Optimizing radio resource management decisions in o-ran networks using predicted information

Abstract

Systems, methods, and computer program products for a system configured to execute Artificial Intelligence and Machine Learning models. A scheduler metric of each logical channel, where for an Ith User Equipment, Channel State Information (CSI) at time instant t is given by CSli t and predicted CSI for next m slots is given by PredCSIi <t,t+m>, wherein the CSI comprises a Channel Quality Indicator and a Rank Indicator.

Description

64K3052 OPTIMIZING RADIO RESOURCE MANAGEMENT DECISIONS IN O-RAN NETWORKS USING PREDICTED INFORMATION BACKGROUND

[0001] The present disclosure is related to Open Radio Access Network (O-RAN)wireless networks and radio resource management (RRM) policies in O-RAN Networks. SUMMARY

[0002] Described are systems, methods, and computer program products for asystem configured to execute Artificial Intelligence Machine Learning AIML models.

[0003] In an implementation, described is a method for a system executing AIMLmodels, the method comprising: computing scheduler metric of each logical channel PLC using, where for an User Equipment (UE) Channel State Indicator (CSI) at time instant is given by and predicted CSI for next slots is given by,, wherein CSI comprises a Channel Quality Indicator (CQI) and a Rank Indicator (RI). The scheduler metric can further comprise: a Priority metric for each user (or UE) with respect to its ownpredicted CSI for a future time interval ( for user i); and a Priority metric of a userrelative to predicted CSI of all other users in that cell; a Priority metric for each user withrespect to its predicted CSI. The method can further comprise computing g for useri) by calculating an average of predicted CSI over slots for each user, denoted as ,; and computing the Priority metric, as:+=wherein is filter a coefficient and 0 < 1, and wherein (t, t+ ) is thepriority metric for LC i corresponding to UE i with respect to the predicted CSI where the predicted CSI is available for the time interval (t, t+ ). (t, t+ ) can take a higher value when CSI at time t, is greater than or equal to its average predicted CSI for time interval (t,t+m),,, so that the RRM attempts to serve packets for this LC i for the corresponding UE i at time t; or (t, t+ ) can take a lower value, when CSI at64K3052 time t, is less than its average predicted CSI for time interval (t,t+m), , . so that the RRM attempts to delay serving packets for this LC i (for the corresponding UE i) beyond time t.

[0004] The method can further comprise: computing average predicted CSI values,for each user i among the users and computing a minimum value among the users be,corresponding to user and the maximum value be,corresponding to user; and calculating a range as ,,.

[0005] The method can further comprise: computing a predicted CSI for user irelative to other users in the cell , when,is not equal to , ): (, + ) = , ,

[0006] a BufferOccupancy (BO) for each LC in RLC queue as well. For LC corresponding to the UE, BO at time instant and predicted BO for next slots i,, wherein the priority metric for each LC i with respect to predicted BO is computed as: calculating an average of predicted BO over next slots,for LC, and calculating a priority metric for each LC i with respect to its predicted BO aswherein is filter coefficient and 0 < 1.

[0007] (t, t+ ) can take a higher value when BO at time t, is greater thanits average predicted BO for time interval (t, t+ ),,so that the RRM can64K3052 choose to delay serving packets for this LC beyond t; or (t, t+ ) can take a lower value (i.e. less than one) when BO at time t, is less than its average predicted BO for time interval (t, t+ ),,so that the RRM attempts to serve packets for this LC i (for the corresponding UE i) at time t.

[0008] The method can further comprise calculating the Priority metric with respectto predicted BO relative to all the other LCs by: calculating an average predicted BO values among the LCs, wherein a minimum value among the LCs,corresponding to LC and a maximum value,corresponding to LC; calculating a range as, ,; computing a relative predicted BO PBorel when the maximum value of the average predicted B,,is not equal to theminimum value of average predicted BO, , :( , + ) = , ,wherein, ,,.

[0009] The scheduling metric of a logical channel , for LC I can be computed as:,= function of { PLC,i, , , ,is the,, and is the weight corresponding to and is the weight corresponding to for LC i (of UE i).

[0010] The method can comprise: computing quantized values for ,, and .

[0011] The method can further comprise: adjusting the values weights used tocompute an overall priority metric P for LC I using predicted CSI and BO values,and computing g the metrics P , using P and P .64K3052

[0012] The method can further comprise: computing the AIML model is located at aNear-RT-RIC and , , and are computed at the Near-RT-RIC and communicated to a DU. Where the AIML model is located at the CU-CP, the method can further comprise computing , , and at the CU-CP and communicated to a DU via an F1AP protocol running over an F1-C interface between a CU- CP and a DU.

[0013] In an implementation, described is a system configured to execute AIMLmodel, the system being configured to: compute a scheduler metric of each logical channel PLC, where for an User Equipment (UE) Channel State Indicator (CSI) at time instant is given by and predicted CSI for next slots is given by,, wherein CSI comprises a Channel Quality Indicator (CQI) and a Rank Indicator (RI). The scheduler metric can further comprise: a Priority metric for each user (or UE) with respect to its ownpredicted CSI for a future time interval ( for user i); and a Priority metric of a userrelative to predicted CSI of all other users in that cell; a Priority metric for each user with respect to its predicted CSI. The system can be further configured to: compute for user i) by calculating an average of predicted CSI over slots for each user, denoted as , ; and compute the Priority metric, : (, + ) =wherein is a < is thepriority metric for LC i corresponding to UE i with respect to the predicted CSI where the predicted CSI is available for the time interval (t, t+ ).

[0014] The system can be further configured to: predict, by the AIML model, a BufferOccupancy (BO) for each LC in RLC queue as well. For LC corresponding to the UE, BO at time instant and predicted BO for next slots i,, wherein the priority metric for each LC i with respect to predicted BO is computed by:64K3052 calculating an average of predicted BO over next slots,for LC; and calculating a priority metric for each LC i with respect to its predicted BO as (, + ) =whereinBRIEF DESCRIPTION OF THE DRAWINGS

[0015] FIG. 1a is a block diagram of a system architecture.

[0016] FIG. 1b shows an example of a User Plane Stack.

[0017] FIG. 2 shows an example of a Control Plane Stack.

[0018] FIG. 3 shows an example of high-level NG-RAN including a gNB CU and DU.

[0019] FIG.4 shows an example of a Separation of CU-CP (CU-Control Plane) and CU-UP (CU-User Plane) in a 5G gNB.

[0020] FIG. 5 shows a DL (Downlink) Layer 2 Structure.

[0021] FIG. 6 shows an exemplary logical flow for implementing an RB allocationpolicy.

[0022] FIG. 7 shows an L2 Data Flow example.

[0023] FIG. 8a shows an example of an O-RAN architecture.

[0024] FIG. 8b shows a logical flow for an O-RAN architecture.

[0025] FIG. 9 illustrates a PDU Session architecture comprising of multiple DRBs andmultiple QoS Flows.

[0026] FIG. 10 illustrates a PDU Session flow comprising multiple DRBs.64K3052

[0027] FIG. 11 illustrates a CU and DU view on PDU session, DRBs and GTP-Utunnels for a 5G network architecture.

[0028] FIG. 12 illustrates a logical flow for an AIML supported scheduler metric flow.

[0029] FIG. 13 logical flow for quantized values.

[0030] FIG. 14a shows flow for a computation of a quantized value.

[0031] FIG. 14b shows a flow for a computation of a quantized value.

[0032] FIG. 15 illustrates a logical flow for AIML models located at the CU-CP.DETAILED DESCRIPTION

[0033] In the following sections, overview of Next Generation Radio Access Network(NG-RAN) architecture and 5G New Radio (NR) stacks will be discussed.5G NR (New Radio) user and control plane functions with monolithic gNB (gNodeB) are shown in FIGS. 1a, 1b and 2. For the user plane (shown in FIG.1a, which is in accordance with 3GPP TS 38.300), PHY (physical), MAC (Medium Access Control), RLC (Radio Link Control), PDCP (Packet Data Convergence Protocol) and SDAP (Service Data Adaptation Protocol) sublayers originate in the UE 101 and are terminated in the gNB 102 on the network side.

[0034] As shown in FIG. 1b, which is a block diagram illustrating the user planeprotocols stacks for a PDU session, in accordance with 3GPP TS 23.501, PDU layer 9010 corresponds to the PDU carried between the UE 101 and the data network (DN) 9011 over the PDU session. As shown in FIG.1b, UE 101 is connected to the 5G access network (AN) 902, which AN 902 is in turn connected via the N3 interface to the Intermediate UPF (I- UPF) 903a portion of the UPF 903, which I-UPF 903a is in turn connected via the N9 interface to the PDU session anchor 903b portion of the UPF 903, and which PDU session anchor 903b is connected to the DN 9011. The PDU session can correspond to IPv4, IPv6, or both types of IP packets, when the PDU session is of type IPv4, IPv6 or IPv4v6, respectively. GTP-U shown in FIG.1b supports tunnelling user plane data over N3 and N9 interfaces and provides encapsulation of end user PDUs for N3 and N9 interfaces.64K3052

[0035] For the control plane, as shown in FIG. 2 in accordance with 3GPP TS 38.300,RRC (Radio Resource Control), PDCP, RLC, MAC and PHY sublayers originate in the UE 101 and are terminated in the gNB 102 on the network side, and NAS (Non-Access Stratum) originate in the UE 101 and is terminated in the AMF (Access Mobility Function) 103 on the network side.

[0036] NG-Radio Access Network (NG-RAN) architecture from 3GPP TS 38.401 isshown in FIGS.3-4. As shown in FIG.3, the NG-RAN 301 comprises of a set of gNBs 302 connected to the 5GC 303 through the NG interface. Each gNB comprises gNB-CU 304 and one or more gNB-DU 305 (see FIG.3). As shown in FIG.4, which illustrates separation of CU-CP (CU-Control Plane) and CU-UP (CU-User Plane), E1 is the interface between gNB-CU- CP (CU-Control Plane) 304a and gNB-CU-UP (CU-User Plane) 304b, F1-C is the interface between gNB-CU-CP 304a and gNB-DU 305, and F1-U is the interface between gNB-CU-UP 304b and gNB-DU 305. As shown in FIG.4, gNB 302 comprises a gNB-CU-CP 304a, multiple gNB-CU-UPs (or gNB-CU-UP instances) 304b and multiple gNB-DUs (or gNB-DU instances) 305. A gNB-DU 305 is connected to gNB-CU-CP 304a, and gNB-CU-UP 304b is connected to gNB-CU-CP 304a.

[0037] In the example, F1-AP (F1-Application Protocol) running on F1-C is specifiedin 3GPP TS38.473 version 18.1.0, and NR-U (NR User Plane) running on F1-U is specified in 3GPP TS38.425 version 18.0.0.

[0038] In this section, an overview of Layer 2 (L2) of 5G NR is provided inconnection with FIGS.5-7. L2 of 5G NR is split into the following sublayers in accordance with 3GPP TS 38.30:

[0039] 1) Medium Access Control (MAC) 501 in FIGS. 5-7: Logical Channels (LCs) areSAPs (Service Access Points) between the MAC and RLC layers. This layer runs a MAC scheduler to schedule radio resources across different LCs (and their associated radio bearers). For the downlink direction, the MAC layer processes and sends RLC PDUs received on LCs to the Physical layer as Transport Blocks (TBs). For the uplink direction, it64K3052 receives transport blocks (TBs) from the physical layer, processes these and sends to the RLC layer using the LCs.

[0040] 2) Radio Link Control (RLC) 502 in FIGS. 5-7: The RLC sublayer presents RLCchannels to the Packet Data Convergence Protocol (PDCP) sublayer. The RLC sublayer supports three transmission modes: RLC-Transparent Mode (RLC-TM), RLC- Unacknowledged Mode (RLC-UM) and RLC-Acknowledgement Mode (RLC-AM). RLC configuration is per logical channel. It hosts ARQ (Automatic Repeat Request) protocol for RLC-AM mode.

[0041] 3) Packet Data Convergence Protocol (PDCP) 503 in FIGS. 5-7: The PDCPsublayer presents Radio Bearers (RBs) to the SDAP sublayer. There are two types of Radio Bearers: Data Radio Bearers (DRBs) for data and Signaling Radio Bearers (SRBs) for control plane.

[0042] 4) Service Data Adaptation Protocol (SDAP) 504 in FIGS. 5-7: The SDAP mapsQoS flows in a PDU session to a specific Data Radio Bearer.

[0043] FIG. 5 is a block diagram illustrating DL L2 structure, in accordance with3GPP TS 38.300. FIG.6 is a block diagram illustrating UL L2 structure, in accordance with 3GPP TS 38.300. FIG.7 is a block diagram illustrating L2 data flow example, in accordance with 3GPP TS 38.300 (in FIG.7, H denotes headers or sub-headers).

[0044] Open Radio Access Network (O-RAN) is based on disaggregated componentswhich are connected through open and standardized interfaces based on 3GPP NG-RAN. An overview of O-RAN with disaggregated RAN CU (Centralized Unit), DU (Distributed Unit), and RU (Radio Unit), near-real-time Radio Intelligent Controller (RIC) and non-real-time RIC is illustrated in FIG.8.

[0045] As shown in FIG. 8a, the CU (shown split as O-CU-CP 801a and O-CU-UP801b) and the DU (shown as O-DU 802) are connected using the F1 interface (with F1-C for control plane and F1-U for user plane traffic) over a mid-haul (MH) path. One DU can host multiple cells (e.g., one DU can host 24 cells) and each cell can support many users. For64K3052 example, one cell can support 800 Radio Resource Control (RRC)-connected users and out of these 800, there can be 250 Active users (i.e., users that have data to send at a given point of time).

[0046] A cell site can comprise multiple sectors, and each sector can supportmultiple cells. For example, one site can comprise three sectors and each sector can support eight cells (with each cell being on a different frequency band in a given sector). One CU-CP (CU-Control Plane) can support multiple DUs and thus multiple cells. For example, a CU-CP can support 500 cells and around 100,000 User Equipment (UEs). Each UE can support multiple Data Radio Bearers (DRBs) and there can be multiple instances of CU-UP (CU-User Plane) to serve these DRBs. For example, each UE can support 4 DRBs, and 400,000 DRBs (corresponding to 100,000 UEs) can be served by five CU-UP instances (and one CU-CP instance).

[0047] The DU can be located in a private data center, or it can be located at a cell-site. The CU can also be in a private data center or even hosted on a public cloud system. The DU and CU, which are typically located at different physical locations, can be tens of kilometers apart. The CU communicates with a 5G core system, which can also be hosted in the same public cloud system (or can be hosted by a different cloud provider). A RU (Radio Unit) (shown as O-RU 803 in FIG.8) is located at a cell-site and communicates with the DU via a front-haul (FH) interface.

[0048] The E2 nodes (CU and DU) are connected to the near-real-time RIC 132 usingthe E2 interface. The E2 interface is used to send data (e.g., user and / or cell KPMs) from the RAN, and deploy control actions and policies to the RAN at near-real-time RIC 132. The applications or services at the near-real-time RIC 132 that deploys the control actions and policies to the RAN are called xApps. During the E2 setup procedures, the E2 node advertises the metrics it can expose, and an xApp in the near-RT RIC can send a subscription message specifying key performance metrics which are of interest. The near- real-time RIC 132 is connected to the non-real-time RIC 133 (which is shown as part of Service Management and Orchestration (SMO) Framework 805 in FIG.8) using the A1 interface. The applications that are hosted at non-RT-RIC are called rApps. Also shown in64K3052 FIG.8 are O-eNB 806 (which is shown as being connected to the near-real-time RIC 132 and the SMO Framework 805) and O-Cloud 804 (which is shown as being connected to the SMO Framework 805).

[0049] As in FIG. 8b, E2 node (which is DU or CU) and Near-RT-RIC establish E2session using E2 SETUP REQUEST and E2 SETUP RESPONSE. Near-RT-RIC can subscribe to certain parameters from the E2 node (on behalf the xApp running at Near-RT-RIC) using the RIC SUBSCRIPTION REQUEST and E2 node acknowledges this message by sending RIC SUBSCRIPTION RESPONSE to the Near-RT-RIC. As part of this, xApp running at the Near- RT-RIC also provides the event triggers to E2 node, e.g. it can ask E2 node to REPORT subscribed parameters periodically to the xApp or to REPORT these subscribed parameters based on certain events to the xApp. E2 node communicates subscribed parameters to Near-RT-RIC (and the xApp) using RIC INDICATION as shown in FIG.8b. After analyzing received parameters from the E2 nodes (and based on network operator policies), Near- RT-RIC can send RIC CONTROL REQUEST to take an action at the E2 node (e.g. influence mobility decision). E2 node acknowledges this message by sending RIC CONTROL ACKNOWLEDGE to Near-RT-RIC while E2 node takes action as asked by the Near-RT-RIC.

[0050] In this section, PDU sessions, DRBs, and Quality of Service (QoS) flows EWare described. In 5G networks, PDU connectivity service is a service that provides exchange of PDUs between a UE and a Data Network (DN) identified by a Data Network Name (DNN). The PDU Connectivity service is supported via PDU sessions that are established upon request from the UE. The DNN defines the interface to a specific external data network. One or more QoS flows can be supported in a PDU session. All the packets belonging to a specific QoS flow have the same 5QI (5G QoS Identifier). A PDU session includes the following: Data Radio Bearers which are between UE and CU in RAN; and an NG-U GTP tunnel which is between CU and UPF (User Plane Function) in the core network. FIG.9 illustrates an example PDU session (in accordance with 3GPP TS 23.501) comprising multiple DRBs, where each DRB can include of multiple QoS flows. In FIG.9, three components are shown for the PDU session 901: UE 101; access network (AN) 902; and UPF 903, which includes Packet Detection Rules (PDRs) 9031.64K3052

[0051] A 3GPP 5G network architecture is illustrated in FIG. 10. In the context ofmultiple PDU sessions involving multiple DRBs and QoS Flow Identifiers (QFIs), which PDU sessions are implemented involving UE 101, gNodeB 102, UPF 903, and DNNs 9011a and 9011b) and FIG.11 (in the context of Radio Resource Management (RRM) for connecting UE 101 to the network via RU 306 with a MAC Scheduler 1001):

[0052] 1) The transport connection between the base station (i.e., CU-UP 304b ofFIG.11) and the UPF 903 uses a single GTP-U tunnel per PDU session, as shown in FIGS.10 and 11. The PDU session is identified using GTP-U TEID (Tunnel Endpoint Identifier).

[0053] 2) The transport connection between the DU 305 and the CU-UP 304b of FIG.11 uses a single GTP-U tunnel per DRB (see also FIG.10 and FIG.11). The DU is provided with an UL GTP-U TEID and the CU is provided with the corresponding DL GTP-U TEID to allow for data communication for that DRB between DU and CU-UP.

[0054] 3) SDAP:

[0055] a) The SDAP (Service Adaptation Protocol) 504 Layer receives downlink datafrom the UPF 903 across the NG-U interface (see FIG.11).

[0056] b) The SDAP 504 maps one or more QoS Flow(s) onto a specific DRB.

[0057] c) The SDAP header is present between the UE 101 and the CU (whenreflective QoS is enabled) and includes a field to identify the QoS flow within a specific PDU session.

[0058] 4) GTP-U protocol includes a field to identify the QoS flow and is presentbetween CU and UPF 903 (in the core network).

[0059] 5) One (logical) DU (or RLC) queue exists per DRB (or per logical channel) forRLC PDUs that are to be transmitted for the first time, as shown in FIG.11. Separate logical queues can exist in DU for packets that are to be retransmitted to UE.64K3052

[0060] In this section, standardized 5QI to QoS characteristics mapping arediscussed. As per 3GPP TS 23.501, the one-to-one mapping of standardized 5QI values to 5G QoS characteristics is specified in Table 1 shown below. The first column represents the 5QI value. The second column lists the different resource types, i.e., as one of Non-GBR, GBR, Delay-critical GBR. The third column (“Default Priority Level”) represents the priority level Priority5QI, for which lower the value the higher the priority of the corresponding QoS flow. The fourth column represents the Packet Delay Budget (PDB), which defines an upper bound for the time that a packet can be delayed between the UE and the N6 termination point at the UPF. The fifth column represents the Packet Error Rate (PER). The sixth column represents the maximum data burst volume for delay-critical GBR types. The seventh column represents averaging window for GBR, delay critical GBR types. Note that only a subset of 5QI values defined in 3GPP TS 23.501 are shown in Table 1 below.

[0061] For example, as shown in Table 1, 5QI value 1 is of resource type GBR withthe default priority value of 20, PDB of 100ms, PER of 0.01, and averaging window of 2000 ms. Conversational voice falls under this category. Similarly, as shown in Table 1, 5QI value 7 is of resource type Non-GBR with the default priority value of 70, PDB of 100ms and PER of 0.001. Voice, video (live streaming), and interactive gaming fall under this category.64K3052 Table 1

[0062] In this section, Radio Resource Management (RRM)is discussed. A blockdiagram for an example RRM with a MAC Scheduler is shown in FIG. ). L2 methods (such as64K3052 MAC scheduler) play a critical role in allocating radio resources to different UEs in a cellular network. For example, the scheduling priority of a logical channel (PLC) can be determined as part of MAC scheduler using one of the following (or some other variant) for a LC sending data in the downlink direction: PLC = W5QI*P5QI+ WGBR*PGBR+WPDB* PPDB +WBO*PBO + WPF*PPF, or PLC= (W5QI*P5QI) *(WPF*PPF) * (WGBR*PGBR)*(WPDB* PPDB), or PLC= (W5QI*P5QI+ WPF*PPF) * maximum (WGBR*PGBR, WPDB* PPDB) + WBO*PBO, or PLC= (W5QI*P5QI+ WPF*PPF) + maximum (WGBR*PGBR, WPDB* PPDB) + WBO*PBOOnce one of the above methods is used to compute scheduling priority of a logical channel corresponding to a UE in a cell, the same method is used for all other UEs and these scheduling priorities are used to determine the resources to be allocated to each LC in each cell.

[0063] In the above expressions, the parameters are defined as follows:

[0064] P5QI is the priority metric corresponding to the QoS class (5QI) of the logicalchannel. Incoming traffic from a DRB is mapped to Logical Channel (LC) at RLC level. P5QI is a function of the default 5QI priority value, Priority5QI, of a QoS flow that is mapped to the current LC. The lower the value of Priority5QIthe higher the priority of the corresponding QoS flow. For example, Voice over New Radio (VoNR) (with 5QI of 1) will have a higher P5QIcompared to web browsing (with 5QI of 9).

[0065] PGBR is the priority metric corresponding to the target bit rate of thecorresponding logical channel. The GBR metric PGBR represents the fraction of data that must be delivered to the UE within the time left in the current averaging window Tavg_win (as per 5QI table, default is 2000 msec.) to meet the UE’s GBR requirement. PGBR is calculated as follows: PGBR = remData / targetData64K3052 where targetData is the total data bits to be served in each averaging window Tavg_win in order to meet the GFBR (Guaranteed Flow Bit Rate) of the given QoS flow; remData is the amount of data bits remaining to be served within the time left in the current averaging window; PGBRis reset to 1 (or some other suitable value) at the start of each averaging window Tavg_win, and should go down to 0 towards the end of this window if the GBR criterion is met; and PGBR= 0 for non-GBR flows. For GBR DRB m corresponding to UE h, remData at time t is denoted as remData(h, m; t) and targetData at time t is denoted as targetData(h, m; t).

[0066] PPDB is the priority metric corresponding to the packet delay budget at DU forthe corresponding logical channel. PPDB = 1 if PDBDU<=QDelayRLC and PPDB = 1 / (PDBDU- QDelayRLC) if PDBDU> QDelayRLC where both PDBDU(Packet Delay Budget at DU) and RLC Queuing delay, QDelayRLC, are measured in terms of (time) slots. For example, each (time) slot can be equal to 1 ms or 0.5 ms.

[0067] ‘Slot’ and ‘time slot’ are used interchangeably in this document.

[0068] QDelayRLC = (t –TRLC) is the delay of the oldest RLC packet in the QoS flowthat has not been scheduled yet, and it is calculated as the difference in time between the SDU insertion in RLC queue to current time where t := current time instant, TRLC := time instant when oldest SDU was inserted in RLC.

[0069] Packet delay budget at DU is denoted as PDBDU. Waiting time for HoL RLCpacket for DRB m corresponding to UE h at time t (i.e. in slot t) is denoted as QDelayRLC(h, m; t).64K3052

[0070] PPF is the priority metric corresponding to proportional fair metric of the UE.PPF is the PF Metric, calculated on a per UE basis as =where r: It is the UE’s achievable data rate. DU considers CSI (Channel Status Information) which also includes CQI (Channel Quality Indication), reported by UE to compute this; Ravg = a *Ravg + (1-a) *b , UE’s exponentially weighted moving average throughput, where b>=0 is the number of bits scheduled in the current slot and parameter ‘a’ is selected such that 0 < a <= 1. 1and 1 are configurable parameters. For example, if one sets 1=1 and 1 = 0, thepriority metric, PPF, works in greedy way and favors UEs in good channel conditions. This helps to improve cell throughput but need not be fair to individual logical channels and some of these LCs may not meet their QoS requirements.

[0071] For some existing systems, 1 and 1 are in the range of 0 to 1. Value canbe be upper bounded by 1_max (and lower bounded by zero). As a LC is eventually selected by the overall scheduling priority of a logical channel (PLC) which has multiple other factors (and not only the PPF metric), 1_max can be even higher than one (for example, 1_max = 1.2) to help design and enforce various type of policies (and associated service level agreements at per-cell, per-DU and per-logical channel level). Similarly, 1, isupper bounded by 1_ , and lower bounded by zero.

[0072] For UE h, achievable data rate in slot t is denoted as r(h; t) and this isinflucned by CSI reported by UE h for slot t which is denoted as CSI(h; t). Also, UE’s weighted average throughput for UE h at the beginning of slot t or at the end of the slot (t- 1) is denoted as Ravg(h; t-1).

[0073] BO is the (normalized) buffer occupancy in the RLC queue (e.g. in the RLCqueue at DU for traffic in downlink direction for a DRB). PBO is the normalized value of buffer occupancy across all DRBs which can be proportional to the value of BO.64K3052

[0074] RLC queue (normalized) BO for DRB m corresponding to UE h at time t isdenoted as RLCBO(h, m; t).

[0075] In addition, the following weights are defined: W5QI is the weight of P5QI; WGBRis the weight of PGBR; WPDB is the weight of PPDB; WBO is the weight of PBO and WPF is the weight of PPF. For example, each of the above weights can be set to a value between 0 and 1 though other suitable set of values can be chosen too.

[0076] For an LC sending data in the uplink direction, scheduling priority of thatlogical channel, PLCcan be computed using one of the following methods (or some other variant): PLC= W5QI*P5QI+ WGBR*PGBR+WBSR*PBSR+ WPF*PPF, or PLC= (W5QI*P5QI) *(WPF*PPF) * (WGBR*PGBR)*(WBSR* PBSR).

[0077] For UL, UE reports BSR (Buffer Status Report) to DU based on the data whichis waiting in the UL queues at the UE and this BSR can be used to estimate the (UL) BO. Different LCGs (Logical Channel Groups) can be used depending on the type of traffic. For example, one LCG can be used for GBR traffic and another GBR can be used for non-GBR traffic. A separate LCG can be used for signalling traffic.

[0078] For each GBR DRB m (corresponding to UE h) which is sending data in theuplink direction, remData(h,m;t) and targetData(h,m;t) at time t are computed by monitoring data received in the UL direction from the UE at the DU.

[0079] The PF metric for a UE can be computed using the MCS and thecorresponding TB (Transport Block) size for the UL direction for that UE. The weighted average throughput (Ravg) for the UL traffic from each UE can be computed at the DU by monitoring UL traffic from that UE.

[0080] In this section, basic structures and characteristics of neural networks will bediscussed. Recurrent neural networks (RNNs) are a family of neural networks that are64K3052 suited for handling sequential data. RNNs use a hidden state associated with each time-step and the output at each time step is computed using the input and the previous hidden state.

[0081] RNN architectures suffer from some limitations. First, RNNs fail to storeinformation for a long period of time and thus do not handle the situations well in which a reference to certain information stored quite a long time ago is required to predict the current output. Second, there is no fine control to select which part of the context needs to be carried forward and or can be “forgotten”. Also, in RNNs, gradients in early layers are computed as the product of terms from later layers. With this, gradients in early layers can grow exponentially large (e.g. if the terms in later layers are large enough) or can exponentially decrease (e.g. if terms in later layers are small).

[0082] Long Short-Term Memory (LSTM) networks, which are an extension ofrecurrent neural networks (RNNs), have been introduced to handle situations where RNNs do not work well. The basic difference between the architectures of RNNs and LSTMs is that the hidden layer of LSTM is a gated unit or a gated cell, which has four layers that interact with one another in a way to produce the output of that cell along with the cell state. The output and the cell state are then passed onto the next hidden layer. Unlike RNNs which have only a single neural net layer of tanh (hyperbolic tangent), LSTMs comprise three logistic sigmoid gates and one tanh layer. It should be noted that output of sigmoid activation function is in the range (0 to 1) for any real value as input, while the output of tanh is in the range (-1 to 1).

[0083] Gates are provided in LSTM in order to limit the information that is passedthrough the cell, i.e., the gates determine which part of the information will be needed by the next cell and which part is to be discarded. The output is usually in the range of 0 to 1, where “0” means “reject all”, and “1” means “include all”. Information is retained by the cells, and the memory manipulations are done by the gates. Three gates are provided in LSTM: 1) forget gate; 2) input gate; and 3) output gate.

[0084] In this section, basic working on Deep Neural Networks (DNNs) such as MLP(Multi-Layer Perceptron) with higher number of hidden layers is briefly described. DNNs64K3052 train the model to learn the dependencies between the target and the independent variables from a dataset. These architectures have three types of layers: input layer, hidden layers and output layer. DNNs can map nonlinear input to output by extracting subtle patterns and multiple features from a dataset through each layer.

[0085] The input layer processes the input data and passes it to the hidden layer.Each hidden layer processes the output from the previous layer and passes it to the next layer. Activation functions such as ReLU, sigmoid and tanh are used. The output layer produces the DNN result. Weights used in the DNN architecture are usually updated using gradient descent process where the gradient indicates how the weights should change in order to reduce the error or loss (i.e. the gap between the output produced by the DNN model based on its current outputs and the correct output). The training process is repeated iteratively to continuously reduce the overall error (or loss) until the error (or loss) is below a predefined threshold. Once a DNN is trained, it can compute the output of the network using the weights derived during the training process.

[0086] In a wireless network environment, channel conditions can vary dependingon many factors such as atmospheric conditions, multi-path propagation, and mobility. So, mere interpolation of channel estimates will not promise optimal performance.

[0087] Also, traffic conditions vary for services like video streaming, live streamingand the like.

[0088] It is possible to use AIML techniques (such as LSTM or Transformer basedDNN architectures) to predict traffic patterns and thus predict BO for RLC queues in DU for each DRB and use this for radio resource management related decisions. One such method is presented in Indian Patent application number 202321076223 (having a filing date November 8, 2023), incorporated in its entirety by reference hereby.

[0089] It is also possible to use AIML techniques (such as deep neural networks) topredict channel state information (CSI).3GPP TR38.843 version 18.0.0 considers some such methods.64K3052

[0090] CSI and BO play a key role in RRM related decisions for UEs in a cell. Withpredicted channel and traffic conditions, there is scope to enhance the computation of scheduler metric and to optimize RRM decisions.

[0091] It is also desirable to reduce computation effort with these AIML-assistedRRM methods especially for the cases when base station does not support sufficient software and hardware resources to run such methods.

[0092] In some scenarios, AIML module may not be able to predict values of some ofthese parameters very accurately. For example, it can happen when the AIML model is not trained with sufficient data (e.g. during the initial deployments). Also, values of the predicted parameters can become stale, and it is important to take right near-optimal RRM decisions for such scenarios also. Implementations as described herein advantageously address these issues.

[0093] RRM Optimizations

[0094] METHOD IA

[0095] As discussed earlier, AIML based models are used to predict BO for nextslots and CSI for next slots. A scheduler metric of each logical channel (or the corresponding DRB), PLC, is computed using current and past information. FIG.12 shows an implementation of an AIML supported scheduler metric flow configured take into account predicted BO and predicted CSI to optimize RRM in O-RAN systems. As shown in FIG.12, RRM is enhanced to take into account predicted BO and predicted CSI to optimize radio resource management in O-RAN systems.

[0096] For UE, CSI at time instant is given by and predicted CSI for nextslots is given by,.

[0097] CSI comprises CQI (Channel Quality Indicator), RI (Rank Indicator) and otherparameters. In general, CSI as used herein denotes CQI part of CSI, but can also denote another number derived from CQI, RI and optionally other parameters.64K3052

[0098] This method considers the following when computing scheduler metricrelated to predicted CSI: -(a) Priority metric for each user (or UE) with respect to its own predicted CSI fora future time interval -(b) Priority metric of a user relative to predicted CSI of all other users in that cell

[0099] (a) Priority metric for each user with respect to its predicted CSI (denoted asfor user i) is computed as follows: - Calculate average of predicted CSI over slots for each user, denoted as , . - Priority metric, , is computed as given below: (, + ) =Where

[0100] Note that (t, t+ ) is the priority metric for LC i (corresponding to UEi) with respect to predicted CSI where predicted CSI is available for the time interval (t, t+ ). This metric, (t, t+ ), is also denoted as in this description here.

[0101] Also, note that (t, t+ ) takes higher value (i.e., greater than or equalto one), when CSI at time t, isthan (or equal to) its average predicted CSI for time interval (t,t+m),,. In such cases, the optimized RRM method given here attempts to serve packets for this LC i (for the corresponding UE i) at time t if possible, as average CSI for user i is expected to become worse beyond time t.

[0102] Conversely, (t, t+ ) takes a lower value (i.e. less than one), when CSIat time t, is less than its average predicted CSI for time interval (t,t+m), , . In such cases, this optimized RRM method given here attempts to64K3052 opportunistically delay serving packets for this LC i (for the corresponding UE i) beyond time t if possible, as average CSI is expected to improve beyond time t.

[0103] (b) Priority metric with respect to predicted CSI relative to all the other usersis computed as follows: -Sort out average predicted CSI values (i.e. , considering eachuser i) among the users. -Let the minimum value among the users be , correspondingto user and the maximum value be,corresponding to user. -Calculate range as ( , , )

[0104] Normalize predicted CSI for user i relative to other users in the cell, denotedas , is computed as follows when,is not equal to , ): += , ,

[0105] , + ,,)

[0106] This metric, (t,t+m), gives an indication of average value of predictedCSI for user i compared to other users in the cell. If ( , + )> ( , + ), itgives an indication to the optimized RRM methodto be better than the average CSI for user q during the time interval (t, t+m).64K3052

[0107] Note that a UE can have multiple logical channels. ( , + ) is for UE iand applies to all (active) logical channels being supported by the UE i. User and UE are used interchangeably in the methods here.

[0108] Each LC in the UE has its own (logical) RLC queue and thus its own BO. Theoptimized RRM method is described here assuming that there is one LC per-UE. Thus, LC refers to UE in the description here. Note that this method is equally applicable for the case when there are multiple LCs communicating data in that UE. For example, there can be three LCs, denoted as i(1), i(2),i(3), for UE (or user) i and this method works for such cases also.

[0109] AIML models are used to predict Buffer Occupancy (BO) for each LC in RLCqueue as well. For LC (which corresponds to UE with the notation being used here), buffer occupancy (BO) at time instant is given by and predicted BO for next slots is given by,.

[0110] Priority metric for each LC i with respect to its predicted BO, denoted asis computed as follows: -Calculate average of predicted BO over next slots, denoted as,for LC. -Calculate priority metric for each LC i with respect to its predicted BO as below:+ =where is filter coefficient and 0 < 1.

[0111] Note that (t, t+ ) is the priority metric for LC i with respect topredicted BO where predicted BO is available for the time interval (t, t+ ). This metric, (t, t+ ), is also denoted as in this description here.64K3052

[0112] Also, note that (t, t+ ) takes higher value (i.e., greater than or equal toone), when BO at time t, , is greater than its average predicted BO for time interval (t, t+ ),,. In such cases, the optimized RRM method given here attempts to serve packets for this LC i (for the corresponding UE i) at time t if possible or it can opportunistically delay serving packets for this LC beyond t as somewhat lower number of new RLC packets expected during the time interval (t, t+ ) for this LC in the DU.

[0113] Conversely, (t, t+ ) takes a lower value (i.e. less than one), when BOat time t, is less than its average predicted BO for time interval (t, t+ ), , . In such cases, this optimized RRM method given here attempts to serve packets for this LC i (for the corresponding UE i) at time t if possible as BO expected to further increase during the time interval (t, t+ ) for this LC in the DU.

[0114] (b) Priority metric with respect to predicted BO relative to all the other LCs iscomputed as follows:

[0115] Sort out average predicted BO values among the LCs. Let the minimum valueamong the LCs be,corresponding to LC and the maximum value be , corresponding to LC.

[0116] Calculate range as ( , , ).

[0117] Normalize,as followswhen maximum value of average predicted BP,,is not equal to minimum value of average predicted BO,,= , ,( , + ) is,,

[0118] This metric, (t,t+ ), gives an indication of average value of predictedBO for LC i compared to other LCs in the cell. If ( , + )> ( , + ), it gives an64K3052 indication to the optimized RRM method that average BO for i is expected to be higher than the average BO for user q during the time interval (t, t+ ).

[0119] Note that this method used predicted BO for applications such as videostreaming (e.g., for 5QI 9, 8 or 6 bearers in 5G network), video conferencing and AR / VR (Augmented Reality / Virtual Reality). It does not need to use predicted BO for low data rate applications such as voice over 5G NR.

[0120] With above, scheduling metric of a logical channel is enhanced, denoted by,for LC i, and is computed as a function of various terms as follows: , = function of { PLC,i, , , ,corresponding to , is the weight corresponding to and is the weight corresponding to for LC i (of UE i).

[0122] For example, , can be computed as, = PLC,i + ( ) + ( ) + ( ) +

[00123] As another example, , can be computed as, = PLC,i * ( ) * ( ) * ( )*current and past information were described earlier. This method computes an enhanced scheduling metric,,, using PLC,i and other factors computed using predicted CSI and predicted BO as described above. As before certain number of LCs with the maximum value of,are selected in each time slot and resources are distributed for these LCs.64K3052

[0125] This RRM method opportunistically tries to delay serving RLC packets for alogical channel based on certain conditions, such as CSI of this UE expected to improve in next few time slots or less number of new RLC packets expected in next few time slots (and thus BO of this LC not expected increase significantly) or this UE expected to get better CSI in next few slots compared to other UEs in the cell or the BO in next the few slots expected to be less compared to other LCs (belonging to different UEs) in the cell.

[0126] At the same time, this RRM method opportunistically tries to serve RLCpackets for a logical channel at current slot based on certain conditions, such as CSI of this UE expected to degrade in the next few slots or higher number of new RLC packets expected in next few time slots (and thus BO of this LC expected to increase significantly) or this UE expected to get worse CSI in next few slots compared to other UEs in the cell or the BO in next the few slots expected to be more compared to other LCs (belonging to different UEs) in the cell.

[0127] Also, note that CSI is predicted over the time interval (t, t+m) for UE i and BOis predicted for LC i (corresponding to UE i) over the time interval (t, t+ ). Here, ‘m’ and ‘can be different. In one policy, RRM optimization methods specified here compute and use average of CSI for each UE i and average of BO for the corresponding LC i over a future timeinterval (t, t+d) where ‘d’ is minimum of ‘m’ and ‘ , i.e. d = min(m, ).

[0128] In another policy, average CSI for UE i is computed using predicted CSI valuesfor the time interval (t, t+m), and average BO for LC i (corresponding to UE i) is computedusing the time interval (t, t+ ) even if ‘m’ and ‘ ’ are different. In this case, it is also ensuredthat each of ‘m’ and ‘ ’ are less than a pre-specified threshold.

[0129] METHOD IB

[0130] This method uses quantized values for , , and.

[0131] In certain deployment options, AIML models to predict CSI and BO can bedeployed at other nodes (such as as CU-CP or at Near-RT-RIC or another aalytics server).64K3052 For such cases, using this method with quantized values can reduce communication overhead over the interface between the node where AIML models are run and the DU. Using quantized values can also simplify computation of the overall scheduling metric at the DU and result in reduced computation effort at the DU.

[0132] An example with three steps of quantized values for CSI and BO is consideredhere. With three steps quantization, each of these can take value as Low (L), Medium (M), or High (H).

[0133] As shown in Table , predicted CSI Priority metric, denoted as , for LC iis computed using quantized values of and .

[0134] For example, this method sets =L when = L and =L.

[0135] = L and =M results in this method setting =L.

[0136] = L and =H results in this method setting =L.

[0137] = M and =L results in this method setting =M.

[0138] = M and =M results in this method

[0139] = M and =H results in this method setting =M or H. Inthis case, value of is chosen as M or H randomly.

[0140] = H and =L results in method setting =H.

[0141] = H and =M results in this method setting =H.

[0142] = H and =H results in this method setting =H

[0143] As can be seen, (which indicates current CSI for UE i relative to aweighted average notion of CSI which includes average predicted value of CSI for this UE i for next few TTIs) takes a more dominant role compared to (which indicates64K3052 average of predicted CSI of this UE i compared to other UEs in the cell) for the computation of when quantization is used to reduce computational effort.

[0144] Similarly, BO priority metric for each LC i, denoted as , is computedusing quantized values of and as shown in Table .

[0145] For example, this method sets = L if = L and =L.

[0146] = L and =M results in this method setting = L.

[0147] = L and =H results in this method setting = L.

[0148] = M and =L results in this method setting = M.

[0149] = M and =M results in this method setting = M.

[0150] = and =H results in this method setting = M or H.In this case, value of is chosen as M or H randomly.

[0151] = H and =L results in this method setting = H.

[0152] = H and =M results in this method setting = H.

[0153] = H and =H results in this method setting = H.

[0154] As can be seen, (which indicates current BO for UE i relative to aweighted average notion of BO which includes average predicted value of BO for this UE i for next few time slots) takes a more dominant role compared to (which indicates average of predicted BO of this LC i (of UE i) compared to other LCs in the cell) for the computation of when quantization is used to reduce computational effort.64K3052 Average Normalized CSI Average Relative BO Predicted Relative Priority Predicted Predicted Priority CSI predicted CSI metric Buffer Buffer metric priority priority Occupancy Occupancy metric metric. priority priority metric metric L L L L L L L M L L M L L H L L H L M L M M L M M M M M M M M H M / H M H M / H H L H H L H H M H H M H H H H H H H Table 2: Computation of and for each LC i using quantized values of,computed as a function of various terms as follows: , = function of { PLC,i, , }

[0156] In this case, quantized values of and are used for the64K3052 computation of,.

[0157] Here, is a weight corresponding to and is a weightcorresponding to .

[0158] FIG. 13 shows a logical flow for quantized values for , ,and .

[0159] Step (1) in FIG. 13 shows computation of quantized value of andStep (2) shows computation of . Step (3) of FIG.13 shows computation of quantized value of and FIGS.14a-14b show computation of quantized value of . Step (5) in FIG.13 shows computation of quantized value of and Step (6) shows computation of . Overall scheduling metric,,, is computed as part of Step (7).

[0160] PLC,i is the scheduling metric for LC i as defined earlier. As described earlier,this depends on current and past values of some parameters. New scheduling metric for LC i,,, depends on past, current and predicted values of certain parameters. METHOD IC

[0162] This method further reduces the communication overhead of previousmethods between the node where AIML models are hosted and the DU. It is also useful for scenarios where required hardware and software resources to run previous methods are not available at the DU. This method also considers the errors in prediction and takes corrective action while computing overall scheduling metric for each LC.

[0163] In some scenarios, some of the above predicted values can go stale or theprediction accuracy may not be very good for certain time intervals. Base station and other modes provide feedback to the AI-ML models to improve the accuracy of predicted CSI and predicted BO but it can take some time to update the models to improve the accuracy. The optimized RRM method described here adjusts the values of certain weights which are used to compute an overall priority metric using predicted CSI and BO values. This overallpriority metric is denoted as P for LC i.64K3052

[0164] This method computes the metric, P , using P and P asshown in Table 3.

[0165] For example, this method sets the value of P to be+ + if P = L and P = L.

[0166] Here, can be either positive or negative depending on the predictionaccuracy of BO and is a step value to influence the value of P . For example, ifpredicted BO is not very accurate for a time interval (e.g. when AIML model is not trained with sufficient data or when a new type of application starts communicating data), a negative value of is used to reduce the impact of the value of predicted BO on the metric,P . Also, value of can be varied to further increase or decrease impact of on themetric, P .

[0167] Similarly, can be either positive or negative depending on the predictionaccuracy of CSI and is a step value to influence the value of P . For example, ifpredicted CSI is not very accurate for a time interval (e.g. when AIML model is not trained with sufficient data), a negative value of is used to reduce the impact of the value ofpredicted CSI on the metric, P . Also, value of can be varied to further increase ordecrease impact of on the metric, P .

[0168] ( , , ) are used to reduce or increase the impact of thecorresponding predicted parameters while computing value of P and eventuallythe overall scheduling metric for each LC.

[0169] In addition, value of weight, , associated with can also bechanged dynamically depending on the observed error in predicting CSI for user i. For example, value of weight can be reduced (or even made zero) if this error in predicting CSI for user i is above a pre-defined threshold for d1 time slots out of d2 time slots. Here, d1 and d2 are cofigurable parameters.64K3052

[0170] Similarly, value of weight, , associated with can also bechanged dynamically depending on the observed error in predicting BO for LC i (for UE i). For example, value of weight can be reduced (or even made zero) if this error in predicting BO for LC i (for UE i) is above a pre-defined threshold for d3 time slots out of d4 time slots. Here, d3 and d4 are cofigurable parameters.

[0171] Note that this method of handling prediction inaccuracies is applicable to theprevious methods specific in this document also. Also, note that errors due to quantization can also be optionally be handled with this method if needed.

[0172] For P = L and P = M, this method sets the value of P tobe + + .

[0173] For P = L and P = H, this method sets the value of P tobe + + .

[0174] For P = M and P = L, this method sets the value of P tobe + + .

[0175] For P = M and P = M, this method sets the value of P tobe + + .

[0176] For P = M and P = H, this method sets the value of P tobe + + .

[0177] For P = H and P = L, this method sets the value of P tobe + + .

[0178] For P = H and P = M, this method sets the value of P tobe + + .

[0179] For P = H and P = H, this method sets the value of P tobe + + .64K3052 Priority metric for Priority metric for Overall Priority metric predicted CSI for LC i predicted BO for LC i due to predicted CSI and predicted BO for LC i PL L + +L M + +L H + +M L + +M M + +M H + +H L + +H M + +H H + +Table 3: Computation of overall priority metric due to predicted CSI and predicted BO

[0180] Note that if value of ( + + ) exceeds the value of quantized level “M”,value of ( + + ) is taken as value of quantized level “M”. Similarly, if value of( + + ) exceeds the quantized value of “H”, value of ( + + ) is taken asvalue of quantized level “H”.

[0181] If value of ( + + ) goes below the value of quantized level “L”, valueof ( + + ) is taken as value of quantized level “L”. Similarly, if value of ( + +64K3052 )goes below the value of quantized level “M”, value of ( + + ) is taken as valueof quantized level “M”.

[0182] Overall scheduling metric of the logical channel i, denoted as , , iscomputed as a function of various terms as follows: , = function of { PLC,i , }

[0183] In thisis used for the computation of,.

[0184] is the weight corresponding to .

[0185] PLC,i is the scheduling metric for LC i as defined earlier. As described earlier,this depends on current and past values of some parameters. New scheduling metric for LC i,,, depends on past, current and predicted values of certain parameters.

[0186] Note that number of quantization levels that are used in a system depends onits hardware and software capabilities (e.g. compute resources available at the DU, communication overhead which can be supported between the node where the AIML models are deployed and the DU for the case where it is not possible to directly deploy AIML models at the DU). In addition, this can also depend on the accuracy of the AIML models used. Also, number of quantized levels for can be different than thenumber of quantized levels used for P and P .

[0187] Values of ( , , ) can be configured or derived using some policiesat the DU. Alternatively,be derived using operator defined policies at the CU-CP or at the Near-RT-RIC and communicated to the DU.

[0188] METHOD 1D64K3052

[0189] AIML models to predict BO and CSI can be located at the DU or CU-CP orNear-RT-RIC or another entity.

[0190] For the case when these models are located at the DU, predicted BO,predicted CSI and other parameters to compute overall scheduling metric are available at the DU itself.

[0191] For the case, where these AIML models are located at the Near-RT-RIC (forexample as xApps): , , and are computed at the Near-RT- RIC and communicated to DU by enhancing the E2 interface. This is shown in FIG.14a.

[0192] Values of ( , , ) can be derived using operator-defined policies atthe Near-RT-RIC and communicated from the Near-RT-RIC to the DU.

[0193] Quantization can also be done at the Near-RT-RIC for the case wherequantized values are used to compute the overall scheduling metric of a LC. In this case, and are computed at the Near-RT-RIC and communicated to DU by enhancing the E2 interface. This is shown in FIG.14b.

[0194] For the case, where these AIML models are located at the CU-CP: ,, and are computed at the CU-CP and communicated to DU by enhancing the F1AP protocol running over the F1-C interface between CU-CP and DU. This is shown in FIG.15.

[0195] Quantization can also be done at the CU-CP for the case where quantizedvalues are used to compute the overall scheduling metric of a LC. In this case, and are computed at the CU-CP and communicated to DU by enhancing the F1AP running over the F1-C interface between CU-CP and DU. This is also shown in FIG.15.

[0196] With this deployment model, values of ( , , ) can also becommunicated from the CU-UP to the DU.

[0197] As shown in FIG. 15, AIML training can be done at SMO.64K3052

[0198] Reference is made to Third Generation Partnership Project (3GPP), O-RANAlliance and the Internet Engineering Task Force (IETF) and related standards bodies in accordance with embodiments of the present disclosure. The present disclosure employs abbreviations, terms and technology defined in accord with Third Generation Partnership Project (3GPP), O-RAN Alliance and / or Internet Engineering Task Force (IETF) technology standards and papers, including the following standards and definitions.3GPP, O-RAN and IETF technical specifications (TS), standards (including proposed standards), technical reports (TR), RFCs and other papers are incorporated by reference in their entirety hereby, define the related terms and architecture reference models that follow. 3GPP TS 23.501 V 18.1.02024-06-26 3GPP TS 38.300 V 18.1.04-03-2024 3GPP TS 38.401 V 18.1.02024-03-29 3GPP TS 38.473 version 18.1.02024-03-29 3GPP TS 38.425 version 18.0.02024-01-12 O-RAN Near-RT-Architecture 6.0, O-RAN.WG3.RICARCH-R003-v06.00, R003, June 2024 O-RAN E2 Service Model (E2SM) KPM 5.0, O-RAN.WG3.E2SM-KPM-R003-v05.00, R003, June 2024 O-RAN E2 Application Protocol (E2AP) 5.0, O-RAN.WG3.E2AP-R003-v05.00, R003, February 2024 O-RAN A1 Interface: Application Protocol 4.02, O-RAN.WG2.A1AP v04.02, R003, June 2024

[0199] A list of abbreviations used in the present specification is provided below:5GC: 5G Core Network 5G NR: 5G New Radio64K3052 5QI: 5G QoS Identifier ACK: Acknowledgement AI: Artificial Intelligence AI / ML (or AIML): Artificial Intelligence and Machine Learning AM: Acknowledged Mode APN: Access Point Name ARP: Allocation and Retention Priority BO: Buffer Occupancy BS: Base Station BSR: Buffer Status Report CNN: Convolution Neural Network CP: Control Plane CSI: Channel State Information CQI: Channel Quality Indicator CU: Centralized Unit CU-CP: Centralized Unit – Control Plane CU-UP: Centralized Unit – User Plane DL: Downlink DDDS: DL Data Delivery Status DNN: Data Network Name DNN: Deep Neural Network DQN: Deep Q Network DRB: Data Radio Bearer DU: Distributed Unit64K3052 eNB: evolved NodeB EPC: Evolved Packet Core EN-DC: E-UTRAN New Radio Dual Connectivity GBR: Guaranteed Bit Rate gNB: gNodeB GTP-U: GPRS Tunnelling Protocol – User Plane IP: Internet Protocol L1: Layer 1 L2: Layer 2 L3: Layer 3 L4S: Low Latency, Low Loss and Scalable Throughput LC: Logical Channel LESS: Low Energy Scheduler Solution LSTM: Long Short-Term Memory MAC: Medium Access Control MDP: Markov Decision Process MIB: Master Information Block ML: Machine Learning MR-DC: Multi-RAT Dual Connectivity NACK: Negative Acknowledgement NAS: Non-Access Stratum NG-RAN: Next Generation Radio Access Network NR-U: New Radio – User Plane NSI: Network Slice Instance64K3052 NSSI: Network Slice Subnet Instance NWDAF: Network Data Analytics Function O-RAN: Open Radio Access Network OAM: Operations, Administration Maintenance PDB: Packet Delay Budget PDCP: Packet Data Convergence Protocol PDU: Protocol Data Unit PER: Packet Error Rate PF: Proportional Fair PHY: Physical Layer PRB: Physical Resource Block QCI: QoS Class Identifier QFI: QoS Flow Identifier QoS : Quality of Service RAN: Radio Access Network RAT: Radio Access Technology RB: Resource Block RDI: Reflective QoS Flow to DRB Indication RL: Reinforcement Learning RLC: Radio Link Control RLC-AM: RLC Acknowledged Mode RLC-UM: RLC Unacknowledged Mode RNN: Recurrent Neural Networks RQI: Reflective QoS Indication64K3052 RRC: Radio Resource Control RRM: Radio Resource Management RTP: Real-Time Transport Protocol RTCP: Real-Time Transport Control Protocol RU: Radio Unit SCTP: Stream Control Transmission Protocol SD: Slice Differentiator SDAP: Service Data Adaptation Protocol SIB: System Information Block SLA: Service Level Agreement S-NSSAI: Single Network Slice Selection Assistance SST: Slice / Service Type TB: Transport Block TCP: Transmission Control Protocol TEID: Tunnel Endpoint Identifier TTI: Transmission Time Interval UE: User Equipment UP: User Plane UL: Uplink UM: Unacknowledged Mode UPF: User Plane Function

[0200] It will be understood that implementations and embodiments can beimplemented by computer program instructions. These program instructions can be provided to a processor to produce a machine, so that the instructions, which execute on the processor, create means for implementing the actions specified herein. The computer64K3052 program instructions can be executed by a processor to cause a series of operational steps to be performed by the processor to produce a computer-implemented process so that the instructions, which execute on the processor to provide steps for implementing the actions specified. Moreover, some of the steps can also be performed across more than one processor, such as might arise in a multi-processor computer system or even a group of multiple computer systems. In addition, one or more blocks or combinations of blocks in the flowchart illustration can also be performed concurrently with other blocks or combinations of blocks, or even in a different sequence than illustrated without departing from the scope or spirit of the disclosure.

Claims

64K3052 CLAIMS 1. A method for a system executing Artificial Intelligence Machine Learning (AIML) models, the method comprising: computing a scheduler metric of each logical channel PLC, where for an User Equipment (UE) Channel State Indicator (CSI) at time instant is given by and predicted CSI for next slots is given by,, wherein CSI comprises a Channel Quality Indicator (CQI) and a Rank Indicator (RI).

2. The method of claim 1, wherein the scheduler metric further comprises: a Priority metric for each user (or UE) with respect to its own predicted CSI for a future time interval ( for user i); anda Priority metric of a user relative to predicted CSI of all other users in that cell; a Priority metric for each user with respect to its predicted CSI.

3. The method of claim 2, further comprising: computing the metric for user i) by calculating an average of predicted CSI over slots for each user, denoted as,; and computing the Priority metric, , as: +=wherein a <wherein (t, t+ ) is the priority metric for LC i corresponding to UE i with respect to the predicted CSI where the predicted CSI is available for the time interval (t, t+ ).

4. The method of claim 3, wherein:64K3052 (t, t+ ) takes a higher value when CSI at time t, is greater than or equal to its average predicted CSI for time interval (t,t+m),,, so that the RRM attempts to serve packets for this LC i for the corresponding UE i at time t; or (t, t+ ) takes a lower value, when CSI at time t, is less than its average predicted CSI for time interval (t,t+m),,. so that the RRM attempts to delay serving packets for this LC i (for the corresponding UE i) beyond time t.

5. The method of claim 3, further comprising: computing average predicted CSI values,for each user i among the users; computing a minimum value among the users be , corresponding to user and the maximum value be , corresponding to user; and calculating a range as, ,.

6. The method of claim 1, further comprising: computing a predicted CSI for user i relative to other users in the cell , when , is not equal to,): ,,7. The method of claim 6, further comprising: predicting, by the AIML-model, a Buffer Occupancy (BO) for each LC in RLC queue as well. For LC corresponding to the UE, BO at time instant and predicted BO for next slots i,, wherein64K3052 the Priority metric for each LC i with respect to predicted BO is computed as: calculating an average of predicted BO over next slots,for LC; and calculating the Priority metric for each LC i with respect to its predicted BO as (, + ) =8. The method of claim 7, wherein: (t, t+ ) takes higher value when BO at time t, is greater than its averageinterval (t, t+ ),,so that the RRM can choose to delay serving packets for this LC beyond t; or (t, t+ ) takes a lower value (i.e. less than one) when BO at time t, is less than its average predicted BO for time interval (t, t+ ),,so that the RRM attempts to serve packets for this LC i (for the corresponding UE i) at time t.

9. The method of claim 8, further comprising calculating the Priority metric with respect to predicted BO relative to all the other LCs by: calculating an average predicted BO values among the LCs, wherein a minimum value among the LCs,corresponding to LC and a maximum value , corresponding to LC; calculating a range as, ,;computing a relative predicted BO PBorelwhen the maximum value of the average predicted B,,is not equal to the minimum value of average predicted BO, ,:64K3052 (, + ) = , ,wherein ( ,, .

10. The method of claim 9, wherein the scheduling metric of a logical channel,for LC i, is computed as: , = function of { PLC,i, , , ,where is the weight corresponding to , is the weight corresponding to , , and is the weight corresponding to and is the weight corresponding to for LC i (of UE i).

11. The method of claim 9, comprising computing quantized values for , , and .

12. The method of claim 10, further comprising: adjusting the values of the weights used to compute an overall priority metricP for LC I using predicted CSI and BO values, andcomputing g the metrics P , using P and P .

13. The method of claim 9, further comprising: wherein the AIML model is located at a Near-RT-RIC and , , and are computed at the Near-RT-RIC and communicated to a DU.64K3052 14. The method of claim 9, wherein the AIML model is located at the CU-CP and , , and are computed at the CU-CP and communicated to DU via an F1AP protocol running over an F1-C interface between a CU-CP and a DU.

17. A system configured to execute AIML models, the system being configured to: compute a scheduler metric of each logical channel PLC, where for an User Equipment (UE) Channel State Indicator (CSI) at time instant is given by and predicted CSI for next slots is given by,, wherein CSI comprises a Channel Quality Indicator (CQI) and a Rank Indicator (RI).

18. The system of claim 17, wherein the scheduler metric further comprises: a Priority metric for each user (or UE) with respect to its own predicted CSI for a future time interval ( for user i); anda Priority metric of a user relative to predicted CSI of all other users in that cell; a Priority metric for each user with respect to its predicted CSI.

19. The system of claim 18, the system being further configured to: compute for user i) by calculating an average of predicted CSI over slots for each user, denoted as , ; and compute the Priority metric, : +=wherein a <wherein (t, t+ ) is the Priority metric for LC i corresponding to UE i with respect to the predicted CSIthe predicted CSI is available for the time interval (t, t+ ).64K3052 20. The system of claim 19, the system being further configured to: predict, by the AIML-model, a Buffer Occupancy (BO) for each LC in RLC queue as well. For LC corresponding to the UE, BO at time instant and predicted BO for next slots i,, wherein the Priority metric for each LC i with respect to predicted BO is computed by: calculating an average of predicted BO over next slots,for LC; and calculating a priority metric for each LC i with respect to its predicted BO as (, + ) =wherein is

Images