Systems and methods for artificial intelligence / machine learning based quality of service flow to data radio bearer mapping design
AI/ML models in wireless communication systems address the adaptability and efficiency issues of QoS frameworks by dynamically mapping IP flows to DRBs, improving QoS guarantees and reducing performance loss.
Patent Information
- Application Number
- PCT/CN2024/101044
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-06-24
- Publication Date
- 2026-01-02
AI Technical Summary
Existing QoS frameworks in wireless communication systems face challenges in adaptability to ever-changing applications and services due to semi-static NAS and AS level mappings, which can lead to QoS guarantee mismatches and inefficiencies in managing dynamic radio channels and UE-specific information.
Implementing AI/ML models for dynamic QoS flow to DRB mapping that adapt to real-time changes in applications, radio conditions, and UE mobility, reducing performance loss by predicting QoS flow arrival times and optimizing DRB allocation.
Enhances QoS guarantee by reducing packet errors, delays, and increasing throughput through timely and adaptive mapping, while maintaining UE privacy and minimizing signaling overhead.
Smart Images

Figure CN2024101044_02012026_PF_FP_ABST
Abstract
Description
SYSTEMS AND METHODS FOR ARTIFICIAL INTELLIGENCE / MACHINE LEARNING BASED QUALITY OF SERVICE FLOW TO DATA RADIO BEARER MAPPING DESIGNTECHNICAL FIELD
[0001] This application relates generally to wireless communication systems, including the implementation of artificial intelligence (AI) / machine learning (ML) models within various quality of service (QoS) frameworks.BACKGROUND
[0002] Wireless mobile communication technology uses various standards and protocols to transmit data between a base station and a wireless communication device. Wireless communication system standards and protocols can include, for example, 3rd Generation Partnership Project (3GPP) Long Term Evolution (LTE) (e.g., 4G) , 3GPP New Radio (NR) (e.g., 5G) , and Institute of Electrical and Electronics Engineers (IEEE) 802.11 standard for Wireless Local Area Networks (WLAN) (commonly known to industry groups as ) .
[0003] As contemplated by the 3GPP, different wireless communication systems' standards and protocols can use various radio access networks (RANs) for communicating between a base station of the RAN (which may also sometimes be referred to generally as a RAN node, a network node, or simply a node) and a wireless communication device known as a user equipment (UE) . 3GPP RANs can include, for example, Global System for Mobile communications (GSM) , Enhanced Data Rates for GSM Evolution (EDGE) RAN (GERAN) , Universal Terrestrial Radio Access Network (UTRAN) , Evolved Universal Terrestrial Radio Access Network (E-UTRAN) , and / or Next-Generation Radio Access Network (NG-RAN) .
[0004] Each RAN may use one or more radio access technologies (RATs) to perform communication between the base station and the UE. For example, the GERAN implements GSM and / or EDGE RAT, the UTRAN implements Universal Mobile Telecommunication System (UMTS) RAT or other 3GPP RAT, the E-UTRAN implements LTE RAT (sometimes simply referred to as LTE) , and NG-RAN implements NR RAT (sometimes referred to herein as 5G RAT, 5G NR RAT, or simply NR) . In certain deployments, the E-UTRAN may also implement NR RAT. In certain deployments, NG-RAN may also implement LTE RAT.
[0005] A base station used by a RAN may correspond to that RAN. One example of an E-UTRAN base station is an Evolved Universal Terrestrial Radio Access Network (E-UTRAN) Node B (also commonly denoted as evolved Node B, enhanced Node B, eNodeB, or eNB) . One example of an NG-RAN base station is a next generation Node B (also sometimes referred to as a g Node B or gNB) .
[0006] A RAN provides its communication services with external entities through its connection to a core network (CN) . For example, E-UTRAN may utilize an Evolved Packet Core (EPC) while NG-RAN may utilize a 5G Core Network (5GC) .
[0007] BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS
[0008] To easily identify the discussion of any particular element or act, the most significant digit or digits in a reference number refer to the figure number in which that element is first introduced.
[0009] FIG. 1 illustrates an example of a quality of service (QoS) flow based QoS framework.
[0010] FIG. 2A illustrates an example of a QoS mapping protocol stack.
[0011] FIG. 2B illustrates an example of a downlink (DL) service data adaption protocol (SDAP) protocol data unit (PDU) and an uplink (UL) SDAP PDU used in the QoS mapping protocol stack.
[0012] FIG. 3 illustrates an example of a 5G QoS model.
[0013] FIG. 4 illustrates a diagram of SDAP use.
[0014] FIG. 5 illustrates various examples of AI / ML based mapping, according to embodiments herein.
[0015] FIG. 6 illustrates a first example of AI / ML based internet protocol (IP) flow to QoS flow mapping using an AI / ML model, according to embodiments herein.
[0016] FIG. 7 illustrates a second example of AI / ML based IP flow to QoS flow mapping using an AI / ML model, according to embodiments herein.
[0017] FIG. 8 illustrates a third example of AI / ML based IP flow to QoS flow mapping using an AI / ML model, according to embodiments herein.
[0018] FIG. 9 illustrates an example of inputs used by and AI / ML model performing IP flow to QoS flow mapping, according to embodiments herein.
[0019] FIG. 10 illustrates an example of AI / ML based IP flow to QoS flow mapping through the use of an AI / ML correlation detector, according to embodiments herein.
[0020] FIG. 11 illustrates an example of AI / ML based PDU to QoS flow mapping, according to embodiments herein.
[0021] FIG. 12 illustrates a flow diagram for using a UE-sided AI / ML model to map various IP flows to QoS flows, according to embodiments herein.
[0022] FIG. 13 illustrates a flow diagram of using a network-sided AI / ML model to map various IP flows to QoS flow (s) , according to embodiments herein.
[0023] FIG. 14A, FIG. 14B, and FIG. 14C illustrate examples of AI / ML based QoS flow to DRB mapping, according to embodiments herein.
[0024] FIG. 15 illustrates a flow diagram for using a network-sided AI / ML model for QoS flow to DRB mapping, according to embodiments herein.
[0025] FIG. 16 illustrates a flow diagram of using a two-sided AI / ML model for QoS flow to DRB mapping, according to embodiments herein.
[0026] FIG. 17 illustrates an example of a bipartite network, according to embodiments herein.
[0027] FIG. 18 illustrates an example of connections between QFIs to DRBs at different instances, according to embodiments herein.
[0028] FIG. 19 illustrates an example of reinforced learning (RL) based AI / ML model training, according to embodiments herein.
[0029] FIG. 20 illustrates an example of QFI-DRB mapping when no input features are detected, according to embodiments herein.
[0030] FIG. 21 illustrates a flow diagram of triggering AI / ML model inference, according to embodiments herein.
[0031] FIG. 22 illustrates an example of AI / ML based QoS flow data arrival time prediction, according to embodiments herein.
[0032] FIG. 23 illustrates a flow diagram for using a UE-sided AI / ML model for QoS flow to DRB mapping based on predicted QoS arrival times, according to embodiments herein.
[0033] FIG. 24 illustrates an example of an AI / ML model used for QoS flow to DRB mapping, according to embodiments herein.
[0034] FIG. 25 illustrates an example of using QoS flow data arrival time probabilities for QoS flow to DRB mapping, according to embodiments herein.
[0035] FIG. 26 illustrates a flow diagram for AI / ML based QoS flow to DRB mapping where a QoS flow data arrival prediction may be used in QFI-DRB mapping, according to embodiments herein.
[0036] FIG. 27 illustrates an example of AI / ML based QoS flow to DRB mapping using an AI / ML model correlation detector for QoS flows, according to embodiments herein.
[0037] FIG. 28 illustrates an example of AI / ML model based IP flow to DRB mapping, according to embodiments herein.
[0038] FIG. 29 illustrates an example of a protocol stack for IP flow to DRB mapping, according to embodiments herein.
[0039] FIG. 30 illustrates a flow diagram of using a network-sided AI / ML model for mapping various IP flows to a current set of DRB (s) , according to embodiments herein.
[0040] FIG. 31 illustrates a flow diagram of using a two-sided AI / ML model for mapping various IP flow to a current set of DRB (s) , according to embodiments herein.
[0041] FIG. 32 illustrates an example of AI / ML model based packet / IP flow to DRB mapping that generates predicted DRB information, according to embodiments herein.
[0042] FIG. 33 illustrates a flow diagram of using a UE-sided AI / ML model for mapping various IP flows to a future set of DRB (s) , according to embodiments herein.
[0043] FIG. 34 illustrates an end-to-end example of a QoS flow framework, according to embodiments herein.
[0044] FIG. 35 illustrates a method of a UE of a wireless communication system, according to embodiments herein.
[0045] FIG. 36 illustrates a method of a UE of a wireless communication system, according to embodiments herein.
[0046] FIG. 37 illustrates a method of a UE of a wireless communication systems, according to embodiments herein.
[0047] FIG. 38 illustrates a method of a base station of a wireless communication system, according to embodiments herein.
[0048] FIG. 39 illustrates a method of a base station of a wireless communication system, according to embodiments herein.
[0049] FIG. 40 illustrates a method of a base station of a wireless communication system, according to embodiments herein
[0050] FIG. 41 illustrates a method of a UE of a wireless communication system, according to embodiments herein.
[0051] FIG. 42 illustrates a method of a UE of a wireless communication system, according to embodiments herein
[0052] FIG. 43 illustrates a method of a UE of a wireless communication system, according to embodiments herein.
[0053] FIG. 44 illustrates a method of a base station of a wireless communication system, according to embodiments herein.
[0054] FIG. 45 illustrates a method of a base station of a wireless communication system, according to embodiments herein.
[0055] FIG. 46 illustrates a method of a base station of a wireless communication system, according to embodiments herein.
[0056] FIG. 47 illustrates a method of a UE of a wireless communication system, according to embodiments herein.
[0057] FIG. 48 illustrates a method of a UE of a wireless communication system, according to embodiments herein.
[0058] FIG. 49 illustrates a method of a base station of a wireless communication system, according to embodiments herein.
[0059] FIG. 50 illustrates a method of a base station of a wireless communication system, according to embodiments herein.
[0060] FIG. 51 illustrates an example architecture of a wireless communication system, according to embodiments disclosed herein.
[0061] FIG. 52 illustrates a system for performing signaling between a wireless device and a network device, according to embodiments disclosed herein.DETAILED DESCRIPTION
[0062] Various embodiments are described with regard to a UE. However, reference to a UE is merely provided for illustrative purposes. The example embodiments may be utilized with any electronic component that may establish a connection to a network and is configured with the hardware, software, and / or firmware to exchange information and data with the network. Therefore, the UE as described herein is used to represent any appropriate electronic component.
[0063] FIG. 1 illustrates an example of a quality of service (QoS) flow based QoS framework 100.
[0064] In some wireless communication systems, a QoS flow based QoS framework 100 may be used. In the QoS flow based QoS framework 100, various IP flows 102 may each be made up of / use one or more packets that are meant for transmission. Various types of IP flows 102 are possible. FIG. 1 illustrates, by way of example, a best effort IP flow 156, a first service video IP flow 158, a second service video IP flow 160, a third service stream IP flow 162, a fourth service stream IP flow 164, a voice IP flow 166, and a video IP flow 168.
[0065] These IP flows 102 may be organized into different protocol data units (PDUs) . As illustrated in the example of FIG. 1, there is an internet PDU 104 that includes the best effort IP flow 156, the first service video IP flow 158, the second service video IP flow 160 and the third service stream IP flow 162; a service specific PDU 106 that includes the fourth service stream IP flow 164 (where the service specific PDU 106 is specific to the fourth service) ; and an IP multimedia core network subsystem (IMS) PDU 108 that includes the voice IP flow 166 and the video IP flow 168.
[0066] Then, at a user plane function (UPF) 110, using a service data flow (SDF) / traffic flow template (TFT) 112 for QoS flow identifier (QFI) insertion 126, the various IP flows may be mapped to various QoS flows. For example, the best effort IP flow 156 may be mapped to a first QoS flow 114 (having a QFI of 1) . The first service video IP flow 158 and the second service video IP flow 160 may be mapped to the second QoS flow 116 (having a QFI of 2) . The third service stream IP flow 162 may be mapped to the third QoS flow 118 (having a QFI of 3) . The fourth service stream IP flow 164 may be mapped to the fourth QoS flow 120 (having a QFI of 4) . The voice IP flow 166 may be mapped to the fifth QoS flow 122 (having a QFI of 5) . The video IP flow 168 may be mapped to the sixth QoS flow 124 (having a QFI of 6) .
[0067] At the base station 128 (e.g., at the access stratum (AS) layer) , the various QoS flows may then undergo data radio bearer (DRB) mapping 136, where the various QoS flows are mapped to various DRBs using service data adaption protocol (SDAP) configurations. For example, the first QoS flow 114 may be mapped to a first DRB 138 using a first SDAP configuration 130. The second QoS flow 116 and the third QoS flow 118 may be mapped to a second DRB 140 using the first SDAP configuration 130. The fourth QoS flow 120 may be mapped to a third DRB 142 using the second SDAP configuration 132. The fifth QoS flow 122 may be mapped to a fourth DRB 144 using a third SDAP configuration 134. The sixth QoS flow 124 may be mapped to a fifth DRB 146 using the third SDAP configuration 134. Note that the various DRBs may be used for further transmission of the IP flow from the base station 128 to the UE 148.
[0068] Then, at the UE 148, the various DRBs may have appropriate SDAP + TFT processing applied to them to retrieve back the initial IP flows 102. For example, the first DRB 138 and the second DRB 140 may have the first SDAP + TFT processing 150 applied to them to retrieve back the best effort IP flow 156, the first service video IP flow 158, the second service video IP flow 160 and the third service stream IP flow 162. The third DRB 142 may have the second SDAP + TFT processing 152 applied to it to retrieve back the fourth service stream IP flow 164. The fourth DRB 144 and the fifth DRB 146 may have the third SDAP + TFT processing 154 applied to them to retrieve back the voice IP flow 166 and the video IP flow 168.
[0069] As has been shown in FIG. 1, it may be that a QoS flow is the finest granularity of QoS differentiation within in PDU session, and that a QoS flow is identified by a QFI within a PDU session.
[0070] In some cases, there are two levels of mapping: a non-access stratum (NAS) level and an AS level. For the NAS level, the user plane (UP) traffic (i.e., IP flow) may be mapped to the QoS flow (e.g., as in the QFI insertion 126) based on a core-network-provided QoS profile and QoS rules. The core network may provide the QoS profile to the RAN via an N2 communication. Also, the core network may provide the QoS rules to the UE via an N1 communication. Note that an unknown IP flow may be mapped a QoS flow associated with the default QoS rules.
[0071] At the AS level, a QoS flow may be mapped to a DRB (e.g., as in the DRB mapping 136) . The DRB may be the finest treatment of granularity in the AS. Note that both explicit way (s) of mapping (radio resource control (RRC) configuration) and implicit way (s) of mapping (e.g., using an SDAP configuration) may be supported.
[0072] Further, 1: 1 and M: 1 mapping may be allowed for both levels of mapping. Note that M: 1 mapping may be allowed corresponding to cases where the number of QoS flows may be larger than the number of DRBs (e.g., as illustrated with respect to the second QoS flow 116, the third QoS flow 118, and the second DRB 140 of FIG. 1) .
[0073] In some wireless communication mechanisms, a 5G QoS identifier (5QI) is associated with QoS characteristics giving guidelines for setting node-specific parameters for each QoS flow. Standardized or pre-configured QoS characteristics may be derived from the 5QI value and may not be explicitly signaled. Note that signaled QoS characteristics may be included as part of the QoS profile. Further, the QoS characteristics may include, for example, a priority level, a packet delay budget (including a core network packet delay budget) , a packet error rate, an averaging window, and / or a maximum data burst volume. Other possible kinds of QoS characteristics are provided in, for example, 3GPP Technical Specification (TS) 23.501 V. 15.0.0, December 2017.
[0074] FIG. 2A illustrates an example of a QoS mapping protocol stack 200.
[0075] In some mechanisms, an SDAP layer supports various functionalities in the protocol stack for QoS mapping. It should be understood that the protocol stack includes an SDAP layer, a packet data convergence protocol (PDCP) layer, a radio link control (RLC) layer, and a medium access control (MAC) layer, where the SDAP layer uses SDAP SDUs 202 and corresponding headers, the PDCP layer uses PDCP SDUs 204 and a corresponding headers, the RLC layer uses RLC SDUs 206 and a corresponding headers, and the MAC layer uses MAC SDUs 208 and corresponding headers.
[0076] Supported functionalities in various embodiments may include, for example, mapping between a QoS flow and a DRB and marking a QFI in both downlink (DL) and uplink (UL) IP flows. Note that a single protocol entity of the SDAP may be configured for each individual PDU session.
[0077] FIG. 2B illustrates an example of a DL SDAP PDU 210 and a UL SDAP PDU 218 used in the QoS mapping protocol stack.
[0078] At an SDAP layer of a QoS mapping protocol stack, a DL SDAP PDU 210 may be used. A DL SDAP PDU 210 may include a one bit reflective QoS flow to DRB mapping indication (RDI) 212, a one bit reflective QoS indication (RQI) 214 and a six bit QFI 216. When the RDI 212 is one, the UE may update the QoS flow to DRB mapping for uplink. When the RQI 214 is one, the UE informs the NAS that the service data flow (SDF) to QoS mapping rules have been updated.
[0079] Additionally or alternatively, a UL SDAP PDU 218 may be used at the SDAP layer. The UL SDAP PDU 218 includes a one bit D / C 220 field, a one bit R 222 field, and a six bit QFI 224. The one bit D / C 220 field may be set to one to indicate a data PDU.
[0080] Note that the aforementioned mapping may be configured by the network and the UE may not be allowed to autonomously change the mapping configured by the network. One exception may be that the UE may map a QoS flow which is not configured with a mapping rule to a default DRB.
[0081] FIG. 3 illustrates an example of a 5G QoS model 300.
[0082] The 5G QoS model 300 may include a UE 302, an access network (AN) 304 and a UPF 306. Data IP flows from application (s) may reside at both the UE 302 (e.g., UL data packets 314) and the UPF 306 (e.g., DL data packets 316) . Further, QoS rules for mapping UL packets to QoS flows and apply QoS flow marking may be applied at the UE 302 and the UPF 306. Additionally, QoS flows for all IP flows marked with the same QFI may be used at the UE 302 and the UPF 306. A mapping of QoS flows to AN resources (e.g., DRB (s) ) may be used between the UE 302 and the AN 304. Note that while FIG. 3 illustrates a flow of packets between the UPF 306 and the AN 304 that correspond to a same PDU session, multiple PDU sessions are possible within the 5G QoS model 300.
[0083] Corresponding to a NAS level, the UE 302 may use 308 the QoS rules (corresponding to the packet detection rules (PDRs) ) to map UL IP flows to QoS flows. Additionally, there may be a QoS flow to DRB mapping 310 at the UE 302 corresponding to an AS level. Note that the UPF 306 may use PDRs 312 which classify IP flows for QoS flow marking and other actions in the DL.
[0084] In some wireless communication systems, a 5G QoS model 300 may use one or more of a PDU session, a QoS flow identifier (ID) and a QoS binding. The PDU session (e.g., one multiple possible PDU sessions) may contain multiple QoS flows. The QFI may be used as a marker for traffic, and traffic that is marked with the same QFI receives the same QoS treatment. The QFI is used as a user-plane marking on N3 / N9 communication and may be unique within a PDU session. It may also be carried in a radio header. Note that a range of QFIs may map directly to standardized 5QIs and a default allocation and retention priority (ARP) value. The QoS binding may include DL and UL aspects where the DL aspects are performed by the network and the UL aspects are performed by the UE either via explicit QoS signaling or via reflective QoS based on DL data QoS marking.
[0085] In some mechanisms, QoS models may perform QoS mapping using a two-step approach. In a first step, QoS flow classification may be performed according to QoS rules / PDRs. The UE NAS and unified data function (UDF) may perform IP flow / packet to QoS flow mapping. Additionally, this first step classifies IP flows as part of a traffic detection (or application detection) process for QoS enforcement in the wireless communication system.
[0086] In a second step, QoS flow to DRB mapping may be performed. The AS (the SDAP layer) may perform mapping where different QoS flows may be mapped to the same or different DRBs (e.g., N: 1 or 1: 1) . Several QoS flows belonging to the same PDU session may be mapped to the same DRB, while QoS flows belonging to different PDU sessions may not be mapped to the same DRB. Note that the RAN may add new DRBs with corresponding QFI mappings to fulfill the QoS characteristics of a QoS flow. Further, the UE may determine UL data QoS binding either via explicit QoS signaling or via reflective QoS based on DL data QoS marking.
[0087] In some wireless communication systems, the core network may be in control of IP flow filters in the wireless communication system. For a given SDF, the binding of service requirements and QoS flows may be based on policy and charging control (PCC) rules provided by a policy control function (PCF) (e.g., as provided in 3GPP TS 23.501 V15.0.0 (December 2017) , 3GPP TS 23.502 V15.0.0 (December 2017) and / or 3GPP TS 23.503 V15.0.0 (December 2017) ) . A system management function (SMF) may perform the binding of PCC rules to QoS flows based on the QoS and service requirements. Further, the SMF provides QoS rules to the UE and PDRs (as provided in, for example, 3GPP TS 29.244 V15.0.0 (December 2017) ) to the UPF. These may also be known as packet filters. For pre-defined PCC rules, traffic detection filters (e.g., packet filters) used in the UPF may be configured either in the SMF and may be provided to the UPF, as SDF filter (s) , or be configured in the UPF, as an application detection filter identified by an application identifier.
[0088] In some mechanisms, for QoS flow to DRB mapping at the AS level, mapping between a QoS flow and a DRB may be decided by the network. In some cases, explicit mapping may be used for UL traffic. In such cases, RRC signaling may be used to indicate which QoS flow is sent over which DRB. In some other cases, reflective mapping may be used for UL traffic, where the UE sends UL traffic over the same DRB in which it received associated / corresponding DL traffic. In such cases, the UE may monitor DL traffic for which DRB a QoS flow was received in and update the mapping table accordingly. In some such cases, the UE may update the table when the DL packet has an RQI bit that is set.
[0089] Such mechanisms make it possible for the network to send DL data for a QoS flow in different DRBs, and may be useful for the network to provide a different priority for some IP flows with a QoS flow.
[0090] FIG. 4 illustrates a diagram 400 of SDAP use. In some wireless communication systems, a QoS model may include QoS flow to DRB mapping, where the QoS flow to DRB mapping is an explicit mapping using an RRC signaling QFI-DRB mapping rule (an SDAP configuration) to indicate which QoS flow is sent over which DRB. Optionally, a reflective UL mapping may be used, and may be based on a received DL QFI. Note that the QoS flow to DRB mapping may use a default DRB if no mapping rule is defined.
[0091] In some instances SDAP data PDU (e.g., under the SDAP configuration) includes a header (e.g., made up of an RDI 426, an RQI 428, and / or a QFI 430) depending on network configuration. A UL SDAP header may be used if more than one flow is mapped to a DRB and a DL SDAP header may be needed if the optional reflective mapping is used. Separate RQI 428 and RDI 426 bits for NAS and AS signaling may be used to update reflective mapping.
[0092] In some instances, an SDAP control PDU (e.g., under the SDAP configuration) includes a header (e.g., made up of a D / C 432, reserved bits R 434, and / or a QFI 436) may use UL end-marker generation for flow re-mapping.
[0093] In some cases, there may be one SDAP entity for each individual PDU session when mapping between a QoS flow and a DRB. The SDAP entity may optionally perform marking of QFI in both UL and DL IP flows. Further, data IP flows may include, if configured, a QoS marking (corresponding to QFI and RQI, RDI) in the SDAP header.
[0094] An example diagram 400 of the use of an SDAP is illustrated in FIG. 4. The diagram 400 illustrates a transmitting SDAP entity 404 mapping 408 of a QoS flow 402 to a DRB. If an SDAP header is configured 410 for use, an SDAP header may be added 414 to the traffic, which is then sent over a radio interface (e.g., Uu) 416 to a receiving SDAP entity 406 on the mapped DRB. If an SDAP header is not configured 412 for use, the traffic is sent over a radio interface (e.g., Uu) 416 to a receiving SDAP entity 406 on the mapped DRB (without the use of an SDAP header) .
[0095] Then, at the receiving SDAP entity 406, if the SDAP is configured 418 for use, reflective QoS flow to DRB mapping 422 may be performed. Then, the SDAP header being removed 424, the remaining traffic is associated with the QoS flow 402. If the SDAP header is not configured 420 the result is that the traffic is associated with the QoS flow 402 at the receiving SDAP entity 406.
[0096] Various issues may be encountered when using existing QoS frameworks. One issue may be that NAS level mapping is configured semi-statically by a core network configuration, and therefore not sufficiently adaptable to ever-changing new applications / services. For example, many new applications / services may frequently become used, each of which will be mapped to default QoS flow if the core network does not identify them. Further, it may be difficult for the core network to maintain the real-time updates of these applications / services. Additionally, for NAS level mapping, the use of a static M: 1 mapping may sacrifice some of an IP flow's QoS guarantee. In principle, all the IP flows mapped to same QoS flow may be expected to have similar QoS parameters, but the realities of the use of static mapping rule by the core network may incur some mismatch.
[0097] With respect to AS level mapping, it may be difficult for existing QoS frameworks to allow for / fully consider UE sensitive information (e.g. mobility orientation, UE contextual information) in the mapping. The UE may not be willing to share some information with the network due to privacy concerns (e.g., moving trajectory) . Additionally, AS level mapping may be semi-static as configured by a network configuration, meaning that it may not be able to timely adapt to channel and / or loading changes.
[0098] Further, in some cases, the use of two level (AS level and NAS level) mapping may sacrifice some QoS guarantee, due to overhead managing across two levels. For example, in cases not implementing reflective QoS (which may not be widely deployed) , QoS adaptation may incur extra signaling exchanges between the NAS level and the AS level. Further, due to layered design principles (e.g., as used in 3GPP) , NAS level mapping may not take fading channel information into account.
[0099] Embodiments herein introduce different levels of artificial intelligence (AI) / machine learning (ML) use for various possible QoS frameworks for wireless communication. Instead of static NAS / AS level mapping (as may be utilized in current mechanisms) , the UE and / or the network may use a trained AI / ML model to derive the best mapping at different levels (i.e., at the AS level and at the NAS level) .
[0100] FIG. 5 illustrates various examples of AI / ML based mapping, according to embodiments herein.
[0101] In some embodiments, an AI / ML model may be used to perform mapping from various IP flows to QoS flows 506. For example, as illustrated in FIG. 5, it may be that an AI / ML model 502 takes various IP flows 504 (such as voice IP flows, video IP flows and internet IP flows) from varying services as inputs . The AI / ML model 502 may then be used to perform mapping to the QoS flows 506. As illustrated, the AI / ML model 502 maps a first IP flow to a first QoS flow, a second and third IP flow to a second QoS flow, and a fourth IP flow to a third QoS flow. Such AI / ML model (s) may utilize classification with supervised learning and / or unsupervised learning.
[0102] In some other embodiments, an AI / ML model may be used to assist in performing mapping from QoS flows to DRB (s) . For example, as illustrated in FIG. 5, it may be that an AI / ML model 508 may take various QoS flows 510 as inputs and may predict confidence levels / probabilities for arrival times of data of the QoS flows 512. Then, such predicted arrival times of data of the QoS flows 512 may be used to update an SDAP configuration that may be used for QoS flow to DRB mapping. Depending on whether the current DRB set is used for mapping or a future DRB set is used for mapping, a real-time mapping from QoS flow to the current DRB set may be used or a prediction for QoS flow data arrival times may be used.
[0103] In yet some other embodiments, an AI / ML model may be used to perform a single mapping from various IP flows to DRB (s) . For example, as illustrated in FIG. 5, it may be that an AI / ML model 514 uses various IP flows 516 (such as voice IP flows, video IP flows and internet IP flows) from varying services as inputs, and additionally may use radio conditions and UE mobility information as inputs. Then, the AI / ML model 514 may map the various IP flows 516 to DRB (s) 518. As illustrated, the AI / ML model 514 maps the first, second, and third IP flows to a first DRB and a fourth IP flow to a second DRB. Depending on whether a current DRB set is used or a future DRB set is used, real-time mapping from IP flows to a current DRB set may be used or a predicted mapping from IP flows to a future DRB set may be used.
[0104] Note that embodiments herein also discuss procedures and signaling of data collection, AI / ML model training and performance monitoring for AI / ML models used for mapping.
[0105] Beneficially, embodiments herein reduce the performance loss due to QoS flow mapping mismatch, and thereby a reduced packet error rate, an decreased delay, and / or an increased throughput may be achieved.
[0106] As compared with current mechanisms (such as mechanisms using statically configured QoS rule (s) ) , a UE using AI / ML models as introduced here may more timely map ever-changing new applications / services to the most suitable QoS flow. Accordingly, performance loss due to QoS flow mapping mismatch thus may be alleviated according to embodiments for IP flow to QoS flow and IP flow to DRB mapping.
[0107] Further embodiments herein may allow for better adaptability to quickly-changing radio channel (s) and loading statuses. For example, the UE may report the prediction of QoS flow arrival information to the network, which may reduce the mismatch between buffer status report (BSR) reporting and network reconfiguration of QoS flow to DRB mapping.
[0108] Additionally, embodiments herein may reflect UE privacy-sensitive information in the mapping, without exposure of UE privacy, may reduce procedure / signaling duplication of two level mapping, may alleviate the QoS / quality of experience (QoE) loss cause by M: 1 mapping and / or may ensure aligned mapping between the network and the UE.
[0109] Various embodiments herein may use the concept (s) of QoS flow and SDAP protocol stack, and / or may use a design that one QoS flow is associated with one standardized 5QI or non-standardized metric set (e.g., packet loss, throughput, delay) .
[0110] In some embodiments, an AI / ML model may be used to perform real-time mapping from various packets (i.e., IP flow (s) ) to QoS flow (s) . The core network may configure various configurations according to various cases.
[0111] FIG. 6 illustrates a first example of AI / ML based IP flow to QoS flow mapping using an AI / ML model 602, according to embodiments herein.
[0112] The AI / ML model 602 may be used to perform mapping for all known IP flows. For example, the AI / ML model 602 may use as inputs: a first IP flow 604 (a voice IP flow) , a second IP flow 606 (a video IP flow from a first service) , a third IP flow 608 (a video IP flow from a second service) , a fourth IP flow 610 (an internet IP flow) , and optionally assistance information 612 including RAN and / or UE specific data. Then, the AI / ML model 602 may map the various input IP flows to QoS flow (s) . For example, the first IP flow 604 may be mapped to a first QoS flow 614. The second IP flow 606 and the third IP flow 608 may be mapped to a second QoS flow 616. The fourth IP flow 610 may be mapped to the third QoS flow 618.
[0113] FIG. 7 illustrates a second example of AI / ML based IP flow to QoS flow mapping using an AI / ML model 718, according to embodiments herein.
[0114] The AI / ML model 718 may be used to perform mapping for a configured set of IP flows, as configured by the network or by the UE. The core network may configure various sets of IP flows (i.e., filtering IP flows into various sets) where a first set of IP flows is mapped according to QoS rules and a second set of IP flows is mapped through the use of the AI / ML model 718. For example, a first IP flow 702 (a voice IP flow) , a second IP flow 704 (a video IP flow from a first service) , a third IP flow 706 (a sensing IP flow) , and a fourth IP flow 708 (an extended reality (XR) IP flow) , may be filtered 710 based on coding 726 from a core network into a first set 712 and a second set 714. The first set 712 includes the first IP flow 702 and the second IP flow 704, while the second set 714 includes the third IP flow 706 and the fourth IP flow 708. The first set 712 may have QoS rules 716 applied to it for mapping to QoS flows (as these IP flows apply to the QoS rules 716) , while the second set 714 is taken as inputs to the AI / ML model 718 for mapping to QoS flows. As a result, the first IP flow 702 is mapped to the first QoS flow 720 and the second IP flow 704 is mapped to the second QoS flow 722 according to the QoS rules 716, while the third IP flow 706 is mapped to the second QoS flow 722 and the fourth IP flow 708 is mapped to the third QoS flow 724 by the AI / ML model 718.
[0115] FIG. 8 illustrates a third example of AI / ML based IP flow to QoS flow mapping using an AI / ML model 814, according to embodiments herein.
[0116] The AI / ML model 814 may be used to perform mapping for any IP flows that are not matched with any QoS rules. The core network may provide a set of QoS rules which may be used for mapping some IP flows. For all the IP flows that are not covered by the QoS rules, the AI / ML model 814 may be instead used to perform the mapping to QoS flows. Note that this represents a change from existing mechanisms, where such IP flows are rather mapped as a matter of routine to a default QoS flow associated with a default QoS rule.
[0117] For example, a first IP flow 802 (a voice IP flow) , a second IP flow 804 (a video IP flow from a first service) , a third IP flow 806 (a sensing IP flow) , and a fourth IP flow 808 (an XR IP flow) , may be analyzed to determine whether each IP flow is covered by the QoS rules 810. If the IP flow does not match 812 the QoS rules 810 (in FIG. 8, this is the case for the third IP flow 806 and the fourth IP flow 808) , then the IP flows may be taken as inputs to an AI / ML model 814. If the IP flow does match the QoS rules 810, then the QoS rules 810 may be used for mapping to QoS flows. As illustrated by way of example in FIG. 8, the first IP flow 802 is covered by the QoS rules 810, thus the QoS rules 810 are used to map the first IP flow 802 to the first QoS flow 816. Similarly, the second IP flow 804 is covered by the QoS rules 810 and thus, the QoS rules 810 are used to map the second IP flow 804 to the second QoS flow 818. The third IP flow 806 and the fourth IP flow 808 do not match 812 the QoS rules 810, thus they are passed to the AI / ML model 814 to be taken as inputs. Then, the AI / ML model 814 maps the third IP flow 806 to the second QoS flow 818 and the fourth IP flow 808 to the third QoS flow 820, as shown.
[0118] It should be understood that using an AI / ML model for mapping IP flows to QoS flows may be understood as a classification problem, which may be mitigated via the use of supervised learning or unsupervised learning. In addition, the AI / ML model mapper may use UE-and / or RAN-specific data as input features in addition to the packet / IP flows.
[0119] With respect to the training an AI / ML model for IP flow to QoS flow mapping, the AI / ML model may be either UE-sided (trained at the UE) or network-sided (trained at the network and then transferred to the UE) .
[0120] A UE-sided AI / ML model may have some benefits over a network-sided AI / ML model, as the UE’s local traffic patterns may be well-trained in such cases (as the QoS mapping may occur at the UE) . However, a network-sided AI / ML model (where the trained model is transferred from the network to the UE) may benefit over the UE-sided model as the network is more aware of the ever-changing applications / services network due to its insight into communications with multiple UEs. In network-sided models, the UE may need to report any observed new applications / services and corresponding performance metrics to the network via RRC or NAS signaling (as part of data collection) .
[0121] Additionally, AI / ML model monitoring may be performed at either the UE or at the network according to various metrics. Some metrics may include a match factor of the QoS flow classification, while other metrics may include a system QoE or system QoS guarantee key performance indicators (KPIs) .
[0122] Note that network-sided monitoring may be used in cases where the UE is not be fully aware of system QoE. Alternatively, in some cases it may be that the network monitors uplink signaling while the UE monitors downlink signaling.
[0123] FIG. 9 illustrates an example of inputs used by and AI / ML model performing IP flow to QoS flow mapping, according to embodiments herein.
[0124] Various possibilities for input features may be used in the implementation of an AI / ML model 902 (e.g., whether UE-or network-sided) for mapping various IP flows to the QoS flows. In addition to the various IP flows (i.e., first IP flow 904, second IP flow 906, third IP flow 908, and fourth IP flow 910) and the assistance information 912 , other possible inputs to the AI / ML model 902 may further include other uplink packet / IP flows, a QoS flow candidate list 914, a QoS Identifier (e.g., a 5QI, or a 6G QoS identifier (6QI) ) , a flow / bearer identifier (e.g., a QFI 916) , QoE based indications, and / or QoS and application layer performance measurements 918 (e.g., an average of packet loss rate, a packet drop rate, a packet delay, and / or throughput, etc. ) . Further possible inputs may include radio channel and RAN level data (e.g., reference signal received power (RSRP) , signal to noise ratio (SINR) , channel quality indicator (CQI) , reference signal received quality (RSRQ) , congestion, cell load, etc. ) , UE contextual information, (i.e., UE mobility information, device telemetry, etc. ) , and historical grant information.
[0125] Possible output features for the AI / ML model 902 may include a packet / IP flow-QoS flow mapping table.
[0126] FIG. 10 illustrates an example of AI / ML based IP flow to QoS flow mapping through the use of an AI / ML correlation detector, according to embodiments herein.
[0127] In some embodiments, an AI / ML model may be used to identify a correlation between IP flows that are provided as inputs to the AI / ML model. The correlation may be identified based on a specific application that the IP flows belong to. Additionally or alternatively, the AI / ML model may try to identify correlation among the IP flows based on their historical pattern of arrival at the UE and / or at the network. Once the AI / ML model classifies IP flows as correlated (e.g., within a correlation threshold) then they may be categorized under the same correlated set i. A correlated set i typically contains more than one IP flow, hence reducing the number of IP flows that may need to be assigned to QoS flows. The correlated set i, that is the bundle of IP flows, may be used as an input to the QoS rules, where the corresponding bundle may be mapped to a particular QoS flow.
[0128] Note that it is also contemplated that in some cases, the QoS rules may (also) be based on AI / ML model implementation, rather than statically configured settings.
[0129] For example, at the application layer 1006, a first IP flow 1008 (a voice IP flow) , a second IP flow 1010 (a video IP flow corresponding to a first service) , a third IP flow 1012 (a sensing IP flow) , and a fourth IP flow 1014 (an XR IP flow) may be taken as input to an AI / ML model 1002 that is acting as a correlation detector. Then, the AI / ML model 1002 may generate correlated sets (using coding 1024 from a core network) where a first correlated set 1016 includes the first IP flow 1008 and the second IP flow 1010 and a second correlated set 1018 includes the third IP flow 1012 and the fourth IP flow 1014. Subsequently, QoS rules 1004 may be applied to the correlated sets, resulting in a mapping of the various IP flows to QoS flows based on their correlated sets. For example, the QoS rules 1004 maps first IP flow 1008 and the second IP flow 1010 of the first correlated set 1016 to a first QoS flow 1020 and the third IP flow 1012 and the fourth IP flow 1014 of the second correlated set 1018 to the second QoS flow 1022.
[0130] Note that such embodiments may result in a packet / IP flow that may be mapped to a QoS flow with a higher priority as compared to what would otherwise be the case using existing mapping mechanisms. For example, according to such embodiments, a sensing IP flow (e.g., the third IP flow 1012) , may be mapped to the second QoS flow 1022 instead of the first QoS flow 1020 due to its correlation with an XR IP flow (e.g., the fourth IP flow 1014) , which generally / frequently has a relatively higher QoS requirement than other possible IP flow types.
[0131] FIG. 11 illustrates an example of AI / ML based PDU to QoS flow mapping, according to embodiments herein.
[0132] In some embodiments, PDUs generated by an application / service may consist of various PDU sets and data bursts. The PDU sets and the data bursts may enable the RAN to identify the PDUs which carry content that the application processes as a single unit (e.g., a portion of an image or an audio frame) and the duration of a data transmission. Using a PDU set integrated handling indication (PSIHI) , the PDU classifier may know whether all PDUs of the PDU set are needed for the effective / substantive usage of data corresponding to the PDU set by the application layer. In some cases, if this is not set, then the PDUs may be split into ratios such that the minimum number of PDUs for correct usage of the PDU set by the application layer are identified as higher priority than the rest of the PDUs belonging to the PDU set when presented to the AI / ML model mapper. Each PDU set may be signaled with an importance (i.e., priority) , which identifies the relative importance (or priority) of a PDU set compared to other PDU sets. The importance of PDU sets may be provided to the AI / ML model mapper as an input for its mapping to appropriate QoS flows.
[0133] For example, a first PDU 1102 (a voice PDU) , a second PDU 1104 (a video PDU corresponding to a first service) , a third PDU 1106 (i.e., a sensing PDU) , and a fourth PDU 1108 (i.e., an XR PDU corresponding to a PDU set / data burst) may be inputted into a PDU classifier 1110. Then, the PDU classifier 1110 may classify the PDUs based on coding 1122 from a core network. In some cases, the PDU classifier 1110 may split a PDU (e.g., the fourth PDU 1108) into split PDU sets 1114 including a first PDU set and a second PDU set (e.g., PDU Set 4a and PDU Set 4b, as illustrated) . Then the various PDUs / split PDU sets may be taken as inputs to the AI / ML model 1112, which maps them to QoS flows. In the example of FIG. 11, the first PDU 1102 and the second PDU 1104 may be mapped to the first QoS flow 1116. The third PDU 1106 and the first set of the fourth PDU 1108 may be mapped to the second QoS flow 1118. The second set of the fourth PDU 1108 may be mapped to the third QoS flow 1120.
[0134] FIG. 12 illustrates a flow diagram 1200 for using a UE-sided AI / ML model to map various IP flows to QoS flows, according to embodiments herein.
[0135] The flow diagram 1200 beings with the UE 1204 transmitting 1208 a UE capability report to the base station 1206. Then, the base station 1206 may send 1210 a training configuration to the UE 1204 of one or more AI / ML models. Subsequently, the UE 1204 may collect 1212 data to be used for the AI / ML model training. Then, the UE 1204, in combination with the UE server 1202, performs 1214 offline training at the UE server 1202 of the one or more AI / ML models.
[0136] Once the AI / ML model (s) are trained, the UE 1204 may transmit 1216 a notification message to the base station 1206 where the UE 1204 notifies, to the base station 1206, which AI / ML model (s) are available in UE 1204 and their model IDs, AI / ML model applicability condition (s) (e.g. mobility speed, indoor or outdoor) or optimal AI / ML model ID. Such notification message may take the form of, for example, a MAC control element (CE) message, a UE assistance information (UAI) message or an RRC message.
[0137] Subsequently, based on UE 1204 notification message, the base station 1206 may signal 1218 which AI / ML model to activate and use, and may signal the UE to activate said AI / ML model. The signaling may take the form of, for example, downlink control information (DCI) , a MAC CE message, or an RRC message.
[0138] Then, the UE 1204 may perform 1220 inference of QoS flow mapping, according to base station 1206 configurations, and may report 1222 the inferred IP flow to QoS flow mapping to the base station 1206.
[0139] The UE 1204 and / or the base station 1206 may optionally perform 1224 performance monitoring of the AI / ML model.
[0140] Further, the base station 1206 may optionally perform 1226 life cycle management (LCM) signaling corresponding to the AI / ML model and may perform corresponding signaling to the UE 1204 (e.g., to indicate a new AI / ML model activation, a current AI / ML model deactivation, and / or AI / ML model switching) .
[0141] The UE 1204 may switch and / or deactivate 1228 the AI / ML model, (e.g., possibly in response to LCM signaling received from the base station 1206, as described) .
[0142] FIG. 13 illustrates a flow diagram 1300 of using a network-sided AI / ML model to map various IP flows to QoS flow (s) , according to embodiments herein.
[0143] The flow diagram 1300 may begin with the UE 1302 transmitting 1308 a UE capability report to the base station 1304. In response, the base station 1304 may send 1310 an configuration for training message to the UE 1302 indicating that one or more AI / ML model (s) are to be configured for training.
[0144] Then, the UE 1302 may report 1312 observations of new IP flow (s) and any related metric (s) to the base station 1304. Note that the UE 1302 may need to report 1312 its observations of new IP flows and their performance metrics (e.g., latency requirement) to base station 1304 for AI / ML model training purposes. Accordingly, the base station 1304, in combination with the network server 1306, performs 1314 offline training of one or more AI / ML models at the network server 1306. Then, the base station 1304 may transfer 1316 the trained AI / ML model (s) to the UE 1302. Note that the AI / ML model training is performed at the base station 1304, and then transferred to the UE 1302, where the AI / ML model transfer signaling may take the form of, for example, RRC signaling or DRB signaling.
[0145] The base station 1304 may then signal 1318 which AI / ML model to activate (e.g., similarly to UE-sided AI / ML procedure described previously) . The UE 1302 may perform 1320 inference of QoS flow mapping using the indicated AI / ML model. Then, the UE 1302 may report 1322 the QoS flow mapping to the base station 1304.
[0146] Further, the UE 1302 and / or the base station 1304 may optionally perform 1324 AI / ML model performance monitoring on the AI / ML model.
[0147] In some cases, the base station 1304 may perform 1326 LCM monitoring corresponding to the AI / ML model and may perform corresponding signaling to the UE 1302 (e.g., to indicate a new AI / ML model activation, current AI / ML model deactivation, and / or AI / ML model switching) .
[0148] The UE 1302 may switch and / or deactivate 1328 the AI / ML model (e.g., possibly in response to LCM signaling received from the base station 1304, as described) .
[0149] Initially, it should be assumed that various embodiments herein may use the concepts of QoS flow, SDAP protocol stacks and DRB design (as discussed elsewhere herein) .
[0150] In some embodiments, real-time QoS flow (s) may be mapped to a current set of DRB(s) . In such embodiments, UE contextual information / radio channel conditions may be used as inputs to an AI / ML model for training and inference. The radio channel conditions and / or the UE contextual information are examples of possible assistance information that may be used.
[0151] FIG. 14A, FIG. 14B, and FIG. 14C illustrate examples of AI / ML based QoS flow to DRB mapping, according to embodiments herein.
[0152] FIG. 14A illustrates a case were an AI / ML model 1412 is used to perform mapping for all known QoS flows. In the embodiment of FIG. 14A, a first QoS flow 1402, a second QoS flow 1404, and a third QoS flow 1406 (along with (optionally) radio channel conditions 1408 and / or UE contextual information 1410) may be taken as inputs to an AI / ML model 1412 used for QoS flow to DRB mapping. As mapped by the AI / ML model 1412, the first QoS flow 1402 and the second QoS flow 1404 may be mapped to a first DRB 1414 and the third QoS flow 1406 may be mapped to a second DRB 1416.
[0153] FIG. 14B illustrates a case where an AI / ML model 1428 is used to perform mapping for a configured set of QoS flows. In the embodiments of FIG. 14B, a first QoS flow 1418, a second QoS flow 1420 and a third QoS flow 1422 may be passed through a filter 1424 that uses filtering information to determine if the QoS flow is to be mapped to DRBs using an SDAP configuration 1426 or through the use of the AI / ML model 1428. The filtering information may define a configured set of QoS flows for use with the AI / ML model 1428.
[0154] FIG. 14B illustrates that the first QoS flow 1418 is not part of the configured set of QoS flows for use with the AI / ML model 1428, and accordingly is matched to the first DRB 1434 using the SDAP configuration 1426. Further, the second QoS flow 1420 and the third QoS flow 1422 are part of the configured set of QoS flows for use with the AI / ML model 1428, and accordingly are taken as inputs for the AI / ML model 1428 (along with (optionally) radio channel conditions 1430 and / or UE contextual information 1432) . The radio channel conditions 1430 and / or the UE contextual information UE contextual information 1432 are examples of possible assistance information that may be used. The AI / ML model 1428 then maps the second QoS flow 1420 to the first DRB 1434 and the third QoS flow 1422 to the second DRB 1436, as illustrated.
[0155] FIG. 14C illustrates a case where an AI / ML model is used to perform mapping for all QoS flows that are not matched with a configured QoS flow to DRB mapping (e.g., QoS flows that would otherwise be mapped to a default DRB in existing mechanisms) . For example, a first QoS flow 1438, a second QoS flow 1440, and a third QoS flow 1442 may initially be handled by an SDAP configuration 1444. In the case that QoS flow is covered by / matches a QoS flow to DRB mapping rule found in the SDAP configuration 1444, then that QoS flow may be mapped to a DRB according to the rule. Accordingly, FIG. 14C illustrates that the first QoS flow 1438 is mapped to the first DRB 1454 according to a QoS flow to DRB mapping rule of the SDAP configuration 1444. However, the second QoS flow 1440 and the third QoS flow 1442 do not match / are not covered 1452 by any QoS flow to DRB mapping rule of the SDAP configuration 1444. Accordingly, the second QoS flow 1440 and the third QoS flow 1442 are taken as inputs at the AI / ML model 1450 (along with (optionally) radio channel conditions 1446 and / or UE contextual information 1448) . The radio channel conditions 1446 and / or the UE contextual information 1448 are examples of possible assistance information that may be used. The AI / ML model 1450 then maps the second QoS flow 1440 to the first DRB 1454 and the third QoS flow 1442 to the second DRB 1456.
[0156] In still further cases, the network may configure the UE to trigger an AI / ML-based QFI-to-DRB mapping behavior when certain conditions are met. The conditions may include, for example, that the AI / ML model (trained by the UE) becomes available and ready to use, and / or that the UE observes some performance loss. It may be that by default for such cases, if the UE is not triggered, or if the AI / ML model is not available / conditions are not met, the UE may apply a fixed mapping.
[0157] With respect to training an AI / ML model for QoS flow to DRB mapping the AI / ML model may be either UE-sided (trained at the UE) or network-sided (trained at the network and then transferred to the UE)
[0158] With respect to the training of a UE-sided AI / ML model for QoS flow to DRB mapping according to embodiments herein, it may be understood that that such UE-sided AI / ML models benefit over the network-sided AI / ML models for QoS flow to DRB mapping in certain ways. For example, the training of UE-sided models can be carried out such that the UE’s local traffic patterns and other privacy-sensitive information do not leave the UE.
[0159] With respect to such UE-sided models, a training distortion metric may need to consider tradeoffs with respect to DRB usage. For example, a formula such as:
[0160] a *QoE -b *bandwidth (BW) of DRB
[0161] may be used, where a and b are weights that are set / tuned to keep the UE from greedily occupying over-qualified DRB (s) ) .
[0162] Further, when using a UE-sided model, it may be that the network sends possible DRBs, or a DRB list, to the UE that the UE may use for training and / or for inferencing a QoS flow-DRB mapping. This may allow for some DRBs to remain unoccupied or reserved.
[0163] With respect to training a network-sided AI / ML model for QoS flow to DRB mapping (where the trained AI / ML model is transferred from the network to the UE) according to embodiments herein, it may be that the network-sided AI / ML model benefits over the UE-sided AI / ML model in certain ways. For example, the network may take parameters from multiple UEs into consideration, thereby improving overall system optimization (rather than just a particular UE optimization) . To facilitate network side training, the UE may need to report observed QoS flows to the network as part of data collection.
[0164] It is also contemplated that an AI / ML model for QoS flow to DRB mapping may be trained in part at the UE and in part at the network. In one such example, the UE may first train the model based on its local observations, and then the UE may send the resulting UE-trained AI / ML model to the network for further training. The network may reflect DRB loading and the impact of other UEs using the network within the final version of the trained model, and send back the finally trained AI / ML model to the UE for use in inferencing.
[0165] In some cases corresponding to the use of an AI / ML model for QoS flow to DRB mapping, model monitoring may be performed by the network. Such monitoring may be according to one or more metrics, such as system QoE or system QoS guaranteed KPI. Such monitoring at the network is facilitated by the fact that the network can observe the loading status of different DRBs and different UEs.
[0166] FIG. 15 illustrates a flow diagram 1500 for using a network-sided AI / ML model for QoS flow to DRB mapping, according to embodiments herein.
[0167] The flow diagram 1500 begins with the UE 1502 transmitting 1508 a UE capability report to the base station 1504. Then, the base station 1504 may send 1510 a configuration message indicating an AI / ML model to configure for training. In response, the UE 1502 may report 1512 observations of new IP flow (s) and related metric (s) to the base station 1504. Then, the base station 1504, in combination with the network server 1506, perform 1514 offline training of one or more AI / ML models at the network server 1506.
[0168] Then, the base station 1504 may transfer 1516 the AI / ML model (s) to the UE 1502. Subsequently, the base station 1504 may signal 1518, to the UE 1502, which AI / ML model to activate. The UE 1502 may perform 1520 inference of QoS flow to DRB mapping using the activated AI / ML model (as determined by the base station 1504) used for QoS flow to DRB mapping. Then, the UE 1502 may report 1522 the DRB mapping to the base station 1504.
[0169] Accordingly, the base station 1504 may perform 1524 AI / ML model performance monitoring on the AI / ML model.
[0170] In some cases, the base station 1504 may perform 1526 LCM monitoring corresponding to the AI / ML model and may perform corresponding signaling to the UE 1502, (e.g., to indicate a new AI / ML model activation, current AI / ML model deactivation, and / or AI / ML model switching) .
[0171] The UE 1502 may switch and / or deactivate 1528 the AI / ML model, (e.g., possibly in response to LCM signaling received from the base station 1504, as described) .
[0172] FIG. 16 illustrates a flow diagram 1600 of using a two-sided AI / ML model for QoS flow to DRB mapping, according to embodiments herein.
[0173] The flow diagram 1600 begins with the UE 1604 transmitting 1610 a UE capability report to the base station 1606. Then, the base station 1606 may send 1612 a configuration used for AI / ML model training to the UE 1604 indicating that one or more AI / ML model (s) are to be configured for training.
[0174] The UE 1604 may report 1614 observations of QoS flow and corresponding information to the base station 1606. Then, the UE 1604, in combination with the UE server 1602, may perform 1616 offline training of one or more AI / ML models at the UE server 1602. Subsequently, the UE 1604 may transfer 1618 the trained AI / ML model (s) to the base station 1606. Then, the base station 1606, in combination with the network server 1608, may further perform 1620 offline training of the AI / ML model (s) at the network server 1608. The base station 1606 may transfer 1622 the trained AI / ML model (s) to the UE 1604.
[0175] Then, the base station 1606 may determine which of the AI / ML model (s) to activate and signal 1624 the activated AI / ML model to the UE 1604. Accordingly, the UE 1604 may perform 1626 inference of QoS flow to DRB mapping using the indicated AI / ML model. Then, the UE 1604 may report 1628 the DRB mapping to the base station 1606.
[0176] Subsequently, the base station 1606 may perform 1630 AI / ML model performance monitoring on the AI / ML model.
[0177] The base station 1606 may perform 1632 LCM monitoring corresponding to the AI / ML model and may perform corresponding signaling to the UE 1604, (e.g., to indicate a new AI / ML model activation, current AI / ML model deactivation, and / or AI / ML model switching) .
[0178] Accordingly, the UE 1604 may switch and / or deactivate 1634 the AI / ML model (e.g., possibly in response to LCM signaling received from the base station 1606, as described) .
[0179] It should be understood that, in some embodiments, a first AI / ML model may be used for IP flow to QoS flow mapping and a second AI / ML model may be used for QoS flow to DRB mapping, to accomplish IP flow to DRB mapping.
[0180] FIG. 17 illustrates an example of a bipartite network 1700, according to embodiments herein.
[0181] In some cases, sequential reinforcement learning may be used to find optimal bipartite network 1700 mapping between the QFI 1702 and the DRBs 1704. Consider two examples. In a first example, dynamic adaptation of QFI 1702 to DRB 1704 mapping may be used due to uncertain environments where the QoS values may change dynamically. In a second example, establishment or release of the QFI 1702 or the DRB 1704 may be used.
[0182] Additionally, consider the following Markov decision process (MDP) where state s is a set of defining features of mapping such as an adjacency matrix which defines the mapping between QFIs 1702 and DRBs 1704 when the entry is one. In the adjacency matrix, Nqos may be defined as the number of QFIs 1702 while Ndrb may be defined as the number of DRBs. Further, the input features of DRBs 1704 and QFIs 1702 may be i ∈ {1, ..., Ndrb} and j ∈ {1, ..., Nqos} , respectively. For example, DRB 1704 features may be an average historical measurement, channel measurement while QFI 1702 features may be a QoS requirement. Note that there may be multiple QoS attributes such as throughput and / or latency. Further, s0 ∈ s may be an initial state in which there are no edges (i.e., edge prediction 1706) , and sf ∈ s may be a final state in which every QFI 1702 is connected / mapped to DRBs 1704. Note that reward value r may be understood as normalized QoS measurements such that Here, wi is the weight across various QoS attributes, and qi, j is QoS measurement for QFI 1702 j, and attribute i.
[0183] FIG. 18 illustrates an example of connections between QFIs to DRBs at different instances, according to embodiments herein.
[0184] Additionally, the action space a may be considered as a potential connection between QFIs to DRBs. Consider a sequential connection starting from a first QFI to QFI Nqos which limits the action space at each state to Ndrb. Therefore a deep Q-network (DQN) model may be considered as where the various values Wk are neural network weights. Further, the DQN model may select the best action which maximizes the Q-value.
[0185] For example, in a first instance 1818, no QFIs 1802 may be connected to DRBs 1804, however, in a second instance 1820, after an action 1810 to maximize the Q-value, a first QFI 1806 may be connected to various DRBs 1808. Then, in a third instance 1822, a second QFI 1816 may be connected to various DRBs, with the first QFI 1812 being already connected to the first DRB 1814.
[0186] FIG. 19 illustrates an example of reinforced learning (RL) based AI / ML model training, according to embodiments herein.
[0187] In some cases, for training of the AI / ML model, deep Q-learning may be used. For example, an RL agent may learn to find optimal QFI-DRB mapping through interactions with the network and the observations (i.e., a reward value r which quantifies how the network responds) . In the UE centric QFI-DRB mapping selection, the RL agent at the UE may independently discover the sequence of actions that may maximize the reward value. In this framework, note that the RL agent resides at UE with a nominal model. To improve the performance, the UE may initiate RL based training to optimize the DQN model through a MAC CE message. In some examples, the training of the AI / ML model may be initiated through RRC reconfiguration performed by the network.
[0188] After the RL based model AI / ML model training 1914 has been initiated 1906 (as initiated by the UE 1902 and indicated to the base station 1904) , the UE 1902 may explore various actions (e.g., a QFI-DRB mapping change 1908 Q (s, a) , si ∈ s action a) . Additionally, the network may detect 1910 a QFI mapping function change (from a QFI 1916 connected to a first DRB 1918 to the QFI 1916 connected to a second DRB 1920) and the base station 1904 may respond 1912 with a reward value r (as previously discussed) . Note that there may be no need to inform the time of the change in QFI mapping as after any configuration change (as detected by the network) , the network may respond 1912 with a reward value r. Note that various iterations 1922 of said process may be performed as to optimize the reward value r.
[0189] FIG. 20 illustrates an example of QFI-DRB mapping when no input features are detected, according to embodiments herein.
[0190] For embodiments herein, there are various cases for QFI-DRB mapping depending on the availability of input features for action selection. In some cases, if there are recent input features per DRB or QFI due to an existing QFI-DRB mapping, then, the UE may consider the existing features to select a new action. In some other cases, if there is no recent input features per DRB or QFI mapping, for example, due to a new QFI or DRB establishment, the UE may request these measurements from the network.
[0191] For example, a UE 2002 may request 2006 features, a specific DRB 2014 and / or QFI 2012 from the base station 2004. Then, the base station 2004 may respond 2008 to the UE 2002 with a feature vector. Accordingly, the UE may select 2010 the best action for the node according to the received feature vector. Note that various QFIs 2012 may be mapped to various DRBs 2014, where, in some instances, a DRB may have no input features 2016, thus impacting the selection performed at the UE 2002.
[0192] FIG. 21 illustrates a flow diagram 2100 of triggering AI / ML model inference, according to embodiments herein.
[0193] The flow diagram 2100 begins with monitoring 2102 QoS values, QFIs and / or DRBs. Inference may be triggered if there is a change 2104 in the QoS values or if the QFI or DRB is released or established 2106. If the QFI or DRB is released or established 2106 then, an additional request 2108 may be made for input features of the AI / ML model for inference. Subsequently, an AI / ML model may be used 2110 for QFI-DRB mapping (i.e., a DQN model with (Q (s, a) with Ap) . Then, a change 2112 in mapping may be detected and if, yes, a change 2112 is detected the new mapping may be applied 2114.
[0194] There may be various cases that triggers AI / ML model inference. In some cases, inference may be triggered when a change in QoS values is detected. In some other cases, inference may be triggered when a QFI or a DRB is released or established. After inference, if there is any change in mapping, the UE may apply the new changes.
[0195] FIG. 22 illustrates an example of AI / ML based QoS flow data arrival time prediction, according to embodiments herein.
[0196] In some embodiments, an AI / ML model may be used to generate predictions of data arrival times for QoS flow (s) that are used to update the SDAP configuration (s) used for QoS flow to DRB mapping. In some such embodiments, for such QoS flows (s) the UE may rely on network SDAP configuration information (e.g., the network updates the QoS flow to DRB mapping configuration based on a UE-sided prediction) , where the SDAP configuration is updated using the predicted data arrival times for the QoS flows.
[0197] Beneficially, such a use of prediction may reduce the mismatch between BSR reporting and the network reconfiguration of the QoS flow to DRB mapping. Additionally, the use of such prediction may enable QoS prioritization that may be beneficial when the system is under resource constraints, using resource optimization, and / or using load balancing. For a UE-sided model (a model trained at the UE) , each QoS flow’s predicated data arrival timing (s) and associated confidence level (s) / probability (s) may be used as a training input.
[0198] Note that under such a scheme, outcomes for different future times may be different, considering variability in time with respect to the QoS flow data arrival time predictions (e.g., as shown in FIG. 22) .
[0199] FIG. 22 illustrates that various previous arrival times of data of a first QoS flow 2202, various previous arrival times of data of a second QoS flow 2204, and various previous arrival times of data of a third QoS flow 2206 (e.g., t1, t2, ..., tn) (along with (optionally) radio channel conditions 2208 and UE contextual information 2210) may be taken as inputs to the AI / ML model 2212. The radio channel conditions 2208 and / or the UE contextual information 2210 are examples of possible assistance information that may be used.
[0200] Then, the AI / ML model 2212 may be used to predict various future data arrival time probabilities for data of the QoS flow (s) . For example, FIG. 22 illustrates a case where the AI / ML model 2212 predicts that there is a 90%probability of data arriving on the first QoS flow 2202 at time Tm 2214, that there is an 80%probability of data arriving on the first QoS flow 2202 at time Tm+1 2216, and that there is an 80%probability of data arriving on the first QoS flow 2202 at time Tm+2 2218.
[0201] Further, the AI / ML model 2212 predicts that there is a 2%probability of data arriving on the second QoS flow 2204 at time Tm 2214, that there is a 60%probability of data arriving on the second QoS flow 2204 at time Tm+1 2216, and that there is a 70%probability of data arriving on the second QoS flow 2204 at time Tm+2 2218.
[0202] Still further, the AI / ML model 2212 predicts that there is an 80%probability of data arriving on the third QoS flow 2206 at Tm 2214, that there is a 3%probability of data arriving on the third QoS flow 2206 at time Tm+1 2216, and that there is a 95%probability of data arriving on the third QoS flow 2206 at Tm+2 2218.
[0203] These predicted QoS flow data arrival times may then be used, by the network, to update the SDAP configuration. As a result, the updated SDAP configuration may be used to perform a more optimal mapping of QoS flow (s) to DRB (s) .
[0204] Once the inference is made by the AI / ML model at UE (resulting in prediction (s) in the manner just described) , the UE reports the corresponding prediction (s) to the network, and waits for possible SDAP reconfiguration (e.g., via reflective QoS or based on RRC signaling) .
[0205] In some cases corresponding to the use of an AI / ML model for QoS flow to DRB mapping model monitoring may be performed by the network. Such monitoring may be according to one or more metrics, such as system QoE or system QoS guaranteed KPI. Such monitoring at the network is facilitated by the fact that the network can observe the loading status of different DRBs and different UEs.
[0206] FIG. 23 illustrates a flow diagram 2300 for using a UE-sided AI / ML model for QoS flow to DRB mapping based on predicted QoS arrival times, according to embodiments herein.
[0207] The flow diagram 2300 begins with the base station 2306 transmitting 2308 an SDAP configuration to the UE 2304. Then, the UE 2304 may transmit 2310 a UE capability report to the base station 2306. In response, the base station 2306 may send 2312 a configuration message to the UE 2304 for training of an AI / ML model indicating that one or more AI / ML model (s) are to be configured for training.
[0208] Then, the UE 2304 may perform 2314 data collection. Subsequently, the UE 2304, in combination with the UE server 2302, may perform 2316 offline training of the one or more AI / ML models (with the result being stored at the UE server 2302) . Then, the UE 2304 may send 2318 a notification message to the base station 2306 notifying the base station 2306 that the AI / ML model (s) have been trained. Accordingly, the base station 2306 may signal 2320 the activate one of the AI / ML model (s) and indicate such activation of the AI / ML model to the UE 2304. Then, the UE 2304 may perform 2322 inference of QoS flow to DRB mapping using the AI / ML model used for predicting the QoS flow data arrival information (i.e., QoS flow data arrival timing) and may report 2324 the predicted QoS flow arrival information to the base station 2306.
[0209] Accordingly, the base station 2306 may update 2326 the SDAP configuration, based on the predicted QoS flow arrival information, and indicate such updating of the SDAP configuration to the UE 2304.
[0210] Additionally, the base station 2306 may perform 2328 AI / ML model performance monitoring on the AI / ML model.
[0211] The base station 2306 may perform 2330 LCM monitoring corresponding to the AI / ML Model and may perform corresponding signaling to the UE 2304, (e.g., to indicate a new AI / ML model activation, current AI / ML model deactivation, and / or AI / ML model switching) .
[0212] Then, the UE 2304 may switch and / or deactivate 2332 the AI / ML model (e.g., possibly in response to LCM signaling received from the base station 2306, as described) .
[0213] FIG. 24 illustrates an example of an AI / ML model 2402 used for QoS flow to DRB mapping, according to embodiments herein.
[0214] Input features for the AI / ML model 2402may include QoS flows (e.g., the first QoS flow 2404, the second QoS flow 2406, and the third QoS flow 2408) , a candidate DRB list 2412, a QoS identifier (e.g., 5QI / 6QI 2414) , a flow / bearer identifier (e.g., QFI 2416) , a reflective QoS (e.g., RQI 2418) , a QFI to DRB mapping, QoE based indications, and / or QoS flow, DRB, and application layer performance measurements 2420 (such as, for example, an average of packet loss rate, a packet drop rate, a packet delay, and / or throughput, etc. ) . The possible input features for the AI / ML model may further include radio channel and RAN level data (e.g., RSRP, SINR, CQI, RSRQ, congestion, cell load) and / or UE contextual information 2410 (e.g., UE mobility and UE device telemetry, etc. ) .
[0215] Possible output features for the AI / ML model 2402 may include a QoS flow-DRB mapping table. In the example shown in FIG. 24, the first QoS flow 2404 and the second QoS flow 2406 are mapped to the first DRB 2422 and the third QoS flow 2408 is mapped to the second DRB 2424.
[0216] FIG. 25 illustrates an example of using QoS flow data arrival time probabilities for QoS flow to DRB mapping, according to embodiments herein.
[0217] In some embodiments, various AI / ML models may be contemplated in view of QoS flow data arrival time prediction and the QoS flow data arrival time prediction confidence level. For example, a two stage AI / ML model may be contemplated. In the first stage, each QoS flow may have an associated prediction model which predicts the next QoS flow arriving timing along with a corresponding confidence level / probability. For example for a first QoS flow the corresponding first QFI related attributes 2502 may have a first AI / ML model 2508 to predict a first probability 2514. For a second QoS flow, the corresponding second QFI related attributes 2504 may have a second AI / ML model 2510 to predict a second probability 2516. For a third QoS flow, the corresponding third QFI related attributes 2506 may have a third AI / ML model 2512 to predict a third probability 2518. The first probability 2514, the second probability 2516 and the third probability 2518 may be considered the probability of the QoS flow data arrival time between QFIs 2520 and DRBs 2522.
[0218] In the second stage, the AI / ML model 2524 developed for QFI-DRB mapping may take into account these predictions to identify future set of QFIs (i.e., QoS flows) , and may incorporate prediction confidence / probability to decide the QFI-DRB mapping in order to maximize an average weighted sum of QoS metrics. For example, the probabilities predicted in the first stage (e.g., the first probability 2514, the second probability 2516 and the third probability 2518) are used as inputs to an AI / ML model 2524 used for QFI-DRB mapping (i.e., a DQN model for QFI-DRB mapping with function Q (s, a) and adjacency matrix Ap) .
[0219] In some cases, each QFI may have an offline trained AI / ML model per QoS flow such as long short-term memory (LSTM) , recurrent neural network (RNN) or transformer to predict QoS flow data arrivals. Such models may take historical QoS flow data arrival profile (s) and predict the future QoS flow data arrival time (s) whenever there is a QoS flow data arrival (i.e., each QoS flow data arrival triggers a prediction (i.e., an AI / ML model inference) ) .
[0220] Note that, due to various prediction errors, the AI / ML model may also predict the probability of future QoS flow (s) arriving. In such cases, the adjacency matrix which defines the mapping between QFI and DRBs, may be modified. In these cases, alternatively weighted edges between QFIs and DRBs may be used in the adjacency matrix. The weight of the edges may be given by the confidence of QFI arrival prediction (s) .
[0221] FIG. 26 illustrates a flow diagram 2600 for AI / ML based QoS flow to DRB mapping where a QoS flow data arrival prediction may be used in QFI-DRB mapping, according to embodiments herein.
[0222] The flow diagram 2600 includes QFI-DRB mapping 2614 as discussed herein (e.g., in FIG. 21) , however the QFI-DRB mapping 2614 may be supplemented with a QoS flow data prediction 2608. The QoS flow data prediction 2608 may begin with determining when there is a QoS arrival 2602 in the QFI i, where i ∈ 1 : Nqos. Then, the QoS flow arrival prediction for the QFI i may be performed 2604. Accordingly, the AI / ML model (i.e., DQN model) parameters (e.g., Ap) may be updated 2606 and passed to the AI / ML model 2612 (i.e., DQN model) and used by the AI / ML model 2612 as input 2610. Otherwise, the flow diagram 2600 follows the same flow as was described in relation to the flow diagram 2100 of FIG. 21.
[0223] FIG. 27 illustrates an example of AI / ML based QoS flow to DRB mapping using an AI / ML model correlation detector for QoS flows, according to embodiments herein.
[0224] In some embodiments, an AI / ML model 2702 may be used to identify a correlation between QoS flows. The correlation may be identified based on a specific PDU session that the QoS flows belong to. Alternatively, the AI / ML may try to identify correlation among the QoS flows based on their historical pattern of data arrival at the UE and / or at the network. For example, the QoS flow related to haptic data in XR session may be correlated to video haptic data within the same session. As a result, the QoS flows may be treated in the same DRBs to reduce cognitive burden.
[0225] For example, as illustrated in FIG. 27, a first QoS flow 2708, a second QoS flow 2710, a third QoS flow 2712 and a fourth QoS flow 2714 corresponding to an application layer 2706 may be used as inputs to an AI / ML model 2702 that is used for correlation detection. Once the AI / ML model 2702 classifies the QoS flows as correlated (i.e., within a correlation threshold) , for example based on coding 2724 from a core network, the QoS flows may be categorized under the same correlated set: i. For example, the first QoS flow 614 and the second QoS flow 2710 may be in a first correlated QoS flow set 2716 and the third QoS flow 2712 and the fourth QoS flow 2714 may be in a second correlated QoS flow set 2718. Such a correlated set i typically contains more than one QoS flow, thus reducing the number of QoS flows that may need to be assigned to the DRBs at the SDAP functionality.
[0226] A correlated set i that is the bundle of QoS flows may have the SDAP configuration 2704 applied to them for mapping purposes, where the corresponding bundle may be mapped to a particular DRB. For example, FIG. 27 illustrates a case where the first correlated QoS flow set 2716 is mapped to a first DRB 2720 and the second correlated QoS flow set 2718 is mapped to a second DRB 2722. In some cases, the SDAP configuration 2704 may also be based on a (separate) AI / ML model implementation, rather than static model based solutions.
[0227] Such embodiments may result in a QoS flow that is mapped to a DRB that may support a higher QoS requirement as might otherwise be the result under other existing mapping mechanisms. For example, in the context of FIG. 27, a QoS flow (e.g., second QoS flow 2710) , may be mapped to the second DRB 2722 instead of the first DRB 2720 due to its correlation with the fourth QoS flow 2714, where the fourth QoS flow 2714 needs to meet relatively high (er) QoS requirements.
[0228] FIG. 28 illustrates an example of AI / ML model based IP flow to DRB mapping, according to embodiments herein.
[0229] Initially, note that for some embodiments herein it may be assumed that some embodiments herein may use the DRB design as discussed herein but without using the concepts of QoS flow (s) and / or an SDAP protocol stack. Note that in such cases a direct mapping from IP flows to DRB (s) may be performed (without the intermediate use of QoS flow (s) ) . Further, in some such embodiments, a new protocol stack may be introduced to account for the fact that there is no QoS flow to DRB mapping, where the SDAP layer may be removed.
[0230] In some cases, the network may configure the UE with a QoS association rule for the mapping from IP flows to DRB, where the QoS association rule may be a mapping from ports to DRB (s) . Additionally or alternatively, an AI / ML model may be used for the mapping inference.
[0231] FIG. 28 illustrates an example AI / ML model 2814 that takes a first IP flow 2802 (a voice IP flow) , a second IP flow 2804 (a video IP flow corresponding to a first service) , a third IP flow 2806 (a video IP flow corresponding to a second service) , a fourth IP flow 2808 (an internet IP flow) (along with (optionally) radio channel conditions 2810, and / or UE contextual information 2812 (e.g., mobility information, etc. ) ) as inputs. The radio channel conditions 2810 and / or the UE contextual information 2812 are examples of possible assistance information that may be used. Based on these inputs, the AI / ML model 2814 generates a mapping of IP flows to DRB (s) . As a result of using the AI / ML model 2814, the first IP flow 2802, the second IP flow 2804, and the third IP flow 2806 are mapped to the first DRB 2816 and the fourth IP flow 2808 is mapped to the second DRB 2818.
[0232] FIG. 29 illustrates an example of a protocol stack 2900 for IP flow to DRB mapping, according to embodiments herein.
[0233] For example, the protocol stack 2900 may include a PDCP layer, a RLC layer and a MAC layer. The PDCP layer may use a first PDCP SDU 2908 and a second PDCP SDU 2912, where the first PDCP SDU 2908 includes data of a first IP flow 2902, a second IP flow 2904 and a third IP flow 2906 (with a header) , and the second PDCP SDU 2912 includes data of a fourth IP flow 2910 (with a header) . The RLC layer may use a RLC SDUs (with headers) . The MAC layer may use MAC SDUs (with headers) . Note that the protocol stack 2900 does not include an SDAP layer but includes a PDCP layer, an RLC layer and a MAC layer, that use corresponding headers.
[0234] In some embodiments, real-time mapping from IP flows to a current DRB set may be performed.
[0235] With respect to the training of a UE-sided AI / ML model for IP flow to DRB mapping according to embodiments herein, it may be understood that that such UE-sided AI / ML models benefit over the network-sided AI / ML models for IP flow to DRB mapping in certain ways. For example, the training of UE-sided models can be carried out such that the UE’s local traffic patterns and other privacy-sensitive information do not leave the UE.
[0236] With respect to such UE-sided models, a training distortion metric may need to consider tradeoffs with respect to DRB usage. For example, a formula such as:
[0237] a *QoE -b *BW of DRB
[0238] may be used, where a and b are weights that are set / tuned to keep the UE from greedily occupying over-qualified DRB (s) ) . In some cases, as a result of embodiments discussed herein using direct mapping from IP flows to DRB (s) , radio channel condition (s) and / or UE mobility information may also be used as inputs for the AI / ML model in addition to the IP flows.
[0239] With respect to training a network-sided AI / ML model for IP flow to DRB mapping (where the trained AI / ML model is transferred from the network to the UE) according to embodiments herein, it may be that the network-sided AI / ML model may benefit over the UE-sided AI / ML model in certain ways. For example, the network may consider parameters from multiple UEs into consideration, thereby improving overall system optimization (rather than just a particular UE optimization) . To facilitate network side training, the UE may need to report observed IP flows to the network as part of data collection.
[0240] It is also contemplated that an AI / ML model for IP flow to DRB mapping may be trained in part at the UE and in part at the network. In one such example, the UE may first train the model based on its local observations, and then the UE may send the resulting UE-trained AI / ML model to the network for further training. The network may reflect DRB loading and the impact of other UEs using the network within the final version of the trained model, and send back the finally trained AI / ML model to the UE for use in inferencing.
[0241] In some cases corresponding to the use of an AI / ML model for QoS flow to DRB mapping model monitoring may be performed by the network. Such monitoring may be according to one or more metrics, such as system QoE or system QoS guaranteed KPI. Such monitoring at the network is facilitated by the fact that the network can observe the loading status of different DRBs and different UEs.
[0242] FIG. 30 illustrates a flow diagram 3000 of using a network-sided AI / ML model for mapping various IP flows to a current set of DRB (s) , according to embodiments herein.
[0243] The flow diagram 3000 beings with the UE 3002 transmitting 3008 a UE capability repot to the base station 3004. Then, the base station 3004 may send 3010 a configuration for training AI / ML model (s) to the UE 3002 indicating that one or more AI / ML model (s) are to be configured for training.
[0244] The UE 3002 may report 3012 observations of IP flow arrival information and IP flow metrics to the base station 3004. Then, the base station 3004, in combination with the network server 3006, may perform 3014 offline training of one or more AI / ML model (s) . Subsequently, the base station 3004 may transfer 3016 the trained AI / ML model (s) to the UE 3002. Then, the base station 3004 may signal 3018, to the UE 3002 an activation of one of the AI / ML model (s) . Accordingly, the UE may perform 3020 inference of IP flow to current set of DRB mapping using the indicated AI / ML model used for IP flow to the current set of DRB mapping.
[0245] The base station 3004 may additionally perform 3022 AI / ML model performance monitoring on the AI / ML model.
[0246] In some cases, the base station 3004 may perform 3024 LCM monitoring corresponding to the AI / ML model and may perform corresponding signaling to the UE 3002, (e.g., to indicate a new AI / ML model activation, current AI / ML model deactivation, and / or AI / ML model switching) .
[0247] The UE 3002 may switch and / or deactivate 3026 the AI / ML model (e.g., possibly in response to LCM signaling received from the base station 3004, as described) .
[0248] FIG. 31 illustrates a flow diagram 3100 of using a two-sided AI / ML model for mapping various IP flow to a current set of DRB (s) , according to embodiments herein.
[0249] The flow diagram 3100 begins with the UE 3104 transmitting 3110 a UE capability report to the base station 3106. Then, the base station 3106 may send 3112 a configuration for AI / ML model training to the UE 3104 indicating that one or more AI / ML model (s) are to be configured for training. In response, the UE 3104 may report 3114 observations of QoS flow (s) and corresponding information to the base station 3106. Then, the UE 3104, in combination with the UE server 3102, may perform 3116 offline training of one or more AI / ML model (s) . Subsequently, the UE 3104 may transfer 3118 the trained AI / ML model (s) to the base station 3106. Then, the base station 3106, in combination with the network server 3108, may further perform 3120 offline training of the AI / ML model (s) . Then, the base station 3106 may transfer 3122 the trained AI / ML model (s) to the UE 3104 and may signal 3124 an activation of one of the AI / ML model (s) and indicate such activation to the UE 3104.
[0250] Accordingly, the UE 3104 may perform 3126 inference of IP flow to current set of DRB mapping using the indicated AI / ML model. Subsequently, the base station 3106 may perform 3128 AI / ML model performance monitoring on the AI / ML model.
[0251] In some cases, the base station 3106 may perform 3130 LCM monitoring corresponding to the AI / ML model and may perform corresponding signaling to the UE 3104, (e.g., to indicate a new AI / ML model activation, current AI / ML model deactivation, and / or AI / ML model switching) .
[0252] Then, the UE 3104 may switch and / or deactivate 3132 the AI / ML model (e.g., possibly in response to LCM signaling received from the base station 3106, as described) .
[0253] FIG. 32 illustrates an example of AI / ML model based packet / IP flow to DRB mapping that generates predicted DRB information, according to embodiments herein.
[0254] In some embodiments, an AI / ML model may be used to predict DRB information using various IP flows as inputs, where the DRB information indicates a predicted mapping of the IP flow to the future set of DRB (s) .
[0255] In some cases, a UE-sided AI / ML model may be trained with each IP flow, its arrival timing, channel condition and UE mobility information.
[0256] Note that, beneficially, the use of prediction-based DRB mapping in this manner is intended to reduce the effects that would otherwise occur (e.g., in alternative cases using QoS flows) based on a mismatch between BSR reporting and the network reconfiguration of the QoS flow to DRB mapping. Further note that outcomes / mappings for a given IP flow may be different for different future times may be different
[0257] FIG. 32 illustrates an AI / ML model 3214 that generates predicted DRB information for a first IP flow 3202 (a voice IP flow) , a second IP flow 3204 (a video IP flow corresponding to a first service) , a third IP flow 3206 (a video IP flow corresponding to a second service) , and a fourth IP flow 3208 (an internet IP flow) that have been observed over various time instances (e.g., time t1, t2, …tn) . As illustrated, the AI / ML model 3214 takes these IP flows (and (optionally) to radio channel conditions 3210 and / or UE contextual information 3212) as inputs. The radio channel conditions 3210 and / or the UE contextual information 3212 are examples of possible assistance information that may be used.
[0258] The AI / ML model 3214 then predicts future mapping (s) of the IP flows to various DRB (s) at various times. For example, for a first future time (i.e., time tm) , the AI / ML model 3214 predicts that the first IP flow 3202 is to be mapped to a first DRB 3216, the second IP flow 3204 and the third IP flow 3206 are to be mapped to a second DRB 3218 and the fourth IP flow is to be mapped to a third DRB 3220.
[0259] Further, as illustrated, for a second future time (i.e., time tm+1 3226) , the AI / ML model 3214 predicts that the first IP flow 3202 is to be mapped to the first DRB 3216, the second IP flow 3204 is to be mapped to the second DRB 3218, the third IP flow 3206 is to be mapped to a fourth DRB 3222 and the fourth IP flow 3208 is to be mapped to the third DRB 3220.
[0260] This predicted information may be sent to the network to update QoS association rules which may be indicated back to the UE to be used to map the IP flows / packets to the DRB (s) .
[0261] Additionally, inference of the AI / ML model used for mapping may be performed at UE, which may report the prediction to the network, and then wait for a possible reconfiguration of the QoS association rules. In some cases corresponding to the use of an AI / ML model for IP flow to DRB mapping model monitoring may be performed by the network. Such monitoring may be according to one or more metrics, such as system QoE or system QoS guaranteed KPI. Such monitoring at the network is facilitated by the fact that the network can observe the loading status of different DRBs and different UEs.
[0262] FIG. 33 illustrates a flow diagram 3300 of using a UE-sided AI / ML model for mapping various IP flows to a future set of DRB (s) , according to embodiments herein.
[0263] The flow diagram 3300 begins with the base station 3306 transmitting 3308 a QoS association rule to the UE 3304. In response, the UE 3304 may transmit 3310 a UE capability report to the base station 3306. Then, the base station 3306 may send 3312 a configuration for AI / ML model training to the UE 3304 indicating that one or more AI / ML model (s) are to be configured for training. Subsequently, the UE 3304 may perform 3314 data collection at the UE 3304. Then, the UE 3304, in combination with the UE server 3302, may perform 3316 offline AI / ML model training at the UE server 3302 of one or more AI / ML model (s) . Accordingly, the UE may send 3318 a notification message to the base station 3306 that the AI / ML model (s) have been trained and, in response, may signal 3320, to the UE 3304 an indication of activation of one of the AI / ML model (s) .
[0264] Then, the UE 3304 may perform 3322 inference using the indicated AI / ML model to predict the IP flow to set of DRB mapping. Subsequently, the UE 3304 may report 3324 the prediction of the IP flow to set of DRB (s) mapping, as predicted by the AI / ML model, to the base station 3306.
[0265] Then, the base station 3306 may update 3326 the QoS association rule and transmit the updated QoS association rule to the UE 3304 to be used for IP flow to DRB mapping.
[0266] Further, the base station 3306 may perform 3328 AI / ML model performance monitoring on the AI / ML model.
[0267] In some cases, the base station 3306 may perform 3330 LCM monitoring corresponding to the AI / ML model and may perform corresponding signaling to the UE 3304, (e.g., to indicate a new AI / ML model activation, current AI / ML model deactivation, and / or AI / ML model switching) .
[0268] The UE 3304 may switch and / or deactivate 3332 the AI / ML model (e.g., possibly in response to LCM signaling received from the base station 3306, as described) .
[0269] Embodiments herein, from the perspective of the UE, discuss different levels of AI / ML based QoS mapping for wireless communication systems. For example, the UE may use AI / ML model (s) to derive / predict an optimal QoS mapping at different levels. In some embodiments, an AI / ML model may be used to perform mapping from IP flows to QoS flows. Input (s) to the AI / ML model, output (s) of the AI / ML model and metrics of / for performance monitoring for such cases are discussed herein. Further, procedure (s) and signaling of data collection, model training and performance monitoring for such cases are discussed herein. In some other embodiments, an AI / ML model may be used to perform mapping from QoS flow (s) to DRB (s) . Input (s) to the AI / ML model, output (s) of the AI / ML model and metrics of / for performance monitoring for such cases are discussed herein. Further, procedure (s) and signaling of data collection, model training and performance monitoring for such cases are discussed herein. In yet some other embodiments, an AI / ML model may be used to perform a single mapping from IP flows to DRB (s) . Such cases may include details on an introduced protocol stack. Further, input (s) to the AI / ML model, output (s) of the AI / ML model and metrics of / for performance monitoring for such cases are discussed herein. Additionally, procedure (s) and signaling of data collection, model training and performance monitoring for such cases are discussed herein.
[0270] FIG. 34 illustrates an end-to-end example 3400 of a QoS flow framework, according to embodiments herein. The end-to-end example 3400 illustrates a QoS flow framework from an application / service to an application / service (including various intermediate steps) . The end-to-end example 3400 illustrates the optimization of various system impacts according to embodiments herein.
[0271] As illustrated, at the UE stack, IP flow filters 3402 may be based on an AI / ML based QoS mapping. NAS signaling 3404 may include a new AI / ML model configuration described herein. Additionally, at the base station, the base station may treat QoS flows 3406 according to AI / ML model configuration received from a SMF. Also, the base station may receive N2 signaling 3408 from the access and mobility management function (AMF) where the N2 signaling 3408 may further include an AI / ML model configuration used for mapping. From the perspective of the UPF, the UPF may receive N4 signaling 3414 from an SMF where the N4 signaling 3414 may include an AI / ML model configuration as discussed herein for mapping. Further, the N11 signaling 3410 from the SMF to the AMF may include an AI / ML configuration used for mapping. Additionally, the PCC rules 3412 transmitted over N7 signaling to the SMF may include an AI / ML model configuration also used for mapping.
[0272] Note that 5G nomenclature may be discussed in various locations herein for the purpose of straightforwardly referencing various network functions (NFs) . However, because is it understood that future networks (e.g., 6G networks) may implement functionalities with respective to NF equivalents, embodiments discussed herein should be understood to also apply in cases / networks where such NF equivalents are used.
[0273] Further note that it may be that a QoS flow is the finest granularity for a QoS forwarding treatment in a wireless communication system and may be applied by various nodes. Without any adjustment to the current mechanisms in the network, the network or UPF may discard some packets, such as packets associated with a QFI received on an “unexpected” DRBs / QoS flows. As a result, the core network (CN) may configure the UP nodes in the chain accordingly.
[0274] FIG. 35 illustrates a method 3500 of a UE of a wireless communication system, according to embodiments herein. The illustrated method 3500 includes generating 3502, using an IP flow mapping AI / ML model at the UE, a mapping of one or more IP flows of the UE to one or more QoS flows. The method 3500 further includes communicating 3504, with a base station, data of the one or more IP flows using the one or more QoS flows according to the mapping of the one or more IP flows to the one or more QoS flows.
[0275] In some embodiments, the method 3500 further comprises, receiving, from the base station, IP flow filtering information, and identifying the one or more IP flows for use with the IP flow mapping AI / ML model based on the IP flow filtering information.
[0276] In some embodiments, the method 3500 further comprises, identifying the one or more IP flows for use with the IP flow mapping AI / ML model based on a determination that each of the first one or more IP flows is not covered by any IP flow to QoS flow mapping rule at the UE.
[0277] In some embodiments, the method 3500 further comprises, receiving IP flow mapping AI / ML model configuration information from the base station, and identifying, at the UE, the IP flow mapping AI / ML model based on the AI / ML mapping model configuration information.
[0278] In some embodiments of the method 3500, the IP flow mapping AI / ML model is a UE-trained AI / ML model.
[0279] In some embodiments of the method 3500, the IP flow mapping AI / ML model is a network-trained AI / ML model.
[0280] In some embodiments, the method 3500 further comprises, monitoring a performance of the IP flow mapping AI / ML model using one or more of a match factor metric and a KPI metric.
[0281] In some embodiments of the method 3500, the IP flow mapping AI / ML model uses assistance information to generate the mapping, wherein the assistance information comprises one or more of: a QoS flow candidate list that includes the one or more QoS flows, QoS identifier information that identifies one or more QoS identifiers for the one or more QoS flows, QoS flow identifier information that identifies the one or more QoS flows, QoE information, performance measurements for the one or more QoS flows, radio channel condition information, UE contextual information, and historical grant information.
[0282] In some embodiments, the method 3500 further comprises, receiving, from the base station, an activation indication for the IP flow mapping AI / ML model; wherein the mapping is generated using the IP flow mapping AI / ML model in response to the activation indication.
[0283] In some embodiments, the method 3500 further comprises, reporting the mapping to the base station.
[0284] FIG. 36 illustrates a method 3600 of a UE of a wireless communication system, according to embodiments herein. The illustrated method 3600 includes correlating 3602, using an IP flow correlation AI / ML model at the UE, one or more IP flows of the UE into one or more correlated sets of the IP flows. The method 3600 further includes generating 3604, using one or more QoS rules, a mapping of the one or more correlated sets of the IP flows to one or more QoS flows. The method 3600 further includes communicating 3606, with a base station, data of the one or more IP flows using the one or more QoS flows according the mapping of the one or more correlated sets of the IP flows to the one or more QoS flows.
[0285] In some embodiments of the method 3600, the IP flow correlation AI / ML model performs the correlating according to an identification of one or more applications for the one or more IP flows.
[0286] In some embodiments of the method 3600, the IP flow correlation AI / ML model performs the correlating according to a historical pattern of arrival for the data of the one or more IP flows.
[0287] In some embodiments of the method 3600, the QoS rules are implemented by a QoS rules AI / ML model at the UE that generates the mapping of the one or more correlated sets of the IP flows to the one or more QoS flows.
[0288] In some embodiments, the method 3600 further comprises, receiving, from the base station, an activation indication for the IP flow correlation AI / ML model; wherein the one or more correlated sets of the IP flows is generated using the IP flow correlation AI / ML model in response to the activation indication.
[0289] In some embodiments, the method 3600 further comprises, reporting the mapping to the base station.
[0290] FIG. 37 illustrates a method 3700 of a UE of a wireless communication systems, according to embodiments herein. The illustrated method 3700 includes generating 3702, using a PDU set mapping AI / ML model at the UE, a mapping of one or more PDU sets of one more PDUs of the UE to one or more QoS flows according to relative importance information for the one or more PDU sets. The method 3700 further includes communicating 3704, with a base station, data of the one or more PDUs using the one or more QoS flows according to the mapping of the one or more PDU sets to the one or more QoS flows.
[0291] In some embodiments, the method 3700 further comprises, determining that a PSIHI for an initial PDU set of the UE is not set, and splitting the initial PDU set of the UE into a first PDU set and a second PDU set based on the determination that the PSIHI for the initial PDU set is not set, wherein the one or more PDU sets mapped by the PDU set mapping AI / ML model comprises the first PDU set and the second PDU set. In some such embodiments, the UE provides the first PDU set a higher priority than the second PDU set within the relative importance information.
[0292] In some embodiments, the method 3700 further comprises, receiving, from the base station, an activation indication for the PDU set mapping AI / ML model; wherein the mapping is generated using the PDU set mapping AI / ML model in response to the activation indication.
[0293] In some embodiments, the method 3700 further comprises, reporting the mapping to the base station.
[0294] FIG. 38 illustrates a method 3800 of a base station of a wireless communication system, according to embodiments herein. The illustrated method 3800 includes sending 3802, to a UE, an activation indication activating an IP flow mapping AI / ML model at the UE. The method 3800 further includes receiving 3804, from the UE, in response to sending the activation indication, a mapping of one or more IP flows of the UE to one or more QoS flows. The method 3800 further includes communicating 3806, with the UE, data of the one or more IP flows using the one or more QoS flows according to the mapping of the one or more IP flows to the one or more QoS flows.
[0295] In some embodiments, the method 3800 further comprises, monitoring a performance of the IP flow mapping AI / ML model using a KPI metric. Some such embodiments further comprise indicating, to the UE, based on the monitoring, to deactivate the IP flow mapping AI / ML model.
[0296] In some embodiments, the method 3800 further comprises, receiving, from a CN, IP flow filtering information, and sending, to the UE, the IP flow filtering information.
[0297] In some embodiments, the method 3800 further comprises, sending, to the UE, one or more IP flow to QoS flow mapping rules.
[0298] In some embodiments, the method 3800 further comprises, sending, to the UE, IP flow mapping AI / ML model configuration information.
[0299] FIG. 39 illustrates a method 3900 of a base station of a wireless communication system, according to embodiments herein. The illustrated method 3900 includes sending 3902, to a UE, an activation indication activating an IP flow correlation AI / ML model at the UE. The method 3900 further includes receiving 3904, from the UE, in response to sending the activation indication, a mapping of one or more correlated sets of IP flows of the UE to one or more QoS flows. The method 3900 further includes communicating 3906, with the UE, data of the one or more IP flows using the one or more QoS flows according the mapping of the one or more correlated sets of the IP flows to the one or more QoS flows.
[0300] In some embodiments, the method 3900 further comprises, monitoring a performance of the IP flow correlation AI / ML model using a KPI metric.
[0301] In some embodiments, the method 3900 further comprises, indicating, to the UE, based on the monitoring, to deactivate the IP flow correlation AI / ML model.
[0302] FIG. 40 illustrates a method 4000 of a base station of a wireless communication system, according to embodiments herein. The illustrated method 4000 includes sending 4002, to a UE, an activation indication activating a PDU set mapping AI / ML model at the UE. The method 4000 further includes receiving 4004, from the UE, in response to sending the activation indication, a mapping of one or more PDU sets for one or more PDUs of the UE to one or more QoS flows. The method 4000 further includes communicating 4006, with the UE, data of the one or more PDUs using the one or more QoS flows according to the mapping of the one or more PDU sets to the one or more QoS flows.
[0303] In some embodiments, the method 4000 further comprises, monitoring a performance of the PDU set mapping AI / ML model using a KPI metric. Some such embodiments further comprise indicating, to the UE, based on the monitoring, to deactivate the PDU set mapping AI / ML model.
[0304] In some embodiments, the method 4000 further comprises, setting a PSIHI for an initial PDU set of the UE, and sending the PSIHI for the initial PDU set to the UE.
[0305] FIG. 41 illustrates a method 4100 of a UE of a wireless communication system, according to embodiments herein. The illustrated method 4100 includes generating 4102, using a QoS flow mapping AI / ML model at the UE, a mapping of one or more QoS flows used for data of one or more IP flows of the UE to one or more DRBs. The method 4100 further includes communicating 4104, with a base station, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more QoS flows to the one or more DRBs.
[0306] In some embodiments, the method 4100 further comprises, receiving QoS flow filtering information, and identifying the one or more QoS flows for use with the QoS flow mapping AI / ML model based on the QoS flow filtering information.
[0307] In some embodiments, the method 4100 further comprises, identifying the one or more QoS flows for use with the QoS flow mapping AI / ML model based on a determination that each of the first one or more QoS flows is not covered by any QoS flow to DRB mapping rule at the UE.
[0308] In some embodiments, the method 4100 further comprises, triggering a use of the QoS flow mapping AI / ML model to generate the mapping based on a trigger condition, the trigger condition comprising at least one of: that the QoS flow mapping AI / ML model has become available, and that the UE has observed a performance loss associated with the one or more QoS flows.
[0309] In some embodiments, the method 4100 further comprises, receiving QoS flow mapping AI / ML model configuration information from the base station, and identifying, at the UE, the QoS flow mapping AI / ML model based on the AI / ML mapping model configuration information.
[0310] In some embodiments of the method 4100, the QoS flow mapping AI / ML model is a UE-trained AI / ML model.
[0311] In some embodiments of the method 4100, the QoS flow mapping AI / ML model is a network-trained AI / ML model.
[0312] In some embodiments of the method 4100, the QoS flow mapping AI / ML model is a network-and-UE-trained AI / ML model.
[0313] In some embodiments of the method 4100, the QoS flow mapping AI / ML model uses assistance information to generate the mapping, wherein the assistance information comprises one or more of: radio channel condition information, and UE contextual information.
[0314] In some embodiments, the method 4100 further comprises, receiving, from the base station, an activation indication for the QoS flow mapping AI / ML model; wherein the mapping is generated using the QoS flow mapping AI / ML model in response to the activation indication.
[0315] In some embodiments, the method 4100 further comprises, reporting the mapping to the base station.
[0316] FIG. 42 illustrates a method 4200 of a UE of a wireless communication system, according to embodiments herein. The illustrated method 4200 includes generating 4202, using a QoS flow data arrival time AI / ML model at the UE, based on known data arrival time information for one or more QoS flows used for data of one of more IP flows of the UE, predicted data arrival time information for the one or more QoS flows. The method 4200 further includes sending 4204, to a base station, the predicted data arrival time information for the one or more QoS flows. The method 4200 further includes receiving 4206, from the base station, in response to sending the predicted data arrival time information, a mapping of the one or more QoS flows to one or more DRBs. The method 4200 further includes communicating 4208, with the base station, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more QoS flows to the one or more DRBs.
[0317] In some embodiments of the method 4200, the predicted data arrival time information comprises a first probability that a first QoS flow of the one or more QoS flows is used for a first part of the data at a first time. In some such embodiments, the predicted data arrival time information further comprises a second probability that the first QoS flow of the one or more QoS flows is used for a second part of the data at a second time. In some other such embodiments, the predicted data arrival time information further comprises a second probability that a second QoS flow of the one or more QoS flows is used for a second part of the data at the first time.
[0318] In some embodiments of the method 4200, the QoS flow data arrival time AI / ML model is a UE-trained AI / ML model.
[0319] In some embodiments of the method 4200, the QoS flow data arrival time AI / ML model uses assistance information to generate the predicted data arrival time information, wherein the assistance information comprises one or more of: a list of the one or more QoS flows, a candidate DRB list, QoS identifier information that identifies one or more QoS identifiers for the one or more QoS flows, QoS flow identifier information that identifies the one or more QoS flows, reflective QoS indicator information for the one or more QoS flows, QoE information, performance measurements for the one or more QoS flows, radio channel condition information, and UE contextual information.
[0320] In some embodiments, the method 4200 further comprises, receiving, from the base station, an activation indication for the QoS flow data arrival time AI / ML model, wherein the predicted data arrival time information is generated using the QoS flow data arrival time AI / ML model in response to the activation indication.
[0321] FIG. 43 illustrates a method 4300 of a UE of a wireless communication system, according to embodiments herein. The illustrated method 4300 includes correlating 4302, using a QoS flow correlation AI / ML model at the UE, one or more QoS flows for IP data flows of the UE into one or more correlated sets of the QoS flows. The method 4300 further includes generating 4304, using SDAP configuration, a mapping of the one or more correlated sets of the QoS flows to one or more DRBs. The method 4300 further includes communicating 4306, with a base station, data of one or more IP flows of the UE using the one or more DRBs according to the mapping of the one or more correlated sets of the QoS flows to the one or more DRBs.
[0322] In some embodiments of the method 4300, the QoS flow correlation AI / ML model performs the correlating according to PDU session information for the one or more QoS flows.
[0323] In some embodiments of the method 4300, the QoS flow correlation AI / ML model performs the correlating according to a historical pattern of arrival for the data within the one or more QoS flows.
[0324] In some embodiments of the method 4300, the SDAP configuration is implemented by a SDAP configuration AI / ML model at the UE that generates the mapping of the one or more correlated sets of the QoS flows to the one or more DRBs.
[0325] In some embodiments, the method 4300 further comprises, receiving, from the base station, an activation indication for the QoS flow correlation AI / ML model; wherein the one or more correlated sets of the QoS flows is generated using the QoS flow correlation AI / ML model in response to the activation indication.
[0326] In some embodiments, the method 4300 further comprises, reporting the mapping to the base station.
[0327] FIG. 44 illustrates a method 4400 of a base station of a wireless communication system, according to embodiments herein. The illustrated method 4400 includes sending 4402, to a UE, an activation indication activating a QoS flow mapping AI / ML model at the UE. The method 4400 further includes receiving 4404, from the UE, in response to sending the activation indication, a mapping of one or more QoS flows used for data of one or more IP flows of the UE to one or more DRBs. The method 4400 further includes communicating 4406, with the UE, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more QoS flows to the one or more DRBs.
[0328] In some embodiments, the method 4400 further comprises, monitoring a performance of the QoS flow mapping AI / ML model using a KPI metric. Some such embodiments further comprise indicating, to the UE, based on the monitoring, to deactivate the QoS flow mapping AI / ML model.
[0329] In some embodiments, the method 4400 further comprises, sending QoS flow filtering information to the UE.
[0330] In some embodiments, the method 4400 further comprises, sending, to the UE, one or more QoS flow to DRB mapping rules.
[0331] In some embodiments, the method 4400 further comprises, sending QoS flow mapping AI / ML model configuration information to the UE.
[0332] FIG. 45 illustrates a method 4500 of a base station of a wireless communication system, according to embodiments herein. The illustrated method 4500 includes receiving 4502, from a UE, predicted data arrival time information for one or more QoS flows used for data of one or more IP flows of the UE. The method 4500 further includes generating 4504, using the predicted data arrival time information, a mapping of the one or more QoS flows to one or more DRBs. The method 4500 further includes sending 4506, to the UE, the mapping of the one or more QoS flows to the one or more DRBs. The method 4500 further includes communicating 4508, with the UE, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more QoS flows to the one or more DRBs.
[0333] In some embodiments of the method 4500, the predicted data arrival time information comprises a first probability that a first QoS flow of the one or more QoS flows is used for a first part of the data at a first time. In some such embodiments, the predicted data arrival time information further comprises a second probability that the first QoS flow of the one or more QoS flows is used for a second part of the data at a second time. In some other such embodiments, the predicted data arrival time information further comprises a second probability that a second QoS flow of the one or more QoS flows is used for a second part of the data at the first time.
[0334] In some embodiments, the method 4500 further comprises, sending, to the UE, an activation indication activating a QoS flow data arrival time AI / ML model at the UE for use to generate the predicted data arrival time information, wherein the predicted data arrival time information is received from the UE in response to sending the activation indication.
[0335] In some embodiments, the method 4500 further comprises, monitoring a performance of a QoS flow data arrival time AI / ML model at the UE using a KPI metric. Some such embodiments further comprise indicating, to the UE, based on the monitoring, to deactivate the QoS flow data arrival time AI / ML model.
[0336] FIG. 46 illustrates a method 4600 of a base station of a wireless communication system, according to embodiments herein. The illustrated method 4600 includes sending 4602, to a UE, an activation indication activating a QoS flow correlation AI / ML model at the UE. The method 4600 further includes receiving 4604, from the UE, in response to sending the activation indication, a mapping of one or more correlated sets of QoS flows used for data of one or more IP flows of the UE to one or more DRBs. The method 4600 further includes communicating 4606, with the UE, the data of one or more IP flows of the UE using the one or more DRBs according to the mapping of the one or more correlated sets of the QoS flows to the one or more DRBs.
[0337] In some embodiments, the method 4600 further comprises, monitoring a performance of the QoS flow correlation AI / ML model using a KPI metric. Some such embodiments further comprise indicating, to the UE, based on the monitoring, to deactivate the QoS flow correlation AI / ML model.
[0338] FIG. 47 illustrates a method 4700 of a UE of a wireless communication system, according to embodiments herein. The illustrated method 4700 includes generating 4702, using an IP flow mapping AI / ML model at the UE, a mapping of one or more IP flows of the UE to one or more DRBs. The method 4700 further includes communicating 4704, with a base station, data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more IP flows to the one or more DRBs.
[0339] In some embodiments of the method 4700, the IP flow mapping AI / ML model is a UE-trained AI / ML model.
[0340] In some embodiments of the method 4700, the IP flow mapping AI / ML model is a network-trained AI / ML model.
[0341] In some embodiments of the method 4700, the IP flow mapping AI / ML model is a network-and-UE-trained AI / ML model.
[0342] In some embodiments of the method 4700, the IP flow mapping AI / ML model uses assistance information to generate the mapping, wherein the assistance information comprises one or more of: radio channel condition information, and UE contextual information.
[0343] FIG. 48 illustrates a method 4800 of a UE of a wireless communication system, according to embodiments herein. The illustrated method 4800 includes generating 4802, using an IP flow mapping AI / ML model at the UE, based on known data arrival time information for data of one or more IP flows of the UE, predicted DRB information for the one or more IP flows, the predicted DRB information indicating a predicted mapping of the one or more IP flows to one or more DRBs to be used at a future time. The method 4800 further includes sending 4804, to a base station, the predicted DRB information. The method 4800 further includes receiving 4806, from the base station, in response to sending the predicted DRB information, a mapping of the one or more IP flows to the one or more DRBs. The method 4800 further includes communicating 4808, with the base station, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more IP flows to the one or more DRBs.
[0344] In some embodiments of the method 4800, the IP flow mapping AI / ML model is a UE-trained AI / ML model.
[0345] In some embodiments, the method 4800 further comprises, monitoring a performance of the IP flow mapping AI / ML model using one or more of a match factor metric and a KPI metric.
[0346] In some embodiments of the method 4800, the IP flow mapping AI / ML model uses assistance information to generate the predicted mapping, wherein the assistance information comprises one or more of: radio channel condition information, and UE mobility information.
[0347] FIG. 49 illustrates a method 4900 of a base station of a wireless communication system, according to embodiments herein. The illustrated method 4900 includes sending 4902, to a UE, an activation indication activating an IP flow mapping AI / ML model at the UE. The method 4900 further includes receiving 4904, from the UE, in response to sending the activation indication, a mapping of one or more IP flows of the UE to one or more DRBs. The method 4900 further includes communicating 4906, with the UE, data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more IP flows to the one or more DRBs.
[0348] In some embodiments, the method 4900 further comprises, monitoring a performance of the IP flow mapping AI / ML model using a KPI metric. Some such embodiments further comprise indicating, to the UE, based on the monitoring, to deactivate the IP flow mapping AI / ML model.
[0349] FIG. 50 illustrates a method 5000 of a base station of a wireless communication system, according to embodiments herein. The illustrated method 5000 includes receiving 5002, from a UE, predicted DRB information for of one or more IP flows of the UE, the predicted DRB information indicating a predicted mapping of the one or more IP flows to one or more DRBs to be used at a future time. The method 5000 further includes generating 5004, using the predicted DRB information, a mapping of the one or more IP flows to the one or more DRBs. The method 5000 further includes sending 5006, to the UE, the mapping of the one or more IP flows to the one or more DRBs. The method 5000 further includes communicating 5008, with the UE, data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more IP flows to the one or more DRBs.
[0350] In some embodiments, the method 5000 further comprises, sending, to the UE, an activation indication activating a IP flow mapping AI / ML model at the UE for use to generate the predicted data arrival time information, wherein the predicted DRB information is received from the UE in response to sending the activation indication.
[0351] In some embodiments, the method 5000 further comprises, monitoring a performance of an IP flow mapping AI / ML model at the UE using a KPI metric. Some such embodiments further comprise indicating, to the UE, based on the monitoring, to deactivate the IP flow mapping AI / ML model.
[0352] FIG. 51 illustrates an example architecture of a wireless communication system 5100, according to embodiments disclosed herein. The following description is provided for an example wireless communication system 5100 that operates in conjunction with the LTE system standards and / or 5G or NR system standards as provided by 3GPP technical specifications.
[0353] As shown by FIG. 51, the wireless communication system 5100 includes UE 5102 and UE 5104 (although any number of UEs may be used) . In this example, the UE 5102 and the UE 5104 are illustrated as smartphones (e.g., handheld touchscreen mobile computing devices connectable to one or more cellular networks) , but may also comprise any mobile or non-mobile computing device configured for wireless communication.
[0354] The UE 5102 and UE 5104 may be configured to communicatively couple with a RAN 5106. In embodiments, the RAN 5106 may be NG-RAN, E-UTRAN, etc. The UE 5102 and UE 5104 utilize connections (or channels) (shown as connection 5108 and connection 5110, respectively) with the RAN 5106, each of which comprises a physical communications interface. The RAN 5106 can include one or more base stations (such as base station 5112 and base station 5114) that enable the connection 5108 and connection 5110.
[0355] In this example, the connection 5108 and connection 5110 are air interfaces to enable such communicative coupling, and may be consistent with RAT (s) used by the RAN 5106, such as, for example, an LTE and / or NR.
[0356] In some embodiments, the UE 5102 and UE 5104 may also directly exchange communication data via a sidelink interface 5116. The UE 5104 is shown to be configured to access an access point (shown as AP 5118) via connection 5120. By way of example, the connection 5120 can comprise a local wireless connection, such as a connection consistent with any IEEE 802.11 protocol, wherein the AP 5118 may comprise a router. In this example, the AP 5118 may be connected to another network (for example, the Internet) without going through a CN 5124.
[0357] In embodiments, the UE 5102 and UE 5104 can be configured to communicate using orthogonal frequency division multiplexing (OFDM) communication signals with each other or with the base station 5112 and / or the base station 5114 over a multicarrier communication channel in accordance with various communication techniques, such as, but not limited to, an orthogonal frequency division multiple access (OFDMA) communication technique (e.g., for downlink communications) or a single carrier frequency division multiple access (SC-FDMA) communication technique (e.g., for uplink and ProSe or sidelink communications) , although the scope of the embodiments is not limited in this respect. The OFDM signals can comprise a plurality of orthogonal subcarriers.
[0358] In some embodiments, all or parts of the base station 5112 or base station 5114 may be implemented as one or more software entities running on server computers as part of a virtual network. In addition, or in other embodiments, the base station 5112 or base station 5114 may be configured to communicate with one another via interface 5122. In embodiments where the wireless communication system 5100 is an LTE system (e.g., when the CN 5124 is an EPC) , the interface 5122 may be an X2 interface. The X2 interface may be defined between two or more base stations (e.g., two or more eNBs and the like) that connect to an EPC, and / or between two eNBs connecting to the EPC. In embodiments where the wireless communication system 5100 is an NR system (e.g., when CN 5124 is a 5GC) , the interface 5122 may be an Xn interface. The Xn interface is defined between two or more base stations (e.g., two or more gNBs and the like) that connect to 5GC, between a base station 5112 (e.g., a gNB) connecting to 5GC and an eNB, and / or between two eNBs connecting to 5GC (e.g., CN 5124) .
[0359] The RAN 5106 is shown to be communicatively coupled to the CN 5124. The CN 5124 may comprise one or more network elements 5126, which are configured to offer various data and telecommunications services to customers / subscribers (e.g., users of UE 5102 and UE 5104) who are connected to the CN 5124 via the RAN 5106. The components of the CN 5124 may be implemented in one physical device or separate physical devices including components to read and execute instructions from a machine-readable or computer-readable medium (e.g., a non-transitory machine-readable storage medium) .
[0360] In embodiments, the CN 5124 may be an EPC, and the RAN 5106 may be connected with the CN 5124 via an S1 interface 5128. In embodiments, the S1 interface 5128 may be split into two parts, an S1 user plane (S1-U) interface, which carries traffic data between the base station 5112 or base station 5114 and a serving gateway (S-GW) , and the S1-MME interface, which is a signaling interface between the base station 5112 or base station 5114 and mobility management entities (MMEs) .
[0361] In embodiments, the CN 5124 may be a 5GC, and the RAN 5106 may be connected with the CN 5124 via an NG interface 5128. In embodiments, the NG interface 5128 may be split into two parts, an NG user plane (NG-U) interface, which carries traffic data between the base station 5112 or base station 5114 and a user plane function (UPF) , and the S1 control plane (NG-C) interface, which is a signaling interface between the base station 5112 or base station 5114 and access and mobility management functions (AMFs) .
[0362] Generally, an application server 5130 may be an element offering applications that use internet protocol (IP) bearer resources with the CN 5124 (e.g., packet switched data services) . The application server 5130 can also be configured to support one or more communication services (e.g., VoIP sessions, group communication sessions, etc. ) for the UE 5102 and UE 5104 via the CN 5124. The application server 5130 may communicate with the CN 5124 through an IP communications interface 5132.
[0363] FIG. 52 illustrates a system 5200 for performing signaling 5234 between a wireless device 5202 and a network device 5218, according to embodiments disclosed herein. The system 5200 may be a portion of a wireless communications system as herein described. The wireless device 5202 may be, for example, a UE of a wireless communication system. The network device 5218 may be, for example, a base station (e.g., an eNB or a gNB) of a wireless communication system.
[0364] The wireless device 5202 may include one or more processor (s) 5204. The processor (s) 5204 may execute instructions such that various operations of the wireless device 5202 are performed, as described herein. The processor (s) 5204 may include one or more baseband processors implemented using, for example, a central processing unit (CPU) , a digital signal processor (DSP) , an application specific integrated circuit (ASIC) , a controller, a field programmable gate array (FPGA) device, another hardware device, a firmware device, or any combination thereof configured to perform the operations described herein.
[0365] The wireless device 5202 may include a memory 5206. The memory 5206 may be a non-transitory computer-readable storage medium that stores instructions 5208 (which may include, for example, the instructions being executed by the processor (s) 5204) . The instructions 5208 may also be referred to as program code or a computer program. The memory 5206 may also store data used by, and results computed by, the processor (s) 5204.
[0366] The wireless device 5202 may include one or more transceiver (s) 5210 that may include radio frequency (RF) transmitter circuitry and / or receiver circuitry that use the antenna (s) 5212 of the wireless device 5202 to facilitate signaling (e.g., the signaling 5234) to and / or from the wireless device 5202 with other devices (e.g., the network device 5218) according to corresponding RATs.
[0367] The wireless device 5202 may include one or more antenna (s) 5212 (e.g., one, two, four, or more) . For embodiments with multiple antenna (s) 5212, the wireless device 5202 may leverage the spatial diversity of such multiple antenna (s) 5212 to send and / or receive multiple different data streams on the same time and frequency resources. This behavior may be referred to as, for example, multiple input multiple output (MIMO) behavior (referring to the multiple antennas used at each of a transmitting device and a receiving device that enable this aspect) . MIMO transmissions by the wireless device 5202 may be accomplished according to precoding (or digital beamforming) that is applied at the wireless device 5202 that multiplexes the data streams across the antenna (s) 5212 according to known or assumed channel characteristics such that each data stream is received with an appropriate signal strength relative to other streams and at a desired location in the spatial domain (e.g., the location of a receiver associated with that data stream) . Certain embodiments may use single user MIMO (SU-MIMO) methods (where the data streams are all directed to a single receiver) and / or multi user MIMO (MU-MIMO) methods (where individual data streams may be directed to individual (different) receivers in different locations in the spatial domain) .
[0368] In certain embodiments having multiple antennas, the wireless device 5202 may implement analog beamforming techniques, whereby phases of the signals sent by the antenna (s) 5212 are relatively adjusted such that the (joint) transmission of the antenna (s) 5212 can be directed (this is sometimes referred to as beam steering) .
[0369] The wireless device 5202 may include one or more interface (s) 5214. The interface (s) 5214 may be used to provide input to or output from the wireless device 5202. For example, a wireless device 5202 that is a UE may include interface (s) 5214 such as microphones, speakers, a touchscreen, buttons, and the like in order to allow for input and / or output to the UE by a user of the UE. Other interfaces of such a UE may be made up of transmitters, receivers, and other circuitry (e.g., other than the transceiver (s) 5210 / antenna (s) 5212 already described) that allow for communication between the UE and other devices and may operate according to known protocols (e.g., and the like) .
[0370] The wireless device 5202 may include a QoS mapping module 5216. The QoS mapping module 5216 may be implemented via hardware, software, or combinations thereof. For example, the QoS mapping module 5216 may be implemented as a processor, circuit, and / or instructions 5208 stored in the memory 5206 and executed by the processor (s) 5204. In some examples, the QoS mapping module 5216 may be integrated within the processor (s) 5204 and / or the transceiver (s) 5210. For example, the QoS mapping module 5216 may be implemented by a combination of software components (e.g., executed by a DSP or a general processor) and hardware components (e.g., logic gates and circuitry) within the processor (s) 5204 or the transceiver (s) 5210.
[0371] The QoS mapping module 5216 may be used for various aspects of the present disclosure, for example, aspects of FIG. 1 to FIG. 50. The QoS mapping module 5216 is may configure the wireless device 5202 to operate according to any embodiment for AI / ML model-based IP flow to QoS flow mapping, AI / ML model-based QoS flow to DRB mapping, and / or AI / ML model-based IP flow to DRB mapping as discussed herein.
[0372] The network device 5218 may include one or more processor (s) 5220. The processor (s) 5220 may execute instructions such that various operations of the network device 5218 are performed, as described herein. The processor (s) 5220 may include one or more baseband processors implemented using, for example, a CPU, a DSP, an ASIC, a controller, an FPGA device, another hardware device, a firmware device, or any combination thereof configured to perform the operations described herein.
[0373] The network device 5218 may include a memory 5222. The memory 5222 may be a non-transitory computer-readable storage medium that stores instructions 5224 (which may include, for example, the instructions being executed by the processor (s) 5220) . The instructions 5224 may also be referred to as program code or a computer program. The memory 5222 may also store data used by, and results computed by, the processor (s) 5220.
[0374] The network device 5218 may include one or more transceiver (s) 5226 that may include RF transmitter circuitry and / or receiver circuitry that use the antenna (s) 5228 of the network device 5218 to facilitate signaling (e.g., the signaling 5234) to and / or from the network device 5218 with other devices (e.g., the wireless device 5202) according to corresponding RATs.
[0375] The network device 5218 may include one or more antenna (s) 5228 (e.g., one, two, four, or more) . In embodiments having multiple antenna (s) 5228, the network device 5218 may perform MIMO, digital beamforming, analog beamforming, beam steering, etc., as has been described.
[0376] The network device 5218 may include one or more interface (s) 5230. The interface (s) 5230 may be used to provide input to or output from the network device 5218. For example, a network device 5218 that is a base station may include interface (s) 5230 made up of transmitters, receivers, and other circuitry (e.g., other than the transceiver (s) 5226 / antenna (s) 5228 already described) that enables the base station to communicate with other equipment in a core network, and / or that enables the base station to communicate with external networks, computers, databases, and the like for purposes of operations, administration, and maintenance of the base station or other equipment operably connected thereto.
[0377] The network device 5218 may include a QoS mapping module 5232. The QoS mapping module 5232 may be implemented via hardware, software, or combinations thereof. For example, the QoS mapping module 5232 may be implemented as a processor, circuit, and / or instructions 5224 stored in the memory 5222 and executed by the processor (s) 5220. In some examples, the QoS mapping module 5232 may be integrated within the processor (s) 5220 and / or the transceiver (s) 5226. For example, the QoS mapping module 5232 may be implemented by a combination of software components (e.g., executed by a DSP or a general processor) and hardware components (e.g., logic gates and circuitry) within the processor (s) 5220 or the transceiver (s) 5226.
[0378] The QoS mapping module 5232 may be used for various aspects of the present disclosure, for example, aspects of FIG. 1 to FIG. 50. The QoS mapping module 5232 may configure the network device 5218 to operate according to any embodiment for AI / ML model-based IP flow to QoS flow mapping, AI / ML model-based QoS flow to DRB mapping, and / or AI / ML model-based IP flow to DRB mapping as discussed herein.
[0379] Embodiments contemplated herein include an apparatus comprising means to perform one or more elements of any of the method 3500, method 3600, method 3700, method 4100, method 4200, method 4300, method 4700, and method 4800. This apparatus may be, for example, an apparatus of a UE (such as a wireless device 5202 that is a UE, as described herein) .
[0380] Embodiments contemplated herein include one or more non-transitory computer-readable media comprising instructions to cause an electronic device, upon execution of the instructions by one or more processors of the electronic device, to perform one or more elements of any of the method 3500, method 3600, method 3700, method 4100, method 4200, method 4300, method 4700, and method 4800. This non-transitory computer-readable media may be, for example, a memory of a UE (such as a memory 5206 of a wireless device 5202 that is a UE, as described herein) .
[0381] Embodiments contemplated herein include an apparatus comprising logic, modules, or circuitry to perform one or more elements of any of the method 3500, method 3600, method 3700, method 4100, method 4200, method 4300, method 4700, and method 4800. This apparatus may be, for example, an apparatus of a UE (such as a wireless device 5202 that is a UE, as described herein) .
[0382] Embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform one or more elements of any of the method 3500, method 3600, method 3700, method 4100, method 4200, method 4300, method 4700, and method 4800. This apparatus may be, for example, an apparatus of a UE (such as a wireless device 5202 that is a UE, as described herein) .
[0383] Embodiments contemplated herein include a signal as described in or related to one or more elements of any of the method 3500, method 3600, method 3700, method 4100, method 4200, method 4300, method 4700, and method 4800.
[0384] Embodiments contemplated herein include a computer program or computer program product comprising instructions, wherein execution of the program by a processor is to cause the processor to carry out one or more elements of any of the method 3500, method 3600, method 3700, method 4100, method 4200, method 4300, method 4700, and method 4800. The processor may be a processor of a UE (such as a processor (s) 5204 of a wireless device 5202 that is a UE, as described herein) . These instructions may be, for example, located in the processor and / or on a memory of the UE (such as a memory 5206 of a wireless device 5202 that is a UE, as described herein) .
[0385] Embodiments contemplated herein include an apparatus comprising means to perform one or more elements of any of the method 3800, method 3900, method 4000, method 4400, method 4500, method 4600, method 4900, and method 5000. This apparatus may be, for example, an apparatus of a base station (such as a network device 5218 that is a base station, as described herein) .
[0386] Embodiments contemplated herein include one or more non-transitory computer-readable media comprising instructions to cause an electronic device, upon execution of the instructions by one or more processors of the electronic device, to perform one or more elements of any of the method 3800, method 3900, method 4000, method 4400, method 4500, method 4600, method 4900, and method 5000. This non-transitory computer-readable media may be, for example, a memory of a base station (such as a memory 5222 of a network device 5218 that is a base station, as described herein) .
[0387] Embodiments contemplated herein include an apparatus comprising logic, modules, or circuitry to perform one or more elements of any of the method 3800, method 3900, method 4000, method 4400, method 4500, method 4600, method 4900, and method 5000. This apparatus may be, for example, an apparatus of a base station (such as a network device 5218 that is a base station, as described herein) .
[0388] Embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform one or more elements of any of the method 3800, method 3900, method 4000, method 4400, method 4500, method 4600, method 4900, and method 5000. This apparatus may be, for example, an apparatus of a base station (such as a network device 5218 that is a base station, as described herein) .
[0389] Embodiments contemplated herein include a signal as described in or related to one or more elements of any of the method 3800, method 3900, method 4000, method 4400, method 4500, method 4600, method 4900, and method 5000.
[0390] Embodiments contemplated herein include a computer program or computer program product comprising instructions, wherein execution of the program by a processing element is to cause the processing element to carry out one or more elements of any of the method 3800, method 3900, method 4000, method 4400, method 4500, method 4600, method 4900, and method 5000. The processor may be a processor of a base station (such as a processor (s) 5220 of a network device 5218 that is a base station, as described herein) . These instructions may be, for example, located in the processor and / or on a memory of the base station (such as a memory 5222 of a network device 5218 that is a base station, as described herein) .
[0391] For one or more embodiments, at least one of the components set forth in one or more of the preceding figures may be configured to perform one or more operations, techniques, processes, and / or methods as set forth herein. For example, a baseband processor as described herein in connection with one or more of the preceding figures may be configured to operate in accordance with one or more of the examples set forth herein. For another example, circuitry associated with a UE, base station, network element, etc. as described above in connection with one or more of the preceding figures may be configured to operate in accordance with one or more of the examples set forth herein.
[0392] Any of the above described embodiments may be combined with any other embodiment (or combination of embodiments) , unless explicitly stated otherwise. The foregoing description of one or more implementations provides illustration and description, but is not intended to be exhaustive or to limit the scope of embodiments to the precise form disclosed. Modifications and variations are possible in light of the above teachings or may be acquired from practice of various embodiments.
[0393] Embodiments and implementations of the systems and methods described herein may include various operations, which may be embodied in machine-executable instructions to be executed by a computer system. A computer system may include one or more general-purpose or special-purpose computers (or other electronic devices) . The computer system may include hardware components that include specific logic for performing the operations or may include a combination of hardware, software, and / or firmware.
[0394] It should be recognized that the systems described herein include descriptions of specific embodiments. These embodiments can be combined into single systems, partially combined into other systems, split into multiple systems or divided or combined in other ways. In addition, it is contemplated that parameters, attributes, aspects, etc. of one embodiment can be used in another embodiment. The parameters, attributes, aspects, etc. are merely described in one or more embodiments for clarity, and it is recognized that the parameters, attributes, aspects, etc. can be combined with or substituted for parameters, attributes, aspects, etc. of another embodiment unless specifically disclaimed herein.
[0395] It is well understood that the use of personally identifiable information should follow privacy policies and practices that are generally recognized as meeting or exceeding industry or governmental requirements for maintaining the privacy of users. In particular, personally identifiable information data should be managed and handled so as to minimize risks of unintentional or unauthorized access or use, and the nature of authorized use should be clearly indicated to users.
[0396] Although the foregoing has been described in some detail for purposes of clarity, it will be apparent that certain changes and modifications may be made without departing from the principles thereof. It should be noted that there are many alternative ways of implementing both the processes and apparatuses described herein. Accordingly, the present embodiments are to be considered illustrative and not restrictive, and the description is not to be limited to the details given herein, but may be modified within the scope and equivalents of the appended claims.
Claims
1.A method of a user equipment (UE) of a wireless communication system, comprising:generating, using a quality of service (QoS) flow mapping artificial intelligence (AI) / machine learning (ML) model at the UE, a mapping of one or more QoS flows used for data of one or more internet protocol (IP) flows of the UE to one or more data radio bearers (DRBs) ; andcommunicating, with a base station, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more QoS flows to the one or more DRBs.2.The method of claim 1, further comprising:receiving QoS flow filtering information; andidentifying the one or more QoS flows for use with the QoS flow mapping AI / ML model based on the QoS flow filtering information.3.The method of claim 1, further comprising identifying the one or more QoS flows for use with the QoS flow mapping AI / ML model based on a determination that each of the first one or more QoS flows is not covered by any QoS flow to DRB mapping rule at the UE.4.The method of claim 1, further comprising triggering a use of the QoS flow mapping AI / ML model to generate the mapping based on a trigger condition, the trigger condition comprising at least one of:that the QoS flow mapping AI / ML model has become available; andthat the UE has observed a performance loss associated with the one or more QoS flows.5.The method of claim 1, further comprising:receiving QoS flow mapping AI / ML model configuration information from the base station; andidentifying, at the UE, the QoS flow mapping AI / ML model based on the AI / ML mapping model configuration information.6.The method of claim 1, wherein the QoS flow mapping AI / ML model is a UE-trained AI / ML model.7.The method of claim 1, wherein the QoS flow mapping AI / ML model is a network-trained AI / ML model.8.The method of claim 1, wherein the QoS flow mapping AI / ML model is a network-and-UE-trained AI / ML model.9.The method of claim 1, wherein the QoS flow mapping AI / ML model uses assistance information to generate the mapping, wherein the assistance information comprises one or more of:radio channel condition information; andUE contextual information.10.The method of claim 1, further comprising receiving, from the base station, an activation indication for the QoS flow mapping AI / ML model; wherein the mapping is generated using the QoS flow mapping AI / ML model in response to the activation indication.11.The method of claim 1, further comprising reporting the mapping to the base station.12.A method of a user equipment (UE) of a wireless communication system, comprising:generating, using a quality of service (QoS) flow data arrival time artificial intelligence (AI) / machine learning (ML) model at the UE, based on known data arrival time information for one or more QoS flows used for data of one of more internet protocol (IP) flows of the UE, predicted data arrival time information for the one or more QoS flows;sending, to a base station, the predicted data arrival time information for the one or more QoS flows;receiving, from the base station, in response to sending the predicted data arrival time information, a mapping of the one or more QoS flows to one or more data radio bearers (DRBs) ; andcommunicating, with the base station, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more QoS flows to the one or more DRBs.13.The method of claim 12, wherein the predicted data arrival time information comprises a first probability that a first QoS flow of the one or more QoS flows is used for a first part of the data at a first time.14.The method of claim 13, wherein the predicted data arrival time information further comprises a second probability that the first QoS flow of the one or more QoS flows is used for a second part of the data at a second time.15.The method of claim 13, wherein the predicted data arrival time information further comprises a second probability that a second QoS flow of the one or more QoS flows is used for a second part of the data at the first time.16.The method of claim 12, wherein the QoS flow data arrival time AI / ML model is a UE-trained AI / ML model.17.The method of claim 12, wherein the QoS flow data arrival time AI / ML model uses assistance information to generate the predicted data arrival time information, wherein the assistance information comprises one or more of:a list of the one or more QoS flows;a candidate DRB list;QoS identifier information that identifies one or more QoS identifiers for the one or more QoS flows;QoS flow identifier information that identifies the one or more QoS flows;reflective QoS indicator information for the one or more QoS flows;quality of experience (QoE) information;performance measurements for the one or more QoS flows;radio channel condition information; andUE contextual information.18.The method of claim 12, further comprising receiving, from the base station, an activation indication for the QoS flow data arrival time AI / ML model; wherein the predicted data arrival time information is generated using the QoS flow data arrival time AI / ML model in response to the activation indication.19.A method of a user equipment (UE) of a wireless communication system, comprising:correlating, using a quality of service (QoS) flow correlation artificial intelligence (AI) / machine learning (ML) model at the UE, one or more QoS flows for internet protocol (IP) data flows of the UE into one or more correlated sets of the QoS flows;generating, using service data adaptation protocol (SDAP) configuration, a mapping of the one or more correlated sets of the QoS flows to one or more data radio bearers (DRBs) ; andcommunicating, with a base station, data of one or more IP flows of the UE using the one or more DRBs according to the mapping of the one or more correlated sets of the QoS flows to the one or more DRBs.20.The method of claim 19, wherein the QoS flow correlation AI / ML model performs the correlating according to protocol data unit (PDU) session information for the one or more QoS flows.21.The method of claim 19, wherein the QoS flow correlation AI / ML model performs the correlating according to a historical pattern of arrival for the data within the one or more QoS flows.22.The method of claim 19, wherein the SDAP configuration is implemented by a SDAP configuration AI / ML model at the UE that generates the mapping of the one or more correlated sets of the QoS flows to the one or more DRBs.23.The method of claim 19, further comprising receiving, from the base station, an activation indication for the QoS flow correlation AI / ML model; wherein the one or more correlated sets of the QoS flows is generated using the QoS flow correlation AI / ML model in response to the activation indication.24.The method of claim 19, further comprising reporting the mapping to the base station.25.A method of a base station of a wireless communication system, comprising:sending, to a user equipment (UE) , an activation indication activating a quality of service (QoS) flow mapping artificial intelligence (AI) / machine learning (ML) model at the UE;receiving, from the UE, in response to sending the activation indication, a mapping of one or more QoS flows used for data of one or more internet protocol (IP) flows of the UE to one or more data radio bearers (DRBs) ; andcommunicating, with the UE, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more QoS flows to the one or more DRBs.26.The method of claim 25, further comprising monitoring a performance of the QoS flow mapping AI / ML model using a key performance indicator (KPI) metric.27.The method of claim 26, further comprising indicating, to the UE, based on the monitoring, to deactivate the QoS flow mapping AI / ML model.28.The method of claim 25, further comprising sending QoS flow filtering information to the UE.29.The method of claim 25, further comprising sending, to the UE, one or more QoS flow to DRB mapping rules.30.The method of claim 25, further comprising sending QoS flow mapping AI / ML model configuration information to the UE.31.A method of a base station of a wireless communication system, comprising:receiving, from a user equipment (UE) , predicted data arrival time information for one or more quality of service (QoS) flows used for data of one or more internet protocol (IP) flows of the UE;generating, using the predicted data arrival time information, a mapping of the one or more QoS flows to one or more data radio bearers (DRBs) ;sending, to the UE, the mapping of the one or more QoS flows to the one or more DRBs; andcommunicating, with the UE, the data of the one or more IP flows using the one or more DRBs according to the mapping of the one or more QoS flows to the one or more DRBs.32.The method of claim 31, wherein the predicted data arrival time information comprises a first probability that a first QoS flow of the one or more QoS flows is used for a first part of the data at a first time.33.The method of claim 32, wherein the predicted data arrival time information further comprises a second probability that the first QoS flow of the one or more QoS flows is used for a second part of the data at a second time.34.The method of claim 32, wherein the predicted data arrival time information further comprises a second probability that a second QoS flow of the one or more QoS flows is used for a second part of the data at the first time.35.The method of claim 31, further comprising sending, to the UE, an activation indication activating a QoS flow data arrival time artificial intelligence (AI) / machine learning (ML) model at the UE for use to generate the predicted data arrival time information, wherein the predicted data arrival time information is received from the UE in response to sending the activation indication.36.The method of claim 31, further comprising monitoring a performance of a QoS flow data arrival time artificial intelligence (AI) / machine learning (ML) model at the UE using a key performance indicator (KPI) metric.37.The method of claim 36, further comprising indicating, to the UE, based on the monitoring, to deactivate the QoS flow data arrival time AI / ML model.38.A method of a base station of a wireless communication system, comprising:sending, to a user equipment (UE) , an activation indication activating a quality of service (QoS) flow correlation artificial intelligence (AI) / machine learning (ML) model at the UE;receiving, from the UE, in response to sending the activation indication, a mapping of one or more correlated sets of QoS flows used for data of one or more internet protocol (IP) flows of the UE to one or more data radio bearers (DRBs) ; andcommunicating, with the UE, the data of one or more IP flows of the UE using the one or more DRBs according to the mapping of the one or more correlated sets of the QoS flows to the one or more DRBs.39.The method of claim 38, further comprising monitoring a performance of the QoS flow correlation AI / ML model using a key performance indicator (KPI) metric.40.The method of claim 39, further comprising indicating, to the UE, based on the monitoring, to deactivate the QoS flow correlation AI / ML model.41.An apparatus comprising means to perform the method of any of claim 1 to claim 40.42.A computer-readable media comprising instructions to cause an electronic device, upon execution of the instructions by one or more processors of the electronic device, to perform the method of any of claim 1 to claim 40.43.An apparatus comprising logic, modules, or circuitry to perform the method of any of claim 1 to claim 40.44.A baseband processor for a user equipment (UE) that is configured to cause the UE to perform one or more elements of any one of claim 1 to claim 24.45.A baseband processor for a base station that is configured to cause the base station to perform one or more elements of any one of claim 25 to claim 40.
Citation Information
Patent Citations
Method and apparatus for provisioning quality of service in next radio
CN109923897A
Stream-based end-to-end quality of service management
CN117397286A
METHOD FOR TRANSMITTING UL PACKET BASED ON QUALITY OF SERVICE (QoS) FLOW IN WIRELESS COMMUNICATION SYSTEM AND A DEVICE THEREFOR
US20200128430A1
QOS profile adaptation
US20230328580A1