Deciding on a Machine Learning Architecture for Network Slicing

Abstract

To provide a method and a device that determine a machine learning architecture for network slicing.SOLUTION: In a wireless network environment 100 in which user equipment (UE) 110 and a network-slice manager 190 communicate with each other to determine a machine learning (ML) architecture, the UE 110 selects a machine learning architecture that provides the quality of service level required by an application. The network-slice manager 190 accepts or rejects requests on the basis of one or more available end-to-end machine learning architectures associated with network slices that support the quality of service level required by the application.SELECTED DRAWING: Figure 1

Description

[Background technology]

[0001] background The evolution of wireless communication networks is often driven by changing user needs or required quality-of-service (QoS) levels. A quality-of-service level is defined by one or more quality-of-service parameters, such as latency, throughput, reliability, or error rate. Over time, wireless communication networks have evolved from meeting quality-of-service levels associated with voice calls to meeting quality-of-service levels associated with web access and video streaming. Due to the increasing amount of applications and different capabilities of user equipment (UE), it may be difficult to meet the diverse quality-of-service levels required in a single wireless communication network. Summary of the Invention [Problem to be solved by the invention]

[0002] overview Techniques and apparatus for determining a machine-learning (ML) architecture for network slicing are described. A user equipment (UE) and a network slice manager communicate with each other to determine a machine-learning (ML) architecture, after which the UE employs the machine-learning architecture to wirelessly communicate data for an application. In particular, the UE selects a machine-learning architecture that provides a quality of service level required by the application. The network slice manager accepts or rejects the request based on one or more available end-to-end machine-learning architectures associated with a network slice that supports the quality of service level required by the application. The UE and the network slice manager can cooperate to determine an appropriate machine-learning architecture that meets the quality of service level associated with the application and forms part of an end-to-end machine-learning architecture that meets the quality of service level required by the application. [Means for solving the problem]

[0003] Aspects described below include a method performed by a user equipment to determine a machine learning architecture for network slicing. The method comprises executing a first application associated with a first requested quality of service level and selecting a first machine learning architecture based on the first requested quality of service level. The method further comprises sending a first machine learning architecture request message to a network slice manager of a wireless network requesting permission to use the first machine learning architecture. The method further comprises receiving a first machine learning architecture response message from the network slice manager granting permission to use the first machine learning architecture based on the first network slice. The method further comprises wirelessly communicating data for the first application using the first machine learning architecture.

[0004] The machine learning architecture implemented by the UE, the machine learning architecture implemented by the base station, and the machine learning architecture implemented by the 5G Core Network (5GC) entities are operable together to form an end-to-end machine learning architecture. A machine learning interface passes information between the machine learning architectures. In this way, the end-to-end machine learning architecture is a single distributed It may function as a machine learning architecture and may provide an interface between the UE, at least one base station, and 5GC entities.

[0005] A machine learning architecture may represent a particular structure of one or more layers and the connections between these layers.

[0006] Network slicing can instantiate one or more parallel and independent networks that meet different quality of service levels.

[0007] The quality of service level may be defined by one or more quality of service parameters, such as latency, throughput, reliability or error rate.

[0008] Selecting the first machine learning architecture may include receiving an available machine learning architectures message from the network-slice manager identifying available machine learning architectures associated with the first network slice. Selecting the first machine learning architecture may further include selecting the first machine learning architecture from the available machine learning architectures.

[0009] Selecting the first machine learning architecture may include receiving a machine learning architecture selection message from the network-slicing manager including one or more selection rules. Selecting the first machine learning architecture may further include executing the one or more selection rules to determine the first machine learning architecture.

[0010] The method may further comprise wirelessly communicating other data for the first application using a second machine learning architecture. Selecting the second machine learning architecture may be based on the first requested quality of service level. Selecting the second machine learning architecture may be responsive to the user device moving to a geographic location associated with a different base station. Selecting the second machine learning architecture may be responsive to the user device moving to a geographic location associated with a different tracking area. Selecting the second machine learning architecture may be responsive to the user device roaming between a home network and a destination network.

[0011] The home network may be a network in which the UE is registered, and the visited network may be a network in which the UE is not registered.

[0012] The method may further comprise executing a second application associated with the second requested quality of service level. The method may further comprise selecting a second machine learning architecture based on the second requested quality of service level. The method may further comprise sending a second machine learning architecture request message to a network slice manager requesting permission to use the second machine learning architecture. The method may further comprise receiving a second machine learning architecture response message from the network slice manager granting permission to use the second machine learning architecture based on the first network slice or the second network slice.

[0013] The method may further comprise wirelessly communicating other data for the second application using the second machine learning architecture in response to receiving the second machine learning architecture response message. Selecting the second machine learning architecture may be based on the second requested second quality of service level. Running the second application may occur during at least an overlapping portion of the time that the first application runs. Using the first machine learning architecture and using the second machine learning architecture may be based on the second requested second quality of service level. Both uses may occur during at least some overlapping portion of the time.

[0014] The method may further comprise executing a third application associated with the third requested quality of service level. The method may further comprise selecting a third machine learning architecture based on the third requested quality of service level. The method may further comprise sending a third machine learning architecture request message to the network slice manager requesting permission to use the third machine learning architecture. The method may further comprise receiving a third machine learning architecture response message from the network slice manager that does not grant permission to use the third machine learning architecture. The method may further comprise selecting a default machine learning architecture in response to receiving the third machine learning architecture response message. The method may further comprise sending a fourth machine learning architecture request message to the network slice manager requesting permission to use the default machine learning architecture. The method may further comprise receiving a fourth machine learning architecture response message from the network slice manager granting permission to use the default machine learning architecture based on the third network slice. The method may further comprise wirelessly communicating additional data for the third application using the default machine learning architecture.

[0015] The aspects described below comprise a user equipment having a radio frequency transceiver, the user equipment further comprising a first processor and a first memory system configured to perform any of the described methods.

[0016] The aspects described below comprise a method performed by a network slice manager to determine a machine learning architecture for network slicing. The method comprises creating a first network slice. The method further comprises receiving a first machine learning architecture request message from a user device requesting permission to use the first machine learning architecture. The method further comprises determining that the first machine learning architecture is associated with at least one first end-to-end machine learning architecture of the first network slice. The method further comprises sending a first machine learning architecture response message to the user device granting permission to use the first machine learning architecture.

[0017] The method may further comprise storing information relating the first network slice to the at least one available end-to-end machine learning architecture. Determining that the first machine learning architecture is associated with the at least one first end-to-end machine learning architecture may include determining that the first machine learning architecture forms a part of the at least one available end-to-end machine learning architecture.

[0018] The method may further comprise transmitting an available machine learning architectures message to the user device identifying at least one available end-to-end machine learning architecture.

[0019] The method may further comprise transmitting a machine learning architecture selection message to the user device including one or more selection rules that enable the user device to determine the first machine learning architecture.

[0020] The method may further comprise creating a second network slice. The method may further comprise: The method may include receiving a second end-to-end machine learning architecture request message from the user device. The method may further include determining that the second machine learning architecture is not associated with any second end-to-end machine learning architecture of the second network slice. The method may further include sending a second machine learning architecture response message to the user device that does not grant permission to use the second machine learning architecture.

[0021] The second machine learning architecture response message may include a third machine learning architecture associated with at least one of the second end-to-end machine learning architectures.

[0022] The aspects described below further comprise a network-slice manager including a processor and a memory system configured to perform any of the described methods.

[0023] The aspects described below comprise a system including a user device and a network slice manager.

[0024] The aspects described below further comprise a system having means for determining a machine learning architecture.

[0025] The described methods may be stored as instructions on a computer readable storage medium. The apparatus and techniques for determining a machine learning architecture for network slicing are described with reference to the following drawings, in which the same numbers are used throughout to reference like features and components, and in which: [Brief description of the drawings]

[0026] [Figure 1]FIG. 1 illustrates an example of a wireless network environment in which the determination of a machine learning architecture for network slicing may be performed. [Diagram 2] FIG. 1 illustrates an example of an end-to-end machine learning architecture for network slicing. [Diagram 3] FIG. 1 illustrates an example of properties of a machine learning architecture for network slicing. [Figure 4] A diagram showing an example of a network slice created by a network slice manager. [Diagram 5] A figure showing an example of a device diagram of a user equipment and a base station for determining a machine learning architecture for network slicing. [Figure 6] FIG. 1 illustrates an example of a device diagram of a network slice manager for determining a machine learning architecture for network slicing. [Figure 7] FIG. 1 illustrates an example block diagram of a wireless network stack model in which various aspects of determining a machine learning architecture for network slicing may be implemented. [Figure 8] A diagram showing example messaging transactions between a user equipment and a network slice manager for determining a machine learning architecture for network slicing. [Figure 9] FIG. 1 illustrates an example of a method performed by a user equipment to determine a machine learning architecture for network slicing. [Figure 10] FIG. 1 illustrates an example of a method performed by a network-slice manager to determine a machine learning architecture for network slicing. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] Detailed Description Summary Network slicing enables wireless communication networks to satisfy various quality of service (QoS) levels. In particular, network slicing enables wireless communication network infrastructure to be dynamically partitioned and shared to provide flows with different quality of service. In general, network slicing instantiates one or more parallel and independent virtual networks that satisfy different quality of service levels. A quality of service level is defined by one or more quality of service parameters, such as latency, throughput, reliability, or error rate.

[0028] Techniques and apparatus are described for determining a machine learning (ML) architecture for network slicing to achieve a requested quality of service level associated with an application. A user equipment (UE) and a network slice manager communicate with each other to determine a machine learning (ML) architecture, after which the UE employs the machine learning architecture to wirelessly communicate data for the application. In particular, the UE selects a machine learning architecture that provides a quality of service level requested by the application. The network slice manager accepts or rejects the request based on one or more available end-to-end machine learning architectures associated with a network slice that supports the quality of service level requested by the application. The UE and the network slice manager can cooperate to determine an appropriate machine learning architecture that meets the quality of service level associated with the application and forms part of an end-to-end machine learning architecture that meets the quality of service level requested by the application.

[0029] Example Environment FIG. 1 illustrates an example environment 100 including a plurality of user equipments 110 (UE 110), illustrated as UE 111, UE 112, and UE 113. Each UE 110 can communicate with a base station 120 (shown as base stations 121 and 122) via one or more wireless communication links 130 (radio links 130), illustrated as wireless links 131 and 132. For simplicity, the UEs 110 are implemented as smartphones, but may be implemented as any suitable computing or electronic device, such as a mobile communication device, a modem, a mobile phone, a gaming device, a navigation device, a media device, a laptop computer, a desktop computer, a tablet computer, a smart appliance, a vehicle-based communication system, or an Internet-of-Things (IoT) device, such as a sensor or actuator. The base station 120 (e.g., an Evolved Universal Terrestrial Radio An Access Network Node B (E-UTRAN Node B, evolved Node B, eNodeB, eNB, Next Generation Evolved Node B, ng-eNB, next generation Node B, gNode B, gNB or ng-eNB, etc.) may be implemented as a macro cell, a micro cell, a small cell, a pico cell, etc., or any combination thereof.

[0030] Base station 120 communicates with UE 110 using radio links 131 and 132, which may be implemented as any suitable type of radio link. Radio links 131 and 132 include control and data communications, such as a downlink of data and control information communicated from base station 120 to UE 110, an uplink of other data and control information communicated from UE 110 to base station 120, or both. Radio link 130 is implemented as a 3GPP standard. Any suitable communications standard, such as 3rd Generation Partnership Project Long-Term Evolution (LTE), Enhanced Long-Term Evolution (eLTE), 5G NR (Fifth-Generation New Radio), or 4G (4th-Generation) standards. The wireless links 130 may include one or more radio links (e.g., radio links) or bearers implemented using a protocol or standard, or a combination of communication protocols or standards. Multiple wireless links 130 may be aggregated in carrier aggregation to provide higher data rates to the UE 110. Multiple wireless links 130 from multiple base stations 120 may be configured for Coordinated Multipoint (CoMP) communication with the UE 110.

[0031] The base stations 120 collectively comprise a radio access network 140 (e.g., RAN, Evolved Universal Terrestrial Radio Access The RAN 140 is a 5G Core Network (5G RAN), E-UTRAN, 5G NR RAN, or NR RAN. The base stations 121 and 122 of the RAN 140 are connected to a 5G Core Network (5GC) 150. The base stations 121 and 122 connect to the 5GC 150 at 102 and 104, respectively, using an NG2 interface for control plane signaling and an NG3 interface for user plane data communication. The base stations 121 and 122 can communicate at 106 using an Xn Application Protocol (XnAP) over an Xn interface or an X2 Application Protocol (X2AP) over an X2 interface to exchange user plane data and control plane data. Although not shown, the UE 110 may connect to a public network, such as the Internet, via the 5GC 150 to interact with remote services.

[0032] The 5G NR network includes an Access and Mobility Management Function 152 (AMF 152) that provides control plane functions such as registration, authentication, authorization, and mobility management of multiple UEs 110 in the 5G NR network. The AMF 152 communicates with base stations 120 in the RAN 140 and uses the base stations 120 to communicate with the multiple UEs 110.

[0033] The 5GC 150 also includes a network slice manager 190. The network slice manager 190 may be implemented by a server of the 5GC 150. In general, the network slice manager 190 uses network slicing to provide different quality of service flows through the wireless communication network (e.g., provide different quality of service flows between at least one UE 110, at least one base station 120, and the 5GC 150).

[0034] Each quality of service flow is provided by a network slice that supports a quality of service level required by an application on the UE 110. In some cases, a network slice provides multiple quality of service flows. As described above, a quality of service level specifies one or more quality of service parameters, such as latency, throughput (e.g., bandwidth or data rate), reliability, or error rate (e.g., bit error rate). Examples of other quality of service parameters include availability, packet loss, or jitter. In addition to the quality of service level, the network slice can provide a certain security level through encryption. To provide the quality of service level, the network-slice manager 190 associates each network slice with one or more end-to-end machine learning architectures capable of satisfying the quality of service level, as further described with respect to FIG. 2.

[0035] FIG. 2 illustrates an example of an end-to-end machine learning architecture 200 for network slicing. In general, the end-to-end machine learning architecture 200 provides an interface between a UE 110, at least one base station 120, and a 5GC entity 150. The 5GC entity may include a server or an entity providing a user plane function in the 5GC. In particular, the UE 110 may implement a 5GC-based ... The machine learning architecture 210 implemented by the base station 120, the machine learning architecture 220 implemented by the entity of the 5GC 150, and the machine learning architecture 230 implemented by the entity of the 5GC 150 work together to form the end-to-end machine learning architecture 200. The machine learning interface 240 passes information between the machine learning architecture 210, the machine learning architecture 220, and the machine learning architecture 230. In this manner, the end-to-end machine learning architecture 200 functions as a single distributed machine learning architecture. Although not shown, the end-to-end machine learning architecture 200 may also include an application server in close proximity to the 5GC 150 for edge computing.

[0036] The machine learning architectures 210, 220, and 230 provide solutions to complex processing, such as complex functionality used in wireless communication systems. The machine learning architectures 210, 220, and 230 can replace conventional complex functionality in various ways, such as replacing some or all of the conventional processing blocks used in end-to-end processing of wireless communication signals, replacing individual processing chain blocks, etc. As an example, each of the machine learning architectures 210, 220, and 230 can perform one or more of the following functions: channel coding, channel decoding, modulation, demodulation, signal processing (e.g., noise cancellation), encryption, or decoding. As described in more detail with respect to FIG. 3, the machine learning architectures represent a particular structure of one or more layers and the connections between these layers. The design of these machine learning architectures can be adjusted based on the capabilities of the respective entities. These capabilities can include available memory, available power, and computational power. In general, a larger amount of available memory, a larger amount of available power, and / or a higher level of computational power allows larger and more complex machine learning architectures to be implemented.

[0037] In some cases, the UE 110, the base station 120, and / or the 5GC 150 entities may use a variety of different machine learning architectures. The UE 110 may implement, for example, machine learning architectures 211, 212, and 213. Similarly, the base station 120 may implement machine learning architectures 221, 222, and 223, and the 5GC 150 entities may implement machine learning architectures 231, 232, and 233. The design of these various machine learning architectures may differ in terms of the amount of layers, the types of layers, the order of the layers, and the connections between the layers, as will be further described with respect to FIG. 3. The differences in the design of the machine learning architectures may affect the service quality levels achievable using the end-to-end machine learning architecture 200, as will be further described with respect to FIG. 4.

[0038] The network-slice manager 190 communicates with entities of the UE 110, the base station 120, and the 5GC 150 to select an appropriate machine learning architecture. In this manner, the network-slice manager 190 can selectively activate different end-to-end machine learning architectures 200 with different combinations of the machine learning architectures 210, 220, and 230. General characteristics of the machine learning architectures 210, 230, and 230 are further described with respect to FIG.

[0039] 3 illustrates an example block diagram 300 of a machine learning architecture, such as machine learning architecture 210, 220, or 230 of FIG. 2. Machine learning architecture 210, 220, or 230 may be implemented using any combination of software, hardware, and / or firmware. In general, machine learning architecture 210, 220, or 230 may include one or more layers of structure and processing, as described further below. The machine learning architectures 210, 220, and 230 define the connections (e.g., node connections or layer connections) between these layers. The machine learning architectures 210, 220, and 230 do not define or specify parameter configurations that affect the operation or performance of the machine learning architectures 210, 220, and 230. Examples of parameter configurations include coefficients (e.g., weights or biases), kernel sizes or parameters, activation functions, and stride / pooling configurations. The parameter configurations may also specify nodes to be utilized or layers to be skipped. Changing various parameter configurations in training or adaptive learning provides the ability to adapt to changing operating conditions. Even as the parameter configurations change, the machine learning architectures 210, 220, and 230 remain the same. In other words, the machine learning architectures 210, 220, and 230 are considered to be distinct from the parameter configurations.

[0040] Each machine learning architecture 210, 220, or 230 includes at least one layer 310 (e.g., layers 311, 312, and 313). A layer 310 includes one or more nodes (e.g., neurons and / or perceptrons). Each node accepts an input and calculates an output based on the input. In calculating the output, a node may use one or more coefficients.

[0041] Within each machine learning architecture 210, 220, or 230, the amount of layers 310 may vary depending on the complexity of the function that the machine learning architecture 210, 220, or 230 implements. Also, the machine learning architecture 210, 220, or 230 may include different types of layers 310 with nodes that perform different types of calculations (e.g., perform different algorithms or activation functions). In some cases, the machine learning architecture 210, 220, or 230 implements a deep neural network with an input layer, an output layer, and at least one hidden layer. Other types of layers 310 include, for example, a convolutional layer, a pooling layer, a rectified linear unit layer, a loss layer, or a filter layer.

[0042] The machine learning architectures 210, 220, and 230 also specify one or more connections 320 between the layers 310. These connections 320 may vary among various machine learning architectures. Examples of types of connections 320 include full connections 321 or partial connections 322, which vary in terms of the percentage of nodes connected between the two layers 310. As an example, the first layer 311 is fully connected (321) to the second layer 312 such that all of the nodes in both the first layer 311 and the second layer 312 are connected together. In other words, each node in the first layer 311 is connected to at least one other node in the second layer 312. As another example, the second layer 312 is partially connected (322) to the third layer 313. In particular, a subset of the nodes in the second layer 312 is connected to a subset of the nodes in the third layer 313.

[0043] The connections 320 may also specify the direction or order in which the layers 310 are connected. The feedforward connections 323, for example, propagate information in a forward direction toward the output. As an example, the feedforward connections 323 connect the first layer 311 and the second layer 312, and the second layer 312 and the third layer 313. In contrast, the feedback connections 324 propagate information in a backward direction to the previous layer. As an example, the feedback connections 324 connect at least one node of the third layer 313 to another node of the first layer 311. Using the feedback connections 324, the machine learning architecture 210, 220, or 230 may implement a recurrent neural network, such as a long short-term memory (LSTM) neural network.

[0044] The inputs and outputs of the machine learning architecture 210, 220 or 230 may vary depending on the functionality implemented. As an example, the inputs may include a downlink signal received at the UE 110. The sampled signal may include samples of a signal, such as samples of a signal or samples of an uplink signal received at base station 120.

[0045] The output may represent a modified sample of the signal. To implement a regression model, the machine learning architecture 210, 220 or 230 may be designed to output one or more continuous values. Alternatively, to implement a classification model, the machine learning architecture 210, 220 or 230 may be designed to output one or more discrete values.

[0046] A variety of different training techniques can be used to determine the coefficients used by the machine learning architectures 210, 220, or 230. These training techniques include supervised learning, unsupervised learning, or reinforcement learning. In general, the network-slice manager 190 associates each machine learning architecture 210, 220, and 230 with one or more network slices, as further described with respect to FIG.

[0047] 4 illustrates an example of a network slice 400 created by the network slice manager 190. The network slice manager 190 designs the network slice 400 to meet quality of service levels 410 associated with one or more applications. Thus, each network slice 400 can meet different quality of service levels 410 as dictated by different applications or use cases.

[0048] Typically, each application is associated with an application type and a quality of service class identifier (QCI). By way of example, application examples Hulu®, Netflix®, or YouTube® may be associated with a video streaming application type and the quality of service identifier may include 4, 6, 8, or 9. Other application examples Skype®, Facetime®, or Google Hangouts® may be associated with a video conferencing application type and the quality of service identifier may include 2 or 7. Yet other application examples such as Minecraft® or PUBG Mobile® may be associated with a type of gaming application and the quality of service identifier may include 3 or 7.

[0049] 4, the network slice manager 190 creates a first network slice 401, a second network slice 402, and a third network slice 403. Each of the network slices 401, 402, and 403 provides at least one quality of service flow through the wireless communication network. In some circumstances, two or more of the network slices 401, 402, and 403 occur in parallel during overlapping portions of time.

[0050] The network slices 401, 402, and 403 are designed to satisfy different quality of service levels 410 that can be associated with different use cases or applications. For example, the first network slice 401 provides low latency and is associated with ultra-reliable low-latency communication (URLLC) and may be used for factory automation, autonomous driving, or remote surgery. Meanwhile, the second network slice 402 provides high throughput and is associated with enhanced mobile broadband (eMBB). As an example, the second network slice 402 handles high-definition video streaming. In contrast, the third network slice 403 is designed to handle a large number of UEs 110 to support a large amount of machine-to-machine communication. Other types of network slices 400 are designed to handle a large amount of UEs 110 to support online gaming, mobile phone services, and other applications. The video may be associated with a live broadcast, one-way streaming, or two-way streaming (eg, video conferencing).

[0051] The network-slice manager 190 determines at least one end-to-end machine learning architecture 200 (FIG. 2) that enables the network slice 400 to meet the service quality level 410. The network-slice manager 190 associates (e.g., associates or maps) the determined end-to-end machine learning architecture 200 with the network slice 400. As an example, the network-slice manager 190 associates the network slice 400 with the first end-to-end machine learning architecture 201, the second end-to-end machine learning architecture 202, and the default end-to-end machine learning architecture 203. The first end-to-end machine learning architecture 201 includes the machine learning architecture 211, the machine learning architecture 221, and the machine learning architecture 231 of FIG. 2. Although not shown, the second end-to-end machine learning architecture 202 includes, for example, the machine learning architecture 212, the machine learning architecture 222, and the machine learning architecture 232 of FIG. 2. The default end-to-end machine learning architecture 203 includes machine learning architecture 213, machine learning architecture 223, and machine learning architecture 233. Other end-to-end machine learning architectures 200 are possible, such as end-to-end machine learning architecture 200 including machine learning architecture 211, machine learning architecture 222, and machine learning architecture 233.

[0052] Although the end-to-end machine learning architectures 201, 202, and 203 may differ in terms of performance, the end-to-end machine learning architectures 201, 202, and 203 at least enable the network slice 400 to meet the quality of service level 410. Consider an example where the quality of service level 410 specifies a high level of reliability. The default end-to-end machine learning architecture 203 may provide an entry-level level of performance, such as a medium level of latency, a medium level of throughput, and a high level of reliability. The performance of the end-to-end machine learning architectures 201 and 202 may be similar to or better than the default end-to-end machine learning architecture 203. For example, the first end-to-end machine learning architecture 201 provides a low level of latency instead of a medium level of latency, and also provides a medium level of throughput and a high level of reliability. In contrast, the second end-to-end machine learning architecture 202 provides a high level of throughput instead of a medium level of throughput, and also provides a medium level of latency and a high level of reliability.

[0053] In some implementations, the network-slice manager 190 determines the relationship between different offline network slices 400 and different end-to-end machine learning architectures 200. The network-slice manager 190 stores information describing this relationship for future reference. As an example, the network-slice manager 190 may store one or more lists or tables that map each network slice 400 to one or more end-to-end machine learning architectures 200. Relatedly, this relationship mapping also extends to the individual machine learning architectures 210, 220, and 230 that form the end-to-end machine learning architecture 200.

[0054] By associating the network slice 400 with two or more end-to-end machine learning architectures 200, the network slice 400 can be implemented for various entities having different capabilities. These entities include different UEs 110, different base stations 120, or different entities of the 5GC 150. For example, a first UE 111 with limited computational capabilities may be implemented with the machine learning architecture 211. The first end-to-end machine learning architecture 201 can operate with the end-to-end machine learning architecture 212 but cannot operate with the machine learning architecture 212. Therefore, the first end-to-end machine learning architecture 201 can be activated by the network slice manager 190 instead of the end-to-end machine learning architecture 202 to satisfy the quality of service level 410 associated with the network slice 400. In another example, the second UE 112 with limited available power can operate with the machine learning architecture 212 but cannot operate with the machine learning architecture 211. Therefore, the second end-to-end machine learning architecture 202 can be activated by the network slice manager 190 instead of the end-to-end machine learning architecture 201 to similarly satisfy the quality of service level 410 associated with the network slice 400. The selection of the end-to-end machine learning architecture 200, and more specifically the machine learning architecture 210, is further described with respect to FIG. 8.

[0055] Device Examples FIG. 5 illustrates an example of an apparatus diagram 500 of a UE 110 and a base station 120. The UE 110 and the base station 120 may include additional functionality and interfaces that are omitted from FIG. 5 for clarity. The UE 110 includes an antenna 502, a radio frequency (RF) front end 504 (RF front end 504), an LTE transceiver 506, and a 5G NR transceiver 508 for communicating with a base station 120 in the 5G RAN 141 and / or the E-UTRAN 142. The RF front end 504 of the UE 110 may couple or connect the LTE transceiver 506 and the 5G NR transceiver 508 to the antenna 502 to facilitate various types of wireless communication. The antenna 502 of the UE 110 may include an array of multiple antennas that are configured similarly or differently from one another. The antenna 502 and the RF front end 504 may be configured according to 3GPP The antenna 502, RF front end 504, LTE transceiver 506, and / or 5G NR transceiver 508 may be tuned and / or tunable to one or more frequency bands defined by the LTE and 5G NR communications standards and implemented by the LTE transceiver 506 and / or the 5G NR transceiver 508. Additionally, the antenna 502, RF front end 504, LTE transceiver 506, and / or the 5G NR transceiver 508 may be configured to support beamforming (e.g., analog or digital), duplex operation (e.g., full duplex or half duplex operation), or in-phase and quadrature (I / Q) operation (e.g., I / Q modulation or demodulation operation) for transmitting and receiving communications with the base station 120. By way of example and not limitation, the antenna 502 and RF front end 504 may be implemented for operation in sub-gigahertz bands, sub-6 GHZ bands, and / or frequency bands having frequencies between about 6 and 500 GHz defined by the 3GPP LTE and 5G NR communications standards. In some cases, the antenna 502 and RF front end 504 may be implemented for operation in the ultra-high frequency spectrum (eg, for frequencies between about 54 and 500 GHz).

[0056] The UE 110 also includes a processor(s) 510 and a computer readable storage medium 512 (CRM 512). The processor 510 may be a single-core processor or a multi-core processor constructed of various materials such as silicon, polysilicon, high-dielectric and copper. The computer readable storage medium described herein excludes propagating signals. The CRM 512 may include any suitable memory or storage device, such as random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), non-volatile RAM (NVRAM), read only memory (ROM), or flash memory, that can be used to store device data 514 of the UE 110. The device data 514 includes user data, multimedia data, beamforming codebooks, applications, and / or an operating system of the UE 110 that can be executed by the processor(s) 510 to enable user plane communications, control plane signaling, and user interaction with the UE 110. It is Noh.

[0057] The CRM 512 also includes or stores instructions associated therewith a machine learning manager 516. Alternatively or additionally, the machine learning manager 516 may be implemented, in whole or in part, as hardware logic or circuitry integrated with or separate from other components of the UE 110. In at least some aspects, the machine learning manager 516 configures the LTE transceiver 506 or the 5G NR transceiver 508 for communication with the base station 120. In this manner, the machine learning manager 516 may send messages to or receive messages from the network-slice manager 190 through the base station 120, as further described with respect to FIG.

[0058] The CRM 512 further includes at least one machine learning architecture 210 (FIG. 2), which may be implemented based on the capabilities of the UE 110. The machine learning manager 516 selects the machine learning architecture 210 based on one or more quality of service levels associated with an application, as further described with respect to FIG. 8. The machine learning manager 516 may also initiate a learning procedure to determine coefficients that are applied to enable wireless communication using the machine learning architecture 210.

[0059] The equipment diagram of the base station 120 shown in FIG. 5 includes a single network node (e.g., gNode B). The functions of the base station 120 may be distributed among multiple network nodes or devices, and may be distributed in any manner suitable for performing the functions described herein. The base station 120 includes an antenna 552 for communicating with the UE 110, a radio frequency (RF) front end 554 (RF front end 554), one or more LTE transceivers 556, and / or one or more 5G NR transceivers 558. The RF front end 554 of the base station 120 may couple or connect the LTE transceiver 556 and the 5G NR transceiver 558 to the antenna 552 to facilitate various types of wireless communication. The antenna 552 of the base station 120 may include an array of multiple antennas configured similarly or differently from each other. The antenna 552 and the RF front end 554 may be tuned or tunable to one or more frequency bands defined by the 3GPP LTE and 5G NR communications standards and implemented by the LTE transceiver 556, and / or the 5G NR transceiver 558. Additionally, the antenna 552, the RF front end 554, the LTE transceiver 556, and / or the 5G NR transceiver 558 may be configured to support beamforming, such as Massive-MIMO, for transmitting and receiving communications with the UE 110.

[0060] The base station 120 also includes a processor(s) 560 and a computer readable storage medium 562 (CRM 562). The processor 560 may be a single-core processor or a multi-core processor constructed of various materials such as silicon, polysilicon, high-k dielectrics, and copper. The CRM 562 may include any suitable memory or storage device, such as random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), non-volatile RAM (NVRAM), read-only memory (ROM), or flash memory, that can be used to store device data 564 of the base station 120. The device data 564 includes network scheduling data, radio resource management data, beamforming codebooks, applications, and / or an operating system of the base station 120, which are executable by the processor(s) 560 to enable communication with the UE 110.

[0061] The CRM 562 also includes or stores instructions relating to a machine learning (ML) manager 566. Alternatively or additionally, the machine learning manager 566 may be, in whole or in part, may be implemented as hardware logic or circuitry integrated with or separate from other components of the base station 120. In at least some aspects, the machine learning manager 566 configures the LTE transceiver 556 and the 5G NR transceiver 558 for communication with the UE 110. In this manner, the machine learning manager 566 may forward messages from the UE 110 to the network-slice manager 190 or from the network-slice manager 190 to the UE 110, as further described with respect to FIG.

[0062] The CRM 562 further includes at least one machine learning architecture 220 (of FIG. 2), which may be implemented based on the capabilities of the base station 120. In some implementations, the machine learning manager 566 selects the machine learning architecture 220 based on available end-to-end machine learning architectures 200 associated with the network slice 400, as further described with respect to FIG. 8. The machine learning manager 566 may also initiate a training procedure or provide coefficients to the machine learning architecture 220.

[0063] The base station 120 includes an inter-base station interface 568, such as an Xn and / or X2 interface, that configures a base station manager (not shown) to exchange user plane data and control plane data between different base stations 120 to manage the base station's 120 communications with the UE 110. The base station 120 also includes a core network interface 570 that configures the machine learning manager 566 to exchange information with core network functions and entities, such as the network slice manager 190. In this manner, the machine learning manager 566 can provide information to or obtain information from the network slice manager 190. The network slice manager, the machine learning manager 516 of the UE 110, and the machine learning manager 566 of the base station 120 together can at least partially determine the machine learning architecture for network slicing as described herein. The network slice manager 190 is further described with respect to FIG. 3.

[0064] 6 illustrates an example apparatus diagram 600 of a network-slice manager 190 for determining a machine learning architecture for network slicing. In the illustrated configuration, the network-slice manager 190 includes a processor 602 and a computer-readable storage medium (CRM) 604. Like the CRM 512 and the CRM 562, the CRM 604 may include any suitable memory or storage device, such as random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), non-volatile RAM (NVRAM), read-only memory (ROM), or flash memory.

[0065] CRM 604 stores one or more available machine learning architectures 606 associated with at least one network slice 400. The available machine learning architectures 606 specify end-to-end machine learning architecture 200 and / or individual machine learning architectures (e.g., machine learning architecture 210, machine learning architecture 220, and / or machine learning architecture 230) that satisfy a quality of service level 410 of the network slice 400.

[0066] The CRM 604 also stores machine learning architecture selection rules 608. The machine learning architecture selection rules 608 provide instructions that enable the UE 110 to select an appropriate machine learning architecture 210 based on an application that the UE 110 executes. In some cases, the machine learning architecture selection rules 608 enable the UE 110 to derive a requested quality of service level from a given performance requirement of the application. The machine learning architecture selection rules 608 may also include instructions that enable an entity at the UE 110, the base station 120, or the 5GC 150 to select an appropriate machine learning architecture based on the available machine learning architectures 606.

[0067] The network slice manager 190 may also include a base station interface 610 that configures the network slice manager 190 to exchange information with the base station 120. In this manner, the network slice manager 190 can provide information directly to the base station 120 or obtain information from the base station 120.

[0068] User Plane and Control Plane Signaling FIG. 7 illustrates an example of a block diagram 700 of a wireless network stack model 700 (stack 700). The stack 700 characterizes a communication system for the example environment 100 in which various aspects determining a machine learning architecture for network slicing may be implemented. The stack 700 includes a user plane 702 and a control plane 704. The upper layers of the user plane 702 and the control plane 704 share common lower layers in the stack 700. A wireless device, such as the UE 110 or the base station 120, implements each layer as an entity for communicating with another device using a protocol defined for the layer. For example, the UE 110 uses a Packet Data Convergence Protocol (PDCP) entity to communicate to a peer PDCP entity in the base station 120 using PDCP.

[0069] The shared lower layers include a physical (PHY) layer 706, a media access control (MAC) layer 708, a radio link control (RLC) layer 710, and a PDCP layer 712. The PHY layer 706 provides the hardware specifications for devices to communicate with each other. As such, the PHY layer 706 helps establish how devices connect to each other, manage how communication resources are shared between devices, and so on.

[0070] The MAC layer 708 specifies how data is transferred between devices. In general, the MAC layer 708 provides how transmitted data packets are encoded and decoded into bits as part of a transmission protocol.

[0071] The RLC layer 710 provides data transfer services to higher layers in the stack 700. In general, the RLC layer 710 provides error correction, packet segmentation and reassembly, and management of data transfer in various modes, such as acknowledged, unacknowledged, or transparent modes.

[0072] The PDCP layer 712 provides data transport services to the upper layers of the stack 700. In general, the PDCP layer 712 provides transport, header compression, encryption, and integrity protection of user plane 702 and control plane 704 data.

[0073] Above the PDCP layer 712, the stack splits into a user plane 702 and a control plane 704. The user plane 702 layers include an optional Service Data Adaptation Protocol (SDAP) layer 714, an Internet Protocol (IP) layer 716, a Transmission Control Protocol / User Datagram Protocol (TCP / UDP) layer 718, and an application layer 720 that transfers data using the radio links 106. The optional SDAP layer 714 is present in 5G NR networks. The SDAP layer 714 maps quality of service flows per data radio bearer and provides uplink and downlink quality of service for each packet data session. The IP layer 716 specifies how data from the application layer 720 is forwarded to a destination node. The TCP / UDP layer 718 is used to verify that data packets intended for forwarding to a destination node have reached the destination node, using either TCP or UDP for data forwarding by the application layer 720. In some implementations, the user plane 702 may also include a data services layer (not shown) that provides data forwarding services for forwarding application data, such as IP packets containing web browsing content, video content, image content, audio content, or social media content.

[0074] The control plane 704 includes a Radio Resource Control (RRC) layer 724 and a Non-Access Stratum (NAS) layer 726. The RRC layer 724 establishes and releases connections and radio bearers, broadcasts system information, or performs power control. The RRC layer 724 also controls the resource control state of the UE 110 and causes the UE 110 to perform operations according to the resource control state. Examples of the resource control state include a connected state (e.g., an RRC connected state) or a disconnected state, such as an inactive state (e.g., an RRC inactive state) or an idle state (e.g., an RRC idle state). Generally, when the UE 110 is in a connected state, the connection with the base station 120 is active. When in an inactive state, the connection with the base station 120 is interrupted. When the UE 110 is in an idle state, the connection with the base station 120 is released. Generally, the RRC layer 724 supports 7GPP access but not non-3GPP access (e.g., WLAN communication).

[0075] The NAS layer 726 provides support for mobility management (e.g., using a 5th-Generation Mobility Management (5GMM) layer 728) and packet data bearer contexts (e.g., using a 5th-Generation Session Management (5GSM) layer 730) between the UE 110 and entities or functions in the core network, such as the AMF 152 of the 5GC 150. The NAS layer 726 supports both 3GPP and non-3GPP access.

[0076] In the UE 110, each layer in both the user plane 702 and control plane 704 of the stack 700 interacts with corresponding peer layers or entities in the base station 120, core network entities or functions, and / or remote services to support and control the operation of user applications of the UE 110 in the RAN 140.

[0077] Deciding on a Machine Learning Architecture for Network Slicing 8 illustrates an example of messaging transactions 800 between the UE 110 and the network slice manager 190 to determine the machine learning architecture 210 for network slicing. Although not shown, the base station 120 may facilitate forwarding of messages from the network slice manager 190 to the UE 110 or from the UE 110 to the network slice manager 190.

[0078] At 805, the UE 110 executes an application, such as any of the applications described above. Generally, the application depends on the ability of the UE 110 to wirelessly communicate data. The application is associated with a requested quality of service level 806.

[0079] At 810, the network-slice manager 190 creates a network slice 400 that supports a quality of service level 410. The quality of service level 410 meets a requested quality of service level 806 associated with the application, as described above with respect to FIG. 4. In some cases, the network-slice manager 190 receives an indication of the requested quality of service level 806, such as an index (5QI), from the UE 110. Based on the indication, the network-slice manager 190 can select an appropriate network slice 400.

[0080] At 815, the network-slice manager 190 sends an available machine learning architectures message 816 to the UE 110. The available machine learning architectures message 816 identifies available machine learning architectures 606 (e.g., available machine learning architectures 210) that form part of the available end-to-end machine learning architectures 200 associated with the network slice 400, as described above with respect to FIG. 6. By sharing this information with the UE 110, the network-slice manager 190 and the UE 110 can efficiently determine an appropriate machine learning architecture 210 that satisfies the requested quality of service level 806 of the application and forms part of the end-to-end machine learning architecture 200 (e.g., machine learning architectures 210 of FIGS. 2 and 3) that satisfies the requested quality of service level 806. The available machine learning architectures message 816 can be implemented as a NAS message.

[0081] In the above example, the available machine learning architectures message 816 is sent after the network slice 400 is created. Alternatively, the network slice manager 190 sends the available machine learning architectures message 816 before creating the network slice 400. In this case, the available machine learning architectures message 816 identifies the available machine learning architectures 606 associated with multiple network slices 400 (e.g., network slices 401, 402, and 403 of FIG. 4).

[0082] At 820, the network-slice manager 190 sends a machine learning architecture selection message 822 to the UE 110. The machine learning architecture selection message 822 includes the machine learning architecture selection rules 608 (of FIG. 6) that enable the UE 110 to select a machine learning architecture based on the application (e.g., select one of the machine learning architectures 210 of FIG. 2). Like the available machine learning architectures message 816, the machine learning architecture selection message 822 may also be implemented as a NAS message. In general, each of steps 815 and 820 may occur at any time prior to 825 or in response to the network-slice manager 190 rejecting the UE 110's request to use a particular machine learning architecture 210, as described further below.

[0083] At 825, the UE 110 selects a machine learning architecture 210 based on the requested quality of service level 806. The UE 110 may select the machine learning architecture 210 from a set of possible machine learning architectures that the UE 110 can implement based on its current capabilities. In some cases, the UE 110 applies the machine learning architecture selection rules 608 to derive the requested quality of service level 806 from the performance requirements of the application and select an appropriate machine learning architecture 210. In other cases, the UE 110 applies the machine learning architecture selection rules 608 to select the machine learning architecture 210 from the available machine learning architectures 606.

[0084] At 830, the UE 110 may select the selected machine learning architecture 2 (of FIGS. 2 and 3). 10 to the network-slice manager 190 requesting permission to use the machine learning architecture request message 832.

[0085] At 835, the network-slice manager 190 determines whether to accept or reject the request. In particular, the network-slice manager 190 determines whether the selected machine learning architecture 210 is associated with one of the end-to-end machine learning architectures 200 of the network slice 400. For example, the network-slice manager 190 determines whether one of the available end-to-end machine learning architectures 200 associated with the network slice 400 can be formed with the selected machine learning architecture 210.

[0086] At 840, the network-slice manager 190 sends a machine learning architecture response message 842 to the UE 110 determining whether to grant or not grant permission to use the selected machine learning architecture 210. If the network-slice manager 190 determines that the selected machine learning architecture 210 is not associated with any of the available end-to-end machine learning architectures 200 associated with the network slice 400, the machine learning architecture response message 842 does not grant the UE 110 permission to use the selected machine learning architecture 210. Sometimes, the network-slice manager 190 sends an available machine learning architectures message 816 and / or a machine learning architecture selection message 822 in response to rejecting the UE 110's request (not shown).

[0087] If permission is not granted, the UE 110 may select a machine learning architecture associated with the default end-to-end machine learning architecture 203 (e.g., machine learning architecture 213 of FIG. 2) and wirelessly communicate data for the application using the machine learning architecture 213. Alternatively, the process may be repeated at 825 such that the UE 110 selects a second machine learning architecture 210.

[0088] Alternatively, if the network-slice manager 190 determines that the selected machine learning architecture 210 can be used to form one of the end-to-end machine learning architectures 200 associated with the network slice 400, the machine learning architecture response message 842 grants the UE 110 permission to use the selected machine learning architecture 210.

[0089] If permission is granted, the UE 110 wirelessly communicates data for the application using the selected machine learning architecture 210. For example, the UE 110 wirelessly communicates with the base station 120 using the selected machine learning architecture 210.

[0090] After granting the UE 110 permission to use the selected machine learning architecture 210, the network-slice manager 190 may select an available end-to-end machine learning architecture 200 that includes the selected machine learning architecture 210. Based on the selected end-to-end machine learning architecture 200, the network-slice manager 190 may instruct the base station 120 and / or 5GC 150 entities to use the machine learning architecture 220 and the machine learning architecture 230, respectively, associated with the selected end-to-end machine learning architecture 200.

[0091] In some circumstances, the UE 110 may run multiple applications in parallel, and the network slice manager 190 may create multiple network slices 400 to serve one or more UEs 110. The process can be repeated at 805 such that another UE 110) runs a second application and the network slice manager 190 can create a second network slice 400 or provide another quality of service flow using the first network slice 400.

[0092] When the UE 110 connects to a different base station 120 or a different entity of the 5GC 150, the available end-to-end machine learning architectures 200 that satisfy the quality of service level 410 may change. This change may occur if the current base station 120 or the current entity has different capabilities compared to a previous base station or previous entity of the 5GC 150. Due to the different capabilities, the first set of machine learning architectures 220 that the current base station 120 can implement may be different from the second set of machine learning architectures 220 that the previous base station can implement. Similarly, the first set of machine learning architectures 230 that the current entity can implement may be different from the second set of machine learning architectures 230 that the previous entity can implement. In this way, the available machine learning architectures 606 may depend on the end-to-end machine learning architectures 200 that are realizable using the machine learning architectures 220 supported by the current base station 120 and the machine learning architectures 230 supported by the current entity.

[0093] Changes to the available machine learning architectures 606 may occur in response to a variety of different circumstances. For example, these changes may occur when the UE 110 moves to a different location or tracking area. In another example, these changes occur when the UE 110 roams between a home network (e.g., a network in which the UE 110 is registered) and a visited network (e.g., a network in which the UE 110 is not registered). The network-slice manager 190 may send a second available machine learning architectures message 816 to inform the UE 110 of the latest available machine learning architectures 606 associated with the network slice 400.

[0094] Consider an example where the base station 121 is capable of implementing the machine learning architecture 221. In this example, the UE 110 switches from communication with the base station 121 to communication with the base station 122. However, the capabilities of the base station 122 do not allow the base station 122 to implement the machine learning architecture 221. Therefore, the end-to-end machine learning architecture 201 (of FIG. 2) is no longer a viable option for meeting the quality of service level 410 of the network slice 400. Relatedly, the machine learning architecture 211 is also no longer available. The network-slice manager 190 sends an available machine learning architecture message 816 to inform the UE 110 of an update to the available machine learning architecture 606. Based on this update, the UE 110 selects the machine learning architecture 212 associated with the end-to-end machine learning architecture 202. In this case, the machine learning architecture 212 allows the implementation of the end-to-end machine learning architecture 202 to meet the requested quality of service level 806.

[0095] In some cases, the selection process at 825 occurs in response to the UE 110 performing a tracking area update (TAU) procedure. The TAU procedure informs the wireless network which tracking area corresponds to the current location of the UE 110.

[0096] Although described with respect to the UE 110, similar operations may occur between the network slice manager 190 and other entities in the wireless communication network, such as entities in the base station 120 or the 5GC 150. The network-slice manager 190 determines an appropriate machine learning architecture based on the entity's capabilities, the requested quality of service level 806, and the available end-to-end machine learning architectures 200.

[0097] Example of how to FIG. 9 illustrates an example of a method 900 performed by a UE 110 to determine a machine learning architecture for network slicing. The method 900 is illustrated as a series of operations (or acts) performed, but the operations are not necessarily limited to the order or combination shown. Furthermore, any of one or more of the operations may be repeated, combined, rearranged, skipped, or linked to provide a wide range of additional and / or alternative methods. In some of the following discussion, reference may be made to the environment 100 of FIG. 1 and entities detailed in FIG. 2, but such references are for illustrative purposes only. The present technology is not limited to execution by one entity or multiple entities operating on one device.

[0098] At 902, the UE executes a first application associated with the requested quality of service level. For example, the UE 110 executes the first application associated with the requested quality of service level 806, as described above at 805 in FIG.

[0099] At 904, the UE selects a first machine learning architecture based on the requested quality of service level. For example, the UE 110 selects the machine learning architecture 210 based on the requested quality of service level 806. The machine learning architecture 210 may be one of the machine learning architectures 210 that the UE 110 is capable of implementing, such as one of the machine learning architectures 211, 212, or 213 of FIG.

[0100] At 906, the UE sends a first machine learning architecture request message to a network-slice manager of the wireless network requesting permission to use the first machine learning architecture. For example, the UE 110 sends a machine learning architecture request message 832 to the network-slice manager 190 requesting permission to use the machine learning architecture 210.

[0101] In response to receiving the machine learning architecture request message 832, the network-slice manager 190 determines whether the machine learning architecture 210 requested by the UE 110 is associated with at least one end-to-end machine learning architecture 200 of the network slice 400, such as one of the end-to-end machine learning architectures 201, 202, or 203 of FIG. 4. Consider an example in which the end-to-end machine learning architectures 201, 202, and 203 include machine learning architectures 211, 212, and 213, respectively. In this case, the network-slice manager 190 grants the UE 110 permission to use the machine learning architecture 210 if the machine learning architecture 210 is one of the machine learning architectures 211, 212, or 213 associated with the network slice 400. However, if the machine learning architecture 210 is not associated with any of the machine learning architectures 211, 212, and 213, the network-slice manager 190 does not grant the UE 110 permission to use the machine learning architecture 210. The network slice manager 190 sends a machine learning architecture response message 842 to the UE 110 conveying whether permission has been granted or not.

[0102] At 908, the UE receives a first machine learning architecture response message from the network slice manager granting permission to use the first machine learning architecture based on the first network slice. For example, the UE 110 may receive a first machine learning architecture response message from the network slice manager at 840 of FIG. A machine learning architecture response message 842 as shown is received from the network slice manager 190. In this example, the machine learning architecture response message 842 grants the UE 110 permission to use the machine learning architecture 210 based on the network slice 400 (e.g., based on the machine learning architecture 210 forming part of one or more of the end-to-end machine learning architectures 200 associated with the network slice 400). In other examples, the machine learning architecture response message 842 does not grant the UE 110 permission to use the machine learning architecture 210.

[0103] At 910, the UE wirelessly communicates data for a first application using a first machine learning architecture. For example, the UE 110 wirelessly communicates data for the first application using the machine learning architecture 210. The machine learning architecture may perform, for example, channel coding, channel decoding, modulation, demodulation, encryption, or decoding.

[0104] FIG. 10 illustrates an example of a method 1000 performed by the network slice manager 190 to determine a machine learning architecture for network slicing. The method 1000 is illustrated as a series of operations (or acts) performed, but is not necessarily limited to the order or combination of operations shown. Furthermore, any of one or more operations may be repeated, combined, rearranged, skipped, or linked to provide a wide range of additional and / or alternative methods. In some of the following discussion, reference may be made to the environment 100 of FIG. 1 and entities detailed in FIG. 3, but these references are made for purposes of example only. The present technology is not limited to execution by one entity or multiple entities operating on one device.

[0105] At 1002, the network slice manager creates a first network slice. For example, the network slice manager 190 creates the network slice 400 at 810 in FIG.

[0106] At 1004, the network-slice manager receives a first machine learning architecture request message from the UE requesting permission to use the first machine learning architecture. For example, the network-slice manager 190 receives a machine learning architecture request message 832 from the UE 110, as shown at 830 in FIG. 8. The machine learning architecture request message 832 requests permission for the UE 110 to use the machine learning architecture 210.

[0107] At 1006, the network-slice manager determines that the first machine learning architecture is associated with at least one first end-to-end machine learning architecture of the first network slice. For example, the network-slice manager 190 determines that the machine learning architecture 210 is associated with at least one of the end-to-end machine learning architectures 200 of the network slice 400, as described above at 835 of FIG.

[0108] At 1008, the network-slice manager, based on the determination, sends a first machine learning architecture response message to the UE granting permission to use the first machine learning architecture. For example, the network-slice manager 190 sends a machine learning architecture response message 842 to the UE 110 at 840 of FIG. 8. In this example, the machine learning architecture response message 842 grants the UE 110 permission to use the machine learning architecture 210. The network-slice manager 190 The base station 120 or a 5GC entity may also be instructed to use a machine learning architecture associated with the end-to-end machine learning architecture 200, which includes the machine learning architecture 210. In another example, the machine learning architecture response message 842 does not grant the UE 110 permission to use the machine learning architecture 210.

[0109] The UE 110 and the network slice manager 190, working together, are able to determine an appropriate machine learning architecture 210 that satisfies the requested quality of service level 806 associated with the application and forms part of the end-to-end machine learning architecture 200 that satisfies the requested quality of service level 806.

[0110] conclusion Although the techniques for determining a machine learning architecture for network slicing have been described in feature and / or method specific language, it should be understood that the subject matter of the appended claims is not necessarily limited to the particular features or methods described. Rather, the particular features and methods are disclosed as example implementations of determining a machine learning architecture for network slicing.

[0111] Some examples are described below. Example 1: A method performed by a user device, the method comprising: executing a first application associated with a first requested quality of service level; selecting a first machine learning architecture based on a first requested quality of service level; sending a first machine learning architecture request message to a network slice manager of the wireless network requesting permission to use the first machine learning architecture; receiving a first machine learning architecture response message from the network-slice manager granting permission to use the first machine learning architecture based on the first network slice; and wirelessly communicating data for a first application using a first machine learning architecture.

[0112] Example 2: Selecting a first machine learning architecture receiving an available machine learning architectures message from a network-slice manager identifying available machine learning architectures associated with a first network slice; and selecting a first machine learning architecture from among the available machine learning architectures.

[0113] Example 3: Selecting a first machine learning architecture receiving a machine learning architecture selection message from a network-slice manager, the machine learning architecture selection message including one or more selection rules; and executing one or more selection rules to determine the first machine learning architecture.

[0114] Example 4: Selecting a second machine learning architecture; sending a second machine learning architecture request message to the network slice manager requesting permission to use the second machine learning architecture; The method of any one of the preceding examples, further comprising receiving a second machine learning architecture response message from the network slice manager granting permission to use the second machine learning architecture based on the second network slice.

[0115] Example 5: A method for wirelessly communicating other data for the first application using a second machine learning architecture, Selecting the second machine learning architecture is based on the first requested quality of service level; Choosing a second machine learning architecture the user equipment moves to a geographic location associated with a different base station or a different tracking area; or 5. The method of example 4, wherein the user device is responsive to at least one of roaming between a home network and a visited network.

[0116] Example 6: Executing a second application associated with a second requested quality of service level; selecting a second machine learning architecture based on a second requested quality of service level; sending a second machine learning architecture request message to the network slice manager requesting permission to use the second machine learning architecture; The method of any one of Examples 1 to 3, further comprising receiving a second machine learning architecture response message from the network slice manager granting permission to use the second machine learning architecture based on the first network slice or the second network slice.

[0117] Example 7: The method further comprising: in response to receiving the second machine learning architecture response message, wirelessly communicating other data for the second application using the second machine learning architecture; Selecting the second machine learning architecture is based on a second requested quality of service level; executing the second application occurs during at least an overlapping portion of the time that the first application is executed; The method of Example 6, wherein the use of the first machine learning architecture and the use of the second machine learning architecture both occur during at least an overlapping portion of time.

[0118] Example 8: The method further comprises performing a tracking area update procedure. The method of any one of the preceding examples, wherein selecting the second machine learning architecture occurs in response to performing a tracking update procedure.

[0119] Example 9: Executing a third application associated with a third requested quality of service level; selecting a third machine learning architecture based on the third requested quality of service level; and sending a third machine learning architecture request message to the network slice manager requesting permission to use the third machine learning architecture; receiving a third machine learning architecture response message from the network slice manager that does not grant permission to use the third machine learning architecture; selecting a default machine learning architecture in response to receiving the third machine learning architecture response message; and sending a fourth machine learning architecture request message to the network-slice manager requesting permission to use a default machine learning architecture; receiving a fourth machine learning architecture response message from the network-slice manager granting permission to use a default machine learning architecture based on the third network slice; The method of any one of the preceding examples, further comprising wirelessly communicating additional data for a third application using the default machine learning architecture.

[0120] Example 10: At least one first quality of service level is Latency level, Throughput level, or The method of any one of the preceding examples comprising at least one of a confidence level.

[0121] Example 11: The first machine learning architecture is Many layers, One or more types of layers, One or more types of layer order, or The method of any one of the preceding examples, further comprising specifying at least one of: one or more types of connections between layers.

[0122] Example 12: A radio frequency transceiver; A user device comprising a first processor and a first memory system configured to execute the method according to any one of Examples 1 to 11.

[0123] Example 13: A method performed by a network slice manager, comprising: Creating a first network slice; receiving a first machine learning architecture request message from a user device requesting permission to use the first machine learning architecture; determining that a first machine learning architecture is associated with at least one first end-to-end machine learning architecture of a first network slice; A method comprising: sending a first machine learning architecture response message to a user device granting permission to use the first machine learning architecture.

[0124] Example 14: The method further comprises storing information associating the first network slice with at least one available end-to-end machine learning architecture; 14. The method of example 13, wherein determining that the first machine learning architecture is associated with at least one first end-to-end machine learning architecture includes determining that the first machine learning architecture forms a part of at least one available end-to-end machine learning architecture.

[0125] Example 15: The method of Example 14, wherein the at least one available machine learning architecture includes at least one end-to-end machine learning architecture, and the at least one end-to-end machine learning architecture includes a first user device machine learning architecture.

[0126] Example 16: The method of Example 14 or Example 15, further comprising sending an available machine learning architectures message to a user device identifying at least one available end-to-end machine learning architecture.

[0127] Example 17: The method of example 16, wherein the available machine learning architecture messages include non-access stratum messages.

[0128] Example 18: A method as described in any one of Examples 13 to 17, further comprising sending to the user device a machine learning architecture selection message including one or more selection rules that enable the user device to determine a first machine learning architecture.

[0129] Example 19: Creating a second network slice; receiving a second machine learning architecture request message from a user device requesting permission to use the second machine learning architecture; determining that the second machine learning architecture is not associated with any second end-to-end machine learning architecture of the second network slice; The method of any one of Examples 13-18, further comprising: sending a second machine learning architecture response message to the user device that does not grant permission to use the second machine learning architecture.

[0130] Example 20: The method of example 19, wherein the second machine learning architecture response message includes a third machine learning architecture associated with at least one of the second end-to-end machine learning architectures.

[0131] Example 21: A network-slice manager comprising a processor and a memory system configured to execute the method according to any one of Examples 12-20.

[0132] Example 22: A user device according to example 12; A system comprising the network slice manager of Example 21, wherein the processor and memory system of the network slice manager includes a second processor and a second memory system.

Claims

1. 1. A method performed by a user device, comprising: sending a first machine learning architecture request message to a network slice manager of the wireless network requesting permission to use the first machine learning architecture; receiving from the network-slice manager a first machine learning architecture response message granting permission to use the first machine learning architecture based on a first network slice, the first machine learning architecture forming part of at least one first end-to-end machine learning architecture associated with the first network slice, the at least one first end-to-end machine learning architecture being configured to process wireless communication signals and being a distributed machine learning architecture formed by the first machine learning architecture implemented by the user equipment, a machine learning architecture implemented by a base station, and a machine learning architecture implemented by an entity of a core network, the method further comprising:

4. A method comprising wirelessly communicating data using the first machine learning architecture.

2. The method further comprising selecting the first machine learning architecture based on a first requested quality of service level; Selecting the first machine learning architecture comprises: receiving an available machine learning architectures message from the network slice manager identifying available machine learning architectures associated with the first network slice; and selecting the first machine learning architecture from the available machine learning architectures.

3. Selecting the first machine learning architecture comprises: receiving a machine learning architecture selection message from the network-slice manager, the machine learning architecture selection message including one or more selection rules; and executing the one or more selection rules to determine the first machine learning architecture.

4. selecting a second machine learning architecture; and sending a second machine learning architecture request message to the network-slice manager requesting permission to use the second machine learning architecture; 3. The method of claim 2, further comprising receiving a second machine learning architecture response message from the network-slice manager granting permission to use the second machine learning architecture based on a second network slice.

5. and wirelessly communicating other data for a first application using the second machine learning architecture. selecting the second machine learning architecture based on the first requested quality of service level; Selecting the second machine learning architecture comprises: the user equipment moves to a geographic location associated with a different base station or a different tracking area; or The method of claim 4 , occurring in response to at least one of the user device roaming between a home network and a visited network.

6. executing a second application associated with a second requested quality of service level; selecting a second machine learning architecture based on the second requested quality of service level; and sending a second machine learning architecture request message to the network-slice manager requesting permission to use the second machine learning architecture; The method of any one of claims 1 to 3, further comprising receiving a second machine learning architecture response message from the network slice manager granting permission to use the second machine learning architecture based on the first network slice or the second network slice.

7. and in response to receiving the second machine learning architecture response message, wirelessly communicating other data for the second application using the second machine learning architecture. selecting the second machine learning architecture based on the second requested quality of service level; executing the second application occurs during at least an overlapping portion of the time that the first application is executed; 7. The method of claim 6, wherein the use of the first machine learning architecture and the use of the second machine learning architecture both occur during at least an overlapping portion of the time.

8. executing a third application associated with a third requested quality of service level; selecting a third machine learning architecture based on the third requested quality of service level; and sending a third machine learning architecture request message to the network-slice manager requesting permission to use the third machine learning architecture; receiving a third machine learning architecture response message from the network-slice manager not granting permission to use the third machine learning architecture; selecting a default machine learning architecture in response to receiving the third machine learning architecture response message; and sending a fourth machine learning architecture request message to the network-slice manager requesting permission to use the default machine learning architecture; receiving a fourth machine learning architecture response message from the network-slice manager granting permission to use the default machine learning architecture based on a third network slice; and 8. The method of claim 1, further comprising wirelessly communicating additional data for the third application using the default machine learning architecture.

9. A radio frequency transceiver; A user equipment comprising a first processor and a first memory system configured to perform the method of any one of claims 1 to 8.

10. A method performed by a network slice manager, comprising: creating a first network slice to support a first quality of service level; and associating the first network slice with at least one available end-to-end machine learning architecture capable of satisfying the first quality of service level, each of the at least one available end-to-end machine learning architecture being configured for processing wireless communication signals and being a distributed machine learning architecture formed by a machine learning architecture implemented by a user equipment, a machine learning architecture implemented by a base station, and a machine learning architecture implemented by an entity of a core network, the method further comprising: receiving a first machine learning architecture request message from the user equipment requesting permission to use a first machine learning architecture, the first machine learning architecture configured to process wireless communication signals, the method further comprising: determining whether the first machine learning architecture forms part of the at least one available end-to-end machine learning architecture associated with the first network slice; and in response to determining that the first machine learning architecture forms part of the at least one available end-to-end machine learning architecture, sending a first machine learning architecture response message to the user device granting permission to use the first machine learning architecture.

11. The method of claim 10 , further comprising: sending an available machine learning architectures message to the user device identifying the at least one available end-to-end machine learning architecture.

12. 12. The method of claim 10 or 11, further comprising sending a machine learning architecture selection message to the user device, the machine learning architecture selection message including one or more selection rules that enable the user device to determine the first machine learning architecture.

13. Creating a second network slice; and receiving a second machine learning architecture request message from the user device requesting permission to use a second machine learning architecture; determining that the second machine learning architecture is not associated with any second end-to-end machine learning architecture of the second network slice; and The method of any one of claims 10 to 12, further comprising: sending a second machine learning architecture response message to the user device not granting permission to use the second machine learning architecture.

14. The method of claim 13 , wherein the second machine learning architecture response message includes a third machine learning architecture associated with at least one of the second end-to-end machine learning architectures.

15. A network-slice manager comprising a processor and a memory system configured to perform the method of any one of claims 10 to 14.

Images