Chapter 6

Coordinated MultiPoint (CoMP) Systems

Mauro Boldi, Antti Tölli, Magnus Olsson, Eric Hardouin, Tommy Svensson, Federico Boccardi, Lars Thiele and Volker Jungnickel

6.1 Overview of CoMP

Future cellular networks will need to provide high data rate services for a large number of User Equipments (UEs), which requires a high spectral efficiency over the entire cell area. Hence it is important that the radio interface is robust to interference. In particular Inter-Cell Interference (ICI) appears when the same radio resource is reused in different cells in an uncoordinated way. Naturally, ICI particularly degrades the performance of UEs located in the cell-edge areas, which creates a performance discrepancy between cell-edge and inner-cell UEs.

Over the years, several different methods have been studied in order to mitigate ICI. Interference averaging techniques (WINNER-II 2007a) aim at averaging the interference over all UEs, thereby reducing the interference experienced by some UEs. Frequency hopping (Carneheim 1994), which is used in GSM, is a well-known example of an interference averaging technique. Interference avoidance techniques (WINNER-II 2007b), on the other hand, aim at explicitly coordinating and avoiding interference, for example, by setting restrictions on how the radio resources are used. An example of this is Inter-Cell Interference Coordination (ICIC), which is available in the 3GPP Long Term Evolution (LTE) standard (see e.g. Dahlman et al. 2008).

Recently, even tighter interference coordination has gained significant interest under the name Coordinated Multipoint transmission or reception (CoMP). This refers to a system where the transmission and/or reception at multiple, geographically separated antenna sites is dynamically coordinated in order to improve system performance. The idea in CoMP is to proactively manage the interference to improve the performances of UEs, especially for those experiencing poor QoS.

Interest in the coordination among nodes either in transmission or reception has significantly grown in recent years. A comprehensive literature review of the studies that have led to the proposal of CoMP in future systems is given in section 6.1.3, and has been documented in several European Union (EU) research projects (ARTIST4G 2010; WINNER+ 2009a,b).

The adopted acronym to describe this coordinated approach (CoMP) directly drives the attention to the coordination that takes place among the involved entities, but it has to be reported that at the moment many different definitions are given to systems generally considered under the framework of CoMP. In the literature it is also possible to find CoMP systems labeled as “Network MIMO” or “Multicellular MIMO” or “Multicellular cooperation” (Gesbert et al. 2010). The acronym CoMP will be used in this chapter due to its more generic meaning and its adoption in the 3GPP standard development organization (3GPP 2010).

In this section an introduction to CoMP types and the most feasible architectures to implement CoMP will be provided, together with the constraints that have to be taken into account. Closer view to specific CoMP techniques will be given in following sections. The chapter is completed by a description of a practical implementation of CoMP in a trial environment.

6.1.1 CoMP Types

The coordination among multiple entities in radio access networks can occur either as a transmission coordination, namely in the Downlink (DL) of the radio systems, or as a reception coordination, namely in the uplink. As a general view two main families of coordination methods have emerged:

Joint Processing means a coordination in which multiple entities are simultaneously transmitting/receiving to/from UEs located in the coordination area.

Coordinated scheduling or beamforming means a coordination where each UE is communicating with a single transmission/reception point, but the transmission/reception is made with an exchange of control information among the coordinated entities.

Downlink coordinated multipoint transmission implies dynamic coordination among multiple geographically separated transmission points, and the two main families described above can be further detailed as follows:

  • In Joint processing/transmission schemes, data to a single UE is simultaneously transmitted from multiple transmission points in the coordination area, for example, to (coherently or noncoherently1) improve the received signal quality and/or cancel actively interference from transmissions intended for other UEs. This category of schemes put high requirements on the coordination links and the backhaul since the UE data need to be made available at the multiple coordinated transmission points. The amount of control data to be exchanged over the coordination links is also large, for example, channel knowledge and computed transmission weights. Joint processing/transmission is illustrated in the right panel of Figure 6.1. Fundamentally, multiple transmit antennas can be used to improve the DL performance in two ways:

    1. the power of the desired signal received at each UE can be increased, that is, signals transmitted from multiple antennas are formed such that they add constructively at the desired UE;

    2. the interference experienced by each UE can be suppressed, that is, signals transmitted to one UE from multiple antennas are formed such that they add destructively at the other UEs.

    Joint processing/transmission aims to either accomplish one or both of the ways stated above. Note that the former inherently is a single-user transmission technique and can be referred to as single-user CoMP(SU-CoMP), while the latter is inherently a multiuser transmission technique and can be referred to as multiuser CoMP(MU-CoMP).

  • In coordinated scheduling and/or beamforming schemes, data to a single UE is instantaneously transmitted from one of the transmission points in the CoMP set (the set of points/cells that are coordinated), and that scheduling decisions and/or generated beams are coordinated in order to control the interference created. An illustration is provided in the left panel of Figure 6.1. The main advantages of these schemes compared to schemes involving joint processing/transmission (see above) are that the requirements on the coordination links and on the backhaul network are much reduced, because typically

    – only information on scheduling decisions and/or generated beams (and information needed for their generation) need to be coordinated; and

    – the UE data do not need to be made available at the multiple coordinated transmission points, since there is only one serving transmission point for one particular UE.

Figure 6.1 Coordinated beamforming/scheduling (on the left) and JP (on the right) exemplary schemes

img

Similarly, uplink coordinated multipoint reception can be categorized as follows:

  • Joint reception/processing of signals at multiple, geographically separated points. The basic idea is to utilize multiple antennas at geographically separated sites to form a virtual receive array that is used to receive the signal transmitted by each UE, and then to combine and process the signals received at the different reception points. A simple variant of this is already used in 3GPP UTRAN systems and known as soft handover (Holma and Toskala 2004). The main benefits are that energy is collected at several reception points, the obtained macro-diversity gain, and also that advanced combining algorithms can be used in the receive processing in order to cancel out interference. However, as for DL joint processing/transmission described above, this approach puts high requirements on the coordination links since information based on the received signals needs to be exchanged.
  • Coordinated scheduling, with reception of signals at one site only, scheduling decisions can be coordinated among cells to control interference. Again, the main advantage compared to schemes involving joint reception at several reception points (see above) is that the requirements on the coordination links are much reduced because only information on scheduling decisions need to be coordinated.

6.1.2 Architectures and Clustering

The possible introduction of CoMP implies a considerable impact on the architecture of the radio system and the feasibility of CoMP is extensively studied in the research community. Different topologies are generally encompassed under the general definition of CoMP, each of them allowing a different degree of coordination among the involved nodes. In Figure 6.2 the architectures commonly considered when introducing CoMP are schematically represented. The central drawing sketches the most conventional architecture with a set of Base Stations (BSs) that coordinate their resources in order to improve the performances experienced by a set of UEs, in a defined area of cooperation. The fundamental issue in this simple architecture is the inter BS coordination interface. This interface may rely on existing logical interfaces, such as the X2 interface in 3GPP LTE systems, and the physical means adopted to implement this interface are important as well. The deployment of high capacity optical fiber connections in current networks paves the way for a growing adoption of optical fiber in the radio access network as well. Another feasible implementation making use of fiber links in CoMP is again in the Figure 6.2, upper right drawing, where fiber is now used in the front-haul link, following an approach often referred to as Radio over Fiber (RoF). In this scenario a BS Central Unit (CU) is connected by means of RoF links to a set of Radio Remote Units (RRUs) located closer to the UEs, enhances the overall capacity and coverage of a cell. The architecture described herein shows an application of CoMP in a so-called “intracell” or “intrasite” scenario, where the cooperation is performed among nodes belonging to the same cell. The straightforward advantage of an intrasite scenario is the reduction of information needed to be exchanged between the nodes. The case with Relay Node (RN) as cooperating entities is often included in the framework of CoMP architectures (see Figure 6.2, upper left). Depending on the type of RN different CoMP strategies can be applied (see Chapter 7 for more details on RN).

Figure 6.2 CoMP architectures

img

Different time scales are possible for coordination in all the CoMP architectures. The most efficient schemes require the information needed for scheduling to be available at each coordinated BS in the order of a millisecond, which calls for very low-latency information exchanges between cooperating nodes, or between the UE and all the cooperating nodes.

Two extreme approaches can be distinguished regarding how to make this information timely available at distant cooperating nodes: centralized and decentralized cooperation (see Figure 6.3). These approaches are described in the most general case where JP is performed across various nodes but they also apply in the case of coordinated scheduling/beamforming. In a centralized approach the UE estimates the channel information from all the cooperating BSs and feeds it back to a central control unit, where scheduling operations are performed accordingly. This central entity is a logical entity that can be accommodated at one of the collaborating BSs, by establishing a hierarchy between BSs. The need for a central entity entails the following changes upon the conventional architecture of cellular systems: collaborating BSs need to be interconnected via the central entity with low latency links in order to exchange local feedback information. Furthermore this information exchange needs to be coordinated with the use of additional communication protocols. In a decentralized approach, instead, the UE feeds back the channel information to all the cooperating BS. Therefore each BS gathers all the available feedback information, including those related to other BSs. Under this decentralized framework, infrastructure cost and signaling protocol complexity can be minimized, because neither a central entity nor low latency links connecting it with the cooperating BS are required. The main obstacle associated with the decentralized collaborative framework is the handling of errors on the different feedback links; the impact of feedback errors on the system performance is discussed in Papadogiannis et al. (2009), and solutions for enhancing the robustness of decentralized multicell coordination were proposed in WINNER+ (2009a). Two types of solutions are usually introduced: some for reducing malfunction probability, and others for recovering from potential malfunctions, allowing a practical decentralized multicell coordination scheme to be designed. Other alternatives lying between these two extreme approaches are possible where some pieces of information are transmitted to a central entity and others are derived in a decentralized way. For instance, Song et al. (2007) describes a scheme with centralized scheduling but where the precoding weights are locally designed at each BS. Therefore, the amount of information to be exchanged between the collaborating nodes is reduced compared to the fully centralized approach.

Figure 6.3 Centralized and decentralized architecture

img

In any case, whatever the chosen architecture, studies have been performed on how the distributions of UEs in a given ideal cellular layout could improve the performance with respect to an uncoordinated case; as the number of UEs and nodes increase, the signalling overhead required for the interbase information exchange and the amount of feedback needed from the UEs also increase. Therefore, cooperation should be restricted to a limited number of nodes. To achieve this goal, the network is thus divided into clusters of cooperative cells. Cluster selection is considered a key issue to cooperation algorithms performance and has been widely studied in the literature (description and definitions can be found in in WINNER+ (2009b) and in the Third Generation Partnership Project (3GPP) Study Item on CoMP 3GPP (2010). Basically a cluster of cells for CoMP transmission can be formed in a UE-centric, network-centric or a hybrid fashion:

UE centric clustering: each UE chooses a small number of cells that are most suitable for cooperative transmission. In this case UE scheduling is challenging since the clusters for different UEs are chosen in a dynamic way and may overlap.

Network centric clustering: clusters are defined statically for all UEs of a given serving cell based on the neighborhood and try to combine the cells with strongest mutual interference. The performance for the UEs can be compromised at the cluster boundary.

Hybrid clustering: the network can predefine a set of clusters for a given area and the selection of the best cluster is assisted by the UE through feedback information.

Another relevant classification is related to dynamic or static formation of clusters for CoMP: if a cluster is statically formed, no fine tuning is allowed depending on the UE's actual channel and interference conditions, while a dynamic cluster formation can cope with them according to dedicated network algorithms aimed to the evolution of clusters. An ideal example of dynamic clusters is shown in Figure 6.4 where, as an example, three clusters of six cells are modified in the two CoMP scenarios presented. It is worth noting that the clusters could also be overlapped in some possible schemes; many works on dynamic or “semi-static” clustering approaches are now available in the literature; some examples can be found in Brueck et al. (2010), Papadogiannis et al. (2008b) and in Xiao et al. (2010) where thresholds are introduced to dynamically select the cells of the cluster.

Figure 6.4 Example of dynamic clusters in a CoMP scenario

img

6.1.3 Theoretical Performance Limits and Implementation Constraints

In this section the concept of CoMP is presented in the various steps from the introduction of simple relay-based approaches to recent JP schemes. This helps to explain the important enhancements that have been reached and at the same time indicates the main implementation constraints that have still to be overcome in this field.

Since the introduction of the relaying concept (Cover and Gamal 1979; Cover and Thomas 1991; van der Meulen 1977), network node cooperation and coordination has received significant attention in the context of improving the capacity or coverage of wireless communications (see Chapter 7 on Relaying for more details). In the classic relaying problem, UEs may cooperate in routing or relaying each others' data packets. A large number of papers have addressed the relaying problem with different assumptions on the transceiver capabilities. See, for example, (Cover and Gamal 1979; Cover and Thomas 1991; Gupta and Kumar 2000; Kramer et al. 2005; van der Meulen 1977) and the references therein.

The UE cooperation diversity builds upon the classical relay channel model (Cover and Gamal 1979; Cover and Thomas 1991; van der Meulen 1977), where the spatial diversity gains can be achieved both at the transmitters and the receivers using a collection of distributed antennas belonging to multiple nodes. Each node is responsible for transmitting not only its information, but also the information of the other nodes it receives and detects. Thus, spatial diversity is achieved through the joint use of the antennas of all the nodes. Sendonaris et al. (2003a,b) considered the beamforming approach where different nodes adjust their transmissions based on Channel State Information (CSI) knowledge so that the transmitted signals add up coherently at the destination. On the other hand, Laneman and Wornell (2003) and Laneman et al. (2004) assumed no CSI at the transmitters and proposed a variety of low-complexity, cooperative diversity protocols that enable wireless nodes to fully exploit spatial diversity in the channel. However, the cooperation is complicated due to the fact that the (wireless) channel between the nodes is noisy (Sendonaris et al. 2003a,b). Therefore, the UE cooperation often increases the interference level, the protocol overhead and the complexity of the transceivers (Bletsas and Lippman 2006; Sendonaris et al. 2003a).

In the ideal case with a noiseless link between the receiving or the transmitting nodes, the node cooperation is simplified to MIMO MAC or MIMO BC with per-node power constraints, respectively (Sendonaris et al. 2003a). In cellular networks, this can be achieved for example by a wired backbone connection between the distributed BS antenna heads, or by highly directional wireless microwave links (Zhang et al. 2005).

From Wyner Model to Joint Processing

Network infrastructure based coordinated processing across distributed BS antenna heads has received significant interest in the recent literature (Aktas et al. 2006; Bletsas and Lippman 2006; Choi and Andrews 2007; Jungnickel et al. 2009a; Karakayali et al. 2006b,c; Shamai and Zaidel 2001; Somekh et al. 2006; Wyner 1994; Zhang and Dai 2004a; Zhang et al. 2005). Wyner (1994) considered the Uplink (UL) of a cellular network with JP at the centralized controller, which has access to all the received signals at multiple receivers, and optimally decodes all the transmitted data in the entire network. In this simplistic model, each cell senses only the signal radiated from a limited number of neighboring cells, which yields closed-form expressions for the achievable rates and allows, to a certain extent, the analytical treatment of the distributed antenna systems with joint processing. It was shown in inWyner (1994) that a cellular network with such a JP receiver significantly outperforms a traditional network with individual processing per BS. A JP receiver for the UL was further considered in Aktas et al. (2006), Jafar et al. (2004), Liang et al. (2006) and Somekh et al. (2007).

Shamai and Zaidel (2001) were among the first to consider the DL sum rate and spectral efficiency optimization for coordinated Multiple-Input Multiple-Output (MIMO) systems with perfect data cooperation between BSs. They applied the ZF-DPC to the multicell JP in Wyner's scenario (Wyner 1994) with an average system power constraint and showed significant capacity enhancements from BS coordination. For a single-cell scenario, the power constraints are generally imposed on the total power radiated by all the elements of the array. Conversely, per antenna power constraints have to be enforced for distributed antenna systems, as each antenna head is provided with a separate power amplifier. In practice, each antenna element may also have a separate power amplifier.

Jafar and Goldsmith (2002) and Jafar et al. (2004) considered a multicell DL channel with perfect CSI at both ends, where an individual power constraint per BS is imposed. They proposed a heuristic but efficient suboptimal method based on iterative water-filling, which aimed at maximizing the sum-rate throughput of the coordinated system while meeting the individual BS power constraints. The sum capacity and the entire capacity region of the MIMO DL with per antenna or per BS power constraints were discovered in Yu (2006), Weingarten et al. (2006) and Yu and Lan (2007), respectively. These important findings can be utilized for finding the maximum achievable user rates from the rate region of a coordinated cellular MIMO system with practical peak power constraints per antenna or per BS.

Inspired by the pioneering work of Shamai, Jafar et al., BS cooperation has been studied by several other authors (Karakayali et al. 2006b,c; Somekh et al. 2006, 2007; Zhang and Dai 2004a). Somekh et al. (2006, 2007) provided an information theory analysis of distributed antenna systems under the circular Wyner model (Wyner 1994), and derived bounds for the sum-rate capacity supported by the multicell DL under per BS power constraints. Karakayali et al. (2006b,c) studied the coordinated cellular DL using different ZF transmission schemes. In Karakayali et al. (2006b,c), the symmetric (or common) rate maximization in the coordinated cellular DL with Zero Forcing (ZF) transmission subject to per BS power constraints was formulated as a convex optimization problem, which can be efficiently solved.

Assuming linear transmitter processing, a coordinated antenna system with N antennas is ideally able to accommodate up to N streams without becoming interference limited. Both the inter-UE and intercell interference can be controlled or even completely eliminated by a proper precoder selection. This is especially true in the coherent multicell MIMO case, where UE data is conveyed frommultiple antenna heads over a large virtual MIMO channel (Karakayali et al. 2006b).

CoMP Implementation Constraints

The coherent multiuser, multicell precoding techniques, however, have high requirements in terms of signaling and measurements as mentioned earlier. In addition to the complete channel knowledge of all jointly processed links, a tight synchronization across the transmitting nodes and centralized entities performing scheduling and computation of joint precoding weights is required in order to avoid carrier phase drifting at different transmit nodes. The theoretical studies referred above mostly assume perfect and complete CSI for all the transmitters, which is difficult to accomplish in practical cellular networks. In practical TDD MIMO-OFDM cellular systems, for example, UL transmissions from adjacent cells can be significantly more attenuated compared to the own cell UEs. Therefore, the joint channel estimation may be difficult, if not impossible, to implement in practice. In order to reduce the required information exchange between the network nodes, the JP of the transmitted signal from several MIMO BS antenna heads can be restricted to the cell edge region, where the available gains from the JP are the most beneficial (Tölli et al. 2008).

In order to perform joint transmission from all the distributed BS antenna heads, the baseband signals must have a common carrier phase reference. The Radio Frequency (RF) impairments and the impact of the propagation delay from each of the transmitters to the intended UE must be compensated for at the transmitter, for example, by using some feedback from UE. The BS antenna heads cannot fully synchronize the desired and the interfering signals received by different UEs due to different propagation times between the BS antenna heads and the UEs. Zhang et al. (2007) showed that significant performance degradation may follow if the asynchronous nature of the multiuser interference is not taken into account when designing the precoder for the coordinated DL. In MIMO-OFDM systems, however, this problem can be handled efficiently as long as the received signal paths are within the guard interval. Specifically, the increase of the delay spread is not necessarily so dramatic if the coordinated processing is limited to the cell edge region.

The general constraints that have to be taken into account when considering a possible and feasible introduction of CoMP in a radio access system are summarized in Table 6.1, see also (WINNER+ 2009c, p. 28).

Table 6.1 CoMP implementation constraints.

Constraint Main issue
Backhauling The amount of data to be transmitted over the backhaul depends on the chosen CoMP strategy, that is, whether the cooperation is on control plane only (e.g. in case of coordinated scheduling for interference avoidance) or control plane and user plane (for joint transmission from different sites).
Pilot design The difficulty in obtaining the precise DL CSI imposes requirements on the pilot design to enable the channel estimation with sufficient quality and to be able to separate the pilots from different cells. In addition, the UE in some approaches needs to be able to decode a control channel of the neighboring cells. In order to attain the full CSI between all UEs and BS antennas in the cellular network, the UE channels should be jointly estimated at each antenna head.
Feedback design Time Division Duplex (TDD) systems can exploit channel reciprocity to attain most of the required CSI at the transmitters. In FDD systems, the design of suitable feedback channels is an important issue. The information to be exchanged over the feedback link can include, for example, short-term or long-term CSI, preferred precoding matrix indices, received power from all the nodes, long term fading, etc.
Synchronization A tight frequency synchronization, carrier phase synchronization across the transmitting nodes and complete CSI of all jointly processed links is required in coherent approaches. Noncoherent approaches are less demanding, but they could still need centralized resource management mechanisms. Time synchronization is required for all CoMP techniques, with looser requirements for coordinated beamforming than for JP.

6.2 CoMP in the Standardization Bodies

CoMP has been studied widely in 3GPP and IEEE during the feasibility studies of the LTE-Advanced and 802.16m WiMAX, respectively. In this section the focus will be on the standardization activity performed in the framework of LTE-Advanced.

CoMP was one of the main techniques investigated during the 3GPP LTE-Advanced feasibility study (called Study Item in 3GPP) between May 2009 and March 2010. In the end, no consensus was achieved on the maturity of the CoMP technology for being standardized in 3GPP (Rel-10). Nevertheless, several decisions were made on the design of a CoMP functionality for LTE that will be able to serve as a basis for future specification works. Indeed, CoMP will continue to be studied in 3GPP for potential inclusion in future releases. It is worth noting that even though (Rel-10) does not provide any specific support for CoMP, some CoMP schemes can be implemented in a proprietary manner in LTE Rel-10 networks.

This section is organized in two main parts: section 6.2.1 gives an overview of the work carried out on CoMP in 3GPP for LTE-Advanced, and section 6.2.2 describes the main decisions that have been taken on the design of the CoMP functionality.

6.2.1 Overview of CoMP Studies

Both UL and DL CoMP techniques were studied in 3GPP. Most of the attention was devoted to DL techniques because UL techniques were expected to have only a limited impact on the physical layer specifications. The main reason for this situation is that LTE (Rel-8) already allows the UE transmission to be demodulated at several eNodeBs (eNBs) (i.e. BSs). How the eNBs exchange the information to perform the coordination is up to the upper layers (typically the Medium Access Control (MAC) and Radio Resource Control (RRC) layers), which were not part of the initial studies on CoMP.

Nevertheless, some CoMP techniques were proposed for the UL and two families of techniques were identified: joint reception of the transmitted signal at multiple reception points and/or coordinated scheduling decisions among cells to control interference (3GPP 2010). In addition, UL coordinated link adaptation was discussed (Deutsche Telekom AG and Vodafone 2009).

In the DL, coordinated beamforming/scheduling and JP were both studied. For coordinated beamforming, the best companion approach (see also section 6.5.3) and ZF-based precoding approaches received particular attention. Within JP, two subcategories were identified: joint transmission, where the data to a single UE is simultaneously transmitted from multiple transmission points, and dynamic cell selection, where a single transmission point is active at a time within a set of cooperating points (3GPP 2010). For joint transmission, various techniques were considered, assuming different levels of coordination. Many of the proposed DL techniques were assumed to work in a MU-MIMO fashion, that is, where several UEs could be served on the same resources across the same cell or a set of cooperating cells.

As already stated from a general standpoint in section 6.1, for both UL and DL two types of coordination are considered in 3GPP for LTE-Advanced: inter-eNB and intra-eNB coordination. Inter-eNB CoMP in a multivendor environment is more complicated to standardize because standardized interfaces have to be used to exchange the necessary information between the coordinated eNBs. In addition, depending on the used backhaul technology, a non-negligible delay can affect the information availability, and inter-eNB CoMP techniques need to be robust to this. The delay in the information exchange was considered in the 3GPP work on CoMP. At the end of the LTE-Advanced Study Item, it was decided to restrict potential CoMP solutions for Rel-10 to intra-eNB CoMP, in order to reduce the standardization effort. It is worth noting that the absence of a standardized interface would not mean that inter-eNB CoMP is not possible; however, proprietary solutions would then have to be used, which would only be applicable in areas where the BSs are supplied by the same equipment vendor. It is useful to note that the UE does not know whether cells belong to the same or different eNBs, so it cannot distinguish between intra- and inter-eNB CoMP operation.

Even though standard support was not agreed for CoMP in Rel-10, the discussions in 3GPP have led to structuring decisions on the design of a CoMP functionality, which are described in section 6.2.2. Both coordinated beamforming/scheduling and JP are supported in the agreed framework. The main remaining areas to be addressed to complete the definition of the CoMP concept are the measurements needed at the UE and the feedback design principles.

By the end of the work on CoMP in 3GPP, an evaluation campaign evaluated the benefits of CoMP against Multi-User (MU)-MIMO. A majority of earlier performance evaluations had shown that either MU-MIMO or CoMP were necessary to meet the International Mobile Telecommunications Advanced (IMT-Advanced) spectral efficiency requirements (3GPP 2010). The results were not consistent enough between the companies to make conclusions. In fact some companies showed gains, whereas others found no gain but loss. There were also disagreements on the realism of the simulations depending on the overhead assumptions. However, it should be noted that the MU-MIMO versus CoMP comparison was undertaken with the full buffer traffic model only, most companies simulating a fairly high number of UEs per cell. In reality, the traffic is generally rather bursty, so that pairing UEs to serve them on the same resource is more difficult.

Overall, the CoMP technology was not recognized as mature enough to be standardized. Therefore, it was agreed that there would be no standard support for CoMP in Rel-10. One exception to this rule was the design of Reference Signal (RS) targeting CSI estimation (called CSI-RS), which takes into account CoMP needs for future proof (see section 6.2.2). Despite the lack of specific standard support, CoMP techniques may be feasible by proprietary solutions, for example, by taking advantage of the (long-term for Frequency Division Duplex (FDD)) UL-DL channel reciprocity property (Ericsson and ST-Ericsson 2010) to acquire the necessary multicell CSI, in particular when limited to intra-eNB coordination. Studies on CoMP continue within 3GPP in 2011, through a dedicated CoMP Study Item.

6.2.2 Design Choices for a CoMP Functionality

As indicated earlier, UL CoMP was expected to have only a limited impact on the physical layer specifications, and was therefore allocated only a limited attention. As a consequence, all the decisions specific to the CoMP design addressed the DL. The decisions addressed the design areas of reference signals, control signalling design and feedback. These decisions are explained in the following.

Reference Signals

Reference signals (RSs) are a key element of the system design, because they determine the performance of the channel estimation. Channel knowledge is key to allow coherent demodulation at the receiver, and CSI allows advanced MIMO transmitter processing when known at the BS. Compared to single-cell processing, DL CoMP requires channel estimates to be obtained from several cells, which constrains the ability to receive the RS from multiple cells.

LTE Rel-10 has defined a new framework for RSs. Even though most of the related design was not motivated by CoMP, some of the RSs' characteristics are useful for CoMP operation, so it is useful to briefly recall them here. Two types of reference signals are introduced in Rel-10: CSI-RSs are cell-specific and target CSI estimation (to be subsequently fed back to the transmitter), whereas DM RSs are UE-specific and target the demodulation of the scheduled transmissions. The split of the RS types according to the RS functionality was motivated by two objectives: one is to limit the overhead for a high number of transmit antennas (up to eight in Rel-10). Therefore, the CSI-RSs are transmitted sparsely in time and frequency to allow CSI to be acquired on the whole bandwidth with affordable overhead. On the other hand, the DM-RSs are more dense and only transmitted on the resources assigned to the UE. This allows to fine tune the tradeoff between overheads and CSI estimation accuracy compared to a fixed RS design fulfilling both demodulation and CSI estimation constraints. The other objective was to allow powerful transmitter processing at the BS. To this end, the DM-RSs are precoded with the same precoding as the data. Precoded RSs suppress the need to signal the precoding weights (necessarily selected from a predefined set) to the UE, which gives total freedom to the transmitter to apply any type of precoding. Indeed, the UE directly estimates the composite channel formed by the precoding operation and the actual propagation channel. One example of powerful transmitter processing enabled by this approach is ZF precoding. It is therefore useful to remark that the specifications will not mandate a single particular precoding method, but rather will allow a family of methods.

One immediate consequence of the DM-RSs adoption on the CoMP functionality design is that the UE does not need to know what transmission points or cells are involved in the transmission: provided the coordinated cells use the same DM-RS sequence to serve a given UE, the UE will only see a unique channel, formed by the sum of the channels to the cells participating in the transmission. This enables the multicell transmission as well as transmission points to be dynamically switched on/off without any signaling to inform the UE.

In contrast, the CSI-RSs are cell-specific and not precoded in order to allow the UE to estimate and report the CSI related to various neighboring cells. To ensure proper CoMP operation, the CSI-RS need to allow accurate estimation of multiple cells' channels. Even though CoMP is not supported in Rel-10, the CSI-RS design needs to account for this need in Rel-10, because such a design is difficult to change in the future without creating backward compatibility problems. It was therefore agreed that the CSI-RS design in Rel-10 will take into account the CoMP needs, and should in particular allow accurate multi-cell measurements. As a result, Rel-10 provisions the possibility to mute data resource elements (i.e. one subcarrier from one OFDM symbol) at one cell to protect the CSI-RSs transmission on the same resources at a neighboring cell.

Control Signalling

The UE will receive its control channel (called PDCCH in LTE), which carries in particular the scheduling information, from a single cell. This cell is called the serving cell, and is the cell the UE is served by in the case of a single-cell transmission. This choice limits the UE complexity by avoiding the need to monitor the control channels from several cells. In addition, this enforces the principle that the UE is only aware of its serving cell, the other cells potentially participating in the transmission remaining transparent.

CSI Feedback

Feedback is the main remaining open issue for CoMP standardization. Although the complete design was not agreed, a number of decisionswere taken. The feedbackwas an important design issue in Rel-10, since new feedback mechanisms were discussed for SU- and MU-MIMO in addition to CoMP. Therefore, the discussion on feedback for CoMP was embedded into a global feedback framework, and can not be considered isolated.

One design principle for feedback in Rel-10 is scalability, so that the same feedback allows the network to select dynamically between SU-MIMO and MU-MIMO. More specifically to CoMP, the UE feedback supporting CoMP transmission should be such that it also enables the network to switch dynamically to single-point transmission. To facilitate this switch, it was agreed that the UE reception and demodulation of CoMP transmissions is the same as for single-cell SU/MU-MIMO. Furthermore, if a new form of feedback was needed for CoMP, scalable feedback for different CoMP categories (e.g. coordinated beamforming and JP) should be aimed at, which means that feedback in support of CoMP JP should be a superset of the feedback in support of CoMP coordinated beamforming/scheduling.

Two main types of feedback have been identified: the explicit feedback involves reporting the raw channel as observed by the receiver (together with interference information), without assuming any transmission or receiver processing. Examples of explicit feedback include reporting the channel impulse response, or the channel covariancematrix. In contrast, the implicit feedback involves reporting channel-related characteristics assuming hypotheses on a particular transmission and/or reception processing; one typical example of implicit feedback is the Channel Quality Indicator (CQI)/PMI/RI feedback used in Rel-8. In addition, UE transmission of UL Sounding Reference Signal (SRS) was identified as being usable for CSI estimation at multiple cells, in particular via the exploitation of the channel reciprocity property (both for FDD and TDD). More details on the different feedback alternatives within each type can be found in (3GPP 2010). Note that the framework finally adopted for single cellMIMO was the implicit one. In principle, this does not preclude explicit feedback to be adopted for CoMP in the future, but the constraint of easy switch between single-cell and multicell operation tends to render implicit feedback easier to adopt also for CoMP.

Several decisions have been made regarding how to report the CoMP-specific information. When an eNB-to-eNB communication interface (called X2 for LTE) is available and is adequate for CoMP operation in terms of latency and capacity, the starting-point solution is that UE CoMP feedback targets the serving cell. In this case, the reception of UE reports at cells other than the serving cell is possible but is left to implementation. For cases where X2 interface is not available or not adequate (with respect to latency and capacity), the feedback reporting needs further consideration. It was also agreed that the starting point for schemes that need feedback is individual per-cell feedback (i.e. no differential CSI information between two cells). Nevertheless, it was noted that complementary intercell feedback might be needed. In addition, the starting point is that the feedback related to each individual cooperating cell is transmitted only to the serving cell. For the purpose of measurements, the CoMP measurement set was defined as the set of cells about which the UE performs CSI measurement; the cells effectively reported are possibly a subset of the CoMP measurement set.

The feedback design for CoMP is still an open discussion topic in 3GPP. The choice of a feedback strategy, is in particular between explicit and implicit feedback, or a combination of both, in order to take a tradeoff between the UL overhead and the DL performance. Nevertheless, although it was not completed in Rel-10, the standardization of CoMP has progressed substantially. For an overview of the latest CoMP views within 3GPP, the reader can be referred to 3GPP (2010).

6.3 Generic System Model for Downlink CoMP

In order to have a common framework for the CoMP schemes a system model is briefly outlined in this section. A cellular Orthogonal Frequency Division Multiplexing (OFDM) DL is considered where a center site is surrounded by multiple tiers of sites. Each site is partitioned into three 120° sectors, that is, a set img consisting of L sectors (also called a cell or a BS) in total. Consider that each BS has NT transmit antennas and the UE m is equipped with img antennas. A set img with size img includes all UEs active at the given time instant, while a subset img includes the UEs allocated to BS l. Each UE m can be served by img BSs which define the cluster set img for the UE m, and img. The signal img received by the UE m consists of the desired signal, intracluster and intercluster interference, and can be expressed as

(6.1) equation

where the vector img is the transmitted signal from the l'th BS to UE m, img represents the additive noise sample vector with noise power density img, img is the inter-cluster interference plus noise, and img is the channel matrix from BS l to UE m. The transmitted vector for UE m is generated at BS l as

(6.2)equation

where img is the pre-coding matrix, img is the vector of normalized complex data symbols, and img denotes the number of active data streams within the cluster set img. Here, img is the data symbol on the s'th data stream designated to UE m. It should be noted that the first term in Equation (6.1) includes interstream interference.

6.3.1 SINR for Linear Transmissions

We focus on linear transmission schemes, where the L BS transmitters send altogether img independent data streams. Per data stream processing is considered, where for each data stream img the scheduler unit associates an intended UE img, with the channel matrices img, img.

Let img (a column of img) and img be arbitrary transmit and receive beamformers for the stream s. Hence the achievable data rate of the stream s intended to the UE img is given by img, where img is the DL Signal-to-Interference-plus-Noise Ratio (SINR) of the data stream s and can be expressed as

(6.3)equation

Note that in case of nonlinear receivers, such as the Successive Interference Cancellation receiver, Equation 6.3 is not valid.

6.3.2 Compact Matricial Model

For simplicity, let us designate K the cluster size and assume it is not UE specific. Hence, BSs are grouped in subsets of img clusters. As as an example, Figure 6.4 depicts the case of img clusters where each of them contains img sectors. Further, let NR and NS be the number of receive antennas and maximum number of data streams per UE, respectively. Let img be the number of active UEs in a cluster.

Let img be the img precoding matrix of a defined cluster for UE m. Moreover, let img be the img channel matrix between all the transmit antennas of a cluster and the NR receive antennas of UE m. Further, define the transmit data vector, transmit precoding matrix, channel matrix, inter-cluster interference plus noise vector, and receive data vector, respectively as follows:

(6.4)equation

(6.5)equation

(6.6)equation

(6.7)equation

(6.8)equation

Finally, the receive data vector, on a per subcarrier basis for a cluster, can be written in the following compact matrix form:

(6.9)equation

In this model, the precoding matrix W inherently carries the transmit powers of each data stream. Another common practice is to decompose the precoding matrix into a product of two matrices, where the first one includes the spatial transmit directions as unit-norm column vectors, and the second one is a diagonal matrix that indicates the power weighting of each stream.

6.4 Joint Processing Techniques

Motivated by the predicted theoretical gains, joint transmission or JP between BSs has been identified as one of the key techniques for mitigating intercell interference in future broadband communication systems (Karakayali et al. 2006a). As discussed in section 6.2 JP techniques are included in the more general framework of CoMP transmission schemes within the 3GPP standardization activities concerning LTE-A (3GPP 2010).

As discussed earlier, from a practical point of view one of the major drawbacks related to JP is its high complexity, that is, large backhaul and signaling overhead. To reduce the complexity, clustering solutions that restrict JP techniques to a limited number of BSs have been proposed. In these approaches, the network is statically or dynamically divided into clusters of cells (Boccardi and Huang 2007a; Papadogiannis et al. 2008c; Thiele et al. 2009). Moreover, the cluster formation may be performed and optimized by a central entity (network-centric), or in a per-UE way (user-centric). In the following, JP is only allowed between BSs belonging to the same cluster, whereas BSs belonging to different clusters are not coordinated. The clusters are considered disjoint, that is, a given BS cannot belong to more than one cluster.

In this section, some of the main recent achievements on JP for CoMP are presented. Section 6.4.1 complements the state-of-the art overview of JP CoMP in the downlink. In section 6.4.2 results from a system level performance evaluation of coherent JP CoMP are shown and compared with a baseline LTE Rel-8 system configuration. Section 6.4.3 investigates possible dynamic JP approaches for user-centric, and BS-centric based clustering. Finally, in section 6.4.4, an insight about uplink JP CoMP is given.

6.4.1 State of the Art

A comprehensive description of the state of the art of CoMP has been presented in section 6.1.3. In this section some further references are given for some JP-specific aspect of CoMP.

An interesting topic extensively investigated in the literature is the case of JP with BSs equipped with antenna arrays. Motivated by current antenna array designs where a separate amplifier is provided for each antenna, Boccardi and Huang (2006) and Wei and Lan (2007) consider the transmitter optimization problem subject to per-antenna power constraints. These results can be extended to the case when a single antenna array is partitioned into antenna groups and a per-group power constraint is imposed, which is equivalent to a fully coordinated multi-cell system with per-BS power constraints. The power minimization problem for CoMP systems was addressed in Botella et al. (2008) and Dahrouj and Wei (2008). The main contribution of (Dahrouj and Wei 2008) is a practical algorithm that is capable of finding the jointly optimal beamformers for all BSs globally and efficiently.

Recent works about CoMP systems can be divided into two main areas of research: on one hand, some authors give practical insights for the implementation of coordinated multicell systems (Hongyuan et al. 2008; Papadopoulos and Sundberg 2008; Zarikoff and Cavers 2008) and, on the other hand, others addressed the increased complexity of implementing multiuser MIMO schemes and the large signaling overhead associated with these systems (Boccardi and Huang 2007a; Papadogiannis et al. 2008a,c; Skjevling et al. 2008; Venkatesan 2007).

In a CoMP system, the signal arriving at the UE experiences different propagation delays from each BS, that is, it is fundamentally asynchronous. Most previous studies simplified the system model by assuming synchronous reception at the UE. In Hongyuan et al. (2008), a mathematical model for this synchronicity is proposed, whereas in Botella et al. (2008) a spatial equalizer is assumed at the receiver of each UE. On the other hand, perfect CSI at the BSs is one of the requirements for implementing transmit beamforming from the set of all coordinated antennas. In Zarikoff and Cavers (2008), the estimation of the carrier frequency offset is addressed using training sequences taking into account the possible loss of orthogonality. Finally, asynchrony-resilient space-time codes are proposed in Papadopoulos and Sundberg (2008) for coordinated multicell systems.

Another relevant topic in this field, as stated in section 6.1, is the adoption of clusters, formed either in a static, dynamic or hybrid way. Some recent achievements concerning clustering are reported in section 6.1.2. The concept of clustering is not actually restricted to JP CoMP schemes only, and an important implementation of clustering in a system encompassing both joint processing in an inner cluster and coordinated processing in an outer cluster can be found in Björnson et al. (in press).

6.4.2 Potential of Joint Processing

A system-level performance evaluation of JP CoMP is presented here in order to estimate the possible expected benefits. Coherent joint transmission CoMP based on traditional ZF precoding (Caire and Shamai 2003) is compared with a baseline LTE Rel-8 system configuration. Then the goal is to show the potential of JP CoMP.

The overall assumptions (WINNER+ 2009a) follow to a large extent the deployment scenario and system parameters as specified for the 3GPP case 1 scenario, which is an urban macro scenario defined in 3GPP 2006. Each sector is equipped with an antenna array comprising four elements separated by ten wavelengths. CoMP transmission is carried out over a number of traditional cell sites, which are controlled by one eNB. The underlying assumption is that the coordinated cell sites are connected to the eNB via high-speed connections in order to allow for the fast coordination. Clusters of nine and 21 cells are taken into consideration.

The UEs have two antenna elements each separated by half a wavelength and employ ideal Minimum Mean Square Error (MMSE) receivers with additional Successive Interference Cancellation (SIC) functionality between the two streams of own UE. The DL transmission scheme is 4×2MIMO on all links, and as reference case LTE Rel-8 with fast codebook switching (Dahlman et al. 2008) is simulated. The simulations assume perfect channel and interference estimation at the UE. Furthermore, all UEs are assumed to have full buffers, and the average number of UEs per cell is varied from 0.1 to five. The UEs are uniformly distributed across the simulated area, and each UE moves at thespeed of 3 km/h and is assumed to be indoors (modeled with a 20 dB additional path loss).

Figure 6.5 presents the average throughput versus 5th percentile UE bit rate for coordination over nine cells and 21 cells, compared to LTE Rel-8, with the UE density varied along the curves. Overall it can be seen that the JP based on ZF provides a significant gain both in terms of average system throughput as well as in cell-edge UE throughput (5th percentile UE bit rate). It can be seen that the performance is clearly improved when the coordination is carried out over 21 cells compared to nine cells (also for less significant impact of edge effects on the borders). This is explained by the fact that in the case with 21 cells a larger effective antenna array is considered, and hence higher degrees of freedom can be exploited.

Figure 6.5 Simulation results for ZF based JP CoMP over 9 and 21 cells compared to LTE Rel-8

img

6.4.3 Dynamic Joint Processing

Motivated by the need to reduce the complexity of JP with full cooperation, in this section JP CoMP within a cluster area with a dynamic utilization of the involved BSs is taken into consideration.

Partial Joint Processing: a User-Centric Clustering

In the Partial Joint Processing (Partial Joint Processing (PJP)) scheme presented here, the UE receives its data from a dynamically selected subset of the K BSs or from an active set (Botella et al. 2008). From the system point of view, three benefits are provided adopting PJP with respect to more conventional JP methods: reduced feedback from the UEs, lower interbase information exchange and transmit power saving. However, the PJP scheme introduces multiuser interference in the system, because less CSI is available at the CU to design the linear precoding matrix W.

In order to define the subset of BSs transmitting to a given UE, assume that the UE is assigned to a master BS, which is the one with the highest channel gain. The UE estimates the average channel gain fromthe remaining BSs in the cluster, K−1, and compares it to the channel gain from the master BS. Base stations are included in the active set only if their channel gains are above a relative threshold with respect to the master BS. By doing so, BSs related to poor quality channels do not transmit to the UE and the cluster becomes partially coordinated. The threshold value is specified by the cluster management, and different degrees of JP can be obtained by modifying its value. Therefore, the PJP scheme includes as a particular case full cooperation as defined in Equation 6.9, here denoted as Centralized Joint Processing (CJP).

Both the CJP and the PJP approaches imply the need of a CU to design the precoding scheme. For comparison purposes, a Decentralized Joint Processing (DJP) scheme is considered, where only local CSI is available at each BS and the power allocation and the precoding are locally implemented at each BS. However, the UE may receive its data from several BSs, depending on its given channel conditions. Hence, the cardinality of the set of spatially separated UEs that can be served by each BS in the cluster is reduced to img, and a multibase scheduling algorithm is required in order to assign UEs to BSs. Here, the multibase scheduling problem is solved allowing each BS in the cluster to serve its set of img UEs. As shown in Skjevling et al. (2008), with this solution, each of the M UEs can be served by a number of BSs that ranges from zero to K. Hence, the DJP scheme implies that a certain number of UEs in the cluster may remain without service.

To evaluate the PJP scheme, single-antenna UEs (img) are considered, whereas a ZF precoder is jointly designed by the BSs, img. The maximum available transmit power at each BS is restricted to a img value, and hence, under an equal received UE power allocation assumption, the final precoder matrix can be written as img, where img, in which img are the rows of matrix img related to the kth BS (Zhang and Dai 2004b).

The Quality of Service (QoS) experienced by a UE should preferably not be location dependent – that is, the JP scheme should provide a uniform performance over the cluster area. For this reason, we characterize and compare the performance of the CJP, PJP and DJP schemes over the whole cluster area. A cluster is considered of img BSs, each one equipped with an array of img antennas, and img single-antenna UEs. The cluster radius is img m. The channel vector between the mth UE and the kth BS is modeled as img, where the shadow fading is a random variable described by a log-normal distribution, img, the pathloss follows the 3GPP Long Term Evolution Long Term Evolution (LTE) model, img, and img includes the small-scale fading coefficients, which are i.i.d. complex Gaussian values according to img(WINNER+ 2009b).

In this case, img UEs are placed in each position. The PJP plots stand for active set threshold values of 10, 20 or 40 dB, respectively. For low mobility UEs, the backhaul overhead related to exchanging the UE data between the BSs is higher than that required for exchanging the channel coefficients. Then, the combined amount of backhaul exchange and feedback from the UEs can be roughly estimated by means of the average number of BSs that are transmitting to a UE, img. In the CJP and DJP schemes, this parameter remains fixed regardless of the location of the UEs in the cluster area. However, for the PJP scheme, img depends both on the active set threshold value and the UE position over the cluster area. This is shown in Figure 6.6, where the average number of BSs used along a line from one BS towards the far most point in the cluster is plotted. The median Throughput (TP) gain of CJP, DJP and PJP is shown in Table 6.2. As it can be seen from the table and from Figure 6.6 the active set threshold value is a suitable parameter to define the PJP scheme, in order to trade-off the cooperation gain versus the backhaul load to locations in the cluster area where the gain is substantial. Additional results are included in Appendix C.1.1.

Table 6.2 Gains with respect to 1BS case and at Distance/R = 1.

img

Figure 6.6 Average number of BSs assuming PJP transmitting to each UE, imgBS, for different threshold values versus normalized distance. M = 6 UE and an edge-of-cell SNR of 15 dB (Thiele et al. 2010a)

img

Distributed Joint Processing: a User-Centric Clustering with Multi-antenna Receivers

This section presents a distributed JP scheme, where K BSs perform the DL CoMP transmission to M UEs each equipped with img receive antennas. In order to use the same ZF beamformer, appropriate virtual receive antennas out of the img antennas have to be selected. At maximum, the cluster can provide img coherently transmitted data streams.

A total number of img Multiple-Input Single-Output (MISO) channels, selected from a sufficiently large set of UEs, are composed to form a compound MIMO channel matrix of size img. The basic idea is to enable each UE to generate and provide CSI feedback by selecting a preferred receive strategy img, which can differ from the equalizer img used in (6.3). Therefore, each UE can choose its desired receive strategy according to its own computational capabilities and knowledge about CSI at the Receiver (CSIR) including interference, independently from other UEs.

Each UE is assumed to use linear receive filters to transform the MIMO channel into an effective MISO channel (Thiele et al. 2010b, 2009) img. We limit the evaluation to a Multiuser Eigenmode Transmission (MET)-based (Boccardi and Huang 2007b) approach: each UE decomposes its channel img by a Singular Value Decomposition (SVD) into orthogonal eigenspaces, that is, img. Further, each UE is applying for a single data stream only. Thus, it is favorable to select the dominant mode, that is, the singular vector corresponding to the highest singular value. The effective channel after decomposition using the dominant left singular vector, that is, img is given by img. The scheme maximizes the signal power transferred from a cluster to the UE. UEs should preferably be grouped such that their modes show highest orthogonality. This keeps the costs in received power reduction due to ZF precoding as small as possible. The approach has two major advantages: first, the multiple receive antennas are efficiently used for suppression of interference at the UE side. Second, by reducing the number of data streams per UE, the system is enabled to serve a larger set of active UEs instantaneously and thus benefiting from multiuser diversity.

The scheme has been evaluated with a system simulator using img multiantenna BS sectors and a wraparound technique using the 3GPP Spatial Channel Model Extended (SCME) channel model (Baum et al. 2005). The system performance is determined by assuming a dynamic and UE-driven clustering method. However, the system does not utilize any additional gains from multiuser scheduling, that is, active set of UEs is selected according to the following metric. A set of active multiantenna UEs is uniformly distributed in a cluster of the cellular environment. The UE selection for each cell is done by a round-robin scheduling policy, yielding a set of UEs img of size img. Note that the UEs in img experience highest channel gain to the k-th BS, that is, all UEs are connected to a master BS. In addition, all UE sets img are disjoint for different BSs img. Further, we emulate a cluster selection which is UE-centric and dynamic over frequency: the K strongest channel gains of the UEs in img are the ones of the K BSs within the cluster.

Results are provided for different cluster sizes of img. All results in Figure 6.7 are based on an equal per-beam power constraint with a Per-Antenna Power Constraint (PAPC) (Zhang and Dai 2004a), which is aligned with the assumptions made in 3GPP LTE. For reference purpose, the performance results for interference-limited Single Input Single Output (SISO) as well as a MIMO img transmission are included. For img, two active fixed beams are sent to img different UEs in a round-robin manner.

Figure 6.7 Performance of distributed DL CoMP results as a function of the cluster size K. The normalized MSE is given per subchannel, that is, in case of K = 10 sectors in the cluster, the UE estimates NT = 2 times 10 = 20 subchannels in total with an i.i.d. Gaussian normalized MSE. Reproduced by permission of © 2009 IEEE

img

As a next step, CSI feedback is introduced from each UE to its serving BS. In particular, each UE is assumed to decompose its MIMO channel matrix into its eigenspaces, where only the dominant one is used as feedback (Boccardi and Huang 2007b; Noda et al. 2007). Based on this feedback, the BS can serve its UEs using ZF beamforming. From Figure 6.7 rather small gains are observed from channel adaptive precoding in a single sector, that is, img. This is mainly caused by the following fact: a simplified PAPC is assumed, which leads to a suboptimal power allocation where only one antenna transmits with full power and all others are scaled accordingly (Zhang and Dai 2004a). In contrast, in the case of fixed PMI-based precoding all BS antennas transmit with full power.

Now, the cluster size is increased from img to img. Figure 6.7 depicts the achievable Shannon information rate per sector as a function of the cluster size K and accuracy of CSI feedback, that is, in case of error free and erroneous CSI. For erroneous feedback, an additive Gaussian i.i.d. normalized Mean Square Error (MSE) is considered per subchannel – that is, per antenna port. The error variance μ is normalized with respect to channel power. Throughput results are provided for different values of μ, where the precoding weights are obtained by the erroneous CSI estimates. From this figure it is obvious that an MSE of img dB would restrict the useful cluster size to img. In essence, the CoMP gains as function of the cluster size show less saturation behavior for improved multicell channel knowledge.

In conclusion, it can be observed that the median sector spectral efficiencies are increased by 220 %, 300 % and 430 % for coordinating 3, 5 and 10 cells for error-free CSI feedback, respectively, compared to a noncoordinated SISO setup. These numbers are reduced to 190 %, 230 % and 300 % in case of erroneous feedback with an MSE of img dB.

Dynamic Base Station Clustering

The previous two methods presented here were JP user-centric based clustering. Dynamic BS clustering as its name indicates is a dynamic JP BS-centric clustering method. In fact, in the PJP method a threshold was chosen to limit the cluster size in order to keep the overhead rate reasonable. As a further extension the cluster can be created in a dynamic way in order to maximize a given objective function, for example, img which is function of img the BS clusters, img the UEs scheduled in each cluster, img the beamforming coefficients, and img the power allocation. The algorithm is described in details in Appendix C.1.2.

In Table 6.3 the performance of this algorithm is summarized in terms of average rate per cell and cell-edge rate (5 % tile) respectively for a cluster size of 10 BSs, which corresponds to a 50 % reduction in the number of BSs sharing the data of the UEs scheduled for transmission in a given frame. Four different techniques are compared: noncooperating BSs, static coordination, that is, clusters of cooperating BSs are kept fixed during all the simulation and in each cluster the UEs are selected for transmission using a Proportional Fair (PF) scheduler, dynamic coordination and full coordination, that is, all the K BSs cooperate together and up to M UEs are scheduled for transmission in each frame with a PF approach. As seen in Table 6.3, a substantial fraction of the gains with full cooperation can be achieved with this technique. Note that the simulation assumption are shown in Appendix C.1.2.

Table 6.3 Performance in terms of gains with respect to the non-CoMP case. The cell-edge performance is measured at the 5 %-tile of the UE throughput CDF.

img

6.4.4 Uplink Joint Processing

As already stated in section 6.1 the concept of CoMP encompasses both DL and uplink of a radio access network. We have so far put an emphasis in this chapter on Downlink techniques, but JP CoMP has the same potential in the uplink. JP CoMP in the uplink is more straightforward to implement, with respect to backward compatibility, since it can be made transparent for the UEs. Intrasite CoMP can also be efficiently implemented, and there are large gains in average spectral efficiency and cell-edge performance to obtain, 22 % and 26 % respectively, as shown in an LTE-like evaluation in Frank et al. (2010).

Intersite CoMP can provide additional gains, but the backhaul traffic between BSs is a challenge. Limited central JP in the form of joint detection is performed today by coordinated BSs in Third Generation (3G) Wideband Code Division Multiple Access (WCDMA) networks, denoted soft handover for multiple participating sites and softer handover for cells/sectors of the same site. Decentralized implementations and quantization (Marsch and Fettweis 2007) are suggested as the main approaches for achieving lower backhaul traffic for JP. Decentralized implementations include distributed decoding (Aktas et al. 2008), with local interaction between neighboring BSs, and distributed iterative detection Khattak et al. (2008), where each BS performs single UE detection and exchanges data in an iterative manner. In Marsch and Fettweis (2007), quantized data is exchanged among BSs. More basic analysis and approaches can be found in Yifan and Goldsmith (2006) and Yifan et al. (2006).

To illustrate the potential gains of intersite JP CoMP in an LTE uplink, the system level simulation results in Hoymann et al. (2009) and Frank et al. (2010) are highlighted. In Hoymann et al. (2009), a straightforward algorithm is considered, where the received complex baseband signals from cooperating BSs are distributed in a decentralized manner over the backhaul links to the serving BS, provided the received signal strength difference is within a certain range. The results are reproduced in Table 6.4, cf. Hoymann et al. (2009) for further simulation assumptions.

Table 6.4 Average cell throughput and 5 %-percentile UE throughput for different cooperation parameters (Hoymann et al. 2009)

img

Similar gains for intersite cooperation are reported in (Frank et al. 2010). In addition, in this work an intersite joint detection scheme with reduced backhaul load is proposed and evaluated within an LTE system. The idea is to restrict the cooperation to a subset of the available subcarriers per Physical Resource Block (PRB), and the scheme is combined with a threshold on received signal strength difference as in Hoymann et al. (2009). The results show that performance close to full cooperation can be reached if only eight out of 12 subcarriers per PRB are considered for cooperation, with a corresponding reduction of 33 % in the backhaul.

6.5 Coordinated Beamforming and Scheduling Techniques

As shown in section 6.4, JP is a very effective technique to enhance both cell-average and cell-edge throughput. Nevertheless, from section 6.4 it appears clear that many problems remain to be solved to apply the JP paradigm to realistic systems. In this section a different class of techniques is considered, where each UE may connect to a single BS at the same time.

Coordination is either used to jointly schedule UEs belonging to BSs in the same coordinated cluster (coordinated scheduling) or to calculate beamforming coefficients and power allocation, jointly in the set of coordinated cells (coordinated beamforming).

Due to the reduced constraints in terms of amount of exchanged data, coordinated scheduling/beamforming is likely to be a candidate for the first implementations in realistic systems. In this section, after a summary of the state of the art, two methods are described in more details, the first one based on coordinated beamforming (section 6.5.2), the second on coordinated scheduling (section 6.5.3).

6.5.1 State of the Art

The idea of coordinated beamforming comes from the mid-1990s, mainly targeting a so-called SINR leveling problem, that is, the power levels and the beamforming coefficients are calculated to achieve some common SINRs in the system or to maximize the minimum SINR. Usually in these first implementations there is no scheduling involved, as the set of UEs is fixed and the SINRs are optimized for this fixed set of UEs.

In Rashid-Farrokhi et al. (1998) an UL system is considered where power and receive beamforming coefficients are calculated in order to achieve a minimum target SINR for each UE and minimize the sum of the powers. Two algorithms can be used to calculate the powers and beamforming coefficients: a centralized one, and a distributed iterative one that updates powers and coefficients based on the last interference measurement.

In Bengtsson (2001), an algorithm to determine the jointly optimal DL beamformers and assignment of UEs to BSs for a set of cochannel UEs is presented. The algorithm is based on global knowledge about the channels from all BSs to all UEs.

Referring to coordinated beamforming methods, an alternative method where the optimal minimum power beamformers can be obtained locally relying on some coupled parameters exchanged between adjacent BSs is shown in Tölli et al. (2011). More details on this method are reported in section 6.5.2.

Coordinated scheduling is a relatively new idea and the theoretical limits of this approach have been explored only in the last past years. In the following, some of the main works in this field are recalled.

In Gesbert and Kountouris (2011) and Choi and Andrews (2008) the problem of joint multicell power control and scheduling to maximize the sum rate is considered. In Gesbert and Kountouris (2011) it is shown that, when the number of UEs goes to infinity,multicell interference (no matter how strong) does not affect the asymptotic scaling of the network, if a rate-optimal scheduling is applied. Moreover, a simple scheduler based on each cell measuring a worst-case SINR and not requiring any exchange of information between the cells results in a quasi-optimal behavior. For example, when UEs have the same average SINR the capacity has the same scaling law as in a single-cell broadcast channel with random beamforming. In Choi and Andrews (2008), the effect of intercell scheduling on multiuser diversity is studied. A full-reuse system with intercell scheduling is compared with a system with frequency reuse. It is proved that the former system has a higher multiuser diversity order due to the effect of intercell scheduling.

From a more practical perspective, different works have studied the possibility of exploiting information about the intercell interference in the scheduling process. In the following, two of the major contributions in this area are recalled. In Das et al. (2003) the entire network is divided into clusters and centralized scheduling is assumed for cells belonging to the same cluster such that only the long term average channel conditions are assumed available to the central scheduling entity. Scheduling involves determining which BSs in the cluster should transmit in each time slot and to which UEs. In Alcatel-Lucent (2008) a technique is considered where the UE sends a feedback to the serving BS which includes also information about the worst interfering beams belonging to nonserving BSs. A distributed scheduler is used that prioritizes the beam/UE selection across BSs belonging to the same coordination cluster.

It is worth noting that coordinated beamforming/scheduling techniques present another important advantage with reference to JP methods: the enhanced robustness to UEs mobility that could be achieved in case of slower update change of coordination data. In particular, studies reported in (WINNER+ 2009b) have shown that coordinated beamforming is competitive compared to non-CoMP and JP for the International Telecommunication Union (ITU) Urban Macro model scenario.

6.5.2 Decentralized Coordinated Beamforming

In this section, a decentralized coordinated beamforming method first introduced inTölli et al. (2009,2011) is outlined. In this method the optimal minimum power beamformers are considered and they are obtained locally at the BSs relying on some coupled parameters exchanged between adjacent BSs. The problem of finding such optimal beamformers is reformulated such that the BSs are coupled by real-valued intercell interference terms, that are handled by taking local copies of the terms at each BS and enforcing consistency between them. Thus, the coupling in the interference terms is transferred to a coupling in the consistency constraints, which can then be decoupled by a standard dual decomposition approach allowing a distributed algorithm. The consistency is enforced by exchanging the local intercell interference terms between adjacent BSs as depicted in Figure 6.8.

Figure 6.8 Distributed implementation of interference aware beamformer design Tölli et al. (2011). Reproduced by permission of © 2011 IEEE

img

The proposed approach is able to guarantee feasible solutions for intermediate number of iterations. The UE specific SINR targets can be guaranteed even if the update rate of the coupled interference terms between BSs is relatively low, at the possible cost of increased sum power. In addition, the dual decomposition approach allows for a number of special cases, where the backhaul information exchange is further reduced at the cost of somewhat suboptimal performance.

The general system optimization objective considered here is to minimize the total transmitted power subject to fixed UE-specific SINR constraints img. This can be formulated as

(6.10)equation

where L is number of BSs, img is the total power transmitted by the l-th BS and img is the m-th UE's SINR.

A decentralized method was proposed in Tölli et al. (2011) for the minimum power beamforming problem in (6.10). The problem can be solved by using a dual decomposition approach, which is appropriate when the optimization problem has a coupling constraint such that the problem decouples into several subproblems after relaxing the constraint. Thus, the original one-level optimization problem can be separated into two-level optimization, that is, subproblems for fixed dual variables (prices) and a master dual problem in charge of updating the dual variables (Boyd et al. 2007). In other words, the master problem sets the price for the resources in each subproblem, which in turn has to decide the amount of used resources depending on the price. By using this approach a decentralized solution of (6.10) can be found where the beamformer vectors are obtained locally relying on coupled real-valued intercell interference parameters exchanged between adjacent BSs. A more detailed description and the mathematical derivations can be found in Appendix C.2.1.

Table 6.5 Main performance results of decentralized coordinated beamforming.

Transmission scheme Average sum power [dB]: k = 0 dB Average sum power [dB]: k = 10 dB
Coherent JP-CoMP −1.45 9.71
CB-CoMP: UE-specific ICI constraint 3.90 19.44
ZF for ICI 8.92 24.81
ZF for intra- and intercell interference 15.21 25.21

A detailed performance evaluation of the decentralized coordinated beamforming with a wide range of numerical results, for example in time-correlated fading scenario, is provided in Tölli et al. (2011). Some of the numerical examples from (Tölli et al. 2011) are provided in the following. Table 6.2 presents the average sum power of a system with img UEs, img BSs and img transmit antennas. The fixed SINR target per UE are 0 dB and 10 dB. Different coordinated beamforming cases and two ZF (interference nulling) approaches are compared with coherent transmission at the cell edge, where each UE has similar large-scale fading properties. As expected, coherent CoMP greatly outperforms the coordinated beamforming cases at the cell edge. The coordinated beamforming case required about 5-6 dB more power than the JP CoMP case in order to meet the 0 dB SINR target. There is a large gain from the optimal intracell beamformer design (ZF for ICI) as compared to the channel inversion (ZF for both intra- and inter-cell interference). It shall be noted that without coordination, the required power is infinite (i.e. SINR constraints can not be guaranteed).

6.5.3 Coordinated Scheduling via Worst Companion Reporting

In Alcatel-Lucent (2008), an approach to realize coordinated scheduling is proposed for LTE. The basic idea behind this proposal, nicknamed as worst companion reporting, is to allow the UE to provide feedback information about how the interference level could be reduced by forbidding the interfering eNBs to use one or more beams, belonging to a codebook of possible transmit beams.

The scheme can be summarized as follows. Each UE measures the channel of the serving eNB and of neighboring eNBs; the feedback signaling from UE to the serving eNB is designed such that in addition to the index of the preferred precoding codeword and the associated CQI (as for example in LTE), the UE also reports the index of the most interfering eNBs. Based on this additional information, the eNBs can coordinate to serve the UEs using appropriated precoding weights in order to minimize the inter-cell interference. The principle of the scheme is summarized in Figure 6.9.

Figure 6.9 Principle of the interference reduction via worst companion signaling

img

Coordinated scheduling can be realized either with a centralized or with a distributed processing approach. An effective distributed scheduling technique, nicknamed as cyclically prioritized scheduling, has been proposed in Alcatel-Lucent (2010). The principle is that a scheduling priority is assigned to each eNB, and scheduling between the eNBs of the same coordination cluster is realized starting from the eNBs with high priority and ending to eNBs with low priority. More in details, an eNB with a given priority schedules a set of UEs based on

  • the PMI report and the associated CQI;
  • the worst companion report and the associated img-CQI, which is the CQI when the worst companion is not used in the neighbor cell;
  • the constraints on the precoding codewords coming from eNBs with a higher priority.

For example an eNB with a priority i could decide, based on the worst companion report, to forbid an eNB with priority img to use the precoding codeword corresponding to the worst companion report.

The scheduling constraints (or forbidden codewords) must be passed to eNB with a lower priority (for example through the X2 interface in LTE). Besides this, no other information exchange is assumed between the cooperating BSs. The required bandwidth on the backhaul is therefore small. From the latency point of view, the backhaul and the schedulers must be designed taking into account that overall latency is given by the time to realize the entire coordination cycle from the eNBs with the highest priority to the eNBs with the lowest priority.

The performance is assessed considering jointly cell edge and cell average performance, as shown in Figure 6.10. As the main goal of the worst companion scheme described herein is to improve the cell-edge performance, a possible way to assess the achieved gain is by fixing the cell-average throughput and measure the gain in cell-edge performance. For instance, by choosing an appropriate PF scheduling α factor, we can achieve a gain of about 24 %. Further results and the simulation assumptions of this coordinated scheduling method are shown in Appendix C.2.2.

Figure 6.10 Performance of the worst companion reporting, in terms of cell average and cell-edge performance

img

6.6 Practical Implementation of CoMP in a Trial Environment

In this section an initial real-time implementation and testing of distributed JP CoMP in the Downlink of an LTE-Advanced trial network, taking place in Berlin in the framework of the LTE-Advanced testbed, is described. Enabling features such as distributed synchronization (Jungnickel et al. 2008), cell- and UE-specific pilots have been implemented and tested in real-time; a fast backbone network has been realized to transport data for the JP coordination. The trial layout shown in Figure 6.11 has been implemented; it is based on two BSs and two UEs having two antennas each at 2.68 and 2.53 GHz, respectively. The duplex is based on FDD mode, using 20 MHz bandwidth in the DL and 10 MHz bandwidth for each UE in the UL. For more implementation details, please refer to Jungnickel et al. (2010, 2009c).

Figure 6.11 Distributed DL CoMP (Jungnickel et al. 2009c). © 2009 IEEE

img

Transmission experiments, limited by interference, have been conducted using three multicell transmission modes:

  • mode with no coordination between the cells, considered as a lower bound, applying per-antenna rate control with interference-aware equalization at the UE;
  • mode with distributed DL CoMP, where the mutual interference between the cells is canceled;
  • mode with isolated cells, considered as an upper bound where the interference from other cells is switched off.

Trials over the air are described here for indoor and outdoor-to-indoor scenarios covering both intrasite and intersite CoMP. For an outdoor-to-outdoor scenario refer to Jungnickel et al. (2010). Quite impressive observations regarding the reduced outage probability at the cell edge are described and demonstrate the huge overall performance improvements when using CoMP JP instead of interference-limited transmission. Clearly these gains are obtained only in the absence of interference from non-coordinated cells. The performance in presence of residual Co-Channel Interference (CCI) were shown in section 6.4. Nonetheless, the following experiments demonstrate that the implementation challenges of DL CoMP can be overcome and that similarly high gains as predicted by the theory can be realized in practice.

6.6.1 Setup and Scenarios

Scenarios comprise indoor and outdoor-to-indoor configurations, refer to Figure 6.12(a) for the geography of the trial setup. In the indoor test, both BSs are located in the same lab. In the outdoor-to-indoor scenarios, two BS sites transmit and two sectors are selected either at the same site or at two different sites in order to realize intra- and inter-site cooperation, respectively. Sites are located on the Deutsche Telekom Laboratories (TLabs) building at Ernst-Reuter-Platz (84 m antenna height) and on the Technical University of Berlin (TUB) main building, Straβe des 17 Juni (43 m, see Figure 6.12(a)) in Berlin. The estimated height of the buildings in the area is between 25 and 35 m. For more insights, refer to Jungnickel et al. (2009b). Sites are interconnected by optical fibers deployed in the campus with a length of 4.5 km. The X2 signaling over the fiber is based on 1 Gbps Ethernet.

Figure 6.12 Scenario for CoMP trials in Berlin LTE-Advanced testbed (Thiele et al. 2010b). Reproduced by permission of © 2010 IEICE

img

For indoor and outdoor-to-indoor scenarios, both UEs are located on the 11th floor at the Heinrich Hertz Institute (HHI). Both UEs are placed at the south front of the building with the windows facing towards both BSs either in the same lab or in two different labs which are 25 m separated. UE2 is at a fixed location. In order to capture the local fading statistics, UE1 moves at low speed with approximately 3 cm/s. In our implementation, UE1 is always assigned to BS1 and UE2 to BS2, that is, handover is not performed. For performance evaluation, the overall statistics from UEs placed in the same lab and different labs are considered (see 6.12(b) for an example of results in the interference limited and CoMP case for both UEs).

6.6.2 Measurement Results

Indoor and outdoor-to-indoor results are plotted in Figure 6.13. In the setup implemented in practice, a single UE is assigned to BS1 and a single UE to BS2. In the case in which both UEs are at the cell border between BS1 and BS2, both interference scenarios are very similar and similar throughput statistics are then obtained. In Figure 6.13 the throughput is shown together with reference bounds. As explained above, some different transmission modes are taken into consideration in Figure 6.13:

  • as a lower bound, interference-limited transmission is introduced using an identity matrix as an independent precoder for BS1 and BS2, in combination with Optimum Combining at the receiver side; in this case, a single UE is assigned to each BS and, due to limited hardware, the trial system cannot utilize multiuser diversity gains in the frequency selective scheduling, as discussed in section 6.4.3, which implies higher outage probability;
  • CoMP JP transmission is used from both sites with a fixed number of four streams on the air;
  • as an upper bound, there is the case of isolated cells where the interference from the other cell is not present, that is, either BS1 or BS2 is switched off.

Figure 6.13 Performance results of CoMP trials in Berlin LTE-Advanced testbed (Jungnickel et al. 2010). Reproduced by permission of © 2010 IEEE

img

In the indoor scenario (i.e. Scenario 1 in Figure 6.13), both BSs are received with the same average power in a rich scattering environment. The average SINR is around 0 dB in both cells simultaneously. Due to multiple reflections in the room, however, both the signal and the interference experience fading. Statistically independent fading of both signals creates a crucial throughput situation for a UE: when moving the UE by a few cm only, we can realize situations where either the signal channel is strong while the interference is in a fade, and correspondingly the serving BS assigns data transmission in a certain part of the whole frequency band, as well as the reverse situation where the interference channel is strong while the signal is in a fade so that no more data is usually transmitted. Note that the BS assignment is always kept fixed, that is, BS1 serves UE1 and BS2 serves UE2. As a consequence, the UE suffers from bad SINR conditions and thus the outage probability amounts to 50 %. Thus, the data traffic is not continuous but frequently interrupted when moving through the lab and the UE experience may be quite poor. If CoMP is enabled in such a bad interference scenario, significant improvements are observed for the data throughput. Despite the critical interference situation and although the data rate still varies, CoMP removes outage completely. Using CoMP in such indoor deployments provides 18 times higher data rates for both UEs in both cells simultaneously with respect to the interference limited setup. The CoMP setup realizes approximately 78 % of the rate achievable in the isolated cell scenario.

Next consider the intrasite scenarios where the intrasite interference is canceled (see Scenarios 2a and 2b in Figure 6.13). It is typical in the distributed multicell network that path losses are not equal for different pairs of BSs and UEs. Nonetheless, the basic observations remain similar. In the interference-limited case, again there is significant outage. Using CoMP, in contrast, both UEs can realize 30 % of the peak data rate on average. Note that in this scenario both UEs are placed in the same building, while both serving BSs are situated at the same site, that is, the same antenna pole. In this case, signals transmitted from both sectors show high correlation (Jaeckel et al. 2009).

Finally consider the intersite scenarios, called Scenarios 2c and 2d in Figure 6.13, where the intersite interference from the other site is suppressed. These scenarios provide superior performance compared to the intrasite scenarios, despite distributed synchronization of both BSs. Studies of the underlying channel correlations suggest that the higher data rates observed with CoMP in the intersite scenarios may also be attributed to the lower transmit antenna correlation (Jaeckel et al. 2009) if eNB antennas as well UEs are sufficiently separated in the deployment. However, in (Jungnickel et al. 2010) the difference between intrasite and intersite CoMP throughput compared to isolated cell transmission is rather small when UEs are placed at different buildings or streets in the outdoor-to-outdoor scenario. The mobility of the UE in the field was rather limited due to the time variance of the channel. The inevitable feedback delay causes a mismatch between the precoder and the actual channel as soon as the UE is moved. Compensation is possible by using channel prediction to make the interference suppression more robust.

6.7 Future Directions

The CoMP systems represent a topic of wide interest in the research community and are widely studied towards their introduction in practical Radio Access Networks. Nevertheless CoMP is still in a phase of uncertainty due to the complexity of its implementation and some unresolved issues, such as the CSI estimation accuracy, and at the moment the fundamental step from theoretical research to practical industrial exploitation has yet to be done. The information and the encouraging results shown in this chapter, especially regarding the trial, have been aimed at illustrating CoMP issues.

As explained earlier, CoMP adoption would have implications for the networks, both from a technical and also from a financial standpoint. Additional costs have to be expected for the significant exchange of information that some practical implementations of CoMP could require, leading even to a complete update of the backhaul technologies adopted for this information sharing. In this sense the choice of centralized or decentralized solutions for CoMP, as well as of the level of coordination, are not a minor issue, together with a more pervasive deployment of fiber back-haul and front-haul connections to support higher capacity data exchange among cooperating nodes.

Currently the CoMP studies are mostly related to two main areas, and standardization activities have followed this classification as well: coordinated beamforming/scheduling and JP schemes. Considering DL, these are the most suitable practical implementations of CoMP. It has been pointed out how JP performances are significantly better than coordinated scheduling/beamforming, at the expense of greater, and often excessive, complexity. As for uplink almost the same applies, even if it has to be added that the somewhat lower complexity could allow a possible quicker adoption of uplink CoMP, currently foreseen as a proprietary implementation in cooperative nodes.

It is expected that the future directions will focus on several aspects of CoMP: architectural, theoretical and practical.

There are many architectural aspects to be tackled, like the backhaul design and the adopted simplifications to be sought: decentralized approaches, distributing complexity among nodes, and clustering. In particular, dynamic clustering is considered a viable solution for future CoMP methods. Otherwise, at least in the first implementations, cooperation among sectors of the same cell could be attempted. The introduction of Heterogeneous Network(HetNet), then, could be an option to exploit coordination also among different cells in a multilayer environment.

From a theoretical point of view, the feedback design and robust channel estimation algorithms are of crucial importance. In fact, the question to be solved, regardless of the adopted coordination solution, is the achievable tradeoff between the amount of information exchanged among the nodes and the expected performance gain. Another related important tradeoff, with achievable performance, is between the amount of feedback and performance gain, discussed in Section 6.2.

From a practical perspective, future research trends should consider reference signal design, CSI impairments, and hardware limitations such as synchronization. In fact, the performance of any CoMP system must take into consideration the imperfect CSI, addressing scenarios less ideal than those analyzed so far in the first stages of CoMP studies. Upper bounds about achievable capacity with realistic assumptions on CSI are being searched. A possible further direction of research about CoMP could be also the performance estimation in case of different mobility scenarios; especially for JP the high mobility scenario could be very challenging for CoMP. Finally, synchronization (and the cyclic prefix design) of the cooperative entities in CoMP schemes is another research topic that shall deserve attention in the future, especially for the schemes requiring higher coherence. There are already functionalities aiming to distribute clock signals among the coordinated nodes or relying on GPS, but further efforts in this field are foreseen in the coming years.

Note

1. The coherence in the transmission methods is a relevant issue in the CoMP techniques, depending on the level of synchronization among the coordinated entities.

References

3GPP 2006 Physical Layer Aspects for Evolved Universal Terrestial Radio Access (UTRA) (Release 7). Technical Report 25.814 v7.1.0, 3rd Generation Partnership Project (3GPP), http://www.3gpp.org/ftp/Specs/2006-09/Rel-7/.

3GPP 2010 Further Advancements for E-UTRA Physical Layer Aspects. Technical Report 36.814 v9.0.0, 3rd Generation Partnership Project (3GPP), http://www.3gpp.org/ftp/Specs/2010-03/Rel-9/.

Aktas D, Bacha M, Evans J and Hanly S 2006 Scaling results on the sum capacity of cellular networks with MIMO links. IEEE Trans. Inform. Theory 52(7), 3264–3274.

Aktas E, Evans J and Hanly S 2008 Distributed decoding in a cellular multiple-access channel. IEEE Trans. Wireless Commun. 7(1), 241–250.

Alcatel-Lucent 2008 UE PMI feedback signalling for user pairing/coordination. Technical Report R1-083759, 3GPP Third Generation Partnership Project, Working Group RAN1, meeting 54, Prague, Czech Republic.

Alcatel-Lucent 2010 Performance of coordinated beamforming with multiple PMI feedback. Technical Report R1-100944, 3GPP Third Generation Partnership Project, Working Group RAN1, meeting 60, San Francisco, USA.

ARTIST4G 2010 Definitions and Architecture Requirements for Supporting Interference Avoidance Techniques (ed. D'Amico V. and Gresset N.). Public Deliverable D1.1, Advanced Radio InTerface TechnologIes for 4G SysTems - ARTIST4G, August 2010.

Baum DS, Salo J, Milojevic M, Kyösti P and Hansen J 2005 MATLAB implementation of the interim channel model for beyond-3G systems (SCME), http://radio.tkk.fi/en/research/rf_applications_in_mobile_communication/radio_channel/scm.html.

Bengtsson M 2001 Jointly optimal downlink beamforming and base station assignment in Proc. IEEE Int. Conf. Acoust., Speech, Signal Processing, Salt Lake City, Utah, USA.

Björnson E, Jaldén N, Bengtsson M and Ottersten B In press Optimality properties, distributed strategies, and measurement-based evaluation of coordinated multicell OFDMA transmission. IEEE Trans. Signal Processing.

Bletsas A and Lippman A 2006 Implementing cooperative diversity antenna arrays with commodity hardware. IEEE Commun. Mag. 44(12), 33–40.

Boccardi F and Huang H 2006 Zero-forcing precoding for the mimo broadcast channel under per-antenna power constraints Signal Processing Advances in Wireless Communications, 2006. SPAWC '06. IEEE 7th Workshop on, pp. 1– 5.

Boccardi F and Huang H 2007a Limited downlink network coordination in cellular networks. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications.

Boccardi F and Huang H 2007b A near-optimum technique using linear precoding for the MIMO broadcast channel. IEEE International Conference on Acoustics, Speech and Signal Processing.

Botella C, Piñero G, González A and de Diego M 2008 Coordination in a multi-cell multi-antenna multi-user W-CDMA system: a beamforming approach. IEEE Trans. Wireless Commun. 7 (11), 4479–4485.

Boyd S, Xiao L, Mutapcic A and Mattingley J 2007 Notes on decomposition methods. Course reader for convex optimization II, Stanford, http://www.stanford.edu/class/ee364b/.

Brueck S, Zhao L, Giese J and Awais Amin M 2010 Centralized scheduling for joint transmission coordinated multi-point in LTE-Advanced. International ITG Workshop on Smart Antennas (WSA), pp. 177– 184.

Caire G and Shamai S 2003 On the achievable throughput of a multiantenna gaussian broadcast channel. IEEE Trans. Inform. Theory 49(7), 1691–1706.

Carneheim 1994 FH-GSM Frequency Hopping GSM Proceedings of the 44th IEEE Vehicular Technology Conference.

Choi W and Andrews J 2007 Downlink performance and capacity of distributed antenna systems in a multicell environment. IEEE Trans. Wireless Commun. 6(1), 69–73.

Choi W and Andrews J 2008 The capacity gain from intercell scheduling in multi-antenna systems. IEEE Trans. Wireless Commun. 7(2), 714–725.

Cover TM and Gamal AAE 1979 Capacity theorems for the relay channel. IEEE Trans. Inform. Theory 25(5), 572–584.

Cover TM and Thomas JA 1991 Elements of Information Theory. John Wiley & Sons, Inc., New York.

Dahlman E, Parkvall S, Sköld J and Beming P 2008 3G Evolution – HSPA and LTE for Mobile Broadband. Academic Press, Amsterdam.

Dahrouj H and Wei Y 2008 Coordinated beamforming for the multi-cell multi-antenna wireless system Information Sciences and Systems, 2008. CISS 2008. 42nd Annual Conference on, pp. 429– 434.

Das S, Viswanathan H and Rittenhouse G 2003 Dynamic Load Balancing Through Coordinated Scheduling in Packet Data Systems IEEE Infocom, San Francisco.

Deutsche Telekom AG and Vodafone 2009 Coordinated link adaptation based on multi-cell channel estimation in the LTE-A uplink. Technical Report R1-095067, 3GPP Third Generation Partnership Project, Working Group RAN1, meeting 59, Jeju, South Korea.

Ericsson and ST-Ericsson 2010 Channel reciprocity in FDD systems including systems with large duplex distance. Technical Report R1-100853, 3GPP Third Generation Partnership Project, Working Group RAN1, meeting 60, San Francisco, USA.

Frank P, Müller A and Speidel J 2010 Inter-site joint detection with reduced backhaul capacity requirements for the 3GPP LTE uplink Proc. IEEE VTC-Fall 2010, pp. 1–5.

Gesbert D, Hanly S, Huang H, Shamai Shitz S, Simeone O and Yu W 2010 Multi-Cell MIMO Cooperative Networks: A New Look at Interference. IEEE J. Selected Areas in Commun. 28(9), 1380–1408.

Gesbert D and Kountouris M 2011 Rate Scaling Laws in Multicell Networks Under Distributed Power Control and User Scheduling. IEEE Trans. Inform. Theor. 57(1), 234–244.

Gupta P and Kumar PR 2000 The capacity of wireless networks. IEEE Trans. Inform. Theory 46(3), 388–404.

Holma H and Toskala A 2004 WCDMA for UMTS: Radio Access for Third Generation Mobile Communications 3rdf edn. John Wiley & Sons, Ltd, Chichester.

Hongyuan Z, Mehta NB, Molisch AF, Jin Z and Huaiyu D 2008 Asynchronous interference mitigation in cooperative base station systems. IEEE Trans. Wireless Commun. 7(1), 155–165.

Hoymann C, Falconetti L and Gupta R 2009 Distributed uplink signal processing of cooperating base stations based on IQ sample exchange Proc. IEEE ICC 2009, pp. 1– 5.

Jaeckel S, Jiang L, Jungnickel V, Thiele L, Jandura C, Sommerkorn G and Schneider C 2009 Correlation properties of large and small-scale parameters from multicell channel measurements European Conference on Antennas and Propagation (EuCAP 2009), Berlin, Germany.

Jafar SA and Goldsmith AJ 2002 Transmitter optimization for multiple antenna cellular systems Proc. IEEE ISIT 2002, vol. 1, p. 50, Lausanne, Switzerland.

Jafar SA, Foschini GJ and Goldsmith AJ 2004 PhantomNet: Exploring optimal multicellular multiple antenna systems. EURASIP J. Applied Signal Processing 5, 591–604.

Jungnickel V, Forck A, Jaeckel S, Bauermeister F, Schiffermller S, Schubert S, Wahls S, Thiele L, Haustein T, Kreher W, Müller J, Droste H and Kadel G 2010 Field trials using coordinated multi-point transmission in the downlink Proc. 3rd International Workshop on Wireless Distributed Networks (WDN), held in conjunction with IEEE PIMRC 2010, Istanbul, Turkey.

Jungnickel V, Jaeckel S, Thiele L, Jiang L, Krüger U, Brylka A and Helmolt C 2009a Capacity measurements in a cooperative multicell mimo network. IEEE Trans. Veh. Technol. 58, 2392–2405.

Jungnickel V, Schellmann M, Thiele L, Wirth T, Haustein T, Koch O, Zirwas W and Schulz E 2009b Interference-aware scheduling in the multiuser MIMO-OFDM downlink. IEEE Communicaions Magazine 47(6), 56–66.

Jungnickel V, Thiele L, Wirth T, Haustein T, Schiffermüller S, Forck A, Wahls S, Jaeckel S, Schubert S, Juchems C, Luhn F, Zavrtak R, Droste H, Kadel G, Kreher W, Mueller J, Stoermer W and Wannemacher G 2009c Coordinated multipoint trials in the downlink Proc. 5th IEEE Broadband Wireless Access Workshop (BWAWS), Co-located with IEEE GLOBECOM 2009, Honolulu, Hawai.

Jungnickel V, Wirth T, Schellmann M, Haustein T and Zirwas W 2008 Synchronization of cooperative base stations. IEEE International Symposium on Wireless Communication Systems 2008 (ISWCS2008).

Karakayali M, Foschini G and Valenzuela R 2006a Network coordination for spectrally efficient communications in cellular systems. IEEE Trans. Wireless Commun. 13(4), 56–61.

Karakayali MK, Foschini GJ and Valenzuela RA 2006b Network coordination for spectrally efficient communications in cellular systems. IEEE Trans. Wireless Commun. Mag. 3(14), 56–61.

Karakayali M, Foschini GJ, Valenzuela RA and Yates RD 2006c On the maximum common rate achievable in a coordinated network Proc. IEEE ICC 2006, vol. 9, pp. 4333– 4338, Istanbul, Turkey.

Khattak S, Rave W and Fettweis G 2008 Distributed iterative multiuser detection through base station cooperation. EURASIP Journal on Wireless Communications and Networking. Article ID 390489.

Kramer G, Gastpar M and Gupta P 2005 Cooperative strategies and capacity theorems for relay networks. IEEE Trans. Inform. Theory 51(9), 3037–3063.

Laneman JN, Tse DNC and Wornell GW 2004 Cooperative diversity in wireless networks: Efficient protocols and outage behavior. IEEE Trans. Inform. Theory 50(12), 3062–3080.

Laneman JN and Wornell GW 2003 Distributed space-time-coded protocols for exploiting cooperative diversity in wireless networks. IEEE Trans. Inform. Theory 49(10), 2415–2425.

Liang Y, Yoo T and Goldsmith A 2006 Coverage spectral efficiency of cellular systems with cooperative base stations Proc. IEEE GLOBECOM 2006, pp. 1– 5, San Francisco, CA, USA.

Marsch P and Fettweis G 2007 A framework for optimizing the downlink performance of distributed antenna systems under a constrained backhaul 13th European Wireless Conference.

Noda M, Muraguchi M, Khanh TG, Sakaguchi K and Araki K 2007 Eigenmode tomlinson-harashima precoding for multi-antenna multi-user MIMO broadcast channel Information, Communications Signal Processing, 2007 6th International Conference on, pp. 1– 5.

Papadogiannis A, Bang HJ, Gesbert D and Hardouin E 2008a Downlink overhead reduction for multi-cell cooperative processing enabled wireless networks Personal, Indoor and Mobile Radio Communications, 2008. PIMRC 2008. IEEE 19th International Symposium on, pp. 1– 5.

Papadogiannis A, Gesbert D and Hardouin E 2008b A dynamic clustering approach in wireless networks with multi-cell cooperative processing Proc. IEEE ICC 2008, pp. 4033– 4037.

Papadogiannis A, Gesbert D and Hardouin E 2008c A dynamic clustering approach in wireless networks with multi-cell cooperative processing. IEEE International Conference on Communications.

Papadogiannis A, Hardouin E and Gesbert D 2009 Decentralising multi-cell cooperative processing on the downlink: a novel robust framework. EURASIP J. Wireless Commun. Networking, Special issue on broadband wireless access.

Papadopoulos H and Sundberg CEW 2008 Space-time codes for mimo systems with non-collocated transmit antennas. IEEE J. Selected Areas in Commun. 26(6), 927–937.

Rashid-Farrokhi F, Tassiulas L and Liu K 1998 Joint optimal power control and beamforming in wireless networks using antenna arrays. IEEE Trans. Commun. 46(10), 1313–1324.

Sendonaris A, Erkip E and Aazhang B 2003a User cooperation diversity–part I: System description. IEEE Trans. Commun. 51(11), 1927–1938.

Sendonaris A, Erkip E and Aazhang B 2003b User cooperation diversity–part II: Implementation aspects and performance analysis. IEEE Trans. Commun. 51(11), 1939–1948.

Shamai S and Zaidel B 2001 Enhancing the cellular downlink capacity via co-processing at the transmitting end Proc. IEEE VTC Spring 2001, pp. 1745– 1749, Rhodes, Greece.

Skjevling H, Gesbert D and Hjimgrungnes A 2008 Low-complexity distributed multibase transmission and scheduling. EURASIP Journal on Advances in Signal Processing.

Somekh O, Simeone O, Bar-Ness Y and Haimovich M 2006 Distributed multi-cell zero-forcing beamforming in cellular downlink channels Proc. IEEE GLOBECOM 2006, pp. 1– 5, San Francisco, CA, USA.

Somekh O, Simeone O, Bar-Ness Y, Haimovich A, Spagnolini U and Shamai S 2007 An Information Theoretic View of Distributed Antenna Processing in Cellular Systems, in Distributed Antenna Systems: Open Architecture for Future Wireless Communications first edn. Auerbach Publications, CRC Press.

Song Y, Cai L, Wu K and Yang H 2007 Collaborative mimo based on multiple base station coordination. Contribution to IEEE 802.16m.

Thiele L, Boccardi F, Botella C, Svensson T and Boldi M 2010a Scheduling-Assisted Joint Processing for CoMP in the Framework of the WINNER+ Project Proc. Future Network & Mobile Summit 2010, Florence, Italy.

Thiele L, Jungnickel V and Haustein T 2010b Interference management for future cellular OFDMA systems using Coordinated Multi-Point transmission. IEICE Transactions on Wireless Distributed Networks.

Thiele L, Wirth T, Haustein T, Jungnickel V, Schulz E and Zirwas W 2009 A unified feedback scheme for distributed interference management in cellular systems: Benefits and challenges for real-time implementation. 17th Euopean Signal Processing Conference (EUSIPCO2009).

Tölli A, Codreanu M and Juntti M 2008 Cooperative MIMO-OFDM cellular system with soft handover between distributed base station antennas. IEEE Trans. Wireless Commun. 7(4), 1428–1440.

Tölli A, Pennanen H and Komulainen P 2009 Distributed coordinated multi-cell transmission based on dual decomposition Proc. IEEE GLOBECOM 2009, pp. 1– 6, Honolulu, Hawai'i, USA.

Tölli A, Pennananen H and Komulainen P 2011 Decentralized minimum power multi-cell beamforming with limited backhaul signaling. IEEE Trans. Wireless Commun. 10(2), 570–580.

van der Meulen EC 1977 A survey of multiway channels in information theory: 1961–1976. IEEE Trans. Inform. Theory 23(1), 1–37.

Venkatesan S 2007 Coordinating base stations for greater uplink spectral efficiency in a cellular network. IEEE International Symposium on Personal, Indoor and Mobile Radio Communications.

Wei Y and Lan T 2007 Transmitter optimization for the multi-antenna downlink with per-antenna power constraints. IEEE Trans. Signal Processing 55(6), 2646–2660.

Weingarten H, Steinberg Y and Shamai S 2006 The capacity region of the Gaussian multiple-input multiple-output broadcast channel. IEEE Trans. Inform. Theory 52(9), 3936–3964.

WINNER+ 2009a Celtic Project CP5-026 Initial Report on Advanced Multiple Antenna Systems (ed. Komulainen P. and Boldi M.). Public Deliverable D1.4, Wireless World Initiative New Radio - WINNER+, January 2009, http://projects.celtic-initiative.org/winner+/index.html (accessed June 2011).

WINNER+ 2009b Celtic Project CP5-026 Intermediate Report on CoMP (Coordinated Multi-Point) and Relaying in the Framework of CoMP (ed. Mayrargue S.). Public Deliverable D1.8, Wireless World Initiative New Radio - WINNER+, November 2009, projects.celtic-initiative.org/winner+/index.html (accessed June 2011).

WINNER+ 2009c Celtic Project CP5-026 Preliminary WINNER+ System Concept (ed. Osseiran A. and Gouraud A.). Public Deliverable D2.1, Wireless World Initiative New Radio – WINNER+, May 2009, http://projects.celtic-initiative.org/winner+/index.html (accessed June 2011).

WINNER-II 2007a IST-4-027756 Interference averaging concepts (ed. Bublin M.). Public Deliverable D4.7.1, Wireless World Initiative New Radio – WINNER, 2007, http://projects.celtic-initiative.org/winner+/index.html (accessed June 2011).

WINNER-II 2007b IST-4-027756 Interference avoidance concepts (ed. Mange G.). Public Deliverable D4.7.2, Wireless World Initiative New Radio – WINNER, 2007, http://projects.celtic-initiative.org/winner+/index.html (accessed June 2011).

Wyner AD 1994 Shannon–theoretic approach to a Gaussian cellular multiple-access channel. IEEE Trans. Inform. Theory 40(6), 1713–1727.

Xiao S, Zhang Z and Shi Z 2010 Clustered multi-point coordinating transmission systems with the reduced overhead. IEEE Trans. Commun. 5(6), 493–500.

Yifan L and Goldsmith A 2006 Symmetric rate capacity of cellular systems with cooperative base stations Global Telecommunications Conference, 2006. GLOBECOM '06. IEEE, pp. 1– 5.

Yifan L, Taesang Y and Goldsmith A 2006 Coverage spectral efficiency of cellular systems with cooperative base stations Global Telecommunications Conference, 2006. GLOBECOM '06. IEEE, pp. 1– 5.

Yu W 2006 Uplink-downlink duality via minimax duality. IEEE Trans. Inform. Theory 52(2), 361–374.

Yu W and Lan T 2007 Transmitter optimization for the multi-antenna downlink with per-antenna power constraints. IEEE Trans. Signal Processing 55(6, part 1), 2646–2660.

Zarikoff B and Cavers J 2008 Multiple frequency offset estimation for the downlink of coordinated mimo systems. Selected Areas in Communications, IEEE Journal on 26(6), 901–912.

Zhang H and Dai H 2004a Cochannel interference mitigation and cooperative processing in downlink multicell multiuser MIMO networks. EURASIP J. Wireless Comm. and Netw. 2004(2), 222–235.

Zhang H and Dai H 2004b Cochannel interference mitigation and cooperative processing in downlink multicell multiuser MIMO networks. EURASIP J. Wireless Comm. and Netw. 2004(2), 222–235.

Zhang H, Mehta NB, Molisch AF, Zhang J and Dai H 2007 On the fundamentally asynchronous nature of interference in cooperative base station systems Proc. IEEE ICC 2007, vol. 4, pp. 6073– 6078, Glasgow, Scotland, UK.

Zhang P, Tao X, Zhang J, Wang Y, Li L and Wang Y 2005 A vision from the future: beyond 3G TDD. IEEE Commun. Mag. 43(1), 38–44.

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset
3.129.195.215