Method and apparatus for ascertaining user plane traffic anomalies or forbidden / malicious traffic
A data analytics server in communication systems uses ML models to analyze user plane traffic, addressing inefficiencies in existing mechanisms by effectively identifying and managing abnormal and malicious traffic patterns, thereby enhancing network performance.
Patent Information
- Application Number
- PCT/IB2026/054596
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-09
- Filing Date
- 2026-05-08
- Publication Date
- 2026-11-12
AI Technical Summary
Existing mechanisms for managing user plane traffic in communication systems are inefficient and lack the use of Artificial Intelligence Machine Learning (AIML) capabilities, particularly in identifying and mitigating abnormal traffic patterns and malicious or forbidden sites, which impact user plane performance.
A data analytics server receives parameters from a consumer node to analyze user plane traffic, generates analytics information or ML models to identify anomalies and malicious or forbidden traffic, and sends responses to the consumer node for appropriate actions.
Enhances the ability to detect and mitigate abnormal user plane traffic, improving network performance by providing efficient analytics and enabling proactive management of malicious or forbidden sites.
Smart Images

Figure IB2026054596_12112026_PF_FP_ABST
Abstract
Description
[0001] METHOD AND APPARATUS FOR ASCERTAINING USER PLANE TRAFFIC ANOMALIES OR FORBIDDEN / MALICIOUS TRAFFIC
[0002] Related Application
[0003] [1] This patent application claims priority from EP provisional patent application no. 25382467.6 entitled “Method and Apparatus for Ascertaining User Plane Traffic Anomalies or Forbidden I Malicious Traffic” filed on May 9, 2025, which is incorporated by reference in its entirety.
[0004] Field of the Disclosure
[0005] [2] This disclosure relates to communication systems, and more particularly to communication systems that assess user plane traffic.
[0006] Background
[0007] [3] 3GPP (Third Generation Partnership Project) has published various TS (Technical Standard) and TR (Technical Report) documents. Some of these documents referenced herein are identified below:
[0008] • 3GPP TS 23.288 entitled “Architecture enhancements for 5G System (5GS) to support network data analytics services” version 19.2.0 (March 2025), hereinafter “3GPP TS 23.288”;
[0009] • 3GPP TR 23.700-04 entitled “Study on Core Network Enhanced Support for Artificial Intelligence (Al) I Machine Learning (ML); Phase 2” version 0.1.0 (April 2025), hereinafter “3GPP TR 23.700-04”; and
[0010] • 3GPP TS 29.564 entitled “5G System; User Plane Function Services; Stage 3” version 19.2.0 (March 2025), hereinafter “3GPP TS 29.564”.
[0011] [4] Referring first to Figure 1, shown is a block diagram of a 5GC (5th Generation Core) reference architecture 100 with policy and charging control framework as defined by 3GPP. The 5GC reference architecture 100 includes:
[0012] • a UDR 101 (Unified Data Repository)
[0013] • an NEF 102 (Network Exposure Function),
[0014] • an NWDAF 103 (Network Data Analytics Function),• an AF 104 (Application Function),
[0015] • a PCF 105 (Policy Control Function),
[0016] • a CHF 106 (Charging Function),
[0017] • an AMF 107 (Access and Mobility Management Function),
[0018] • an SMF 108 (Session Management Function), and
[0019] • a UPF 109 (User Plane Function).
[0020] The relevant architectural aspects for this invention are described below.
[0021] [5] The NWDAF 103 represents operator managed network analytics logical function. The NWDAF 103 is part of the 5GC architecture 100 and uses the mechanisms and interfaces specified for 5GC and 0AM (Operations, Administration, and Maintenance).
[0022] [6] The NWDAF 103 may contain the following logical functions:
[0023] • AnLF (Analytics logical function): A logical function in the NWDAF 103, which performs inference, derives analytics information (i.e. derives statistics and / or predictions based on Analytics Consumer request) and exposes analytics service, e.g. Nnwdaf_AnalyticsSubscription or Nnwdaf_Analyticslnfo.
[0024] • MTLF (Model Training logical function): A logical function in the NWDAF 103, which trains ML (Machine Learning) models and exposes new training services (e.g. providing trained ML model).
[0025] • ADRF (Analytics Data Repository Function) stores the collected data and analytics.
[0026] [7] The NWDAF 103 interacts with different entities for different purposes:
[0027] • Data collection based on event subscription, provided by the AMF 107, the SMF 108, the PCF 105, the UDR 101 via a UDM (Unified Data Management), the AF 104 (directly or via the NEF 102), and the 0AM;
[0028] • Retrieval of information from data repositories (e.g. the UDR 101 via the UDM for subscriber-related information);
[0029] • Retrieval of information about NFs (Network Functions), e.g. an NRF (Network Resource Function) for NF-related information, and an NSSF (Network Slice Selection Function) for slice-related information;
[0030] • On demand provision of analytics to consumers.o Storage in ADRF (Analytics Data Repository Function) for two types of data:
[0031] ■ collected data (e.g., event exposure data).
[0032] ■ analytics reports.
[0033] [8] The AF 104 interacts with the 3GPP Core Network 100, and specifically in the context of this disclosure, allows external parties to use the Exposure APIs (Application Programming Interfaces) offered by the network operator.
[0034] [9] The NEF 102 supports different functionality, and specifically in the context of this disclosure, the NEF 102 supports different Exposure APIs.
[0035]
[0010] The PCF 105 supports a unified policy framework to govern the network behavior. Specifically, the PCF 105 provides PCC (Policy and Charging Control) rules to the PCEF (Policy and Charging Enforcement Function), i.e. the SMF 108 I the UPF 109 that enforces policy and charging decisions according to provisioned PCC rules.
[0036]
[0011] The SMF 108 supports different functionalities, e.g. the SMF 108 receives the PCC rules from the PCF 105 and configures the UPF 109 accordingly.
[0037]
[0012] The UPF 109 supports handling of user plane traffic, including packet inspection, packet routing and forwarding, traffic usage reporting, QoS (Quality of Service) handling.
[0038]
[0013] Regarding 3GPP TR 23.700-04, R19 AIML_CN has focused on analytics for control plane. For R20 AIML_CN Ph2, many companies, including many relevant MNOs (Mobile Network Operators), have proposed to study enhancements on AI / ML related to user plane the UPF who handles user plane data.
[0039]
[0014] Key lssue#2 of 3GPP TR 23.700-04 has been approved for release 20:
[0040] o Kl#2: AI / ML-assisted user plane traffic pattern and behavior analysis to support efficient performance of User Plane.
[0041]
[0015] Use of AI / ML can be studied to provide analytics on user plane traffic patterns and how to use the analytics to improve user plane performance. In this context, it is proposed to study:o How to provide analytics (e.g. enhancing existing in 3GPP TS 23.288 or introducing new) including the aspects of required inputs / outputs to improve performance of user plane.
[0042] o Which consumers can subscribe / request the analytics and how the consumers use the analytics to take actions to improve user plane performance.
[0043] o Whether and how to train ML models and perform the inference for the user plane analytics, considering solutions that minimize the load on Control Plane NFs.
[0044]
[0016] Content filtering and parental control are mechanisms used to manage and restrict access to digital content based on predefined criteria or settings, often to protect users from harmful or inappropriate material.
[0045]
[0017] Content Filtering: Content filtering refers to the process of screening and blocking access to certain content on the internet or within a network. It is typically implemented through software tools or hardware devices that analyze the data being transmitted and determine whether it should be allowed or blocked based on specific criteria. These criteria can include keywords, URLs, file types, or content categories (such as adult content, violence, or hate speech). Content filtering is used in various settings, including schools, workplaces, and homes, to ensure compliance with policies or to enhance security by preventing access to malicious sites.
[0046]
[0018] Parental Control: Parental control is a subset of content filtering specifically designed to help parents manage and restrict their children's access to digital content. Parental control tools offer features such as website blocking, app restrictions, screen time limits, and monitoring of online activity. These tools enable parents to create a safer online environment for their children by preventing access to inappropriate content, managing usage times, and monitoring online interactions. Parental controls can be implemented on individual devices, such as smartphones and tablets, or through network-level solutions that cover all devices connected to a home network.
[0047] Summary of the Disclosure
[0048]
[0019] According to an aspect, there is provided a method for execution by a data analytics server. The method involves the data analytics server receiving, from aconsumer node, a request message including at least one parameter concerning user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden. The method also involves the data analytics server acquiring, from a user plane node, user plane traffic based on the at least one parameter of the request message. The method involves the data analytics server generating, based on the user plane traffic acquired using the at least one parameter, analytics information or ML (Machine Learning) model from which the consumer node can ascertain whether there is any user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden.
[0049]
[0020] Finally, the method involves the data analytics server sending, to the consumer node, a response based on the analytics information or ML model. Based on this response, the consumer node can ascertain whether there is any user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden.
[0050]
[0021] In a first embodiment, the at least one parameter concerns user plane traffic anomalies, and the analytics information or ML model is configured to identify whether there is any user plane traffic anomalies. In some implementations, the at least one parameter comprises at least one of volume and / or rate, traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets, and for each of these parameters a corresponding threshold for the parameter to be an anomaly. In some implementations, the request message further comprises at least one of DNN (Data Network Name), S-NSSAI (Serving Network Slice Selection Assistance Information) and / or area of interest, a list of SUPIs (Subscription Permanent Identifiers), a group of SUPIs or any UE (User Equipment).
[0051]
[0022] In some implementations, the data analytics server acquires the user plane traffic by acquiring at least one of metrics or KPIs (Key Performance Metrics) and / or packet parameters and / or classification related information. In some implementations, the data analytics server acquires the user plane traffic based on the request message by acquiring the user plane traffic whenever any of the corresponding thresholds is surpassed.
[0052]
[0023] In some implementations, the data analytics server generates the analytics information or ML model by generating a list of exception IDs (Identifiers)corresponding to abnormal UP (User Plane) traffic detected, and indicating for which SUPI(s), application identifier(s) and / or server IP (Internet Protocol) addresses to which the abnormal UP traffic applies. In some implementations, the data analytics server generates the analytics information or ML model further by generating an indication if and how abnormal UP traffic has any impact on UPF (User Plane Function) load.
[0053]
[0024] In a second embodiment, the at least one parameter concerns user plane traffic with sites that are malicious and / or forbidden, and the analytics information or ML model is configured to identify whether there is user plane traffic with sites that are malicious and / or forbidden. In some implementations, the at least one parameter comprises at least one of malicious and / or forbidden sites indication, content filtering categories and / or parental control categories. In some implementations, the request message further comprises at least one of vendor ID (Identifier), ML model filter information, target of ML model reporting, ML model target period, time when model is needed, inference input data information, ML model monitoring information, ML model interoperability information, indication of support for multiple ML models, and number of ML models and accuracy level(s) of interest.
[0054]
[0025] In some implementations, the method involves the data analytics server determining whether an existing trained ML model can be used, and / or determining whether triggering further training for the existing trained ML model is needed. In some implementations, upon determining that further training for the existing trained ML model is needed, the data analytics server initiates data collection to retrieve information relative to malicious and / or forbidden sites, and trains the existing trained ML model.
[0055]
[0026] In some implementations, the analytics information or ML model comprises analytics information, and the response includes at least some of the analytics information.
[0056]
[0027] In other implementations, the analytics information or ML model comprises an ML model, and the response includes an ML model identifier and ML model information. In some implementations, the analytics information or ML model further comprises ML model filter information and / or target of ML model reporting. In some implementations, the analytics information or ML model further comprises at leastsome of a validity period, a spatial validity, training input data information, and ML model accuracy Information.
[0057]
[0028] According to another aspect, there is provided a non-transitory computer readable medium having recorded thereon statements and instructions that, when executed by a processor of a data analytics server, configure the data analytics server to implement a method as summarized above.
[0058]
[0029] According to another aspect, there is provided a data analytics server. The data analytics server has a network interface configured to communicate with other network nodes, and data analytics circuitry coupled to the network interface. The data analytics circuitry is configured to receive, from a consumer node, a request message including at least one parameter concerning user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden. The data analytics circuitry is also configured to acquire, from a user plane node, user plane traffic based on the at least one parameter of the request message. The data analytics circuitry is configured to generate, based on the user plane traffic acquired using the at least one parameter, analytics information or ML (Machine Learning) model from which the consumer node can ascertain whether there is any user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden.
[0059]
[0030] Finally, the data analytics circuitry is configured to send, to the consumer node, a response based on the analytics information or ML model. Based on this response, the consumer node can ascertain whether there is any user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden.
[0060]
[0031] In some implementations, the data analytics circuitry is also configured to implement a method as summarized above.
[0061]
[0032] Other aspects and features of the present disclosure will become apparent, to those ordinarily skilled in the art, upon review of the following description of the various embodiments of the disclosure.Brief Description of the Drawings
[0062]
[0033] Embodiments will now be described with reference to the attached drawings in which:
[0063] Figure 1 is a block diagram of a 5G reference architecture with policy and charging control framework;
[0064] Figure 2 is a block diagram of a communication system, in accordance with an embodiment of the disclosure;
[0065] Figure 3 is a sequence drawing of a method of ascertaining user plane traffic anomalies or forbidden I malicious traffic;
[0066] Figure 4 is a sequence drawing of a method of ascertaining user plane traffic anomalies; and
[0067] Figure 5 is a sequence drawing of a method of ascertaining user plane traffic with sites that are malicious and / or forbidden.
[0068] Detailed Description of Embodiments
[0069]
[0034] It should be understood at the outset that although illustrative implementations of one or more embodiments of the present disclosure are provided below, the disclosed systems and / or methods may be implemented using any number of techniques. The disclosure should in no way be limited to the illustrative implementations, drawings, and techniques illustrated below, including the exemplary designs and implementations illustrated and described herein, but may be modified within the scope of the appended claims along with their full scope of equivalents.
[0070] Introduction
[0071]
[0035] Referring first to Figure 2, shown is a block diagram of a communication system 200, in accordance with an embodiment of the disclosure. The communication system 200 has a data analytics server 220 operatively coupled to consumer node 230 and a user plane 210 via at least one network 202. The communication system 200 also has a UE group 204a-c coupled to the network 202. The UE group 204a-c can include numerous communication devices, but only three communication devices areshown for simplicity. There may be other UE groups as well. The communication system 200 may have other components that are not shown.
[0072]
[0036] The data analytics server 220 has a network interface 215 configured to communicate with other nodes of the communication system 200, a CRM 219, and data analytics circuitry 216 coupled to the network interface 215 and the CRM 219. In some implementations, the data analytics circuitry 226 includes a processor 217 that executes software, which can stem from a memory 228. However, other implementations are possible and are within the scope of this disclosure. The data analytics server 220 can have additional components, but these are not shown for simplicity.
[0073]
[0037] Abnormal user plane traffic, specifically too much traffic (in terms of volume and / or throughput) from application servers, abnormal traffic bursts, wrong formed packets, too much retransmitted packets, as well as too much fragmentation & reassembly processes, have a relevant impact in user plane performance. Also, MNOs today apply different traffic management actions, e.g. content filtering (to block traffic towards malicious or forbidden sites) and parental control. However, the existing mechanisms are not efficient in terms of performance and lack the use of AIML (Artificial Intelligence Machine Learning) capabilities.
[0074]
[0038] Embodiments of the disclosure set out to address or mitigate one or more of these deficiencies with the existing mechanisms. More specifically, the data analytics circuitry 216 of the data analytics server 220 operates to implement a method for providing analytics information or ML (Machine Learning) model from which the consumer node 230 can ascertain whether there is any user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden. Such operation will be described below with reference to Figure 3. Although the method of Figure 3 is described below with reference to the communication system 200 shown in Figure 2, it is to be understood that the method of Figure 3 is applicable to other communication systems. In general, the method of Figure 3 is applicable to any appropriately configured communication system.
[0039] At step 301 , the data analytics server 220 receives, from the consumer node 230, a request message including at least one parameter concerning user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden.
[0075]
[0040] At step 302, the data analytics server 220 acquires, from the user plane node, user plane traffic based on the at least one parameter of the request message. Such acquisition can include multiple messages, including a request and a response, although such multiple messages are not shown.
[0076]
[0041] At step 303, the data analytics server 220 generates, based on the user plane traffic acquired using the at least one parameter, analytics information or ML (Machine Learning) model from which the consumer node 230 can ascertain whether there is any user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden.
[0077]
[0042] Finally, at step 304, the data analytics server 220 sends, to the consumer node 230, a response based on the analytics information or ML model. Based on this response, the consumer node 230 can ascertain whether there is any user plane traffic anomalies and / or user plane traffic with sites that are malicious and / or forbidden.
[0078]
[0043] In a first embodiment, the at least one parameter concerns user plane traffic anomalies, and the analytics information or ML model is configured to identify whether there is any user plane traffic anomalies. An example of this is described later with reference to Figure 4.
[0079]
[0044] In a second embodiment, the at least one parameter concerns user plane traffic with sites that are malicious and / or forbidden, and the analytics information or ML model is configured to identify whether there is user plane traffic with sites that are malicious and / or forbidden. An example of this is described later with reference to Figure 5.
[0080]
[0045] In some implementations, the analytics information or ML model comprises analytics information, and the response includes at least some of the analytics information. Such implementations may be applicable to both the first embodiment and the second embodiment. An example implementation with the first embodiment is described later with reference to Figure 4.
[0046] In other implementations, the analytics information or ML model comprises an ML model, and the response includes an ML model identifier and ML model information. Such implementations may be applicable to both the first embodiment and the second embodiment. An example implementation with the second embodiment is described later with reference to Figure 5.
[0081]
[0047] There are many possibilities for the network 202. In some implementations, the network 202 includes a 5G (Fifth Generation) network. The network 202 can include a RAN (Radio Access Network) and a CN (Core Network). The RAN includes base stations and antennas for wirelessly connecting the UE group 204a-c to the CN. The CN can have components such as those shown in Figure 1 , and can manage data routing, authentication, and connectivity between users and external networks like the internet. Each NF of the CN may be implemented either as a network element on a dedicated hardware, as a software instance running on a dedicated hardware, or as a VNF (Virtualized Network Function) instantiated on an appropriate platform, e.g. a cloud infrastructure. Other implementations for the network 202 are possible other than 5G. For example, in other implementations, the network 202 includes a 6G (Sixth Generation) network.
[0082]
[0048] According to another embodiment of the disclosure, there is provided a non-transitory CRM having recorded thereon statements and instructions that, when executed by the processor 217 of the data analytics server 220, implement a method as described herein. The non-transitory computer readable medium can be the memory 218 and / or the CRM 219 of the data analytics server 220 shown in Figure 2, or some other non-transitory CRM. Examples of a non-transitory CRM include memory, an SSD (Solid State Drive), a hard disk drive, a CD (Compact Disc), a DVD (Digital Video Disc), a BD (Blu-ray Disc), a memory stick, etc. Other non-transitory CRMs are also possible.
[0083]
[0049] The illustrated examples described herein focus on software implementations. However, other implementations are possible and are within the scope of this disclosure. It is noted that other implementations can include additional or alternative hardware components, such as any appropriately configured FPGA (Field-Programmable Gate Array), ASIC (Application-Specific Integrated Circuit), and / or microcontroller, for example. Thus, the data analytics circuitry 216 of the data analyticsserver 220 can instead be implemented with any suitable combination of hardware, software and / or firmware.
[0084]
[0050] Further example details are provided in the following sections. It is to be understood that the following sections are very specific and are provided merely for exemplary purposes, such that other implementations are possible and within the scope of the disclosure.
[0085] First Embodiment
[0086]
[0051] Abnormal user plane traffic, specifically too much traffic (in terms of volume and / or throughput) from application servers, abnormal traffic bursts, wrong formed packets, too much retransmitted packets, as well as too much fragmentation & reassembly processes, have a relevant impact in user plane performance.
[0087]
[0052] The first embodiment includes a 5GC enhancement to support analysis of abnormal user plane traffic, specifically a new analytics related to abnormal user plane traffic. These are the high-level principles of some embodiments:
[0088] o Assumption: To identify abnormal user plane traffic, NWDAF needs to have information relative to normal user plane traffic. It is proposed 0AM to provide such information to NWDAF. 0AM might either have this information locally configured or the proposed analytics can be extended, by 0AM acting as analytics consumer and providing as input parameter an indication to NWDAF to request normal user plane traffic behavior.
[0089] o Consumer (e.g. PCF, SMF) subscribes to NWDAF for a new analytic relative to abnormal user plane traffic, including as input (among other inputs) the user plane traffic anomalies of interest (too much data and / or rate, abnormal traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets).
[0090] o NWDAF triggers data collection as follows:
[0091] o Data collection from UPF to retrieve user plane traffic parameters (specifically, metrics or KPIs like volume, data rate, etc and / or packet parameters like 5-tuple(s), SNI(s), domain name(s), etc and / or classification related information like application identifier(s), traffic burst(s) and / or malformed, duplicate, unknown, retransmitted and / or fragmented packets).o Data collection (e.g. from OAM, NRF, UPF) to retrieve information relative to UPF load. This can be used by NWDAF to determine if and how the abnormal user plane traffic has any significant impact on the UPF load.
[0092] o NWDAF runs analytics based on collected data and derives the analytic result, which can be used by the consumer to apply actions to improve user plane performance, e.g. to avoid abnormal traffic.
[0093]
[0053] In summary, some implementations provide a 5GC enhancement to support analysis of abnormal user plane traffic, specifically a new analytics related to abnormal user plane traffic.
[0094]
[0054] The first embodiment can allow the network operator to detect abnormal user plane traffic and to take actions to mitigate it. The first embodiment can provide new analytics related to abnormal user plane traffic to be included as a new solution in 3GPP TR 23.700-04.
[0095] Abnormal UP traffic Analytics
[0096]
[0055] The consumer of this analytics shall indicate in the request:
[0097] o Analytics ID = 'Abnormal UP traffic";
[0098] o Target of Analytics Reporting can be a list of SUPIs, a group of SUPIs or any UE;
[0099] o An Analytics target period indicates the time period over which the statistics or predictions are requested;
[0100] o Analytics Filter Information optionally including:
[0101] o analytics type (or list of Exception IDs), where the analytics type refers to the traffic anomalies of interest (too much data and / or rate, abnormal traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets);
[0102] o Optionally, when analytics type refers to abnormal traffic bursts, information relative to the definition of a traffic burst.
[0103] o Optionally, information relative to normal UP traffic or associated thresholds (for each Exception Level)
[0104] o Area of interest;
[0105] o Application ID;o DNN;
[0106] o S-NSSAI.
[0107] Input Data
[0108]
[0056] NWDAF triggers data collection from UPF to retrieve user plane traffic parameters. The detailed input data is described in Table 1 , where UL is uplink and DL is downlink.
[0109] Table 1: Input data
[0110] Information Source Description
[0111] Application ID UPF Identifies the application.
[0112] Application Traffic Flow UPF Per Application Traffic flow related information for Information (1. . max) the application and optionally, for a DNN, S-NSSAI combination.
[0113] > IP 5-tuple UPF Identifies a service flow of the UE that uses the application.
[0114] > Start time UPF Start time of traffic detection for the flow.
[0115] > End time UPF End time of traffic detection for the flow.
[0116] > UL Data volume UPF Measured UL data traffic volume for the flow. > DL Data volume UPF Measured DL data traffic volume for the flow. > UL Data Rate UPF Measured UL data rate for the flow.
[0117] > DL Data Rate UPF Measured DL data rate for the flow.
[0118] > URL list UPF List of URLs extracted from the inspected user plane packets in the flow.
[0119] > Domain Name list UPF List of domain names extracted from the inspected user plane packets in the flow.
[0120] > UL Traffic Bursts UPF Number of detected UL traffic bursts in the flow. > DL Traffic Bursts UPF Number of detected DL traffic bursts in the flow. > UL Malformed packets UPF Number of UL malformed packets in the flow. > DL Malformed packets UPF Number of DL malformed packets in the flow. > UL Duplicate packets UPF Number of UL duplicate packets in the flow. > DL Duplicate packets UPF Number of DL duplicate packets in the flow. > UL Unknown packets UPF Number of UL unknown packets in the flow. > DL Unknown packets UPF Number of DL unknown packets in the flow.
[0121]
[0122] > UL Retransmitted UPF Number of UL retransmitted packets in the flow. packets
[0123] > DL Retransmitted UPF Number of DL retransmitted packets in the flow. packets
[0124] > UL Fragmented packets UPF Number of UL fragmented packets in the flow. > DL Fragmented packets UPF Number of DL fragmented packets in the flow.
[0125]
[0126]
[0057] NWDAF also triggers data collection (e.g. from OAM, NRF, UPF) to retrieve information relative to UPF load via existing mechanisms.
[0127] Output Analytics
[0128]
[0058] Corresponding to the 'Abnormal UP traffic" Analytics ID, the analytics result provided by the NWDAF is defined in Table 2 and Table 3. When the level of an exception trespasses above or below the threshold, the NWDAF shall notify the consumer with the exception ID associated with the exception if the exception ID is within the list of exception IDs indicated by the consumer or matches the expected analytics type indicated by the consumer. The NWDAF shall provide the Exception Level and determine which of the other lEs (information elements) to provide, depending on the observed exception.
[0129]
[0059] Abnormal UP traffic statistics information is defined in Table 2.
[0130] Table 2: Abnormal UP traffic statistics
[0131] Information Description
[0132] Exceptions (1. . max) List of observed exceptions
[0133] > Exception ID The risk detected by NWDAF
[0134] > Exception Level Scalar value indicating the severity of the abnormal UP traffic
[0135] > Exception trend Measured trend (up / down / unknown / stable)
[0136] > SUPI list (1. . SUPImax) SUPI(s) of the UE(s) affected with the Exception
[0137] > Ratio Estimated percentage of UEs affected by the Exception within the Target of Analytics Reporting
[0138] > Amount Estimated number of UEs affected by the Exception (applicable when the Target of Analytics Reporting = "any UE")
[0139] > Additional information Specific information for each risk
[0140]
[0141]
[0060] Abnormal UP traffic predictions information is defined in Table 3.
[0142] Table 3: Abnormal UP traffic predictions
[0143] Information Description
[0144] Exceptions (1. . max) List of predicted exceptions
[0145] > Exception ID The risk detected by NWDAF
[0146] > Exception Level Scalar value indicating the severity of the abnormal behaviour
[0147] > Exception trend Measured trend (up / down / unknown / stable)
[0148] > SURI list (1. . SUPImax) SUPI(s) of the UE(s) affected with the Exception
[0149] > Ratio Estimated percentage of UEs affected by the Exception within the Target of Analytics Reporting
[0150] > Amount Estimated number of UEs affected by the Exception (applicable when the Target of Analytics Reporting = "any UE")
[0151] > Additional information Specific information for each risk
[0152] > Confidence Confidence of this prediction
[0153]
[0154]
[0061] Exception ID can be: too much data and / or rate, abnormal traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets or fragmented packets.
[0155]
[0062] Regarding Additional information, it includes specific information for each risk, e.g. to indicate, for the Exception ID, for which server IP address(es) and / or application ID(s) it applies to. It might also indicate if the exception ID has any significant impact on the UPF load.
[0156] Procedure
[0157]
[0063] Referring now to Figure 4, shown is a sequence drawing of a procedure for analytics on abnormal user plane traffic.
[0158]
[0064] At step 401, the consumer (e.g. PCF, SMF) subscribes to NWDAF analytics on abnormal user plane traffic, by triggering Nnwdaf_AnalyticsSubscription_Subscribe request message including as analytics identifier (AbnormalUPTraffic), including as input the user plane traffic anomalies of interest (too much data and / or rate, abnormal traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets), as analytic filter the DNN, S-NSSAI and / or Area of Interest, as analytic targeta list of SUPIs, a group of SUPIs or any UE, and this either for the traffic of the whole PDU session or for a certain application identif ier(s) .
[0159]
[0065] At step 402, the NWDAF triggers data collection (e.g. from 0AM, NRF, UPF) to retrieve information relative to UPF load via existing mechanisms.
[0160]
[0066] At step 403, the NWDAF triggers data collection from UPF to retrieve user plane traffic parameters, by triggering a Nupf_EventExposure_Subscribe request message, including as eventType="USER_DATA_USAGE_MEASURES", a new measurementType="TRAFFIC_DATA" and a list of traffic data of interest (volume and / or rate, traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets), and for each a threshold (e.g. to report only when the percentage of fragmented packets goes over a certain threshold), and the target SUPI(s) and application identifier(s). Optionally, when traffic data of interest refers to traffic bursts, information relative to the definition of a traffic burst.
[0161] NOTE1 : It is assumed both UPF event subscription and reporting applies per SUPI (and application identifier), but optimizations (to minimize signaling) as discussed by CT4 can also be applied.
[0162]
[0067] At step 404, the UPF collects data for the above event, specifically the UP traffic parameters of interest: metrics or KPIs like volume, data rate, etc and / or packet parameters like 5-tuple(s), SNI(s), domain name(s), etc and / or classification related information, if available, like application identifier(s), traffic burst(s) and / or malformed, duplicate, unknown, retransmitted and / or fragmented packets.
[0163]
[0068] At step 405, the UPF notifies the NWDAF by triggering a Nupf_EventExposure_Notify request message, including the eventType="USER_DATA_USAGE_MEASURES" and the event information, specifically the UP traffic parameters indicated above (when the corresponding thresholds are surpassed).
[0164]
[0069] At step 406, the NWDAF runs analytics based on collected data and derives the analytics result, specifically a list of Exception IDs corresponding to abnormal UP traffic detected, e.g. when the number and / or percentage of fragmented packets is over the normal behavior, indicating for which SUPI(s), applicationidentifier(s) and / or server IP addresses it applies to. The NWDAF can also determine and indicate to the consumer if and how the abnormal UP traffic has any impact on the UPF load.
[0165]
[0070] At step 407, the NWDAF notifies the analytic consumer (e.g. PCF, SMF) by triggering Nnwdaf_AnalyticsSubscription_Notify request message including the analytics identifier (UPAbnormalTraffic) and the analytics result. Based on the analytics result, the consumer can apply actions to improve user plane performance, e.g. to avoid abnormal traffic, by PCF setting policies, e.g. for UPF to block traffic from / to servers (server IP addresses) from which abnormal behavior has been detected (e.g. too much packet fragmentation) or SMF to trigger UPF re-selection.
[0166] Technical
[0167]
[0168]
[0071] 3GPP TS 23.288: New NWDAF analytic related to abnormal user plane traffic.
[0169]
[0072] 3GPP TS 29.564: As part of UPF event exposure service (Nupf_EventExposure), extension of the existing UPF event (eventType="USER_DATA_USAGE_MEASURES") with a new measurementType="TRAFFIC_DATA" and a list of data anomalies of interest (too much data and / or rate, abnormal traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets). As an alternative, it is also possible to define a new event.
[0170]
[0073] Regarding 3GPP TR 23.700-04, FS_AIML_CN_Ph2 / Rel-20: (Kl#2) Solution based on analysis of abnormal user plane traffic. See the following text proposal.
[0171] 6.X Solution #X: Analysis of abnormal user plane traffic
[0172] 6.X.1 High level principles
[0173] These are the high-level principles of the proposed solution:
[0174] A new AnalyticsID “Abnormal user plane traffic” is proposed.
[0175] Input data retrieved from UPF on user plane traffic parameters as defined in Table 6.X.2.2-1.Output, a list of exceptions and the list of SUPIs and / or Application ID(s) affected per exception in the list, as defined in Table 6.X.2.3-1 and Table 6.X.2.3-2. The list of exceptions are: too much data sent / received and / or too high or low rate, abnormal traffic bursts, too many malformed packets, too many duplicate packets, too many unknown packets, too many retransmitted packets and / or too many fragmented packets.
[0176] The analytics consumer may be SMF or PCF.
[0177] The consumer may request the type of exception is interested in when subscribing to this new AnalyticsID. The consumer takes actions based on the type of exception reported by NWDAF, such as blocking traffic reported as abnormal.
[0178] Assumption: To identify abnormal user plane traffic, NWDAF may provide a threshold to the UPF when retrieving the input data, then any traffic that do not reach the threshold is not considered “abnormal” traffic by NWDAF and by UPF.
[0179] 6.X.2 Description
[0180] This solution proposes a 5GC enhancement to support analysis of abnormal user plane traffic, specifically a new AnalyticsID related to abnormal user plane traffic.
[0181] 6.X.2.1 Abnormal user plane traffic AnalyticsID
[0182] The consumer of this AnalyticsID shall indicate in the request:
[0183] - Analytics ID = "Abnormal user plane traffic";
[0184] - Target of Analytics Reporting can be a list of SUPIs, a group of SUPIs or any UE;
[0185] - An Analytics target period indicates the time period over which the statistics or predictions are requested;
[0186] - Analytics Filter Information optionally including:
[0187] - list of Exception ID, where each exception ID refers to a traffic data anomaly of interest (too much data and / or rate, abnormal traffic bursts, too many malformed packets, too many duplicate packets, too many unknown packets, too many retransmitted packets and / or too many fragmented packets);
[0188] - Optionally, when analytics type refers to abnormal traffic bursts, information relative to the definition of a traffic burst;
[0189] - Optionally, associated thresholds (for each Exception Level);
[0190] - Area of interest;
[0191] - Application ID;
[0192] - DNN;- S-NSSAI.
[0193] 6.X.2.2 Input Data
[0194] NWDAF triggers data collection from UPF to retrieve user plane traffic parameters. The detailed input data is described in Table 6.X.2.2-1.
[0195] Table 6. X.2.2-1: Input data
[0196] Information Source Description
[0197] Application ID UPF Identifies the application.
[0198] Application Traffic Flow UPF Per Application Traffic flow related information for the application Information (1..max) and optionally, for a DNN, S-NSSAI combination.
[0199] > IP 5-tuple UPF Identifies a service flow of the UE that uses the application. > Start time UPF Start time of traffic detection for the flow.
[0200] > End time UPF End time of traffic detection for the flow.
[0201] > UL Data volume UPF Measured UL data traffic volume for the flow.
[0202] > DL Data volume UPF Measured DL data traffic volume for the flow.
[0203] > UL Data Rate UPF Measured UL data rate for the flow.
[0204] > DL Data Rate UPF Measured DL data rate for the flow.
[0205] > URL list UPF List of URLs extracted from the inspected user plane packets in the flow.
[0206] > Domain Name list UPF List of domain names extracted from the inspected user plane packets in the flow.
[0207] > UL Traffic Bursts UPF Number of detected UL traffic bursts in the flow.
[0208] > DL Traffic Bursts UPF Number of detected DL traffic bursts in the flow.
[0209] > UL Malformed packets UPF Number of UL malformed packets in the flow.
[0210] > DL Malformed packets UPF Number of DL malformed packets in the flow.
[0211] > UL Duplicate packets UPF Number of UL duplicate packets in the flow.
[0212] > DL Duplicate packets UPF Number of DL duplicate packets in the flow.
[0213] > UL Unknown packets UPF Number of UL unknown packets in the flow.
[0214] > DL Unknown packets UPF Number of DL unknown packets in the flow.
[0215] > UL Retransmitted packets UPF Number of UL retransmitted packets in the flow.
[0216] > DL Retransmitted packets UPF Number of DL retransmitted packets in the flow.
[0217] > UL Fragmented packets UPF Number of UL fragmented packets in the flow.
[0218] > DL Fragmented packets UPF Number of DL fragmented packets in the flow.
[0219]
[0220] NWDAF also triggers data collection (e.g. from 0AM, NRF, UPF) to retrieve information relative to UPF load via existing mechanisms.6.X.2.3 Output Analytics
[0221] Corresponding to the "Abnormal user plane traffic" Analytics ID, the analytics result provided by the NWDAF is defined in Table 6.X.2.3-1 and Table 6.X.2.3-2. When the level of an exception passes above or below the threshold, the NWDAF shall notify the consumer with the exception ID associated with the exception if the exception ID is within the list of exception IDs indicated by the consumer or matches the expected analytics type indicated by the consumer. The NWDAF shall provide the Exception Level and determine which of the other information elements to provide, depending on the observed exception.
[0222] Abnormal user plane traffic statistics information is defined in Table 6.X.2.3-1.
[0223] Table 6.X.2.3-1 : Abnormal user plane traffic statistics
[0224] Information Description
[0225] Exceptions (1..max) List of observed exceptions
[0226] > Exception ID The risk detected by NWDAF
[0227] > Exception Level Scalar value indicating the severity of the abnormal UP traffic
[0228] > Exception trend Measured trend (up / down / unknown / stable)
[0229] > SUPI list (1..SUPImax) SUPI(s) of the UE(s) affected with the Exception. When the consumer requests the analytics for "any UE”, this information is not present.
[0230] > Application ID list (LAppIDmax) Application ID(s) affected with the Exception
[0231] > Ratio Estimated percentage of UEs affected by the Exception within the Target of Analytics Reporting
[0232] > Amount Estimated number of UEs affected by the Exception (applicable when the Target of Analytics Reporting = "any UE")
[0233] > Additional information Specific information for each risk
[0234]
[0235] Abnormal user plane traffic predictions information is defined in Table 6.X.2.3-2.
[0236] Table 6.X.2.3-2: Abnormal user plane traffic predictions Information Description
[0237] Exceptions (1..max) List of predicted exceptions
[0238] > Exception ID The risk detected by NWDAF
[0239] > Exception Level Scalar value indicating the severity of the abnormal behaviour
[0240] > Exception trend Measured trend (up / down / unknown / stable)
[0241]
[0242] > SUPI list (T.SUPImax) SUPI(s) of the UE(s) affected with the Exception. When the consumer requests the analytics for "any UE”, this information is not present.
[0243] > Application ID list (UAppIDmax) Application ID(s) affected with the Exception
[0244] > Ratio Estimated percentage of UEs affected by the Exception within the Target of Analytics Reporting
[0245] > Amount Estimated number of UEs affected by the Exception (applicable when the Target of Analytics Reporting = "any UE")
[0246] > Additional information Specific information for each risk
[0247] > Confidence Confidence of this prediction
[0248]
[0249] Exception ID can be: too much data and / or rate, abnormal traffic bursts, too many malformed packets, too many duplicate packets, too many unknown packets, too many retransmitted packets or too many fragmented packets.
[0250] Regarding Additional information, it includes specific information for each risk, e.g. to indicate, for the Exception ID, for which server IP address(es) it applies to. It might also indicate if the exception ID has any significant impact on the UPF load.
[0251] 6.X.2.4 Procedure for Analytics ID on abnormal user plane traffic Figure 4 depicts the procedure for the proposed solution.
[0252] 1 Consumer (e.g. PCF, SMF) subscribes to NWDAF analytics on abnormal user plane traffic, by triggering Nnwdaf_AnalyticsSubscription_Subscribe request message including as analytics identifier (“Abnormal user plane traffic”), including as input the user plane traffic data anomalies of interest (too much data and / or rate, abnormal traffic bursts, too many malformed packets, too many duplicate packets, too many unknown packets, too many retransmitted packets and / or too many fragmented packets), and for each, optionally, a threshold (e.g. in case the number and / or percentage of malformed packets is above the threshold, this needs to be provided as analytic output to the consumer), as analytic filter the DNN, S-NSSAI and / or Area of Interest, as analytic target a list of SUPIs, a group of SUPIs or any UE, and this either for the traffic of the whole PDU session or for certain application identified s).
[0253] 2 NWDAF triggers data collection (e.g. from OAM, NRF, UPF) to retrieve information relative to UPF load via existing mechanisms.NWDAF triggers data collection from UPF to retrieve user plane traffic parameters, by triggering a Nupf_EventExposure_Subscribe request message, including as eventType="USER_DATA_USAGE_MEASURES", a new measurementType="TRAFFIC_DATA ANOMALIES" and a list of traffic data anomalies of interest (volume and / or rate, traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets), and for each a threshold, e.g. to report only when the percentage of fragmented packets goes over a certain threshold (when included in step 1 above or locally provisioned at NWDAF; when not included, it is assumed the threshold is locally provisioned at UPF), and the target SUPI(s) and application identifier(s). Optionally, when traffic data of interest refers to traffic bursts, information relative to the definition of a traffic burst.
[0254] N0TE1: It is assumed both UPF event subscription and reporting applies per SUPI (and application identifier), but optimizations (to minimize signaling) as discussed by CT4 can also be applied.
[0255] UPF collects data for the above event, specifically the UP traffic parameters of interest: metrics or KPIs like volume, data rate, etc and / or packet parameters like 5-tuple(s), SNI(s), domain name(s), etc and / or classification related information, if available, like application identified s), traffic burst(s) and / or malformed, duplicate, unknown, retransmitted and / or fragmented packets.
[0256] UPF notifies NWDAF by triggering a Nupf_EventExposure_Notify request message, including the eventType="USER_DATA_USAGE_MEASURES" and the event information, specifically the UP traffic parameters indicated above (when the corresponding thresholds are surpassed).
[0257] NWDAF runs analytics based on collected data and derives the analytics result, specifically a list of Exception IDs corresponding to abnormal UP traffic detected, e.g. when the number and / or percentage of fragmented packets is over the threshold, indicating for which SUPI(s), application identifier(s) and / or server IP addresses it applies to. NWDAF can also determine and indicate to the consumer if and how the abnormal UP traffic has any impact on the UPF load.
[0258] NWDAF notifies the analytic consumer (e.g. PCF, SMF) by triggering Nnwdaf_AnalyticsSubscription_Notify request message including the analytics identifier (“Abnormal user plane Traffic”) and the analytics result. Based on theanalytics result, the consumer can apply actions to improve user plane performance, e.g. to avoid abnormal traffic, by PCF setting policies (via SMF) based on existing policy framework, e.g. for UPF to block traffic from / to servers (server IP addresses) from which abnormal behavior has been detected (e.g. too much packet fragmentation) or SMF to trigger UPF re-selection for PDU Sessions for the SUPIs from which abnormal behavior has been detected.
[0259] 6.X.3 Impacts on services, entities and interfaces
[0260] NWDAF:
[0261] - New analyticsID relative to abnormal user plane traffic.
[0262] UPF:
[0263] Extension of the existing USER DATA USAGE MEASURES event with a new measurement type (TRAFFIC DATA ANOMALIES) which allows UPF to report information related with traffic data anomalies, specifically traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets.
[0264] SMF, PCF:
[0265] Act as consumer of the new analyticsID relative to abnormal user plane traffic, and based on the received analytics result, to apply actions, e.g. based on existing policy framework, PCF setting policies (via SMF) for UPF to block traffic from / to servers (server IP addresses) from which abnormal behavior has been detected (e.g. too much packet fragmentation) or SMF to trigger UPF re-selection for PDU Sessions for the SUPIs from which abnormal behavior has been detected.
[0266] Second Embodiment
[0267]
[0074] MNOs today apply different traffic management actions, e.g. content filtering (to block traffic towards malicious or forbidden sites) and parental control. However, the existing mechanisms are not efficient in terms of performance and lack the use of AIML capabilities.
[0268]
[0075] Regarding 3GPP TR 23.700-04, Rel20 AIML_CN_Ph2 has agreed to study whether and how to train ML models and perform the inference for the user plane analytics, considering solutions that minimize the load on Control Plane NFs.
[0076] This IvD proposes a 5GC enhancement to support analysis of user plane traffic towards websites or services that are deemed harmful, illicit, or restricted, specifically:
[0269] o Malicious Sites: These are sites that host malware, phishing or other harmful activities designed to compromise user devices, steal information, or disrupt services. Traffic directed towards such sites can lead to security breaches, data theft, or system infections.
[0270] o Forbidden Sites: These are sites that are restricted by policy, law, or regulation.
[0271] Examples include sites that host illegal content, violate intellectual property rights, or are restricted due to geographical or organizational policies.
[0272]
[0077] The principles of some implementations are as follows:
[0273] o A consumer (e.g. UPF, 0AM) requests / subscribes to a ML Model which allows UPF to detect UP traffic towards malicious and / or forbidden sites. The training of this ML Model can be done either by 0AM or by NWDAF containing MTLF and can be based both on AF data (e.g. IWF database) and UPF data (e.g. UP traffic parameters).
[0274] o When UPF retrieves the trained ML Model, it uses it to detect in real-time (e.g.
[0275] on a per traffic flow basis) UP traffic towards malicious and / or forbidden sites. Based existing policy framework, UPF applies the corresponding actions (e.g. to block this traffic).
[0276]
[0078] New procedure for UPF (acting as NWDAF Service Consumer) to request a ML Model for UPF-based detection of traffic towards malicious and / or forbidden sites, where the ML Model is trained by an NWDAF containing MTLF. As an alternative, UPF might also fetch the model via 0AM (and even 0AM doing the ML Model training, without involving an NWDAF containing MTLF.
[0277]
[0079] Some implementations can allow the network operator to support an efficient solution based on AIML which allows to detect and mitigate (in real time) user plane traffic towards websites or services that are deemed harmful, illicit, or restricted.
[0080] The second embodiment is proposed to be included as a new solution in 3GPP TR 23.700-04.
[0278]
[0081] Figure 5 depicts the procedure for the proposed solution, for the specific case UPF (acting as NWDAF Service Consumer) requests a ML Model for UPF-based detection of traffic towards malicious and / or forbidden sites, where the ML Model is trained by an NWDAF containing MTLF. As an alternative, it is also possible that UPF fetches the model via 0AM (and even 0AM doing the ML Model training, without involving an NWDAF containing MTLF).
[0279]
[0082] At step 501 , the NWDAF Service Consumer (e.g. UPF) subscribes to the NWDAF containing MTLF for a trained ML Model related to detection of user plane traffic towards malicious and / or forbidden sites, by invoking the Nnwdaf_MLModelProvision_Subscribe request service operation, including the following input data:
[0280] o UPF-based detection of traffic towards malicious and / or forbidden sites indication.
[0281] o Traffic of interest (e.g. content filtering categories and / or parental control categories like adult, violence, etc).
[0282] o Optionally, Vendor ID, ML Model Filter Information (e.g. Area of Interest), Target of ML Model Reporting, ML Model Target Period, Time when model is needed, Inference Input Data information and ML Model Monitoring Information.
[0283] o If vendor specific information is required, then the ML Model Interoperability Information is included.
[0284] o If the UPF supports multiple AI / ML Models, indication of support for multiple ML Models, optionally with Number of ML Models and Accuracy level(s) of Interest.
[0285]
[0083] When a ML Model subscription is received, the NWDAF containing MTLF may:
[0286] o determine whether existing trained ML Model(s) can be used for the subscription;
[0287] or
[0288] o determine whether triggering further training for the existing trained ML Models is needed for the subscription.If the service invocation is for a subscription modification or subscription cancelation, the NWDAF service consumer includes an identifier (Subscription Correlation ID) to be modified in the invocation of Nnwdaf_MLModelProvision_Subscribe.
[0289]
[0084] At step 502, if the NWDAF containing MTLF determines that further training is needed, this NWDAF may initiate data collection from the AF (via NEF if AF is external / untrusted) to retrieve information relative to malicious and / or forbidden sites (e.g. IWF database).
[0290]
[0085] At step 503, the NWDAF containing MTLF may also collect input data from the UPF to retrieve user plane traffic information, by invoking the Nupf_EventExposure_Subscribe service operation, including as eventType="USER_DATA_USAGE_MEASURES", and the target SUPIs.
[0291]
[0086] At steps 504 and 505, the UPF collects data for the above event and notifies the NWDAF by invoking the Nupf_EventExposure_Notify service operation, including the eventType="USER_DATA_USAGE_MEASURES" and the event information, e.g. 5-tuple(s), SNI(s), URL(s) for the target SUPIs.
[0292]
[0087] At step 506, the NWDAF containing MTLF, based on collected data, trains the ML Model which allows UPF to detect user plane traffic towards malicious and / or forbidden sites.
[0293]
[0088] At step 507, the NWDAF containing MTLF notifies the NWDAF Service Consumer (e.g. UPF) by invoking the Nnwdaf_MLModelProvision_Notify service operation, including the following information:
[0294] o ML Model identifier and ML Model Information.
[0295] o Optionally, ML Model Filter Information and / or Target of ML Model Reporting, if the ML Model provisioning request includes multiple ML Model Filter Information and / or Target of ML Model Reporting;
[0296] o Optionally, Indication of whether the ML Model identifier is updated (e.g.
[0297] retrained ML model).o Optionally, Validity period, Spatial validity, Training Input Data Information, ML Model accuracy Information.
[0298]
[0089] At step 508, the NWDAF Service Consumer (e.g. UPF), based on the above information, retrieves the ML Model and uses it to detect in real-time (e.g. on a per traffic flow basis) user plane traffic towards malicious and / or forbidden sites. It is assumed UPF applies the corresponding policies (e.g. block traffic towards malicious and / or forbidden sites) based on existing policy framework.
[0299] Technical
[0300]
[0301]
[0090] 3GPP TS 23.288: New procedure for UPF (acting as NWDAF Service Consumer) to request a ML Model for UPF-based detection of traffic towards malicious and / or forbidden sites, where the ML Model is trained by an NWDAF containing MTLF. As an alternative, UPF might also fetch the model via 0AM (and even 0AM doing the ML Model training, without involving an NWDAF containing MTLF).
[0302]
[0091] NWDAF containing MTLF, 0AM impacts:
[0303] o New ML Model which allows UPF to detect UP traffic towards malicious and / or forbidden sites, which is trained based both on AF data (e.g. IWF database) and UPF data (e.g. UP traffic parameters).
[0304] o New procedure to retrieve from AF (via NEF if AF is external / untrusted) information relative to malicious and / or forbidden sites (e.g. IWF database).
[0305]
[0092] UPF impacts:
[0306] o Retrieval (either from NWDAF containing MTLF or via 0AM) of the trained ML Model to detect traffic towards malicious and / or forbidden sites.
[0307] o Use the trained ML Model to detect traffic towards malicious and / or forbidden sites and enforcement of policies (e.g. block traffic towards malicious and / or forbidden sites) based on existing policy framework.
[0308]
[0093] NEF impacts:
[0309] o New procedure to enable NWDAF containing MTLF or 0AM to retrieve from AF information relative to malicious and / or forbidden sites (e.g. IWF database).
[0094] Regarding 3GPP TR 23.700-04, FS_AIML_CN_Ph2 I Rel-20: (Kl#2) Solution based on analysis of user plane traffic towards malicious and / or forbidden sites. See the following text proposal.
[0310] 6.X Solution #X: Analysis of user plane traffic towards malicious and / or forbidden sites 6.X.1 High Level principles
[0311] The high level principles of this solution are as follows:
[0312] - A consumer (i.e. UPF) retrieves a trained ML Model which allows detection of UP traffic towards malicious and / or forbidden sites either from NWDAF (MTLF) or from 0AM. - The training of this ML Model is performed based both on AF data (e.g. Internet Watch Foundation or IWF database) and UPF data (e.g. UP traffic parameters).
[0313] - The UPF uses this ML Model for detection of user plane traffic to malicious or forbidden sites in real-time (e.g. on a per flow basis).
[0314] - The UPF applies corresponding actions (e.g. block) for the detected traffic, based on existing policy framework.
[0315] 6.X.2 Description
[0316] This solution proposes a 5GC enhancement to support analysis of user plane traffic towards websites or services that are deemed harmful, illicit, or restricted, specifically:
[0317] - Malicious Sites: These are sites that host malware, phishing or other harmful activities designed to compromise user devices, steal information, or disrupt services. Traffic directed towards such sites can lead to security breaches, data theft, or system infections. - Forbidden Sites: These are sites that are restricted by policy, law, or regulation. Examples include sites that host illegal content, violate intellectual property rights, or are restricted due to geographical or organizational policies.
[0318] 6.X.2 Procedures
[0319] Figure 5 depicts the procedure for the proposed solution, for the specific case UPF (acting as NWDAF Service Consumer) requests a ML Model for UPF -based detection of traffic towards malicious and / or forbidden sites, where the ML Model is trained by an NWDAF containing MTLF. As an alternative, it is also possible that UPF fetches the model via 0AM (without involving an NWDAF containing MTLF).
[0320] 1 NWDAF Service Consumer (UPF) subscribes to NWDAF containing MTLF for a trained ML Model related to detection of user plane traffic towards malicious and / or forbidden sites, by invoking the Nnwdaf_MLModelProvision_Subscribe service operation, including the following input data:- Indication to request a model for detection of traffic towards malicious and / or forbidden sites, including the traffic of interest (e.g. content filtering categories and / or parental control categories like adult, violence, etc).
[0321] Optionally, Vendor ID, ML Model Filter Information (e.g. Area of Interest), Target of ML Model Reporting, ML Model Target Period, Time when model is needed and ML Model Monitoring Information.
[0322] If vendor specific information is required, then the ML Model Interoperability Information is included.
[0323] If the UPF supports multiple AI / ML Models, indication of support for multiple ML Models, optionally with Number of ML Models and Accuracy level(s) of Interest. When a ML Model subscription is received, the NWDAF containing MTLF may: determine whether existing trained ML Model(s) can be used for the subscription then steps 2 to 6 are skipped; or
[0324] determine whether triggering further training for the existing trained ML Models is needed for the subscription.
[0325] If the service invocation is for a subscription modification or subscription cancelation, the NWDAF service consumer includes an identifier (Subscription Correlation ID) to be modified in the invocation of Nnwdaf MLModelProvision Subscribe.
[0326] If the NWDAF containing MTLF determines that further training is needed, this NWDAF may initiate data collection from AF (via NEF if AF is external / untrusted) to retrieve information relative to malicious and / or forbidden sites (e.g. IWF database).
[0327] The NWDAF containing MTLF may also collect input data from UPF to retrieve user plane traffic information, by invoking the Nupf EventExposure Subscribe service operation, including as eventType="USER_DATA_USAGE_MEASURES", and the target SUPIs. This data might be relevant for ML Model training, e.g. to know the weight relative to the information retrieved from AF.
[0328] NOTE: It is assumed both UPF event subscription and reporting applies per SUPI, but optimizations (to minimize signaling) as discussed by CT4 can also be applied.
[0329] -5 The UPF collects data for the above event and notifies NWDAF by invoking the Nupf_EventExposure_Notify service operation, including theeventType="USER_DATA_USAGE_MEASURES" and the event information, e.g.
[0330] 5-tuple(s), SNI(s), URL(s) for the target SUPIs.
[0331] 6 The NWDAF containing MTLF, based on collected data and e.g. by correlating the information about the list of forbidden or malicious sites and the usage per application retrieved from the UPF, trains the ML Model which allows UPF to detect user plane traffic towards malicious and / or forbidden sites.
[0332] 7 NWDAF containing MTLF notifies the NWDAF Service Consumer (e.g. UPF) by invoking the Nnwdaf_MLModelProvision_Notify service operation, including the following information:
[0333] - ML Model identifier and ML Model Information;
[0334] - Optionally, ML Model Filter Information and / or Target of ML Model Reporting, if the ML Model provisioning request includes multiple ML Model Filter Information and / or Target of ML Model Reporting;
[0335] - Optionally, Indication of whether the ML Model identifier is updated (e.g. retrained ML model);
[0336] - Optionally, Validity period, Spatial validity, ML Model accuracy Information. 8 The NWDAF Service Consumer (e.g. UPF), based on the above information, retrieves the ML Model and uses it to detect in real-time (e.g. on a per traffic flow basis) user plane traffic towards malicious and / or forbidden sites. It is assumed UPF applies the corresponding policies (e.g. block traffic towards malicious and / or forbidden sites) based on existing policy framework (e.g. via a PCC rule for an application ID relative to malicious and / or forbidden sites).
[0337] 6.X.3 Impacts on services, entities and interfaces
[0338] NWDAF containing MTLF :
[0339] - New ML Model which allows UPF to detect UP traffic towards malicious and / or forbidden sites, which is trained based both on AF data (e.g. IWF database) and UPF data (e.g. UP traffic parameters).
[0340] - New procedure to retrieve from AF (via NEF if AF is external / untrusted) information relative to malicious and / or forbidden sites (e.g. IWF database) and from UPF to retrieve input data related to the usage of an application traffic.
[0341] Correlate the information about the list of forbidden or malicious sites and the usage per application retrieved from the UPF.OAM:
[0342] 0AM can provide a trained ML Model for “detection of user plane traffic towards malicious and / or forbidden sites” to UPF.
[0343] NOTE: Coordination with SA5 is required to request them to train and provide this ML Model to consumers.
[0344] UPF:
[0345] Retrieval (either from NWDAF containing MTLF or via OAM (depending on coordination with SA5) of the trained ML Model to detect traffic towards malicious and / or forbidden sites,
[0346] - Use the trained ML Model to detect traffic towards malicious and / or forbidden sites. - Enforcement of policies (e.g. block traffic towards malicious and / or forbidden sites) based on existing policy framework (e.g. via a PCC rule for an application ID relative to malicious and / or forbidden sites).
[0347] NEF:
[0348] New procedure to enable NWDAF containing MTLF to retrieve from AF information relative to malicious and / or forbidden sites (e.g. IWF database).
[0349]
[0095] Any appropriate steps, methods, features, functions, or benefits disclosed herein may be performed through one or more functional units or modules of one or more virtual apparatuses. Each virtual apparatus may comprise a number of these functional units. These functional units may be implemented via processing circuitry, which may include one or more microprocessor or microcontrollers, as well as other digital hardware, which may include DSPs (Digital Signal Processor), special-purpose digital logic, and the like. The processing circuitry may be configured to execute program code stored in memory, which may include one or several types of memory such as ROM (Read Only Memory), RAM (Random Access Memory), cache memory, flash memory devices, optical storage devices, etc. Program code stored in memory includes program instructions for executing one or more telecommunications and / or data communications protocols as well as instructions for carrying out one or more of the techniques described herein. In some implementations, the processing circuitry may be used to cause the respective functional unit to perform corresponding functions according one or more embodiments of the present disclosure.
[0096] While processes in the figures may show a particular order of operations performed by certain embodiments of the present disclosure, it should be understood that such order is exemplary (e.g. , alternative embodiments may perform the operations in a different order, combine certain operations, overlap certain operations, etc.).
[0350]
[0097] Numerous modifications and variations of the present disclosure are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims, the disclosure may be practised otherwise than as specifically described herein.
Claims
Claims:
1. A method for execution by a data analytics server (220), comprising:receiving (301), from a consumer node, a request message including at least one parameter concerning user plane traffic anomalies;acquiring (302), from a user plane node, user plane traffic based on the at least one parameter of the request message;generating (303), based on the user plane traffic acquired using the at least one parameter, analytics information or ML (Machine Learning) model from which the consumer node can ascertain whether there is any user plane traffic anomalies; andsending (304), to the consumer node, a response based on the analytics information or ML model.
2. The method of claim 1, wherein the analytics information or ML model is configured to identify whether there is any user plane traffic anomalies.
3. The method of claim 2, wherein the at least one parameter comprises at least one of volume and / or rate, traffic bursts, malformed packets, duplicate packets, unknown packets, retransmitted packets and / or fragmented packets, and for each of these parameters a corresponding threshold for the parameter to be an anomaly.
4. The method of claim 3, wherein the request message further comprises at least one of DNN (Data Network Name), S-NSSAI (Serving Network Slice Selection Assistance Information) and / or area of interest, a list of SUPIs (Subscription Permanent Identifiers), a group of SUPIs or any UE (User Equipment).
5. The method of claim 3 or claim 4, wherein acquiring the user plane traffic comprises acquiring at least one of metrics or KPIs (Key Performance Metrics) and / or packet parameters and / or classification related information.
6. The method of any one of claims 3 to 5, wherein acquiring the user plane traffic based on the request message comprises acquiring the user plane traffic whenever any of the corresponding thresholds is surpassed.
7. The method of any one of claims 3 to 6, wherein generating the analytics information or ML model comprises generating a list of exception IDs (Identifiers) corresponding to abnormal user plane traffic detected, and indicating for which SUPI(s), application identifier(s) and / or server IP (Internet Protocol) addresses to which the abnormal user plane traffic applies.
8. The method of any one of claims 3 to 7, wherein generating the analytics information or ML model further comprises generating an indication if and how abnormal user plane traffic has any impact on UPF (User Plane Function) load.
9. The method of any one of claims 1 to 8, wherein the analytics information or ML model comprises analytics information, and the response includes at least some of the analytics information.
10. The method of any one of claims 1 to 8, wherein the analytics information or ML model comprises an ML model, and the response includes an ML model identifier and ML model information.
11. The method of claim 10, wherein the analytics information or ML model further comprises ML model filter information and / or target of ML model reporting.
12. The method of claim 10 or claim 11, wherein the analytics information or ML model further comprises at least some of a validity period, a spatial validity, training input data information, and ML model accuracy Information.
13. A non-transitory computer readable medium (219) having recorded thereon statements and instructions that, when executed by a processor (217) of an data analytics server (220), configure the data analytics server to implement a method according to any one of claims 1 to 12.
14. A data analytics server (220) comprising:a network interface (215) configured to communicate with other network nodes;data analytics circuitry (216) coupled to the network interface and configured to:receive (301), from a consumer node, a request message including at least one parameter concerning user plane traffic anomalies;acquire (302), from a user plane node, user plane traffic based on the at least one parameter of the request message;generate (303), based on the user plane traffic acquired using the at least one parameter, analytics information or ML (Machine Learning) model from which the consumer node can ascertain whether there is any user plane traffic anomalies; andsend (304), to the consumer node, a response based on the analytics information or ML model.
15. The data analytics server of claim 14, wherein the data analytics circuitry (216) is further configured to implement a method according to any one of claims 2 to 12.
16. A method for execution by a data analytics server (220), comprising:receiving (301), from a consumer node, a request message including at least one parameter concerning user plane traffic with sites that are malicious and / or forbidden;acquiring (302), from a user plane node, user plane traffic based on the at least one parameter of the request message;generating (303), based on the user plane traffic acquired using the at least one parameter, analytics information or ML (Machine Learning) model from which the consumer node can ascertain whether there is any user plane traffic with sites that are malicious and / or forbidden; andsending (304), to the consumer node, a response based on the analytics information or ML model.
17. The method of claim 16, wherein the analytics information or ML model is configured to identify whether there is user plane traffic with sites that are malicious and / or forbidden.
18. The method of claim 17, wherein the at least one parameter comprises at least one of malicious and / or forbidden sites indication, content filtering categories and / or parental control categories.
19. The method of claim 18, wherein the request message further comprises at least one of vendor ID (Identifier), ML model filter information, target of ML model reporting, ML model target period, time when model is needed, inference input data information, ML model monitoring information, ML model interoperability information, indication of support for multiple ML models, and number of ML models and accuracy level(s) of interest.
20. The method of any one of claims 17 to 19, further comprising:determining whether an existing trained ML model can be used; and / ordetermining whether triggering further training for the existing trained ML model is needed.
21. The method of claim 20, comprising:upon determining that further training for the existing trained ML model is needed, initiating data collection to retrieve information relative to malicious and / or forbidden sites, and training the existing trained ML model.
22. The method of any one of claims 16 to 21, wherein the analytics information or ML model comprises analytics information, and the response includes at least some of the analytics information.
23. The method of any one of claims 16 to 21, wherein the analytics information or ML model comprises an ML model, and the response includes an ML model identifier and ML model information.
24. The method of claim 23, wherein the analytics information or ML model further comprises ML model filter information and / or target of ML model reporting.
25. The method of claim 23 or claim 24, wherein the analytics information or ML model further comprises at least some of a validity period, a spatial validity, training input data information, and ML model accuracy Information.
26. A non-transitory computer readable medium (219) having recorded thereon statements and instructions that, when executed by a processor (217) of an data analytics server (220), configure the data analytics server to implement a method according to any one of claims 16 to 25.
27. A data analytics server (220) comprising:a network interface (215) configured to communicate with other network nodes;data analytics circuitry (216) coupled to the network interface and configured to:receive (301), from a consumer node, a request message including at least one parameter concerning user plane traffic with sites that are malicious and / or forbidden;acquire (302), from a user plane node, user plane traffic based on the at least one parameter of the request message;generate (303), based on the user plane traffic acquired using the at least one parameter, analytics information or ML (Machine Learning) model from which the consumer node can ascertain whether there is any user plane traffic with sites that are malicious and / or forbidden; andsend (304), to the consumer node, a response based on the analytics information or ML model.
28. The data analytics server of claim 27, wherein the data analytics circuitry (216) is further configured to implement a method according to any one of claims 17 to 25.
