Dynamic Resource Assignment and Cooperative Relaying in Cellular Networks: Concept and Performance Assessment

The wide area test scenario is an urban macrocellular deployment. It aims at providing ubiquitous coverage in an urban environment resulting in rather large cells, having a radius up to several kilometers. Base stations (BSs) are consequently expected to provide high power outputs, in each of the three sectors, equipped with four antennas. All BSs are deployed above rooftop ( ), possibly requiring additional masts for their installation. This implies that the site selection and rental costs will probably be dominant with respect to the other costs such as the backhaul infrastructure. We further consider RSs which are deployed below rooftop ( ) with a single antenna and a significantly lower output power in order to keep costs low and allow for a flexible deployment of multiple RSs per sector. For the same reason the RS is equipped with a single antenna.

We distinguish between a carefully planned RS deployment with a high probability of line-of-sight (LOS) and a not carefully planned RS deployment without LOS to the BS. For both cases and the BS-MS link we assume an urban macro-cell model whereas we assume for the RS-MS a non-LOS (NLOS) microcell model (see Appendix B for details).

The cells form a regular grid with a hexagonal layout and an intersite distance (ISD) of 1000 meters. We study this scenario with three RSs per sector according to the deployment in Table 1. It provides, as shown in Figure 2, a good coverage for MSs at the cell border. Moreover, we consider also the scenario with only one RS per sector (also shown in Figure 2) for comparison purpose. The exact RS deployments are outlined in Table 1 for both scenarios.

The metropolitan area test scenario is an urban micro-cellular scenario modeled by a two-dimensional Manhattan grid consisting of streets (width 30 m) and buildings ( block size). The BS deployment follows the UMTS 30.03 recommendation [13] with 73 BS deployed below rooftop level (10 m height) and placed in the midpoint between two crossroads. Two sectors are formed with bore-sight along the street direction and one antenna per sector. The added relays extend the coverage area of these BSs and distribute the cell capacity more evenly.

A single RS (5 m height) is added to each BS site in the midpoint between two BSs, as depicted in Figure 3. Thereby the number of radio access points is doubled. Adding a second relay per BS site increases the cell throughput only slightly and does not justify the additional costs [25]. The RSs are equipped with two antennas, a directional antenna to communicate with the serving BS and an omnidirectional antenna to serve its MSs. The power masks assigned to each BS and RS in Figure 3 are used by an interference coordination scheme based on soft-frequency reuse which is described in Section 4. The transmit power of the RS is 7 dB lower than the transmit power of the BS to enable a smaller physical size.

A LOS link is assumed for nodes in the same street and a NLOS link for nodes in different streets and MSs are located inside a building or in a street. Details of the propagation model and additional simulation parameters can be found in Appendix C.

The local area test scenario is defined as an isolated hot-spot-like indoor area with high user density where the users are either stationary or slowly moving. It is characterized by high shadowing and considerable signal attenuation due to the existence of rooms separated by walls. As a result of the isolated characteristics the interference is much lower compared to the previous two scenarios. The scenario consists of one floor (3 m high) in a building with two corridors ( ) and 40 rooms ( ).

A deployment with two single antenna BSs (dark gray nodes) is presented in Figure 4. They are located in the middle of the corridors, halfway from the left/right side of the building. Each of them is assisted by four single antenna RSs (light gray nodes): two on the left and two on the right side, respectively (i.e., 10, 30, 70, and 90 meters from the left or the right side of the building) as depicted in Figure 4. All the area marked with gray color benefits from the use of (cooperating) RSs.

A LOS or NLOS office propagation model is employed depending on the presence of walls between the BS, RSs, and MSs. Details of the propagation model and additional simulation parameters can be found in Appendix D.

A fixed and static resource assignment will not allow to exploit the full potential of relay-based deployments since the relay deployments can have very different properties as illustrated in Section 3. Therefore, we propose that the BS flexibly assigns parts or all of the available system resources to itself and to each RS in the relay enhanced cell. In particular the BS assigns the frames in which the RS communicates with the BS (act as an MS) or serves its MS (act as a BS). Further, it assigns the OFDMA resource units (chunks) that the RS can use in the frames for which it acts as a BS. The assigned resources are then available for autonomous scheduling at each individual radio access point. Figure 5 illustrates an example resource allocation for a BS with three RS in its cell.

The actual resource assignment strategy depends on the utilized interference coordination and multiantenna techniques. In the wide area test scenario beamforming has been shown to be an effective way to improve the cell capacity [37]. We propose to coordinate the interference from the subcells formed by the BS to the subcell formed by RSs by using at the BS beams with low interference to the RS subcell for resources that have been assigned to the RS. The amount of resources for the RS is dynamically adjusted depending on the traffic and interference situation. We refer to this approach as Dynamic Resource Sharing (DRS) [38]. DRS uses logical beams which can be seen as a dynamic version of sectors. The Dynamic Resource Sharing (DRS) acts in three steps: the creation of the beams, grouping of the beams, and the actual resource assignment [38]. For the assessment results presented in Section 5 we utilize the resource assignment that we proposed in [23] which aims to achieve the maximum possible cell throughput by allocating an OFDMA resource unit (chunk) to the group of beams that can reach the highest total rate.

In the metropolitan area scenario we study an interference coordination scheme based on soft frequency reuse. It assigns power masks (in the frequency domain) to neighboring radio access points to coordinate the mutual interference. Thereby, soft frequency reuse enables frequency reuse one and at the same time each radio access point has high power resources with reduced interference available to schedule MS located at the border area. Soft frequency reuse is better suited for the metropolitan area than beamforming because the radio signal propagates very well in the street canyons making it difficult to separate different beams. Further, interference coordination is mainly needed at street crossings and in the border area between radio access points, whereas the border area is smaller than in a wide area deployment.

In the local area scenario we make use of the fact that the BSs and RSs located in different corridors are separated by at least three walls which can be perceived as a natural means of suppressing interference. Due to the physical separation, sharing of the same resources may be possible for multiple transmissions. In cases where it is not possible to share the resources, the users are either served cooperatively by multiple radio access points or exclusive resources are assigned.

Table 2 summarizes the essential elements of the resource assignment. The MS does not need to perform additional measurements to support the resource assignment. The BS uses the received signal strength from neighboring radio access points (BS or RS) reported by the MS as an input, which are anyway required for handover purposes. Please note that the logical beams are a dynamic version of sectors and therefore also measurements for the logical beams will be available.

Real world deployments are not as regular as the presented test scenarios and due to the small size of the subcells formed by BSs and its RSs the traffic density can vary significantly in these subcells. The proposed dynamic resource assignment scheme offers sufficient flexibility to adapt better to real world situations than a static resource assignment.

In WINNER we propose a distributed scheduling, that is, the BS assigns resources to itself and the RSs in the relay enhanced cells but it does not centrally schedule the transmissions to the MSs. The RSs can then independently allocate these resources to its associated MSs. Thus, frequency adaptive transmissions and multiantenna transmission schemes can be supported without forwarding channel state information, precoding weight feedback, and so forth to the BS. This decision can be justified by the results in [14, 15] which indicate a performance loss of less than 10% compared to a centralized scheduler even without considering the signaling overhead for a centralized scheduler.

However, when utilizing distributed scheduling the BS should be aware of the buffer status of each MS or flow at the RS. If it forwards too much data to the RS eventually the buffer of the RS will overflow and if it forwards too little data the MS will be starved. Even if the buffer at the RS is large enough to store all the data for the MS, the resources on the BS-RS link have been wasted when the MS performs a handover to another RS or BS. In our work we have considered two different approaches to flow control: connection-based scheduling (CbS) and stop-and-go signaling. The results in Section 5 show that both schemes are well suited for the considered deployments with a maximum of two hops.

The CbS is a resource request and allocation strategy proposed in [24] for controlling the resources and delays of multihop communications with different numbers of hops. Each RS requests to the BS not only the needed resources for data transmission on the access link between the RS and the MSs but also on the multihop links from/to the BS. Every RS computes the resources required for each end-to-end connection served by the RS instead of only the next link towards the destination. The BS collects the resource requests and grants resources on each hop for each connection (uplink and/or downlink) between the BS and each RS.

The stop-and-go flow control requires less signaling than the CbS but CbS is better suited for deployments with more than two hops. It depends on the rate of the RS-MS link. The RS sends a stop signal for an MS to the BS when the queue size for the MS exceeds . The queue size depends on the current channel quality of the RS-MS link and is calculated as

where denotes the predicted rate (based on channel quality feedback) when the MS is assigned the full bandwidth and is a parameter that can depend on the number of users served by the RS and the amount of frames where the RS serves its MSs. For the numerical assessment results in the metropolitan area we have used a fixed parameter and compare the performance of the proposed flow control to a scenario without flow control.

Next to the flexible resource assignment, we propose cooperative relaying to further enhance the capacity of a relay enhanced cell. In the DL of single-path relaying, the data is first transmitted from the BS to the RS and then the RS forwards this data to the MS. (We refer in the following to noncooperative relaying as single-path relaying, because only a single transmission path between source and destination is exploited.) To gain on large-scale spatial diversity, most cooperative relaying protocols proposed in literature, for example, [39‐41] benefit from a combination of the transmissions in two phases, first from the BS and then from the RS. An overview and classification of different cooperative relaying protocols can be found in [42‐44].

As the transmission from a BS is received by the MS and the RS, dedicated multiantenna techniques (beamforming and other space division multiple access (SDMA) algorithms) can be applied only partially, because one stream is only optimized forone destination. Furthermore, as we assume an intelligent deployment, the achievable data rate on the BS-RS link is likely to exceed the data rate on the RS-MS links. However, to enable cooperation on the physical layer the same modulation and coding scheme or only a limited set of specialized and sophisticated modulation and coding schemes can be used.

Thus, we do not only consider cooperative relaying that exploits large-scale spatial diversity but we investigate mainly cooperative relaying, where multiple radio access points form a Virtual Antenna Array (VAA) [45]. Any multiantenna transmission technique, including spatial multiplexing, can then be applied, for example, to the BS antennas augmented by the antennas of an RS. In Section 5 we present results for a cooperative multiuser MIMO relaying scheme that we proposed in [26]. It utilizes distributed LQ precoding which has been introduced for cooperating BSs in [46] and a dirty paper coding technique as proposed in [47].

In our cooperative relaying proposal the first common node in the tree topology schedules the cooperative transmission. Thus, in a network that is limited to two hops, the BS allocates resources to all cooperative transmissions in a similar way as in single hop networks using similar feedback information. The BS then sends the resource allocation and the selected transmission mode (MIMO transmission scheme, precoding weights, modulation and coding scheme for different streams, etc.) together with the data to the RS(s). Both BS-RS cooperation and RS-RS cooperation are supported. Figure 6 illustrates restrictions at the RS resulting from cooperatively served MSs. The RS has to take these restrictions into account when allocating resources to the MSs served solely by the RS within the resources assigned from the BS.

When calculating the precoding weights for a cooperative (multiuser) MIMO transmission scheme the channel matrices of all the cooperating nodes have to be forwarded to the BS and the precoding weights have to be transmitted to the RS before the cooperative transmission. Due to this high amount of data which has to be communicated between BS and RS(s), MIMO cooperative relaying is more affected by a limited BS-RS link capacity than single path relaying. Hence, the proposed MIMO cooperative relaying solution requires a high capacity BS-RS link which can be guaranteed by a line-of-sight assumption between BS and RSs.

The highest gain from cooperative relaying is obtained if the signals received from the cooperating radio access points are of similar strength. Therefore we base the decision which radio access points (BS or RS) should form the VAA on the received signal strength reported by the MS and RS. In particular we propose the use of a static version of the REACT algorithm [48]. The original algorithm was developed for mobile ad hoc networks with relays and due to the fact that in the scenario under investigation the RSs are located at fixed positions there is no need to perform periodic neighbor discovery and topology recognition. The static version of REACT is executed by the BS and exploits information about power levels of the signals received by MSs from different radio access points (BS or RS) as well as the power levels of the signals received by RSs from the BS. Thus, the BS has a good overview of the topology to select the cooperation type.

Next to data transmissions the MSs have to receive control information. In our cooperative relaying proposal the control information is not transmitted cooperatively but each MS has a serving RAP which can be the BS or an RS. In either case, the serving RAP performs retransmissions, transmits the broadcast channel, receives feedback from the MS, and signals the resource allocation to the MS.

The IEEE802.16j draft standard [49] allows already a dynamic resource assignment in the time domain by adjusting the duration of the relay zone but no mechanism has been standardized for the frequency domain. In the case of dynamic resource sharing the resource assignment in frequency domain can simply be done by signaling chunks (subchannels in WiMAX terminology) that should not be used by an RS. For soft-frequency reuse, in addition the power mask to be applied for chunks has to be signaled. Thus, with small additional signaling the 802.16j standard can support the flexible resource assignment proposed in this paper.

The draft 802.16j standard [49] also specifies the possibility for cooperative BS-RS transmissions. It mentions two basic possibilities: cooperative source diversity (repetition coding) and cooperative transmit diversity established through distributed space-time block coding (STBC) and a combination of both. We propose a much more flexible scheme that supports also RS-RS cooperation and any MIMO scheme that is used in a system. Thus, major additions would be required to the standard in order to support our concept.

In this analysis the deployment positions of RSs are not optimized with respect to the propagation conditions to the BS. Therefore an NLOS model is assumed and the path-loss between BS-RS is calculated as in (B.2).

The wide area results on DRS in Table 3 show that the DRS outperforms the BS only deployment. By utilizing this approach the cell throughput is increased by 25% with only one RS per sector and by almost 50% assuming 3 RSs per sector. Table 3 also shows results for a Fixed Resource Partitioning (FRP) without coordinating the beams at the BS with RS transmissions. The static resource partitioning is based on the following considerations. The relay coverage area is about one forth of the sector area, as shown in Figure 2. The throughput of the relay link (BS-RS) is assumed to be twice the average throughput of the RS-MS links. Further, the throughput per user in the coverage area of a BS is assumed to be the same as in the coverage area of an RS. To avoid interference the BS does not serve its MSs while the RS serves its MSs. Hence, the resource demand for the different links was estimated to be for the BS-MS links, for the BS-RS links, and 2/9 for the RS-MS links. With this static resource partitioning we can observe that the average cell throughput is reduced by 30% compared to the BS only scenario. Thus, without properly assigning the resources inside the cell the potential benefits of relaying are lost and the performance might even degrade.

In the metropolitan area we compare the performance of a relay deployment using the flexible resource assignment proposed in Section 4 and soft frequency reuse (SFR) to a BS only deployment. These studies assume a slowly changing resource assignment for the studied part of the network which remains constant during the simulated 70 seconds of network operation. This models a flexible resource assignment that adapts to slow variations, for example, depending on the time of the day, and the same assignment is used for all cells in this part of the network.

Table 3 shows the results both for users located indoors and in the streets. The outdoor to indoor coverage of the BS only deployment is limited and adding relays is especially beneficial for users with low throughput in the BS only deployment. As a result the fifth percentile of the user throughput CDF more than quadruples. However, for users in the street the BS only scenario is already interference limited and adding RSs does neither increase the cell throughput nor the fifth percentile of the user throughput CDF.

We allow both the RS and BS to serve its MSs at the same time which achieves significantly better results compared to BS and RS serving MSs in separate frames. The amount of frames within a superframe where the RS is serving MSs depends on the capacity of the BS-RS link and the RS-MS links. As the capacity of the BS-RS link is very high, the best result was achieved when the RS serves its MSs in five out of eight frames. Thus, three out of eight frames are sufficient for the BS-RS communication. Selecting the optimal number of frames for the RS transmission improves the fifth percentile of the user throughput CDF by 38% and the average cell throughput by 4% compared to an assignment where the RS serves its MSs in every other frame. This indicates that the performance of relay deployments strongly depends on the proper balance between the resources spent on the first hop, between BS and RS, and on the second hop, between RS and MS.

We also studied the impact of flow control on the overall performance of the network. For the case without flow control we set the stop limit to 25 Mbit per flow which corresponds to about 8  seconds of data for an MS. Without flow control the average cell throughput decreases by less than 1% and the fifth percentile of the user throughput CDF by 3%. The impact of flow control is rather limited in this scenario since the BSs transmit data to the RSs only in 38% of the frames and RSs are only present in every second sector. The conclusions will likely be different in a scenario with more relays and more than two hops. Especially for more than two hops a flow control based on connection-based scheduling is likely the better option.

Cooperative relaying can further enhance the performance of a relay deployment. To evaluate the potential benefits of cooperative relaying we compare the cooperative multiuser MIMO relaying scheme with single-path relaying and a system using only direct links between BSs and MSs (BS only). The path loss for the BS-RS link assumes a careful relay deployment and is calculated as in (B.3).

Figure 7 presents the CDF of the expected user throughput . We can clearly observe from the CDF of the throughput that the number of users with low throughput is significantly reduced, compared to a system without relay stations. Besides, we can observe a major performance advantage of cooperative relaying in comparison to single-path relaying. This is of course at the cost of additional signaling and control overhead. Nonetheless, the coordinated and joint transmission of BSs and RSs seems to be a viable option especially in those areas where an MS experiences similar channel conditions to both radio access points.

In the local area test scenario, we assess the performance of cooperative relaying for the deployment given in Figure 4 [27]. We compare two different possibilities. The MSs are served by the BS or by RS using either single-path relaying (BS-RS-MS) or cooperative relaying (BS-VAA-MS), where a Virtual Antenna Array (VAA) is formed by a pair of cooperating RSs. The RSs forming the VAA are chosen with the use of a static version of the REACT algorithm as described in Section 4.3.

The results were obtained for a fixed modulation and coding scheme based on QPSK modulation and the ( ) convolutional code with the use of the fixed resource assignment in [27]. Further, an AWGN channel model was assumed. The presence of an outdoor network is modeled by setting an average interference power level of  dBm per subcarrier.

The results are depicted in Figure 8 and show improvements of up to 30% for regions where RS-RS cooperation can be applied compared to MSs that are served only by the BS. Analyzing these results one can see that the performance is largely unaffected in the central part of the scenario and in the corridors that can be covered by the BS. However, direct transmission is only advantageous in the central part and nearby the horizontal walls, halfway between the vertical ones, where it offers a gain of 10% over cooperative relaying. As it was already mentioned, the benefits of deploying relays in the local area scenario can be explained by numerous walls providing additional shadowing. Hence, it is advantageous to have additional radio access point. The BSs serve the MSs in their immediate vicinity. The other MSs are served cooperatively and experience a higher throughput.

Table 4 presents selected simulation parameters and Table 5 the OFDMA parameters of the WINNER system in both FDD and TDD mode.

The numerical evaluations have been carried out for a carrier frequency of 3.95 GHz which has not been allocated to IMT systems at the World Radio Communications Conference 2007. Nevertheless, for example, 200 MHz have been allocated between 3.4 and 3.6 GHz and changing the carrier frequency, for example, to 3.55 GHz will not significantly change the results or our conclusions. The 5 GHz carrier frequency chosen for the local area is close to the license exempt Universal-Networking Information Infrastructure (U-NII, 5.15–5.35 GHz) band and the upper Industry Science and Medical (ISM, 5.725–5.825) band.

The path-loss equation [35] for the BS to MS link, assuming an urban macro-cell model (C2) with non-line-of-sight (NLOS), is given by

where is the BS antenna height in meters, the carrier frequency in GHz, the transmitter-receiver separation in meters, and  dB the standard deviation of the shadow fading. The link BS to RS is expected to be 3 dB better with respect to the link BS to MS due to some intelligence when selecting the RS location:

For the interfering link from other BS, the path loss is obtained as in (B.1).

Alternatively, a more careful planning of the relay locations with line-of-sight (LOS) or obstructed LOS can be assumed. In that case the path-loss equation for the BS-RS link can be found from a stationary feeder model (B5f):

where  dB.

The path loss for the RS to MS link is based on an urban micro-cell model (B1) and can be found in LOS situations as

where denotes the speed of light and  dB. For  m and  m, where and refer to the heights of RS and MS antennas, the breakpoint is at  m; is the path loss in free space. In NLOS situations the path loss can be found as

with  dB. We assume a geometry for and , where the RS and MS are located in perpendicular streets. Furthermore, is the distance from the RS to the midpoint of the crossing and is the distance from the midpoint of the crossing to the MS whereas .

A LOS link is assumed for nodes in the same street and a NLOS link for nodes in different streets and MSs are located inside of a building or in a street. The corresponding channel and path-loss models for all links are specified in [35]: urban micro-cell B1 LOS for nodes in the same street (B.4), urban micro-cell B1 NLOS for nodes in different streets (B.5), and the outdoor to indoor urban micro-cell model (B4).

The outdoor to indoor path-loss model consists of three components, the outdoor path-loss as defined by the urban micro-cell (B1 LOS/B1 NLOS) model, the penetration loss into the building , and the indoor path-loss . The path-loss equation is given as

where the path loss is calculated using the four points , and of the outer walls of the building blocks that are closest to the indoor MS, and

Moreover, denotes the distance to the closest point in all four streets surrounding the building block. Please note that is the distance traveled in the streets to reach these points and that the B1 path-loss model distinguishes whether the two nodes are in the same street or not, that is, line-of-sight (LOS) or NLOS path-loss model is used. Furthermore, denotes the angle relative to the normal of the wall under which the signal is entering the building at the points closest to the MS, and denotes the distance to the MS inside the building block.

A LOS or NLOS office propagation model (A1) [35] is employed depending on the presence of walls between the BS, RSs, and MSs. In particular LOS propagation model is used between BS and RS as well as between BS/RS and MS if there is a direct visibility. The LOS model is given by the following formula:

where denotes the distance in meters between the transmitter and the receiver and represents the standard deviation of the shadow fading and is equal to  dB. The NLOS propagation model is defined as

where denotes the number of walls on a direct line between the transmitter and the receiver and  dB. For the simulations light walls of the same type for all walls were assumed.