Method for providing machine learning models to mobile devices

Abstract

A method for providing a machine learning model to a mobile terminal of a mobile communication system is described, the method comprising a mobile radio communication network of the communication system determining that a machine learning model should be distributed to the mobile terminal to be used by the mobile terminal for a machine learning task, and providing the requested machine learning model to the mobile terminal via control plane communication.

Description

[Technical Field]

[0001] The present disclosure relates to a method for providing machine learning models to a mobile terminal. [Background technology]

[0002] In communication systems, such as 5G mobile communication systems, it is important to ensure that a certain quality of service can be maintained. To this end, various information, such as load, resource usage, available components, user mobility, component status, etc., can be monitored and taken into account when controlling the communication system, e.g., taking measures when overload is imminent to avoid degradation of service quality. Collecting and / or evaluating this information can be performed using machine learning models trained accordingly for the corresponding predictive tasks. For example, using artificial intelligence (AI) and machine learning (ML) techniques is a promising approach to improving the performance of NG-RAN (Next Generation Radio Access Network) air interfaces in several use cases, such as improved channel status information (CSI) feedback, beam management optimization, and positioning accuracy. However, this requires that, at least in some cases, the machine learning model be used on the terminal side, but due to the effort required for training, model training on the terminal side is difficult (due to constrained power and computational resources), and model training should typically be offloaded to the network, i.e., the model should be trained on the network side (e.g., a network function or an application function) and provided to the mobile terminal when needed. Therefore, an efficient technique for providing machine learning models to mobile terminals in a mobile communication system is desired. Summary of the Invention

[0003] A method for providing a machine learning model to a mobile terminal of a mobile communication system is provided, the method including a mobile radio communication network of the communication system determining that a machine learning model should be distributed to the mobile terminal to be used by the mobile terminal for a machine learning task, and providing the machine learning model to the mobile terminal via control plane communication. [Brief explanation of the drawings]

[0004] [Figure 1] FIG. 1 illustrates a communication system according to one embodiment. [Figure 2] A flow diagram 200 showing a (network-side) ML (machine learning) model search procedure is shown. [Figure 3] 3 shows a flow diagram 300 illustrating an embodiment in which an Access and Mobility Management Function (AMF) acts as a model provider. [Figure 4] FIG. 1 is a flow diagram illustrating an embodiment in which a Session Management Function (SMF) acts as a model provider. [Figure 5] FIG. 10 is a flow diagram illustrating an embodiment in which a model is requested and provided via RRC (Radio Resource Control) signaling. [Figure 6] FIG. 1 is a flow diagram illustrating a method for providing a machine learning model to a mobile terminal in a mobile communication system according to one embodiment. [Figure 7] FIG. 1 is a flow diagram illustrating a method for acquiring a machine learning model from a mobile communication system according to one embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0005] In the drawings, like reference numbers generally refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention. In the following description, various aspects are described with reference to the following drawings:

[0006] The following detailed description refers to the accompanying drawings, which show, by way of example, specific details and aspects of the present disclosure in which the invention may be practiced. Other aspects may be utilized, and structural, logical, and electrical changes may be made, without departing from the scope of the present invention. Various aspects of the present disclosure are not necessarily mutually exclusive, as some aspects of the present disclosure may be combined with one or more other aspects of the present disclosure to form new aspects.

[0007] Various examples corresponding to aspects of the present disclosure are described below.

[0008] Example 1 is a method for providing a machine learning model to a mobile terminal of a mobile communication system, the method including: a mobile radio communication network of the communication system determining that a machine learning model should be distributed to the mobile terminal for use by the mobile terminal for a machine learning task; and providing the machine learning model to the mobile terminal via control plane communication.

[0009] Example 2 is the method of example 1, including determining that the machine learning model should be distributed to the mobile terminal based on one or more of temporal and / or spatial variations in operating conditions, degradation of machine learning model performance, configuration of the mobile terminal, and operator policies of the mobile communications network.

[0010] Example 3 is the method of example 1, including: determining, by the mobile wireless communications network, in response to receiving a request for the machine learning model from the mobile terminal, that the machine learning model should be delivered to the mobile terminal; and providing, in response to the request, the machine learning model to the mobile terminal.

[0011] Example 4 is the method of example 3, wherein the request includes an indication that a machine learning model is requested for a machine learning task at the mobile terminal.

[0012] Example 5 is the method of example 3 or 4, including receiving the request at a core network of the communication network.

[0013] Example 6 is the method of Examples 3 or 4, including receiving the request at a radio access network of the communications network; generating a second request for the machine learning model at the radio access network; sending the second request to a core network of the communications network; providing the machine learning model from the core network to the radio access network; and providing the machine learning model from the radio access network to the mobile terminal.

[0014] Example 7 is the method of example 6, including receiving a request and providing the machine learning model from the radio access network to the mobile terminal via a Radio Resource Control message.

[0015] Example 8 is the method of example 6 or 7, including sending a second request to the core network and providing the machine learning model from the core network to the radio access network via a Next Generation Application Protocol message.

[0016] Example 9 is the method of any one of Examples 1 to 8, including the core network component discovering a storage location of the machine learning model, retrieving the machine learning model from the discovered storage location, and providing the machine learning model to the mobile terminal.

[0017] Example 10 is the method of example 9, wherein the core network component is an Access and Mobility Management Function or a Session Management Function.

[0018] Example 11 is the method of any one of Examples 1 to 10, wherein the machine learning model is used by the mobile terminal for prediction tasks and / or for compression and / or encoding and decompression and / or decoding tasks.

[0019] Example 12 is a communications network configured to perform the method of any one of Examples 1 to 11.

[0020] Example 13 is a method of obtaining a machine learning model from a mobile communication system, comprising: sending a request from the mobile terminal to a mobile radio communication network of the communication system for a machine learning model to be used by the mobile terminal for a machine learning task; and receiving the machine learning model via control plane communication in response to the request.

[0021] Example 14 is the method of example 13, including determining, by the mobile terminal, that the machine learning model should be requested based on one or more of temporal and / or spatial changes in operating conditions, degradation of performance of the machine learning model, configuration of the mobile terminal, and operator policies of the mobile communications network.

[0022] Example 15 is a mobile terminal configured to perform the method of example 13 or 14.

[0023] It should be noted that one or more features of any of the above examples may be combined with any one of the other examples, and in particular, embodiments described in the context of an apparatus are equally valid for a method.

[0024] According to further embodiments, there is provided a computer program and computer readable medium comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of the above embodiments.

[0025] Various examples are described in more detail below.

[0026] FIG. 1 illustrates a communication system according to one embodiment.

[0027] The communication system in this example comprises a Radio Access Network (RAN) 101, a core network 103 and a transport (communication) network 102 connecting the RAN 101 to the core network 103. Subscriber terminals (denoted UE according to 3GPP) 104 are connected to distributed units 105 of the RAN 101 which are connected to a centralized unit 106 (for implementing a base station of the RAN 101). Application functions 107 are connected to the core network 103. The core network 103, for example a 5G core network (5GC), includes various network functions (NFs) such as an AMF (Access and Mobility Management Function) 108, an SMF (Session Management Function) 109, an NEF (Network Exposure Function) 110, an UPF (User Plane Function) 111, an NWDAF (Network Data Analytics Function) 112, and a PCF (Policy Control Function) 113.

[0028] The core network 103 is coupled to an OAM (Operation, Administration and Maintenance) system 114 .

[0029] In communication systems such as those shown in Figure 1, machine learning (ML) models are used to perform prediction tasks for various purposes, e.g. Enhanced CSI (Channel State Information) feedback, e.g., reducing overhead and improving accuracy of predictions Spatial frequency domain CSI compression using two-sided AI models Time-domain CSI prediction using UE-side model • Beam management, e.g., beam prediction in time and / or spatial domains for reduced overhead and latency, improved beam selection accuracy, e.g. Spatial domain downlink beam prediction for the first set of beams A based on the measurement results of the second set of beams Time downlink beam prediction for the first set of beams based on historical measurements of the second set of beams ●Improved positioning accuracy for different scenarios, including those with heavy NLOS (No Line of Sight) conditions 〇Direct AI / ML positioning AI / ML-assisted positioning to optimize the performance of the radio link between the subscriber terminal 104 and the RAN 101. Additionally, AI / ML techniques, e.g. ●Network energy saving, for example, ML models Input: UE mobility / trajectory, current energy efficiency, UE measurement report Output: predicted energy efficiency, handover strategy Used in Load balancing, for example for ML models, Input: UE trajectory, UE traffic, RAN resource status Output: Predicted resource status information Used in ● Mobility optimization, for example, ML models, Input: UE location information, radio measurement, UE handover Output: Predicted handover, UE traffic prediction Used in In particular, it can be used to improve RAN performance, for example.

[0030] For one of the above purposes, the corresponding ML model may be trained by an entity, for example, a RAN entity, in particular, a base station (denoted as gNB in ​​5G), the OAM 114, a component (e.g., a network function) of the core network 103, or a third party (e.g., corresponding to the AF 107 or connected to the core network 103 via the AF 107). This training entity is hereinafter referred to as a Model Training NF (MTF). The MTF may be, for example, the NWDAF 112 including an MTLF (Model Training Logical Function) with an extension supporting training models for RAN use cases (e.g., CSI prediction models) that provide models to NFs other than the AnLF (Analytical Logical Function), such as the AMF, UPF, or SMF. The model training entity, i.e., the MTF, collects training data and trains the respective ML model. The trained model (i.e., the trained model's specifications, e.g., for neural network weights) is then stored in a storage device, hereinafter referred to as a Model Repository NF (MRF), that can store the trained model and provide the model to other components and entities (e.g., NFs). The MRF is, for example, an Analytics Data Repository Function (ADRF) that has the ability to provide stored models to NFs other than the AnLF, such as AMF, UPF, and SMF.

[0031] The trained model can then be used at multiple inference locations, which may include, among others, one or more mobile terminals. Thus, according to various embodiments, techniques are provided for distributing ML models from the network side (via the RAN), e.g., the RAN, 5GC, OAM, or a third party or AF, to the UE.

[0032] According to various embodiments, model distribution is enabled by extending the interfaces in the UE and 5GC NF for model distribution in the control plane (CP). Note that training locations inside and outside the core network 103 (e.g., 5GC) are supported. The ML model may be a RAN model, i.e., an ML model trained specifically for a RAN task, i.e., to make predictions regarding RAN operation.

[0033] In the embodiments described below, it is assumed that there is a registration and discovery procedure whereby an MTF registers its supported models along with associated ML model filter information with an NRF (Network Repository Function), and other NFs discover and select MTFs via querying the NRF for specific models and corresponding filters.

[0034] FIG. 2 shows a flow diagram 200 illustrating the ML model search procedure (network side).

[0035] The flow involves a consumer NF 201 (i.e., an NF (Network Function) acting as a consumer, i.e., retrieving a trained ML model), such as an AMF, SMF, UPF, or NEF), an NRF 202, an MTF 203, and an MRF 204.

[0036] At 205, the MTF 203 registers with the NRF 202, informing the NRF 202, among other things, that the MTF 203 has ML model training and provisioning capabilities.

[0037] At 206, the consumer NF 201 sends a discovery request to the NRF 202, and the NRF 202 responds at 207 with a discovery request response to the NRF 220 that informs the consumer NF 201 about the MTF 203.

[0038] At 208, the consumer NF 201 requests (by get or subscribe) an ML model from the MTF. The request may include various parameters related to the requested model. In particular, if the ML model is to be used on the terminal (UE) side, the request may include information about the UE type, UE vendor, and UE location.

[0039] If the requested ML model is stored in the MTF 203, the MTF 203 provides the requested model to the consumer NF 201 at 209.

[0040] If the requested ML model is not stored in the MTF 203 but is stored in the MRF 204, the MTF 203 informs the consumer NF 201 of the identity of the MRF 204 at 210, the consumer NF 201 requests the ML model from the MRF at 211, and the MRF provides the requested ML model at 212.

[0041] Both the terminal and the network should be capable of ML model distribution from the network to the terminal, i.e., the terminal should be able to request and accept models from the network. The network should also be able to train and distribute models to the terminal. The mobile terminal should be able to know its capabilities and characteristics regarding ML models, e.g., UE model provisioning capability (i.e., the UE has the capability to obtain an AIML model from the network (e.g., a 5G communication system (5GS))) A list of supported ML models (e.g., ML models for CSI) User device information (e.g., UE type, vendor, hardware, ...) The network may be notified about

[0042] The mobile terminal may provide information about its capabilities and characteristics, for example, as part of a UE capability transfer (i.e., in a UECapabilityEnquiry message to the network in response to a UECapabilityInformation message from the network) or on its registration request (which it sends to the RAN to register with the network).

[0043] The network may perform a Network ML Model Provisioning Capability Advertisement, i.e., inform the UE of its capabilities for ML model provisioning, e.g., through system information in a System Information Block (SIB), e.g., a System Information message (e.g., in response to a SystemInformationRequest). The network may provide, for example, the following system information: Network Model Provisioning Capability (i.e., the network has the capability to provide ML models to the UE) A list of supported ML models (e.g., ML models for CSI)

[0044] In the following, an embodiment will be described in which the ML model is provided to the mobile terminal from the network side via control plane communication.

[0045] FIG. 3 shows a flow diagram 300 illustrating an embodiment in which AMF acts as a model provider.

[0046] The flow involves a mobile terminal 301, a RAN 302, an AMF 303, an NRF 304, an MTF 305, and an MRF 306.

[0047] In 307, the mobile terminal 301 requests an ML model from the AMF 303. The request may include, for example, an ML model provisioning content according to 3GPP, e.g. List of Analytics IDs: Identifies the analytics in which the ML model will be used Use case context: Provides the context for the use of analytics to select the most relevant ML model. ●ML Model Interoperability Information ●ML Model Target Period ●Accuracy level(s) of Interest ●Time when the model is needed It includes ML model information that may be:

[0048] Furthermore, according to various embodiments, the ML model information is extended with an indication that the requested model should support that the UE is the ultimate model consumer (i.e., the ML model can be used on the terminal side).

[0049] At 308, the AMF 303 performs MTF discovery and selection, followed by ML model search at 309 (eg, as described with reference to FIG. 2).

[0050] At 310, the AMF 303 provides the requested ML model (i.e., the trained model specifications, e.g., in terms of neural network weights) to the mobile terminal via an ML model provisioning response.

[0051] FIG. 4 shows a flow diagram 400 illustrating an embodiment in which an SMF acts as a model provider.

[0052] The flow involves a mobile terminal 401, a RAN 402, an AMF 403, an SMF 404, an NRF 405, an MTF 406, and an MRF 407.

[0053] At 408, the mobile terminal 401 requests an ML model from the AMF 403. The request includes ML model information, which may be, for example, ML model provisioning content according to 3GPP (see description of FIG. 3), extended according to various embodiments with an indication that the requested model should support that the UE is the ultimate model consumer (i.e., that the ML model can be used on the terminal side).

[0054] At 409, the AMF 403 sends a corresponding request for the ML model to the AMF 404.

[0055] At 410, the SMF 404 performs MTF discovery and selection, followed by ML model search at 411 (eg, as described with reference to FIG. 2).

[0056] At 412, the SMF 404 provides the AMF 403 with the requested ML model (i.e., the specifications of the trained model, e.g., in terms of neural network weights).

[0057] At 413, the AMF 403 provides the requested ML model (i.e., the trained model specification) to the mobile terminal via an ML model provisioning response.

[0058] For communication between the AMF 403 and the SMF 404 at 409 and 412, for example, a service-based architecture (SBA) service for ML model provisioning may be introduced.

[0059] FIG. 5 shows a flow diagram 500 illustrating an embodiment in which a model is requested and provided via Radio Resource Control (RRC) signaling.

[0060] The flow involves a mobile terminal 501, a RAN 502, an AMF 503, an NRF 504, an MTF 505, and an MRF 506.

[0061] At 507, the mobile terminal 501 requests an ML model from the RAN 502. The request includes ML model information, which may be, for example, ML model provisioning content according to 3GPP (see description of FIG. 3), extended according to various embodiments with an indication that the requested model should support that the UE is the ultimate model consumer (i.e., that the ML model can be used on the terminal side).

[0062] At 508, the RAN 502 sends a corresponding request for the ML model to the AMF 503.

[0063] At 509, the AMF 503 performs MTF discovery and selection, followed by ML model search at 510 (eg, as described with reference to FIG. 2).

[0064] At 511, the AMF 503 provides the RAN 502 with the requested ML model (i.e., the trained model specification, e.g., in terms of neural network weights), which the RAN 102 forwards to the mobile terminal 501 at 512.

[0065] For communication between the mobile terminal 501 and the RAN 502 at 507 and 511, additional RRC (Radio Resource Control) messages for ML model request and delivery may be introduced. For communication between the RAN 502 and the AMF 503 at 508 and 511, the NGAP (Next Generation Application Protocol) protocol may be extended by additional messages for ML model request and delivery.

[0066] Note that all request messages (to request an ML model) and response messages (to provide an ML model) in the above examples can be implemented as pairs of subscribe and notify messages. This enables network-triggered ML model distribution, where a mobile terminal subscribes to a particular model and then the network provides customized / updated versions of the model, taking into account, for example, operator policies, monitoring of ML model performance, mobile terminal mobility / handover, etc.

[0067] It is further noted that the various approaches for ML model provisioning are not exclusive, and multiple approaches may be implemented and used according to operator policies, mobile terminal configuration, etc.

[0068] In summary, according to various embodiments, a method is provided as shown in FIGS.

[0069] FIG. 6 shows a flow diagram 600 illustrating a method for providing a machine learning model to a mobile terminal in a mobile communication system, according to one embodiment.

[0070] In 601, a mobile radio communication network of a communication system determines that a machine learning model should be distributed to a mobile terminal to be used by the mobile terminal for a machine learning task (e.g., inference, e.g., a prediction task, e.g., a prediction task for controlling radio access network operation or data compression or encoding (e.g., as an autoencoder)). For example, this is determined in response to receiving a request from the mobile terminal for the machine learning model. The request may be a subscription (i.e., a subscribe message) to a corresponding model providing service.

[0071] At 602, the mobile wireless communication network provides the machine learning model to the mobile terminal via control plane communication (in response to determining that the machine learning model should be distributed to the mobile terminal).

[0072] According to various embodiments, a communications network configured to perform the method described with reference to FIG. 6 is provided.

[0073] The communication network may correspond, for example, to the RAN 101, the transport network 102, and / or the core network 103 of the communication system of FIG.

[0074] FIG. 7 illustrates a flow diagram 700 illustrating a method for obtaining a machine learning model from a mobile communication system according to one embodiment.

[0075] At 701, a mobile terminal transmits a request to a mobile radio communication network of a communication system for a machine learning model to be used by the mobile terminal for an ML task.

[0076] At 702, the mobile terminal receives the requested machine learning model via control plane communication in response to the request.

[0077] According to various embodiments, there is provided a mobile terminal configured to perform the method described with reference to FIG.

[0078] The mobile terminal may correspond, for example, to one of the UEs 104 of the communication system of FIG.

[0079] In other words, according to various embodiments, the ML model is provided to the mobile terminal via the control plane, which allows for example to use the ML model on the terminal side, e.g., on the air interface, for communication optimization.

[0080] As mentioned above, many variations are possible that provide a flexible set approach.

[0081] Providing a machine learning model via control plane communication may be understood as transmitting the specifications of the machine learning model, i.e., the values ​​of the trained parameters of the machine learning model (such as neural network weights if the machine learning model is a neural network), via control plane communication (i.e., via the air interface) from the network side (i.e., the mobile radio communication network) to the mobile terminal, i.e., using communication channels, interfaces and / or reference points belonging to the control plane (of the mobile communication system).

[0082] The communication network and mobile terminal may include and / or be implemented by data processing components such as one or more processors, memories, interfaces, receivers, transmitters, antennas, etc., and may be implemented, for example, by one or more circuits. A "circuit" may be understood as any kind of logic implementation entity, and may be a dedicated circuit or a processor that executes software, firmware, or any combination thereof stored in memory. Thus, a "circuit" may be a hardwired logic circuit or a programmable processor, e.g., a programmable logic circuit such as a microprocessor. A "circuit" may also be a processor that executes software, e.g., any kind of computer program. Any other kind of implementation of each of the above-mentioned functions may also be understood as a "circuit."

[0083] While particular embodiments have been described, it should be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the embodiments of the present disclosure as defined by the appended claims. The scope is therefore indicated by the appended claims, and all changes that come within the meaning and range of equivalents of the claims are therefore intended to be embraced.

Claims

1. 1. A method for providing a machine learning model to a mobile terminal in a mobile communication system, comprising: determining, by a mobile radio communication network of the communication system, that a machine learning model should be distributed to a mobile terminal to be used by the mobile terminal for a machine learning task; providing the machine learning model to the mobile terminal via control plane communication. method.

2. determining that the machine learning model should be distributed to the mobile terminal based on one or more of temporal and / or spatial variations in operating conditions, degradation of machine learning model performance, configuration of the mobile terminal, and operator policies of the mobile communications network; The method of claim 1.

3. 2. The method of claim 1, comprising: determining, by the mobile radio communications network, in response to receiving a request for the machine learning model from the mobile terminal, that the machine learning model should be distributed to the mobile terminal; and providing the machine learning model to the mobile terminal in response to the request.

4. The method of claim 3 , wherein the request includes an indication that the machine learning model is requested for a machine learning task at the mobile terminal.

5. 5. The method of claim 3 or 4, comprising receiving the request in a core network of the communications network.

6. 5. The method of claim 3, comprising receiving the request at a radio access network of the communications network, generating a second request for the machine learning model at the radio access network, transmitting the second request to a core network of the communications network, providing the machine learning model from the core network to the radio access network, and providing the machine learning model from the radio access network to the mobile terminal.

7. 7. The method of claim 6, comprising receiving the request and providing the machine learning model from the radio access network to the mobile terminal via a Radio Resource Control message.

8. 8. The method of claim 6 or 7, comprising sending the second request to the core network and providing the machine learning model from the core network to the radio access network via a Next Generation Application Protocol message.

9. 9. The method of claim 1, comprising: a core network component discovering a storage location of the machine learning model; retrieving the machine learning model from the discovered storage location; and providing the machine learning model to the mobile terminal.

10. The method of claim 9 , wherein the core network component is an Access and Mobility Management Function or a Session Management Function.

11. 11. The method according to claim 1, wherein the machine learning model is used by the mobile terminal for prediction tasks and / or for compression and / or encoding and decompression and / or decoding tasks.

12. A communications network configured to perform a method according to any one of claims 1 to 11.

13. 1. A method for obtaining a machine learning model from a mobile communication system, comprising: sending a request from the mobile terminal to a mobile radio communications network of the communications system for a machine learning model to be used by the mobile terminal for a machine learning task; and receiving the machine learning model via control plane communication in response to the request.

14. 14. The method of claim 13, comprising determining that the mobile terminal should request the machine learning model based on one or more of temporal and / or spatial changes in operating conditions, degradation of machine learning model performance, configuration of the mobile terminal, and operator policies of the mobile communications network.

15. A mobile terminal configured to perform the method according to claim 13 or 14.

Images