WO2015120569A1

WO2015120569A1 - Coordination of resource-allocation in cellular communications network

Info

Publication number: WO2015120569A1
Application number: PCT/CN2014/000631
Authority: WO
Inventors: Zhenning Shi
Original assignee: Orange
Priority date: 2014-06-30
Filing date: 2014-06-30
Publication date: 2015-08-20

Abstract

The allocation of resources in different cells of a group of cells in a cellular communications network is coordinated using a two-level approach. A centralized controller (CONT_CENT) determines resource-allocation coordination decisions (G-RCD) that optimize the performance of the group of cells. The centralized controller also calculates, for each cell of the group, local coordination assistance information (L-CAI) indicating how the performance of the other cooperating cells would be affected if this cell were to depart from the centrally-set resource-allocation coordination decisions. A local controller (CONT_LOC) updates the centrally-set resource-allocation coordination decisions (G-RCD) in respect of a given cell based on up-to-date local knowledge of user channel conditions and / or interference values for this cell, provided that the updating will produce a net performance gain for the cooperating cells considered as a group. The local controller evaluates whether the updating will produce a globalperformance gain or loss using the L-CAI.

Description

COORDINATION OF RESOURCE-ALLOCATION IN CELLULAR

COMMUNICATIONS NETWORK

The present invention relates to the field of resource allocation in a cellular communications network and, in particular, to methods and devices configured to coordinate resource allocation in different cells of a group.

In cellular communications networks it is routine for decisions regarding the allocation of radio resources to users in a cell to be taken locally to that cell (e.g. at a local base station, for example at a NodeB, or an eNodeB). For example, a control unit may determine how to assign resource blocks/sub-channels to different UEs (user equipment) within a cell. Typically, the control unit which takes these decisions is referred to as a "scheduler", even in contexts where the control unit is not exercising control over timing. For example, according to the LTE specification (the "Long Term Evolution" specification established by the 3rd Generation Partnership Project (3GPP): and the variants thereof, such as LTE-Advanced), an eNodeB scheduler may assign radio resources to the UEs in the cell(s) it controls (or there may be an individual scheduler for each sector/cell). Typically the scheduler assigns resources to UEs in its sector(s)/cell(s) with the aim of maximizing the system performance obtained for the sector(s)/cell(s) in question.

Various proposals have been made to increase coordination between cells, particularly with the aim of mitigating inter-cell interference (ICI) (and, especially, downlink ICI). Some proposals seek to mitigate ICI by coordinating how resources are allocated in different cells of a group even when those cells have different local schedulers, This extended coordination may be beneficial, for example so that resource reuse can be optimized while keeping ICI to a low level. This type of approach is called "coordinated scheduling" (CS).

Figure 1 is a diagram illustrating an example of a centralized architecture in which coordinated scheduling may be implemented. The centralized architecture of Fig.l includes a central scheduling control unit (labelled SCHED_CONT in Fig.l) and multiple cooperating cells (labelled CLi to CL_N in Fig.l). In this system the cooperating cells periodically report user channel condition information and interference measurements to the central scheduling control unit, for coordination purposes. This provides the central control unit SCHED_CONT with a global view of user distribution, and of channel and interference conditions, i.e. a view which is "global" in the sense that it takes into account conditions in the whole group of cooperating cells. Based on this global view, the central controller calculates an allocation of resources, in this example radio resources, within each cell of the group, which should result in optimum overall performance of this group of cells. The central control unit then notifies the cooperating cells of its resource coordination decision (G- CD) and the individual cells perform scheduling to implement the resource allocation specified by the received G-RCD. Here the letter G is used in G-RCD to recall that the centralized controller's resource-allocation decisions are generated based on a global perspective of optimizing performance for the relevant group of cells. The parameter (or parameters) which are set by the G- CD can vary according to the application, and examples include, but are not limited to: resource block/sub-channel assignment, power allocation, user scheduling, and so forth.

In the example of Fig.l there is a one-to-one relationship between cells and base stations (local nodes). However, this is not essential: a single base station/local node may control resource allocation in plural cells.

Fig.3 is a diagram illustrating how functions may be divided between a centralized controller and local schedulers for cells, in an example of a coordinated scheduling system implemented in an LTE network. In the example illustrated in Fig.3, the centralized controller sends its globally-determined resource coordination decisions, G-RCD, to eNodeB schedulers of the cells in the LTE network. The centralized controller CONT_CENT generates the G-RCD and the eNodeB schedulers perform scheduling for their cells according to the relevant directions in the G-RCD (i.e. according to that part of the G-RCD which applies to this eNodeB's cell(s)).

When coordinated scheduling is implemented using a centralized architecture such as those discussed above, problems can arise due to the time taken for channel state information to reach the central control unit from the local nodes and the time taken for the G-RCDs from the central control unit to reach the local nodes. These time delays can be a particular issue when this signalling takes place over legacy networks (i.e. non-ideal backhauls producing noticeable delays). The problem may be understood from consideration of Fig.2.

Figure 2 shows an example of signalling that is exchanged between the cells and centralized controller of Fig.l. In Fig.2 the signalling is arranged ordered in the time domain. According to this example, the centralized scheduler repeatedly determines, according to a global perspective, resource decisions for successive time periods, or cycles. The resource decision RCD_N for cycle N is based on cell-reported information from the previous cycle.

According to this example, the cells send the centralized scheduler information {h_N_ , I_N_ } relating to the user channel state information h and measured interference / applicable during the cycle N-l. The cycle duration T_cs + D_RTT \s determined by the centralized processing time T_cs and the backhaul round trip delay D_RTT.

In legacy networks the RTT latency of backhaul inter-connections can be quite noticeable (e.g. 10-20 ms) and so, when signalling between the centralized scheduler and the group of cells takes place over such inter-connections, the resource decisions made by the centralized scheduler are based on channel/interference information that is relatively out of date. This leads to suboptimal CDs (e.g. RCDs that do not, in fact, optimize resource reuse) and compromises the benefits introduced by having a centralized scheduler.

Coordinated scheduling can be implemented using a de-centralized architecture but, in this case, the decisions regarding coordination of resource allocation tend not to optimize the global performance of the group of cooperating cells considered as a whole, because the local decisions are taken without adequate knowledge of the global situation.

The present invention has been made in view of the above-mentioned disadvantages.

Embodiments of the present invention provide a method of coordinating allocation of resources in different cells of a group of plural cells in a cellular communications network, the method comprising:

receiving first information at a central controller, wherein the received first information is indicative of user channel condition information and/or interference in the group of cells;

generating decision data by the central controller, wherein the decision data is indicative of a first allocation of resources in the cells of said group, and the central controller selects the first resource-allocation based on the first information and on a selection criterion prioritizing the global performance level for the group of cells;

the central controller generating, for each cell of the group, coordination assistance information indicating the expected effect, on the performance of the other cells in the group, produced by a variation by said cell of said first resource-allocation;

transmitting the decision data and coordination assistance information to the cells of the group;

generating variant data by a local controller associated with a cell of the group, based on second user channel condition information and/or interference data obtained by said local controller in respect of said cell, wherein the second user channel condition information/interference data is more up-to-date than the first information used by the central controller in generating the decision data, and wherein the variant data is indicative of a variant allocation of resources in said cell different from the resource allocation in said cell indicated by the decision data;

evaluating, by said local controller, based on coordination assistance information generated by the central controller in respect of said cell, the effect that application of the variant resource allocation would produce on the global performance of said group of cells compared to application of the resource allocation in said cell indicated by the decision data; based on the result of the global-performance evaluation, selecting between the resource allocation specified by the variant data and the resource allocation specified by the decision data provided by the central controller; and

commanding allocation of resources in said cell according to the selected resource allocation

In the above-mentioned embodiments of the invention a two-level approach is applied to coordinated scheduling of resource-allocation in a group of cells in a cellular network. A centralized controller functions to generate decisions on resource allocation for the group of cells but the centralized controller also generates coordination assistance information which enables local controllers to determine how departure, by them, from the centrally-specified resource-allocation for their cell would affect the performance of the other cells in the group. The local controllers can thus modify the centrally-specified resource allocation for their cell, when up-to-date local cell information suggests that this would be beneficial, depending on whether or not the global-performance evaluation (by taking into account the coordination assistance information) indicates that the modification would produce a net improvement in the global performance of the group of cells. Thus, resource reuse can be optimized taking into account up-to-date local cell information without degrading overall performance of the group of cells.

In some applications the method may be implemented so that the variant resource allocation is selected whenever the global-performance evaluation indicates that the variant resource allocation would produce an improvement in global performance of the group of cells (or an improvement in excess of a certain threshold amount).

In certain applications, as well as including an evaluation of the effect that the variant resource allocation in the cell would have on the global performance of the overall group of cells the method also includes an assessment of the effect the variant resource allocation would have on resource usage in the cell. The two factors may then both play a role in the process for selecting which resource allocation should be chosen for application in this cell.

Methods which assess the resource-usage of the variant resource allocation as well as its effect, on global performance can take these two factors into account in different ways when selecting which resource allocation should be applied in the cell. However, in general, advantages can be obtained when the selection - between the variant resource allocation and the resource allocation specified in the decision data - produces a reduction in resource-usage and/or an improvement in global system performance.

Thus, for example, if the variant resource allocation for this cell would use fewer resources than the resource allocation specified by the central controller, and the local controller calculates (based on the coordination assistance information) that the global performance of the group of cells would not be reduced by using the variant resource allocation instead of the allocation specified by the decision data, then the variant resource allocation is selected for application. This approach produces savings in resources. As another example, the selection may be designed to select the variant resource allocation over the resource allocation specified in the decision data, even in a case where this results in increased resource usage, provided that this is predicted to result in an improvement (of a desired degree) in global performance.

Embodiments of the invention further provide a group-controller apparatus configured to control coordination of resource-allocation in different cells of a group of plural cells in a cellular communications network, the group-controller apparatus comprising:

a first calculation unit configured to receive information indicative of user channel condition information and/or interference in the group of cells, and to generate decision data indicative of a first allocation of resources in the cells of said group, wherein the first calculation unit is configured to evaluate the first resource-allocation as producing a specified global performance level for the group of cells; and

a second calculation unit configured to generate, for each cell of the group, coordination assistance information indicating the expected effect, on the performance of the other cells in the group, produced by a variation by said cell of said first resource-allocation;

wherein the group-control apparatus is configured to transmit the decision data and coordination assistance information to the cells of the group.

Embodiments of the invention still further provide a local controller apparatus associated with a cell in a cellular communications network, wherein said cell is one of a group of plural cells and the local controller apparatus is configured to control allocation of resources in said cell in coordination with resource-allocation in the other cells of said group, the local controller apparatus comprising:

a receiver configured to receive, from a group controller apparatus: decision data indicative of a first allocation of resources in the cells of said group, and coordination assistance information indicating the expected effect, on the performance of the other cells in the group, produced by a variation by said cell of said first resource-allocation;

a variant-data generator configured to generate variant data indicative of a variant allocation of resources in said cell different from the resource allocation in said cell indicated by the decision data, wherein the variant data generator is configured to generate the variant data based on user channel condition information and/or interference data obtained by the local controller apparatus in respect of said cell, said user channel condition information/interference data being more up-to-date than user channel condition information/interference data used in generation of the decision data at the group controller apparatus;

a variant-data evaluator configured to evaluate, based on received coordination assistance information, the effect that application of the variant resource allocation would produce on the global performance of said group of cells compared to application of the resource allocation in said cell indicated by the decision data;

a selector configured to perform a selection, based on the result of the evaluation performed by the variant-data evaluator, between the resource allocation specified by the variant data and the resource allocation specified by decision data received from the group controller apparatus; and

an output module configured, to output a command commanding allocation of resources in said cell according to the resource allocation selected by the selector.

Further features and advantages of embodiments of the present invention will become apparent from the following description of said embodiments, which is given by way of illustration and not limitation, illustrated by the accompanying drawings, in which:

Fig.1 is a diagram illustrating an example of a centralized architecture for implementing coordinated scheduling in a cellular communications network;

Fig.2 is a diagram illustrating how latency can affect a system according to Fig.1 ; Fig.3 is a diagram illustrating a division of functions between components in an example of a centralized architecture configured to implement coordinated scheduling in an LTE network;

Fig.4 is a diagram illustrating a division of functions between components in an example of an architecture according to an embodiment of the invention configured to implement two- level coordination of scheduling in an LTE network;

Fig.5 is a diagram illustrating components in a central controller and control components for local nodes in an example of a system implementing two-level coordination of scheduling;

Fig.6 is a flow diagram illustrating a resource-allocation coordination method implemented at a central controller according to an embodiment of the invention;

Fig.7 is a flow diagram illustrating a resource-allocation coordination method implemented at a local controller according to an embodiment of the invention;

Fig.8 is a diagram illustrating how piecewise differential values may be obtained for a non-differentiable utility function; and

Fig.9 is a diagram illustrating components in a local controller in an example of a system implementing two-level coordination of scheduling.

Embodiments of the present invention implement a two-level approach to coordinating resource allocation in different cells of a group. According to this two-level approach, "global" resource-allocation decisions that are taken centrally in view of optimizing performance of the group of cells considered as a whole are adapted locally, at the cell level, incorporating fresh local user information, conditional on a favourable assessment of the effect that the local adaptation may have on the overall performance of the group of cells.

Fig.4 illustrates an example of a general architecture that may be used to implement two-level scheduling coordination in certain embodiments of the invention. The example of Fig.4 is in the context of an LTE network to enable a straightforward comparison with the architecture illustrated in Fig.3. Both figures represent a generic centralized scheduler, not limited by reference to the particular resource whose allocation is being controlled.

In the example of Fig.4 the two-level scheduling coordination is implemented using two types of control elements: a centralized controller CONT_CENT operative to generate resource allocation decisions for a group of cooperating cells, and local controllers CONT_LOC for each of the local nodes serving the cells in the group. Once again, the centralized controller is configured to generate a global resource allocation (G- CD) based on user CSI and/or interference values that have been signalled for the cooperating cells and which are subject to delay (especially when the signalling has used legacy networks). However, in this example the centralized controller is also configured to determine information designated L-CAI, which stands for "Local coordination assistance information". The L-CAI indicates what effect there would be on the other cells in the group if a local node were to modify the resource allocation that the centralized controller has decided for a particular cell in the group. The centralized controller generates L-CAI for each cell in the group of cooperating cells and transmits the L- CAI to the local nodes for the relevant cells, as well as transmitting the G- CD.

Incidentally, the signalling between the centralized controller CONT_CENT and the local controllers CONT_LOC may take place over any convenient interface. Thus, for example, the signalling may take place over a new interface adopted specifically for implementation of this invention, or it may take place using an enhanced 3GPP X2 interface (in the context of LTE), and so forth.

The number of cells in the group may vary depending on various factors: for example, the number of cells may depend on the location within the cellular network of the centralized controller. In some embodiments centralized controllers are located at backhaul aggregation points. At a first level aggregation point typically 5-10 sites are connected and, in a case where there are 3 cells per site, the central controller may be coordinating scheduling for a group of 15-30 cells.

As another example, the number of cells in the group may vary depending on the calculation approach used by the centralized controller to determine the G-RCD (because the number of cells should not increase to the extent that the central controller spends a relatively long time for calculating the G-RCD). If the central controller applies a greedy search algorithm (see below) then the calculation complexity increases almost in a linear relationship with the number of cells. However, if an exhaustive search is used for the G-RCD then the increase in computation complexity may be exponential with respect to the number of cells. When deciding the number of cells in the group the system designer may take these considerations into account having regard to the latency of the backhaul connections.

The local controllers CONT_LOC of Fig.4 are configured to compute an updated resource allocation, G-RCD*, for the cells controlled by a given base station (in this example an eNodeB) taking into account up-to-date information that is available locally regarding user channel condition information and/or interference. The local controller is configured to adapt the G-RCD directive(s) applicable to the cell(s) controlled by the associated base station, based on the fresh channel information, selectively or conditionally depending on whether adaptation of the G-RCD directive(s) would produce a net benefit for the group of cooperating cells as a whole. The local controller is configured to use the L-CAI to make the assessment of whether or not the adaptation will bring benefits in terms of overall performance of the group of cooperating cells. In effect the L-CAI gives the local controller a global view of resource coordination. The local controller CONT_LOC passes the updated G-RCD* message to the associated eNodeB scheduler as guidelines to direct user scheduling. The updated G-RCD* message is the same as G- CD if the local controller assessed that adaptation would bring no net benefit, but will be adapted (based on fresh local conditions) if the local controller assessed that adaptation would bring net benefit to the cooperating cells considered as a group.

In the example of Fig.4 there is a one-to-one relationship between local controllers and the node components that assign resources to users (e.g. one local controller per eNodeB). However, in principle, if multiple eNodeBs are inter-connected with ideal backhaul (such that up-to-date user information can be shared instantly between them,) then a single local controller may be provided at one of these eNodeBs to serve all of them. This common local controller can implement if variation in resource allocation of this small group may bring overall performance benefits.

Fig.5 is a simplified diagram illustrating components in a central controller and control components for local nodes in an example of a system implementing two-level coordination of scheduling as describe above. In the example of Fig.5 a centralized controller 10 cooperates with local controllers to manage coordination of resource-allocation in a group of cells G containing three cells 20-i , 20₂ and 20₃. In this example the centralized controller 10 is deployed at a central node (e.g. an aggregation point of the radio access network) and comprises a first calculation unit 12 configured to calculate the global resource-allocation coordination decisions G-RCD, and a second calculation unit 14 to calculate the L-CAI. The first and second calculation units 12, 14 may be implemented, for example, using appropriately-programmed processors, and a single processor may be used to implement both calculation units.

In the example illustrated in Fig.5, each cell has a local scheduler LSc to control the allocation of radio resources to UEs in the cell, and a local controller LC which processes the G- RCD issued by the centralized controller 10 and may modify them before supply to the local scheduler for implementation. The local controller LC and local scheduler LSc may be implemented, for example, using appropriately-programmed processors. Typically, the local scheduler LSc is provided at the local node which controls communications within the cell. When the local controller LC is located at the same geographical location as the local node the local controller LC has access to optimally fresh channel information. The local controller LC and the local scheduler LSc may be integrated into a common module. Thus, for example, in an LTE context, the local controller may be integrated into an eNodeB scheduler.

The specific functions of the centralized and local controllers, and corresponding signalling, are described below.

The centralized controller located at a centralized scheduler node may perform its part in coordinating resource allocation by performing the example method illustrated in the flow diagram of Fig.6. In the example illustrated in Fig.6, the centralized controller starts up in a step SOI and begins to receive periodic transmissions of user channel state information and/or interference measurement information (designated generally by "CSI" in Fig.6) relating to the cooperating cells in the group.

In a step S03 of the method illustrated in Fig.6, the centralized controller computes the global resource coordination decisions (G- CD) to be applied in a cycle N based on the cell signalling of CSIN-I received in step S02 of the method. The centralized controller calculates the G-RCD based on a metric which seeks to maximize the overall performance of the group of cooperating cells, for example the long-term overall network performance. The appropriate metric to use depends on the application (i.e. it depends on the parameter whose allocation is being controlled in a coordinated manner).

In step S04 of the method illustrated in Fig.6, the centralized controller computes the L-CAI in respect of the different cells in the group and in respect of the current cycle N. The centralized controller then transmits both the full set of G-RCD and cell-specific L-CAI information to local controllers associated with the base stations (here, eNodeBs) of the cooperating cells, in step S05 of the method illustrated in Fig.6. Then the flow returns to step S02 for the next cycle, and the centralized controller receives the next set of user channel condition information/interference measurements.

Each local controller associated with a local node of the radio access network may perform its part in coordinating resource allocation for an associated cell Ci by performing the example method illustrated in the flow diagram of Fig.7.

In step 101 of the example method illustrated in Fig.7, the local controller receives the G-RCDIM information for the current cycle N, as well as L-CAI_N for the current cycle and the current cell Ci Next, in step S102, the local controller computes a parameter (a) which serves as an estimate of the performance level that would be achieved by cell Ci during cycle N assuming that the G-RCDN are implemented (by this cell and by the other cooperating cells in the group) and taking into account the up-to-date knowledge that the local controller has regarding user channel conditions and/or interference in the cell Ci.

In a step S103 of the method illustrated in Fig.7, the local controller initializes the value of a counting variable q. Then, in step S04 of the method illustrated in Fig.7, the local controller computes a parameter (b_q). The parameter (b_q) serves as an estimate of the performance level that would be achieved by cell Ci during cycle N assuming that the cell Ci implemented, not the relevant directive(s) specified by G-RCDN , but instead an alternative resource-allocation according to a locally-generated hypothesis L-RCDq . The calculation of (b_q) assumes that the other cooperating cells in the group implement G- CD, and takes into account the up-to-date knowledge that the local controller has regarding user channel conditions and/or interference in the cell Ci.

In a step S105 of the method illustrated in Fig.7, the local controller computes a parameter (c_q). The parameter (c_q) serves as an estimate of the performance loss that would occur in the other cooperating cells in the group during cycle N assuming that the cell Ci implemented the alternative resource-allocation according to L-RCDq . The calculation of (c_q) assumes that the other cooperating cells in the group implement G-RCD, and takes into account the up-to-date knowledge that the local controller has regarding user channel conditions and/or interference in the cell Ci. The calculation of (c_q) is based on the L-CAI_N applicable to cell Ci and received from the centralized controller. If (c_q) has a negative value this indicates that the variant resource-allocation under consideration is expected to produce a performance improvement in the cooperating cells if it were to be implemented.

In a step S106 of the method illustrated in Fig.7, the local controller checks whether the counting variable q has reached its maximum value q_max, that is, whether the local controller has performed computations for the potential resource-allocations according to all the available locally-generated hypotheses. If it is determined in step sl06 that the counting variable q has not reached the maximum value then the value of the counting variable is incremented in step S07 and the flow returns to step S104. On the other hand, if it is determined in step sl06 that the counting variable q has reached the maximum value then the method proceeds to step S08.

It is possible to compute the performance gain that is expected to arise in the local cell Ci if a variant resource-allocation determined by a local hypothesis L-RCDq were to be performed. This performance gain may be computed as (b_q)-(a). If the performance gain of the local cell Ci exceeds the performance loss c_q of neighbour cells, it means the variant resource-allocation corresponding to the local hypothesis L-RCDq may bring additional benefits to the overall performance (i.e. the combined performance of local and neighbour cells in cooperation). So, benefits can be expected if the G-RCD concerning the local cell is updated based on the hypothesis L_RCDqm associated with the positive maximum performance margin. For this reason, in step S08 of the method illustrated in Fig.7, the local controller determines the value (g_qm) of the maximum net performance gain, i.e. the maximum value of [ ((b_q)- (a)) -c_q], and identifies the locally generated hypothesis L-RCDqm which gave rise to this value (g_qm). ln step S109 of the method illustrated in Fig.7, the local controller determines whether this value (g_qm) of the maximum net performance gain is positive or negative. If in step S109 (g_qm) is determined to be negative this is interpreted to mean that application of L-RCDqm, the best locally-generated hypothesis, is expected to produce a net degradation in performance for the cooperating cells considered as a group. Thus it is considered that the centrally-set G-RCD is the best to use for this cell Ci, and the flow proceeds to step SllO of Fig.7 where the G-RCD directive(s) for this cell Ci are output to the scheduler for the cell Ci (in this example, to the scheduler of the associated eNodeB).

On the other hand, if in step S109 (g_qm) is determined to be positive this means that application of L-RCDqm in the local cell is expected to produce a net benefit for the cooperating cells considered as a group. Thus the flow proceeds to step Sill of Fig.7 where the resource allocation G-RCD is updated with L-RCDqm to generate G-RCD* (which includes the locally-taken decision L-RCDqm for the current cell and the centrally-taken decisions for the other cells of the group). The updated G-RCD* is output to the scheduler for the cell Ci.

After scheduling instructions have been output from the local controller to the scheduler the flow returns to step S101 of Fig.7 for implementation of the next cycle in of the method.

It is possible that both ((b_q)-(a)) and the corresponding c_q may be negative, i.e. that the local RCD hypothesis causes a performance loss in the cell of interest but a performance gain in the other cells of the group. In the method illustrated in Fig.7, the local controller is configured to update global decisions as long as the total performance gain for the other cells of the group exceeds the local cell's performance loss. That is, even if the new decision results in performance degradation in the local cell, the loss may be offset by the collective gain of the other cells.

The two-level coordination approach according to embodiments of the invention may be contrasted with a comparative example in which fresh user channel information available at a local cell is used to update global resource-allocation coordination decisions that have been taken at a centralized controller based on outdated information. The approach according to this comparative example has the disadvantage that local cells are lacking in visibility of global channel information; hence the local decisions only aim at enhancing local cell performance. As a result, even on occasions where these decisions may enhance the performance in cells of interest there is a considerable risk that performance of neighbour cells will be degraded more significantly.

In contrast to the comparative example, in the two-level coordination approach according to embodiments of the invention, local decisions which modify the centrally-decided resource-allocation coordination decisions are only taken if it is assessed that the changes are expected to produce net performance improvement for the group of cooperating cells.

In the example discussed above, the factor which determines whether the locally- generated resource allocation is used for the current cell, or the resource allocation specified for the current cell in the centrally-decided G-RCD is used, is whether the local controller determines that the locally-generated resource allocation will produce an improvement in the global performance of the cooperating group of cells. However, the selection may also take into account resource usage in the cell. Thus, for example, in the event that the local controller does not find a local hypothesis which produces a resource variation that improves global system performance, it may nevertheless apply a locally-decided resource allocation if this produces a reduction in resource usage in the cell and there is at least no degradation in the global performance of the group of cooperating cells.

Example Application: Allocation of power values

As an example of implementation of the above-described two-level coordination scheme according to embodiments of the invention, a description will now be given of a power control coordination scheme. In this example, to manage inter-cell interference (ICI) and optimize the overall network performance the two-level coordinated scheduling dictates the RBG-specific transmit power of cooperating cells (i.e. the resource-block-group-specific transmit power).

The following notation is employed below. Let C = {Q,C₂,- - -,C_jy } denote the cooperating cell set, Q stand for the cell of interest (i.e., local cell) and C_;- = C \ Q be the complement set of cell C_i , i.e., all cells excluding C_i (also called "neighbor cells").

The RBG-level transmit power is chosen from a finite set P_G— {Ρ\ , Ρ2 ·_< " ' Ρο ] · ^wni^cn represents the eNodeB's transmit power grade in ascending order, with P ≤ P\ and PQ≤ P_MA . The RBG-specific transmit power of cell C_i is given by P_z- =

; ]^Γ , where the ;^'-th element P_; (/) = _;- (e P_G ) is the transmit power of the / -th RBG and L is the total number of RBGs. Denote by P_c = P₁ χ · · - χ Ρ_ζ· x - - - x V_N the transmit power allocation of N cooperating cells. It can be deduced that

P_c = P^ x P V/ (1). Let H_z- = [hi _z-,h_{2 ζ}· ,··· , h_{£ z}- ]^r and l _t = [^ _ζ· , _{2 ζ}· , · · · , ^ _Ζ· ]^Γ denote user channels and interference measurement of L resource block groups in cell C,- . Correspondingly, H_c = H₁ xH₂ X · Hy and I_C =I₁ xl₂ x~-I_N are the user channel and interference measurement of N cooperating cells.

Let ion of cell Q conditioned on transmit po

the utility realized on RBG /, where utility quantifies the usefulness or desirability of the result achieved using the specified transmit power allocation. Due to the additive property of the utility function, it holds that

L

₉{c_t |P_C , H_C ( , i_c {t)) =∑ p(q , /|P_C , H_C ( , i_c ( ) (2).

1=1

Let >(c|Pc , H_c ( )) denote the sum utility of cells in Cand it holds that

^(C|P_C,H_c( ,I_c( )=∑^(C_; ^.|P_C,H_c( ,I_c( )

,eC

L (3).

=∑ ∑^(c_f,/|P_c,Hc( ,Ic( )

C,eC 1=1

In (3), the second equality is due to the additive property of the utility function

< ( |P H_c(t),I_c( )). It can also be deduced that

,H_c( ,I_c(t)) (4), where ^(c_z- |P_C , H_c (t), I_c (t))= ^ (p{c _j |P_C , H_c (i), I_c (i)) is the sum utility of neighbor

C_j≡C,

cells.

In this example application the centralized controller is configured to determine the cross-cell power allocation to maximize the overall utilities of the group of cooperating cells. Therefore, in this example application, the RCD is the RBG-level transmit power of eNodeBs, which are obtained as

P_c (t) = arg max (c|P _c , H _c (t), I _c (tj)

P_c

L (5).

= arg max∑ ^^(Q,/|P_c,H_c(t),I_c(t))

pc C,eC 1=1

It is known that the joint power allocation P_c is chosen from the space

JV-fold

(?G )^{L x} (?G )^{L x} " ' ^x (?G )^{L anc}' ^tnat' altogether, there are G^NL power allocation hypotheses. Therefore for a practical system (for example, a system having G - 4 , N = 4,and L = 10 ), it not feasible to obtain the optimum power allocation Pc( ) via exhaustive search

(because in this case G —2 « 10 ! )

In practice, it is possible to use numerical methods to arrive at the near-optimum power allocation with a manageable computation complexity. For example, Greedy Search (GS) can be used to solve the joint power allocation problem. GS is well known and so will not be described in detail here. However, a brief description is provided below, for clarification.

Applying the principles of GS, the power allocation is determined cell by cell in sequential order. In each step, the cell power allocation is calculated to maximize the combined performance of the current cell and the cells traversed in previous searching steps. An exemplary algorithm is shown as follows.

Multi-Cell power allocation via Greedy Search

Determine the GS search order according to certain criteria and denote the search order as C2, - - - y without losing generality.

FOR n=2 TO N

Let C_n = {C_n__l , C_n

x P„, H_C„ , I_C J+ <t>{c_n P_c„ , P„, H _CB » ICJ

END FOR OUTPUT P_r = P_r

As illustrated in Figures 4 and 6, the centralized controller also generates the cell- specific local coordination assistant information (L-CAI) that indicates the performance gain/loss of neighbor cells caused by varying the power allocation in local cell Q. To this end, in the present example application the utility derivative of cells in C_;- is used to represent the sensitivity of neighbor cell performance with respect to power variation δΡ,· in cell C_i . Thus, in this exam le the L-CAI information is defined as

(6), ^(C; |P_c , H_c (t),I_c( ) .

where is the partial utility derivative of neighbor cells on BG /

When the new transmit power allocation due to power perturbation ΔΡ_; is defined by

(7),

^(C; , m|P_c , H_c (t),I_c( )

where the second equality holds as = 0, Vm≠ / .

dP l,,i

It may be that in some cases the utility function ^?(C_; |P₍- , H_c (t), I_c (t))may not be differentiable with respect to the power allocation Ρ,· of cell Q . In such a case (as shown in Figure 8), piecewise power differential values ΔΡ_; (k), k— 1,2, · ·^■ and corresponding utility difference values A(p(Cf ) can be used as the L-CAI signaling.

When the centralized controller has determined the G-RCD V_c{t) and L-CAI signaling it transmits this data to connected cells, for example via a

new interface or a known interface (e.g. an enhanced X2 interface).

A description will now be given of the functioning of a local controller in the present example application.

Upon receiving the global decisions P_c( j and L-CAI

, the local

3P;

controller can determine how to update the global decisions with fresh user channel state information H_;(t') where f—t≥D_RTT . (a) The local controller computes local cell performance, given global decision Pc( ), according to the following relation:

L

^(Q|P_c,H_c(t'),I_c(t'))=^^(QJ|P_c,H_c(t'),I_c(t')). (8)

1=1

(b) In this example application the local controller Incorporates fresh channel state

information to update power allocation by implanting the following procedure

Global decision update by local controller

New power allocation is initialized as p^^ew = p_c (multi-cell) and p"^ew = p. (local cell)

FOR 1=1 TO L % traverse over all RBGs

FOR g=l TO G % traverse over power hypotheses

Power allocation hypothesis is initialized as = p"^ew

Let P (l) = P_g and P^ = P₁ x— xP χ-χΡ_λ

Calculate the performance metric of local cell (p{C_j P^ ,H_c(t'),I_c(t') Calculate the corresponding differential performance metric as

Δ^(ς^)=^(ς^ρ ,Η_ε( ,ι_ε( )-^(ς^Ρ ,Η_ε(^ ^( ) Calculate nei hbor cell performance difference due to hypothesis Pj? as

END FOR

Let the survivor hypothesis be H = argmax(A<^(C_;;g) + A<^(c,;g))

g

[NB Summing the two performance difference terms A^( _;; ) and A^(c_;; ) then gives the total net performance gain/loss. So, in this example...]

IF A^(C_;;H) + A^(C_;;H)> 0 THEN

Let Pf ^ew = P a nd P ^ew = P_j x · · · x p x · · · x P^

END END FOR OUTPUT power allocation update P^^ew

Although the above example application concerned coordinated scheduling of power values, the invention is not limited to that application. Fig.9 illustrates an example of functional components that may be provided in a local controller LC. In the example illustrated in Fig.9, the local controller includes a generator module 31 configured to generate the candidates L-RCDq. A performance evaluator module 32 is provided to compute the global performance values ((a), (b_q) and (c_q)) and to determine which of the candidates L-RCDq produces the best global performance. A selector 33 is provided to select which resource allocation (out of the candidates L-RCDq and the resource allocation specified for this cell by the G-RCD) should be applied in the cell. In a case where the selection of the resource allocation takes into account resource usage in the cell, then a resource-usage quantifier 34 may also be provided.

It is to be understood that the various functional components illustrated in Fig.9 may be implemented in different ways (e.g. using hardware, software, firmware). For example, these components may be implemented using one or more suitably programmed processors.

Claims

1 . A method of coordinating allocation of resources in different cells of a group of plural cells in a cellular communications network, the method comprising:

receiving first information at a central controller (CONT_CENT), wherein the received first information is indicative of user channel condition information and/or interference in the group of cells;

generating decision data (G-RCD) by the central controller, wherein the decision data (G-RCD) is indicative of a first allocation of resources in the cells of said group, and the central controller selects the first resource-allocation based on the first information and on a selection criterion prioritizing the global performance level for the group of cells;

the central controller generating, for each cell of the group, coordination assistance information (L-CAI) indicating the expected effect, on the performance of the other cells in the group, produced by a variation by said cell of said first resource-allocation;

transmitting the decision data (G-RCD) and coordination assistance information (L-CAI) to the cells of the group;

generating variant data by a local controller (CONT_LOC) associated with a cell of the group, based on second user channel condition information and/or interference data obtained by said local controller in respect of said cell, wherein the second user channel condition information and/or interference data is more up-to-date than the first information used by the central controller in generating the decision data, and wherein the variant data is indicative of a variant allocation of resources in said cell different from the resource allocation in said cell indicated by the decision data (G-RCD);

evaluating, by said local controller (CONT_LOC), based on coordination assistance information generated by the central controller in respect of said cell, the effect that application of the variant resource allocation would produce on the global performance of said group of cells compared to application of the resource allocation in said cell indicated by the decision data (G-RCD);

based on the result of the global-performance evaluation, selecting between the resource allocation specified by the variant data and the resource allocation specified by the decision data provided by the central controller; and

commanding allocation of resources in said cell according to the selected resource allocation.

2. The resource-allocation coordination method according to claim 1 , wherein the selection chooses the variant resource allocation when the global-performance evaluation indicates that the variant resource allocation would produce an improvement in global performance.

3. The resource-allocation coordination method according to claim 1 , and comprising: quantifying the resources required by the variant resource allocation compared to the resources required by the resource allocation in said cell indicated by the decision data (G- RCD);

wherein the selection between the resource allocation specified by the variant data and the resource allocation specified by the decision data provided by the central controller is based on the result of the resource-quantification as well as the result of the global-performance evaluation.

4. The resource-allocation coordination method according to claim 3, wherein the selection chooses the variant resource allocation in the event that:

the resource-quantification determines that the variant resource allocation requires fewer resources than the resource allocation specified in the decision data received from the central controller, and

the global-performance evaluation indicates that the variant resource allocation would not produce a degradation in global performance.

5. The resource-allocation coordination method according to claim 1 , wherein the step of generating variant data comprises generating plural candidate sets of data (L-RCDq), determining which candidate set of data (L-RCDqm) produces the greatest net improvement in global performance of said group of cells, and selecting as the variant data the candidate set of data (L-RCDqm) determined to produce the greatest net improvement in global performance of the group of cells.

6. A resource-allocation coordination method according to claim 1 , wherein the cellular communications network is an LTE network and the step of commanding allocation of resources in said cell according to the selected resource allocation comprises controlling an ENodeB scheduler to allocate resources in said cell according to the selected resource allocation.

7. The resource-allocation coordination method according to any one of claims 1 to 6, wherein the decision data and variant data define allocations of radio resources.

8. A group-controller apparatus (10) configured to control coordination of resource- allocation in different cells (20-i ,20₂,20₃) of a group of plural cells in a cellular communications network, the group-controller apparatus comprising:

a first calculation unit (12) configured to receive information indicative of user channel condition information and/or interference in the group of cells, and to generate decision data (G- RCD) indicative of a first allocation of resources in the cells of said group, wherein the first calculation unit (12) is configured to evaluate the first resource-allocation as producing a specified global performance level for the group of cells; and a second calculation unit (14) configured to generate, for each cell of the group, coordination assistance information (L-CAI) indicating the expected effect, on the performance of the other cells in the group, produced by a variation by said cell of said first resource- allocation;

wherein the group-control apparatus is configured to transmit the decision data (G-RCD) and coordination assistance information (L-CAI) to the cells of the group.

9. A group-controller apparatus (10) according to claim 8, wherein the first and second calculation unit are provided by a single calculation unit.

10. A local controller apparatus (21 ) associated with a cell (20-i ) in a cellular communications network, wherein said cell (20-i ) is one of a group of plural cells and the local controller apparatus (21 ) is configured to control allocation of resources in said cell (20-i ) in coordination with resource-allocation in the other cells (20₂,20₃) of said group, the local controller apparatus (21 ) comprising:

a receiver configured to receive, from a group controller apparatus: decision data (G- RCD) indicative of a first allocation of resources in the cells of said group, and coordination assistance information (L-CAI) indicating the expected effect, on the performance of the other cells in the group, produced by a variation by said cell of said first resource-allocation;

a variant-data generator configured to generate variant data indicative of a variant allocation of resources in said cell different from the resource allocation in said cell indicated by the decision data (G-RCD), wherein the variant data generator is configured to generate the variant data based on up-to-date user channel condition information and/or interference data obtained by the local controller apparatus in respect of said cell, said user channel condition information/interference data being more up-to-date than user channel condition information/interference data used in generation of the decision data at the group controller apparatus;

a variant-data evaluator configured to evaluate, based on received coordination assistance information, the effect that application of the variant resource allocation would produce on the global performance of said group of cells compared to application of the resource allocation in said cell indicated by the decision data (G-RCD);

an output module configured to output a command commanding allocation of resources in said cell according to the resource allocation selected by the selector.

1 1 . An eNodeB of an LTE cellular communications network, comprising a local controller apparatus (21 ) according to claim 6.