Pair selection for calibration

Abstract

The invention relates to selection of pairs for calibration. The present application relates to a method for calibrating M radio units in a cellular network for data transmission using distributed multiple-input multiple-output (D-MIMO), the method comprising the following steps at a control entity: selecting calibration pairs from the M radio units, and initiating a calibration with respect to the other in each of the selected calibration pairs. The selection of the calibration pair includes providing an undirected graph having M radio units as nodes, the nodes being connected by weighted edges, where the weighted edges represent signal-to-noise ratio parameters between the radio units of the corresponding calibration pair. An optimized sub-graph having M nodes and a first number of edges is determined from the undirected graph, where the sum of functions of the signal-to-noise ratio parameters is optimized, and where the diameter of the optimized sub-graph is limited to a defined number.

Description

Technical Field

[0001] This application relates to a method for calibrating M radio units in a cellular network using a control entity for data transmission using distributed multiple-input multiple-output MIMO (also known as cellular-free MIMO), and a corresponding control entity. Furthermore, a computer program including program code and a carrier including the computer program are provided. Background Technology

[0002] Distributed MIMO (D-MIMO) is a promising radio access network technology in which many radio units (RUs), such as Figure 1 The wireless units RU1 to RU7 shown, including antennas, power amplifiers, mixers, oscillators, etc., are geographically distributed to increase macro diversity and mitigate path loss and shadowing effects. RUs are connected to a central processing unit (CP) via wired / wireless fronthaul links to enable coordination between different RUs and perform highly complex centralized baseband processing. In 5G terminology, the CP can be implemented in either a distributed unit (DU) or a central unit (CU).

[0003] User equipment (UE) connected to a network can be served by multiple RUs to improve spectrum efficiency (SE), reliability, and coverage probability. It is well known that to maximize the benefits of multi-RU joint transmission, time / frequency / amplitude / phase calibration should be performed on the different RUs to enable coherent joint transmission (CJT).

[0004] To calibrate phase misalignment between RU pairs, S. Xu, Y. Cao, C. Li, D. Wang, and L. Yang proposed an air calibration scheme in their paper "Spanning Tree Approach for Air Channel Calibration in 6G Cellular Massive MIMO" published in the *IEEE Transactions on Wireless Communications*. In this scheme, bidirectional pilot signals are transmitted between some RU pairs to calibrate the phase misalignment between these pairs. Each RU should be included in this calibration process to calibrate all networks. Figure 2 The diagram illustrates the over-the-air calibration process for a pair of RUs. During the calibration operation, the RUs form beams to send / receive calibration signals to / from other RUs. These beams are used to amplify the calibration signal strength at the desired RU.

[0005] Figure 2 An example bidirectional calibration process is shown. In this example, the transmit and receive beams for RU 2 and RU 4 are illustrated. During the calibration process for these two RUs, each RU uses the same beam to transmit signals to / receive signals from the other RU.

[0006] Combination Figure 2To explain, phase calibration between RUs in a D-MIMO network presents several potential problems, including calibration error propagation when some RUs are indirectly calibrated using a series of measurements to correct phase misalignment. This is because the calibration process includes errors due to thermal noise at the RU receivers. For example, if RU-3 and RU-5 should be indirectly calibrated using calibration pairs (RU-3, RU-7), (RU-7, RU-6), and (RU-6, RU-5), then the variance of the phase error is the sum of the variances of these three measurements, such as... Figure 2 As shown.

[0007] Therefore, an improved process is needed to select RU pairs and calibrate the selected RU pairs. Summary of the Invention

[0008] The aforementioned requirements for the improved process are satisfied by the features of the independent claim. Other aspects are described in the dependent claims.

[0009] According to a first aspect, a method is provided for calibrating M radio units in a cellular network for data transmission using distributed multiple-input multiple-output (D-MIMO), wherein the method includes the following steps at a control entity: selecting calibration pairs from the M radio units, and initiating calibration relative to the other in each of the selected calibration pairs. The selection of calibration pairs includes the step of providing an undirected graph having M radio units as nodes connected by weighted edges, wherein the weighted edges represent signal-to-noise ratio (SNR) parameters between the radio units of the corresponding calibration pair. An optimized subgraph having M nodes and a first number of edges is determined from the undirected graph, wherein the sum of functions of the SNR parameters is optimized, and the diameter of the optimized subgraph is limited to a defined number. Furthermore, a corresponding control entity including at least one processing unit and a memory is provided, wherein the control entity is configured to operate as described above or as discussed in more detail below.

[0010] Using this invention, calibration pairs with low signal-to-noise ratios can be effectively selected, while ensuring that all nodes of the graph corresponding to the radio unit are included in the calibration process.

[0011] In addition, a computer program including program code is provided, which is executed by at least one processing unit of a control entity, wherein the execution of the program code causes the at least one processing unit to perform the method as described above or further detailed below.

[0012] Last but equally important, a carrier comprising a computer program is provided, wherein the carrier is one of electronic signals, optical signals, radio signals, and computer-readable storage media. Attached Figure Description

[0013] Figure 1An example overview of a MIMO network is shown, where calibrations for different value units are to be performed.

[0014] Figure 2 It shows Figure 1 The diagram shows a network that uses a directional calibration process and indirect calibration between radio units.

[0015] Figure 3 An example schematic diagram of a network with different radio units is shown, wherein calibration incorporating features of the present invention is performed in the control entity.

[0016] Figure 4 A schematic diagram of a flowchart is shown, which includes steps for selecting a subgraph to be used in the method for finding calibration pairs.

[0017] Figure 5 A schematic diagram of a flowchart is shown, which includes steps from using... Figure 4 The steps involved are to find the optimal subgraph within the determined intermediate subgraph.

[0018] Figure 6 A schematic diagram of a method for determining calibration pairs and calibrating corresponding pairs, incorporating features of the present invention, is shown.

[0019] Figure 7 The configuration is shown to select a calibration pair and perform a binding. Figures 3 to 5 A schematic example of the control entity for calibration discussed is shown.

[0020] Figure 8 A schematic diagram of the cumulative distribution function (CDF) for different methods used to find calibration pairs is shown.

[0021] Figure 9 This is a schematic diagram of CDF for the signal-to-noise ratio of different methods.

[0022] Figure 10 The diagram illustrates the percentage of calibration pairs used for different methods.

[0023] Figure 11 The CDF over signal-to-noise ratio is shown, with a comparison of two methods for 64 radio units.

[0024] Figure 12 The graph shows the percentages used by two different methods. Detailed Implementation

[0025] In the following, embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be understood that the following description of the embodiments should not be considered limiting. The scope of the invention is not intended to be limited by the embodiments or drawings described below, which are merely illustrative.

[0026] The accompanying drawings should be considered schematic, and the elements shown are not necessarily to scale. Rather, the various elements are shown such that their function and general purpose will be obvious to those skilled in the art. Any connections or couplings between functional blocks, devices, or components of the physical or functional units shown in the drawings and described below may also be achieved through indirect connections or couplings. Couplings between components may be established through wired or wireless connections. Functional blocks may be implemented using hardware, software, firmware, or a combination thereof.

[0027] In the context of this application, the term "mobile entity" or "user equipment" (UE) refers to a device used, for example, by a person (i.e., a user) for their personal communications. It can be a telephone-type device, such as a telephone or Session Initiation Protocol (SIP) or Voice over IP (VoIP) phone, cellular phone, mobile station, cordless phone, or a personal digital assistant-type device such as a laptop computer, notebook computer, or tablet computer equipped with wireless data connectivity. The UE can also be associated with non-human entities such as animals, plants, or machines. The UE can be equipped with a SIM (Subscriber Identity Module) or an electronic SIM, which includes a unique identity associated with the user using the UE, such as an IMSI (International Mobile Subscriber Identity), TMSI (Temporary Mobile Subscriber Identity), or GUTI (Globally Unique Temporary UE Identity). The presence of a SIM within the UE uniquely customizes the UE through the user's subscription.

[0028] For clarity, it's important to note that while there are differences between users and subscribers, they are also closely related. A user accesses the network by acquiring a network subscription, thus becoming a subscriber within the network. The network then identifies subscribers (e.g., via IMSI, TMSI, or GUTI) and uses the associated subscription to identify the relevant subscriber data. A user is the actual user of the UE, and a user may also be a user who owns a subscription, but the user and the subscription owner may be different. For example, the subscription owner could be a parent, and the actual user of the UE could be that parent's child.

[0029] A method is explained in more detail below that helps to find suitable radio unit pairs that effectively utilize distributed MIMO (e.g., coherent joint transmission) in cellular networks and initiate calibration in the determined calibration pairs. As will be discussed further below, calibration quality in the distributed MIMO network is optimized by determining weights for each edge using a decreasing function of the calibration signal-to-noise ratio values ​​of the radio unit pairs, where the total weight of the selected subgraph is minimized. A method is provided that minimizes error propagation during phase calibration of radio units in a network using the shortest path between nodes, where the algorithm is used to limit the diameter of the resulting subgraph.

[0030] Figure 3A schematic diagram of a distributed MIMO network including radio units RU 201 to 207 is shown, as follows. Figure 3 As shown, these radio units are also referred to as RUs 1 through RUs 7. A central control entity 100 (or central processing unit CP) is provided, which now selects the optimal calibration pair in such a way that all radio units are included in the calibration, and in which the calibration pair is selected such that the SNR is optimized by maximizing the signal-to-noise ratio (SNR) or minimizing the weights between the RUs. Furthermore, the control entity can initiate or trigger calibration in the selected calibration pair. Initiation can mean that the control entity itself performs the calibration, or that it triggers calibration in a different calibration pair that performs the calibration.

[0031] The bidirectional calibration process involves joint estimation of the channels between RUs 201 and 207 and the phase offsets between them. Assuming the effective channels are accurately estimated, the following calibration signal model can be obtained:

[0032]

[0033]

[0034] in It is in the The calibration signal received at each RU, It is the first The RU and the first The effective channel between the two RUs includes the transmit and receive beamformers of both RUs. It is the first The pilot signal sent by each RU for calibration, It is the first The RU and the first Phase offset between RUs It is the first Thermal noise at the RU. For the thermal noise at the RU. The first RU sends a calibration pilot signal and is then sent by the first The terminology is similarly defined for the other direction of reception by a RU. Here, we assume channel reciprocity, that is... And the statistics of the same noise and pilot signals at the two RUs can be used to determine the first The RU and the first The calibration signal-to-noise ratio (SNR) between each RU is defined as follows:

[0035] in It is the effective large-scale path loss between these two RUs (including transmit and receive beamforming gain). It is the calibrated signal power. This refers to the thermal noise power at the two RUs. In the channel... After being estimated, the signal model can be simplified to

[0036] Calibrate SNR value ( This determines the calibration quality between RU pairs.

[0037] To overcome the problem of calibration error propagation, the theoretically best method is to complete the calibration process using a full-to-full approach. However, this method requires... One calibration pair, of which This represents the total number of RUs. Due to the significant overhead of the calibration process, high energy consumption, and high computational complexity, it may be impractical in practice. It can be used... Each pair defines a calibration process (forming a path through each RU when each RU is selected only once) to minimize the total number of calibration pairs. However, this can lead to poor calibration quality because some selected pairs may have lower calibration SNR values.

[0038] Based on the phase misalignment between RU pairs, the RU pair selection method can be used for in-flight phase calibration in D-MIMO. Phase misalignment is defined as a decreasing function of the corresponding calibration SNR. RU pairs can be selected simply by comparing the phase misalignment value with a predetermined threshold. Only RU pairs with phase misalignment values ​​less than the threshold are selected for calibration. This problem can be modeled as a weighted graph edge selection problem, where only edges with sufficiently small weights are selected. The problems with this method can be summarized as follows: The resulting graph may be disconnected. Although it can be decomposed into connected subgraphs, in this case, not all RUs can be used for coherent joint transport, which limits performance. Furthermore, even if the graph is connected, long paths may exist between some vertices, causing calibration errors to propagate between corresponding RUs.

[0039] Therefore, a method is proposed in which (a) the method selects RU pairs with high calibration SNR values, (b) the method selects a small number of RU pairs for calibration, and (c) the proposed method has low complexity and can therefore be implemented in D-MIMO networks with a large number of RUs.

[0040] Through (a), (b) and (c), D-MIMO networks can be calibrated with low energy consumption and in a scalable manner, with low phase error and small calibration overhead.

[0041] This method will combine Figure 4 and Figure 5 To provide a more detailed explanation.

[0042] As explained below, a spanning tree algorithm can be used. This algorithm finds a connected subset of a graph G that contains every node in G and has no cycles. A minimum spanning tree (MST) is a spanning tree with the minimum sum of edge weights. Different algorithms exist to find the minimum spanning tree of a graph. (See reference) Figure 4 In step S41, the MST is initialized as a graph with all vertices and no edges. Kruskal's MST algorithm is a solution for finding the MST. Typically, this algorithm first sorts the edges according to their weights and begins adding edges starting with the smallest weight, as long as it does not cause a cycle in S. The algorithm iterates until all vertices are included in the resulting subgraph. Kruskal's algorithm has a time complexity of O(N log M), where N is the number of edges in graph G and M is the number of vertices in graph G.

[0043] More specifically, the algorithm includes the initialization step S41 described above. In step S42, each vertex is placed in a different set. In step S43, the edges of the graph are sorted in ascending order of their respective weights as e1, e2, ..., e n In step S44, the connection vector is selected. u and v The edges, and check them in step S45. u and v Are they in the same set? If so, discard the corresponding edge in step S46. u and v If they are not in the same set, then in step S47, the corresponding edges are added to the MST and merged. u and v Two sets. In step S48, check if any edges remain, and repeat steps S44 to S47 until all edges have been selected and checked. At the end of step S48, MST contains a subgraph of G that has all vertices and includes only the minimum number of weighted edges to create a minimum spanning tree. Therefore, an intermediate subgraph G1 with all vertices is provided, which means all radio units present in the network.

[0044] exist Figure 5 Continuing in the middle, the subgraph G1 is then further used. Figure 5 The steps shown are related to the binary search, where in finding Figure 4 After the initial or intermediate subgraph shown, the iterative process is executed.

[0045] In step S51, in Figure 4 Subgraph found after the steps The weights of the undrawn edges are sorted in ascending order. Let the corresponding edges be... in It can be added to a subgraph. The number of edges.

[0046] In step S52, set and .

[0047] For the determination in steps S53 and S54 For each value, the following operation is performed: S55: Assessment , S56: Side Add to And evaluate the diameter of the obtained graph. - S58: If the diameter is less than or equal to Then set .

[0048] - S57: Otherwise, set .

[0049] Finally, in the step not shown, by using the edges Add to The final subgraph is then evaluated.

[0050] Figure 5 The dichotomy will New edge added After each iteration, the number of candidate edges to be added is halved. Therefore, at most... After several iterations, it converges. At the convergence point, the diameter condition should be satisfied because at least one subgraph satisfies the diameter condition (the complete graph has a diameter of 1 and...). ).

[0051] The following will provide a more detailed analysis and combination. Figure 4 and Figure 5 The complexity of the methods discussed.

[0052] Steps S41 to S48: Kruskal's algorithm has a time complexity of O(n log n) for solving the MST. ,in It is the number of vertices. It is the number of edges.

[0053] Step-by-step analysis algorithm: - Placing edges in the list requires .

[0054] Sorting the edges requires .

[0055] - Finding the union requires .

[0056] Therefore, the complexity can be calculated as In this example, the initial graph is complete, therefore Therefore, the overall complexity of steps S41 to S48 becomes...

[0057] because .

[0058] The complexity of steps S51 to S58 is as follows: When using a centralized algorithm, the complexity of evaluating the diameter of the graph obtained after each iteration from steps S55 to S58 is . However, when a distributed algorithm is preferred, the complexity can be as low as... However, there is a trade-off in terms of message complexity. Therefore, the overall complexity of step 2 becomes...

[0059] The total complexity of this algorithm is equal to

[0060] in This represents the number of RUs. The results show that the proposed algorithm has low complexity and can be implemented in D-MIMO networks with a large number of RUs.

[0061] The following numerical simulations will be discussed in more detail to demonstrate the effectiveness of the proposed method.

[0062] Randomly placed within a 300m x 300m area (1000 Monte Carlo trials) =6 RUs, and generate calibrated SNR values ​​based on the distance between RU pairs using the following path loss and shading formulas:

[0063] in It is distance (in meters). It is the first The and the first Path loss between RUs (in dB). dB Meters. Assume a log-normal shading pattern with a standard deviation of 8 dB.

[0064] Furthermore, it is assumed that each RU has a transmit power of 0.2W and a receiver noise power of -92 dBm for calibrating signal transmission / reception. Finally, it is assumed that... .

[0065] Three methods were compared in the simulation: Method 1 (Exhaustive Search): This method attempts all possible subgraphs to find the optimal solution. This method is impractical due to the presence of multiple subgraphs. It is only used for comparative purposes to see the performance ceiling.

[0066] Method 2 (discussed and proposed above): This is a two-step method that can find an effective solution. As mentioned above, its complexity is very low.

[0067] Method 3 (All Pairs): This method uses the complete graph directly by selecting all RU pairs.

[0068] Two scenarios were considered: Scenario 1 (Minor Settings): In this case, adopt... Compare methods 1-3. Due to the total number of subgraphs... The size is not too large, so an exhaustive search method can be performed in this case.

[0069] exist Figure 8 In China, targeting The cumulative distribution function of the total weight values ​​obtained by each of the three methods was plotted. It can be observed that selecting all pairs produces a very large total weight, indicating the importance of subgraph selection. On the other hand, the proposed method has a very close sum-weight distribution compared to the optimal solution.

[0070] Figure 9 The calibration SNR distributions obtained by these three methods are shown. It can be observed that the proposed method can select RU pairs with larger calibration SNR values. The proposed method can select subgraphs with a median calibration SNR value greater than 9 dB, where the median corresponds to the calibration SNR value at CDF=0.5. Furthermore, the calibration SNR distribution for the proposed method approaches an exhaustive search, indicating that its performance is close to optimal.

[0071] Ultimately, we can compare the number of RU pairs selected for each of the three methods. From Figure 10 As can be seen, the proposed method selects less than half of the RU pairs on average. Compared to selecting all pairs, using this property can reduce calibration signaling overhead and power consumption by 57%. Compared to the optimal impractical method, it selects only 9% more pairs.

[0072] Scenario 2 (Large Settings): To observe the results for larger settings, consider... Each RU is randomly placed in a 500m x 500m area. Other parameters are selected accordingly. The situation is the same. In such a large setting, the complexity of an exhaustive search method is too high to execute. [The sentence is incomplete and requires further context.] Figure 11 and Figure 12 The results are given in the text.

[0073] exist Figure 11 and 12 In this study, the distribution of calibration SNR and the percentage of RU pairs used by the proposed method were compared with those used by the baseline method, which directly used all RU pairs for calibration. Based on the results, it can be observed that the proposed method selects RU pairs with a median calibration SNR approximately 15 dB higher than the case where all RU pairs are selected. Furthermore, it selects only approximately 25% of all possible RU pairs. The results demonstrate that this method provides an efficient RU pair selection for phase calibration in D-MIMO.

[0074] Figure 6 Summary and combination Figure 4 and Figure 5 The discussion covers some steps performed during the calibration process. In step S60, a calibration pair is selected from M radio units. (As in...) Figure 4 and Figure 5 Specifically, this step includes providing an undirected graph with M nodes in step S61, and determining an optimized subgraph with M nodes and a first number of edges from the undirected graph in step S62, wherein the sum of functions of the signal-to-noise ratio parameters is optimized, and the diameter of the optimized subgraph is constrained within a specified diameter. Any number can be used for the diameter, which also depends on the number of radio units in the network. One possible number is between 3 and 6, but as mentioned above, this number may vary, being larger or smaller depending on the number of radio units. It is well known that the graph diameter is the longest and shortest path between any two graph vertices, or in other words, the maximum number of vertices that must be traversed to get from one vertex to another, without considering backtracking, detours, or loops. In step S62, the optimized subgraph is determined by minimizing the sum of the weights of the edges in the subgraph.

[0075] In step S70, calibration is then initiated for the selected calibration pair.

[0076] Figure 7A schematic architecture diagram of a control entity 100 is shown, which can perform the determination of the calibration pair and the initiation of calibration for the determined pair. The control entity includes an interface 110 provided for sending control messages or signal data between different entities, and wherein the interface is also provided for receiving data such as calibration signals or control messages from other entities. Entity 100 also includes a processing unit 120 responsible for the operation of the control entity 100. The processing entity 120 may include one or more processors and can execute instructions stored on memory 130, wherein the memory may include read-only memory, random access memory, mass storage, hard disk, etc. The memory may also include suitable program code executed by the processing unit 120 to implement the functions described above involving the control entity 100.

[0077] In conclusion, some general conclusions can be drawn.

[0078] An undirected graph comprising M radio units and weighted edges can be constructed, such that the weights of the weighted edges decrease as the signal-to-noise ratio (SNR) parameter increases. Furthermore, to determine an optimal subgraph, the sum of the weights of the first edge in the subgraph can be minimized. Depending on the definition of the SNR parameter, this process can also include maximizing the sum of the weights.

[0079] The step of determining an optimal subgraph may include identifying an intermediate subgraph with M nodes and the minimum sum of the weights of the edges present in the intermediate subgraph, wherein it is determined whether the diameter of the intermediate subgraph is less than a defined number. If not, more edges are added to the intermediate subgraph until the diameter of the intermediate subgraph is less than the defined number. Preferably, an intermediate subgraph with a diameter less than the defined number is an optimal subgraph.

[0080] Spanning tree algorithms, preferably minimum spanning tree algorithms, can be used to determine intermediate subgraphs.

[0081] The calibration process may include phase calibration between different radio units.

[0082] Data transmitted by different radio units in a cellular network can be transmitted using coherent joint transmission.

[0083] Calibration for at least some calibration pairs can be performed using an intermediate radio unit, through which calibration signals are transmitted between the two radio units of the corresponding calibration pair. Therefore, as... Figure 2 As shown, calibration for RU 3 and RU 5 can be performed via RU 7 and RU 6.

[0084] Furthermore, an optimized subgraph containing at least one weighted edge between the two radio units can only be determined when calibration between the two radio units is performed as one of the calibration pairs.

[0085] The steps for determining the intermediate subgraph may include sorting all possible weighted edges by weight and starting with the weighted edge with the smallest weight to add weighted edges to the M radio units, as the weight increases, until all M nodes are included in the intermediate subgraph.

[0086] Furthermore, more edges can be added during the interaction until the diameter of the intermediate subgraph is smaller than the defined diameter.

[0087] In the above solution, it is assumed that the proposed algorithm is executed at control entity 100 using all calibrated signal-to-noise ratios (SNRs) between all radio unit pairs. It can be assumed that the radio units are not mobile, and therefore the radio unit / receiver channel is quasi-static, with the calibrated SNR values, which depend on long-term channel statistics, being slowly varying quantities. Therefore, the calibrated SNRs between radio unit pairs can be obtained in advance without increasing system overhead.

[0088] An approach is outlined for determining a subgraph to reduce calibration overhead. This approach uses a graph theory method to efficiently select suitable radio unit pairs that minimizes error propagation during phase calibration using the shortest path between nodes. The resulting subgraph has a finite diameter, and a weight is determined for each edge such that a decreasing function of the calibration signal-to-noise ratio is applied to the radio unit pairs, and the total weight of the selected subgraph is minimized.

Claims

1. A method for calibrating M radio units (201-207) in a cellular network for data transmission using distributed multiple-input multiple-output MIMO, the method comprising, at a control entity: - Select a calibration pair from the M radio units. - In each of the selected calibration pairs, initiate calibration relative to each other, wherein selecting the calibration pairs includes: - Provide an undirected graph with M radio units as nodes, the nodes being connected by weighted edges, where each weighted edge represents the signal-to-noise ratio parameter between the radio units of a corresponding calibration pair. - Determine an optimized subgraph with M nodes and a first number of edges from the undirected graph, wherein the sum of functions of the signal-to-noise ratio parameters is optimized, and wherein the diameter of the optimized subgraph is restricted to a defined number.

2. The method of claim 1, wherein the weight of the weighted edge decreases as the signal-to-noise ratio parameter increases, and the sum of the weights of the edges in the subgraph is minimized in order to determine the optimal subgraph.

3. The method according to claim 1 or 2, wherein determining the optimized subgraph comprises: - Determine the intermediate subgraph with M nodes and the minimum sum of the weights of the weighted edges present in the intermediate subgraph. - Determine whether the diameter of the intermediate subgraph is less than the defined number. If not, add more edges to the intermediate subgraph until the diameter of the intermediate subgraph is less than the defined number. The intermediate subgraph with a diameter less than the defined number is the optimized subgraph.

4. The method of claim 3, wherein the intermediate subgraph is determined using a spanning tree algorithm.

5. The method according to claim 4, wherein the spanning tree algorithm uses the minimum spanning tree algorithm.

6. The method according to any of the preceding claims, wherein calibration includes phase calibration.

7. The method according to any of the preceding claims, wherein the data is transmitted using coherent joint transmission in the cellular network.

8. The method according to any of the preceding claims, wherein calibration for at least some of the calibration pairs is performed via an intermediate radio unit, through which calibration signals are transmitted between two radio units of the corresponding calibration pair.

9. The method according to any preceding claim, wherein the optimized subgraph is determined such that it contains at least one weighted edge between the two radio units only when calibration between the two radio units is to be performed as one of the calibration pairs.

10. The method according to any one of claims 3 to 9, wherein determining the intermediate subgraph comprises sorting all possible weighted edges according to their weights, and starting with the weighted edge having the smallest weight, adding more weighted edges to the M radio units, as the weights increase, until all M nodes are included in the intermediate subgraph.

11. The method of claim 10, wherein more weighted edges are added during the iteration process until the diameter of the intermediate subgraph is less than the defined number.

12. A control entity configured to calibrate M radio units in a cellular network, the radio units transmitting data using distributed multiple-input multiple-output (MIMO), the control entity comprising at least one processing unit and a memory, the control entity being configured to: - Select a calibration pair from the M radio units. - In each of the selected calibration pairs, initiate calibration relative to each other, wherein selecting the calibration pairs includes: - Provide an undirected graph with M radio units as nodes, the nodes being connected by weighted edges, where each weighted edge represents the signal-to-noise ratio parameter between the radio units of a corresponding calibration pair. - Determine an optimized subgraph with M nodes and a first number of edges from the undirected graph, wherein the sum of functions of the signal-to-noise ratio parameters is optimized, and wherein the diameter of the optimized subgraph is restricted to a defined number.

13. The control entity of claim 12, wherein the weight of the weighted edge decreases as the signal-to-noise ratio parameter increases, and the control entity is further configured to determine the optimized subgraph to minimize the sum of the weights of the edges in the subgraph.

14. The control entity according to claim 12 or 13 is further configured to determine the optimized subgraph, in order to - Determine the intermediate subgraph with M nodes and the minimum sum of the weights of the edges present in the intermediate subgraph. - Determine whether the diameter of the intermediate subgraph is less than the defined number. If not, add more edges to the intermediate subgraph until the diameter of the intermediate subgraph is less than the defined number. The intermediate subgraph with a diameter less than the defined number is the optimized subgraph.

15. The control entity of claim 14 is further configured to use a spanning tree algorithm to determine the intermediate subgraph, and preferably, the spanning tree algorithm uses a minimum spanning tree algorithm.

16. The control entity according to any one of claims 12 to 15 is further configured to calibrate the M radio units to perform phase calibration of the M radio units.

17. The control entity according to any one of claims 12 to 16, wherein the data is transmitted in the cellular network using coherent joint transmission.

18. The control entity according to any one of claims 12 to 17 is further configured to use an intermediate radio unit to calibrate at least some of the calibration pairs, and to transmit calibration signals between two radio units of a corresponding calibration pair via the intermediate radio unit.

19. The control entity according to any one of claims 12 to 18 is further configured to determine the optimized subgraph such that it contains at least one weighted edge between the two radio units only when calibration between the two radio units is to be performed as one of the calibration pairs.

20. The control entity according to any one of claims 14 to 19 is further configured to, in order to determine the intermediate subgraph, sort all possible weighted edges by their weights, and begin adding more weighted edges to the M radio units using the weighted edge with the smallest weight, as the weights increase, until all M nodes are included in the intermediate subgraph.

21. The control entity of claim 20 is further configured to add more weighted edges during the iteration process until the diameter of the intermediate subgraph is less than the defined number.

22. A computer program comprising program code executed by at least one processing unit of a control entity, wherein execution of the program code causes the at least one processing unit to perform the method as claimed in any one of claims 1 to 11.

23. A carrier comprising a computer program according to claim 22, wherein the carrier is one of an electronic signal, an optical signal, a radio signal, and a computer-readable storage medium.

Images