modulation schemes and resource management strategies, to optimize existing radio access techniques; 4. advanced detection schemes and interference mitigation techniques, to increase transmission quality. Among the latter techniques, smart antennas are a powerful means to enhance spectrum efficiency, through the combined use of improved signal detection and interference mitigation. As their applicability is not limited to cellular systems, they can be basic components for the development of the global integrated access network of the future [1]. In this paper we focus on the development of smart antenna arrays located at the base station (BS) of a CDMA (Code Division Multiple Access) cellular system, but many of the concepts discussed in this paper can be also applied to other wireless systems. It is widely recognized [2] that smart antenna systems located at every BS of a cellular system can increase the capacity of CDMA networks well beyond the results obtained with a simple cell sectorization [3]. In general, a smart antenna system is composed of two main parts: an antenna array and a Beam- Forming Network (BFN). A control unit supervises the behavior of the BFN, and control actions can take place on RF signals, IF signals or baseband signals. Ideally, a smart antenna should be able to synthesize very narrow beams and to individually steer each beam towards the direction of arrival of a desired signal. In this case, the beneficial effects are twofold. A first effect consists in the reduction of the interfering power experienced in each link, while a second effect can be the increase in received power in the downlink (or a better sensitivity in the uplink). Thus, the use of a smart antenna leads to a larger spectrum efficiency and/or coverage efficiency. Furthermore, a smart antenna can assist the user localization function, which is a basic requirement for 3G systems. Beamforming algorithms are user-oriented algorithms and, in general, their computational complexity increases with the number of users and can become prohibitive. In addition, due to multipath effects, a beamforming algorithm can steer the beam towards erroneous directions. On the other hand, fixed cell sectorization provides worse radio link performance, but it is robust and much simpler to implement than beamforming. Sectorization is currently used to increase system capacity for second generation cellular systems, and will be included in the UMTS too. In this paper we present a technique to render sectorization smart. We show that a significant outage probability reduction can be obtained both in the uplink and in the downlink by adaptively rotating and/or resizing the sectors. These operations are used in the uplink to equalize the received carrier-tointerference power ratio, C/I, in each sector, and in the downlink to equalize the total transmitted power in each sector. The proposed smart sectorization procedure is analyzed for both the uplink and the downlink transmissions. In evaluating performance, both voice traffic and multi-code CDMA data traffic are considered. Since knowledge is not required of link gains and power levels transmitted by the users in the served area, the technique can be autonomously implemented at any BS (i.e. no coordination among base stations is needed). The application of the proposed algorithm requires C/I measurements at each sector for the uplink and measurements of the total transmitted power at each sector for the downlink. These characteristics render the proposed technique independent of the modulation formats, so that it is

3 SMART ANTENNAS FOR 3G CDMA SYSTEMS 255 applicable to both existing and future cellular systems. The results in this paper were presented in part in Reference [4]. The paper is organized as follows. In Section 2 we review the main characteristics of antenna array systems. In Section 3 we discuss some of the main problems related to the development and implementation of smart antennas in modern CDMA systems. In Section 4 we illustrate the smart sectorization technique for both the uplink and the downlink. In Section 5 we consider the impact of the proposed technique on the architecture of the BS. To show the effectiveness of the proposed approach, and to outline the dependence of the performance on the system parameters simulation results are provided in Section 6. Finally, in Section 7 conclusions are drawn. 2. Overview of Antenna Array Systems The conceptual scheme of a smart antenna system is illustrated in Figure 1. We can identify two main elements: the array, which includes the radiating elements, and the BFN with the associated control subsystem. Frequently, but not necessarily, the antenna elements are identical and have an omnidirectional pattern in the azimuth plane. The geometry of the antenna elements can vary, but it is common to place them on a circle (circular array) or along a straight line (linear array). The BFN parameters are adjusted by the control unit in order to electronically shape the antenna beam and/or to point it along a desired direction. Therefore, a smart antenna is not just an antenna, rather it is a complex transceiver system. Smart antennas are usually physically located at the BS only in order to improve performance both on uplink and downlink. The adoption of a smart antenna in place of a more traditional omnidirectional or fixed sector antenna allows to adapt the antenna pattern to the current radio link and traffic conditions. This provides a link budget improvement. In general, an antenna array containing N elements can provide an N-fold mean power gain in an additive white noise channel, but successful interference suppression depends on the BFN control strategy. Interferers infrequently share the geographical location with the intended user. Therefore, interference is reduced and the overall link quality is enhanced by maximizing the antenna gain in the direction of the desired user and by simultaneously locating radiation function nulls towards the most dangerous interferers. The algorithms used to perform antenna pattern adaptation differ according to the transmission direction (uplink or downlink) and to the modulation format(s) adopted. In general, control algorithms can operate along the spatial (i.e. angular) dimension or both in space and in time [1]. Spatial beamforming alone is effective only when the delay spread is small with respect to the symbol (or chip) time duration. Otherwise equalization is necessary in the time domain. In this case the control unit in the processor in Figure 1 has a hybrid architecture consisting of two stages performing a decoupled spatial and temporal processing. Several space time RAKE receivers (commonly indicated as 2D-RAKE systems) have been proposed for FDD (Frequency Division Duplex) CDMA systems [5]. Joint space time processing can also be used to perform multiuser detection in CDMA systems in order to reduce the intra-cell interference. Some schemes have been proposed also for a TDD (Time Division Duplex) UMTS system to eliminate intra-cell interference by means of space time joint detection [6]. In the following we provide a classification of the main beamforming algorithms presented in the literature. Fig. 1. Conceptual scheme of a smart antenna system.

4 256 R. GIULIANO, F. MAZZENGA AND F. VATALARO 2.1. Beam Pointing Techniques Switched beams It is the simplest technique and it is based on switching between narrow beams obtained from a set of separate and directive antenna elements. When an antenna array with identical and omnidirectional elements is used, fixed beams can also be easily generated by changing the BFN parameters. Cell areas benefiting from the use of a switched beam technique are commonly affected by large propagation losses and/or require high capacity demands due to frequent hot spot conditions, such as in shopping malls, business centers and railway stations and airports. In these cases, due to the presence of the oriented spot beam the required transmitted power is reduced both at the BS and at the mobile terminals, thus resulting in an increased capacity and/or coverage. Mobile terminals can change spot beam during their motion. In this case soft-handoff procedures for terminals located in different spot beams are necessary. Some techniques have been proposed for CdmaOne (IS-95) and for Cdma2000 [7] Dynamically phased array It is based on Direction of Arrival (DoA) algorithms. DoA estimation is used to continuously track the direction of the signal transmitted by the user. The BS needs to dynamically change the direction of the antenna pattern as the terminal moves in the cell area Nulls Positioning in the Antenna Pattern Spatial filtering Many spatial filtering techniques and algorithms have been proposed in the literature. ž DoA based algorithms: in the simple case DoA algorithms are adopted in order to estimate the direction of interferers. The antenna radiation pattern is then adjusted to null-out the interfering signals both in the uplink and in the downlink. In the radiation pattern of a linear array antenna with n e elements, up to n e 2 nulls can be inserted [8]. Therefore, if the number of cochannel users is lower than n e 1 perfect interference cancellation can be ideally obtained. ž Spatial processing algorithms are more general than DoA-based techniques. The radiation pattern is shaped to optimally collect the multipath signals associated with the intended user. Among the optimization criteria to obtain beamforming, the most important is based on the maximization of the signal-to-interference-plus-noise ratio at the receiver output Space division multiple access (SDMA) In this case DoA-based spatial filtering is used to allow more users in the area to share the same communication resource (carrier frequency and/or time slot). In this case the beam pointing toward each terminal contains nulls in the direction(s) of the user(s) transmitting over the same communication resource. 3. Development of Smart Antennas In general, smart antenna systems can be seen as an extension of the commonly used diversity schemes [9], having more than two diversity branches and more complex signal combination algorithms. Their applicability both in the uplink and in the downlink is strictly related to the characteristics of the communication system (modulation format, multiple access technique etc.) and on the considered communication direction (uplink or downlink). The applicability of smart antennas in the downlink is limited by cost, complexity, and weight limitations for the terminals, particularly hand-held ones. Therefore, knowledge of the spatial channel response is not available in downlink, and the terminal cannot adopt any spatial filtering technique to (optimally) extract the desired signal. However, this limitation can be overcome in TDD systems where the optimum antenna parameters calculated in the uplink can be reused by the BS for the downlink, provided that the channel does not significantly change during the switching time between uplink and downlink use. However, this cannot be applied to FDD systems where uplink and downlink frequency separation is usually larger than the channel coherence bandwidth. In this case to steer the downlink beam towards the intended terminal it is necessary to adopt a geometrical approach based on the DoA estimation on uplink and assuming directional reciprocity (i.e. that the direction of arrival for an uplink signal is the same as for the downlink). The suitability of the directional reciprocity assumption has been validated by recent results [10].

5 SMART ANTENNAS FOR 3G CDMA SYSTEMS 257 In the case of a CDMA system other specific differences between uplink and downlink can be evidenced. One refers to the availability of the pilot channels on the two communication directions. In the IS-95 system usually a common pilot channel is broadcasted from the BS in the sectors in order to provide cell identification, phase reference, timing information, etc., to terminals in the area. When a sector is subdivided in multiple narrower spot beams (switched beams) or smart antennas are deployed, a common pilot channel cannot be used because the reference pilot signal used for channel estimation in the terminal must go through the same path (including antennas) of the data. Consequently each antenna beam requires a separate auxiliary pilot. One possible procedure to generate auxiliary pilots has been presented in Reference [7] for Cdma2000. It is based on suitable temporal concatenation of Walsh and conjugate Walsh sequences. In the UMTS WCDMA and TD-CDMA systems each user possesses a dedicated pilot sequence for both uplink and downlink [11] and this renders easier the future deployment of smart antennas. According to References [9] and [12], the use of smart antennas will evolve according to three distinct phases. In the first phase smart antennas will be used for uplink transmission only. In this case the uplink gain is increased and the co-channel interference is suppressed. The increase of uplink gain improves the receiver sensitivity leading to: an improvement in the system capacity, and/or to a reduction of the transmit power of mobile stations, and/or to an enlargement in the cell coverage. On downlink the system capacity is not increased and the extension of cell coverage is only possible by balancing uplink and downlink coverage increasing the downlink transmitted power for each user. In the second phase smart antennas will also be deployed in the downlink direction. In this case the link gain is increased for both uplink and downlink. Link enhancement will be achieved using spatial filtering in both directions and using sophisticated algorithms. In this system, the frequency reuse distance can be shortened and this leads to a sensible increase in the system capacity. The last development phase will lead to the deployment SDMA concept where angularly separated users will use the same physical communication resource (time slot and/or frequency carrier) simultaneously. However, the SDMA technique seems to be inappropriate for CDMA systems since it involves the reuse of spreading/scrambling codes that is not desiderate and practically unnecessary. In CDMA systems a C/I improvement leads directly to a proportional increment of capacity. For this reason smart antennas improving simultaneously both uplink and downlink performance are interesting for CDMA systems Benefits of Smart Antenna Systems The introduction of smart antennas in the access network has a deep impact on system performance. The main improvements and benefits are now listed. ž Capacity increase: smart antennas improve the link budgets thus allowing a significant increase in system capacity, especially for interference limited systems, such as CDMA systems. Experimental results report up to 10 db improvement in the signal-to-interference ratio [13]. This can be useful in densely populated areas, such as urban areas. Since CDMA systems such as the IS-95 and the UMTS are interference limited systems, the expected capacity gain for these systems is even larger than for other narrow-band systems (i.e. FDMA and TDMA). ž Cell range increase: the design criterion for cellular networks in rural areas, and in general for sparsely populated areas, is coverage rather than capacity. Due to the larger gain of smart antennas, the additional link margin can be used to increase the cell size, if the power transmitted by the terminal remains unchanged. ž Localization services: smart antennas allow each BS to accurately locate the user. The accuracy of the position estimate is improved with respect to that achievable with the current techniques. Position location services are used in emergency situations and are becoming an important requirement for cellular systems. ž Security: to violate security in a cellular system using smart antennas, the intruder must be positioned along the same direction of the intended user as seen by its BS. This event can take place with lower probability with respect to the present systems that use fixed sectorization or omnidirectional antennas. ž Reduced multipath propagation: using narrow beams at the BS, multipath can be reduced. The actual reduction depends on the scenario and in general it may not be very significant. However, even if RAKE receivers can cope with multipath, this may not be the case for fast mobile terminals.

6 258 R. GIULIANO, F. MAZZENGA AND F. VATALARO In this case narrow beams obtained with smart antennas allow to reduce multipath effects, and this can help in relaxing the terminal specifications Costs, Critical Factors and Challenges The benefits of a smart antenna system are many, but there are also disadvantages and technical challenges to be solved. A smart antenna transceiver is much more complex than a traditional BS transceiver, since a separate transceiver chain for each array antenna element and accurate calibration are needed. In addition, when adaptive arrays are used, the algorithms to compute the BFN parameters can be rather complex. This requires powerful processors and fast control systems located inside the BS. Smart antennas put problems to the network resource management subsystem and to the mobility management subsystem. As an example, we observe that when a new connection with another BS needs to be established, or the existing connection need to be handed-over, no angular information is available to the new BS and some means to find out the mobile are necessary. This could be obtained by using one beam in the BS to periodically look for new connection candidates, or for a handover request. Another possibility is to use an external system, such as the Global Positioning System (GPS) to locate the users in the area. When SDMA is used in a mobile environment, the probability of angular collision between terminals using the same resources is not null. In this case it is necessary in general to quickly switch one communication on the same channels to another channel. This implies that in an SDMA system the frequency of intra-cell handover can be higher than in a conventional CDMA system. To obtain a reasonable link gain, a smart antenna needs to be composed of several antenna elements. Typically, antennas with 8 10 horizontally separated elements have been considered with an element spacing of 0.5. Therefore, an 8-element antenna will be large: approximately 1.2 m at 900 MHz, and 60 cm at 2 GHz. The separation of antenna elements is another important aspect in smart antennas. It has been observed that, when mounting several elements together (typically less than 0.5 apart), they influence each other. This coupling introduces changes both to the beam-patterns of the single element and to the impedance match, thus degrading performance. Furthermore, in selecting the spacing among elements it is necessary to avoid spatial aliasing problems that can occur when spacing is selected to be much larger than 0.5. Several other technical challenges are yet to be solved for smart antennas implementation. Beamforming in downlink is more difficult than uplink. In fact, DoA-dependent algorithms render downlink performance variable with the rate of change of the radio channel. If the channel suffers large angular spread, i.e. the signal arrives from many directions, it may be difficult to estimate DoA. Another critical aspect is the linearity in transceiver chain. The transfer functions of the up and down converters need to be exactly known for the beamforming to be accurate. This means that smart antenna need frequent on-line calibration of the radio units. The complexity of smart antennas is a great challenge since advanced applications involve simultaneously maximizing the useful signal and nulling out the interferers. In addition, if the BS needs to make beamforming for each user in the area this can be a very challenging task even for the powerful processing units available today. Beamforming algorithms can be trained using information contained in the signals transmitted by each user or they can be blind when the user does not provide any pilot signal. In the first case, the transmission of a known sequence reduces the spectrum efficiency, especially when too frequent updates of the beam parameters are needed to track the selected user(s). In the second case blind algorithms preserve spectrum efficiency but may need long convergence time that may render them unusable when fast tracking is required. Finally, due to multipath effects beamforming algorithms can steer the beam to erroneous directions and blind beamforming algorithms may become unusable and/or can give incorrect results in the presence of multipath. Fixed cell sectorization is not intelligent but it is much simpler to be implemented with respect to beamforming. Fixed sectorization can be seen as a simple case of the beamforming techniques where the BFN parameters are precalculated and stored. Therefore, sectorization doesn t suffer some of the many drawbacks previously described for smart antennas even if performance is inferior. In the following sections, we consider a technique to make sectorization smart, i.e. we present an adaptive sectorization strategy able to adapt to traffic changes. We show that a significant reduction in outage probability for uplink and downlink can be obtained by adaptively rotating and/or resizing the sectors. The proposed technique can be seen as a practical short-term solution toward the full use of

7 SMART ANTENNAS FOR 3G CDMA SYSTEMS 259 smart antennas and can be successfully applied to both second and third generation CDMA communication systems. 4. Smart Sectorization In the following we introduce two algorithms to make cell sectorization adaptive in the uplink and in the downlink, respectively. In both cases adaptation is achieved using rotation and/or resizing of sectors, as illustrated in Figure Uplink Uplink cell capacity is maximized when the ratio of the received carrier power to the total disturbance (interference plus thermal noise) power, C/I t, in each sector is equalized [14]. To illustrate the proposed algorithm we consider a single cell with n s sectors, serving n i u users in each sector, i D 1, 2,.., n s. Assuming perfect uplink power control (i.e., each user s signal received at the BS with fixed power), for each user in the i-th sector the received C/I t is given by: ( ) C i D 1 ( ) Eb I t G c I 0t where: UL P s D 1 G c C n i u 1 P s C P ext i D 1, 2,...,n s, 1 ž G c D B/R is the coding gain (ratio of the bandwidth, B, and the user bit-rate, R); ž E b is the bit energy; ž I 0t is the one-sided power spectral density of the total disturbance; ž is the voice (or data) activity factor; ž P s is the received power; ž is the thermal noise power; ž and, finally, P ext is the power of the external (i.e. other-cell) interference, given by: n ext P ext D n P sˇn, 2 nd1 where n is a binary random value indicating whether the nth user is active or not; n ext is the number of other cell users interfering within the intended sector; ˇn is the propagation coefficient accounting for the interference power in the reference BS due to another user served by a different BS; ˇn is given by: ( ) ) rn ( n0 ˇn D, 3 r n0 where r n and r n0 are the distance of the interfering user from its BS and from the reference BS, respectively; n, n0 are two log-normal random variables accounting for the shadowing experienced by the interfering user with respect to its serving BS and to the reference BS, respectively [3,15]. The parameter is the path loss exponent that in an urban environment typically assumes values between 3.5 and 5.5. We can often consider a nominal uniformly distributed traffic, particularly in urban areas. However, instantaneous traffic situations can be far from nominal conditions. Some of the reasons include: locally congested areas (hot spots), multimedia traffic, lowmedium user density, etc. Therefore, in each sector i, can be significantly different. As a consequence, one (or more) cell sector(s) can be in outage, even if the other sectors can be far from this condition (see Figure 3). Outage is experienced when in C/I t i UL n Fig. 2. Sectors rotation and sectors resizing three sectors.

8 260 R. GIULIANO, F. MAZZENGA AND F. VATALARO Fig. 3. Optimal sector rotation in the presence of hot spots located in one fixed sector. one or more sectors the bit error probability increases above a given threshold. Due to the monotonic relationship between bit error probability and C/I t,an outage event occurs when C/I t i < C/I t d where C/I t d is the threshold value necessary to meet a specified quality of service. When the system is in outage, no more users can be admitted in the system. In this case to decongest the critical sectors, balancing the C/I t is helpful. This can be conveniently achieved using smart sectorization as shown in the following. In the case of Figure 4 we consider a hot spot with a large extended area. The joint use of rotation and resizing leads to a more favorable distribution of the users as indicated in Figure 4(c). In the following, we assume that C/I t balancing is obtained through an adaptive rearrangement of the angular position and/or the resizing of the sectors for each cell according to a min max criterion. We assume that the BS continuously monitors the received C/I t for every user served in the cell. It should be observed that when perfect power control is assumed the measured C/I t is virtually independent of the intended user. However in a practical implementation the measured C/I t may differ among users in the same sector. In this case we can define the reference C/I t i UL in the i-th sector as the average C/I t over the users in the same sector. The proposed adaptive sectorization algorithm can be divided into two adaptation steps: the coarse sector position adjustment and the fine adjustment. To achieve coarse C/I t equalization among sectors, the rotation angle (see Figure 2) is evaluated according to the min max criterion, i.e. we select Ł such that: ( C I t ) ( Ł ) D max 2[0,2/n s { min id1,..,n s ( C I t ) i UL }. 4 Due to the periodicity of the sectorization scheme, maximization is restricted to the angular interval [0, 2/n s. In the ideal case, after min max optimization, the values of C/I in each sector should be similar (ideally identical). In practice, maximization operation is performed varying according to fixed angular steps of 1 degrees. In general it can be difficult or impossible to reach the ideal C/I t balancing condition. However, after rotation we can assume to be close to the optimal position. Now we can perform a fine adjustment of the measured C/I t in each sector using resizing operation. To this aim, the following and simple iterative method can be used. For simplicity we consider the case of n s D 3, but the procedure can be easily extended to any values of n s. From sector rotation we can reach the situation: ( C I t ) 1 UL ( C < I t ) 2 UL ( C < I t ) 3. 5 UL In this case we resize the sector with minimum C/I t until C/I t 1 UL tends to C/I t 3 UL. Now we consider C/I t 2 UL and repeat the resizing operation until we reach convergence. Rotation and resizing can be used independently. Sector rotation can be preferred since it is simpler than resizing, and allows to rapidly reach an (approximate) equilibrium condition. In most cases from simulation we observed that the C/I t fine adjustment is superfluous. Fig. 4. Joint use of sectors rotation and resizing in the case of an extended hot spot.

9 SMART ANTENNAS FOR 3G CDMA SYSTEMS Downlink The downlink scenario is more complex to analyze than uplink and it is not straightforward to identify possible optimization criteria applicable to the rotating sectors. Therefore, we base our reasoning on the following considerations. In the downlink case the C/I t measured by the ith mobile terminal in one sector of the reference cell area can be written as: ( C I t ) i DL D i P i T C Psc i C Poc i, C P Pilot 6 where P i T is the power transmitted for the ith user and i D i/r i is the propagation coefficient accounting for losses due to distance and shadowing, P sc is the power of the intra-cell interference, is the orthogonality factor usually taking values between 0.4 and 0.9, P oc is the outer-cell interference and P Pilot is the interfering power due to the intra-cell and outercell pilots. For simplicity, in the following we neglect the effects of P Pilot. Expressions for P sc and P oc are: P i sc D i n u nd1,n6di n P n T, nc P i oc D md1 i m P m, 7 where P n T is the power transmitted to the nth user in the reference cell serving n u users; P m is the total power transmitted by the n c sectors of the neighboring cells as seen by the reference sector in the reference cell area; m i are the corresponding propagation coefficients. From Equation (6) we observe that to improve the downlink capacity it is necessary to reduce interference effects due to both P sc and P oc. To this aim the proposed adaptive sectorization technique can be helpful. In fact, in the case of a large orthogonal factor, equalizing the power transmitted in each cell-sector first allows to reduce the effects of P sc. In addition when one or more sectors of neighboring interfering cells are crowded (such as in the case of nonuniform distribution of users in the areas) the P oc term becomes important for users located on the border of the reference cell in the proximity of the heavy loaded sectors of the interfering cells. In this case the equalization of the downlink transmitted power in each sector of every BS in the area is helpful. 5. Base Station Architecture The conceptual scheme of a BS using the proposed adaptive sectorization algorithm in the uplink direction is illustrated in Figure 5 in the case of n s D 3 sectors. The sector rotation function and/or the sector resizing function are obtained by controlling the BFN in Figure 5. The task of the BFN is to form the sectors in different angular positions, and to allow resizing of all beams with a small angular extent around each sector position while maintaining the cell fully covered. The control processor in Figure 5 operates on the basis of the measured C/I t and updates the BFN parameters at all times. In the case of a single service cellular system, uniform traffic conditions, and Fig. 5. Conceptual scheme of a base station using adaptive cell sectorization (uplink chain).

10 262 R. GIULIANO, F. MAZZENGA AND F. VATALARO perfect uplink power control, the above C/I t equalization procedure may be entirely based on the measured received power of the CDMA signals in each sector s i t for i D 1, 2, 3 in Figure 5 without resorting to IF processing or to baseband processing. In general, to ensure a user seamless sector rotation/resizing operations the position of each user in the cell needs to be continuously tracked by a sector identification algorithm (SIA). The SIA controls the selector to enable user detection on the CDMA signals, s i t. If antenna patterns identifying the sectors overlap, the SIA can enable detection over more than one sector and the two signals can be suitably combined before the final decision. Considering a BS with two different antenna arrays, one for uplink and the other for downlink, in general the downlink BFN weights are different from the uplink ones. Therefore, the conceptual scheme for the downlink chain of the BS is illustrated in Figure 6. The switching matrix in Figure 6 is used to route the signals of the users to their cell sectors. Routing paths are changed according to the sector positions determined with the criterion illustrated in Section 4.2. A different downlink transmitter (Tx i, i D 1, 2, 3) is used for each sector. In general, to apply the downlink power equalization algorithm illustrated in the previous section, the BS should know the position of each served user. This is necessary to evaluate the total transmitted power in the sectors when we assume that we transmit a different power for each user (downlink power control). Therefore, to apply the proposed algorithm, user localization is a requirement. When users localization is not available, assuming that the downlink power allocated to each user is the same (i.e. no downlink power Fig. 6. Conceptual scheme of a base station using adaptive cell sectorization (downlink chain). control or power control with reduced dynamic) the sector rotation algorithm is simplified, since the only necessary information is the number of users for each sector position. This information could be easily extracted from uplink measurements. When a single antenna BS is considered (as is always the case) and perfect uplink power control is considered, to equalize P sc in the downlink the sectorization layout obtained in uplink can be reused provided that: 1. the ratio between the intracell interference and the external interference in each sector is the same; 2. the same power is transmitted for each channel in the downlink; 3. channels have the same rate. The condition expressed in the above points imply that after sector rotation/resizing, each sector contains an equal number of users. 6. Simulation Results To analyze the performance of the proposed approach and to estimate its effectiveness, simulation results are now provided. The service area comprises 19 hexagonal cells located on two concentric rings. The reference cell is located in the center of the service area. A shadowing process, spatially white with standard deviation s D 6 db is considered. Users were placed randomly according to a uniform distribution. We assume that each user in the area is served by the BS with the larger propagation factor m. To perform optimization, we only consider sector rotation with a finite step of rotation of 1 degrees. We consider a CDMA system with bandwidth B D 1.25 MHz, with basic bit rate of R D 8 kbit s 1 and P s / D 1dB, and P s is assumed to be identical for each BTS in the area [15]. Performance is evaluated in terms of the probability of outage as a function of the average number of active codes in the cell. We maintain that the quality of service objective is met if the received energy per bit-to-noise power density ratio is greater than E b /N 0 u D 7 db for both voice and data uplink transmissions and E b /N 0 d D 5 db for downlink. Multimedia users employ eight basic rate channels with non-orthogonal codes and the power transmitted is eight times the power of a single code user. In our simulations a voice activity factor D 1was assumed for voice users. Considering only voice users, when <1 the number of served users in the cell is approximatively given multiplying the number

11 SMART ANTENNAS FOR 3G CDMA SYSTEMS 263 of codes obtained for D 1 with the inverse of. Different data activity factors ranging from 0.2 to 1 were considered for data users. Perfect power control at the BS was assumed for uplink transmission. To simplify downlink analysis and to speed up simulation, we assume that the BS transmits the same power P T to each terminal in the area it controls. The antenna radiation pattern used for both uplink and downlink transmissions is illustrated in Figure 7. The angular extension of the main lobe is of 2/n s degrees and we assume that the secondary lobe is as in Figure 7 with a constant gain G sec. The assumed antenna model is a conservative (i.e. pejorative) model. However, it gives information on the maximum secondary lobe gain tolerated in order to neglect the intracell and intercell interference due to the interfering sectors. Fig. 7. Radiation diagram of the antenna considered in the simulation Uplink Performance In Figure 8 we plot the outage probability as a function of the average number of active codes in the reference cell and for different values of the angular rotation step 1. Both fixed and rotating sectors have been considered for the basic rate service. Results were obtained by Monte Carlo simulation generating at each trial a distribution of users (snapshot simulation) and rotating the sectors of the reference cell and selecting according to Equation (4). As expected, system performance depends on the selected rotation step 1 and improves with decreasing 1 since the results of maximization in Equation (4) with respect to becomes more accurate. In Figure 9 we plot the outage probability as a function of the average number of active codes and for different values of 1 in the case of multicode traffic. We define the capacity improvement G as the difference between the average numbers of active codes in the mobile sectors case with the active codes in the fixed case normalized to the average number of active codes in the fixed case. The capacity improvement corresponding to the cases in Figures 8 10 are given in Table I for an outage probability P outage D 10 2 and for a different number of sectors. From the results in Table I it can be observed that G is practically independent on the number of sectors but it is influenced by the typology of the traffic (voice, data and mixed). As expected, capacity gain increases passing from voice traffic alone (basic voice rate 8 kbit s 1 )todatatrafficalone(datarate Fig. 8. Probability of outage as a function of the number of active codes in the cell for different values of 1, voice traffic, n s D 3.

12 264 R. GIULIANO, F. MAZZENGA AND F. VATALARO Fig. 9. Probability of outage as a function of the number of active codes in the cell for different values of the sector angular rotation step (1), data traffic, n s D 3. Fig. 10. Probability of outage as a function of the number of active codes in the cell for different values of the sector angular rotation step (1), mixed traffic, n s D 3. Table I. Uplink capacity gain for different types of traffic for an outage probability of 10 2 ; three sectors scenario. P outage D 10 2 Fixed sectors Mobile sectors G (%) Voice traffic Mixed traffic Data traffic kbit s 1 ). In the latter case the sources of interference are stronger and sparse in the coverage area thus approaching the condition of an instantaneous non-uniform user distribution Uplink Performance with Hot Spots The system capacity improvement is more evident considering statistically non uniform conditions such as in the hot spot case. In Figures 11 and 12 we plot the outage probability as a function of the average number of users in the reference cell for different values of the sector step rotation 1. We considered a

13 SMART ANTENNAS FOR 3G CDMA SYSTEMS 265 Fig. 11. Probability of outage as a function of the number of active codes in the cell for different values of the sector angular rotation step (1), hot spot condition voice traffic, n s D 3. Fig. 12. Probability of outage as a function of the number of active codes in the cell for different values of the sector angular rotation step (1), hot spot condition data traffic, n s D 3. single hot spot with its center located in the middle of the cell radius. Hot spot area was a circle entirely contained in the reference cell and having a radius equal to a quarter of the cell radius. In each trial the angular position of the spot s center was randomly generated in accordance to a uniform distribution. The number of the spot s users was a random variable ranging between [0,n hot ]wheren hot is the maximum number of users in the spot. Hot spot users were randomly positioned inside the spot area. We considered n hot D 16 for the voice case and n hot D 4 for data. Results show that fixed sectors are inadequate to cope with non-uniform traffic conditions while smart sectors allow the restoration of the quality of service Downlink Performance We considered the same scenario as in the uplink and we evaluated the downlink performance assuming a uniform user distribution in the area and considering different orthogonality factors. Only sector rotation based on the min max criterion was considered. The formulation of the min max downlink criterion is similar to that in Equation (4) where C/I i UL is

14 266 R. GIULIANO, F. MAZZENGA AND F. VATALARO Fig. 13. Outage probability as a function of active codes for different orthogonality factors downlink case no data users. Fig. 14. Outage probability as a function of active codes for different orthogonality factors downlink case 10 per cent data users with data activity factor D 0.2. replaced with the total power transmitted in the i-th sector. We assume a voice activity factor D 1, we neglect the presence of thermal noise and we consider an ideal antenna sectorization pattern. In Figure 13 we plot the outage probability as a function of the active codes for different orthogonality factors. In Figure 14 we plot the outage probability as a function of the active user codes for different orthogonality factors when both data and voice users are present. The proposed technique is effective to counteract the effects of orthogonality loss in downlink transmissions. However, the capacity gain is lower than the uplink and in general depends on. However, from the results in Table II it can be observed that the capacity gain is practically independent on. Table II. Downlink capacity gain for different orthogonality factors downlink transmission. P outage D 10 2 D 0.4 D 0.9 Voice traffic 13% 16% Mixed traffic 25% 27%

15 SMART ANTENNAS FOR 3G CDMA SYSTEMS Conclusions In this paper we reviewed the main advantages and disadvantages of smart antennas for their use in 3G cellular CDMA systems. Given the complexity of most implementations, we then proposed a technique to render sectorization smart and evaluated its performance for both uplink and downlink transmissions. The proposed smart sectorization technique is based on uplink C/I t equalization in each sector and on the equalization of the transmitted power in each sector for downlink transmissions. Both C/I t and transmitted power equalization are obtained through angular rotation and/or sector resizing driven by a min max optimization criterion. The selection of the optimal angular position is obtained on the basis of the measured C/I t. The effectiveness of the proposed procedure was assessed through simulation. The number of additional served users with respect to the case of fixed sector coverage increases passing from the simple voice traffic to a multimedia traffic where the source of interference is stronger and more sparse in the coverage area, thus exhibiting less uniform traffic conditions. Significant capacity improvement was observed for uplink transmissions, especially in non-uniform traffic conditions that are more likely when data transmissions are prevailing. As capacity in the uplink is frequently the limiting factor, it turns out that the smart sectorization technique tends to level up capacity on both links. References 1. Ponnekanti S. An overview of smart antenna technology for heterogeneous networks. IEEE Comm. Surveys 1999; 2(4): Naguib AF, Paulraj A, Kailath T. Capacity improvement with base-station antenna arrays in cellular CDMA. IEEE Trans. Veh. Tech. 1994; 43(3): Gilhousen KS, Jacobs IM, Padovani R, Viterbi AJ, Weaver A, Wheatley CE. On the capacity of a cellular CDMA system. IEEE Trans. Veh. Tech. 1991; 40(2): Giuliano R, Mazzenga F, Vatalaro F. Adaptive cell sectorization for UMTS third generation CDMA systems. Vehicular Technology Conference, 6 9 May, 2001; Rhodes, Greece. 5. Brunner C et al. On space time Rake receiver structures for WCDMA. Proc. 33rd Asilomar Conf. signals, Systems, Computers, Pacific Grove, CA, October 1999; vol. 2, pp Haardt M et al. Efficient joint detection techniques in the frequency domain. Int. Conf. Third generation wireless communications, San Francisco, CA, June 2000; pp ITU-R Working Document towards submission of RTT candidate to ITU-R, IMT-2000 process, The Cdma2000 RTT candidate submission. 2nd June, Godara C. Application of antenna array to mobile communications. Part II: beam forming and direction of arrival considerations. Proc. IEEE 1997; 85(8): Lehne PH, Pettersen M. An overview of smart antenna technology for mobile communication systems. IEEE Comm. Surveys 1999; 2(4): Mogensen PE et al. Measurements, channel statistics and performance of adaptive base station antennas. Proc. COST 259/260 Joint Workshop, spatial channel models and adaptive antennas, Vienna, Austria, April, GPP, Technical Specification Group, Radio Access Network (RAN), Working Group 1, Physical channels and mapping for transport channels onto physical channels (FDD), TS V2.5.0 ( ). 12. Bull T et al. Technology in smart antennas for universal advanced mobile infrastructure (TSUNAMI R2108) Overview. RACE mobile telecommunication summit 1995, Cascais, Portugal, Nov., 1995; pp Lehne PH et al. Estimating smart antenna performance from directional radio channel measurements. Proc. VTC 99, Amsterdam, the Netherlands, Sept., pp Nettleton RW et al. Power control for spread-spectrum cellular mobile radio system. Proc. IEEE Veh. Tech. Conf., VTC-83, 1983; pp Corazza GE, De Maio G, Vatalaro F. CDMA cellular systems performance with fading, shadowing and imperfect power control. IEEE Trans. Veh. Tech. 1998; 2(47):