A MIMO framework for 4G systems: WINNER Concept And Results Afif Osseiran∗ , Veljko Stankovic† , Eduard Jorswieck‡ , Thorsten Wild§ , Martin Fuchs† and Magnus Olsson∗ ∗ Ericsson Research, Stockholm, Sweden, [email protected], [email protected] † Technische Universit¨at Ilmenau (TUI), Ilmenau, Germany, [email protected], [email protected] ‡ Royal Institute of Technology (KTH), Stockholm, Sweden, [email protected] § Alcatel-Lucent, Stuttgart, Germany, [email protected]

Abstract— In this paper, the MIMO framework within WINNER for fourth generation radio systems is further developed and assessed for various deployment scenarios. The emphasis is on radio network system aspects of multi-antenna techniques where preferred configurations for three basic deployment scenarios are given. In the wide area scenario which aims to provide ubiquitous coverage for rural, suburban and urban areas, the scheme selection depends on the user density of spatially separated users. It ranges from grid of fixed beams (GoB) (TDMA based) in case of few users to SDMA with a fixed linear precoding codebook, and finally to adaptive beams with SDMA for highly dense system. In the metropolitan area scenario which is targeting system deployments in large urban environments, multi-user (MU) MIMO precoding performs very well for slow moving users. For higher velocities, per antenna rate control (PARC) or adaptive linear dispersion codes (LDCs) are better choices. In the local area scenario which is characterized by isolated sites, the combination of SDMA and spatial multiplexing achieved by MU-MIMO precoding provides high spectral efficiency.

Keywords: MIMO, OFDM, spatial processing, spatial user grouping, WINNER. I. I NTRODUCTION The research project WINNER [1] is a cooperation of 41 partners from industry, operators, and academia, which is partly funded by the European Union. WINNER aims at developing a ubiquitous next generation radio system concept providing wireless access to a full range scenarios from indoor isolated hotspots to wide area cellular coverage. Main objectives are increased data rates, low latency, and high system capacity. In order to reach these goals, advanced spatial processing is employed. The use of multiple antennas at transmitters and receivers in a wireless communication system offers additional degrees of freedom which can be used for adaptive directivity, diversity and multiplexing. It implies that reception and transmission in different directions may be controlled, redundancy in the spatial domain may be used, and that the time and frequency resources may be re-used for parallel transmission to one or several users. From a system perspective these basic benefits Part of this work has been performed in the framework of the IST project IST-4-027756 WINNER II, which is partly funded by the European Union. The authors would like to acknowledge the contributions of their colleagues in WINNER II, although the views expressed are those of the authors and do not necessarily represent the project.

can be used in different ways to achieve different objectives such as increased coverage and support of high data rates, increased robustness to interference, increased reliability, and improved spectral efficiency. For this purpose, a unified, generic and flexible MIMO transmission concept based on per stream rate control (PSRC), linear dispersion codes (LDC) and linear precoding (LP) was developed [2] [3]. It aims at choosing the appropriate spatial scheme (i.e. MIMO method) with high efficiency and high quality given a deployment scenario. The spatial scheme is chosen in such a way as to adapt continuously to the spatial properties of the channel. The MIMO concept was later integrated in the overall OFDM-based air interface [4][5] in order to ensure that the WINNER system as a whole can exploit efficiently the benefits of spatial processing. In this paper, the MIMO framework within WINNER is refined and assessed. Consequently we identify suitable MIMO configurations for important deployment scenarios. The outline of this paper is as follows: First, the WINNER MIMO concept is briefly described and some important areas are highlighted. Then we assess and identify suitable MIMO configurations for three different deployment scenarios. Finally, conclusions and an outlook on future work is given. II. T HE WINNER MIMO FRAMEWORK A. Generic Spatial Transmit Processing Figure 1 shows a block diagram of the generic spatial transmit processing. The basic physical transmission unit is called a chunk and consists of several adjacent subcarriers of few consecutive OFDM symbols. The chunk size is chosen so that the channel variations within a chunk is negligible. At the input of the transmitter are the incoming data transport blocks from higher layers. Each of these transport blocks is segmented and channel encoded in a forward error correction (FEC) entity. These encoded segments of transport blocks are referred to as FEC blocks. An important property of the radio interface is that each chunk is partitioned into one or several spatial layers denoted as chunk layers. The above described FEC blocks, i.e. encoded segments of transport blocks, are multiplexed onto these layers. The bits mapped to each chunk layer are separately modulated. The so formed modulated chunk layers are then dispersed or spread onto virtual antenna chunks with a linear

Fig. 1.

Generic Transmitter

dispersion code which is a three dimensional entity spanning the adjacent subcarriers of the consecutive OFDM symbols in time and frequency corresponding to the chunk in addition to the spatial dimension which has been added. All virtual antenna chunks are then subject to a processing technique that is referred to as generalized multicarrier (GMC) processing. The GMC function operates on an OFDM symbol basis in the frequency domain over all chunks allocated to a transport block. More specifically, the layers of virtual antenna chunks are jointly processed by an identity function (when OFDM is used) or a Discrete Fourier Transform (for the case of precoded OFDM) and then split and dispersed over the virtual antenna chunks again. The virtual antenna chunk of each layer is further subject to linear precoding. Finally, the layers’ antenna chunks are summed over the antennas to form a threedimensional antenna chunk, which is passed to assembly and OFDM modulation per antenna. B. Measurement Requirements In order to operate the generic transmitter and adaptively configure it, accurate measurements of channel and interference characteristics are required. We distinguish the following types of measurements required for spatial processing: Channel State Information, Channel Quality Indicators, Effective CSI including the effect of spatial transmit processing (which can all be Short Term or Long Term, where LT CSI is in the form of estimated correlation matrices), scalar Noise plus Interference Power, and frequency dependent Int. Power. In TDD mode, CSI of the downlink channel is obtained by assuming reciprocity of uplink measurements after proper calibration. ECSI requires dedicated pilots to include the effect of spatial processing. In FDD mode, mainly feedback is used. CQIs can be fed back or computed from CSI. Table I lists requirements per spatial processing technique. C. Spatial Adaptation The main idea behind the WINNER MIMO concept is to adapt the transmission to the user needs in different deployments under different conditions of channel and interference knowledge. The process of selecting the spatial scheme may

be called spatial adaptation. The essence of employing advanced spatial user grouping schemes is to exploit the spatial dimension in order to be able to serve additional users where these users are no longer separated by an orthogonal multiple access scheme from the already present one. Hence, the spatial user selection will either group the users according to their spatial correlation or their transmit strategies. User grouping coordinates transport block transmissions in such a way that the experienced intra-cell interference at the mobiles during downlink transmission or at the BS during uplink transmission is reduced. Initial work in this area was reported in [4][5], and a temporarily layered process for selection of the spatial scheme was outlined. Two steps are performed iteratively. The first step separates the users according to their spatial signatures. The users within a set are orthogonally multiplexed (TDMA/FDMA) whereas the different sets are spatially multiplexed (SDMA). The case in which each set consists of one user leads to pure SDMA where more than one user is served in a sector at a certain time instant by exploiting the spatial separation. The other extreme in which all users belong to one set leads to pure TDMA/FDMA. 1) Spatial User Selection/Grouping: The task of spatial user grouping or partitioning is to separate the UTs into spatial sets with low mutual interference. Once such spatial sets are established, we use SDMA for flows of the same set. All user selection and user grouping schemes can be compared to upper bounds, i.e. the sum capacity ! N K X X H max tr log I + SNR Hk,n Qk,n Hk,n n=1

k=1

under power and/or individual QoS requirements. For user selection, a greedy approach is proposed in [6] that can achieve a large portion of the sum capacity. For user grouping, a semi-orthogonal user grouping algorithm is developed in [7]. In WINNER three algorithms are developed: The ProSched [8] algorithm was originally derived for a precoding technique that forces the interference to be zero between all users that are served simultaneously via SDMA. The goal of the algorithm is to group together users with uncorrelated channels. The precoding technique under investigation allowed the use of an efficient approximation technique based on orthogonal projection matrices to estimate the capacity after precoding, which serves as a scheduling metric. Another approach to user grouping is based on successive user insertion for long-term adaptive beams with SDMA using short-term CSI [9]. Long-term adaptive beamforming is used jointly with adaptive scheduling and additionally allows simultaneously transmitting users for one resource block of one or several chunks in the OFDM downlink. This scheme is intended for low angular spreads and moderate to many users. The third scheme is called the overlapping spatial user-grouping approach and aims at achieving a good trade-off between performance and complexity. Overlapping user grouping can provide user-selection with more user-multiplexing forms than

Technique Adaptive linear multi-user MIMO precoding and scheduling (LA downlink adaptive mode) Matrix LDCs (STBCs) (uplink; LA, WA, MA downlink non-adaptive mode) Fixed grid of beams linear precoding and multi-user sched. (WA downlink adaptive mode) Adaptive grid of beams linear precoding and multi-user scheduling (downlink) (using variable codebook size with possible transition to LT CSI based prec.)

CSI LT/ST or mixed -

required at TX (purpose) CQI LT/ST NIP (precoding, scheduling, adaptive mod.cod.), other CQI (sched., technique dependent) can be obtained from CSI, IP (interference avoidance sched.) may be obtained from CSI LT/ST NIP (ad.mod.cod., LDC selection), condition number (LDC sel., technique dependent)

-

ST SINR per user and per beam (adaptive beam selection and scheduling) LT/ST NP (ad.mod.cod., beam selection), obtainable from SINR and CSI, IP (interference avoidance scheduling) may be obtained from CSI

LT/ST or mixed (beam adaptation)

ST SINR per beam (beam selection, scheduling), LT/ST NP (ad.mod.cod., beam sel.) obtainable from SINR and CSI, IP (interference avoidance scheduling) may be obtained from CSI

required at RX (purpose) CSI CQI ECSI LT/ST NIP (spatial equalizer) (spatial equalizer) -

-

ECSI (spatial equalizer) from CSI because of fixed weights ECSI (spatial equalizer)

LT/ST NIP (spatial equalizer)

LT/ST NIP (spatial equalizer)

TABLE I M EASUREMENT REQUIREMENTS OF VARIOUS SPATIO - TEMPORAL PROCESSING TECHNIQUES

tan Area (MA), and Local Area (LA). For further details on the deployment scenarios and the assessments, see [10] [11].

Fig. 2.

Example of Spatial Adaptation

non-overlapping user grouping due to the characteristic that each user can appear in more than one group. 2) Spatial Schemes: In Figure 2, the spatial adaptation mechanism is characterized. The flow is segmented for per stream rate control (PSRC). Three types of precoding can be chosen depending on the QoS requirements, CSI, result from spatial user selection, and other system parameters. Fixed precoding leads to a grid of fixed beams (GoB) approach with one (TDMA) or multiple (SDMA) active beams. A spacetime code is selected for the case without precoding. The two adaptive precoding types are based on short-term or longterm CSI. Short-term (long-term) multiuser adaptive precoding is realized by (long-term) Successive MMSE (SMMSE) precoding [10], otherwise single-user linear precoding, i.e. power allocation and beamforming based on the available spatial CSI is performed. III. A SSESSMENTS AND P REFERRED C ONFIGURATIONS In this section we will assess the WINNER spatial processing concept and identify suitable configurations of it in three different deployment scenarios; Wide Area (WA), Metropoli-

A. Wide Area The WINNER wide area concept shall provide ubiquitous coverage for rural, suburban and urban areas. It is characterized by medium to large cell coverage where the users may move at very high speeds. The considered test scenario is an urban macro-cellular deployment using the FDD physical layer mode, a carrier frequency of 3.7 GHz / 3.95 GHz and 2 x 50 MHz bandwidth. A traditional hexagonal cell layout is considered with 3-sector sites, a site-to-site distance of 1000 m, and antennas mounted above roof top level. In such a scenario, typically the angular spread is low and the antenna correlations are significant, which results in a low rank channel matrix. Spatial schemes for wide area have to be robust and to cope with these typical environmental properties. Shortterm CSI will not be available at the transmitter as user mobility is non-negligible and the feedback bandwidth and delay are limited. On the other hand, for low to medium user velocities scalar CQI information can be used on a fast fading granularity. Due to these limitations, the focus of the assessments has been on different linear precoding (i.e. beamforming) techniques. These included single-stream transmission with a grid of fixed beams (GoB) or adaptive beamforming, higher order sectorisation (HOS), and SDMA based on either GoB or adaptive beamforming. The first investigated form i.e. single-stream transmission GoB in a 3-sector site with 4 transmit (TX) antennas provided a gain in the order of 25-40% compared to a single transmit antenna. Using adaptive weights based on eigen-beamforming instead of GoB gives nearly no performance gain. One way to improve the performance of single-stream transmission was to use higher order sectorisation (HOS). E.g. the number of sectors per site was increased to e.g. 6, 9 or 12. Two ways of implementing HOS are investigated. The first consisted of using 3 ULAs where each of the generated fixed beams formed a logical cell (i.e. sector). This setting improved the performance by 40%, 100% and 140% for 6, 9 and 12 sectors

Number of transmit antennas and scheme 1 TX, TDMA 4 TX, GoB + TDMA 4 TX, GoB + SDMA

10% outage user spectral efficiency [bit/s/Hz] 0.027 0.034 0.036

average site spectral efficiency [bit/s/Hz/site] 4.08 5.00 8.18

TABLE II S PECTRAL EFFICIENCY OF 3- SECTOR

50% User spectral efficiency [bit/s/Hz]

3 km/h 50 km/h

0.240 0.100

X% site spectral efficiency [bit/s/Hz/site]

50% 4.8 2.72

90% 9.27 5.01

TABLE III S PECTRAL EFFICIENCY OF 1- SECTOR SITE FOR MU-MIMO

SITE FOR WIDE - AREA .

E ACH UT IS EQUIPPED WITH 2

UT speed

52 UT S / SITE .

ANTENNAS .

respectively. The second setting consisted of using one ULA per sector in order to generate the sector antenna diagram. This second setting allowed a finer granularity of spatial reuse and extra antenna gain hence improving the performance up to a factor of 3.5 for 12 sectors at the expense of requiring more antenna elements (AEs). A promising possibility for performance improvement is to use SDMA on top of GoB. This combination always outperforms single-stream GoB (or so called TDMA based GoB), see also Table II. GoB based SDMA outperforms HOS with 3 arrays by about 25%. HOS can outperform SDMA at the expense of requiring more AEs per site. In addition, the SDMA performance can be further improved by using proper antenna design (e.g. using tapering for beam design). Furthermore, SDMA based on adaptive beams can additionally increase the performance by 10-20% at the expense of requiring dedicated pilots. B. Metropolitan Area The WINNER metropolitan area concept is targeting system deployments in large urban environments, and should provide contiguous outdoor coverage especially in city centres of large and medium size cities. It is intended to support high user density, high system throughput, and mobility up to reasonable velocities in urban environments, e.g. 50 km/h. The test scenario that we will focus on is an urban microcellular scenario using the TDD physical layer mode and 100 MHz bandwidth at 3.95 GHz. The considered deployment is a two-dimensional regular grid of buildings, so called Manhattan grid [12], where the users are located in the streets only. The base stations are placed in centre of streets in centre of blocks. The streets are 30 m wide, and the blocks are 200x200 m. Spatial schemes in the metropolitan area have to cope with the challenging radio propagation conditions arising in these environments, and must be able to meet the high throughput requirements. The limited mobility facilitates availability of reliable short-term CSI at the transmitter. Under these conditions, investigated schemes have been SMMSE MU-MIMO precoding, adaptive LDC, per antenna rate control (PARC), and various forms of eigen-beamforming (EIGBF). The simulation results show that single-stream EIGBF perform poorly in comparison to PARC. For example, twostream PARC was shown to provide approximately 50% higher site throughput compared to single-stream transmission from 1TX antenna, and about 20% higher site throughput

FOR METROPOLITAN AREA .

16 UT S / SITE . T HE BS

ANTENNAS AND EACH

UT

PRECODING

IS EQUIPPED WITH

IS EQUIPPED WITH

8

2 ANTENNAS .

than single-stream EIGBF. It was shown that multi-stream EIGBF matches the performance of PARC at the expense of requiring substantially more antenna elements. Evaluations of a single user link indicate that adaptive LDCs is a good or better alternative than PARC when it comes to multi-stream transmission. SMMSE MU-MIMO precoding was shown to perform very well for low user velocities (3 km/h), providing 80% higher site throughput than a SISO system and 160% higher user throughput. However, at higher user speeds, e.g. 50 km/h, the performance of SMMSE MU-MIMO precoding degrades and is similar to that of a SISO system. The spectral efficiencies for MU-MIMO precoding are shown in Table III. C. Local Area The WINNER local area concept is focused on isolated sites in indoor and hotspot scenarios. For indoor scenarios the BSs cover one or few cells, hence encounter limited interference from other systems as the houses represent a well protected environment. In hotspot scenarios, the coverage area is higher than the home deployment, hence several BSs may be required. The mobility in these environments is very low, typically in the range 0-5 km/h. The scenario under investigation is an indoor scenario targeting inhome and hotspot scenarios, using the TDD physical layer mode and 100 MHz bandwidth at 5 GHz. The considered deployment consists of one floor (height 3 m) of a building containing two corridors of 5x100 m and 40 rooms of 10x10 m. Spatial schemes in the local area have to be able to deliver high throughput and high user data rates. TDD operation and low mobility allows the estimation of short-term CSI at the transmitter based on the uplink measurements and the reciprocity principle. Hence MU-MIMO processing techniques are considered. These techniques exploit the channel knowledge at the BS to provide very high cell and user throughput. On the downlink the BS performs MU-MIMO precoding using the channel information obtained on the uplink. The UTs perform MMSE detection using the estimates of the effective channel using dedicated pilots per flow. On the uplink the BS performs the MU-MIMO decoding. The channel from the UTs to the BS is estimated using the common pilots per antenna. Since the processing at the BS is not necessarily the same on the uplink and the downlink, it is reasonable to assume that the UTs do not have the exact CSI. We can distinguish two situations. In the first case the users transmit using one of the techniques that do not

Fig. 3. CCDF of the MU-MIMO information sum rate. Solid curves are obtained assuming perfect CSI at the BS. Dashed curves take into account channel estimation and calibration errors. Number of antennas at the BS

10% User spectral efficiency [bit/s/Hz]

8 16 24

0.202 0.333 0.609

X% site spectral efficiency [bit/s/Hz/site]

50% 13.03 16.74 25.05

90% 13.99 16.82 25.26

TABLE IV S PECTRAL EFFICIENCY OF 1- SECTOR SITE FOR MU-MIMO PRECODING FOR LOCAL AREA . 36 UT S / SITE . E ACH UT IS EQUIPPED WITH 2 ANTENNAS .

require CSI at the transmitter. In the second case the BS generates the optimum precoding matrices on the uplink and then feedforwards them to the UTs. In Figure 3 we show the complementary cumulative distribution function (CCDF) of the cell throughput of a WINNER system employing MU-MIMO precoding on the downlink. We show the cell throughput of a SISO system as a reference. In the first case we use one ULA with MT = 16 antenna elements. In the second case we use NBS = 4 antenna arrays that are located in the corridors each with MT = 4 antenna elements. Spatial processing at these antenna arrays is coordinated by a central unit. As we can see from the results, MU-MIMO precoding improves the system throughput by more than 5 times relative to the throughput of a SISO system. By using distributed antenna arrays that can cooperate, system throughput is further improved by more than 50%. In Table IV, we give the spectral efficiency of a system employing MUMIMO precoding as a function of the number of antennas at the BS. IV. C ONCLUSIONS AND FUTURE WORK In this paper, the generic WINNER MIMO concept was revisited and assessed. The emphasis is on radio network

system aspects of multi-antenna techniques in various deployment environments. Preferred configurations, or algorithms, for three basic deployment scenarios were given. For wide area, single-stream transmission GoB in a 3-sector site with 4TX antennas provided a throughput gain in the order of 25-40% compared to a single transmit antenna. Using SDMA on top of GoB outperforms single-stream GoB (or so called TDMA based GoB) by 60%. Furthermore, SDMA based on adaptive beams can additionally increase the performance by 10-20% at the expense of requiring dedicated pilots. Finally, higher order sectorisation (HOS) which consisted of using one ULA per sector outperform SDMA at the expense of requiring more antenna elements per site. HOS improves the performance up to a factor of 3.5 for 12-sector sites, yielding 12 b/s/Hz/site. In metropolitan area, for slow moving users, SMMSE MU-MIMO precoding is a suitable scheme yielding almost 9.3 b/s/Hz/site. For higher velocities, the performance of SMMSE MU-MIMO precoding degrades, and PARC or adaptive LDCs might be better choices. For local area, the combination of SDMA and spatial multiplexing achieved by MU-MIMO precoding provides high spectral efficiency, up to 25 b/s/Hz/site. R EFERENCES [1] “Winner, wireless world initiative new radio,” https://www.istwinner.org. [2] IST-2003-507581 WINNER, “D2.7, Assessment of advanced beamforming and MIMO technologies,” Framework Programme 6, Tech. Rep., 2005. [Online]. Available: https://www.ist-winner.org/ DeliverableDocuments/D2.7.pdf [3] M. Dottling et al., “A Multi-User Spatial Domain Link Adaptation Concept for Beyond 3G Systems,” in Personal, Indoor and Mobile Radio Communications, 2005. PIMRC 2005. IEEE 16th International Symposium on, vol. 2, Sept. 2005, pp. 873–877. [4] ——, “Integration of Spatial Processing in the WINNER B3G Air Interface Design,” in Vehicular Technology Conference, 2006. VTC 2006-Spring. IEEE 63rd, vol. 1, 2006, pp. 246– 250. [5] IST-2003-507581 WINNER, “D2.10, final report on identified ri key technologies, system concept, and their assessment,” Framework Programme 6, Tech. Rep., Dec. 2005. [Online]. Available: https: //www.ist-winner.org/DeliverableDocuments/D2.10.pdf [6] G. Dimic and N. C. Sidiropoulos, “On downlink beamforming with greedy user selection: Performance analysis and a simple new algorithm,” IEEE Trans. on Signal Processing, vol. 53, no. 10, pp. 3857– 3868, Oct. 2005. [7] T. Yoo and A. Goldsmith, “On the optimality of multiantenna broadcast scheduling using zero-forcing beamforming,” Selected Areas in Communications, IEEE Journal on, vol. 24, no. 3, pp. 528–541, March 2006. [8] M. Fuchs et al., “Low complexity space-time-frequency scheduling for mimo systems with sdma,” IEEE Transactions on Vehicular Technology, September 2005, submitted to. [9] T. Wild, “Successive user insertion for long-term adaptive beams using short-term cqi,” in InOWo, Hamburg, Germany, August 2006. [10] IST-4-027756, WINNER II, “D3.4.1, The WINNER II Air Interface: Refined Spatial-Temporal Processing Solutions,” Framework Programme 6, Tech. Rep. v1, 2006. [Online]. Available: https://www.ist-winner.org/ WINNER2-Deliverables/ [11] ——, “D6.13.1, WINNER II Test scenarios and calibration cases issue 1,” Framework Programme 6, Tech. Rep. v1, 2006. [Online]. Available: https://www.ist-winner.org/WINNER2-Deliverables/ [12] ETSI, “Selection procedures for the choice of radio transmission technologies of the umts (umts 30.03 version 3.2.0), v 3.2.0,” Tech. Rep. TR 101 112, April 1998.