Variable setting method and variable setting device thereof

The variable setting method and device use CTSEM and Bayesian optimization techniques to predict and prevent QoS violations, enhancing network efficiency and user experience by proactively adjusting QoS parameters.

JP2024166014AActive Publication Date: 2024-11-28WISTRON CORP
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
JP2023136145
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-18
Filing Date
2023-08-24
Publication Date
2024-11-28
Estimated Expiration
2043-08-24

AI Technical Summary

Technical Problem

Current methods for quality of service (QoS) management in networks are inefficient and lead to negative user experiences due to slow responses to QoS violations, resulting in disruptions.

Method used

A variable setting method and device that utilize continuous time structural equation modeling (CTSEM) for proactive QoS violation prediction, allowing networks to adjust policies before violations occur by determining policy time causal variables using Bayesian optimization (BO), causal Bayesian optimization (CBO), or dynamic causal Bayesian optimization (DCBO) to optimize QoS parameters.

Benefits of technology

Enables proactive adjustment of QoS parameters to prevent violations, improving network efficiency and user experience by anticipating and mitigating potential QoS issues before they arise.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024166014000001_ABST
    Figure 2024166014000001_ABST
Patent Text Reader

Abstract

To provide a variable setting method and a variable setting device thereof in order to improve efficiency and user experience.SOLUTION: A variable setting method for a variable setting device includes the steps of: receiving an incident prediction notification from a network at a first time point; determining at least one variable at a second time point in response to the incident prediction notification; and outputting at least one variable at a third time point. The incident prediction notification is used to instruct the variable setting device to determine the at least one variable in response to a quality of service (QoS) violation prediction at a fifth time point later than a fourth time point.SELECTED DRAWING: Figure 2
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] 1. Field of the invention The present invention relates to a variable setting method and a variable setting device thereof, and more particularly to a variable setting method and a variable setting device thereof, which can improve efficiency and user experience. [Background technology]

[0002] 2. Description of the Prior Art According to the current 3GPP specifications, after the RAG finds that the quality of service requirements cannot be achieved, the RAN issues a PDU session modification notification to the PCF (via the AMF and SMF) and lets the PCF decide the policy to modify the PDU session, which is too late and any disruption would lead to a negative user experience. Summary of the Invention

[0003] Therefore, an object of the present invention is to provide a variable setting method and a variable setting device thereof to improve efficiency and user experience.

[0004] An embodiment of the present invention discloses a variable setting method for a variable setting device, the method including the steps of receiving an incident prediction notification from a network at a first time point; determining at least one variable at a second time point in response to the incident prediction notification; and outputting at least one variable at a third time point, where the incident prediction notification is used to instruct the variable setting device to determine the at least one variable in response to a predicted violation of quality of service at a fourth time point, the third time point being earlier than the fourth time point.

[0005] An embodiment of the present invention discloses a variable setting device, the device including: a memory circuit configured to store program code; and a processing circuit coupled to the memory circuit, the program code including: receiving an incident prediction notification from a network at a first time point; determining at least one variable at a second time point in response to the incident prediction notification; and outputting the at least one variable at a third time point, where the incident prediction notification is used to instruct the variable setting device to determine the at least one variable in response to a predicted violation of quality of service at a fourth time point, the third time point being earlier than the fourth time point; and the processing circuit configured to execute the program code stored in the memory circuit.

[0006] An embodiment of the present invention is configured such that the network performs QoS violation prediction by comparing at least one temporal causal variable at a fourth time point with at least one variable at a fourth time point, the at least one temporal causal variable being a vector η h (t) Disclose that it constitutes:

[0007]

number

[0008] where A is a qualitative matrix, t0 is the initial time point, I is the identity matrix, and ξ h represents a random vector, B represents a transformation matrix, and z h denotes the time-independent predictor vector, M denotes the coefficient matrix, and x h,u represents the time-dependent predictor vector, u represents the pulse time point, and W h (s) represents the random walk vector, and G represents the lower triangular matrix.

[0009] An embodiment of the present invention discloses that determining at least one variable at a second time point includes maximizing a plurality of posterior probabilities for assigning a plurality of sub-data of the grounding data to a plurality of observation functions and a causal structure to generate a causal graph, where one of the plurality of posterior probabilities is:

[0010]

number

[0011] Here, s t-1 represents the multiple states of multiple causal variables at time t-1, T represents a fifth time point that is earlier than the fourth time point, C represents a causal structure, and f i represents one of the multiple observation functions that corresponds to the i-th causal variable among the multiple causal variables, and w i,t represents one of multiple sub-data at time t that corresponds to the i-th causal variable.

[0012] An embodiment of the present invention discloses a causal structure of a causal graph between a loss function and a plurality of causal variables, where the plurality of causal variables of the causal graph are determined jointly, and the plurality of causal variables including at least one variable are selected from a plurality of independent variables.

[0013] An embodiment of the present invention discloses that the loss function at the fifth time point is a function of a plurality of causal variables at the fifth time point, a plurality of causal variables at at least one time point earlier than the fifth time point (e.g., the sixth time point), or a loss function at at least one time point (e.g., the sixth time point). These and other objectives of the present invention will no doubt become obvious to those of ordinary skill in the art after reading the following detailed description of the preferred embodiment that is illustrated in the various figures and drawings. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a schematic diagram of a variable setting device according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a schematic diagram of a variable setting method according to an embodiment of the present invention. [Diagram 3] FIG. 3 is a sequence diagram of a variable setting method according to an embodiment of the present invention. [Figure 4] FIG. 4 is a schematic diagram of the relationship between the vector formed / constructed by the policy time causal variables at a particular point in time and the loss function according to an embodiment of the present invention. [Diagram 5] FIG. 5 is a schematic diagram of BO for a one-dimensional problem according to an embodiment of the present invention. [Figure 6] FIG. 6 is a schematic diagram of a portion of a causal graph according to an embodiment of the present invention. [Figure 7] FIG. 7 is a schematic diagram of grounding data and a portion of a causal graph according to an embodiment of the present invention. [Figure 8] FIG. 8 is a schematic diagram of causal variables and sub-data according to an embodiment of the present invention. [Figure 9] FIG. 9 is a schematic diagram of BO, CBO, and DCBO. [Figure 10] FIG. 10 is a schematic diagram of a communication system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] 1 is a schematic diagram of a variable setting device 10 according to an embodiment of the present invention. The variable setting device 10 may be, but is not limited to, a network element (e.g., a Policy Charging Function (PCF)). The variable setting device 10 may include a processing circuit 100 and a memory circuit 110. The memory circuit 110 is configured to store a program code 114, and the processing unit 100 may use the memory circuit 110 to read and execute the program code 114.

[0016] 2 is a schematic diagram of a variable setting method 20 according to an embodiment of the present invention. The variable setting method 20 may be compiled into a program code 114 for the variable setting device 10 and may include the following steps:

[0017] Step S200: Start

[0018] Step S202: Receive an incident prediction notification (eg, a failure prediction notification for predicting a quality of service (QoS) violation) from a network (eg, a radio access network (RAN)) at a first time point.

[0019] Step S204: In response to the incident prediction notification, determine at least one variable at a second time point that is later than the first time point.

[0020] Step S206: Output the determined at least one variable (e.g., policy time causal variable) at a third time point after the second time point. The incident prediction notification is configured to instruct (the variable setting device 10) to determine / determine at least one variable to respond to the QoS violation prediction occurring at a fourth time point after the third time point.

[0021] Step S208: End

[0022] In other words, instead of waiting until the network finds that the QoS requirements cannot be satisfied (and sends a notification of PDU session modification), the network can first predict (e.g., perform QoS violation prediction) in step S202 and issue an incident prediction notification (different from the notification of PDU session modification) to the variable setting device 10 to instruct the variable setting device 10 to perform modification of the protocol data unit (PDU) session. In step S204, the variable setting device 10 can determine how to set / configure a policy (i.e., at least one variable or policy time causal variable) for the PDU session. In step S206, the variable setting device 10 sends the (determined / determined) policy, so that the network can modify the QoS parameters or QoS features of the QoS flows of the associated PDU session to avoid the QoS violation predicted by the network in the QoS violation prediction. For example, if a video call fails to satisfy the QoS requirements of the QoS parameters or QoS features, a direct slowdown (e.g., reducing the video or audio bandwidth) may affect the user experience. Therefore, in the present invention, the network can predict whether / when a video call will be unable to satisfy the QoS requirements in the QoS violation prediction before the failure to satisfy the QoS requirements of the video call actually occurs, and accordingly, the variable setting device 10 can predefine / predetermine a policy for modifying the PDU session, thereby avoiding the occurrence of an event where the video call fails to satisfy the QoS requirements of the QoS parameters or QoS features.

[0023] 3 is a sequence diagram of a variable setting method 30 according to an embodiment of the present invention. The variable setting method 30 may include the following steps:

[0024] Step S301: The radio access network 3RAN performs a QoS violation prediction at time t0 and can predict that a QoS violation (e.g., a video call that cannot meet the QoS requirements) may occur at time t5 (e.g., the fourth time).

[0025] In one embodiment, the RAN 3RAN can utilize a model (e.g., a Recurrent Neural Network (RNN), a Long Short-Term Memory (LSTM), a Transformer, an RNN-like model, or other similar discrete-time models) in step S301 to perform predictive analysis of temporal causal variables. However, this type of model is a discrete-time model that can only handle specific time steps (e.g., time steps t, t+Δt, t+2×Δt) as opposed to time quantities: for example, a discrete-time model cannot handle specific points in time:

[0026] (eg, at time t, lim rt→0 t+rt, or lim rt→0 t+2×rt)

[0027] A discrete-time model predicts what is most likely to happen in the future, but not when it will happen -- it cannot predict exactly when it will happen (which makes it difficult to modify a PDU session if QoS requirements cannot be achieved). Also, choosing an appropriate time scale for a discrete-time model is difficult, considering that the time scale of the discrete-time model (e.g., the length of the time step, Δt) should match the time scale of the dynamic random process being modeled.

[0028] In another embodiment, the RAN 3RAN can utilize continuous time structural equation modeling (CTSEM) in step S301 to perform preventive analysis of time causal variables and predict the (precise) time point (e.g., t5) when QoS requirements will not be met, as opposed to the time steps used in discrete time models. (Thus, the present invention can predict the time point and modify PDU sessions in advance to achieve QoS requirements. CTSEM belongs to continuous time operations, as it may include (random / stochastic) differential equations, and captures / extracts time causal relationships to make accurate predictions for better insight and decisions on PDU session modifications.)

[0029] In CTSEM, the time causal variable is the vector η h (t), which is a function of time and can satisfy the following equation:

[0030]

number

[0031] The matrix A is divided into vectors η h It is possible to qualitatively characterize / capture the temporal relationship of (t). The matrix I is the identity matrix. The random vector ξ h is the vector η h It is possible to determine the long-term trend / level of (t) and the distribution ξ h ~N(κ,φ ξ ), where the vector κ represents the continuous time intercept and the matrix φ ξ can represent covariance (e.g., covariance for QoS flows). The matrix B can be expressed as a vector η h(fixed) time-independent predictor vector z on (t) h where the number of rows of matrix B may differ from the number of columns of matrix B. h Although x may typically be a variable that differs between different QoS flows, it may be constant within different QoS flows for the time horizon of interest. h,u can be observed at time u, and only at time u can the vector η h (t), each of which can be treated as affecting x h,u Vector η, formed / described by δ(tu) h The effect of the impulse on (t) can be represented by a matrix M. The vector W h (s) may be independent random walks in continuous time (e.g., Wiener processes), and dW h (s) can be taken as the random error term. The lower triangular matrix G is the vector η h It is possible to express the effect of a change in (t) as Q=GG T The matrix Q that satisfies the above can represent the variance-covariance matrix of a diffusion process in continuous time. h ,z h ,x h,u ,W h (s), time u, or initial time t0 can be obtained / determined / calculated by applying Equation 1 or Equation 2 to data received, transmitted, or stored by the RAN 3RAN. In one embodiment, the vector η h The t in (t) may be the time point t5 when using the substitution method.

[0032] CTSEM is focused on providing an accessible work flow for full information maximum likelihood estimation of continuous-time multivariate autoregressive models (with random intercepts) for both time series and panel data. In one embodiment, the dynamic random process being modeled is one process (i.e., time series and single device) as opposed to multiple processes (i.e., panel data and multiple devices); correspondingly, the vector η h h in (t) may be equal to 1, and there is only one vector η(t). In another embodiment, vectors η(t) with h ranging from 1 to n may correspond to different QoS flows, and thus the n QoS flows are represented by vectors η(t) through η(t), respectively. n (t), where h and n are positive integers.

[0033] In an embodiment, the time causal variable may be, for example, the lowest flow bit rate (LFBR) for uplink (UL) and / or downlink (DL) in case of guaranteed bit rate (GBR) QoS flows, The highest flow bit rate (HFBR) for UL and / or DL ​​in case of GBR QoS flows, Packet loss rate (PLR) for UL and / or DL ​​in case of GBR QoS flows; Session-aggregate bit rate (Session-ABR) in case of non-guaranteed bit rate (non-GBR) QoS flows (at the PDU session level), User equipment aggregate bit rate (UE-ABR) in case of non-GBR QoS flows, Packet delay time (PDT), Packet error rate (PER), or It may be the data burst amount (only for delay-critical GBR resource types). For the h-th QoS flow of a PDU session, we denote the vector η h (t) may be a multidimensional vector composed of LFBR, HFBR, PLR, Session-ABR, UE-ABR, PDT, or PER, and the vector of time causal variables η h (t) (for monitoring QoS violations) is (LFBR h ,HFBR h ,PLR h ,Session-ABR,UE-ABR,PDT h ,PER h ) may also be expressed as

[0034] In one embodiment, in step S301, the RAN 3RAN can compare the relationship (e.g., whether the time causal variable is greater than or less than its corresponding policy time causal variable) between the time causal variables at a certain time point (e.g., t5) and its / their corresponding pre-configured / predetermined thresholds (e.g., policy time causal variables) to monitor QoS violations or perform QoS violation prediction. For example, If the LFBR at time t5 is less than the Guaranteed Flow Bit Rate (GFBR), If the HFBR at time t5 is greater than the maximum flow bit rate (MFBR), If the PLR ​​at time t5 is greater than the maximum packet loss rate (Maximum-PLR), If the UE-ABR at time t5 is greater than the UE aggregate maximum bit rate (UE-AMBR), If the PDT at time t5 is greater than the Packet Delay Budget (PDB), or If the PER at time t5 is greater than the maximum packet error rate (Maximum-PER), then the RAN 3 may determine in the QoS violation prediction that a QoS violation occurs at time t5. That is, over time, the vector η h (t) is “LFBR<GFBR”,“HFBR> If one or more of the conditions "MFBR", "PLR>Maximum-PLR", "Session-ABR>Session-AMBR", "UE-ABR>UE-AMBR", "PDT>PDB", and "PER>Maximum-PER" are true, a QoS violation is monitored / detected in the QoS violation prediction.

[0035] In one embodiment, a RAN 3RAN Central Unit (CU) (e.g., as shown in FIG. 10) or a resource management unit of the CU may perform CTSEM prediction or QoS violation prediction to predict a possible QoS violation, predict whether a QoS violation is likely to occur, or determine a time when a QoS violation is likely to occur in step S301.

[0036] Step S302a: The RAN 3RAN may send a notification regarding the PDU session modification (eg, an incident prediction notification) to the Session Management Function 3SMF (SMF) via an Access and Mobility Management Function (AMF) 3AMF.

[0037] Step S302b: The SMF 3SMF can notify the PCF 3PCF at time t1 (e.g., a first time point) to determine a policy (e.g., at least one variable, policy time causal variable, QoS profile, alternative profile, or Session-AMBR) for modifying the PDU session (after receiving a notification from the RAN 3RAN) (using the incident prediction notification).

[0038] In other words, instead of waiting until the RAN 3RAN finds that the QoS requirements cannot be achieved, the RAN 3RAN issues an incident prediction notification for PDU session modification to the PCF 3PCF, and lets the PCF 3PCF decide a policy for modifying the PDU session, and a new preventive analysis tool for better insight and decision of PDU session modification is proposed. Specifically, the RAN 3RAN can send an incident prediction notification in steps S302a and S302b (corresponding to step S202) to inform / inform / instruct the PCF 3PCF (via the AMF 3AMF or the SMF 3SMF) to determine / decide a policy for modifying the PDU session, the policy being at least for the prediction of a (coming) QoS violation at the (coming) time t5. That is, the RAN 3RAN can indicate that the policy / strategy for modifying the PDU session should be re-determined / re-evaluated before the (predicted) time t5 to ensure the fulfillment of the QoS requirements at least at the time t5.

[0039] In an embodiment, the policy (e.g., at least one variable, or at least one policy time causal variable of the at least one variable) may include, for example, a QoS profile, an alternative profile, or a Session-AMBR. Fifth Generation Mobile Network (5G) QoS Identifier (5QI), Allocation and retention priority (ARP), GFBR for UL and / or DL ​​in case of GBR QoS flows; MFBR for UL and / or DL ​​in case of GBR QoS flows; Maximum PLR for UL and / or DL ​​in case of GBR QoS flows, Delay-critical resource type for GBR QoS flows, Notification control in case of GBR QoS flows, Reflective QoS attribute (RQA) in the case of non-GBR QoS flows. Session-AMBR (at the PDU session level) for Non-GBR QoS flows, The QoS parameters may include resource type (GBR, delay-critical GBR or non-GBR), priority level, PDB, Maximum-PER, averaging window (for GBR and delay-critical GBR resource types only), or maximum data burst size (for delay-critical GBR resource types only). A PDU session may include 1 st to n th QoS flows, and thus the policy may include (TCV1, TCV2, ..., TCV n For the h-th QoS flow of a PDU session, the vector TCV n may include m policy time causal variables (e.g., Session-AMBR), where m is a positive integer. The policy time causal variables of the h-th QoS flow are denoted by a vector TCV n , which is (QP h ),(QP h ,Session-AMBR), or (GFBR h ,MFBR h ,Maximum-PLR h ,UE-AMBR,PDB h ,Maximum-PER h , Session-AMBR), where QP h It is possible to express the QoS profile of the h-th QoS flow.

[0040] Step S303: The PCF 3PCF can determine a policy (in response to the incident prediction notification) at time t2 (e.g., a second time). In one embodiment, the PCF 3PCF can determine the policy by taking into account the inputs it receives and utilizing an algorithm.

[0041] For example, in one embodiment, the PCF 3PCF may receive input from a network data analytics function (NWDAF) (e.g., various network data analytics), input from a unified data repository (UDR) function (e.g., information about subscription (e.g., usage limit), Nudr_DM_Notify, or application specific information), input from an application function (e.g., Application / Service Info), input from an application server, or input from a video streaming server before or during step S303 and determine the policy according to the input. In one embodiment, the usage limit may be, for example, unlimited or may have a specific data limit (e.g., 1 GB limit).

[0042] In one embodiment, the algorithm used by the PCF 3PCF in step S303 may be, for example, Bayesian optimization (BO), causal Bayesian optimization (CBO), or dynamic causal Bayesian optimization (DCBO). The PCF 3PCF (at time t2) may determine / calculate / select an optimal value / outcome of a policy (e.g., at least one policy time causal variable acting as an independent variable) by utilizing BO, CBO, or DCBO (or by applying BO, CBO, or DCBO to at least one data (e.g., data extracted from a signal transmitted, received, or stored by the PCF 3PCF)) in step S303 so as to achieve a minimum value of a loss function at any given time (e.g., t4). That is, in step S303 (corresponding to step S204), the PCF 3PCF calculates an optimal policy time causal variable (e.g., vector TCV) for one particular time point (e.g., t4). h ) can be determined. The loss function L() (which can serve as an objective function) can be related to the difference between the time causal variables at a particular time point (e.g., t4) and the policy time causal variables; for example, the loss function L() can satisfy the following equation:

[0043]

number

[0044] We use BO as an example. BO is a black-box optimization algorithm for solving extremum problems of functions, where the representation of the function is unknown. For example, L(TCV1,...,TCV n ) = uef(TCV1, ,TCV n ) where L() represents the loss function of the model and uef() represents a function whose expression is unknown. For example, FIG. 4 illustrates a vector TCV1-TCV formed / constructed by policy time causal variables at a particular time (e.g., t4) according to one embodiment of the present invention. n and the loss function L(). However, FIG. 4 shows the relationship between the loss function L() and the vector TCV1-TCV n This is merely an example to show that there is a relationship between the loss function L() and the vector TCV1-TCV, which is formed / constructed by the policy time causal variables. n The expression for the relationship function uef() between is unknown in most cases. The coordinates of one single point on the horizontal axis of Figure 4 may correspond to a point in a multidimensional space, which is represented by the vector TCV1-TCV n is formed / constructed by policy time causal variables.

[0045] Since the expression of the relationship function uef() is unknown, BO can roughly fit the relationship function uef() by using partial / finite sampling points and leverage information of previous sampling points to determine the next sampling point to find the extreme point. For example, FIG. 5 is a schematic diagram of BO for a one-dimensional problem according to an embodiment of the present invention, where the thick solid line indicates the estimated function value of the loss function L(), each of the black dotted points P1 to P5 represents the found sampling point, the area surrounded by two dotted lines represents the fluctuation range of the loss function L() at each point (centered on the average value and proportional to the standard deviation), and the thin solid line represents the acquisition function. The concept of BO is to firstly select a set of initial candidate solutions (e.g., vectors TCV1-TCV nand the loss function L() corresponding to the black-filled point P1), then determine / search for the next sampling point (e.g., the black-filled point P2) that may have an extreme value based on the set of initial candidate solutions (e.g., by estimating the mean and variance of the true loss function based on the function values ​​of the found sampling points (e.g., the loss function corresponding to the black-filled point P1), iteratively search for the next sampling points (e.g., the black-filled points P3-P5) that may have an extreme value until the iterations are terminated, and add all the searched sampling points (e.g., the black-filled points P1-P5) to the set of candidate solutions. Finally, a (global) extreme value is found / selected from the sampling points of the set of candidate solutions (found since the beginning) to serve as the extremum of the loss function L(), and then (e.g., the optimal policy time causal variables Session-AMBR or QP1-QP n Optimal policy that minimizes the loss function L() (such as the minimum loss function depending on the time causal variables Session-AMBR or QP1-QP n (e.g., vector TCVh containing policy time causal variables GFBRh, MFBRh, Maximum-PLRh, PDBh, or Maximum-PERh) is a solution to the problem (e.g., vector TCV1-TCV n The policy for a particular time point (e.g., t4) is then returned as the loss function L() corresponding to the policy time causal variables and the black point P5, thereby discovering / determining the policy for a particular time point (e.g., t4) in step S303.

[0046] According to the results of the Gaussian process regression, an acquisition function (used to measure the degree to which each point of the loss function is worth investigating) may be constructed to find the (relative) extreme points of the acquisition function to determine the next sampling point of the loss function. The acquisition function represented by the thin solid line in FIG. 5 may be constructed according to the mean and variance; that is, the acquisition function can be a function of the mean and variance. An estimate of the likelihood that a point (e.g., one of the solid black points P1-P5) is an extreme point of the loss function reflects the degree to which the point is worth exploring. A relative extreme point of the acquisition function may correspond to the next sampling point of the loss function: for example, point P1 represented by the square box in FIG. 5. * 5 is the maximum point of the acquisition function, (corresponding vector TCV1-TCV n It is possible that the next sampling point of the loss function (i.e., the black point P5) corresponds to the next sampling point of the loss function (according to the above). The acquisition function may be, for example, a knowledge gradient (KG), an entropy search (ES), or a predictive entropy search (PES).

[0047] The algorithm of the present invention can use Gaussian process regression to predict the probability distribution of the function value of the loss function L() at any point based on the function value of the loss function L() at a set of sampling points. Gaussian process regression can be extended to observations with independent normally distributed noise of known variance. Since the variance may be unknown, it is possible to assume that the noise is due to common variance and that the noise includes the variance as a hyperparameter. The present invention uses the posterior mean of the noisy Gaussian process, which is the drift value, rather than the noise of the signal-to-interference-and-noise ratio (SINR). In one embodiment, environmental factors such as temperature and humidity or measurement errors (e.g., measurement errors of LFBR) can cause the drift value of the loss function with respect to some policy time causal variables. In other words, the selected sampling point (e.g., the black point P5) may not select / correspond to the desired / expected extreme value of the functional relationship uef(), but may select / correspond to a relatively optimized extreme value that is closer to the desired / expected extreme value of the functional relationship uef() (i.e., an extreme value of the functional relationship uef() disturbed by noise).

[0048] In one embodiment, (policy time causal variables QP1-QP n ,Session-AMBR,and independent variable 01-O q There may be many independent variables considered by the algorithm of the present invention (in addition to TCV1-TCV2). As the spatial dimension increases, the performance of BO may degrade exponentially. Therefore, in step S303, the algorithm of the present invention can be extended to CBO. In other words, the present invention considers the loss function L() as a function of the vector TCV1-TCV2. n and / or other independent variables, the loss function L(), the vector TCV1-TCV n and / or other independent variables. CBO can perform optimization only on causal variables that are directly related to the loss function L().

[0049] For example, Figure 6 is a schematic diagram of a portion of a causal graph CG1 according to an embodiment of the present invention. In Figure 6, (a) and (b) respectively show two possibilities of the causal graph CG1. Vectors TCV1-TCV function as causal variables. n Policy time causal variables, independent variables O1-O q (Optional and may be omitted) The loss function L() constructs the causal graph CG1 used by the causal model, where q is a positive integer. As shown in Figure 6, the causal graph CG1 shows the causal relationships between the causal variables. The vector TCV1-TCV of the causal graph CG1 n Independent variables 01-0 q Since the policy-time causal variables directly point or affect the loss function L(), the causal intrinsic dimension found by the CBO may be n×m+q (or n×m), and there are n×m+q independent variables (or n×m independent variables) that are selected / searched for optimization and fed into the loss function L() of the CBO. Thus, the values / outcomes of n×m+q causal variables (or n×m causal variables) that can minimize the loss function L() are calculated / discovered. In other words, the causal graph CG1 can significantly improve the ability to reason about optimal decision-making strategies, thereby reducing the optimization cost and avoiding suboptimal solutions. Furthermore, the causal intrinsic dimension of the CBO can be calculated / discovered by the independent variables 01-O that are causes / parents of the loss function L(). q and vector TCV1-TCV n The policy time is given by the number of causal variables (e.g., n×m or n×m+q), which is the independent variables 01-O q and vector TCV1-TCV n rather than the number of causal variables (not shown) that are responsible for the difference, thereby improving our ability to reason about optimal decision-making strategies.

[0050] In one embodiment, the causal model for optimization is selected based on maximum a posterior (MAP) and point estimates, and the causal graph of the loss function, vector TCV1-TCV nIt is possible to obtain / derive the policy time causal variables of the causal graph of the causal model (e.g., cv (i-1) ,cv i ,cv (j-1) ,cv j Since the causal variables (e.g., which causal variables are selected, the number of causal variables, which attributes the causal variables have, or the number of attributes of the causal variables) and the causal structure (e.g., C) of the causal graph (e.g., how the causal variables or their attributes are connected to each other) are obtained / determined / discovered together (i.e., because the causal variables are learned with / together with the causal structure), the causal variables and the causal structure may interact / influence / constrain each other. Determining the causal variables and the causal structure simultaneously / in parallel can avoid problems caused by first determining the causal variables and then determining the causal structure (e.g., the causal model can distinguish between (a) and (b) in FIG. 7 and vice versa, thereby avoiding the problem of indistinguishability).

[0051] For example, FIG. 7 is a schematic diagram of a portion of grounded data gD and causal graph CG2 according to an embodiment of the present invention. In FIG. 7, (a) and (b) respectively show two possibilities of grounded data gD and causal graph CG2. The causal graph CG2 may function as the causal graph CG1. In FIG. 7, the causal structure of the causal graph CG2 is a set of causal variables (e.g., cv (i-1) ,cv i ,cv (j-1) ,cv j ) it is possible to show the relationship between the observation function f (i-1) ,f i ,f (j-1) ,f j Using the sub-data w of the grounding data gD, (i-1) ,w i ,w (j-1) ,w j , the causal variable cv (i-1) ,cv i ,cv (j-1) ,cvj and the causal variable cv (i-1) ,cv i ,cv (j-1) ,cv j and the subdata w of the grounding data gD (i-1) ,w i ,w (j-1) ,w j where i,j are positive integers. The mapping here is done by using the corresponding sub-data (e.g., w (i-1) ,w i ,w (j-1) ,w j ) (e.g., the boxed area in Figure 7). For example, the causal variables cv (i-1) If corresponds to the policy time causal variables in vector TCV1, then the sub-data w (i-1) is the causal variable cv (i-1) (attribute of) (i-1) is associated with all data on policy-time causal variables in the vector TCV1). In other words, the Bayesian probability mechanism using the Bayesian network uses the number of causal variables (i.e., the vector TCV1-TCV n Which policy time causal variable or independent variable 01-0 q The causal variables gD may be combined with the causal structures (e.g., which of the causal variables gD are selected / explored), the states of the causal variables (e.g., including causal variables cv(i-1), cvi, cv(j-1), and cvj), the causal structure of the causal variables, or the observation functions for the causal variables (e.g., including observation functions f(i-1), fi, f(j-1), and fj) to draw relevant joint inferences to explain / interpret the grounded data gD, thereby constructing the causal graph CG2.

[0052] In one embodiment, the sub-data w of the grounding data gD i Let f be the observation function iThe posterior probability P(fi,C|wi) of causal structure C in the causal graph CG is calculated by the sub-data w i Based on (i.e., vector TCV1-TCV n The policy causal variables, loss function L(), and independent variables O1-O q By extracting the policy time causal variables from the grounded data gD, we obtain the corresponding causal structure C and the corresponding causal variables cv i It is possible to maximize to determine / derive / select

[0053] In one embodiment, the posterior probability P(f i ,C|wi,Int) can satisfy the following equation according to Bayes' theorem:

[0054]

number

[0055] where f i denotes the corresponding observation function, C denotes the corresponding causal structure, and w i denotes a part (e.g., sub-data) of the grounding data gD, and Int denotes an intervention (which may be omitted). In one embodiment, the posterior probability P(fi,C|w i )teeth,

[0056]

number

[0057] can be proportional to, where w i,t is the causal variable cv at time t i represents the sub-data corresponding to, C represents the causal structure, and s t-1 represents the state at time t-1 (of all causal variables), T represents the current / now time, and γ may be a real number such as, but not limited to, 0 or 0.5. i |f i, C) may be expressed as follows:

[0058]

number

[0059] In one embodiment, P(wi,t|st-1,C,fi) may be expressed as:

[0060]

number

[0061] Here, s i,t is the causal variable cv at time t i indicates the state (e.g., the corresponding value) of s t denotes the state at time t (of all causal variables), and Ncv denotes the state of all causal variables (e.g., causal variables cv (i-1) ,cv i ,cv (j-1) ,cv j Ncv is a positive integer, and R st is the causal variable cv in the grounded data gD. i Status i Subdata that matches w i In one embodiment, the present invention provides a data amount R st The causal variable cv that minimizes i It is possible to select / discover frequently used data (e.g. sub-data w i ) can be cut into smaller pieces than those that are rarely used.

[0062] In one embodiment, the grounding data gD may be obtained or derived from the space of all observable samples and may therefore be referred to as observed data. In one embodiment, the grounding data gD may include or relate to data extracted from signals (e.g., inputs from the NWDAF) transmitted or received by the PCF 3PCF at any time and in any manner.

[0063] As shown in Figure 7, each causal variable (e.g., cv i ) is the observation function (e.g., f i In one embodiment, the high-dimensional environmental variables (e.g., grounded data gD) can be converted into low-dimensional state attributes (e.g., causal variables cv i A causal semantic generative (CSG) model is used to predict the state or attributes of an observation function (e.g., f i ) can be calculated / derived. Causal variables (e.g., independent variables 01-O acting as causal variables) q ,The vector TCV1-TCV acts as a causal variable. n Policy time causal variables, or causal variables cv (i-1) ,cv i ,cv (j-1) ,cv j ) are manually defined, each causal variable (e.g., cv i ) maps the causal variables to the corresponding subdata (e.g., w i ) (e.g., the boxed area in Figure 7). (e.g., the causal variables are defined manually (non-automatically and individually) by domain experts; alternatively, the causal variables are defined automatically using a program that uses rules written manually by domain experts.) This allows the system to select sub-data (e.g., the sub-data corresponding to the boxed area in Figure 7) based on the definition of specific causal variables from the domain experts. (i-1) ,w i ,w (j-1) ,wj ) is determined. Furthermore, the CSG model converts the variation factor into a causal variable (e.g., cv i ) and can avoid regarding semantic factors as causal variables (e.g., cv i ) can be accurately determined to be the cause of the observed data. In one embodiment, the variance and semantic factors may constitute / be attributed to the observed data. In one embodiment, the CSG model is primarily based on the causal invariance principle and involves variational Bayes.

[0064] In one embodiment, the observation function f i s i,t =f i (w i,t ) can be satisfied. In one embodiment, the observation function f i may be implemented using a multivariate Gaussian distribution: for example, the observation function f i It is possible to satisfy the following equation:

[0065]

number

[0066] Alternatively, the observation function f i may be related to the following equation:

[0067]

number

[0068] Here, z denotes the subdata in the grounded data gD (those that do not contribute to the causal variables cvi), and μ wi ,μ ν denotes a fixed mean as a 0 vector, and Σ is, for example, Σ=LL TThe matrix L may be a lower triangular matrix with positive diagonal elements, and may be parameterized, for example, to satisfy:

[0069]

number

[0070] matrix L wiwi ,L zz can be taken to be a smaller lower triangular matrix. zwi can be any arbitrary matrix. Each matrix L wiwi ,L zz can be parameterized by a sum of positive diagonal elements (guaranteed via the exponential map) and a lower triangular matrix (no positive diagonal elements). In one embodiment, the causal variables (e.g., cv i ) and subdata (e.g., w i ) may be unknown, the causal variables can be predicted / inferred from the sub-data using the CSG model. For example, FIG. 8 shows the causal variables cv y And sub data w x Schematic diagram of the CSG model, where (a), (b), (c), and (d) respectively show the structurability of the CSG model, sc represents the semantic factor, v represents the variance factor, and the solid arrows represent the causal mechanism p(w x |sc,v) and p(cv y |v sc ), and the dashed arrow represents the inference model q(sc,v|w x In Fig. 8(a), the solid undirected line between semantic factor sc and variability factor v represents the domain-specific prior probability p(sc,v). Compared with the solid undirected line between semantic factor sc and variability factor v in Fig. 8(a), Fig. 8(b) shows the independent prior probability p ⊥We introduce (sc,v):=p(sc)p(v) to reflect the intervention and improve the out-of-distribution generalization performance. Compared with the solid undirected line between the semantic factor sc and the variability factor v in Fig. 8(a), Fig. 8(c) shows the prior probability p ~ (sc,v) is introduced to reflect the intervention of the causal invariance principle and to leverage unsupervised data. In one embodiment, the present invention estimates the CSG model p:= <p(sc,v),p(w x |sc,v),p(cv y We fit the |sc> to the subdata and perform calculations using variational inference and the evidence lower bound (ELBO), using Monte Carlo, and estimating the expectation value after applying reparameterization tricks.

[0071] Vector TCV1-TCV n The policy time causal variables may be functions of time (i.e., their values ​​at different times are different), and the CBO computes the output causal variables (e.g., the loss function L()) and the input causal variables (e.g., the vector TCV1-TCV n policy time causal variables and / or independent variables O1-O q) as invariant independent variables and ignore the existence of time evolution in both output and input causal variables (i.e., ignore whether the output and input causal variables change over time), thus breaking the time-dependent structure that exists between the causal variables. Ignoring time may greatly simplify the problem, but it prevents the identification of optimal interventions at all time points and may result in suboptimal solutions instead of providing the current optimal solution at any time point (especially in non-stationary scenarios). Therefore, the present invention can be extended to DCBO in step S303, which provides / explains causal relationships between causal variables, which can evolve / change over time, thus facilitating scenarios where all causal effects in the causal graph change over time.

[0072] For example, Figure 9 is a schematic diagram of BO, CBO, and DCBO. In Figure 9(a) or (b), X1 to X3 represent a causal variable (e.g., X) at three different time points TT1, TT2, and TT3. Y1 to Y3 represent another causal variable (e.g., Y) at time points TT1, TT2, and TT3. Z1 to Z3 represent another causal variable (e.g., Z) at time points TT1, TT2, and TT3. However, the present invention is not limited thereto and can be extended to more time points or more causal variables. DCBO combines BO and CBO to explain the causal relationships between causal variables, and the causal relationships can evolve over time. For example, in DCBO, as shown in FIG. 9, the causal variable Y1 at time TT1 is a function of the causal variable Z1 at time TT1; therefore, the extreme values ​​of the causal variable Y1 (which may represent or correspond to the loss function L() at time TT1) can be found using only the causal variable Z1 that is directly related to the causal variable Y1, and the causal intrinsic dimensionality is 1. Similarly, the causal variable Y2 at time TT2 is a function of the causal variable Z2 at time TT2 and the causal variable Y1 at time TT1; therefore, the extreme values ​​of the causal variable Y2 (which may represent or correspond to the loss function L() at time TT2) can be found using only the causal variables Y1 and Z2 that are directly related to the causal variable Y2, and the causal intrinsic dimensionality is 2. Similarly, the causal variable Y3 at time TT3 (e.g., t4) is a function of the causal variable Z3 at time TT3 and the causal variable Y2 at time TT2; therefore, the extreme values ​​of the causal variable Y3 (which may represent or correspond to the loss function L() at time TT3) can be found using only the causal variables Y2 and Z3 that are directly related to the causal variable Y3, and the causal intrinsic dimensionality is 2. In other words, since the causal variable that functions as a dependent variable at a time is a function of the causal variables (which function as either dependent or independent variables) at the previous time, the extreme values ​​of the former (the causal variable that functions as a dependent variable at a time) can be found using only the latter that is directly related to the former (i.e., the causal variable that functions as a dependent or independent variable at the previous time).

[0073] In step S303, the optimal / best alternative for the QoS parameters or QoS features (e.g., GFBR, MFBR, Maximum-PLR, PDB, or Maximum-PER) calculated / discovered by the algorithm (e.g., DCBO) is calculated based on the policy time causal variable QP h Step S303 may avoid waiting until the network discovers that the QoS requirements cannot be achieved; instead, corrective action may be taken before the network actually discovers that the QoS requirements cannot be met, i.e., the QoS parameters or features (e.g., GFBR, PDB, or Maximum-PER) used for the alternative profile may not be considered / used / needed.

[0074] Step S304a: The PCF 3PCF may send the determined policy (e.g., the policy determined for time t4) at time t3 (e.g., the third time) to the SMF 3SMF (e.g., by sending an Npcf_SMPolicyControl_UpdateNotify request). Step S304a may correspond to step S206.

[0075] Step S304b: The SMF 3SMF can accordingly send a message (eg, Namf_Communication_N1N2MessageTransfer) to the AMF 3AMF.

[0076] Step S304c: The AMF 3AMF can accordingly send a message (e.g., an N2 message or an N2 session request) to the RAN 3RAN to modify the QoS parameters or QoS characteristics of the QoS flows in the PDU session in question.

[0077] Step S305: The RAN 3 determines, at time t4 (e.g., the fifth time point), the QoS parameters (e.g., QoS parameters of the QoS profile) or QoS features (e.g., PDB, Maximum-PER, or other QoS features) of the QoS flows in the PDU session in question in accordance with the (determined) policy (e.g., policy time causal variables QP1-QP2 for time t4). n , or Session-AMBR) to control the flow rate, traffic, or speed of the data flow to avoid a predicted QoS violation due to the QoS violation prediction at time t5.

[0078] In one embodiment, the storage circuitry configured to store image data or instructions may be, but is not limited to, a read-only memory (ROM), a flash memory, a random access memory (RAM), a hard disk, a non-volatile storage device, a non-transitory computer readable medium, etc. In one embodiment, the processing circuitry configured to execute the instructions (stored in the storage circuitry) may be, but is not limited to, a central processing unit (CPU), a microprocessor, or an application specific integrated circuit (ASIC).

[0079] The use of ordinal terms such as "first" and "second" does not, in and of itself, imply any priority, precedence, or order of one element relative to another; a chronological order in which the operations of a method are performed; or a requirement that all elements be present at the same time; and these terms are merely used as labels to distinguish one element having a certain name from another element having the same name. The technical features described in the following embodiments can be mixed or combined in various ways unless there is a contradiction between them.

[0080] In summary, the network can first predict whether a QoS violation will occur, and before the predicted QoS violation occurs, issue an incident prediction notification to the variable setting device to instruct the variable setting device to determine a policy (i.e., how to set / configure the QoS parameters or QoS features of the QoS flow), which enables the network to set or modify the QoS parameters or QoS features of the QoS flow according to the policy.

[0081] Those skilled in the art will readily appreciate that numerous modifications and variations of the devices and methods may be made while retaining the teachings of the present invention. Accordingly, the above disclosure should be construed as limited only by the metes and bounds of the appended claims.

Claims

1. 1. A variable setting method for a variable setting device, comprising: receiving a notification of a predicted incident from the network at a first time; determining at least one variable at a second time in response to the incident prediction notification; and outputting the at least one variable at a third time; wherein the incident prediction notification is used to instruct the variable setting device to determine the at least one variable in response to a Quality of Service (QoS) violation prediction at a fourth time point, the third time point being earlier than the fourth time point.

2. 2. The variable setting method of claim 1, wherein the network is configured to predict that at least one QoS requirement cannot be satisfied at the fourth time point according to a continuous-time structural equation modeling that outputs the incident prediction notification.

3. The variable setting method according to claim 1, The at least one time causal variable includes a minimum flow bit rate, a maximum flow bit rate, a packet loss rate, a session aggregate bit rate, a user equipment aggregate bit rate, a packet delay time, a packet error rate, or a data burst amount; and A variable setting method, wherein the at least one variable includes a guaranteed flow bit rate, a maximum flow bit rate, a maximum flow bit rate, a maximum packet loss rate, a session aggregate maximum bit rate, a user device aggregate maximum bit rate, a packet delay budget, a maximum packet error rate, or a maximum data burst amount.

4. 2. The variable setting method of claim 1, wherein the network is configured to modify at least one QoS parameter or at least one QoS feature according to the at least one variable at a fifth time point earlier than the fourth time point, and the at least one variable is mapped to a minimum value of a loss function at the fifth time point.

5. 1. A variable setting device including: a memory circuit configured to store program code; and a processing circuit coupled to the memory circuit, the program code comprising: receiving a predictive incident notification from the network at a first time; determining at least one variable at a second time in response to the incident prediction notification; and outputting the at least one variable at a third time; the incident prediction notification is used to instruct the variable setting device to determine the at least one variable in response to a Quality of Service (QoS) violation prediction at a fourth time point, the third time point being earlier than the fourth time point; The processing circuitry is configured to execute the program code stored in the storage circuitry.

Citation Information

Patent Citations

  • Methods and functions of network performance monitoring and service assurance

    US20200044943A1

  • Method for managing qos, relay terminal, PCF network element, SMF network element, and remote terminal

    US20220322135A1