Control device for radio access network

Abstract

To suppress reduction in inference speed of a trained model which is used for controlling a radio access network (RAN).SOLUTION: A RAN control device comprises: generation means that generates a trained model by updating, on the basis of training data, a base trained model which has been received from another control device; first control means that performs inference on the basis of the trained model generated by the generation means and that controls the RAN on the basis of an inference result; measurement means that measures an inference speed of the trained model generated by the generation means; and determination means that determines whether or not to update the trained model to increase the inference speed of the trained model, on the basis of a reference value of the inference speed and one or more measurements of the inference speed which have been measured by the measurement means.SELECTED DRAWING: Figure 7

Description

Technical Field

[0001] This disclosure relates to control technologies for radio access networks (RANs).

Background Art

[0002] FIG. 1 shows the control configuration of a RAN proposed by the Open Radio Access Network (O-RAN) Alliance. As shown in FIG. 1, a first control function for performing long-term control and a second control function for performing short-term control are defined. In O-RAN, the first control function is called a Non-Real-Time RAN Intelligent Controller (Non-RT RIC), and the second control function is called a Near-Real-Time RAN Intelligent Controller (Near-RT RIC).

[0003] The first control function and the second control function are connected via an A1 interface. The second control function controls components of the RAN, such as a Central Unit (CU) and a Distributed Unit (DU), via an E2 interface. Note that controlling the CU and DU includes notifying / setting various parameter values used by the CU and DU in their processing to the CU and DU, and instructing the CU and DU to execute some operation. Furthermore, in the following description, the term "RAN" is used as a general term for its components. Therefore, for example, "controlling the RAN" means controlling components of the RAN, such as the CU and DU. The first control function, the second control function, and the RAN are further connected via an O1 interface. The O1 interface can be used for transmitting traffic data, performance data, fault data, etc. measured / detected and accumulated by the RAN.

[0004] Non-Patent Document 1 discloses a configuration that utilizes machine learning for RAN control. Specifically, Non-Patent Document 1 discloses performing machine learning based on various learning data collected from RAN and the like to generate a learning model, making inferences using the learning model, and controlling the RAN according to the inference results. In one of the multiple configurations disclosed in Non-Patent Document 1, a first control function generates a learning model. Then, a second control function controls the RAN using the learning model generated by the first control function. Furthermore, in one of the multiple configurations disclosed in Non-Patent Document 1, the first control function generates a learning model and distributes it to the second control function. The second control function updates the learning model received from the first control function by machine learning and controls the RAN using the learning model.

[0005] Non-Patent Documents 2 and 3 disclose various RAN control contents based on the inference results of the learning model. As an example, the RAN control contents include beamforming control, radio resource allocation, traffic prediction, CU and DU placement control by virtualization technology, and the like.

Prior Art Documents

Non-Patent Documents

[0006]

Non-Patent Document 1

Non-Patent Document 2

Non-Patent Document 3

Summary of the Invention

Problems to be Solved by the Invention

[0007] For example, when the second control function controls the RAN based on the learning model received from the first control function or the learning model obtained by modifying the learning model, it is necessary to complete the inference within the target time according to the control content. However, in the learning model received from the first control function or the learning model obtained by modifying the learning model, there may be cases where the inference cannot be completed within the target time. For example, the second control function executes various RAN controls using a plurality of learning models generated for each of the plurality of control contents. Therefore, due to the limitation of the computer resources of the second control function, the inference speed using each learning model may vary. Furthermore, the inference speed may also vary depending on changes in the situation of the RAN to be controlled, etc. In the present disclosure, the inference speed may be defined as the time from when data is input to the learning model until the inference result is obtained. However, it is also possible to define the inference speed as the time from when data is input to the learning model until the inference result is obtained and the control based on the inference result is completed.

[0008] The present disclosure provides a technique for suppressing a decrease in the inference speed due to the learning model used for RAN control.

Means for Solving the Problems

[0009] According to one aspect of the present invention, a control device for a radio access network (RAN) includes a generation means for generating a learning model by updating a basic learning model received from another control device based on learning data, a first control means for performing an inference based on the learning model generated by the generation means and controlling the RAN based on the inference result, a measurement means for measuring an inference speed of the learning model generated by the generation means, and a determination means for determining whether to update the learning model so that the inference speed of the learning model becomes faster based on one or more measured values of the inference speed measured by the measurement means and a reference value of the inference speed.

Effects of the Invention

[0010] According to the present disclosure, it is possible to suppress a decrease in the inference speed of a learning model used for controlling the RAN.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Modes for Carrying Out the Invention

[0012] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the invention according to the claims, and not all combinations of the features described in the embodiments are essential for the invention. Two or more of the features described in the embodiments may be arbitrarily combined. Also, the same or similar configurations are given the same reference numerals, and duplicate descriptions are omitted.

[0013] <First Embodiment> FIG. 2 shows the control configuration of the RAN according to this embodiment. The configuration of FIG. 2 is basically the same as the configuration of FIG. 1. The first control device 100 performs long-term control of the RAN, for example, a device that implements the function of the Non-RT RIC of O-RAN. The second control device 200 is a device that controls the RAN with a shorter period than the first control device 100, for example, a device that implements the function of the Near-RT RIC of O-RAN. The first control device 100 is connected to one or more second control devices 200. In the example of FIG. 2, the first control device 100 is connected to three second control devices 200, but this is an illustration, and the number of second control devices 200 connected to the first control device 100 can be any number of 1 or more.

[0014] In this embodiment, the first control device 100 performs machine learning to generate a learning model, and distributes the generated learning model to each second control device 200. Note that the first control device 100 generates a learning model corresponding to each of one or more control contents by each second control device 200 and distributes it to each second control device 200. Each second control device 200 is associated with a geographical area, and uses one or more learning models acquired from the first control device 100 to control the components of the RAN arranged in the associated geographical area. Note that what the second control device 200 controls is the part of the geographical area associated with the second control device 200 within the RAN of the mobile communication network, but the part of the geographical area associated with the second control device 200 within the RAN of this mobile communication network is also denoted as the RAN.

[0015] FIG. 3 is a configuration diagram of the first control device 100. The communication unit 14 performs communication processing with the second control device 200 via the A1 interface and communication processing with the second control device 200 and the RAN via the O1 interface. The storage unit 12 stores the learning data collected from the RAN via the O1 interface. In this embodiment, it is assumed that the learning data is collected from the RAN, but a configuration in which the learning data is collected from the second control device 200 may be used. When collecting the learning data from the second control device 200, the O1 interface or the A1 interface can be used. The learning unit 11 performs machine learning based on the learning data stored in the storage unit 12 to generate a learning model. The learning unit 11 transmits the generated learning model to each second control device 200 via the A1 interface of the communication unit 14. Note that the O1 interface can also be used to transmit the learning model to each second control device 200. The control unit 13 transmits and receives control messages to and from each second control device 200 via the communication unit 14. In this embodiment, the control unit 13 transmits and receives control messages to and from each second control device 200 using the O1 interface, but a configuration using the A1 interface may be used. Further, the control unit 13 controls the learning unit 11.

[0016] FIG. 4 is a configuration diagram of the second control device 200. The inference control unit 21 performs inference based on the learning model received from the first control device 100 via the communication unit 24 and controls the RAN via the E2 interface. The control unit 23 transmits and receives control messages to and from the first control device 100 via the communication unit 24. For example, when the control unit 23 receives a measurement request from the first control device 100, the control unit 23 causes the measurement unit 22 to measure the inference speed using the learning model by the inference control unit 21. When the control unit 23 receives measurement data indicating the measurement result of the inference speed from the measurement unit 22, the control unit 23 transmits the measurement result to the first control device 100 by a control message.

[0017] The measurement unit 22 measures the inference speed based on the measurement instruction from the control unit 23 and outputs measurement data indicating the measurement result to the control unit 23. The measurement unit 23 is configured to measure the inference speed based on at least one of, for example, measurement method A and measurement method B. Measurement method A measures the inference speed when the inference control unit 21 actually controls the RAN. Measurement method B measures the inference speed by causing the inference control unit 21 to execute inference with the test data as the input to the learning model. The test data used in measurement method B is, for example, stored in the measurement unit 22 in advance. Alternatively, the test data used in measurement method B is, for example, received from the first control device 100 together with the measurement request. In this embodiment, the first control device 100 acquires learning data from the RAN. However, when the first control device 100 acquires learning data from the second control device 200, the second control device 200 acquires learning data from the RAN via, for example, the O1 interface or the E2 interface, and transmits the learning data to the first control device 100 via the A1 interface or the O1 interface.

[0018] Figure 5 is a sequence diagram of the method according to this embodiment. In S1, the control unit 13 of the first control device 100 transmits a measurement request message instructing the measurement of the inference speed by the learning model to the second control device 200. When the second control device 200 uses a plurality of learning models corresponding to a plurality of control contents, the measurement request message may include information indicating the "target model", which is the learning model for which the inference speed is to be measured. Note that the number of "target models" may be one or more. Further, when the second control device 200 is capable of executing both measurement method A and measurement method B, the measurement request message may include information specifying the measurement method. When specifying measurement method B, the measurement request message may include test data. Furthermore, the measurement request message may include information indicating the measurement period of the inference speed.

[0019] In S2, the control unit 23 of the second control device 200 causes the measurement unit 22 to measure the inference speed of the target model according to the measurement request message. In S3, the control unit 23 of the second control device 200 transmits the measurement result to the first control device 100.

[0020] FIG. 6 shows an example of the measurement result transmitted in S3 when the measurement requests for the inference speeds of the two learning models, target model #1 and target model #2, are received in S1. The measurement unit 22 repeatedly measures the inference speed during the measurement period indicated by the measurement request message. According to FIG. 6, for target model #1, the inference speed is measured at times T#11, T#12, T#13, ···. In FIG. 6, the measured values of the inference speed at times T#11, T#12, and T#13 are 120 ms, 114 ms, and 150 ms, respectively. Similarly, according to FIG. 6, for target model #2, the inference speed is measured at times T#21, T#22, T#23, ···. In FIG. 6, the measured values of the inference speed at times T#21, T#22, and T#23 are 40 ms, 48 ms, and 50 ms, respectively.

[0021] For example, in the case of measurement method A, the measurement unit 22 measures the inference speed in the inference that the inference control unit 21 executes every control cycle. Also, in the case of measurement method B, the measurement unit 22 measures the inference speed using test data at a predetermined cycle.

[0022] Furthermore, the measurement result may also include information indicating the reference values of the inference speeds of target model #1 and target model #2, respectively. According to FIG. 6, the reference value of the inference speed by target model #1 is 100 ms, and the reference value of the inference speed by target model #2 is 50 ms. The reference value of the inference speed may be stored in advance in the control unit 23, for example. Alternatively, the reference value of the inference speed may be dynamically determined by the control unit 23 based on some index, for example. Note that when the learning model is associated with the control content and the reference value of the inference speed determined according to the control content is also set in the first control device 100, it is not necessary to include the reference value of the inference speed in the measurement result.

[0023] In S4, based on the measurement results received in S3, the control unit 13 of the first control device 100 determines whether it is necessary to distribute to the second control device 200 another learning model with a faster inference speed for each of the target model #1 and the target model #2 with the same control content. This determination can be made based on a comparison between a predetermined measurement value among one or more measurement values of the inference speed of the target model acquired in S3, or the average value of the one or more measurement values, and the reference value of the inference speed of the target model.

[0024] As an example, the predetermined measurement value among the one or more measurement values is the minimum measurement value among the one or more measurement values. In this case, the control unit 13 may determine that it is necessary to distribute to the second control device 200 another learning model with a faster inference speed for the same control content when the minimum measurement value among the one or more measurement values exceeds the reference value. Also, as another example, the predetermined measurement value among the one or more measurement values is the maximum measurement value among the one or more measurement values. In this case, the control unit 13 may determine that it is necessary to distribute to the second control device 200 another learning model with a faster inference speed for the same control content when the maximum measurement value among the one or more measurement values exceeds the reference value.

[0025] As yet another example, the predetermined measurement value among the one or more measurement values is the measurement value of a predetermined rank in descending order of the one or more measurement values, or the measurement value corresponding to a predetermined percentile value among the one or more measurement values. Also in this case, the control unit 13 may determine that it is necessary to distribute to the second control device 200 another learning model with a faster inference speed for the same control content when the predetermined measurement value exceeds the reference value.

[0026] Furthermore, instead of the predetermined measurement value among the one or more measurement values, a configuration may be adopted in which the control unit 13 determines that it is necessary to distribute to the second control device 200 another learning model with a faster inference speed for the same control content when a value obtained by performing a predetermined calculation on the one or more measurement values, such as the average value of the one or more measurement values, exceeds the reference value.

[0027] For example, according to FIG. 6, the reference value of the inference speed of the target model #2 is 50 ms, and the measured values of the inference speed are 40 ms, 48 ms, 50 ms, ···, and there are measured values below the reference value. For example, when it is determined that it is necessary to distribute another learning model with a faster inference speed to the second control device 200 with the same control content when the minimum measured value exceeds the reference value, the control unit 13 determines that for the target model #2, it is not necessary to distribute another learning model with a faster inference speed to the second control device 200 with the same control content.

[0028] On the other hand, according to FIG. 6, the reference value of the inference speed of the target model #1 is 100 ms, and the measured values of the inference speed are 120 ms, 114 ms, 150 ms, ···, and there are measured values exceeding the reference value. For example, when it is determined that it is necessary to distribute another learning model with a faster inference speed to the second control device 200 with the same control content when the minimum measured value exceeds the reference value, the control unit 13 determines that for the target model #1, it is necessary to distribute another learning model with a faster inference speed to the second control device 200 with the same control content. In this case, the control unit 13 instructs the learning unit 11 to send another learning model with a faster inference speed to the second control device 200 with the same control content as the target model #1. The learning unit 11 sends another learning model to the second control device 200 in S5 based on the instruction.

[0029] In addition, in the learning unit 11, a plurality of learning models for a certain control content can be generated in advance. The inference speeds of the plurality of learning models for one control content are different from each other. Also, in the learning unit 11, one learning model for a certain control content can be generated in advance, and another learning model with a faster inference speed than the said learning model can be generated by a compression technique, for example, symmetric static quantization technique. Note that by compressing the learning model by the compression technique, the inference speed of the learning model becomes faster than that of the original learning model, but the inference performance generally deteriorates. Therefore, the control unit 13 may be configured to select, as the learning model to be transmitted to the second control device 200 in S5, for example, the learning model with the highest inference performance within a range where the average value, minimum value, and maximum value of the inference speed do not exceed the reference value.

[0030] With the above configuration, it is possible to suppress a decrease in the inference speed of the learning model used for RAN control.

[0031] <Second Embodiment> Subsequently, regarding the second embodiment, the differences from the first embodiment will be mainly described. The RAN control configuration in this embodiment is as shown in FIG. 2, and the configuration of the first control device 100 is as shown in FIG. 3.

[0032] FIG. 7 is a configuration diagram of the second control device 200 according to this embodiment. The second control device 200 of this embodiment includes a storage unit 26 that stores learning data, and a learning unit 25. The storage unit 26 stores learning data collected from the RAN via the O1 interface or the E2 interface. The learning unit 25 updates the basic learning model received from the first control device 100 via the communication unit 24 by machine learning based on the learning data to generate a learning model. The inference control unit 21 performs inference based on the learning data learned by the learning unit 25 and controls the RAN via the E2 interface.

[0033] In this embodiment, even when there is no measurement request from the first control device 100, the control unit 23 causes the measurement unit 22 to measure the inference speed at a predetermined timing. Then, based on the measured value of the inference speed and the reference value, for example, compression processing of the learning model is performed so that the inference speed satisfies a predetermined condition. The predetermined condition can be, for example, the same as the condition for determining in S4 of FIG. 5 of the first embodiment that it is not necessary to distribute another learning model with a faster inference speed to the second control device 200 with the same control content. That is, for example, the predetermined condition can be that a predetermined measured value or average value among one or more measured values of the measured inference speed does not exceed the reference value. Further, the control unit 23 executes the processing of FIG. 5 triggered by the reception of a measurement request from the first control device 100.

[0034] Different from the first embodiment, in this embodiment, in the learning unit 25, the learning model can be updated using, for example, compression processing so that the inference speed satisfies a predetermined condition. However, depending on the basic learning model first received from the first control device 100, there may be a case where the learning model cannot be updated in the processing in the learning unit 25 so that the inference speed satisfies a predetermined condition. In that case, according to the sequence of FIG. 5, the first control device 100 can determine that the inference speed in the second control device 200 does not satisfy a predetermined condition. When the first control device 100 determines that the inference speed in the second control device 200 does not satisfy a predetermined condition, in S5, it transmits another basic learning model with a faster inference speed to the second control device 200 with the same control content as the learning model that does not satisfy the predetermined condition.

[0035] With the above configuration, it is possible to suppress a decrease in the inference speed by the learning model used for RAN control. Note that, as in the first embodiment, in this embodiment as well, based on one or more measured values of the inference speed received from the second control device 200, the first control device 100 determines whether it is necessary to transmit another basic learning model with a higher inference speed to the second control device 200. However, when it is determined that the learning model cannot be updated to satisfy a predetermined condition with the basic learning model received by the second control device 200 from the first control device 100, the second control device 200 may be configured to request the first control device 100 for another basic learning model with a higher inference speed.

[0036] Note that the first control device 100 and the second control device 200 can be implemented by one device, for example, one computer. Further, the first control device 100 and the second control device 200 can be implemented by a plurality of devices capable of communicating with each other, for example, a plurality of computers.

[0037] Furthermore, according to the present disclosure, a program executable by one or more processors is provided. When the program is executed by one or more processors of a device, the program includes instructions for causing the device to function as the first control device 100 or the second control device 200. Further, according to the present disclosure, a non-transitory computer-readable storage medium storing the above program is provided. Further, according to the present disclosure, a method executed by the first control device 100 and a method executed by the second control device 200 for suppressing a decrease in the inference speed by the learning model used for RAN control, that is, the method shown in FIG. 5 and the like are provided. Further, according to the present disclosure, a program for causing a device having one or more processors to execute the method executed by the first control device 100 and the method executed by the second control device 200, and a non-transitory computer-readable storage medium storing the program are provided.

[0038] The invention is not limited to the above embodiments, and various modifications and changes are possible within the scope of the gist of the invention.

[0039] With the above configuration, it is possible to suppress a decrease in the inference speed by the learning model used for RAN control. Therefore, it becomes possible to contribute to Goal 9 of the Sustainable Development Goals (SDGs) led by the United Nations, "Build resilient infrastructure, promote sustainable industrialization and foster innovation."

Explanation of Signs

[0040] 25: Learning Unit, 21: Inference Control Unit, 22: Measurement Unit, 23: Control Unit

Claims

1. A control device for a radio access network (RAN), comprising: a generation means for generating a learning model by updating a basic learning model received from another control device based on learning data; a first control means for performing inference based on the learning model generated by the generation means and controlling the RAN based on the inference result; a measurement means for measuring the inference speed of the learning model generated by the generation means; a determination means for determining whether to update the learning model so that the inference speed of the learning model becomes faster based on one or more measured values of the inference speed measured by the measurement means and a reference value of the inference speed; A control device comprising the above.

2. The control device according to claim 1, wherein when the determination means determines to update the learning model, the generation means updates the learning model by compression processing of the learning model so that the inference speed of the learning model becomes faster.

3. The control device according to claim 1, wherein the determination means determines to update the learning model when a predetermined measured value among the one or more measured values exceeds the reference value, or when a value based on the one or more measured values exceeds the reference value.

4. The control device according to claim 1, further comprising a second control means for controlling to transmit the one or more measured values measured by the measurement means to the another control device in response to a measurement instruction from the another control device.

5. The control device according to claim 4, wherein when the generation means receives another basic learning model from the another control device as a response to transmitting the one or more measured values measured by the measurement means to the another control device, the generation means generates a learning model by updating the another basic learning model based on learning data.

6. The control device according to claim 4, wherein the second control means transmits the reference value of the inference speed to the another control device together with the one or more measured values.

7. The control device according to claim 1, wherein the another control device is a device implementing Non-RT RIC.

8. The control device according to claim 1, wherein the control device is a device implementing Near-RT RIC.

9. A program that, when executed by one or more processors of a device having one or more processors, causes the device to function as the control device according to any one of claims 1 to 8.

Images