Control system, control method, and control program

Abstract

A first control system (10) comprises: a collection unit (11) for collecting identification data that is used for identification by an identification model that identifies a control pertaining to a wireless network, such as a wireless access network; and a transmission unit (12) for transmitting the identification data collected by the collection unit (11) to a second control system (20) that carries out training by means of a training model that learns a control pertaining to a wireless network, such as a wireless access network, said identification data being transmitted as training data used in the training model.

Description

Technical Field

[0001] The present disclosure relates to a control system, a control device, a control method, a control program, and a non-temporary computer-readable medium.

Background Art

[0002] In recent years, as a wireless communication technology for realizing high capacity, low latency, and multi-connectivity, the introduction of 5G (5th Generation) has been promoted. In next-generation wireless communication systems including 5G, in order to cope with the sophistication and complexity of the system, the openization of the RAN (Radio Access Network) has been promoted, and in the O-RAN (Open Radio Access Network) Alliance, the openization and intelligentization of the RAN have been discussed.

[0003] Related Patent Document 1 and Non-Patent Document 1 describe that Non-RT (Real Time) RIC and Near-RT RIC are provided as RICs (RAN Intelligent Controllers) that utilize AI / ML (Artificial Intelligence / Machine Learning) to intelligently control the RAN. The Near-RT RIC is arranged near the E2 node including the O-DU (O-RAN Distributed Unit) and the O-CU (O-RAN Central Unit) and controls the RAN in quasi-real time. The Non-RT RIC is arranged at a location away from the E2 node and controls the RAN in non-real time.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Non-Patent Documents

[0005] [Non-Patent Document 1] O-RAN ALLIANCE, O-RAN Working Group 2,"AI / ML workflow description and requirements", Technical Report, O-RAN.WG2.AIML-v01.03, 2021.07.20 [Overview of the project] [Problems that the invention aims to solve]

[0006] According to Patent Document 1 and Non-Patent Document 1, it is possible to place an inference model for inferring RAN control and a learning model for training to build the inference model in either a Near-RT RIC or a Non-RT RIC, or to distribute them. This allows for control by the inference model to be performed while the learning model is being trained using data from the operational environment. However, since it is necessary to collect training data in order to train the learning model, this may place a load on the transmission lines and equipment used to transmit the data.

[0007] In view of these challenges, one of the objectives of this disclosure is to provide a control system, a control device, a control method, and a non-temporary computer-readable medium that can reduce the load. [Means for solving the problem]

[0008] The control system relating to this disclosure comprises a specific model for identifying control relating to a wireless network, a collection means for collecting specific data used for such identification, and a transmission means for transmitting the specific data to another control system that performs learning by the learning model, as learning data used by the learning model for learning control relating to the wireless network.

[0009] The control system relating to this disclosure includes receiving means for receiving identification data collected by another control system to identify control relating to a wireless network using a specific model, as learning data from the other control system, and learning means for learning control relating to the wireless network using a learning model with respect to the learning data.

[0010] The control device according to this disclosure comprises a collection means for collecting identification data used for the identification of a specific model that identifies control relating to a wireless network, and a transmission means for transmitting the identification data to another control system that performs learning by the learning model, as learning data used by the learning model that learns control relating to the wireless network.

[0011] The control device according to this disclosure comprises a receiving means for receiving identification data collected by another control system to identify control relating to a wireless network using a specific model, as learning data from the other control system, and a learning means for learning control relating to the wireless network using a learning model with respect to the learning data.

[0012] The control method relating to this disclosure involves a specific model that identifies control relating to a wireless network, which collects specific data used for the identification, and transmits the specific data to another control system that performs learning by the learning model, as learning data used by the learning model that learns control relating to the wireless network.

[0013] The control method relating to this disclosure involves receiving identification data collected by another control system to identify control related to a wireless network using a specific model, as training data from the other control system, and using the training data to learn control related to the wireless network using a learning model.

[0014] The non-temporary computer-readable medium relating to this disclosure is a non-temporary computer-readable medium that stores a control program that causes a computer to execute a process in which a specific model that identifies controls relating to a wireless network collects specific data used for such identification, and transmits the specific data as training data used by a learning model that learns controls relating to the wireless network to another control system that learns using the learning model.

[0015] The non-temporary computer-readable medium relating to this disclosure is a non-temporary computer-readable medium on which a control program is stored that causes a computer to execute a process in which it receives identification data collected by another control system to identify controls relating to a wireless network using a specific model, as training data, and uses the training data to learn controls relating to the wireless network using a learning model. [Effects of the Invention]

[0016] This disclosure provides a control system, control device, control method, control program, and non-temporary computer-readable medium that can reduce the load. [Brief explanation of the drawing]

[0017] [Figure 1] This is a configuration diagram showing an overview of the first control system according to an embodiment. [Figure 2] This is a configuration diagram showing an overview of the second control system according to the embodiment. [Figure 3] This is a configuration diagram showing an overview of the first control device according to the embodiment. [Figure 4] This is a configuration diagram showing an overview of the second control device according to the embodiment. [Figure 5] This is a flowchart showing an overview of the first control method according to the embodiment. [Figure 6] This is a flowchart showing an overview of the second control method according to the embodiment. [Figure 7]It is a configuration diagram showing a configuration example of a RAN system according to Embodiment 1. [Figure 8] It is a diagram for explaining the data flow in the comparative example. [Figure 9] It is a diagram for explaining the data flow in Embodiment 1. [Figure 10] It is a configuration diagram showing a configuration example of a Near-RT RIC according to Embodiment 1. [Figure 11] It is a configuration diagram showing a configuration example of a Non-RT RIC according to Embodiment 1. [Figure 12] It is a flowchart showing an overview of the learning process in the RAN system according to Embodiment 1. [Figure 13] It is a flowchart showing an operation example of the learning process in the RAN system according to Embodiment 1. [Figure 14] It is a sequence diagram showing an operation example of the learning process in the RAN system according to Embodiment 1. [Figure 15] It is a configuration diagram showing a configuration example of a Near-RT RIC according to Embodiment 2. [Figure 16] It is a configuration diagram showing a configuration example of a Near-RT RIC according to Embodiment 3. [Figure 17] It is a configuration diagram showing a configuration example of a Near-RT RIC according to Embodiment 4. [Figure 18] It is a configuration diagram showing a configuration example of a RAN system according to Embodiment 5. [Figure 19] It is a configuration diagram showing an overview of the hardware of a computer according to the embodiment.

Modes for Carrying Out the Invention

[0018] Hereinafter, embodiments will be described with reference to the drawings. In each drawing, the same elements are denoted by the same reference numerals, and redundant explanations are omitted as necessary.

[0019] For example, if an inference model is placed on a Near-RT RIC and a training model is placed on a Non-RT RIC, one possible method is for the Near-RT RIC to collect the necessary data from the E2 node for inference, and for the Non-RT RIC to collect the necessary data from the E2 node for training. The inventors of this disclosure have investigated such collection methods and found that there is a problem that the E2 node and the transmission path that transmits the data may be overloaded. Therefore, the embodiment makes it possible to reduce the load on the E2 node and the transmission path that transmits the data.

[0020] (Summary of the embodiment) First, an overview of the embodiment will be described. Figure 1 shows the general configuration of the first control system 10 according to the embodiment, and Figure 2 shows the general configuration of the second control system 20 according to the embodiment. For example, the first control system 10 and the second control system 20 constitute a system for controlling a wireless network such as a RAN. For example, the first control system 10 includes a Near-RT RIC, and the second control system 20 includes a Non-RT RIC, but is not limited to these.

[0021] As shown in Figure 1, the first control system 10 includes a data collection unit 11 and a data transmission unit 12. The data collection unit 11 collects identification data used by an identification model that identifies controls related to the wireless network. The data collection unit 11 collects identification data from the RAN, including O-DUs and O-CUs. Controls related to the wireless network include, for example, the operation of the RAN, and controls such as the wireless resource allocation scheduler, beam, and handover, which can be made possible by setting the O-DUs and O-CUs. The identification model is an inference model that infers controls related to the wireless network, and the identification data can also be said to be inference data used for inference. For example, the identification model identifies controls for the RAN according to data such as wireless quality collected from the RAN. The identification model is included in the first control system 10, for example, but may be located outside the first control system 10.

[0022] The transmitting unit 12 transmits the identification data collected by the collection unit 11 to the second control system 20, which performs learning using the learning model, as learning data to be used in the learning model for learning control related to the wireless network. For example, the learning model is a model for constructing the identification model. The learning data transmitted by the transmitting unit 12 may include identification result data, which is the result identified by the identification model. Identification result data may include, for example, analysis information obtained by analyzing the identification data or control information for controlling the RAN. The transmitting unit 12 may also transmit learning data depending on the communication status between the first control system 10 and the second control system 20. For example, the transmitting unit 12 may transmit learning data depending on the availability of the interface connecting to the second control system 20.

[0023] As shown in Figure 2, the second control system 20 includes a receiving unit 21 and a learning unit 22. The receiving unit 21 receives learning data transmitted from the transmitting unit 12 of the first control system 10. That is, it receives the identification data collected by the first control system 10 to identify control related to the wireless network using a specific model, as learning data.

[0024] The learning unit 22 uses the learning data received by the receiving unit 21 to learn control related to the wireless network using a learning model. That is, it learns control of the RAN according to data such as wireless quality. The learning model is included in the second control system 20, for example, but may also be located outside the second control system 20. The learning unit 22 applies the learned learning model, which has been trained using the learning data, to a specific model of the first control system 10.

[0025] The first control system 10 and the second control system 20 may each be composed of one device or multiple devices. Figure 3 shows an example configuration of the first control device 30 according to an embodiment, and Figure 4 shows an example configuration of the second control device 40 according to an embodiment. As shown in Figure 3, the first control device 30 may include the collection unit 11 and the transmission unit 12 shown in Figure 1. The collection unit 11 and the transmission unit 12 may be implemented in separate devices, and this is not limited to this example. As shown in Figure 4, the second control device 40 may include the receiving unit 21 and the learning unit 22 shown in Figure 2. The receiving unit 21 and the learning unit 22 may be implemented in separate devices, and this is not limited to this example. Similar to the first control system 10 and the second control system 20, for example, the first control device 30 may be a Near-RT RIC, and the second control device 40 may be a Non-RT RIC.

[0026] Furthermore, some or all of the first control system 10 and the second control system 20 may be deployed at the edge or in the cloud using virtualization technology or the like. They may be deployed in a specific location or distributed across multiple locations. The edge is a location or infrastructure on the base station side, including O-DU and O-CU. The cloud is a location or infrastructure on the core network side, away from the base station. For example, the collection unit 11 and transmission unit 12 may be deployed at the edge, and the receiving unit 21 and learning unit 22 may be deployed in the cloud. Alternatively, the collection unit 11, transmission unit 12, receiving unit 21, and learning unit 22 may be distributed across different locations.

[0027] Figure 5 shows a first control method according to an embodiment, and Figure 6 shows a second control method according to an embodiment. For example, the first control method is performed by the first control system 10 in Figure 1 and the first control device 30 in Figure 3. The second control method is performed by the second control system 20 in Figure 2 and the second control device 40 in Figure 4.

[0028] As shown in Figure 5, the collection unit 11 collects identification data used by a specific model that identifies control related to the wireless network (S11). Next, the transmission unit 12 transmits the collected identification data to a second control system 20, which performs learning by the learning model, as learning data used by the learning model to learn control related to the wireless network (S12).

[0029] Next, as shown in Figure 6, the receiving unit 21 receives training data transmitted from the transmitting unit 12 of the first control system 10, that is, training data collected as specific data to be used by a specific model (S21). Next, the learning unit 22 uses the received training data to learn control related to the wireless network using a learning model (S22). Furthermore, the learning unit 22 applies the learned learning model to the specific model of the first control system 10.

[0030] In this embodiment, a first control system, such as a Near-RT RIC, transmits specific data collected for a specific model to a second control system, such as a Non-RT RIC, as training data to be used for the learning model. The second control system then performs training using the training data received from the first control system. As a result, the first control system can transfer data collected from the wireless network to the second control system, and the second control system can train the learning model using the transferred data. This reduces the load on the nodes providing data in the wireless network and the transmission lines that transmit the data.

[0031] (Embodiment 1) Next, Embodiment 1 will be described. In this embodiment, an example will be described in which the data used by the Near-RT RIC for inference is transferred to the Non-RT RIC as training data.

[0032] Figure 7 shows an example configuration of the RAN system 1 according to this embodiment. As shown in Figure 7, the RAN system 1 includes Near-RT RIC100, Non-RT RIC200, O-DU300, and O-CU400.

[0033] Non-RT RIC200 and Near-RT RIC100, and Non-RT RIC200 and E2 nodes including O-DU300 and O-CU400, are communicated via the O1 interface. The O1 interface is primarily used for sending and receiving data and messages necessary for operation and management. An interface is a connection interface defined by a communication protocol for sending and receiving data and messages, and includes logical transmission paths and networks, as well as physical transmission paths and networks.

[0034] The Non-RT RIC200 and Near-RT RIC100 are connected via the A1 interface. The Near-RT RIC100 and E2 nodes, including O-DU300 and O-CU400, are connected via the E2 interface. The A1 and E2 interfaces are primarily for sending and receiving data and messages necessary for control. The O-DU300 and O-CU400 are connected via the F1 interface.

[0035] O-DU300 and O-CU400 constitute the RAN. nodeIt is also called an E2 node. The RAN is a wireless network accessed by UEs (User Equipment) and is connected to core networks such as 5GC (5G Core network) and EPC (Evolved Packet Core). The RAN may also include O-RUs (O-RAN Remote Units) that constitute the antenna. The UE is a terminal device that connects to the RAN and performs wireless communication, and may be a mobile phone, smartphone, tablet, IoT (Internet of Things) terminal, etc. The UE may also be an application device such as a robot, drone, or autonomous vehicle that implements the functions of a terminal.

[0036] The O-DU300 and O-CU400 provide base station functionality. A base station is, for example, a gNB (next Generation Node B) or an eNB (evolved Node B), but is not limited to these. Note that the O-DU300 and O-CU400 are just examples of nodes providing base station functionality; other network nodes may also be used.

[0037] The O-DU300 is a logical node that provides base station radio signal control and Layer 2 control functions. The O-DU300 accommodates O-RUs and controls the radio signals (beams) of the antennas in the O-RUs it accommodates, as well as performing necessary protocol processing such as MAC (Media Access Control) and RLC (Radio Link Control) between the O-RUs and the O-CU400.

[0038] The O-CU400 is a logical node that provides base station radio resource control functions and data processing functions above Layer 2. The O-CU400 accommodates the O-DU300 and performs data transmission and reception, QoS (Quality of Service) control, cell / UE management, handover control, and protocol processing such as PDCP (Packet Data Convergence Protocol), SDAP (Service Data Adaptation Protocol), and RRC (Radio Resource Control) required between the O-DU300 and the core network.

[0039] RAN system 1 may have one or any number of O-DU300s and O-CU400s as E2 nodes. That is, it may include multiple base stations. The number of O-DU300s and O-CU400s does not necessarily have to be equal. The O-DU300s and O-CU400s may be located in different locations or in the same location. Furthermore, the O-DU300s and O-CU400s may be implemented by different virtual machines running on the edge virtualization infrastructure, or by the same virtual machine. The O-DU300s and O-CU400s may also be vDUs (virtualized Distributed Units) and vCUs (virtualized Central Units), and may constitute a virtual base station. Note that the O-DU300s and O-CU400s may also be physical DUs and CUs. Furthermore, an E2 node may be a base station device that includes the functions of O-DU300s and O-CU400s.

[0040] The Near-RT RIC100 is a logic function that controls and optimizes the RAN in near real-time. The Near-RT RIC100 controls the RAN with short control cycles, for example, between 10ms and less than 1s. The Near-RT RIC100 collects and analyzes RAN data from E2 nodes, including either or both of the O-DU300 and O-CU400, via the E2 interface, and controls the E2 nodes according to the RAN data. For example, the Near-RT RIC100 performs control according to the RAN data according to a control policy obtained from the Non-RT RIC200 via the A1 interface. The control policy is a policy concerning the control of the RAN, for example, the A1 policy. The A1 policy is guidance for RAN optimization defined on the A1 interface. The RAN data is data collected from either or both of the O-DU300 and O-CU400. The RAN data is radio-related data concerning the RAN's radio, including radio quality data and location information for each UE, and may include the number of active UEs per base station (cell). For example, wireless quality data may be obtained from the O-DU300, or handover information may be obtained from the O-CU400. The Near-RT RIC100 is located in the same location as either or both of the O-DU300 and O-CU400, or in a location close to either or both of them. For example, the Near-RT RIC100 may be implemented in a virtual machine at the same edge as either or both of the O-DU300 and O-CU400.

[0041] Several functions of the Near-RT RIC100 are implemented through xApps (Near-RT RIC Applications). xApps include applications that perform tasks such as analyzing RAN data and controlling the RAN. For example, an xApp might include a pre-trained inference model (also called an inference engine), which performs analysis of inference data, including RAN data, and controls the RAN. The Near-RT RIC100 may include multiple xApps as needed.

[0042] Non-RT RIC200 is a logical function that controls and optimizes the RAN in a non-real-time manner. Non-RT RIC200 controls the RAN with long control cycles, for example, 1 second or more. Non-RT RIC200 manages control policies, the operation of E2 nodes including O-DU300 and O-CU400, and Near-RT RIC100, as well as training learning models and updating inference models. For example, Non-RT RIC200 generates control policies and notifies Near-RT RIC100 of the generated policies via the A1 interface. Furthermore, Non-RT RIC200 manages and configures the configuration information of E2 nodes based on data obtained from E2 nodes and Near-RT RIC100 via the O1 interface. Non-RT RIC200 is deployed in the SMO (Service Management and Orchestration) that manages and orchestrates the RAN. The SMO is located in a location separate from the O-DU300, O-CU400, and Near-RT RIC100, for example, on the cloud. The Non-RT RIC200 may also include SMO functionality.

[0043] Some functions of Non-RT RIC200 are implemented by rApps (Non-RT RIC Applications). rApps include applications that perform tasks such as generating control policies and managing inference models for Near-RT RIC100. For example, an rApp might include a training model (also called a training engine), generate a training model that learns RAN control using training data acquired from E2 nodes and Near-RT RIC100 via the O1 interface, and then apply the generated, trained training model to the xApp on Near-RT RIC100. Applying the trained training model to the inference model is also referred to as deployment. Deployment means placing and deploying the model to the application's execution environment, making the model executable. Non-RT RIC200 may include multiple rApps as needed.

[0044] Figure 8 shows the data flow for inference and training in the comparative example, and Figure 9 shows the data flow for inference and training in this embodiment. The arrows shown in Figures 8 and 9 indicate the direction of data collection for explanatory purposes and are not intended to limit the transmission and reception of data associated with such data collection.

[0045] As shown in Figure 8, in the comparative example, when the Non-RT RIC200's learning model (rApp) performs training, the Non-RT RIC200 collects RAN data (D1, C1) from the O-DU300 and O-CU400 via the O1 interface as training data. The Near-RT RIC100 also collects RAN data (D2, C2) from the O-DU300 and O-CU400 via the E2 interface as inference data. The Near-RT RIC100 uses the collected RAN data for its inference model (xApp) to perform inference, and controls the O-DU300 and O-CU400 based on the inference results. Note that the Near-RT RIC100 may collect RAN data from either the DU300 or O-CU400 and control either the DU300 or O-CU400. The Near-RT RIC100 transmits the inference result data (N1) it has inferred as training data to the Non-RT RIC200 via the O1 interface. The Non-RT RIC200's learning model uses the RAN data (D1, C1) collected from the O-DU300 and O-CU400, and the inference result data (N1) collected from the Near-RT RIC100 as training data via the O1 interface. Note that, like the Near-RT RIC100, the Non-RT RIC200 may collect RAN data from either the DU300 or the O-CU400. In this example, the RAN data (D1, C1) and the RAN data (D2, C2) are the same data.

[0046] In the comparative example, either or both of the O-DU300 and O-CU400 transmit RAN data via two interfaces, the O1 interface and the E2 interface. This doubles the communication load on either or both of the O-DU300 and O-CU400 compared to transmitting via a single interface. Consequently, in the comparative example, the communication load solely for the purpose of acquiring and collecting training data occurs on either or both of the O-DU300 and O-CU400, requiring increased communication resources for occasional training sessions.

[0047] Furthermore, in the comparative example, RAN data from either or both of the O-DU300 and O-CU400, along with inference result data from the Near-RT RIC100, are transmitted via the O1 interface, resulting in a heavy load on the O1 interface. In particular, if the transmit buffers of the O-DU300 and O-CU400 are small, the transmission timing cannot be adjusted, which is likely to increase the communication load on the O1 interface. Consequently, in the comparative example, the communication load on the O1 interface is concentrated during the acquisition and collection of training data, requiring the O1 interface's communication resources to be increased for training that is performed occasionally.

[0048] Therefore, in this embodiment, when acquiring and collecting training data used by the learning model installed in Non-RT RIC200 from either or both of the O-DU300 and O-CU400, it is possible to reduce the communication load on either or both of the O-DU300 and O-CU400, and to prevent the concentration of communication load on the O1 interface.

[0049] Specifically, in this embodiment, as shown in Figure 9, when the Non-RT RIC200 learning model (rAPP) performs learning, the Near-RT RIC100 collects RAN data (D2, C2) as inference data via the E2 interface, and the inference model (xApp) performs inference using the collected RAN data. Similar to the comparative example, the Near-RT RIC100 may collect RAN data from either the DU300 or the O-CU400 and control either the DU300 or the O-CU400. In this embodiment, the Near-RT RIC100 stores data necessary for learning from the data used in inference as logs in a database (DB). For example, it stores the collected RAN data (D2, C2) and the inference result data (N1) that has been inferred.

[0050] The Near-RT RIC100 transfers the accumulated RAN data (D2, C2) and inference result data (N1) to the Non-RT RIC200 as training data via the O1 interface. For example, the Near-RT RIC100 transmits to the Non-RT RIC200 when the O1 interface is available. The Non-RT RIC200's learning model uses the RAN data (D2, C2) and inference result data (N1) collected from the Near-RT RIC100 via the O1 interface as training data to perform its training.

[0051] As a result, in this embodiment, compared to the comparative example, either or both of the O-DU300 and O-CU400 transmit RAN data only through the E2 interface, thus reducing the load on either or both of the O-DU300 and O-CU400. Furthermore, since data is transmitted only from the Near-RT RIC100 on the O1 interface, the load on the O1 interface can also be reduced.

[0052] Furthermore, the comparative example in Figure 8 and the embodiment in Figure 9 may be combined. For example, some RAN data may be transferred to the Non-RT RIC200 via the O1 interface from either or both of the O-DU300 and O-CU400, as shown in Figure 8, while other RAN data may be transferred from the Near-RT RIC100 to the Non-RT RIC200 via the O1 interface, as shown in Figure 9.

[0053] Figure 10 shows an example configuration of the Near-RT RIC100 according to this embodiment. As shown in Figure 10, the Near-RT RIC100 includes an E2 communication unit 101, an O1 communication unit 102, an A1 communication unit 103, a data acquisition unit 111, a data analysis unit 112, a control content identification unit 113, a model storage unit 120, a learning data storage control unit 131, a learning database 132, and a learning data transmission unit 133. Note that this configuration is just one example, and other configurations are acceptable as long as they enable the operation described later in this embodiment. Furthermore, configurations necessary to realize the functions required for the Near-RT RIC100 may also be included.

[0054] The E2 communication unit 101 is a communication unit that communicates with the O-DU300 and O-CU400 via the E2 interface. For example, the E2 communication unit 101 sends and receives various data, including RAN data and control messages, to and from the O-DU300 and O-CU400 according to the communication method defined as the E2 interface.

[0055] The O1 communication unit 102 is a communication unit that communicates with the Non-RT RIC200 via the O1 interface. For example, the O1 communication unit 102 sends and receives various data, including training data and control messages, to and from the Non-RT RIC200 according to the communication method defined as the O1 interface.

[0056] The A1 communication unit 103 is a communication unit that communicates with the Non-RT RIC200 via the A1 interface. For example, the A1 communication unit 103 sends and receives various data and control messages, including control policies, to and from the Non-RT RIC200 according to the communication method defined as the A1 interface.

[0057] The data acquisition unit 111 collects RAN data from either or both of the O-DU300 and O-CU400 via the E2 interface through the E2 communication unit 101. For example, the data acquisition unit 111 and the E2 communication unit 101 correspond to the acquisition unit 11 in Figure 1. The data acquisition unit 111 periodically collects RAN data as inference data used by the inference models of the data analysis unit 112 and the control content identification unit 113 for inference. The data acquisition unit 111 may also instruct either or both of the O-DU300 and O-CU400 on the data to be collected and the collection period. The data acquisition unit 111 outputs the collected RAN data to the data analysis unit 112 and the learning data storage control unit 131.

[0058] The model storage unit 120 stores inference models 121 used by the data analysis unit 112 and the control content identification unit 113 for analysis and control processing. The inference model 121 is a trained model that infers the control of E2 nodes, including either or both of the O-DU300 and O-CU400, according to the RAN data. The inference model 121 is, for example, a model capable of analyzing and predicting time-series data. The inference model 121 may be a CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), LSTM (Long-Short Term Model), or any other neural network. The inference model 121 is not limited to a neural network; it may be any other machine learning model. The model storage unit 120 may store multiple inference models 121. For example, the model storage unit 120 stores an inference model 121a for data analysis used by the data analysis unit 112, and an inference model 121b for control content identification used by the control content identification unit 113. Inference model 121a is an analytical model that analyzes data, while inference model 121b is a specific model that identifies the control content.

[0059] The data analysis unit 112 analyzes the RAN data, i.e., inference data, collected by the data collection unit 111. The functions of the data analysis unit 112 may also be realized by executing an xApp for data analysis processing, i.e., wireless analysis processing. The data analysis unit 112 analyzes the RAN data using the data analysis inference model 121a stored in the model storage unit 120. The data analysis unit 112 inputs the collected RAN data into the inference model 121a and performs analysis of the RAN data. The data analysis unit 112 outputs the analysis results, i.e., the inference results, of the RAN data as wireless analysis information. For example, if the RAN data is wireless quality data, the inference model 121a predicts future wireless quality from the input wireless quality data and outputs the prediction result of wireless quality as wireless analysis information. The data analysis unit 112 outputs the wireless analysis information to the control content identification unit 113 and the learning data storage control unit 131.

[0060] The control content identification unit 113 identifies the control content of an E2 node, including either or both of the O-DU300 and O-CU400, based on the analysis results of the RAN data analyzed by the data analysis unit 112, and performs control on the E2 node. The function of the control content identification unit 113 may be realized by executing an xApp for control content identification, i.e., for wireless control processing. Alternatively, the control content identification unit 113 may control the E2 node according to a control policy notified via the A1 interface through the A1 communication unit 103. The control content identification unit 113 identifies the control content (control information) of the E2 node using an inference model 121b for control content identification stored in the model storage unit 120. The control content identification unit 113 inputs the wireless analysis information, which is the analysis result of the data analysis unit 112, into the inference model 121b and identifies the control content corresponding to the wireless analysis information. For example, multiple control contents may be inferred, and the control content to be used for control may be identified according to the control policy. The control content identification unit 113 outputs the identification result, i.e., the inference result, of the identified control content as wireless control information. For example, if the data analysis unit 112 predicts future wireless quality, it identifies the wireless strength and modulation method to be set on the E2 node according to the predicted wireless quality, and outputs the control information to be set on the corresponding E2 node as wireless control information. The control content identification unit 113 transmits the wireless control information indicating the identified control content to either or both of the O-DU300 and O-CU400 via the E2 interface through the E2 communication unit 101. Furthermore, the control content identification unit 113 outputs the wireless control information to the learning data storage control unit 131.

[0061] The learning database 132 is a learning data storage unit that stores learning data. The learning data storage control unit 131 controls the learning data stored in the learning database 132. Specifically, the learning data storage control unit 131 acquires and selects the data necessary for learning, and stores the acquired and selected data in the learning database 132. The learning data storage control unit 131 also performs integration processing to format the stored data for transmission. The learning data storage control unit 131 can also be said to include a learning data acquisition unit, a learning data selection unit, and a learning data integration unit.

[0062] For example, the learning data storage control unit 131 starts controlling the storage of learning data in response to instructions from the Non-RT RIC200. The learning data storage control unit 131 acquires RAN data, which is inference data collected by the data collection unit 111, wireless analysis information generated by the data analysis unit 112, and wireless control information generated by the control content identification unit 113, and selects, i.e., extracts, learning data from the collected and acquired data. The learning data storage control unit 131 may collect and acquire all of the RAN data, wireless analysis information, and wireless control information, or it may collect and acquire any of them, or arbitrarily selected data. The wireless analysis information and wireless control information are also inference result data inferred by the inference model. The learning data to be selected is the data used by the Non-RT RIC200's learning model for learning, and may be instructed by the Non-RT RIC200, or may be predetermined. For example, the learning data to be selected may be determined for each learning model or application. The learning data storage control unit 131 may narrow down the learning data under predetermined conditions. For example, if the RAN data includes UE location information, the training data may be filtered based on the location information. For instance, RAN data contained in a specific area for which wireless quality is to be acquired may be extracted. Alternatively, RAN data contained in other areas may be extracted, excluding areas for which data has already been acquired.

[0063] Furthermore, all collected and acquired RAN data, wireless analysis information, and wireless control information may be stored, and when transmitting to the Non-RT RIC200, training data to be transmitted may be selected from the stored data.

[0064] The learning data storage control unit 131 stores the selected learning data in the learning database 132. For example, the learning data may be stored in a predetermined format, or in a format instructed by the Non-RT RIC200. The learning data storage control unit 131 may also store RAN data collected from multiple O-DU300s and multiple O-CU400s. Furthermore, it may store multiple wireless analysis information and multiple wireless control information analyzed and controlled for multiple O-DU300s and multiple O-CU400s.

[0065] The learning data storage control unit 131 terminates the storage of learning data after it has started, in predetermined cases. For example, it may terminate the storage of learning data at a timing or predetermined timing instructed by the Non-RT RIC 200, at a specified time or period, or when a specified amount or predetermined amount of data has been stored. Once storage is complete, the learning data storage control unit 131 integrates the learning data stored in the learning database 132 into a file for transmission. The integration of learning data includes integrating RAN data from multiple O-DU300s and multiple O-CU400s, and integrating RAN data from multiple O-DU300s and multiple O-CU400s with multiple wireless analysis information and multiple wireless control information. The integrated file may be referred to as a learning data file or integrated data.

[0066] The learning data storage control unit 131 may process or format the training data as needed when integrating the training data. For example, the learning data storage control unit 131 may compress the training data to reduce its size. Compression may be performed on the data before integration or on the data after integration. For example, the learning data storage control unit 131 may reduce the data by dimensionality reduction. In dimensionality reduction, multidimensional data is reduced to information of fewer dimensions, to the extent that it does not affect the learning of the learning model. For example, the number of features in the training data input to the learning model during learning may be reduced to the extent that it does not reduce the accuracy of data analysis (wireless analysis) or control content identification (wireless control). In addition, deduplication may be performed to remove duplicate data, such as data from the same area, or the data may be compressed using a predetermined encoding. Not limited to compression, the integrated data may be divided into multiple data files. For example, multiple training data files may be transmitted.

[0067] The learning data transmission unit 133 transmits the learning data, i.e., the learning data file, integrated by the learning data storage control unit 131, to the Non-RT RIC200 via the O1 interface through the O1 communication unit 102. For example, the learning data transmission unit 133 and the O1 communication unit 102 correspond to the transmission unit 12 in Figure 1. The learning data transmission unit 133 may control the transmission timing and transfer rate of the learning data file. The learning data transmission unit 133 may control the transmission timing and transfer rate in response to instructions from the Non-RT RIC200, or in response to the communication status of the O1 interface. For example, if there is available transmission bandwidth in the O1 interface, the learning data file may be transmitted, and if there is no available transmission bandwidth, the transmission timing of the learning data file may be delayed. The learning data file may be transmitted when the available transmission bandwidth is greater than a predetermined threshold, or when the available transmission bandwidth is greater than the bandwidth required for the file size to be transmitted. The transfer rate may also be controlled according to the available transmission bandwidth. The available transmission bandwidth may be the bandwidth obtained by the O1 communication unit 102 monitoring the communication volume of the O1 interface, or it may be the bandwidth predicted from the monitoring results.

[0068] Furthermore, when the training data transmission unit 133 transmits multiple training data, i.e., multiple training data files, it may control the transmission order of the training data files. For example, the training data contained in the training data files may be transmitted in an order according to their priority. The priority may be set according to the order of the data used by the training model for training, or whether the data is essential or optional for training the training model, or it may be specified by Non-RT RIC200. The transmission order of the training data files may be determined based on the highest priority among the training data contained in the training data files. Alternatively, the training data may be integrated according to its priority, and the transmission order of the training data files may be determined based on the priority of the integrated training data file. Also, when integrating and transmitting training data for each training model, the training data files may be transmitted in an order according to the priority of the training model (application). For example, the transmission order of the training data files may be determined based on the importance of the application or the urgency of training the training model. Furthermore, if there is traffic transmitted via the O1 interface other than training data by other applications, the transmission timing and transmission order may be adjusted between the traffic. For example, transmission scheduling may be performed considering desired delivery times for traffic used for multiple purposes.

[0069] Figure 11 shows an example configuration of the Non-RT RIC200 according to this embodiment. As shown in Figure 11, the Non-RT RIC200 includes an O1 communication unit 201, an A1 communication unit 202, a system management unit 211, a learning unit 212, and a model storage unit 220. Note that this configuration is just one example, and other configurations are also acceptable as long as they enable the operation described later in this embodiment. Furthermore, configurations necessary to realize the functions required for the Non-RT RIC200 may also be included.

[0070] The O1 communication unit 201 is a communication unit that communicates with the Near-RT RIC100 via the O1 interface. For example, the O1 communication unit 201 sends and receives various data, including training data, and control messages to and from the Near-RT100 according to the communication method defined as the O1 interface. It can also send and receive necessary data and control messages to and from the O-DU300 and O-CU400 via the O1 interface.

[0071] The A1 communication unit 202 is a communication unit that communicates with the Near-RT RIC 100 via the A1 interface. For example, the A1 communication unit 202 sends and receives various data and control messages, including control policies, to and from the Near-RT RIC 100 according to the communication method defined as the A1 interface.

[0072] The system management unit 211 manages the configuration and operation of the RAN system, including the O-DU300, O-CU400, and Near-RT RIC100. The functions of the system management unit 211 may be implemented by executing an rApp for system management processing. For example, the system management unit 211 is a policy generation unit that generates control policies. The system management unit 211 may generate control policies based on instructions input from operators or external devices, or it may generate control policies based on data obtained from one or all of the O-DU300, O-CU400, and Near-RT RIC100. The system management unit 211 notifies the Near-RT RIC100 of the generated control policies via the A1 interface through the A1 communication unit 202.

[0073] The model memory unit 220 stores the learning model 221 for constructing the inference model 121 of the Near-RT RIC100. The learning model 221 is the same model as the inference model 121, and is, for example, a model that learns to analyze and predict time-series data. The model memory unit 220, like the Near-RT RIC100, can store multiple learning models 221, for example, a learning model 221a for data analysis and a learning model 221b for identifying control content.

[0074] The learning unit 212 performs machine learning using training data, i.e., training data files, received from the Near-RT RIC100 via the O1 interface through the O1 communication unit 201. For example, the learning unit 212 corresponds to the learning unit 22 in Figure 2. The functions of the learning unit 212 may also be realized by executing an rApp for training processing. The learning unit 212 can also be said to include a receiver that receives training data files from the Near-RT RIC100 via the O1 interface. For example, the O1 communication unit 201 and the learning unit 212 also correspond to the receiver 21 in Figure 2. If the received training data files are compressed or split, the learning unit 212 performs decompression or merging processing as needed. The learning unit 212 performs machine learning such as deep learning to generate a trained learning model 221. The learning unit 212 inputs the training data files into the learning model 221a for data analysis and the learning model 221b for control content identification in the model storage unit 220, and trains each model. Received training data fileData processing necessary for inputting the data into learning models 221a and 221b may be performed. Furthermore, similar to the Near-RT RIC100, it may be equipped with a learning unit for data analysis and a learning unit for identifying control content. The learning data in the received learning data file includes RAN data from either or both of the O-DU300 and O-CU400, and inference result data from the Near-RT RIC100 (wireless analysis information and wireless control information). By using this data, analysis and control according to the RAN data can be learned. That is, by using the pre-control RAN data input to the inference model, the inference result data output from the inference model, and the RAN data after control based on the inference result as learning data, accurate training can be achieved. Note that the learning data may be any of the RAN data, wireless analysis information, and wireless control information. The learning unit 212 stores the trained learning models 221a and 221b in the model storage unit 220, and further transmits the trained learning models 221a and 221b to the Near-RT RIC 100 and applies them to the inference models 121a and 121b.

[0075] Figure 12 shows an overview of the learning process in the RAN system 1 according to this embodiment. As shown in Figure 12, first, a learning model is prepared and placed in the RAN system 1 (S101). For example, a trained model is generated by simulation in an external learning device, and the generated model is stored as an inference model (xApp) in the model storage unit 120 of the Near-RT RIC100, and the same model is stored as a learning model (rApp) in the model storage unit 220 of the Non-RT RT200. Not limited to simulation, a model equipped with a predetermined algorithm may be placed in the Near-RT RIC100 as an inference model, and the same model may be placed in the Non-RT RIC200 as a learning model.

[0076] Next, RAN system 1 performs initial training upon system installation (S102). During system installation, the inference model (xApp) of Near-RT RIC100 is run in a real environment, either in production or trial, to collect training data. Furthermore, the training data is transferred from Near-RT RIC100 to Non-RT RIC200, and the trained model (rApp) is trained using the transferred training data. For example, one hour's worth of training data may be collected and used for training. The trained model acquired through the initial training is then applied to Near-RT RIC100 as the inference model. Note that the initial training has little impact on the system load, so as in the comparative example, Non-RT RIC200 may collect RAN data from either or both O-DU300 and O-CU400 via the O1 interface.

[0077] Next, the RAN system 1 determines whether relearning is necessary (S103), and if so, performs the second and subsequent relearning (S104). For example, Non-RT RIC200 determines that relearning is necessary and performs relearning when instructions are input from an operator or external device, when the on-site environment including the UE changes, at regular intervals, or when the accuracy of the inference model deteriorates. Environmental changes may be detected from changes in radio quality, or signals indicating environmental changes such as layout changes may be input. The accuracy of the inference model may be determined from the inference results of the inference model and RAN data, etc. During relearning, Near-RT RIC100 uses the previously trained and applied inference model (xApp) and collects training data during actual operation while running the inference model in the real environment. Furthermore, the training data is transferred from Near-RT RIC100 to Non-RT RIC200, and the training model (rApp) is trained with the transferred training data. For example, one hour's worth of training data may be collected and used for training. The trained model, acquired through relearning, is applied to the Near-RT RIC100 as an inference model. Subsequently, relearning is repeated using steps S103 and S104.

[0078] Figure 13 shows an example of the operation of the learning process in the RAN system 1 according to this embodiment. For example, Figure 13 is an example of the learning process in the retraining (S104) shown in Figure 12, but it may also be applied to the learning process in the initial learning (S102).

[0079] As shown in Figure 13, the Near-RT RIC100 starts accumulating training data (S201). For example, the learning unit 212 instructs the Near-RT RIC100 to start learning. The learning data accumulation control unit 131 starts accumulating training data in the learning database 132 in response to the instruction from the Non-RT RIC200.

[0080] Next, the Near-RT RCI100 determines whether the accumulation of training data has finished (S202), and if the accumulation has finished, it integrates the training data (S203). For example, when the period instructed by the Non-RT RIC200 ends, the training data storage control unit 131 integrates the training data accumulated in the training database 132 into a training data file.

[0081] Next, the Near-RT RIC100 transfers the training data file to the Non-RT RIC200 (S204). For example, the training data transmission unit 133 transfers the integrated training data file to the Non-RT RIC200 at the timing instructed by the Non-RT RIC200 or when the O1 interface is available.

[0082] Next, the Non-RT RIC200 performs training on the training model using the received training data file (S205). When the training unit 212 receives the training data file from the Near-RT RIC100, it trains the training models 221a and 221b using the training data contained in the received training data file. When training is complete, the training unit 212 stores the trained training models 221a and 221b in the model storage unit 220.

[0083] Next, Non-RT RIC200 applies the trained model to Near-RT RIC100 (S206). The learning unit 212 transmits the trained models 221a and 221b stored in the model storage unit 220 to Near-RT RIC100. The data analysis unit 112 and the control content identification unit 113 update the inference models 121a and 121b in the model storage unit 120 using the received trained models 221a and 221b. Near-RT RIC100 performs data analysis and control using the updated inference models 121a and 121b.

[0084] Figure 14 shows a sequence diagram of the learning process in the RAN system 1 according to this embodiment. For example, Figure 14 shows the operation from the storage to the transfer of the learning data shown in Figure 13 (S201 to S204).

[0085] As shown in Figure 14, the Near-RT RIC100 collects RAN data from either or both of the O-DU300 and O-CU400 (S301). The data acquisition unit 111 repeatedly collects RAN data from either or both of the O-DU300 and O-CU400 via the E2 interface. For example, the data acquisition unit 111 may instruct either or both of the O-DU300 and O-CU400 on the data to be collected and the collection period, and either or both of the O-DU300 and O-CU400 may transmit data according to the instruction, or either or both of the O-DU300 and O-CU400 may transmit data according to predetermined data and period. The data analysis unit 112 analyzes the collected RAN data to generate wireless analysis information, and the control content identification unit 113 infers wireless control information according to the wireless analysis information. The control content identification unit 113 transmits wireless control information to either or both of the O-DU300 and O-CU400 via the E2 interface, and controls the operation of either or both of the O-DU300 and O-CU400. Processing from S302 to S307 is performed in parallel with the repeated collection and control of RAN data by S301. That is, the Near-RT RIC100 collects inference data and performs control by the inference model, while accumulating and transferring training data.

[0086] First, the Non-RT RIC200 sends a learning instruction message to the Near-RT RIC100 (S302). The learning unit 212 sends a learning instruction message to the Near-RT RIC100 via the O1 interface in response to instructions from the operator or changes in the environment. The learning instruction message is a message that instructs the start of collecting (storing) training data for training the learning model. For example, the learning instruction message includes the target learning model and the training data collection period. The target learning model indicates identification information that identifies the learning model (learning engine) to be learned, and the training data collection period indicates the time to start and end the collection of training data. The learning instruction message may include information that identifies the training data to be collected, not just the target learning model. The training data collection period may specify the length of time for collecting training data. The learning instruction message may include a period for collecting training data, not just the training data collection period. The target learning model and the training data collection period may be set in advance or set by the operator. For example, if a change in the environment is detected, the target learning model and the learning data collection period may be set according to the change in the environment. If the wireless quality changes, the learning model of the application corresponding to the location of the changed wireless quality may be determined as the target learning model. The learning instruction message may be transmitted via the O1 interface or via the A1 interface. In the Near-RT RIC100, the learning data transmission unit 133 receives the learning instruction message via the O1 or A1 interface, and the learning data storage control unit 131 acquires the received learning instruction message.

[0087] Next, when the Near-RT RIC100 starts accumulating training data as specified in the received training instruction message, it begins accumulating training data (S303). The training data accumulation control unit 131 selects the training data necessary for training the target training model specified in the training instruction message. For example, it may pre-associate training models with training data and identify the training data corresponding to the target training model specified in the training instruction message. If the training data to be collected is specified in the training instruction message, it identifies the specified training data as data to be accumulated. The training data accumulation control unit 131 selects (extracts) the identified training data from one or all of the RAN data collected by the data collection unit 111, the wireless analysis information generated by the data analysis unit 112, and the wireless control information generated by the control content identification unit 113, and accumulates the selected training data in the training database 132. The data acquisition unit 111 may accumulate necessary data each time it collects RAN data, the data analysis unit 112 may generate wireless analysis information, and the control content identification unit 113 may generate wireless control information. If a learning data collection cycle is specified in a learning instruction message, data may be accumulated according to the specified cycle. The Near-RT RIC 100 may also spontaneously start accumulating learning data without receiving a learning instruction message from the Non-RT RIC 200. In this case, predetermined learning data may be accumulated over a predetermined period.

[0088] Next, when the Near-RT RIC100 completes the training data collection period specified in the training instruction message, it integrates the training data (S304). The training data storage control unit 131 formats the training data stored in the training database 132 and integrates it into a single training data file. For example, the training model and the format of the training data file may be pre-associated and set, and the training data may be formatted and integrated into the training data file to match the format corresponding to the target training model specified in the training instruction message. Alternatively, the format may be specified in the training instruction message, and the training data may be integrated to match the specified format. Furthermore, the integrated training data file may be compressed as needed. For example, the training instruction message may instruct a compression method such as dimensionality reduction, and the data may be compressed using the instructed compression method, or it may be compressed using a compression method corresponding to the target training model specified in the training instruction message.

[0089] Next, Near-RT RIC100 sends a data collection completion message to Non-RT RIC200 (S305). When the training data is integrated, the training data transmission unit 133 sends a data collection completion message to Non-RT RIC200 via the O1 interface. The data collection completion message is a message that notifies that the training data is complete, i.e., that the collection (storage) of training data has been completed. The data collection completion message may include the target training model, similar to the training instruction message. The data collection completion message may be sent via the O1 interface or via the A1 interface. In Non-RT RIC200, the training unit 212 receives the data collection completion message via the O1 or A1 interface.

[0090] Next, Non-RT RIC200 sends a transfer instruction message to Near-RT RIC100 (S306). Upon receiving the data collection completion message, the learning unit 212 sends a transfer instruction message to Near-RT RIC100 via the O1 interface. The transfer instruction message is a message instructing the transfer of the collected training data. For example, the transfer instruction message includes timing and bitrate. Timing indicates the timing of transmission of the training data, and may be the transmission time, etc. Bitrate indicates the transmission bitrate (bps) of the training data. The transfer instruction message may also include the target training model, similar to the learning instruction message. The transfer instruction message may be sent via the O1 interface or via the A1 interface. In Near-RT RIC100, the training data transmission unit 133 receives the transfer instruction message via the O1 or A1 interface.

[0091] Next, Near-RT RIC100 sends the training data file to Non-RT RIC200 (S307). The training data transmission unit 133 sends the integrated training data file to Non-RT RIC200 via the O1 interface at the bitrate specified in the transfer instruction message when the timing specified in the received transfer instruction message arrives. For example, the training data file may be sent after the specified timing if the transmission bandwidth of the O1 interface is free. The training data file may be sent via the O1 interface or via the A1 interface. For example, if the transmission bandwidth of the O1 interface is not free, it may be sent via the A1 interface. Near-RT RIC100 may also spontaneously send the training data file without receiving a transfer instruction from Non-RT RIC200. The training data file may also be sent when the accumulation of training data is completed and the training data file is generated, or at a predetermined timing.

[0092] As described above, in this embodiment, when the Near-RT RIC receives a learning instruction, for example, it logs the data necessary for learning from the data used by the inference model, including wireless analysis and wireless control, in parallel with the inference by the inference model. Furthermore, after log collection, the Near-RT RIC transmits the data stored in the Non-RT RIC via the O1 interface, for example, by making effective use of the timing when the O1 interface is idle, and the Non-RT RIC performs learning of the learning model. As a result, since there is no communication load solely for the purpose of acquiring and collecting learning data, the communication load on either or both of the O-DU and O-CU can be reduced, and the required communication resources can be reduced. In addition, since the Non-RT RIC can collect learning data by making effective use of the timing when the O1 interface is idle, for example, the communication load on the O1 interface can be distributed, and the communication resources required for the O1 interface can be reduced.

[0093] (Embodiment 2) Next, Embodiment 2 will be described. In this embodiment, an example of learning using Near-RT RIC will be described. Note that this embodiment can be implemented in combination with Embodiment 1, and the configuration of Embodiment 1 may be used as appropriate.

[0094] Figure 15 shows an example configuration of the Near-RT RIC100 according to this embodiment. For example, as shown in Figure 15, the Near-RT RIC100 according to this embodiment includes a learning unit 141 in addition to the configuration of Embodiment 1. The other configurations are the same as in Embodiment 1.

[0095] The learning unit 141 performs training using inference data and applies the trained learning model to the inference model. The functions of the learning unit 141 may also be realized by executing an xApp for training processing. Similar to the learning unit 212 of the Non-RT RIC200, the learning unit 141 trains the learning model using one or all of the following: RAN data collected by the data acquisition unit 111, wireless analysis information analyzed by the data analysis unit 112, and wireless control information identified by the control content identification unit 113. The learning unit 141 may acquire the necessary data from the data acquisition unit 111, the data analysis unit 112, and the control content identification unit 113, or it may acquire the necessary data from the learning database 132. The learning unit 141 performs training on some of the inference models 121 stored in the model storage unit 120. Training may be performed on either inference model 121a or inference model 121b. For example, in cases where learning can be limited to specific cells, such as in wireless analysis including time series analysis of SINR (Signal-to-Interference plus Noise power Ratio), learning can be performed with relatively light processing, and therefore, learning is carried out by the learning unit 141 of the Near-RT RIC100. For example, the learning unit 141 generates a learning model that has learned wireless analysis using the SINR of a specific cell, and updates the inference model 121a in the model storage unit 120 with the generated, trained learning model.

[0096] Thus, training can be performed on the Near-RT RIC, and the trained model can be applied to the inference model within the Near-RT RIC. This allows for faster training because inference and training can be performed within the Near-RT RIC, and also reduces the load on the O1 interface and the Non-RT RIC.

[0097] (Embodiment 3) Next, Embodiment 3 will be described. In this embodiment, an example will be described in which the data collection frequency and the time granularity of the data to be transferred are adjusted in accordance with changes in the data to be collected. Note that this embodiment can be implemented in combination with either Embodiment 1 or 2, and either the configuration of Embodiment 1 or 2 may be used as appropriate.

[0098] Figure 16 shows an example configuration of the Near-RT RIC100 according to this embodiment. For example, as shown in Figure 16, the Near-RT RIC100 according to this embodiment includes a data change detection unit 142 in addition to the configuration of Embodiment 1. The other configurations are the same as in Embodiment 1.

[0099] The data change detection unit 142 detects changes in RAN data acquired from either or both of the O-DU300 and O-CU400. For example, it detects the magnitude of changes in wireless quality data, or the movement speed of the UE from the UE's location information. Note that the functions of the data change detection unit 142 may be included in the data analysis unit 112 or other xApps. For example, the data analysis unit 112 may output wireless analysis information including the results of the data change detection.

[0100] In this embodiment, the data acquisition unit 111 adjusts the data acquisition frequency according to the detection result of the data change detection unit 142. For example, the data acquisition unit 111 adjusts the acquisition frequency from either or both of the O-DU300 and O-CU400 according to the severity of the time-series changes in the wireless quality data and the movement speed of the UE. As the changes in the wireless quality data become larger, the data acquisition frequency may be increased, i.e., the data acquisition interval may be shortened, and as the changes in the wireless quality data become smaller, the data acquisition frequency may be decreased, i.e., the data acquisition interval may be lengthened. Also, as the movement speed of the UE becomes faster, the data acquisition frequency may be increased, i.e., the data acquisition interval may be shortened, and as the movement speed of the UE becomes slower, the data acquisition frequency may be decreased, i.e., the data acquisition interval may be lengthened.

[0101] Furthermore, the learning data storage control unit 131 may adjust the time granularity of the data stored in the learning database 132, that is, the time granularity of the data transferred as learning data, according to the detection results of the data change detection unit 142. For example, the learning data storage control unit 131 adjusts the time granularity of the data stored and transferred according to the intensity of the time series changes in the wireless quality data and the movement speed of the UE. For example, the time granularity of the data can be made coarser by reducing the number of data points per unit time. As the changes in the wireless quality data become smaller, the time granularity of the data may be made coarser, that is, the interval of the time series data may be made longer. As the movement speed of the UE becomes slower, the time granularity of the data may be made coarser, that is, the interval of the time series data may be made longer.

[0102] Thus, when collecting data with the Near-RT RIC, the data collection frequency and the time granularity of the data to be transferred may be adjusted according to changes in the data being collected. By adjusting the data collection frequency, the load on either or both of the O-DU and O-CU, as well as the load on the E2 interface, can be reduced. Furthermore, by adjusting the time granularity of the data to be transferred, the load on the O1 interface can be further reduced.

[0103] (Embodiment 4) Next, Embodiment 4 will be described. This embodiment describes an example of selecting data to be stored and transferred in response to changes in the environment. This embodiment can be implemented in combination with any of Embodiments 1 to 3, and any of the configurations of Embodiments 1 to 3 may be used as appropriate.

[0104] Figure 17 shows an example configuration of the Near-RT RIC100 according to this embodiment. For example, as shown in Figure 17, the Near-RT RIC100 according to this embodiment includes an environmental change detection unit 143 in addition to the configuration of Embodiment 1. The other configurations are the same as in Embodiment 1.

[0105] The environmental change detection unit 143 detects environmental changes based on RAN data acquired from either or both of the O-DU300 and O-CU400. Environmental changes refer to changes in the UE's wireless environment. For example, these could be changes in the wireless environment that necessitate training a learning model, or changes in the wireless environment caused by layout changes. The environmental change detection unit 143 detects changes in wireless quality and determines that the wireless environment has changed if the wireless quality has changed significantly beyond a predetermined value. It may also identify locations where the wireless quality has changed significantly. The functions of the environmental change detection unit 143 may be included in the data analysis unit 112 or other xApps. For example, the data analysis unit 112 may output wireless analysis information including the results of environmental change detection.

[0106] In this embodiment, the learning data storage control unit 131 selects the data necessary for learning, i.e., the data to be stored, in accordance with the environmental change detection unit 143. For example, locations where wireless quality has changed significantly are likely to be the cause of changes in the wireless environment due to changes in the placement of shielding objects or layout changes, which necessitates training of the learning model. Therefore, the learning data is narrowed down to data from locations (areas) where wireless quality has changed significantly. The learning data storage control unit 131 narrows down the data from the locations identified from the collected RAN data and stores the narrowed-down learning data in the learning database 132, thereby reducing unnecessary data.

[0107] Thus, when collecting and storing data with Near-RT RIC, the data to be stored may be selected according to changes in the environment. This reduces the amount of training data, and therefore further reduces the load on the O1 interface.

[0108] (Embodiment 5) Next, Embodiment 5 will be described. This embodiment describes an example of acquiring data from an external application server. This embodiment can be implemented in combination with any of Embodiments 1 to 4, and any of the configurations of Embodiments 1 to 4 may be used as appropriate.

[0109] Figure 18 shows an example configuration of the RAN system 1 according to this embodiment. As shown in Figure 18, the RAN system 1 according to this embodiment includes an external application server 500 in addition to the configuration of Embodiment 1. Other configurations are the same as, for example, those of Embodiment 1. The Near-RT RIC 100 and the external application server 500 are connected via any interface to enable communication. They may be connected via an interface for a general application server to provide data. The Non-RT RIC 200 and the external application server 500 may also be connected to enable communication.

[0110] The external application server 500 is a server outside the RAN, including at least the O-DU300 and O-CU400. It can also be said that the external application server 500 is a server outside the system, including the O-DU300, O-CU400, Non-RT RIC200, and Near-RT RIC100. Furthermore, the external application server 500 is a data provider that provides application data. For example, the external application server 500 could be a web server or a social networking service (SNS) server, a management server for an automated guided vehicle (AGV) system or an autonomous mobile robot (AMR) system, a management server for a platooning system or an autonomous driving system, a management server for an automated construction system, or a server that executes an application that provides a predetermined service in cooperation with the RAN. The external application server 500 only needs to be able to provide application data to the Near-RT RIC100, and could, for example, be a server on the internet. The external application server 500 may be a physical server or a virtual server on the cloud.

[0111] Application data is data related to applications generated by the external application server 500 and is external data that cannot be collected within the RAN. Application data may include information about the status of the application or information about the communication requirements of the application. Information about the status of the application may include, for example, the location, speed, and trajectory of each UE. Information about the communication requirements of the application may include, for example, the communication quality requirements for each UE. In addition, application data may include various other data that can be used for inference in the Near-RT RIC100 inference model. For example, it may include weather information, traffic information, map information, etc.

[0112] In this embodiment, the Near-RT RIC100 collects application data from an external application server 500 and uses it for inference. Specifically, the Near-RT RIC100 uses RAN data collected from either or both of the O-DU300 and O-CU400, along with application data collected from the external application server 500, as inference data for the inference model to perform inference and control either or both of the O-DU300 and O-CU400.

[0113] Furthermore, Near-RT RIC100 stores the application data used for inference and transfers it to Non-RT RIC200 as training data. Specifically, it stores data selected from the inference data (RAN data and application data) and the inference result data (wireless analysis information and wireless control information) in the training database 132 as training data, and transfers the stored training data to Non-RT RIC200. Near-RT RIC100 acquires the transferred training data and, for example, uses the RAN data, application data, wireless analysis information, and wireless control information contained in the training data to train the learning model.

[0114] Thus, the Near-RT RIC may acquire application data from an external application server and use it for inference along with RAN data from the E2 node. This enables wireless network control suitable for various situations and applications. Furthermore, the application data used for inference by the Near-RT RIC is transferred to the Non-RT RIC as training data along with RAN data, etc. This allows for the creation of training data with consistent RAN data and application data, which can then be transmitted to the Non-RT RIC200. Additionally, since there is no need to collect application data redundantly as both inference data and training data, the communication load on the external application server and the communication lines between the external application server and the system including O-DU300, O-CU400, Non-RT RIC200, and Near-RT RIC100 can be reduced.

[0115] This disclosure is not limited to the embodiments described above, and may be modified as appropriate without departing from its spirit.

[0116] Each configuration in the above-described embodiment may consist of hardware, software, or both, and may consist of one piece of hardware or software, or multiple pieces of hardware or software. Each device and each function (process), including Non-RT RIC and Near-RT RIC, may be realized by a computer 50 having a network interface 51, a processor 52 such as a CPU (Central Processing Unit), and a memory 53 as a storage device, as shown in Figure 19. The network interface 51 may include a network interface card (NIC) for communicating with devices including network nodes. For example, a program for performing the method (control method) in the embodiment may be stored in the memory 53, and each function may be realized by executing the program stored in the memory 53 with the processor 52.

[0117] These programs, when loaded into a computer, include a set of instructions (or software code) for causing the computer to perform one or more of the functions described in the embodiments. The programs may be stored on non-temporary computer-readable media or tangible storage media. Examples, but not limited to, include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drives (SSDs), or other memory technologies, CD-ROMs, digital versatile discs (DVDs), Blu-ray® discs, or other optical disc storage, magnetic cassettes, magnetic tapes, magnetic disk storage, or other magnetic storage devices. The programs may be transmitted over temporary computer-readable media or communication media. Examples, but not limited to, include electrical, optical, acoustic, or other forms of propagating signals.

[0118] Although the present disclosure has been described above with reference to embodiments, the present disclosure is not limited to the embodiments described above. Various modifications to the structure and details of the present disclosure can be understood by those skilled in the art within the scope of the present disclosure.

[0119] Some or all of the above embodiments may also be described as follows, but are not limited to the following: (Note 1) A specific model for identifying control related to a wireless network includes a collection means for collecting specific data used for the identification, A transmission means for transmitting the aforementioned identification data to another control system that performs learning using the learning model, as learning data for use in a learning model that learns control related to the wireless network. A control system equipped with the following features. (Note 2) The aforementioned specific model is constructed using the learned model that has undergone the aforementioned training. The control system described in Appendix 1. (Note 3) The training data includes specific result data which is the result of the specific model using the specific data. The control system described in Appendix 1 or 2. (Note 4) The system includes a selection means for selecting training data to be used by the learning model for training from the aforementioned collected identification data, The transmission means transmits the selected training data. A control system as described in any one of the appendices 1 to 3. (Note 5) The selection means selects the training data based on the location of the user terminal corresponding to the identification data. The control system described in Appendix 4. (Note 6) The selection means selects the training data based on changes in the environment corresponding to the identification data. The control system described in Appendix 4 or 5. (Note 7) The system includes storage means for storing the learning data to be transmitted, The transmission means transmits integrated data which is an integrated version of the accumulated training data. A control system as described in any one of the appendices 1 to 6. (Note 8) The storage means stores learning data, including identification data collected from multiple nodes included in the wireless network. The control system described in Appendix 7. (Note 9) The transmitting means transmits the learning data according to the communication status between the control system and the other control system. A control system as described in any one of the appendices 1 to 8. (Note 10) The transmission means transmits the learning data in response to other traffic transmitted from the control system to the other control system. A control system as described in any one of the appendices 1 to 9. (Note 11) When transmitting multiple training data, the transmitting means transmits the training data according to the priority of each training data. A control system as described in any one of the appendices 1 to 10. (Note 12) The transmission means transmits compressed data obtained by reducing the dimensionality of the training data. A control system as described in any one of the appendices 1 to 11. (Note 13) The collection means adjusts the frequency of collecting the specific data in accordance with changes in the collected specific data. A control system as described in any one of the appendices 1 to 12. (Note 14) The transmission means adjusts the time granularity of the training data to be transmitted in accordance with the changes in the collected identification data. A control system as described in any one of the appendices 1 to 13. (Note 15) The system includes a learning means that generates a learning model trained using the aforementioned specific data and updates the specific model using the generated learning model. A control system as described in any one of the appendices 1 to 14. (Note 16) The aforementioned control system and the other control system include a RIC (RAN Intelligent Controller) that controls the RAN (Radio Access Network). A control system as described in any one of the appendices 1 to 15. (Note 17) The control system includes a Near-RT (Real Time) RIC. The aforementioned other control system includes Non-RT RIC, The control system described in Appendix 16. (Note 18) A receiving means that receives identification data collected by another control system to identify control related to a wireless network using a specific model, as training data from the other control system, A learning means for learning control of the wireless network using the aforementioned training data and a learning model, A control system equipped with the following features. (Note 19) The learning means applies the learned model to the specific model. The control system described in Appendix 18. (Note 20) A specific model for identifying control related to a wireless network includes a collection means for collecting specific data used for the identification, A transmission means for transmitting the aforementioned identification data to another control system that performs learning using the learning model, as learning data for use in a learning model that learns control related to the wireless network. A control device equipped with the following features. (Note 21) A receiving means that receives identification data collected by another control system to identify control related to a wireless network using a specific model, as training data from the other control system, A learning means for learning control of the wireless network using the aforementioned training data and a learning model, A control device equipped with the following features. (Note 22) A specific model that identifies controls related to a wireless network collects specific data used for the identification, The aforementioned identification data is transmitted to another control system that performs learning using the learning model, as learning data to be used by the learning model for learning control related to the wireless network. Control method. (Note 23) The identification data collected by another control system to identify control related to the wireless network using a specific model is received from the other control system as training data. Using the aforementioned training data, the learning model learns control over the wireless network. Control method. (Note 24) A specific model that identifies controls related to a wireless network collects specific data used for the identification, The aforementioned identification data is transmitted to another control system that performs learning using the learning model, as learning data to be used by the learning model for learning control related to the wireless network. A non-temporary, computer-readable medium containing control programs for executing processes on a computer. (Note 25) The identification data collected by another control system to identify control related to the wireless network using a specific model is received from the other control system as training data. Using the aforementioned training data, the learning model learns control over the wireless network. A non-temporary, computer-readable medium containing control programs for executing processes on a computer. [Explanation of symbols]

[0120] 1 RAN System 10 First control system 11 Collection Department 12 Transmitter 20 Second control system 21 Receiving unit 22 Learning Department 30 First control device 40 Second control device 50 Computers 51 Network Interfaces 52 processors 53 memory 100 Near-RT RIC 101 E2 Communications Department 102 O1 Communications Department 103 A1 Communications Department 111 Data Collection Unit 112 Data Analysis Department 113 Control Content Identification Unit 120 Model Memory Unit 121, 121a, 121b Inference Models 131 Learning Data Storage Control Unit 132 Learning Databases 133 Learning Data Transmission Unit 141 Learning Department 142 Data change detection unit 143 Environmental change detection unit 200 Non-RT RIC 201 O1 Communications Department 202 A1 Communications Department 211 System Administration Department 212 Learning Department 220 Model Memory Unit 221, 221a, 221b Learning Models 300 O-DU 400 O-CU 500 External Application Servers

Claims

1. A specific model for identifying control related to a wireless network includes a collection means for collecting specific data used for the identification, A transmission means for transmitting the aforementioned identification data to another control system that performs learning using the learning model, as learning data for use in a learning model that learns control related to the wireless network. A control system equipped with the following features.

2. The aforementioned specific model is constructed using the learned model that has undergone the aforementioned training. The control system according to claim 1.

3. The training data includes specific result data which is the result of the specific model using the specific data. The control system according to claim 1 or 2.

4. The system includes a selection means for selecting training data to be used by the learning model for training from the aforementioned collected identification data, The transmission means transmits the selected training data. The control system according to claim 1 or 2.

5. The selection means selects the training data based on the location of the user terminal corresponding to the identification data. The control system according to claim 4.

6. A receiving means that receives identification data collected by another control system to identify control related to a wireless network using a specific model, as training data from the other control system, A learning means for learning control of the wireless network using the aforementioned training data and a learning model, A control system equipped with the following features.

7. A control method performed by a control system, A specific model that identifies controls related to a wireless network collects specific data used for the identification, The aforementioned identification data is transmitted as training data to a learning model that learns control of the wireless network, to another control system that performs training using the learning model, and which is different from the aforementioned control system. Control method.

8. A control method performed by a control system, The other control system, which is different from the aforementioned control system, receives identification data collected by another control system to identify control related to the wireless network using a specific model, as training data from the other control system. Using the aforementioned training data, the learning model learns control over the wireless network. Control method.

9. A process in a control system, A specific model that identifies controls related to a wireless network collects specific data used for the identification, The aforementioned identification data is transmitted as training data to a learning model that learns control of the wireless network, to another control system that performs training using the learning model, and which is different from the aforementioned control system. A control program that causes the computer of the control system to execute the process.

10. A process in a control system, The other control system, which is different from the aforementioned control system, receives identification data collected by another control system to identify control related to the wireless network using a specific model, as training data from the other control system. Using the aforementioned training data, the learning model learns control over the wireless network. A control program that causes the computer of the control system to execute the process.

Images