You are on page 1of 6

Wireless Access with Smart Antennas and SDMA

Christian Hartmann
Lehrstuhl f. Kommunikationsnetze LKN, Technische Universitt Mnchen Arcisstr. 21, 80290 Mnchen, Germany Tel./Fax +49 89 289-23500/-23523 hartmann@ei.tum.de

Abstract
One of the most important challenges with respect to wireless access is the limited capacity of the air interface which is due to the fact that the available transmission bandwidth is finite. Since the number of wireless subscribers is still growing rapidly and the desire for higher data rates is rising partially due to an ever growing popularity of the Internet a more efficient use of frequency resources is inevitable to meet future capacity needs. Space Division Multiple Access (SDMA) can be used to multiply the capacity given by conventional multiple access techniques such as FDMA, TDMA or CDMA. However, the actual capacity gain which can be achieved with SDMA is highly dependent on the SDMA channel assignment and on the considered scenario (propagation, user distribution, traffic, mobility). Therefore in this paper several channel assignment strategies for SDMA systems are discussed and the problem of modeling different scenarios is addressed. Additionally some simulation results on the performance of SDMA systems are presented. Keywords: Wireless Access, SDMA, Smart Antennas, Dynamic Channel Assignment, Simulation 1. INTRODUCTION The demand for wireless communication capacity is permanently growing. Not only is the number of mobile subscribers ever increasing, additionally higher data rates are needed for applications such as wireless Internet access. In order to meet the future capacity requirements and keeping in mind that bandwidth is a scarce resource it becomes increasingly important to use the frequency spectrum as efficiently as possible. In order for various users to communicate over the air interface with a base station simultaneously, multiple access schemes are needed. The basic methods used so far are Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Code Division Multiple Access (CDMA), and any combinations of the afore mentioned [1]. However, the space domain has not been exploited within a cell so far. Equipping base stations with adaptive (smart) antenna arrays will enable beam steering in contrast to omnidirectional or sectorized single antenna systems. Thus, it becomes possible for base sta-

tions (BS) to radiate directed to specific users on the down-link as well as receive directed on the up-link. This can be exploited by reusing channels within a cell for mobile stations (MS) which are spatially separable by the smart antenna array. This gives an additional orthogonal multiple access component called Space Division Multiple Access (SDMA) which can be used on top of the underlying schemes, be it FDMA, TDMA or CDMA [2][3][4][5]. While it is obvious that capacity can be increased by introducing SDMA, it is not straight forward to predict the quantity of this gain. The actual capacity gain will depend on various aspects such as channel assignment strategies, traffic scenarios, beamforming and properties of the radio channel. This paper is focussing on the impact of channel assignment strategies and traffic scenarios. The remainder of this paper is organized as follows: section 2 briefly discusses the merits of smart antenna systems. Then, in section 3 several channel assignment strategies for SDMA systems are discussed. In section 4 the problem of modeling SDMA systems is addressed. Some simulation results on the performance of SDMA systems are presented in section 5 and section 6 concludes this paper. 2. MERITS OF SMART ANTENNA SYSTEMS Smart antennas at base stations can be used to enhance mobile communication systems in several ways: D increased BS range D less interference within the cell D less interference in neighboring cells D increased capacity by means of SFIR or SDMA The first advantage noted, the increased BS range can be useful in some specific scenarios where the cell sizes are limited not by the traffic needs in the respective area but by the BS range. Here, smart antennas can be used to bundle the radiated power and thus achieve higher BS ranges. However, the aspects of reduced interference within the cell as well as in neighboring cells are of even higher interest. There are basically two ways of transforming the reduced interference in smart antenna systems into capacity gains. The first approach is called Spatial Filtering for Interference Reduction (SFIR). In this case all operating users within the cell are separated by classical

multiple access techniques (F/T/CDMA). However, the decrease in radiated interference due to the use of smart antenna beam steering can be exploited by tightening the frequency reuse in the system, provided all cells in the system use smart antennas. This yields a higher number of channels in each cell and hence an increased capacity. Another method to increase the capacity is SDMA where channels are reused within a single cell as described above. In this paper the focus is on SDMA systems. 3. CHANNEL ASSIGNMENT IN SDMA SYSTEMS In order to get an optimal capacity increase through SDMA, channel assignment is a crucial aspect. The first part of this section deals with the assignment of SDMA channels within a cell exclusively. In subsection 3.3 of this section the possibility of merging SDMA and dynamic inter-cellular channel assignment is roughly discussed. 3.1. Constraints and Costs

distance ratio of the new MS and any of the mobiles operating on that channel. We can now define the costs C associated with the assignment of the new MS to that channel in the following way [7] C i(c, MS j, t) + R, if a min t a th C i(c, MS j, t) + R, if D max u D th a D C i(C, MS j, t) + A a th ) B max , else min D th where A and B are constants that can be used to weight the different constraints. Usually the angular constraint is deemed more important than the distance constraint. 3.1.2 Signal Power and Interference Imposed Constraints In a different approach in [3] the spatial separability is described, taking into account a beamforming algorithm which aims to minimize the radiated power while providing each mobile on the particular channel with a target signal to noise and interference ratio (SNIR). In [3] the so called cross interference loss is proposed as a reasonable measure for the spatial separability of mobiles. The cross interference loss is defined as the ratio P P min where P is the radiated power that would be needed for the new user, as well as the mobiles that are already operating on the channel in question, in order to be accommodated via SDMA on a single channel and P min is the radiated power that would be necessary in order to accommodate all those mobiles each on a separate channel. The higher the powerloss, the less favorable is the assignment in question since a high powerloss indicates insufficient separability in the space domain and furthermore leads to unwanted inter-cell interference. Using this approach, it makes perfectly sense to identify the power loss directly with the costs as [7] C i(c, MS j,t) + P P min The costs C i(c, MS j, t) are deemed infinite if the beamforming with the SNIR constraint fails for that particular channel c. 3.2. Strategies

In order to perform SDMA channel assignment, we must find a way to determine whether a certain assignment is more or less favorable. For this purpose we define the allocation costs C i(c, MS j, t) which are the costs associated with the assignment of channel c to the mobile MS j in cell i at time t by means of SDMA and taking into account the users which are already operating on that channel c via SDMA [7]. Additionally, threshold costs C th are needed to decide if a certain assignment is allowed or not. An assignment is prohibited if C i(c, MS j, t) u C th. The exact definition of the costs C i(c, MS j, t) depends on the beamforming and the requirements concerning link quality. In the following two examples are given. 3.1.1 Geometric Constraints In [6] two constraints are identified which have to be kept for users in order to operate simultaneously on the same channel within a cell. Those can be seen as geometric constraints because they are defined by the user positions. Applying those constraints, two mobiles can operate on the same channel if their angular distance (as seen from the BS) exceeds a threshold value a th and if the ratio of their distances from the BS does not exceed a threshold value D th [2]. The first constraint depends on the beamforming (beamwidth) and the propagation environment (the number and distance of scatterers which have to be considered). The second constraint takes into account the non-zero sidelobes of realistic antenna beams and can be relaxed if power control is used. Let us denote as a min the minimum angular distance of the new MS to any of the mobiles which are already operating on that channel and D max as the maximum

Using the above described allocation costs, several approaches to assign channels within the cell via SDMA can be distinguished. The most important ones are described in the sequel. We assume that a number of N ch channels c is assigned to the specific cell i. Each of the channels can accommodate several users via SDMA. When a new MS desires to establish a connection, two basic assignment approaches can be taken: D no rearrangement of ongoing calls D complete rearrangement of ongoing calls

In the first case the current channel to MS assignments within the cell will stay as they are and only the new MS (MS j) will be assigned to a channel c. In the second case the channel to MS assignment of all connections within the cell can be changed completely depending on the specific assignment strategy. Obviously, allowing complete rearrangement can give somewhat more capacity due to the higher degree of freedom. However, there are some drawbacks to this approach, namely the high computational complexity which is problematic because of the real-time requirements and a drastically increasing number of handovers, which is also undesirable. In the following, the most important strategies for the no rearrangement case are described only: Random First Available A random channel c k is chosen and the costs C i(c k, MS j, t) are determined. If C i(c k, MS j, t) v C th, then MS j is assigned to channel c k. Otherwise another channel c l with l 0 k is chosen randomly. If the assignment costs are higher then C th for all N ch channels, then the connection will be rejected. Sorted First Available The channels are tested in a particular order c 1, . . . , c Nch. The mobile MS j is assigned to the first channel c k for which C i(c k, MS j, t) v C th is valid. If the assignment costs are higher then C th for all N ch channels, then the connection will be rejected. Best Separability (SDMA Last) For all channels c k the costs C i(c k, MS j, t) are determined. Then MS j is assigned to the channel with the lowest costs if those costs are not higher than C th, i.e. if min C i(c, MS j, t) v C th , c c 1, . . . , c N ch. c Otherwise the connection will be rejected. Worst Separability under Constraints (SDMA First) For all channels c k the costs C i(c k, MS j, t) are determined. Then MS j is assigned to the channel with the highest costs for which C i(c k, MS j, t) v C th. If the assignment costs are higher then C th for all N ch channels, then the connection will be rejected. The Random First Available and Sorted First Available strategies do not necessarily require the calculation of the allocation costs for all channels. Statistically the number of required attempts until a valid assignment is found will increase with growing traffic load in the cell. The Random First Available strategy will statistically spread the users over the available channels while the Sorted First Available strategy attempts to serve the users on as few channels as possible. On the other hand the Best Separability and Worst Separability under Constraints strategies require the calculation of the allocation costs for all channels before each channel assignment. The Best Separability approach attempts to keep the radiated

power and thus the interference in neighboring cells to a minimum. The first N ch users will always be served on a separate channel each. The Worst Separability under Constraints strategy attempts to pack users as densely as possible in order not to waste any capacity. However, this might lead to a high number of beam collisions, yielding handovers, depending on the degree of mobility of the users. 3.3. Combining SDMA and DCA

In the previous sections, only SDMA channel assignment within a cell has been discussed. The channel to cell assignment was thus assumed to be permanent, i.e. fixed channel allocation (FCA) has been used for the channel to cell assignment. However, dynamic channel allocation (DCA) can be used to optimize the spatial reuse of channels with respect to the current traffic distribution. Various approaches can be found in the literature to perform DCA in omnidirectional (non-SDMA) systems [8][9][10]. Less work has been done so far in the field of DCA considering SDMA. Several approaches can be taken to apply DCA in SDMA systems, one of which is called Hierarchical SDMA-DCA (HSDCA) [7]. In this strategy the process of assigning a channel to a mobile is hierarchically divided into the channel to BS assignment (inter-cell assignment) and the channel to MS assignment (intra-cell assignment) as indicated in Fig. 1.
Channel Pool


DCA HSDCA

DCA


DCA

SDMA

SDMA

SDMA

Fig. 1: HSDCA principle

In addition to the previously defined intra-cell assignment costs C i(c, MS j, t) we introduce the intercell allocation costs C DCA(c, i, t) representing the costs associated with the assignment of channel c to cell i at time t. The interworking between the two hierarchical levels is based on those costs. The complete HSDCA strategy comprises an acquisition algorithm an release algorithm and a packing algorithm. The acquisition algorithm is triggered when a new connection request arrives in cell i. The algorithm attempts to support the new connection on one of those channels currently assigned to cell i while minimizing the costs C i(c, MS j, t) and C DCA(c, i, t). Note that the respective cost functions will usually become minimal for different channels. In case no appropriate channel assignment can be performed using one of the cell channels, the algorithm aspires to acquire an additional channel via DCA using the upper hierarchical level where the costs C DCA(c, i, t) are minimized. If

this attempt fails, the connection request has to be rejected. The release algorithm controlling the release of channels to the channel pool and the packing algorithm allowing for rearrangements of ongoing connections in order to further optimize the current channel assignment are also operating on the cost functions C i(c, MS j, t) and C DCA(c, i, t) [7]. 4. MODELING SDMA SYSTEMS Determining the capacity of SDMA systems as compared to omnidirectional or sectorized systems is not an easy task. Theoretical approaches taken so far are usually suffering from severe simplifications which are necessary to obtain analytical results. On the other hand montecarlo simulations are a very powerful means to get an insight in the performance of systems. In this paper we are focussing on montecarlo simulations. For this purpose in this section the issue of modeling SDMA systems for computer simulations is addressed. 4.1. Modeling User Mobility, Distribution, and Teletraffic

The purpose of a mobile communication system is obviously to satisfy the communication demands of the subscribers. Thus, in order to evaluate the performance of such systems, the user behavior must be taken into account and be modeled for simulations. For this purpose the concept of user classes (UC) as described in [5] can be used. The overall traffic load A in a scenario is divided into N UC fractions A n each of which is generated by a UC, with:
N UC

proaches to achieve this. In our simulations we use two different models. The first one is based on scattering points which are randomly distributed and grouped in scattering areas [11]. Scattering areas are placed at each MS location and at the BS site, depending on the simulated scenario. Additionally, scattering areas are placed in the system plane to model scatterers (e.g. Buildings) which cause specifically strong scattering. Each scattering point is randomly assigned a complex scattering coefficient. Using those coefficients and the locations of the scatterers, the user- and interference-signals can be determined by combining the propagation paths for all scatterers. This model is very realistic and thus very favorable. However, its computational complexity becomes problematic when large networks are modeled especially in system simulations where much longer time periods must be taken into account than for link level simulations. That is why we use this model for detailed single cell simulations only, while a computationally more efficient model is used for multi-cell simulations. This second model does not take single scattering paths into account but considers one ore more main propagation paths only. All main propagation paths are modeled as log-normal fading. A special characteristic of this model is the spatial correlation of the fading which is obtained by spatial filtering. 4.3. Modeling SDMA Beamforming

An + A
n+1

Each UC is assigned a mobility model, a spatial distribution and a traffic model. Mobility models used are [5]: D Brownian Motion [12][13][14] D Manhattan Grid [15] D Steady Route [16] The spatial distribution of users can be chosen to be [15]: D homogeneous D inhomogeneous Used traffic models are [17][18]: D Erlang D Engset 4.2. Modeling Spatially Selective Channels

The beamforming in SDMA systems can be performed in various ways [3]. However, almost all beamforming approaches have in common that the main power will be steered in the main propagation direction and some sidelobe radiation will cause interference in unwanted directions. Even if the beamforming algorithm attempts to steer nulls in the direction of co-channel users, it has to be considered that the propagation directions are never distinct but are spread over some angular distance.

2p

Fig. 2: Brickwall antenna pattern

Since SDMA exploits the spatial distribution of users, it is inevitable to consider the spatial unisotropy of the mobile radio channel. There are several ap-

In order to be independent from the spatial characteristics of a particular beamforming algorithm, we use a generalized so called brickwall antenna beam pattern with a main beamwidth of b and a non-zero sidelobe attenuation of D as depicted in Fig. 2. This can be seen as a worst case beamforming and it also greatly reduces simulation time.

5. SIMULATION RESULTS In this section, some simulation results are presented. Particularly we are looking at two different scenarios. In the first scenario a single cell is considered only (inter cell interference (ICI) is considered via an additional interference margin). This enables us to investigate the influence of spatially inhomogeneous traffic within a cell on the SDMA capacity. The second scenario is a multicellular network comprising 49 cells. This is a scenario, well suited to get an insight in the performance of dynamic inter-cellular channel assignment in SDMA Systems. 5.1. SDMA and Inhomogeneous Traffic

The simulation results for this scenario in terms of blocking probability are depicted in Fig. 4. To the very left the blocking probability calculated from the Erlang-B formula for N=7 servers is plotted as a reference. From Fig. 4 we find that SDMA gives much lower blocking rates than an omnidirectional system (represented by Erlang-B) and thus higher capacity when a particular blocking rate is required. However, the results also show that inhomogeneous traffic has a negative impact on the capacity gain. With growing inhomogeneity (growing ) the SDMA gain decreases. Although we find that even for a high grade of inhomogeneity a remarkable capacity gain remains, we see that the traffic scenario is very important to be considered.
Blocking Probability B
0.1

As stated before, the capacity gain achievable with SDMA depends not only on the channel allocation strategy chosen but also on the specific scenario. It seems obvious that the highest gain through SDMA can be attained if the generated teletraffic is spatially homogeneous. In order to investigate the influence of inhomogeneous traffic more closely, we look at a single circular cell with two different user classes. The first user class (UC1) generates the part A @ (1 * ) of the over-all offered traffic Load A. Traffic within UC1 is generate homogeneously in the system plane and users are moving according to a manhattan grid mobility model. The second user class (UC2) generates the remaining part A @ of the traffic load in a specific area of the cell only, which we refer to as hotspot. Users in UC2 are moving according to the mobility model Brownian Motion. The combination of those UCs yields an inhomogeneous traffic scenario where describes the grade of inhomogeneity. Fig. 3 shows locations of origins of connection requests that occurred during simulation (within a limited time period) in this scenario where the hot-spot area can be identified by the higher traffic density.
y[km] 5 4 3 2 1 0 -1 -2 -3 -4 -5 -5 -4 -3 -2 -1 0 1 2 3 4 5 x[km]

0.01

0.001

Erlang-B

+ 0.8 + 0.5 + 0.3 + 0.0


0 2 4 6 8 10 12 14 16 18 20 Offered Load A/Erl

0.0001

Fig. 4: Blocking in inhomogeneous scenarios

5.2.

SDMA and Dynamic Channel Allocation

Fig. 3: Origins of connection requests in inhomogeneous scenario

The minimum carrier to interference ratio (CIR min) required for each connection in order to be supported has been chosen 15 dB. As resources, 7 channels are available to the BS all of which can be reused within the cell via SDMA. The chosen channel allocation strategy is sorted first available and a handover is performed whenever the CIR of a connection falls below CIR min or when the SDMA constraints are violated.

In order to investigate assignment strategies involving dynamic inter-cellular channel allocation like HSDCA, a scenario with only one cell is no longer sufficient. Instead, a scenario of 49 cells as depicted in Fig. 5 is used. The arrows in Fig. 5 indicate that a mobile with an ongoing connection that moves out of the scenario will reenter from another side while a handover is performed. For the performance evaluation, only the 9 cells in the center of the scenario (grey cells in Fig. 5) are taken into account in order to rule out edge effects. The minimum CIR for each connection is 15 dB as in the previous scenario. The resources are 49 channels available in the whole system. Three different channel allocation schemes are compared: FCA without SDMA, which we will refer to simply as FCA in the sequel; FCA with SDMA, which we will refer to as SDMA in the sequel; and HSDCA. Handovers (inter-cell and intra-cell) are performed when the CIR of a connection falls below CIR min, when the received signal power of a mobile falls below a threshold, or when the SDMA constraints are violated. No channel rearrangement is performed in the HSDCA case. Only one UC is considered in this simulation generating spatially homogeneous traffic and moving according to the Brownian Motion mobility model.

REFERENCES
[1] [2] [3] Fig. 5: Multicellular scenario Eberspcher, J.; Vgel, H. J.: GSM - Switching, Services and Protocols", Chichester, England: John Wiley & Sons Ltd, 1998 Gerlich, N.:" On the spatial Multiplexing Gain of SDMA for Wireless Local Loop Access". In Proc. 2. EPMCC '97, Bonn, Sept 30.-Okt.2.,1997, pp299-235 Farsakh, C; Nossek, J., A. "Data Detection and Channel Al location on the Uplink of an SDMA Mobile Radio System". In Proc. IEEE Int. Conf. on Telecommunications, Apr. 1997, Melbourne, Australia, Editor W. L. Lavery, Vol. 3, pp 813-818 Palestini, V.; Rolando, A.: Multibeam Antennas for a PCS System Based on GSM/DCS1800 Technology: an Ap proach to Space Division Multiple Access (SDMA)". In: Proc. ICPMSC'96,, Hong Kong, Dec 3 5, 1996; pp. 72 75. Hartmann, C.; Vgel, H. J.: Traffic Analysis of SDMA Systems with Inhomogeneous MS Location Distribution and Mobility", in: Wireless Personal Communications, Special Issue on Space Division Multiple Access (SDMA); Kluwer Academic Publishers, Vol. 11, No. 1, Oct. 1999 Gerlich,Notker;Tangemann,Michael:"TowardsaChannel Allocation Scheme for SDMA based Mobile Communica tion Systems". In: Proc. ITG Fachtagung Mobile Kommu nikation, 26. 28. September 1995 in Neu Ulm. Berlin, Of fenbach: vde verlag, 1995; pp. 109 116. Hartmann, C.: Dynamic Channel Allocation in Cellular Mobile CommunicationSystemswithSDMA",in:Proceed ings EPMCC '99, Paris 1999. Priscoli, F., D.; Magnani, N., P.; Palestini, V. Application of Dynamic Channel Allocation Strategies to the GSM Cellu lar Network". In IEEE Journal on Selected Areas in Com munications, Vol. 15, No 8, Okt 1997, pp. 1558-1566 Katzela,I.;Naghshineh,M.ChannelAssignmentSchemes for Cellular Mobile Telecommunication Systems: a Com prehensive Survey". In IEEE Personal Communications, June 1996, pp. 10-31

The simulation results in terms of blocking and dropping probabilities are depicted in Fig. 6 and Fig. 7. The relative behavior of FCA, SDMA and HSDCA is very similar for the blocking and dropping probabilities. In either case SDMA performs much better than FCA but HSDCA gives an enormous additional gain.
Blocking Probability

[4]

[5]

0.1
[6]

0.01
[7]

0.001
FCA SDMA HSDCA

[8]

0.0001 0 5 10
Fig. 6: Blocking
Dropping Probability

15

20

Offered Load / Cell [Erl]

[9]

0.01

[10] Baiocchi, A.; Delli Priscoli, F.; Grilli, F.; Sestini, F.: The geometric dynamic channel allocation as a practical strat egy in mobile networks with bursty user mobility" IEEE Trans. Veh. Technol., Feb. 1995 [11] Blanz, J. J.; Baier, P. W.; Jung, P.: A Flexibly Configurable Channel Model for Mobile Radio Systems with Directional Diversity". In: Proc. ITG Fachtagung Mobile Kommunika tion, 26. 28. September 1995 in Neu Ulm. Berlin, Offen bach: vde verlag, 1995; pp. 93 100.
FCA SDMA HSDCA

0.001

0.0001

[12] Robertson, H. D.; Hummer, J. E.; Nelson, D. C.:"Manual of Transportation Engineering Studies." Institute of Trans portation Engineers. Prentice Hall. NJ 1994
20

10

15

Offered Load / Cell [Erl]

[13] Guerin, R. A.:"Channel occupancy time distribution in a cellular radio system", IEE Trans. Veh. Technol., vol. VT 35, pp. 89-99, Aug. 1987 [14] Tekinay, S.: Modeling and analysis of cellular networks with highly mobile heterogeneous traffic sources" Ph.D. Dissertion, George Mason Univ., Fairfax, VA, 1994 [15] Markoulidakis, J. G.; Lyberpoulos, G. L.; Tsirkas, D. F.; Sy kas, E. D.: "Mobility Modeling in Third Generation Mo bile Telecommunications Systems". In: IEEE Personal Communications. August 1997; pp. 41 56. [16] Frullone, M.; Grazioso, P.; Serra, A. M.:"Performance Evaluation of a Personal Communication System in a Ur ban Environment" Proc. 12th IEEE VTC, Denver, CO, May 1992 [17] Kleinrock, L.: Queueing Systems - Vol. 1: Theory. New York, London, Sydney, Toronto: J. Wiley & Sons Inc., 1975 [18] Khn, P.: Tables on Delay Systems", University of Stutt gart, 1976, ISBN 3 922403 05 0

Fig. 7: Dropping

6. CONCLUSIONS In this paper, the influence of traffic scenarios and channel allocation schemes on the capacity performance of SDMA systems has been addressed. The presented simulation results emphasize that inhomogeneous traffic causes a degradation of the SDMA capacity gain while dynamic inter-cellular channel allocation combined with SDMA can further increase the capacity as compared to pure SDMA and thus can enhance the efficiency of frequency usage.

You might also like