Artificial intelligence model registration in an open radio access network

Abstract

A method and an apparatus for registering Artificial Intelligence / Machine Learning (AI / ML) model are disclosed. The method includes receiving, by one of a Non-Real Time Radio Access Network (RAN) Intelligence Controller (Non-RT RIC) and a Service Management and Orchestration (SMO), a registration request to register at least one AI / ML model over an R1 interface from at least one of a producer Non-RT RIC application (rApp) and a SMOF. The registration request includes AI / ML model information corresponding to the at least one AI / ML model. The AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute. Further, the inputdatainformation attribute and the outputdatainformation attribute include a data schema parameter and a dataDef parameter for corresponding input data and output data. The method comprises storing, by one of the Non-RT RIC and the SMOF, the received registration request and the AI / ML model information.

Description

ARTIFICIAL INTELLIGENCE MODEL REGISTRATION IN AN OPEN RADIOACCESS NETWORKCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to India provisional application 202411058073, filed on July 31, 2024, and India non-provisoinal application 202411058073, filed April 29, 2025, the entire contents of which are incorporated herein by reference.FIELD

[0002] The present disclosure relates to artificial intelligence model registration in an open radio access network.BACKGROUND

[0003] The information disclosed in this background section is only for enhancement of understanding of the general background of the disclosure and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.

[0004] A Radio Access Network (RAN) is an important component in a telecommunications system. The RAN includes multiple network entities or network components that facilitate connections with end-user devices (user equipment). In recent years, an Open RAN (O-RAN) architecture has been developed that disaggregates functions of the RAN through various logicalnodes. The various components of the O-RAN may include an O-RAN Radio Unit (O-RU), an O- RAN Distributed Unit (O-DU), an O-RAN Centralized Unit (O-CU), a Service Management and Orchestration (SMO), a Near-Real-Time RAN Intelligent Controller (Near-RT RIC), and a Non- Real-Time RAN Intelligent Controller (Non-RT RIC).

[0005] Moreover, the Near-RT RIC or the Non-RT RIC may support an interaction between hosted applications and common functions that may be implemented via Artificial Intelligence / Machine Learning (AI / ML) workflow functions and services. This may further involve model registration and discovery of the corresponding AI / ML models.SUMMARY

[0006] This summary is provided to introduce a selection of concepts, in a simplified format, that are further described in the detailed description of the disclosure. This summary is neither intended to identify key or essential inventive concepts of the disclosure nor is it intended for determining the scope of the disclosure.

[0007] According to one embodiment of the present disclosure, an apparatus is disclosed. The apparatus is configured to receive a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface. The registration request is received from at least one of a producer Non-Real Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) Application (rApp) and a Service Management and Orchestration Function (SMOF). The registration request includes AI / ML model information corresponding to the at least one AI / ML model. The AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute.Further, each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schema parameter and a dataDef parameter for corresponding input data and output data. The apparatus is further configured to store the received registration request and the AI / ML model information.

[0008] According to another embodiment of the present disclosure, a method is disclosed. The method comprises receiving a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface. The registration request is received from at least one of a producer Non-Real Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) Application (rApp) and a Service Management and Orchestration Function (SMOF). The registration request is received by one of a Non-RT RIC and a SMO. The registration request includes AI / ML model information corresponding to the at least one AI / ML model. The AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute. Further, each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schema parameter and a dataDef parameter for corresponding input data and output data. The method further comprises storing, by one of the Non-RT RIC and the SMOF, the received registration request and the AI / ML model information.

[0009] According to one embodiment of the present disclosure, a Non-transitory computer- readable medium is disclosed. The Non-transitory computer-readable medium stores instructions. The instructions comprises one or more instructions that are executed by an apparatus. The apparatus includes one or more processors. The instructions cause the one or more processors to receive, from at least one of a producer Non-Real Time (RT) Radio Access Network (RAN)Intelligent Controller (RIC) Application (rApp) and a Service Management and Orchestration Function (SMOF), a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface. The registration request includes AI / ML model information corresponding to the at least one AI / ML model. The AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute. Further, each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schema parameter and a dataDef parameter for corresponding input data and output data. The one or more instructions further cause the one or more processors to store the received registration request and the AI / ML model information.

[0010] To further clarify the advantages and features of the present disclosure, a more particular description of the disclosure will be rendered by reference to specific embodiments thereof, which are illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the disclosure and are therefore not to be considered limiting of its scope. The disclosure will be described and explained with additional specificity and detail in the accompanying drawings.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Features, aspects, and advantages of embodiments of the disclosure will be described below with reference to the accompanying drawings, in which like reference numerals denote like elements, and wherein:FIG. 1 illustrates a schematic diagram depicting an Open Radio Access Network (0-RAN) architecture associated with a RAN Intelligent Controller (RIC), in accordance with related art;FIG. 2 illustrates a schematic diagram depicting a functional view of an O-RAN standard compliant Non-Real-Time (RT) RIC architecture;FIG. 3 illustrates a schematic diagram depicting a service-based view of the O-RAN standard compliant Non-RT RIC architecture;FIG. 4 illustrates an environment for registering an Artificial Intelligence(AI) / Machine Learning (ML) model, according to an embodiment of the present disclosure;FIG. 5 illustrates a signal flow diagram for registering the AI / ML model, according to an embodiment of the present disclosure;FIGS. 6-7 illustrate flow diagrams depicting methods for registering the AI / ML model, according to an embodiment of the present disclosure; andFIG. 8 illustrates a diagram of example components of a device, according to an embodiment of the present disclosure.DETAILED DESCRIPTION

[0012] The following detailed description of example embodiments refers to the accompanying drawings. The foregoing disclosure provides illustration and description but is not intended to be exhaustive or to limit the implementations to the precise form disclosed. Modifications and variations are possible in light of the present disclosure or may be acquired from the practice of the implementations. Further, one or more features or components of one embodiment may be incorporated into or combined with another embodiment (or one or more features of another embodiment). Additionally, the flowchart and description of operations provided below relate to at least one of the embodiments in the present disclosure. It should be noted that it is possible to make other embodiments that do not exactly match the flowchart and its description. It isunderstood that in other embodiments one or more operations may be omitted, one or more operations may be added, and one or more operations may be performed simultaneously (at least in part). In the present disclosure, specific tasks may be performed using Artificial Intelligence / Machine Learning (ALML) models. An AI / ML model is a model generated using one or more Al technologies, one or more ML algorithms, or both, and generates output data based on input data. This output data is used to perform tasks. Tasks performed using AI / ML models include those generally referred to as intellectual tasks, such as classification, prediction, natural language processing, etc.

[0013] Although Al and ML are explained separately, ML is a technology included in Al. In ML, instead of being explicitly programmed for a specific task, systems can improve their performance over time by identifying patterns and making inferences from training data. Typically, the generation of ML models includes data collection, model training, and model inference. Data collection involves gathering and preprocessing data to be used for training and inference. Model training involves developing and validating models using the collected data. Model inference involves applying the trained models to new data to generate new output data and perform tasks.

[0014] Machine learning includes various types of learning methods such as supervised learning, unsupervised learning, reinforcement learning, semi-supervised learning, self-supervised learning, transductive learning, transfer learning, meta-learning, and the like. These types of learning methods can be appropriately selected according to the embodiments. Unless otherwise specified, the application of types not mentioned in this description is not precluded. Additionally, the structure of ML models may vary depending on the embodiments and learning methods and is not limited to the methods disclosed. Furthermore, ML includes deep learning, which uses models thatinclude neural networks. Deep learning models may include, for example, Deep Neural Networks(DNNs), Convolutional Neural Networks (CNNs), etc.

[0015] It should be noted that the AI / ML models presented hereinafter are examples and are not limited to the illustrated AI / ML models. They can be modified or altered by using different Al or ML algorithms. The configuration of the neural network is not limited to the configuration disclosed in the present disclosure and can be modified.

[0016] It will be apparent that systems and / or methods, described herein, may be implemented in different forms of hardware, software, or a combination of hardware and software. The actual specialized control hardware or software code used to implement these systems and / or methods should not limit their implementations. Thus, the operation and behavior of the systems and / or methods are described herein without reference to specific software code. It is understood that software and hardware may be designed to implement the systems and / or methods based on the description herein.

[0017] Even though particular combinations of features are recited in the claims and / or disclosed in the specification, the particular combinations are not intended to limit the disclosure of implementations. In fact, many of these features may be combined in ways not specifically recited in the claims and / or disclosed in the specification. Even if a dependent claim directly depends on only one claim, the present disclosure may indicate that the dependent claim is dependent on other claims in the claim set.

[0018] No element, act, or instruction used herein should be construed as critical or essential unless explicitly described as such. Also, as used herein, the articles “a” and “an” (in other words, nouns not mentioned in the plural) are intended to include one or more items, and may be usedinterchangeably with “one or more.” Also, as used herein, the terms “has,” “have,” “having,” “include,” “including,” or the like are intended to be open-ended terms. Further, the phrase “based on” is intended to mean “based, at least in part, on” unless explicitly stated otherwise. Furthermore, expressions such as “at least one of [A] and [B],” “[A] and / or [B],” or “at least one of [A] or [B]” are to be understood as including only A, only B, or both A and B.

[0019] The foregoing disclosure provides illustration and description but is not intended to be exhaustive or to limit the implementations to the precise form disclosed. Modifications and variations are possible in light of the above disclosure or may be acquired from the practice of the implementations.

[0020] FIG. 1 illustrates a schematic diagram depicting an Open Radio Access Network (O-RAN) architecture 100 associated with a RAN Intelligent Controller (RIC), in accordance with related art.

[0021] O-RAN alliance, which is a global community of mobile network operators, vendors, and research and academic institutions focused on open RAN standards and interoperability. The O- RAN alliance has defined a set of use cases and applications that the RIC supports. The RIC may be divided into a Non-Real Time (RT) component (for example, a Non-RT RIC 103) and a Near- RT component (for example, a Near-RT RIC 105). The Non-RT RIC 103 may be defined as an element of a centralized Service Management and Orchestration (SMO) framework 101, as defined by the O-RAN Alliance. The Near-RT RIC 105 may reside within a telco edge or a regional cloud. The Telco edge may refer to the deployment of disaggregated RAN functions at edge locations near the radio units to enable low-latency, intelligent, and open radio access networks. The Near-RT RIC 105 may generally enable network optimization actions that take between ten milliseconds to one second to complete.

[0022] In one embodiment, one or more functionalities of the Non-RT RIC 103 may be implemented through applications called Non-RT RIC Applications (rApps). The Non-RT RIC 103 may be connected to the Near-RT RIC 105 over an Al interface. Furthermore, one or more functionalities of the Near-RT RIC 105 may be implemented through applications called Extensible Applications (xApps). The Near-RT RIC 105 may further be connected to E2 nodes (for instance, an eNodeB (O-eNB) 123, gNB) over an E2 interface. The SMO framework 101 may manage and orchestrate various RAN elements. For instance, the SMO framework 101 may orchestrate an O-RAN Cloud (O-Cloud) 111. The SMO framework 101 may interact with the O- Cloud 111 over an 02 interface. The O-Cloud 111 may refer to a collection of physical RAN nodes that host the RICs. The O-RAN architecture 100 may also include an O-Radio Unit (0-RU) 115, an O-Centralized Unit (O-CU) (not shown), and one or more O-Distributed Units (O-DUs) 113 (hereinafter may be referred to as the 0-DU 113). The O-CU may handle non-real-time, higher- layer RAN functions like Packet Data Convergence Protocol (PDCP) layer and Radio Resource Control (RRC) layer. The O-DU 113 may manage real-time, lower-layer functions like Medium Access Control (MAC) layer, Radio Link Control (RLC) layer, and parts of the Physical (PHY) layer. Further, one or more functionalities of various O-RAN components such as the 0-RU 115, and the 0-DU 113, may be implemented through the O-Cloud 111.

[0023] Further, the O-RAN architecture 100 may support one or more Non-RT RIC control loops 107 (hereinafter referred to as the “the Non-RT RIC control loop 107”) and one or more Near-RT RIC control loops 109 (hereinafter referred to as the “the Near-RT RIC control loop 109”). TheNon-RT RIC control loop 107 and the Near-RT RIC control loop 109 may be defined based on a corresponding controlling entity. For example, the Non-RT RIC 103 may act as a control entity for the Non-RT RIC control loop 107. The Non-RT RIC 103 may operate on a timescale greater than 1 second within the SMO framework 101. The timescale may refer to a duration of operation of the rApps. Similarly, the Near-RT RIC 105 may act as a controlling entity for the Near-RT RIC control loop 109. The Near-RT RIC 105 may operate the Near-RT RIC control loop 109 on a timescale between 10 milliseconds and 1 second.

[0024] The 0-RAN architecture 100 further illustrates other logical entities with which the Non- RT RIC 103 and the Near-RT RIC 105 may interact. The Non-RT RIC control loop 107 and the Near-RT RIC control loop 109 may exist at various levels and run simultaneously. The O-RAN alliance defines the use cases for the Non-RT RIC control loop 107 and the Near-RT RIC control loop 109. The 0-RAN alliance also defines an interaction between RICs (for example, the Non- RT RIC 103 and the Near-RT RIC 105). The 0-RAN alliance also specifies interactions for an O- CU-Control Plane (O-CU-CP) 117 and an O-CU-User Plane (O-CU-UP) 121. The O-CU-CP 117 may handle control plane functions like an RRC and Non-Access Stratum (NAS) signaling. The O-CU-UP 121 may manage user plane functions such as packet forwarding and traffic routing in the Non-RT RIC 103. FIG. 1 further illustrates an O-DU control loop 119. The O-DU control loop 119 may correspond to operations such as, call control, mobility, radio scheduling, Hybrid Automatic Repeat Request (HARQ), and beamforming, along with slower functions involving SMO management interfaces.

[0025] Further, the timing of control loops (for example, the Non-RT RIC control loop 107, the Near-RT RIC control loop 109, and the O-DU control loop 119) may depend on different use cases.For example, the Non-RT RIC control loop 107 may take 1 second or more. The Near-RT RIC control loop 109 may operate for about 10 milliseconds or longer. Further, the O-DU control loop119 may operate within 10 milliseconds. For stability, the loop time of the Non-RT RIC control loop 107 and the Near-RT RIC control loop 109 may be much longer than loop time for control entities in the same use case.

[0026] Further, a Service-Based Architecture (SBA) may define roles for service producers and service consumers. The SBA introduces standardized interfaces that ensure smooth interoperability within the SMO framework 101. While the SBA focuses on logical functions, the SBA does not address implementation specifics. When applied correctly, the SBA may offer various key benefits. For example, the SBA may validate services by aligning the services with consumer use cases. The SBA may identify service operations and corresponding semantic behavior through information models. The SBA may specify Application Programming Interfaces (APIs) and data models to ensure syntactic consistency. The SBA may also highlight shared services, like authentication, authorization, and data management, which can be provided by a single producer for multiple internal consumers.

[0027] FIG. 2 illustrates a schematic diagram depicting a functional view of an O-RAN standard compliant Non-RT RIC architecture 200 (hereinafter referred to as the Non-RT RIC architecture 200).

[0028] The Non-RT RIC 103 may be a core part of the SMO framework 101, as explained in reference to FIG. 2. As shown, the Non-RT RIC architecture 200 may include three key components, i.e., one or more rApps 201 (also referred to as the rApps 201), a Non-RT RIC framework 203, and one or more Open APIs, such as Al interface for the rApps 201. The one ormore rApps 201 may be defined as applications configured to be implemented on the Non-RT RIC103. The Non-RT RIC framework 203 may further include various functions to support operations of the Non-RT RIC 103. For example, the R1 interface may connect the rApps 201 to the Non-RT RIC framework 203. The R1 interface may provide services like registration, discovery, AI / ML workflows, and Al -related tasks. The deployment of services across the Non-RT RIC 103 or the SMO framework 101 may remain transparent to the Open APIs. Additionally, a single Non-RT RIC may connect to multiple Near-RT RICs, such as the Near-RT RIC 105.

[0029] The Non-RT RIC framework 203 may include one or more inherent functionalities. The one or more inherent functionalities may be provided as a part of a framework block 205. The inherent functionalities of the framework block 205 may include essential functions necessary for Al interfaces and the rApps 201. The Non-RT RIC framework 203 may define one or more Al functions. The one or more Al functions may serve as a termination point for the Al interface. The one or more Al functions may be considered as inherent Non-RT RIC functions. The one or more Al functions may encompass Al logical termination and may be designed to support one or more Al services. The one or more Al services may include Al policy functions, Al Enterprise Integration (El) functions, and Al ML functions. Further, the inherent functionalities of the framework block 205 may include one or more rApp management functions and one or more R1 service exposure functions that may be vital to manage the rApps 201. The one or more rApp management functions may include rApp conflict mitigation functionality, which may be considered inherent to the Non-RT RIC framework 203. Further, the rApps 201 may also perform rApp orchestration functions. The rApp orchestration functions may involve lifecycle management, policy enforcement, data handling, inter-app coordination, and performance monitoring of RANapplications within the Non-RT RIC 103. However, the rApp orchestration functions may not be part of the one or more rApp management functions. Instead, the rApp orchestration functions may fall under the SMO framework 101. On the other hand, the R1 service exposure functions include key features such as “service registration and discovery function” and “authentication and authorization function”. The R1 service exposure functions help in rApps discovery services provided by both the Non-RT RIC framework 203 and the SMO framework 101. Before accessing the required services, the rApps 201 may undergo authentication and authorization via the relevant function. The inherent functionalities of the framework block 205 may also include other Non-RT RIC framework functions. The other Non-RT RIC framework functions may serve as a placeholder for any additional functions within the Non-RT RIC framework 203 that may be identified in the future.

[0030] The Non-RT RIC framework 203 may further include one or more implementation variability functions 209. The implementation variability functions 209 may either reside within the Non-RT RIC framework 203 or the SMO framework 101, depending on deployment. The deployment of the implementation variability functions 209 may depend on specific implementations. If such a function is deployed within the Non-RT RIC 103 in a given implementation, the function is considered part of the Non-RT RIC framework 203, and its service is accessible to the rApps 201 through the “service registration and discovery function.” Conversely, if the implementation variability functions 209 are deployed within the SMO framework 101 in a particular implementation, the implementation variability functions 209 may not be regarded as part of the Non-RT RIC framework 203, although its service is also accessible to the rApps 201 through the “service registration and discovery function”.

[0031] FIG. 2 further illustrates three types of terminations for external interfaces, which serve as examples of the implementation variability functions 209. The three types of terminations may include an external El termination, an external Al / AIL termination, and a human-machine termination (not shown). The external El termination may connect to one or more external El sources to import enrichment information for the rApps 201. The external AVML termination may link to one or more external AI / ML servers for ML model importation for the rApps 201. The human-machine termination may enable manual input corresponding to RAN intent for the rApps 201. The above-discussed terminations may be located either within the Non-RT RIC 103 or in the SMO framework 101 (but outside of the Non-RT RIC 103). It may be worth noting that the specifications for the above-discussed external interfaces may remain open for further discussion. Additionally, there may be numerous other “implementation variability functions”, such as AI / ML model training, data analytics, and data sharing. In FIG. 2, placeholders labeled “Function 1” through “Function n” may represent additional “implementation variable functions”.

[0032] FIG. 3 illustrates a schematic diagram depicting a service-based view of the 0-RAN standard compliant Non-RT RIC architecture 300 (hereinafter referred to as the Non-RT RIC service-based architecture 300).

[0033] The service-based architectural approach has become a widely adopted standard in the telecommunications industry. Service-based principles allow to describe an architecture not based on fixed functional components and interfaces between them but based on the definition of services as the central point of architecture consideration. While a formal definition is absent, several key principles outline the essence of a service-based architecture. For example, the service-based architecture includes modularity and extensibility. The functionalities may be divided intoappropriate granular services in the modularity. The extensibility ensures the architecture can easily accommodate new services while keeping them discoverable. The service-based architecture may also involve functional abstraction, discoverability, composability, and reusability. The functional abstraction may simplify the complexity of underlying functions. The discoverability ensures services may be located when needed. The composability may allow services to be combined to create new ones. Additionally, reusability may ensure services may be employed across different stages of business processes, and loose coupling enables services to function independently of one another, providing flexibility and adaptability.

[0034] Designing an architecture using a service-based approach may offer significant deployment flexibility and ensure future readiness by allowing the choice of components that produce or consume specific services to be determined at deployment. The service-based approach may also enable multi-vendor interoperability through standardized services and service interfaces. In service-based architecture, services may be modularized and delivered by service producers to service consumers via service endpoints. The service-based architecture may allow modularization of the services rather than the components of a deployment. To ensure security, service consumers may be authenticated and authorized at runtime based on policies, avoiding pre-assigned consumer-producer relationships during standard development. Services may be registered in a registry, enabling consumers to discover available services and their technical parameters, such as endpoints, data formats, and access protocols. Communication may occur directly between consumers and service endpoints or through frameworks like a message bus or service mesh. Adopting a service-based design for functionalities consumed or produced by the rApps 201 may enable seamless operation of the rApps 201. The rApps 201 may function in any deployment thatoffers the required services. This approach removes the need to document implementation variability. Furthermore, the service-based design may support the evolution of the rApps 201 into more generic smart network management applications, as the rApps 201 may easily discover and utilize additional services.

[0035] A service represents a set of capabilities that a service producer offers to consumers through defined endpoints. The primary purpose of the service is to enable access, usage, and control of specific functionalities available within the 0-RAN. While the production and consumption of services are roles assigned to software entities in deployments, the specification does not detail the exact entities involved. A service may be standardized by the 0-RAN Alliance, another organization, or remain proprietary, as long as it adheres to a standardized method for registration, access, and control within the 0-RAN.

[0036] FIG. 3 further illustrates R1 services along with the associated capabilities that belong to the SMO framework 101 and the Non-RT RIC framework 203. These capabilities include communication support tailored to meet service requirements, such as one-to-one, one-to-many, publish-subscribe, and routed communication patterns. Additionally, FIG. 3 highlights how the service-based approach consistently provides access to inherent functionalities of the Non-RT RIC 103 and the SMO framework 101, implementation-dependent functionalities, and even capabilities external to the SMO framework 101.

[0037] The Non-RT RIC service-based architecture 300 may support various deployment options, similar to deployment options illustrated in FIG. 2. This is achieved by appropriately configuring the pairs of authorized producers and consumers to meet specific requirements.

[0038] The conventional solution of the Non-RT RIC 103 has revealed the need for designing an API for AI / ML model registration. However, the existing solution only provides the most basic features necessary for registering the AI / ML models. Additionally, the advanced feature needs to be supported for commercial adoption of AI / ML solutions in the O-RAN.

[0039] Therefore, there is a need to ensure that the registered AI / ML models are utilized in the most accurate and efficient manner to maximize their potential and effectiveness.

[0040] Referring now to the drawings, and more particularly to FIGS. 4 to 7, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments.

[0041] FIG. 4 illustrates an environment 400 for registering at least one Artificial Intelligence(AI) / Machine Learning (ML) model, according to an embodiment of the present disclosure. The environment 400 may include a Management Service (MnS) consumer 410 connected to at least one MnS producer 420 via an R1 interface 430. It may be noted that, although only one MnS producer 420 is depicted, the MnS consumer 410 may be connected to more than one MnS producer 420. Examples of the MnS consumer 410 may include, but are not limited to, the Non-RT RIC 103, and Orchestration and Automation platforms, such as the SMO. Examples of the MnS producer 420 may include, but are not limited to, SMO Functions (SMOFs) and rApps.

[0042] The MnS consumer 410 may be configured to receive a registration request from the MnS producer 420. The registration request may correspond to a request to register the at least one AI / ML model. The registration request may be received over the R1 interface 430. In an embodiment, the registration request may include AI / ML model information corresponding to the at least one AI / ML model. In an embodiment, the AI / ML model information may include, but isnot limited to, an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute. The inputdatainformation attribute may provide information about the input data to train the at least one AI / ML model. The inputdatainformation attribute may also be used to identify the input data required for inferencing of the at least one AI / ML model. Similarly, the outputdatainformation attribute may provide information about the output data to train the at least one AI / ML model. The outputdatainformation attribute may also be used to decode the output data obtained from the inferencing of the at least one AI / ML model. Further, the metadata attribute may provide metadata of the at least one AI / ML model. In an embodiment, the inputdatainformation attribute, the outputdatainformation attribute, and the metadata attribute may be defined as shown in Table 1 :Table 1

[0043] In a further embodiment, the inputdatainformation attribute may include a data schema parameter and a dataDef parameter for corresponding input data. Similarly, theoutputdatainformation attribute may include a data schema parameter and a dataDef parameter for corresponding output data. In an embodiment, the data schema parameter may indicate a data type of the corresponding input data and output data. The data schema parameter may also indicate a range of the corresponding input data and output data. The data schema parameter may further indicate an encoding format of the corresponding input data and output data. In an embodiment, the inputdatainformation attribute and the outputdatainformation attribute may be defined as shown in Table 2:Table 2

[0044] Further, the dataDef parameter may indicate data names and descriptions of the corresponding input data and output data. In an embodiment, the data names may be defined using standardized measurement for standardized metrics and Key Performance Indicators (KPIs). Further, the data names for KPIs (i.e., KPI names) may be defined in TS28.552 and TS 28.554, e.g. RRU.MaxPrbUsedDl.QoS. Further, the data names may be provided along with a description in string for non-standardized data. Further, the description of the data, i.e., the output data or theinput data is required if the data is non-standard based, e.g. detected or predicted anomaly category.In an embodiment, the dataDef parameter may be defined as shown in Table 3:

[0045] Further, in addition to the inputdatainformation and the outputdatainformation, additional metadata information may be provided in the AI / ML model information. In an embodiment, the metadata information may be provided in the metadata attribute. The metadata attribute may indicate a modelResourceRequirement parameter. The modelResourceRequirement parameter may indicate hardware resource details of the at least one AI / ML model. The hardware resource details may include, but are not limited to, minimum Central Processing Unit (CPU) and memory required to deploy the at least one AI / ML model. The metadata attribute may further indicate a modelPerformance parameter. The modelPerformance parameter may indicate at least a performance score of the at least one AI / ML model. The performance score may be defined in terms of metrics like accuracy, precision, Mean Squared Error (MSE), etc. The metadata attribute may also indicate a modelcomplexity parameter. The modelcomplexity parameter may indicatea complexity level of the at least one AI / ML model. For example, the complexity level may include Vapnik-Chervonenkis (VC) dimensioning, a number of parameters of the at least one AI / ML model, model depth for Artificial Neural Networks (ANN), etc. The VC dimension may refer to a concept that helps quantify the complexity of a hypothesis class. The metadata attribute may also indicate a modelUsage parameter. The modelUsage parameter may indicate a type of problem addressed by the at least one AI / ML model. For example, the type of problem addressed by the at least one AI / ML model may include, but is not limited to, regression, classification, and timeseries prediction. The metadata attribute may further indicate a modeiFramework parameter. The modeiFramework parameter may indicate a model framework corresponding to the at least one AI / ML model. It may be noted that the Non-RT RIC framework 203 or the rApps 201 may support some selected frameworks only for model training or inference. Accordingly, the modeiFramework parameter may be used to check the compatibility of the AI / ML framework being used. Further, the metadata attribute may indicate a leamingMechanism parameter. The learningMechanism parameter may indicate a learning mechanism of the at least one AI / ML model. The metadata attribute may further indicate a modelTrainable parameter. The modelTrainable parameter may indicate whether the at least one AI / ML model is trainable. The various parameters of the metadata attribute have been defined in Table 4:Table 4

[0046] In an embodiment, the input and output data information attributes along with the metadata attribute may allow the user of the at least one AI / ML model (e.g. another rApp or SMOF) to retrieve the information during AI / ML model discovery. The various metadata attributes are further defined in the following paragraphs.

[0047] In an embodiment, the modelResourceRequirement parameter may include fields to inform the Non-RT RIC 103 about the required resources by the at least one AI / ML model to deploy the model. Accordingly, the Non-RT RIC 103 may decide whether the at least one AI / ML model may be allowed to register and later get deployed. In an embodiment, the modelResourceRequirement parameter may include, but is not limited to, a CPU parameter and a memory parameter. The CPU parameter may indicate a number of millicores or resource measuring units required to deploy the at least one AI / ML model. The memory parameter may indicate a minimum amount of memory required to deploy the at least one AI / ML model. The various parameters of the modelResourceRequirement parameter may be defined as shown in Table 5:Table 5

[0048] The modelPerformance parameter may include fields indicating the MnS consumer 410, about the performance of the at least one AI / ML model. In an embodiment, the modelPerformance parameter may include, but is not limited to, an inferenceOutputName parameter, a performanceMetricName parameter, a performanceMetricValue parameter, a performanceScore parameter, and a decisionConfidenceScore parameter. The inferenceOutputName may indicate a name of an inference output data corresponding to a dataName provided in the outputdatainformation attribute. In an embodiment, the inferenceOutputName parameter may be defined in accordance with 3rdGeneration Partnership Project (3GPP) Technical Specification (TS) 28.105, clause 7.5.1. The performanceMetricName parameter may indicate a name of a performance metric used to measure the performance of the at least one AI / ML model, such as precision or accuracy for supervised models. In an embodiment, the performanceMetricName parameter may be defined in accordance with 3GPP TS 28.105, clause 7.5.1. The performanceMetricValue parameter may indicate a value of the performance metric. The performanceScore parameter may indicate the performance score of the at least one AI / ML model. In an embodiment, the performanceScore parameter may indicate the performance score (in unitof percentage) of an AI / ML entity when performing inference on a specific data set. The performance metrics may be different for different types of AI / ML models depending on the nature of the at least one AI / ML model. For example, accuracy may serve as the performance metric for a numeric prediction AI / ML model. In contrast, a classification AI / ML model often relies on precision and recall. Accordingly, the performance metric for the classification AI / ML model may be a combination of precision and recall, such as the "Fl score," which balances both precision and recall effectively. In an embodiment, the allowedValues for the performanceScore parameter are {0 ... 100}. In an embodiment, the performanceScore parameter may be defined in accordance with 3GPP TS 28.105, clause 7.5.1. The decisionConfidenceScore parameter may indicate a decision level corresponding to a decision generated by the at least one AI / ML model. In an embodiment, the decisionConfidenceScore parameter may be considered as a numerical value that represents the dependability / quality of a given decision generated by the at least one AI / ML inference function. The lowest value of the decisionConfidenceScore parameter may indicate the lowest level of dependability of the decisions, i.e. that the data is not usable at all. In an embodiment, the allowedValues for the decisionConfidenceScore parameter are {0 ... 100}. In an embodiment, the decisionConfidenceScore parameter may be defined in accordance with 3GPP TS 28.105, clause 7.5.1. The various parameters of the modelPerformance parameter may be defined as shown in Table 6:Table 6

[0049] The modelComplexity parameter may contain fields related to model complexity of the at least one AI / ML model. Accordingly, the modelComplexity parameter may include, but is not limited to, a complexityMetricName parameter and a complexityMetricValue parameter. The complexityMetricName parameter may indicate a name of a complexity metric used to measure the complexity of the at least one AI / ML model. For example, the complexityMetricName parameter may indicate model Size, number of variables, number of floating points operations, etc. The complexityMetricValue parameter may indicate a value of the complexity metric. The various parameters of the modelComplexity parameter may be defined as shown in Table 7:Table 7

[0050] The ModelUsage parameter may include, but is not limited to, a TIMESERIES FORECASTING parameter, a CLASSIFICATION parameter, and a RANKING parameter. The TIMESERIES FORECASTING parameter may indicate time series data used for a forecast function of the at least one AI / ML model. For example, the TIMESERIES FORECASTING parameter may indicate time series data used to forecast the number of users at any given time in a cluster of cells. The CLASSIFICATION parameter may indicate one or more categories corresponding to the input data. The one or more categories may include, but are not limited to, anomaly, not anomaly. The RANKING parameter may indicate a plurality of rank values to rank the input data. The plurality of rank values may include, but are not limited to, certain numerical levels, e.g. priority value of target cells for handover. The various parameters of the ModelUsage parameter may be defined as shown in Table 8:Table 8

[0051] Further, the modeiFramework parameter may include, but is not limited to, a PYTORCH indicator, a TENSORFLOW indicator, a XGBOOST indicator, a SKLEARN indicator, a LIGHTGBM indicator, and a framework indicator (also referred to as OTHER indicator). The PYTORCH indicator may indicate that the at least one AI / ML model is developed using Pytorch.The TENSORFLOW indicator may indicate that the at least one AI / ML model is developed using Tensorflow. The XGBOOST indicator may indicate that the at least one AI / ML model is developed using XGBoost. The SKLEARN indicator may indicate that the at least one AI / ML model is developed using Skleam. The LIGHTGBM indicator may indicate that the at least one AI / ML model is developed using LightGBM. The framework indicator may indicate that the at least one AI / ML model is developed using another framework. Further, it may be noted that the supported model frameworks may be part of the modeiFramework parameter. The various parameters of the modeiFramework parameter may be defined as shown in Table 9:Table 9

[0052] The LearningMechanism parameter may include, but is not limited to, a SUPERVISED indicator, an UNSUPERVISED indicator, a SEMI-SUPERVISED indicator, and a REINFORCEMENT indicator. The SUPERVISED indicator may indicate that the at least one AI / ML model may learn from labeled data and aims to predict future labels. The UNSUPERVISED indicator may indicate that the at least one AI / ML model may learn using patterns or structures from the unlabeled input data without any guidance. The SEMI¬SUPERVISED indicator may indicate that the at least one AI / ML model may combine bothsupervised and unsupervised techniques to learn. Further, in an embodiment, the dataSchema and the dataDef parameters are present when the LearningMechanism parameter includes the SUPERVISED indicator or the SEMI-SUPERVISED indicator. The REINFORCEMENT indicator may indicate that the at least one AEML model may learn from the feedback in the form of rewards or penalties. The various parameters of the modeiFramework parameter may be defined as shown in Table 10:Table 10

[0053] Further, the MnS consumer 410 may be configured to store the received ALML information. The MnS consumer 410 may be further configured to receive a discovery request from at least one consumer rApp. In one embodiment, the discovery request may correspond to a request to discover at least one AEML model. In response, the MnS consumer 410 may transmit the AI / ML model information associated with the at least one AI / ML model.

[0054] FIG. 5 illustrates a signal flow diagram 500 of registering the at least one AI / ML model, according to an embodiment of the present disclosure. The signal flow diagram 500 may illustrate communication between a MnS producer 502, a MnS consumer 504, and rApp / SMOF 506. In oneembodiment, the MnS producer 502 may correspond to the MnS producer 420, and the MnS consumer 504 may correspond to the MnS consumer 410. As shown, at operation 501, the MnS consumer 504 may receive a registration request from the MnS producer 502. In an embodiment, the MnS producer 502 may correspond to at least one of the rApp and SMOF. It may be noted that the received registration request may correspond to the registration request explained in reference to FIG. 4. Accordingly, the registration request has not been explained again for the sake of the brevity of the disclosure. Subsequently, at operation 503, the MnS consumer 504 may store the AI / ML model information received in the registration request. It may be noted that the AI / ML model information may correspond to the AI / ML model information explained in reference to FIG. 4. Hence, the details of the AI / ML model information have been omitted here for the sake of brevity of the disclosure. Further, at operation 505, the MnS consumer 504 may receive a discovery request from at least one of the rApp / SMOF 506. In one embodiment, the discovery request may correspond to a request to discover at least one AI / ML model. In response, at operation 507, the MnS consumer 504 may transmit the AI / ML model information associated with the at least one AI / ML model to the at least one of the rApp / SMOF 506. It may be noted that when the rApp / SMOF 506 transmits the discovery request to the MnS consumer 504, then the rApp / SMOF 506 may act as an MnS consumer, and the MnS consumer 504 may act as the MnS producer. Furthermore, even though the signal flow diagram 500 has been explained with respect to one MnS producer only, the steps of the signal flow diagram 500 may be applicable to more than one MnS producers. In an embodiment, the rApp / SMOF 506 may decide whether, where, and when to use the at least one AI / ML model using the received AI / ML model information. The rApp / SMOF 506 may also prepare the data input for model training and inference using the received AI / MLmodel information. Further, the rApp / SMOF 506 may interpret and process the data output from the model during model training and inference using the received AI / ML model information. The rApp / SMOF 506 may also prepare label data for model training using the received AI / ML model information. The rApp / SMOF 506 may also prepare training and inference software environment using the received AI / ML model information. The rApp / SMOF 506 may also prepare computing resources, i.e. hardware environment, for model training and inference using the received AI / ML model information. Further, the rApp / SMOF 506 may evaluate the performance of the model using the received AI / ML model information.

[0055] FIGS. 6-7 illustrate flow diagrams depicting methods 600, 700, respectively, for registering the at least one AI / ML model, according to an embodiment of the present disclosure. The methods 600 and 700 may be performed by the MnS consumer 410 or the MnS consumer 504.

[0056] At step 601, the method 600 may include receiving the registration request to register the at least one AI / ML model over the R1 interface. The registration request may be received from at least one of the rApp and the SMOF. Further, the registration request may correspond to a request to register the at least one AI / ML model. The registration request may include the AI / ML model information corresponding to the at least one AI / ML model. The AI / ML model information may include the inputdatainformation attribute, the outputdatainformation attribute, and the metadata attribute. Further, each of the inputdatainformation attribute and the outputdatainformation attribute may include the data schema parameter and the dataDef parameter for corresponding input data and output data. The inclusion of the AI / ML information may enable the rApps or SMOF to utilize the registered AI / ML models correctly and most efficiently. Further, the inclusion of the AI / ML information may result in improving the interoperability among multiple vendorsfor model sharing in the SMO / Non-RT RIC system. The inclusion of the AI / ML information may also minimize the development and interoperability cost of the AI / ML models. Furthermore, the inclusion of the AI / ML information may minimize the risk of negative impact on the network performance due to the misuse of the AI / ML models.

[0057] Thereafter, at step 603, the method 600 may include storing the received registration request and the AI / ML model information.

[0058] In an embodiment, the data schema parameter may indicate at least one of the encoding format of the corresponding input data and output data, the data type of the corresponding input data and output data, and the range of the corresponding input data and output data.

[0059] In an embodiment, the dataDef parameter may indicate data names and descriptions of the corresponding input data and output data.

[0060] In an embodiment, the metadata attribute may indicate the modelResourceRequirement parameter. The modelResourceRequirement parameter may indicate hardware resource details of the at least one AI / ML model. The metadata attribute may further indicate the modelPerformance parameter. The modelPerformance parameter may indicate at least a performance score of the at least one AI / ML model. The metadata attribute may also indicate the modelcomplexity parameter. The modelComplexity parameter may indicate the complexity level of the at least one AI / ML model. The metadata attribute may also indicate the modelUsage parameter. The modelUsage parameter may indicate the type of problem addressed by the at least one AI / ML model. The metadata attribute may further indicate the modeiFramework parameter. The modeiFramework parameter may indicate the model framework corresponding to the at least one AI / ML model. Further, the metadata attribute may indicate the learningMechanism parameter. ThelearningMechanism parameter may indicate the learning mechanism of the at least one AI / ML model. The metadata attribute may further indicate the modelTrainable parameter. The modelTrainable parameter may indicate whether the at least one AI / ML model is trainable.

[0061] In an embodiment, the modelResourceRequirement parameter may include the CPU parameter and the memory parameter. The CPU parameter may indicate the number of millicores or resource measuring unit required to deploy the at least one AI / ML model. The memory parameter may indicate the minimum amount of memory required to deploy the at least one AI / ML model.

[0062] In an embodiment, the modelPerformance parameter may include the inferenceOutputName parameter, the performanceMetricName parameter, the performanceMetricValue parameter, the performanceScore parameter, and the decisionConfidenceScore parameter. The inferenceOutputName may indicate the name of the inference output data corresponding to the dataName provided in the outputdatainformation attribute. The performanceMetricName parameter may indicate the name of the performance metric used to measure the performance of the at least one AI / ML model. The performanceMetricValue parameter may indicate the value of the performance metric. The performanceScore parameter may indicate the performance score of the at least one AI / ML model. The decisionConfidenceScore parameter may indicate the decision level corresponding to the decision generated by the at least one AI / ML model.

[0063] In an embodiment, the modelComplexity parameter may include the complexityMetricName parameter and the complexityMetricValue parameter. The complexityMetricName parameter may indicate the name of the complexity metric used tomeasure the complexity of the at least one AI / ML model. The complexityMetricValue parameter may indicate the value of the complexity metric.

[0064] In an embodiment, the ModelUsage parameter may include the TIMESERIES FORECASTING parameter, the CLASSIFICATION parameter, and the RANKING parameter. The TIMESERIES FORECASTING parameter may indicate time series data used for the forecast function of the at least one AI / ML model. The CLASSIFICATION parameter may indicate the one or more categories corresponding to the input data. The RANKING parameter may indicate the plurality of rank values to rank the input data.

[0065] In an embodiment, the modeiFramework parameter may include the PYTORCH indicator, the TENSORFLOW indicator, the XGBOOST indicator, the SKLEARN indicator, the LIGHTGBM indicator, and the framework indicator.

[0066] In an embodiment, the LearningMechanism parameter may include the SUPERVISED indicator, the UNSUPERVISED indicator, the SEMI- SPER VISED indicator, and the REINFORCEMENT indicator.

[0067] Referring to FIG. 7, at step 701, the method 700 may include receiving the discovery request from the at least one consumer rApp. The discovery request may correspond to a request for discovery of the at least one AI / ML model.

[0068] Thereafter, at step 703, the method 700 may include transmitting the AI / ML model information corresponding to the at least one AI / ML model.

[0069] While the above-discussed steps in FIGS. 6 and 7 are shown and described in a particular sequence, the steps may occur in variations to the sequence in accordance with variousembodiments. Further, a detailed description related to the various steps of FIGS. 6 and 7 is already covered in the description related to FIGS. 4-5 and is omitted herein for the sake of brevity.

[0070] FIG. 8 is a diagram of example components of a wireless communication device 800 (also referred to as the device / apparatus), according to an embodiment of the present disclosure. In one or more embodiments, the wireless communication device 800 may correspond to a wireless server, the MnS consumer 410, the MnS producer 420, the MnS producer 502, and / or the MnS consumer 504. As shown in FIG. 8, the device 800 includes a processor 810, a memory 820, a storage component 830, an input component 840, an output component 850, a communication interface 860, and a bus 870.

[0071] The processor 810, as used herein, means any type of computational circuit that may comprise hardware elements and software elements. The processor 810 may be embodied as a multi-core processor, a single-core processor, or a combination of one or more multi-core processors and / or one or more single-core processors, a distributed processing system, or the like. The processor 810 may be a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), an Accelerated Processing Unit (APU), an Application-Specific Integrated Circuit (ASIC), or another type of processing component.

[0072] The memory 820 includes a Non-transitory computer-readable medium. The memory 820 includes a Random-Access Memory (RAM), a Read Only Memory (ROM), and / or another type of dynamic or static storage device (e.g., a flash memory, a magnetic memory, and / or an optical memory) that stores information and / or instructions for use by the processor 810. The memory820 comprises machine-readable instructions which are executable by the processor 810. Thesemachine-readable instructions when executed by the processor 810 cause the processor 810 to perform one or more method steps of an embodiment described above.

[0073] The storage component 830 stores information and / or software related to the operation and use of the device 800. For example, the storage component 830 may include a hard disk (e.g., a magnetic disk, an optical disk, a magneto-optic disk, and / or a solid-state disk), a Compact Disc (CD), a Digital Versatile Disc (DVD), a floppy disk, a cartridge, a magnetic tape, and / or another type of Non-transitory computer-readable medium, along with a corresponding drive.

[0074] The input component 840 is configured to receive information, such as user input. For example, the input component 840 may include, but not be limited to, a touch screen display, a keyboard, a keypad, a mouse, a button, a switch, and / or a microphone. Additionally, or alternatively, the input component 840 may include a sensor for sensing information (e.g., a Global Positioning System (GPS), an accelerometer, a gyroscope, and / or an actuator).

[0075] The output component 850 is configured to provide output information from the device 800. For example, the output component 850 may include, but is not limited to, a display, a speaker, an instruction device to an external device, and / or one or more Light-Emitting Diodes (LEDs).

[0076] The communication interface 860 is an interface that provides a communication connection to other devices, such as external devices and internal devices. The connection by the communication interface 860 can be a wired connection, a wireless connection, or a combination of wired and wireless connections, and can be a direct connection or an indirect connection via a communication network that exists between the device 800 and other devices. In other words, the standard of the communication interface 860 is not limited.

[0077] The bus 870 acts as an interconnect between the processor 810, the memory 820, the storage component 830, the input component 840, the output component 850, and the communication interface 860 of the device 800. The bus 870 may include a wired interconnection or a wireless interconnection.

[0078] The number and arrangement of components shown in FIG. 8 are provided as an example.In practice, the device 800 may include additional components, fewer components, different components, or differently arranged components than those shown in FIG. 8.

[0079] Additionally, or alternatively, a set of components (e.g., one or more components) of the device 800 may perform one or more functions described as being performed by another set of components of the device 800. Further, one or more method steps described in any of the embodiments may be performed utilizing a plurality of devices 800 in communication with one another.

[0080] It is understood that terms including “unit” at the end may refer to the unit for processing at least one function or operation and may be implemented in hardware, software, or a combination of hardware and software.

[0081] In one embodiment, an apparatus is disclosed. The apparatus is configured to receive, from at least one of a producer Non-Real Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) Application (rApp) and a Service Management and Orchestration Function (SMOF), a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface. The registration request includes AI / ML model information corresponding to the at least one AI / ML model. The AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadataatribute. Further, each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schema parameter and a dataDef parameter for corresponding input data and output data. The apparatus is further configured to store the received registration request and the AI / ML model information.

[0082] The apparatus as described in

[0081] , wherein the apparatus is further configured to: receive a discovery request from at least one consumer rApp, wherein the discovery request is associated with the at least one AI / ML model; and transmit the AI / ML model information corresponding to the at least one AI / ML model in response to receiving the discovery request.

[0083] The apparatus as described in any one of

[0081] to

[0082] , wherein the data schema parameter indicates at least one of an encoding format of the corresponding input data and output data, a data type of the corresponding input data and output data, and a range of the corresponding input data and output data.

[0084] The apparatus as described in any one of

[0081] to

[0083] , wherein the dataDef parameter indicates data names and descriptions of the corresponding input data and output data.

[0085] The apparatus as described in any one of

[0081] to

[0084] , wherein the metadata attribute indicates a modelResourceRequirement parameter indicating hardware resource details of the at least one AI / ML model, a modelPerformance parameter indicating at least a performance score of the at least one AI / ML model, a modelComplexity parameter indicating a complexity level of the at least one AI / ML model, a modelUsage parameter indicating a type of problem addressed by the at least one AI / ML model, a modeiFramework parameter indicating a model framework corresponding to the at least one AI / ML model, a learningMechanism parameter indicating alearning mechanism of the at least one AI / ML model, and a model Trainable parameter indicating whether the at least one AI / ML model is trainable.

[0086] The apparatus as described in any one of

[0081] to

[0085] , wherein the modelResourceRequirement parameter includes at least one of a Central Processing unit (CPU) parameter indicating a number of millicores or resource measuring unit required to deploy the at least one AI / ML model and a memory parameter indicating a minimum amount of memory required to deploy the at least one AI / ML models.

[0087] The apparatus as described in any one of

[0081] to

[0086] , wherein the modelPerformance parameter includes at least one of an inferenceOutputName parameter indicating a name of an inference output data corresponding to a dataName provided in the outputdatainformation attribute, a performanceMetricName parameter indicating a name of a performance metric used to measure the performance of the at least one AI / ML model, a performanceMetricValue parameter indicating a value of the performance metric, a performanceScore parameter indicating the performance score of the at least one AI / ML model, and a deci si onConfidence Score parameter indicating a decision level corresponding to a decision generated by the at least one AI / ML model.

[0088] The apparatus as described in any one of

[0081] to

[0087] , wherein the modelComplexity parameter includes at least one of a complexityMetricName parameter indicating a name of a complexity metric used to measure the complexity of the at least one AI / ML model, and a complexityMetricValue parameter indicating a value of the complexity metric.

[0089] The apparatus as described in any one of

[0081] to

[0088] , wherein the ModelUsage parameter includes at least one of a TIMESERIES FORECASTING parameter indicating time series data used for a forecast function of the at least one AI / ML model, a CLASSIFICATIONparameter indicating one or more categories corresponding to the input data, and a RANKING parameter indicating a plurality of rank values to rank the input data.

[0090] The apparatus as described in any one of

[0081] to

[0089] , wherein the modeiFramework parameter includes at least one of a PYTORCH indicator, a TENSORFLOW indicator, a XGBOOST indicator, a SKLEARN indicator, a LIGHTGBM indicator, and a framework indicator.

[0091] The apparatus as described in any one of

[0081] to

[0090] , wherein the LearningMechanism parameter includes at least one of a SUPERVISED indicator, an UNSUPERVISED indicator, a SEMI-SUPERVISED indicator, and a REINFORCEMENT indicator.

[0092] The apparatus as described in any one of

[0081] to

[0091] , wherein the apparatus corresponds to one of a Non-Real Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) and a Service Management and Orchestration (SMO).

[0093] In another embodiment, a method is described. The method comprises receiving, by one of a Non-Real Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) and a Service Management and Orchestration (SMO), a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface, from at least one of a producer Non-Real Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) Application (rApp) and a Service Management and Orchestration Function (SMOF). The registration request includes AI / ML model information corresponding to the at least one AI / ML model. The AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute. Further, each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schemaparameter and a dataDef parameter for corresponding input data and output data. The method further comprises storing, by one of the Non-RT RIC and the SMOF, the received registration request and the AI / ML model information.

[0094] The method as described in

[0093] , wherein the method further comprising: receiving a discovery request from at least one consumer rApp, wherein the discovery request is associated with the at least one AI / ML model; and transmitting the AI / ML model information corresponding to the at least one AI / ML model in response to receiving the discovery request.

[0095] The method as described in any one of

[0093] to

[0094] , wherein the data schema parameter indicates at least one of an encoding format of the corresponding input data and output data, a data type of the corresponding input data and output data, and a range of the corresponding input data and output data and wherein the dataDef parameter indicates data names and descriptions of the corresponding input data and output data.

[0096] The method as described in any one of

[0093] to

[0095] , wherein the metadata attribute indicates a modelResourceRequirement parameter indicating hardware resource details of the at least one AI / ML model, a modelPerformance parameter indicating at least a performance score of the at least one AI / ML model, a modelComplexity parameter indicating a complexity level of the at least one AI / ML model, a modelUsage parameter indicating a type of problem addressed by the at least one AI / ML model, a modeiFramework parameter indicating a model framework corresponding to the at least one AI / ML model, a learningMechanism parameter indicating a learning mechanism of the at least one AI / ML model, and a modelTrainable parameter indicating whether the at least one AI / ML model is trainable; andwherein the modelResourceRequirement parameter includes at least one of a Central Processing unit (CPU) parameter indicating a number of millicores required to deploy the at least one AI / ML model and a memory parameter indicating a minimum amount of memory required to deploy the at least one AI / ML model.

[0097] The method as described in any one of

[0093] to

[0096] , wherein the modelPerformance parameter includes at least one of an inferenceOutputName parameter indicating a name of an inference output data corresponding to a dataName provided in the outputdatainformation attribute, a performanceMetricName parameter indicating a name of a performance metric used to measure the performance of the at least one AI / ML model, a performanceMetricValue parameter indicating a value of the performance metric, a performanceScore parameter indicating the performance score of the at least one AI / ML model, and a deci si onConfidence Score parameter indicating a decision level corresponding to a decision generated by the at least one AI / ML model.

[0098] The method as described in any one of

[0093] to

[0097] , wherein the modelComplexity parameter includes at least one of a complexityMetricName parameter indicating a name of a complexity metric used to measure the complexity of the at least one AI / ML model, and a complexityMetricValue parameter indicating a value of the complexity metric; and wherein the ModelUsage parameter includes at least one of a TIMESERIES FORECASTING parameter indicating time series data used for a forecast function of the at least one AI / ML model, a CLASSIFICATION parameter indicating one or more categories corresponding to the input data, and a RANKING parameter indicating a plurality of rank values to rank the input data.

[0099] The method as described in any one of

[0093] to

[0098] , wherein the modeiFramework parameter includes at least one of a PYTORCH indicator, a TENSORFLOW indicator, a XGBOOST indicator, a SKLEARN indicator, a LIGHTGBM indicator, and a framework indicator; and wherein the LeamingMechanism parameter includes at least one of a SUPERVISED indicator, an UNSUPERVISED indicator, a SEMI-SUPERVISED indicator, and a REINFORCEMENT indicator.

[0100] In one embodiment, a Non-transitory computer-readable medium is disclosed. The Non- transitory computer-readable medium stores instructions. The instructions comprising one or more instructions that are executed by an apparatus. The apparatus comprising one or more processors. The instructions cause the one or more processors to receive, from at least one of a producer NonReal Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) Application (rApp) and a Service Management and Orchestration Function (SMOF), a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface. The registration request includes AI / ML model information corresponding to the at least one AI / ML model. The AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute. Further, each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schema parameter and a dataDef parameter for corresponding input data and output data. The one or more instructions further cause the one or more processors to store the received registration request and the AI / ML model information.

[0101] Accordingly, the present disclosure provides techniques for registering the at least one AI / ML model in the 0-RAN.

[0102] Embodiments of the present disclosure offer several significant commercial and technical advantages, for example:

[0103] Efficient way of registering the AVML model: The disclosed techniques enable the rApps or SMOF to utilize the registered AI / ML models correctly and most efficiently.

[0104] The disclosed techniques result in improving the interoperability among multiple vendors for model sharing in the SMO / Non-RT RIC system.

[0105] The disclosed techniques result in minimizing the development and interoperability cost of the AI / ML models.

[0106] The disclosed techniques result in minimizing the risk of negative impact on the network performance due to the misuse of the AI / ML models.

[0107] While specific language has been used to describe the disclosure, any limitations arising on account of the same are not intended. As would be apparent to a person in the art, various working modifications may be made to the method in order to implement the inventive concept as taught herein.

[0108] The drawings and the forgoing description give examples of embodiments. Those skilled in the art will appreciate that one or more of the described elements may well be combined into a single functional element. Alternatively, certain elements may be split into multiple functional elements. Elements from one embodiment may be added to another embodiment. For example, orders of processes described herein may be changed and are not limited to the manner described herein.

[0109] Moreover, the actions of any flow diagram need not be implemented in the order shown; nor do all of the acts necessarily need to be performed. Also, those acts that are not dependent on other acts may be performed in parallel with the other acts. The scope of embodiments is by no means limited by these specific examples. Numerous variations, whether explicitly given in the specification or not, such as differences in structure, dimension, and use of material, are possible. The scope of embodiments is at least as broad as given by the following claims.

[0110] Benefits, other advantages, and solutions to problems have been described above with regard to specific embodiments. However, the benefits, advantages, solutions to problems, and any component(s) that may cause any benefit, advantage, or solution to occur or become more pronounced are not to be construed as a critical, required, or essential feature or component of any or all the claims.[0U1] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and / or adapt for various applications such specific embodiments without departing from the generic concept. Therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of at least one embodiment, those skilled in the art will recognize that the embodiments herein can be practiced with modification within the spirit and scope of the embodiments as described herein.

Claims

We Claim:

1. An apparatus configured to: receive, from at least one of a producer Non-Real Time (RT) Radio access Network (RAN) Intelligent Controller (RIC) Application (rApp) and a Service Management and Orchestration Function (SMOF), a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface, wherein the registration request includes AI / ML model information corresponding to the at least one AI / ML model, wherein the AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute, and wherein each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schema parameter and a dataDef parameter for corresponding input data and output data; and store the received registration request and the AI / ML model information.

2. The apparatus as claimed in claim 1, further configured to: receive a discovery request from at least one consumer rApp, wherein the discovery request is associated with the at least one AI / Model; and transmit the AI / ML model information corresponding to the at least one AI / ML model in response to receiving the discovery request.

3. The apparatus as claimed in claim 1, wherein the data schema parameter indicates at least one of an encoding format of the corresponding input data and output data, a data type of the corresponding input data and output data, and a range of the corresponding input data and output data.

4. The apparatus as claimed in claim 1, wherein the dataDef parameter indicates data names and descriptions of the corresponding input data and output data.

5. The apparatus as claimed in claim 1, wherein the metadata attribute indicates a modelResourceRequirement parameter indicating hardware resource details of the at least one AI / ML model, a modelPerformance parameter indicating at least a performance score of the at least one AI / ML model, a modelComplexity parameter indicating a complexity level of the at least one AI / ML model, a modelUsage parameter indicating a type of problem addressed by the at least one AI / ML model, a modeiFramework parameter indicating a model framework corresponding to the at least one AI / ML model, a learningMechanism parameter indicating a learning mechanism of the at least one AI / ML model, and a modelTrainable parameter indicating whether the at least one AI / ML model is trainable.

6. The apparatus as claimed in claim 5, wherein the modelResourceRequirement parameter includes at least one of a Central Processing Unit (CPU) parameter indicating a number of millicores or resource measuring unit required to deploy the at least one AI / ML model anda memory parameter indicating a minimum amount of memory required to deploy the at least one AI / ML model.

7. The apparatus as claimed in claim 5, wherein the modelPerformance parameter includes at least one of an inferenceOutputName parameter indicating a name of an inference output data corresponding to a dataName provided in the outputdatainformation attribute, a performanceMetricName parameter indicating a name of a performance metric used to measure the performance of the at least one AI / ML model, a performanceMetricValue parameter indicating a value of the performance metric, a performanceScore parameter indicating the performance score of the at least one AI / ML model, and a decisionConfidence Score parameter indicating a decision level corresponding to a decision generated by the at least one AI / ML model.

8. The apparatus as claimed in claim 5, wherein the modelComplexity parameter includes at least one of a complexityMetricName parameter indicating a name of a complexity metric used to measure the complexity of the at least one AI / ML model, and a complexityMetricValue parameter indicating a value of the complexity metric.

9. The apparatus as claimed in claim 5, wherein the ModelUsage parameter includes at least one of a TIMESERIES FORECASTING parameter indicating time series data used for a forecast function of the at least one AI / ML model, a CLASSIFICATION parameterindicating one or more categories corresponding to the input data, and a RANKING parameter indicating a plurality of rank values to rank the input data.

10. The apparatus as claimed in claim 5, wherein the modeiFramework parameter includes at least one of a PYTORCH indicator, a TENSORFLOW indicator, a XGBOOST indicator, a SKLEARN indicator, a LIGHTGBM indicator, and a framework indicator.

11. The apparatus as claimed in claim 5, wherein the LearningMechanism parameter includes at least one of a SUPERVISED indicator, an UNSUPERVISED indicator, a SEML SPER VISED indicator, and a REINFORCEMENT indicator.

12. The apparatus as claimed in claim 1, wherein the apparatus corresponds to one of a nonReal Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) and a Service Management and Orchestration (SMO).

13. A method comprising: receiving, by one of a non-Real Time (RT) Radio Access Network (RAN) Intelligent Controller (RIC) and a Service Management and Orchestration (SMO) (SMOF), a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface, from at least one of a producer Non-Real Time (RT) Radio access Network (RAN) Intelligent Controller (RIC) Application (rApp) and aService Management and Orchestration Function (SMOF), wherein the registration request includes AI / ML model information corresponding to the at least one AI / ML model, wherein the AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute, and wherein each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schema parameter and a dataDef parameter for corresponding input data and output data; and storing the received registration request and the AI / ML model information.

14. The method as claimed in claim 13, further comprising: receiving a discovery request from at least one consumer rApp, wherein the discovery request is associated with the at least one AI / Model; and transmitting the AI / ML model information corresponding to the at least one AI / ML model in response to receiving the discovery request.

15. The method as claimed in claim 13, wherein the data schema parameter indicates at least one of an encoding format of the corresponding input data and output data, a data type of the corresponding input data and output data, and a range of the corresponding input data and output data; and wherein the dataDef parameter indicates data names and descriptions of the corresponding input data and output data.

16. The method as claimed in claim 13, wherein the metadata attribute indicates a modelResourceRequirement parameter indicating hardware resource details of the at least one AI / ML model, a modelPerformance parameter indicating at least a performance score of the at least one AI / ML model, a modelComplexity parameter indicating a complexity level of the at least one AI / ML model, a modelUsage parameter indicating a type of problem addressed by the at least one AI / ML model, a modeiFramework parameter indicating a model framework corresponding to the at least one AI / ML model, a learningMechanism parameter indicating a learning mechanism of the at least one AI / ML model, and a modelTrainable parameter indicating whether the at least one AI / ML model is trainable; and wherein the modelResourceRequirement parameter includes at least one of a Central Processing Unit (CPU) parameter indicating a number of millicores or resource measuring unit required to deploy the at least one AI / ML model and a memory parameter indicating a minimum amount of memory required to deploy the at least one AI / ML model.

17. The method as claimed in claim 16, wherein the modelPerformance parameter includes at least one of an inferenceOutputName parameter indicating a name of an inference output data corresponding to a dataName provided in the outputdatainformation attribute, a performanceMetricName parameter indicating a name of a performance metric used to measure the performance of the at least one AI / ML model, a performanceMetricValue parameter indicating a value of the performance metric, a performanceScore parameter indicating the performance score of the at least one AI / ML model, and adecisionConfidenceScore parameter indicating a decision level corresponding to a decision generated by the at least one AI / ML model.

18. The method as claimed in claim 16, wherein the modelComplexity parameter includes at least one of a complexityMetricName parameter indicating a name of a complexity metric used to measure the complexity of the at least one AI / ML model, and a complexityMetric Value parameter indicating a value of the complexity metric; and wherein the ModelUsage parameter includes at least one of a TIMESERIES FORECASTING parameter indicating time series data used for a forecast function of the at least one AI / ML model, a CLASSIFICATION parameter indicating one or more categories corresponding to the input data, and a RANKING parameter indicating a plurality of rank values to rank the input data.

19. The method as claimed in claim 16, wherein the modeiFramework parameter includes at least one of a PYTORCH indicator, a TENSORFLOW indicator, a XGBOOST indicator, a SKLEARN indicator, a LIGHTGBM indicator, and a framework indicator; and wherein the LearningMechanism parameter includes at least one of a SUPERVISED indicator, an UNSUPERVISED indicator, a SEMLSPERVISED indicator, and a REINFORCEMENT indicator.

0. A non-transitory computer-readable medium storing instructions, the instructions comprising: one or more instructions that, when executed by an apparatus, the apparatus comprising one or more processors, cause the one or more processors to: receive, from at least one of a producer Non-Real Time (RT) Radio access Network (RAN) Intelligent Controller (RIC) Application (rApp) and a Service Management and Orchestration Function (SMOF), a registration request to register at least one Artificial Intelligence / Machine Learning (AI / ML) model over an R1 interface, wherein the registration request includes AI / ML model information corresponding to the at least one AI / ML model, wherein the AI / ML model information includes at least one of an inputdatainformation attribute, an outputdatainformation attribute, and a metadata attribute, and wherein each of the inputdatainformation attribute and the outputdatainformation attribute includes a data schema parameter and a dataDef parameter for corresponding input data and output data; and store the received registration request and the AI / ML model information.

Images