Patent Application
20240112087
The present disclosure relates AI/ML operation.
In Rel.17, a study item started in 3GPP related to introducing AI/ML intelligence in the NG-RAN architecture. The outcome of the study was captured in TR 37.817. In the TR, a functional framework for RAN Intelligence is provided, shown in
RAN network nodes are expected to host/support, in the future, a wide variety of Machine Learning (ML) based algorithms that are expected to provide inference (Output) to one or more consumers of the inference (i.e. Actor), as shown in
Inference information output by an ML based algorithm (the “output” in
Note that the above Model Framework described in
Generally speaking, statistical inference may have the following exemplary characteristics:
Furthermore the ability of an ML algorithm to provide an inference with a certain characteristics may be impacted by the presence/absence of some input information that may hamper the estimate. For example, lack of traffic in a given time window may affect the ability to predict load on given cell. As another example, lack of input from a neighboring node, e.g. related to UE trajectory may adversely impact making reliable handover decisions.
Also, the ability of an ML algorithm to provide an inference with a certain characteristics may be impacted by a change in the amount of processing resources required or being available for performing the inference.
There may be further potential factors that may affect the ability of an ML algorithm to provide an inference with a certain characteristics than the factors described in the examples above.
The gNB may be divided into two physical entities named CU (Centralized Unit) and DU (Distributed Unit). CU provides support for the higher layers of the protocol stack such as SDAP, PDCP and RRC, while DU provides support for the lower layers of the protocol stack such as RLC, MAC and Physical layer. One CU may control plural DUs.
It is an object of the present invention to improve the prior art.
According to a first aspect of the invention, there is provided an apparatus comprising:
The instructions, when executed by the one or more processors, may further cause the apparatus to perform:
The configuration request may request to configure the first machine learning model and a second model of the machine learning models according to the respective requested configuration;
The configuration request may request to configure the first machine learning model according to a first requested configuration and according to a second requested configuration different from the first requested configuration; and the instructions, when executed by the one or more processors, may further cause the apparatus to perform:
The instructions, when executed by the one or more processors, may further cause the apparatus to perform:
For each of the one or more machine learning models, the capability of the respective machine learning model may comprise at least one input that may be used by the respective machine learning model, at least one output that may be provided by the respective machine learning model, and for each of the at least one output, a characteristic of the respective output dependent on the at least one input.
For each of the one or more machine learning models, the respective requested configuration may comprise a selection of one or more inputs among the at least one input that may be used by the respective machine learning model.
The instructions, when executed by the one or more processors, may further cause the apparatus to perform for the at least one of the one or more machine learning models:
The instructions, when executed by the one or more processors, may further cause the apparatus to perform for the at least one of the one or more machine learning models:
According to a second aspect of the invention, there is provided an apparatus comprising:
The instructions, when executed by the one or more processors, may further cause the apparatus to perform:
The instructions, when executed by the one or more processors, may cause the apparatus to perform
The instructions, when executed by the one or more processors, may cause the apparatus to perform
The instructions, when executed by the one or more processors, may further cause the apparatus to perform:
storing, for each of the one or more machine learning models, the identifier of the respective machine learning model and the at least one capability of the respective machine learning model at the second node if the second node receives the support indication for the respective machine learning model.
For each of the one or more machine learning models, the capability of the respective machine learning model may comprise at least one input that may be used by the respective machine learning model, at least one output that may be provided by the respective machine learning model, and for each of the at least one output, a characteristic of the respective output dependent on the at least one input.
For each of the one or more machine learning models, the respective requested configuration may comprise a selection of one or more inputs among the at least one input that may be used by the respective machine learning model.
According to a third aspect of the invention, there is provided a method comprising:
The method may further comprise:
The configuration request may request to configure the first machine learning model and a second model of the machine learning models according to the respective requested configuration;
The configuration request may request to configure the first machine learning model according to a first requested configuration and according to a second requested configuration different from the first requested configuration; and the method may further comprise:
The method may further comprise:
For each of the one or more machine learning models, the capability of the respective machine learning model may comprise at least one input that may be used by the respective machine learning model, at least one output that may be provided by the respective machine learning model, and for each of the at least one output, a characteristic of the respective output dependent on the at least one input.
For each of the one or more machine learning models, the respective requested configuration may comprise a selection of one or more inputs among the at least one input that may be used by the respective machine learning model.
The method may further comprise, for the at least one of the one or more machine learning models:
The method may further comprise, for the at least one of the one or more machine learning models:
According to a fourth aspect of the invention, there is provided a method comprising:
The method may further comprise:
The monitoring may comprise monitoring whether the second node receives, for the first machine learning model and a second machine learning model of the machine learning models, the respective support indication, wherein the second machine learning model is different from the first machine learning model;
The monitoring may comprise monitoring whether the second node receives, for the first machine learning model, the support indication, wherein the support indication indicates that the first node supports the first machine learning model according to a first capability and according to a second capability;
The method may further comprise:
For each of the one or more machine learning models, the capability of the respective machine learning model may comprise at least one input that may be used by the respective machine learning model, at least one output that may be provided by the respective machine learning model, and for each of the at least one output, a characteristic of the respective output dependent on the at least one input.
For each of the one or more machine learning models, the respective requested configuration may comprise a selection of one or more inputs among the at least one input that may be used by the respective machine learning model.
Each of the methods of the third and fourth aspects may be a method of AI/ML operation.
According to a fifth aspect of the invention, there is provided a computer program product comprising a set of instructions which, when executed on an apparatus, is configured to cause the apparatus to carry out the method according to any of the third and fourth aspects. The computer program product may be embodied as a computer-readable medium or directly loadable into a computer.
According to some embodiments of the invention, at least one of the following advantages may be achieved:
It is to be understood that any of the above modifications can be applied singly or in combination to the respective aspects to which they refer, unless they are explicitly stated as excluding alternatives.
Further details, features, objects, and advantages are apparent from the following detailed description of the preferred embodiments of the present invention which is to be taken in conjunction with the appended drawings, wherein:
Herein below, certain embodiments of the present invention are described in detail with reference to the accompanying drawings, wherein the features of the embodiments can be freely combined with each other unless otherwise described. However, it is to be expressly understood that the description of certain embodiments is given by way of example only, and that it is by no way intended to be understood as limiting the invention to the disclosed details.
Moreover, it is to be understood that the apparatus is configured to perform the corresponding method, although in some cases only the apparatus or only the method are described.
The actor(s) of the AI/ML implementation may or may not be co-located in the same network node as the model inference. If they are not collocated, the interfaces X2/Xn between the RAN nodes and the RAN-Core interface NG-AP may have to support information elements (IE's) to carry information between different network nodes.
According to some example embodiments of the invention, the inference from one node could be queried and used by another node (e.g. a neighbouring node) using a RAN interface (e.g., Xn or X2 interface, F1 interface, E1 interface, to name a few). The querying node may or may not be aware of the dynamic factors that affect the ability of the node to provide an inference with a certain characteristics that does the actual inference.
Some example embodiments of the invention may enable a second network node, interested to receive and utilize Model Inference information, to learn the ML capabilities of a ML function hosted by a first node (e.g. neighboring node, also denoted hosting node) producing the Model Inference. If the second network node knows the host node's capabilities, it may be able to subscribe to the ML model (ML-assisted algorithm) in the hosting node with a certain configuration of the ML model. In this way, the second node may receive, from the hosting node, Model Inference having characteristics as desired by the second node. A solution to this technical problem is relevant for both single and multi-vendor scenarios.
Some example embodiments of the invention enable a network node to reliably share or consume inference from another node (e.g. from a neighboring node). This option is particularly relevant if the two nodes are from different vendors because a hosting node of an ML model would typically not want to share the model implementation details to the other (neighboring) nodes. However, even if both nodes are from the same vendor, the second node might not be interested in the model implementation details of the first one, but only in the characteristics of the Model Inference.
A capability of an ML model (“Model Capability”) comprises a combination of required input parameter(s) to the ML model and characteristics of the output from the ML model, dependent on the input parameters (e.g., dependent on the presence or absence of the input parameter(s) and/or dependent on a respective value range of the input parameter(s)). The characteristics of the output comprise e.g. a level of accuracy, and/or a level of reliability of a prediction, and/or the validity time of a prediction etc. It is to be understood that different Models may have different Model Capabilities depending on the input provided to the ML Model. The ML capability may be restricted to those attributes (input parameters and/or output characteristics) the host node want to make the other node(s) make aware of.
If the capability indicated by the host node fits to the requirements of the other node, the latter may subscribe to AI/ML model predictions (inference) of the host node according to the capability. Then, the host node may provide such predictions (inference) to the other node. For example, in some example embodiments, the host node first has to configure the model according to the capability. Here, configuring means to select a combination of input parameters to be used by the ML model for generating an inference. From the capability, the neighbour node knows the characteristics of the output.
Note that the ML model remains hosted by the host and is not transferred to the neighbor node although the host node may configure the ML model according to the subscription receives from the neighbor node.
In some example embodiment, the host node informs the neighbor node(s) on more than one AI/ML model with a respective capability supported by the host node.
In NG-RAN architecture, this information can be sent in the Xn SETUP REQUEST/RESPONSE procedure over Xn interface or in the NG-RAN NODE CONFIGURATION UPDATE/NG-RAN NODE CONFIGURATION UPDATE ACKNOWLEDGE procedure. For example, original supported ML Model capabilities can be exchanged between neighbour NG-RAN nodes during the Xn SETUP procedure. When an ML Model is updated, the new Model Capabilities can be communicated between neighbour nodes using NG-RAN NODE CONFIGURATION UPDATE/NG-RAN NODE CONFIGURATION UPDATE ACKNOWLEDGE procedures. Note that similar information exchange may be supported over other interfaces, e.g., over F1 interface between CU and DU, or over E1 interface between control plane and user plane of the CU, or over the interface between OAM and RAN.
Each ML model is identified by a respective model ID. Providing of one or more Model Capabilities may be performed in a standardized format. Typically, if the neighbour node receives the capabilities of the model(s), it stores them along with the model ID.
An example of a model ID is as follows. The model identifier may identify the function/purpose of the ML-assisted algorithm ML model). It may be described in the ASN.1 and may be named ML-Model-ID. The ML model IDs may be standardized. A model ID may be unique inside the RAN. For example, part of the Model ID may be related to the entity that has trained the ML model. As an example if the ML Model is trained in the OAM, then model ID could include “OAM” or another binary combination mapped to be interpreted as “OAM”. Similarly if the ML Model is trained in the gNB, in the UE, or in case of split architecture in the gNB-CU, gNB-CU-CP, gNB-CU-UP, or the gNB-DU, it can be indicated through the ML Model name.
A model ID could also indicate a use case that the model is targeting. For example, the AI/ML Model may support Load Balancing, Energy Saving, Mobility Enhancements, Positioning features, to mention a few. The model ID could also indicate a feature it is targeting. For example, is the AI/ML model may be a model for UE Trajectory prediction, Load Prediction, Traffic Prediction, Energy Prediction, etc.
As one example, Model IDs may be managed across vendors or across networks in one of the following possibilities:
The capability of the ML model (ML-assisted algorithm) may be expressed as follows. For example, an ML-assisted algorithm that provides a prediction (such as the amount of air interface resources (e.g. physical resource blocks) for a given time window T) may expose its capabilities as a tuple of {List of inputs, List of outputs, Inference characteristics}. An example is explained with reference to Tables 1 to 3:
The Model capability may comprise a list of inputs that may be used by the ML model and a list of outputs that may be provided as Model inference (prediction). An example of such lists is shown in Table 1. Note that Table 1 shows two separate lists of one column each.
In the example of Table 1, the AI/ML model does not provide the output 0-6. For example, the output 0-6 may be an output provided by a different ML Model. Therefore, “(0-6)” is not marked in bold, differently from the other outputs.
In some example embodiments, based on the lists of inputs and outputs for a given ML model (identified by a respective ML model identifier), an “Inference characteristics” matrix informs on the outputs which the ML model may provide dependent on which inputs are available. An example of such an “inference characteristics” matrix is shown in Table 3. To simplify the “inference characteristics table, in Table 2, different input combinations are listed and labelled as “Input combination i”.
In the left column, the IDs (I-1 to I-7) of the input parameters of table 1 are listed. If a field in the other columns comprises “X”, it means that the respective input has to be provided for the respective input combination. E.g., for the Input Combination 5, the inputs I-2, I-3, and I-6 are used, while the inputs I-1, 1-4, I-5, and I-7 are not used, and for the Input Combination 1, all inputs I-1 to I-7 are used.
Finally, Table 3 shows Inference characteristics for each of the input combinations:
The left column lists the input combinations, e.g. combinations 1 to 5 of Table 2. E.g. “Input combination 5” means that the ML model uses the inputs I-2, I-3, and I-6, while it does not use the inputs I-1, I-4, I-5, and I-7, and “Input combination 1” means that all inputs I-1 to I-7 are used.
The other columns of Table 3 indicate the characteristics of the outputs O-1 to O-5 of Table 1 for each of the input combinations. For example, if Input combination 1 is used (i.e. all input parameters I-1 to I-7 are used by the ML model), the predicted resource status (output parameter O-1) will be indicated by 10 levels increments of 10%, and the predicted resource utilization (output parameter O-2) will be indicated with PRB level granularity in a cell, the validity time is up to 5 seconds (output parameter O-3), the confidence level of the prediction is 95 to 98% (output parameter O-4); and the reliability is 95% or higher (output parameter O-5). Table 3 does not comprise a column for the output parameter O-6 because parameter O-6 is not an output parameter for the ML model.
As can be seen from the table, for different input combinations, the Inference characteristics in the Output may correspond to different levels of granularity in time, e.g., validity time may be up to 5 seconds for input combination 1 but up to 1 second for input combination 5). In addition, for different input combinations resource status reporting can be per PRB level, per beam level or per cell level. Different input combinations may result in different confidence of a ML output e.g., if input combination 1 is used (that utilizes all the supported Model Inputs), confidence (inference characteristic O-4) is very high (95-98%) while for input combination 5 (that utilized the least of the inputs) the confidence is the lowest (60-70%).
As may be seen, the host only exposes input combinations and output inference characteristics but does not expose any details of the ML model (algorithm/training data etc.,) or the ML model itself. This way, confidentiality of the model is maintained. Confidentiality may be crucial in particular for inter-vendor scenarios.
Then, the other node may select one of the capabilities (e.g. X) and subscribe to inference from the AI/ML model with characteristics X.
As described with respect to
After some time, the other node may decide that the characteristics of capability Y are useful. Then, it may subscribe at the host to the AI/ML model with configuration Y. This subscription may either replace the former subscription with characteristics X, or the other node may have two subscriptions to the AI/ML model: one with characteristics X and one with characteristics Y. The host may configure the AI/ML model with capability Y if needed and may than provide inference with the desired characteristics to the other node which may then use the inference.
In some example embodiments, the host may inform the other node(s) on more than one AI/ML models with supported one or more capabilities for each of the AI/ML models.
According to some example embodiments, the host may provide (e.g. by a support indication) a list of ML models that it may support on request but which are not ready for immediate inference calculation. For example, the host may provide a list of one or more ML models that are supported, but not kept up-to-date in training. As another example, the ML model has not had a particular data set used for training for a given capability. In this case, if the neighbour node subscribes to the AI/ML model with the given capability, the host may either reject such a request or accept it. For example, it may accept the request if it received a predefined minimum number of such requests for a specific ML model within a predefined time, otherwise, it may reject the request. If the host accepts the request, it has to make the ML model ready for inference calculation, e.g. the host may train the ML model then.
In some example embodiments, the host may decide for each subscription whether or not it accepts the subscription, regardless of whether the respective ML model is ready for immediate inference calculation.
According to some example embodiments, the messages between the nodes are exchanged directly between the nodes, e.g. via Xn or X2 interfaces. They have an advantage over a manual, OAM based approach, where the OAM indicates the RAN capabilities to the other nodes such that any impacts on Xn or X2 interfaces are avoided. This has the following reasons:
The apparatus comprises means for providing 110, means for monitoring 120, and means for configuring 130. The means for providing 110, means for monitoring 120, and means for configuring 130 may be a providing means, monitoring means, and configuring means, respectively. The means for providing 110, means for monitoring 120, and means for configuring 130 may be a provider, monitor, and configurator, respectively. The means for providing 110, means for monitoring 120, and means for configuring 130 may be a providing processor, monitoring processor, and configuring processor, respectively.
The means for providing 110 provides, to a second node (e.g. gNB), a respective support indication for each of one or more ML models (S110). The respective support indication indicates that a first node (e.g. gNB) supports the respective ML model. The first node is different from the second node. For each of the one or more ML models, the respective support indication comprises an identifier of the respective ML model and at least one capability of the respective ML model.
The means for monitoring 120 monitors whether the first node receives, from the second node for at least one of the one or more ML models, a configuration request (S120). The configuration request requests to configure the respective ML model according to a respective requested configuration.
If the first node receives the configuration request for the at least one of the one or more ML models from the second node (S120=yes), the means for configuring 130 configures the at least one of the one or more ML models according to the respective requested configuration (S130).
The apparatus comprises means for monitoring 210, means for determining 220, and means for providing 230. The means for monitoring 210, means for determining 220, and means for providing 230 may be a monitoring means, determining means, and providing means, respectively. The means for monitoring 210, means for determining 220, and means for providing 230 may be a monitor, determiner, and provider, respectively. The means for monitoring 210, means for determining 220, and means for providing 230 may be a monitoring processor, determining processor, and providing processor, respectively.
The means for monitoring 210 monitors whether a second node receives, for each of one or more ML models, a respective support indication (S210). For each of the one or more ML models, the respective support indication indicates that a first node supports the respective ML model. The first node is different from the second node. For each of the one or more ML models, the respective support indication comprises an identifier of the respective ML model and at least one capability of the respective ML model.
If the second node receives the support indication for each of the one or more ML models (S210=yes), the means for determining 220 determines, for at least one ML model of the one or more ML models, a respective requested configuration of the respective ML model. For the at least one ML model of the one or more ML models, the respective requested configuration is based on the at least one capability of the respective ML model.
The means for providing 230 provides a configuration request to the first node (S230). The configuration request requests to configure the at least one ML model according to the respective requested configuration. The configuration request comprises the identifier(s) of the at least one ML model.
Some example embodiments are explained where the nodes are neighboring nodes. However, in general, the nodes may be arbitrarily arranged in space and need not be neighbored.
Some example embodiments are explained with respect to a 5G network. However, the invention is not limited to 5G. It may be used in other communication networks, too, e.g. in previous of forthcoming generations of 3GPP networks such as 4G, 6G, or 7G, etc. It may be used in non-3GPP communication networks, too.
One piece of information may be transmitted in one or plural messages from one entity to another entity. Each of these messages may comprise further (different) pieces of information.
Names of network elements, network functions, protocols, and methods are based on current standards. In other versions or other technologies, the names of these network elements and/or network functions and/or protocols and/or methods may be different, as long as they provide a corresponding functionality. The same applies correspondingly to the terminal.
If not otherwise stated or otherwise made clear from the context, the statement that two entities are different means that they perform different functions. It does not necessarily mean that they are based on different hardware. That is, each of the entities described in the present description may be based on a different hardware, or some or all of the entities may be based on the same hardware. It does not necessarily mean that they are based on different software. That is, each of the entities described in the present description may be based on different software, or some or all of the entities may be based on the same software. Each of the entities described in the present description may be deployed in the cloud.
According to the above description, it should thus be apparent that example embodiments of the present invention provide, for example, a node, such as a base station (e.g. eNB or gNB) or a NWDAF, or a component thereof, an apparatus embodying the same, a method for controlling and/or operating the same, and computer program(s) controlling and/or operating the same as well as mediums carrying such computer program(s) and forming computer program product(s).
Implementations of any of the above described blocks, apparatuses, systems, techniques or methods include, as non-limiting examples, implementations as hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof. Each of the entities described in the present description may be embodied in the cloud.
It is to be understood that what is described above is what is presently considered the preferred example embodiments of the present invention. However, it should be noted that the description of the preferred example embodiments is given by way of example only and that various modifications may be made without departing from the scope of the invention as defined by the appended claims.
The terms “first X” and “second X” include the options that “first X” is the same as “second X” and that “first X” is different from “second X”, unless otherwise specified. As used herein, “at least one of the following: <a list of two or more elements>” and “at least one of <a list of two or more elements>” and similar wording, where the list of two or more elements are joined by “and” or “or”, mean at least any one of the elements, or at least any two or more of the elements, or at least all the elements.
| Number | Date | Country | Kind |
|---|---|---|---|
| 202241055939 | Sep 2022 | IN | national |
