Timing for channel state information prediction and monitoring
By predicting future CSI using historical data and machine learning models, and aligning monitoring triggers with inference triggers, the challenges of traditional CSI reporting are addressed, enhancing network efficiency and resource utilization in dynamic environments.
Patent Information
- Application Number
- PCT/CN2026/094770
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-09
- Filing Date
- 2026-05-06
- Publication Date
- 2026-11-12
AI Technical Summary
Traditional CSI reporting mechanisms in wireless communication systems are constrained by their reliance on real-time channel measurements, limiting their ability to predict future channel conditions, and introduce challenges in performance monitoring, buffering, and resource allocation, especially in dynamic and high-mobility scenarios.
Implementing methods for predicting future CSI using historical data and machine learning models, determining SGCS metrics by comparing predicted CSI with ground-truth CSI, and aligning monitoring triggers with inference triggers to optimize resource allocation and reduce buffering overhead.
Enables efficient monitoring of CSI prediction accuracy, conserving memory and processing resources, and improving network operations such as beamforming and interference management, while accommodating diverse network configurations and scenarios.
Smart Images

Figure CN2026094770_12112026_PF_FP_ABST
Abstract
Description
TIMING FOR CHANNEL STATE INFORMATION PREDICTION AND MONITORINGCROSS-REFERENCE TO RELATED APPLICATION
[0001] This Patent Application claims priority to PCT Patent Application No. PCT / CN2025 / 093709, filed on May 9, 2025, entitled “ENHANCED ACCESS FOR FIXED WIRELESS ACCESS USER EQUIPMENT, ” and assigned to the assignee hereof. The disclosure of the prior Application is considered part of and is incorporated by reference into this Patent Application. FIELD OF THE DISCLOSURE
[0002] Aspects of the present disclosure generally relate to wireless communication and specifically relate to techniques, apparatuses, and methods associated with timing for channel state information prediction and monitoring. DESCRIPTION OF THE RELATED TECHNOLOGY
[0003] Wireless communication systems are widely deployed to provide various services, which may involve carrying or supporting voice, text, other messaging, video, data, or other traffic. Typical wireless communication systems may employ multiple-access radio access technologies (RATs) capable of supporting communication among multiple wireless communication devices including user devices or other devices by sharing the available system resources (for example, time domain resources, frequency domain resources, spatial domain resources, or device transmit power, among other examples) . Such multiple-access RATs are supported by technological advancements that have been adopted in various telecommunication standards, which define common protocols that enable different wireless communication devices to communicate on a local, municipal, national, regional, or global level. An example telecommunication standard is New Radio (NR) . NR, which also may be referred to as 5G, is part of a continuous mobile broadband evolution promulgated by the Third Generation Partnership Project (3GPP) . As the demand for connectivity continues to increase, further improvements in NR may be implemented, and other RATs, such as 6G and beyond, may be introduced to enable new applications and facilitate new use cases.SUMMARY
[0004] Some aspects described herein relate to a method for wireless communication performed by user equipment (UE) . The method may include predicting future channel state information (CSI) using historical CSI data and a machine learning (ML) model. The method may include determining a squared generalized cosine similarity (SGCS) metric by comparing the predicted CSI with actual ground-truth CSI. The method may include receiving a monitoring trigger associated with an inference report. The method may include transmitting an indication of the SGCS metric in an SGCS report to a network entity.
[0005] Some aspects described herein relate to a UE. The UE may include a processing system. The processing system may include one or more processors and one or more code-storing memories coupled with the one or more processors. The processing system may be configured to cause the UE to predict future CSI using historical CSI data and an ML model. The processing system may be configured to cause the UE to determine an SGCS metric by comparing the predicted CSI with actual ground-truth CSI. The processing system may be configured to cause the UE to receive a monitoring trigger associated with an inference report. The processing system may be configured to cause the UE to transmit an indication of the SGCS metric in an SGCS report to a network entity.
[0006] Some aspects described herein relate to a non-transitory computer-readable medium that stores a set of instructions for wireless communication by a UE. The set of instructions, when executed by one or more processors of the UE, may cause the UE to predict future CSI using historical CSI data and an ML model. The set of instructions, when executed by one or more processors of the UE, may cause the UE to determine an SGCS metric by comparing the predicted CSI with actual ground-truth CSI. The set of instructions, when executed by one or more processor of the UE, may cause the UE to receive a monitoring trigger associated with an inference report. The set of instructions, when executed by one or more processors of the UE, may cause the UE to transmit an indication of the SGCS metric in an SGCS report to a network entity.
[0007] Some aspects described herein relate to an apparatus for wireless communication. The apparatus may include means for predicting future CSI using historical CSI data and an ML model. The apparatus may include means for determining an SGCS metric by comparing the predicted CSI with actual ground-truth CSI. The apparatus may include means for receiving a monitoring trigger associated with an inference report. The apparatus may include means for transmitting an indication of the SGCS metric in an SGCS report to a network entity.
[0008] The systems, methods, and devices of this disclosure each have several innovative aspects, no single one of which is solely responsible for the desirable attributes disclosed herein.
[0009] Aspects of the present disclosure may generally be implemented by or as a method, apparatus, system, computer program product, non-transitory computer-readable medium, user equipment, network node, wireless communication device, or processing system as substantially described in the Detailed Description with reference to, and as illustrated by, the accompanying drawings. Details of one or more implementations of the subject matter described in this disclosure are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages will become apparent from the description, the drawings, and the claims. Note that the relative dimensions of the following figures may not be drawn to scale.BRIEF DESCRIPTION OF THE DRAWINGS
[0010] Fig. 1 is a diagram illustrating an example of a wireless communication network.
[0011] Fig. 2 is a diagram illustrating an example of channel state information (CSI) prediction.
[0012] Fig. 3 is a diagram illustrating an example of squared generalized cosine similarity (SGCS) reporting.
[0013] Fig. 4 is a diagram illustrating an example of buffering times for calculating SGCS.
[0014] Fig. 5 is a diagram illustrating an examples of CSI processing unit durations.
[0015] Fig. 6 is a diagram illustrating examples of restrictions on a timing for monitoring.
[0016] Fig. 7 is a diagram illustrating an example of a timeline for performance monitoring.
[0017] Fig. 8 is a diagram illustrating an example of a timeline for performance monitoring.
[0018] Fig. 9 is a diagram illustrating examples of reporting SGCSs.
[0019] Fig. 10 is a diagram illustrating an example process performed, for example, at a user equipment (UE) or an apparatus of a UE.
[0020] Fig. 11 is a diagram of an example apparatus for wireless communication.DETAILED DESCRIPTION
[0021] Wireless communication systems often require accurate and efficient channel state information (CSI) reporting to maintain robust communication links and optimize network performance. CSI reporting enables a network to assess the wireless channel conditions between a user equipment (UE) and network entities, such as base stations. Accurate CSI reporting can facilitate tasks such as beamforming, resource allocation, and interference management.
[0022] Traditional CSI reporting mechanisms rely on the UE’s ability to measure and report CSI based on configured CSI reference resources, such as CSI reference signal (CSI-RS) occasions. These measurements are performed at specific time intervals and result in periodic or semi-persistent CSI reports. While effective, these methods are constrained by their reliance on real-time channel measurements and are limited in their ability to predict future channel conditions, which can be critical for dynamic and high-mobility scenarios.
[0023] However, the introduction of CSI prediction creates additional challenges. Once predicted CSI is reported to the network, the CSI must be verified for accuracy. This involves comparing the predicted CSI to “ground truth” CSI obtained from subsequent real-time channel measurements. This comparison enables the network to evaluate the reliability of the CSI prediction model, but the comparison necessitates additional operations, such as ground truth measurements and performance monitoring.
[0024] Performance monitoring for CSI prediction typically involves calculating metrics such as the squared generalized cosine similarity (SGCS) , which quantifies the similarity between predicted CSI and ground truth CSI. These metrics provide insight into the accuracy of the predictions, but the metrics also require the UE to buffer predicted CSI data until the ground truth CSI measurements become available. This buffering introduces time and memory constraints, as the UE must maintain the predicted CSI data for extended periods. Furthermore, the timing of performance monitoring triggers from the network can be unpredictable, requiring the UE to buffer data even when the need for monitoring is uncertain.
[0025] The challenges associated with CSI prediction and performance monitoring extend to CSI processing criteria, including the CSI processing unit (CPU) and active resource / port counting mechanisms. These criteria govern the UE’s ability to process and report CSI while adhering to its capability limitations. For example, the CPU duration (or CPU occupation duration) specifies how long a CSI processing unit is occupied, and active resource / port counting defines the number of CSI-RS resources that the UE can handle simultaneously.
[0026] Further complexities arise in scenarios involving multiple prediction instances or multiple ground truth measurement resources. For example, performance monitoring may involve multiple inference reports and monitoring reports within a configured monitoring window. This necessitates the calculation of multiple SGCS values, which may be reported individually or as aggregated statistics, such as mean values or cumulative distribution functions (CDFs) . The increased computational load and memory requirements associated with these operations can strain the UE’s hardware and processing capabilities.
[0027] Moreover, the timing of monitoring triggers relative to inference triggers is a critical factor. Monitoring triggers that occur long after inference triggers can increase the buffering burden on the UE, resulting in inefficiencies and potential delays in reporting. Conversely, monitoring triggers that occur close to inference triggers can mitigate buffering requirements and streamline the implementation of performance monitoring. However, the absence of standardized rules for monitoring trigger timing can lead to inconsistent implementations and suboptimal network configurations.
[0028] In summary, there are significant technical challenges related to performance monitoring, buffering, CSI processing criteria, and monitoring trigger timing in the context of wireless communication systems. These challenges must be addressed in order to maintain efficient and reliable network operations. Various aspects relate generally to improving performance monitoring for CSI prediction in wireless communication systems. Some aspects more specifically relate to methods for predicting future CSI using historical CSI data and machine learning (ML) models, determining SGCS metrics by comparing the predicted CSI with ground-truth CSI, and transmitting SGCS metrics in reports to network entities. These methods enable efficient monitoring of CSI prediction accuracy while accommodating extended CPU durations and active resource counting.
[0029] In some aspects, SGCS metrics can be determined based on configurable monitoring windows associated with extended CPU durations, allowing the UE to buffer predicted CSI data and calculate performance metrics when ground-truth CSI measurements are available. Monitoring triggers may be received within specified time periods, such as shortly after inference triggers, no later than the first symbol of the inference report measurement resource, or prior to a defined deadline, which ensures that performance monitoring operations are performed efficiently and with reduced buffering overhead. Additionally, extended CPU durations accommodate additional time for operations such as receiving measurement resources for historical CSI and ground-truth CSI, performing prediction inference using ML models, and calculating precoding matrix indicators (PMIs) based on codebooks optimized for Doppler-domain compression. The historical CSI is measured from the measurement resource configured in the inference report configuration.
[0030] In some aspects, the SGCS metrics can include calculations for both predicted CSI versus ground-truth CSI and non-predicted CSI versus ground-truth CSI. Such dual calculations may require doubling CPU processing capacities, but the calculations provide more granular insights into prediction accuracy. Furthermore, active resource and port counting mechanisms can be extended to accommodate SGCS calculations and reporting, ensuring that the UE can efficiently allocate memory and processing resources. For periodic, semi-persistent, or aperiodic monitoring scenarios, the number of CPUs or active resources counted can be adjusted based on the number of prediction instances configured in inference reports, enabling scalable performance monitoring aligned with UE capabilities.
[0031] Particular aspects of the subject matter described in this disclosure can be implemented to realize one or more of the following advantages. By predicting future CSI and calculating SGCS metrics, the described techniques can assess the accuracy of CSI predictions, enabling network entities to optimize operations such as beamforming, resource allocation, and interference management. By reducing unnecessary buffering and aligning monitoring triggers with inference triggers, the techniques may conserve memory resources, processing resources, and / or network resources. Additionally, extended CPU durations and active resource counting mechanisms allow for efficient allocation of processing resources while accommodating additional buffering and calculation requirements, which may reduce computational overhead and improve energy efficiency.
[0032] Enabling dual SGCS calculations provides deeper insights into CSI prediction accuracy and helps validate ML models used for CSI prediction. By aligning CPU durations and active resource counting mechanisms with the number of prediction instances in inference reports, the techniques ensure scalability and adaptability for diverse network configurations. Furthermore, the ability to accommodate both periodic and aperiodic monitoring scenarios ensures that the system can flexibly allocate resources to meet varying network requirements, thereby conserving system resources and improving operational efficiency. In this way, the described innovations contribute to robust and reliable wireless communication systems with efficient resource utilization, particularly in dynamic and high-mobility environments.
[0033] 5G New Radio (NR) may support enhanced mobile broadband (eMBB) access, Internet of Things (IoT) networks or reduced capability (RedCap) device deployments, ultra-reliable low-latency communication (URLLC) applications, or massive machine-type communication (mMTC) , among other examples. To support these and other target verticals, a wireless communication system may be designed to implement a modularized functional infrastructure, a disaggregated and service-based network architecture, network function virtualization, network slicing, multi-access edge computing, millimeter wave (mmWave) technologies including massive multiple-input multiple-output (MIMO) , beamforming, IoT device or RedCap device connectivity and management, industrial connectivity, licensed and unlicensed spectrum access, sidelink and other device-to-device direct communication (for example, cellular vehicle-to-everything (CV2X) communication) , frequency spectrum expansion, overlapping spectrum use, small cell deployments, non-terrestrial network (NTN) deployments, device aggregation, advanced duplex communication (for example, sub-band full-duplex (SBFD) ) , multiple-subscriber implementations, high-precision positioning, radio frequency (RF) sensing, network energy savings (NES) , low-power signaling and radios, or artificial intelligence or machine learning (AI / ML) , among other examples.
[0034] The foregoing and other technological improvements may support use cases, such as wireless fronthauls, wireless midhauls, wireless backhauls, wireless data centers, extended reality (XR) and metaverse applications, meta services for supporting vehicle connectivity, holographic and mixed reality communication, autonomous and collaborative robots, vehicle platooning and cooperative maneuvering, sensing networks, gesture monitoring, human-brain interfacing, digital twin applications, asset management, and universal coverage applications using non-terrestrial or aerial platforms, among other examples.
[0035] The methods, operations, apparatuses, and techniques described herein may enable one or more of the foregoing technologies or new technologies or support one or more of the foregoing use cases or new use cases.
[0036] Fig. 1 is a diagram illustrating an example of a wireless communication network 100. The wireless communication network 100 may be or may include elements of a 5G network or a 6G network, among other examples. The wireless communication network 100 may include multiple network nodes 110. For example, in Fig. 1, the wireless communication network 100 includes multiple network nodes 110, including a network node 110a and a network node 110b (each of which also may be referred to herein simply as a “network node 110” ) . The network nodes 110 may support communications with multiple UEs 120. For example, in Fig. 1, the network nodes 110 support communication with a UE 120a, a UE 120b, and a UE 120c (each of which also may be referred to herein simply as a “UE 120” ) . In some examples, a UE 120 also may communicate with other UEs 120 and a network node 110 also may communicate with a core network and with other network nodes 110.
[0037] The network nodes 110 and the UEs 120 of the wireless communication network 100 communicate using the electromagnetic spectrum, which may be subdivided into various licensed or unlicensed operating bands, frequency ranges, component carriers, or channels that define associated frequencies available for communications. In some examples, each of the network nodes 110 and the UEs 120 may communicate using one or multiple component carriers in one or more operating bands or ranges. Typically, various operating bands are defined as frequency range designations FR1 (410 MHz through 7.125 GHz) , FR2 (24.25 GHz through 52.6 GHz) , FR3 (7.125 GHz through 24.25 GHz) , FR4a or FR4-1 (52.6 GHz through 71 GHz) , FR4 (52.6 GHz through 114.25 GHz) , and FR5 (114.25 GHz through 300 GHz) . Although a portion of FR1 is greater than 6 GHz, FR1 is often referred to (interchangeably) as a “sub-6 GHz” band in some documents and articles. Similarly, FR2 is often referred to (interchangeably) as a “millimeter wave” band in some documents and articles.
[0038] A network node 110 or a UE 120 may include one or more devices, components, or systems that enable communication with other devices, components, or systems of the wireless communication network 100. For example, a UE 120 and a network node 110 may each include one or more chips, system-on-chips (SoCs) , chipsets, packages, or devices that individually or collectively constitute or comprise a processing system. As shown in Fig. 1, each UE 120 includes a processing system 140 and each network node 110 includes a processing system 145 with communication manager 155. A processing system (for example, the processing system 140 or the processing system 145) includes processor (or “processing” ) circuitry in the form of one or multiple processors, microprocessors, processing units (such as central processing units (CPUs) , graphics processing units (GPUs) , neural processing units (NPUs) (also referred to as neural network processors or deep learning processors (DLPs) ) , or digital signal processors (DSPs) ) , processing blocks, application-specific integrated circuits (ASICs) , programmable logic devices (PLDs) , or other discrete gate or transistor logic or circuitry (any one or more of which may be generally referred to herein individually as a “processor” or collectively as “the processor” or “the processor circuitry” ) . Such processors may be individually or collectively configurable or configured to perform various functions or operations described herein. A group of processors collectively configurable or configured to perform a set of functions may include a first processor configurable or configured to perform a first function of the set and a second processor configurable or configured to perform a second function of the set. In some other examples, each of a group of processors may be configurable or configured to perform a same set of functions.
[0039] The processing system 140 and the processing system 145 may each include memory circuitry in the form of one or multiple memory devices, memory blocks, memory elements, or other discrete gate or transistor logic or circuitry, each of which may include or implement tangible storage media, such as random-access memory, or read-only memory, or combinations thereof (any one or more of which may be generally referred to herein individually as a “memory” or collectively as “the memory” or “the memory circuitry” ) . One or more of the memories may be coupled (for example, operatively coupled, communicatively coupled, electronically coupled, or electrically coupled) with one or more of the processors. One or more of the memories may individually or collectively store processor-executable code or instructions (such as software) (for example, which may be referred to as “one or more code-storing memories” or “code-storing memory circuitry” ) . For example, “code-storing memory” or “code-storing memory circuitry” refers to memory (or memory circuitry) that is configured to store processor-executable code or instructions. The processor-executable code or instructions, when executed by one or more of the processors, may configure one or more of the processors (or processing circuitry) to perform various functions or operations described herein. Additionally, or alternatively, in some examples, one or more of the processors may be configured to perform various functions or operations described herein without requiring configuration by software. Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executables, threads of execution, procedures, or functions, among other examples, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.
[0040] The processing system 140 and the processing system 145 may each include or be coupled with one or more modems (such as a cellular (for example, a 5G or 6G compliant) modem) . In some examples, one or more processors of the processing system 140 or the processing system 145 may include or implement one or more of the modems. The processing system 140 and the processing system 145 also may include or be coupled with multiple radios (collectively “the radio” ) , multiple RF chains, or multiple transceivers, each of which may in turn be coupled with one or more of multiple antennas. In some examples, one or more processors of the processing system 140 or the processing system 145 may include or implement one or more of the radios, RF chains, or transceivers. An RF chain may include one or more filters, mixers, oscillators, amplifiers, analog-to-digital converters (ADCs) , or other devices that convert between an analog signal (such as for transmission or reception via an air interface) and a digital signal (such as for processing by the processing system 140 or by the processing system 145) .
[0041] A network node 110 and a UE 120 may each include one or multiple antennas or antenna arrays. Typical network nodes 110 and UEs 120 may include multiple antennas, which may be organized or structured into one or more antenna panels, one or more antenna groups, one or more sets of antenna elements, or one or more antenna arrays, among other examples. As used herein, the term “antenna” can refer to one or more antennas, one or more antenna panels, one or more antenna groups, one or more sets of antenna elements, or one or more antenna arrays. The term “antenna panel” can refer to a group of antennas (such as antenna elements) arranged in an array or panel, which may facilitate beamforming by manipulating parameters associated with the group of antennas. The term “antenna module” may refer to circuitry including one or more antennas as well as one or more other components (such as filters, amplifiers, or processors) associated with integrating the antenna module into a wireless communication device, such as the network node 110 and the UE 120.
[0042] A network node 110 may be, may include, or also may be referred to as an NR network node, a 5G network node, a 6G network node, a Node B, a gNB, an access point (AP) , a transmission reception point (TRP) , a network entity, a network element, a network equipment, or another type of device, component, or system included in a radio access network (RAN) . In various deployments, a network node 110 may be implemented as a single physical node (for example, a single physical structure) or may be implemented as two or more physical nodes (for example, two or more distinct physical structures) . For example, a network node 110 may be a device or system that implements a part of a radio protocol stack, a device or system that implements a full radio protocol stack (such as a full gNB protocol stack) , or a collection of devices or systems that collectively implement the full radio protocol stack. For example, and as shown, a network node 110 may be an aggregated network node having an aggregated architecture, meaning that the network node 110 may implement a full radio protocol stack that is physically and logically integrated within a single physical structure in the wireless communication network 100. For example, an aggregated network node 110 may include a single standalone base station or a single TRP that operates with a full radio protocol stack to enable or facilitate communication between a UE 120 and a core network of the wireless communication network 100.
[0043] Alternatively, a network node 110 may be a disaggregated network node 110 (sometimes referred to as a disaggregated base station) , having a disaggregated architecture, meaning that the network node 110 may operate with a radio protocol stack that is physically distributed or logically distributed among two or more nodes in the same geographic location or in different geographic locations. In some deployments, disaggregated network nodes 110 may be used in an integrated access and backhaul (IAB) network, in an open radio access network (O-RAN) (such as a network configuration in compliance with the O-RAN Alliance) , or in a virtualized radio access network (vRAN) , also known as a cloud radio access network (C-RAN) , to facilitate scaling by separating network functionality into multiple units or modules that can be individually deployed.
[0044] The disaggregated network nodes 110 of the wireless communication network 100 may include one or more central units (CUs) , one or more distributed units (DUs) , and one or more radio units (RUs) . A CU may host one or more higher layers, such as a radio resource control (RRC) layer, a packet data convergence protocol (PDCP) layer, and a service data adaptation protocol (SDAP) layer, among other examples. A CU can communicate with a core network either directly (for example, via a backhaul link) or indirectly (for example, via one or more disaggregated control units, such as a non-real-time (Non-RT) RAN intelligent controller (RIC) associated with a Service Management and Orchestration (SMO) framework or a near-real-time (Near-RT) RIC) . A DU may host one or more of a radio link control (RLC) layer, a medium access control (MAC) layer, or one or more higher physical (PHY) layers depending, at least in part, on a functional split, such as a functional split defined by the 3GPP. In some examples, a DU also may host a lower PHY layer that is configured to perform functions, such as a fast Fourier transform (FFT) , an inverse FFT (IFFT) , beamforming, or physical random access channel (PRACH) extraction and filtering, among other examples. An RU may perform RF processing functions or lower PHY layer functions, such as an FFT, an IFFT, beamforming, or PRACH extraction and filtering, among other examples, according to a functional split, such as a lower layer split (LLS) . In such an architecture, each RU can be operated to handle over the air (OTA) communication with one or more UEs 120. A CU may communicate with one or more DUs via respective midhaul links, such as via F1 interfaces. Each of the DUs may communicate with one or more RUs via respective fronthaul links. Each of the RUs may communicate with one or more UEs 120 via respective RF access links. In some deployments, a UE 120 may be simultaneously served by multiple RUs.
[0045] In some examples, a single network node 110 may include a combination of one or more CUs, one or more DUs, or one or more RUs. In some examples, a CU, a DU, or an RU may be implemented as a virtual unit, such as a virtual central unit (VCU) , a virtual distributed unit (VDU) , or a virtual radio unit (VRU) , among other examples, which may be implemented as a virtual network function, such as in a cloud deployment (for example, an open cloud (O-Cloud) platform) . An SMO framework may support RAN deployment and provisioning of non-virtualized and virtualized network elements.
[0046] In some examples, the wireless communication network 100 may be a heterogeneous network that includes network nodes 110 of various types. Different types of network nodes 110 may generally operate on the same or different operating bands, transmit at different power levels, or serve different coverage areas, each of which may be referred to as or associated with a particular cell 130 (for example, a cell 130a and a cell 130b) .
[0047] The UEs 120 may be physically dispersed throughout the coverage area of the wireless communication network 100, and each UE 120 may be stationary or mobile. A UE 120 may be, may include, or also may be referred to as an access terminal, a mobile station, a client device, or a subscriber unit. A UE 120 may be, include, or be coupled with a cellular phone (for example, a smart phone) , a personal digital assistant (PDA) , a wireless modem, a wireless communication device, a handheld device, a laptop computer, a cordless phone, a wireless local loop (WLL) station, a tablet, a camera, a netbook, a smartbook, an ultrabook, a medical device, a biometric device, a wearable device (for example, a smart watch, smart clothing, smart glasses, a smart wristband, or smart jewelry) , a gaming device, an entertainment device (for example, a music device, a video device, or a satellite radio) , an XR device, a vehicular component or sensor, a smart meter or sensor, industrial manufacturing equipment, a Global Navigation Satellite System (GNSS) device (such as a Global Positioning System device or another type of positioning device) , an artificially intelligent robot or other device implementing artificial intelligence, a UE function of a network node, or any other suitable device or function that may communicate in the wireless communication network 100.
[0048] Some UEs 120 may be classified according to different categories in association with different complexities or different capabilities. UEs 120 in a first category may be associated with relatively low complexity or cost such as NB-IoT devices or eMTC UEs. UEs 120 in a second category may include higher complexity or cost devices, such as mission-critical IoT devices, baseline UEs, high-tier UEs, advanced UEs, full-capability UEs, or premium UEs that are capable of URLLC, eMBB, or precise positioning in the wireless communication network 100. A third category of UEs 120 may have mid-tier complexity or capabilities (for example, capabilities between that of the UEs 120 of the first category and the UEs 120 of the second category) . A UE 120 of the third category may be referred to as a reduced capability UE ( “RedCap UE” ) , a mid-tier UE, an NR-Light UE, or an NR-Lite UE, among other examples.
[0049] In some examples, a network node 110 may be, may include, or may operate as an RU, a TRP, or a base station that communicates with one or more UEs 120 via a radio access link (which may be referred to as a “Uu” link) . The radio access link may include a downlink and an uplink. “Downlink” (or “DL” ) refers to a communication direction from a network node 110 to a UE 120, and “uplink” (or “UL” ) refers to a communication direction from a UE 120 to a network node 110. Downlink and uplink resources may include time domain resources (for example, frames, subframes, slots, and symbols) , frequency domain resources (for example, frequency bands, component carriers (CCs) , subcarriers, resource blocks, and resource elements) , and spatial domain resources (for example, particular transmit directions or beams) .
[0050] Frequency domain resources may be subdivided into bandwidth parts (BWPs) . A BWP may be a block of frequency domain resources (for example, a continuous set of resource blocks (RBs) within a full component carrier bandwidth) that may be configured at a UE-specific level. A UE 120 may be configured with both an uplink BWP and a downlink BWP (which may be the same or different) . Each BWP may be associated with its own numerology (indicating a sub-carrier spacing (SCS) and cyclic prefix (CP) ) . A BWP may be dynamically configured or activated (for example, by a network node 110 transmitting a downlink control information (DCI) configuration to the one or more UEs 120) or reconfigured (for example, in real-time or near-real-time) according to changing network conditions in the wireless communication network 100 or specific requirements of one or more UEs 120. An active BWP defines the operating bandwidth of the UE 120 within the operating bandwidth of the serving cell.
[0051] As used herein, a downlink signal may be or include a reference signal, control information, or data. For example, downlink reference signals include a primary synchronization signal (PSS) , a secondary SS (SSS) , an SS block (SSB) (for example, that includes a PSS, an SSS, and a physical broadcast channel (PBCH) ) , a demodulation reference signal (DMRS) , a phase tracking reference signal (PTRS) , a tracking reference signal (TRS) , and a CSI-RS, among other examples. A downlink signal carrying control information or data may be transmitted via a downlink channel. Downlink channels may include one or more control channels for transmitting control information and one or more data channels for transmitting data. Downlink reference signals may be transmitted in addition to, or multiplexed with, downlink control channel communications or downlink data channel communications. A downlink control channel may be specifically used to transmit DCI from a network node 110 to a UE 120. DCI generally contains the information the UE 120 needs to identify RBs in a subsequent subframe and how to decode them, including a modulation and coding scheme (MCS) or redundancy version parameters. Different DCI formats carry different information, such as scheduling information in the form of downlink or uplink grants, slot format indicators (SFIs) , preemption indicators (PIs) , transmit power control (TPC) commands, hybrid automatic repeat request (HARQ) information, new data indicators (NDIs) , among other examples. A downlink data channel may be used to transmit downlink data (for example, user data associated with a UE 120) from a network node 110 to a UE 120. Downlink control channels may include physical downlink control channels (PDCCHs) , and downlink data channels may include physical downlink shared channels (PDSCHs) . Control information or data communications may be transmitted on a PDCCH and PDSCH, respectively. For example, a PDCCH can carry DCI, while a PDSCH can carry a MAC control element (MAC-CE) , an RRC message, or user data, among other examples. Each PDSCH may carry one or more transport blocks (TBs) of data.
[0052] As used herein, an uplink signal may include a reference signal, control information, or data. For example, uplink reference signals include a sounding reference signal (SRS) , a PTRS, and a DMRS, among other examples. An uplink signal carrying control information or data may be transmitted via an uplink channel. An uplink channel may include one or more control channels for transmitting control information and one or more data channels for transmitting data. Uplink reference signals may be transmitted in addition to, or multiplexed with, uplink control channel communications or uplink data channel communications. An uplink control channel may be specifically used to transmit uplink control information (UCI) from a UE 120 to a network node 110. An uplink data channel may be used to transmit uplink data (for example, user data associated with a UE 120) from a UE 120 to a network node 110. Uplink control channels may include physical uplink control channels (PUCCHs) , and uplink data channels may include physical uplink shared channels (PUSCHs) . Control information or data communications may be transmitted on a PUCCH and PUSCH, respectively. For example, a PUCCH can carry UCI, while a PUSCH can carry a MAC-CE, an RRC message, or user data, among other examples. UCI can include a scheduling request (SR) , HARQ feedback information (for example, a HARQ acknowledgement (ACK) indication or a HARQ negative acknowledgement (NACK) indication) , uplink power control information (for example, an uplink TPC parameter) , or CSI, among other examples. CSI can include a channel quality indicator (CQI) (indicative of downlink channel conditions to facilitate selection of transmission parameters, such as an MCS, by a network node 110) , a PMI, a CSI-RS resource indicator (CRI) (for example, indicative of a beam used to transmit a CSI-RS) , an SS / PBCH resource block indicator (SSBRI) (for example, indicative of a beam used to transmit an SSB) , a layer indicator (LI) , a rank indicator (RI) , or measurement information (for example, a layer 1 (L1) -reference signal received power (RSRP) parameter, a received signal strength indicator (RSSI) parameter, a reference signal received quality (RSRQ) parameter, among other examples) which can be used for beam management, among other examples. Each PUSCH may carry one or more TBs of data.
[0053] The information (for example, data, control information, or reference signal information) transmitted by a network node 110 to a UE 120, or vice versa, may be represented as a sequence of binary bits that are mapped (for example, modulated) to an analog signal waveform (for example, a discrete Fourier transform (DFT) -spread-orthogonal frequency division multiplexing (OFDM) (DFT-s-OFDM) waveform or a CP-OFDM waveform) that is transmitted by the network node 110 or UE 120 over a wireless communication channel. In some examples, the network node 110 or the UE 120 (for example, using the processing system 145 or the processing system 140, respectively) may select an MCS (for example, an order of quadrature amplitude modulation (QAM) , such as 64-QAM, 128-QAM, or 256-QAM, among other examples) for a downlink signal or an uplink signal. For example, the network node 110 may select an MCS for a downlink signal in accordance with UCI received from the UE 120 or may transmit, to the UE 120, an indication of an MCS to be applied for an uplink signal.
[0054] A network node 110 or a UE 120 (such as by using the processing system 145 or the processing system 140, respectively, or one or more coupled modems) may perform signal processing on the information (such as filtering, amplification, modulation, digital-to-analog conversion, an IFFT operation, multiplexing, interleaving, mapping, or encoding, among other examples) to generate a processed signal in accordance with the selected MCS. In some examples, the network node 110 or the UE 120 (for example, using the processing system 145 or the processing system 140, respectively, or one or more coupled encoders or modems) may perform a channel coding operation or a forward error correction (FEC) operation to control errors in transmitted information. For example, the network node 110 or the UE 120 may perform an encoding operation to generate encoded information (such as by selectively introducing redundancy into the information, typically using an error correction code (ECC) , such as a polar code or a low-density parity-check (LDPC) code) . The network node 110 or the UE 120 (for example, using the processing system 145 or one or more modems) may further perform spatial processing (for example, precoding) on the encoded information to generate one or more processed or precoded signals for downlink or uplink transmission, respectively. In some examples, the network node 110a or the UE 120a may perform codebook-based precoding or non-codebook-based precoding. Codebook-based precoding may involve selecting a precoder (for example, a precoding matrix) using a codebook. For example, the network node 110a may provide precoding information indicating which precoder, defined by the codebook, is to be used by the UE 120a. Non-codebook-based precoding may involve selecting or deriving a precoder based on, or otherwise associated with, one or more downlink or uplink signal measurements. The network node 110a or the UE 120a may transmit the processed downlink or uplink signals, respectively, via one or more antennas.
[0055] The network node 110a or the UE 120a may receive uplink signals or downlink signals, respectively, via one or more antennas. The network node 110a or the UE 120a (for example, using the processing system 145 or the processing system 140, respectively, or one or more coupled modems) may perform signal processing (for example, in accordance with the MCS) on the received uplink or downlink signals, respectively (such as filtering, amplification, demodulation, analog-to-digital conversion, an FFT operation, demultiplexing, deinterleaving, de-mapping, equalization, interference cancellation, or decoding, among other examples) , to map the received signal (s) to a sequence of binary bits (for example, received information) that estimates the information transmitted by the network node 110 or the UE 120 via the downlink or uplink signals. The network node 110a or the UE 120a (for example, using the processing system 145 or the processing system 140, respectively, or a coupled decoder or one or more modems) may decode the received information (such as by using an ECC, a decoding operation, or an FEC operation) to detect errors or correct bit errors in the received information to generate decoded information. The decoded information may estimate the information transmitted via the downlink or uplink signals.
[0056] In some examples, a UE 120 and a network node 110 may perform MIMO communication. MIMO communication generally refers to transmitting or receiving multiple signals (such as multiple layers or multiple data streams) simultaneously over the same time and frequency resources. A network node 110 or a UE 120 may communicate using single-user MIMO or multi-user MIMO (MU-MIMO) , the latter of which being used by a network node 110 to simultaneously transmit signals to multiple UEs 120. MIMO techniques may involve spatial multiplexing (multi-layer transmission) or beamforming. To implement beamforming, the amplitudes or phases of signals transmitted via antenna elements may be modulated and shifted relative to each other (such as by manipulating a phase shift, a phase offset, or an amplitude) to generate one or more beams. For example, a network node 110 may generate one or more beams 160a, and a UE 120 may generate one or more beams 160b. The term “beam” may refer to a directional transmission of a wireless signal toward a receiving device or otherwise in a desired direction, a directional reception of a wireless signal from a transmitting device or otherwise in a desired direction, a direction associated with such a directional transmission or directional reception, a set of directional resources associated with a signal transmission or signal reception (for example, an angle of arrival, a horizontal direction, or a vertical direction) , or a set of parameters or resources associated with one or more aspects of a directional signal, among other examples.
[0057] In some examples, a network node 110 or a UE 120 may implement massive MIMO, which may be associated with an increased (for example, “massive” ) quantity of antennas at the network node 110 or at the UE 120, such as in a network implementing mmWave technology, which enables more precise beamforming or reduced interference. In some examples, the wireless communication network 100 may implement multi-TRP (mTRP) operation (including redundant transmission or reception on multiple TRPs) or non-coherent joint transmission (NC-JT) .
[0058] The network node 110 and the UE 120 may establish a communication link or beam pair, and otherwise increase reliability, throughput, signal strength, or other signal properties for MIMO communications, by performing beam management operations, such as an initial beam acquisition operation, a beam refinement operation, or a beam recovery operation. For example, an initial beam acquisition operation may involve the network node 110 transmitting signals (for example, SSBs or other signals) via respective beams (for example, of the beams 160 of the network node 110) and the UE 120 receiving and measuring the signal (s) via respective beams of multiple beams (for example, from the beams 160 of the UE 120) to identify a best beam (or beam pair) for communication between the UE 120 and the network node 110. A beam refinement operation may involve a first device (for example, the UE 120 or the network node 110) transmitting signal (s) via a subset of beams (for example, identified based on, or otherwise associated with, measurements reported as part of one or more other beam management operations) . A second device (for example, the network node 110 or the UE 120) may receive the signal (s) via a single beam (for example, to identify the best beam for communication from the subset of beams) . The beam (s) may be identified or defined via one or more spatial parameters, such as a transmission configuration indicator (TCI) state or a quasi co-location (QCL) parameter, among other examples.
[0059] Some aspects and techniques as described herein may be implemented, at least in part, using an artificial intelligence (AI) program (for example, referred to herein as an “AI / ML model” ) , such as a program that includes a machine learning (ML) model or an artificial neural network (ANN) model. The AI / ML model may be deployed at one or more devices 165 (for example, one or more network nodes 110, one or more UEs 120, one or more servers, or one or more components of a cloud computing network, among other examples) . For example, in a deployment in which AI / ML functionality is performed independently at a device 165, sometimes referred to as “overlay AI / ML, ” the AI / ML model (or an instance or portion of the AI / ML model) may be deployed at a UE 120 (for example, by the processing system 140) , a network node 110 (for example, by the processing system 145) , one or more servers, or one or more components of a cloud computing network, among other examples. Additionally, or alternatively, in a deployment where AI / ML functionality is coordinated between different devices 165, sometimes referred to as “coordinated AI / ML, ” or performed at all device and network layers, sometimes referred to as “native AI / ML, ” the AI / ML model (or an instance of the AI / ML model) may be deployed at multiple devices 165 (for example, a first portion of the AI / ML model may be deployed at a UE 120 and a second portion of the AI / ML model may be deployed at a network node 110) . In other examples of coordinated AI / ML or native AI / ML, a first AI / ML model may be deployed at a UE 120 and a second AI / ML model may be deployed at a network node 110. The AI / ML model (s) may be configured to enhance various aspects of the wireless communication network 100 (for example, to increase privacy, reliability, or efficient use of network bandwidth, or to reduce latency, among other examples) . For example, the AI / ML model (s) may be trained to identify patterns or relationships in data corresponding to the wireless communication network 100, a device, or an air interface, among other examples. The AI / ML model (s) may support operational decisions relating to one or more aspects associated with wireless communications devices, networks, or services.
[0060] Accordingly, in some examples, the AI / ML model (s) may enable AI-as-a-Service (for example, an end-to-end AI / ML service via a user plane) for use cases, such as a self-organizing network (SON) , minimization of drive test (MDT) , quality of experience (QoE) , positioning, sensing, predictive mobility, or traffic prediction, among other examples. In some examples, AI-as-a-Service use cases may include measurement collection reporting by a UE 120, device selection criteria (for example, according to a geographical area where measurements are to be collected or UE capabilities to be used to collected measurements) , or reporting configurations (for example, reporting parameters such as location, time, or sensor information, among other examples) . Additionally, or alternatively, the AI / ML model (s) may enable AI / ML procedures (for example, RAN-triggered service establishment, configuration, inferencing using UE-side or network-side models, performance monitoring or management, or capability signaling, among other examples) . Additionally, or alternatively, the AI / ML model (s) may enable RAN-based AI / ML services via one or more application program interfaces (APIs) or management interfaces for use cases, such as beam management, radio resource monitoring (RRM) relaxation, mobility prediction, load prediction, network energy savings, or coverage and capacity improvements, among other examples.
[0061] In some aspects, the UE 120 may include a communication manager 150. As described in more detail elsewhere herein, the communication manager 150 may predict future CSI using historical CSI data and an ML model; determine an SGCS metric by comparing the predicted CSI with actual ground-truth CSI within a configurable monitoring window associated with an extended CPU duration; and transmit an indication of the SGCS metric in an SGCS report to a network entity. Additionally, or alternatively, the communication manager 150 may perform one or more other operations described herein.
[0062] The network node 110, the processing system 145 of the network node 110, the UE 120, the processing system 140 of the UE 120, or any other component (s) of Fig. 1 may implement one or more techniques or perform one or more operations associated with timing for CSI prediction and monitoring, as described in more detail elsewhere herein. For example, the processing system 145 of the network node 110, or the processing system 140 of the UE 120 may perform or direct operations of, for example, process 1000 of Fig. 10, or other processes as described herein (alone or in conjunction with one or more other processors) . Memory of the network node 110 may store data and program code (or instructions) for the network node 110. In some examples, the memory of the network node 110 may store data relating to a UE 120, such as RRC state information or a UE context. Memory of a UE 120 may store data and program code (or instructions) for the UE 120, such as context information. In some examples, the memory of the UE 120 or the memory of the network node 110 may include a non-transitory computer-readable medium storing a set of instructions for wireless communication. For example, the set of instructions, when executed by one or more processors (for example, of the processing system 145 or the processing system 140) of the network node 110, or the UE 120, may cause the one or more processors to perform process 1000 of Fig. 10, or other processes as described herein. In some examples, executing instructions may include running the instructions, converting the instructions, compiling the instructions, or interpreting the instructions, among other examples.
[0063] In some aspects, a UE includes means for predicting future CSI using historical CSI data and an ML model; means for determining an SGCS metric by comparing the predicted CSI with actual ground-truth CSI within a configurable monitoring window associated with an extended CPU duration; or means for transmitting an indication of the SGCS metric in an SGCS report to a network entity. The means for the UE to perform operations described herein may include, for example, one or more of communication manager 150, processing system 140, a radio, one or more RF chains, one or more transceivers, one or more antennas, one or more modems, a reception component (for example, reception component 1102 depicted and described in connection with Fig. 11) , or a transmission component (for example, transmission component 1104 depicted and described in connection with Fig. 11) , among other examples.
[0064] Fig. 2 is a diagram illustrating an example 200 of CSI prediction. Example 200 shows a UE-side model that may predict future CSI. For example, the UE-side model may utilize historical CSI data 204 to perform CSI prediction 202 using an AI / ML model 206. In some aspects, instead of a UE-side model predicting future channel state information, a network-side model may be employed to perform the prediction, using similar historical CSI data. Such a network-side model can leverage centralized computing resources to enhance prediction accuracy and efficiency across multiple UE devices. Additionally, or alternatively, the UE side model 200 may utilize not only historical CSI data 204 but also real-time CSI measurements to perform CSI prediction 202 using an AI / ML model 206, thereby increasing the accuracy of predictions. Real-time measurements allow the model to adjust to immediate changes in channel conditions, providing more precise prediction results. Additionally, or alternatively, in addition to utilizing historical CSI data 204, external data sources such as weather data or user mobility patterns may be incorporated into the AI / ML model 206 to enhance prediction capabilities. These external data sources can influence wireless communication and provide additional context for more informed predictions.
[0065] In example 200, every 4 CSI-RS occasions of the 1st CSI-RS transmission are used to the predict one or more CSIs in the future, such as CSI-RS on slot k, k-5, k-10, k-15 being used to predict the CSI on slot k+4. The CSI can be channel matrix H or precoding vectors V.
[0066] The AI / ML model 206 processes this historical data to predict future CSI 208, such as h (k+4) , where h refers to the channel matrix and k refers to the slot or occasion. In some aspects, the AI / ML model 206 may employ different types of models, such as neural networks or decision trees, rather than a single model type, to process the historical data for predicting future CSI 208. Different model types may offer varying strengths, such as handling complex data relationships in neural networks or performing quick classification tasks with decision trees. Additionally, or alternatively, predicted CSI 208 may be used not only for optimizing network operations like beamforming and resource allocation but also for adjusting transmission power levels or handover decisions in a mobile network. Adjusting transmission power levels may enhance signal quality, while optimized handover decisions may improve connectivity and reduce dropped calls. Additionally, or alternatively, the historical CSI data 204 may include additional metrics such as signal-to-noise ratio (SNR) or interference measurements to enrich the dataset for more comprehensive predictions. Including these metrics can offer deeper insights into channel conditions, aiding in more accurate CSI prediction.
[0067] Predicted CSI 208 is then utilized for optimizing network operations such as beamforming and resource allocation. In some aspects, the prediction of future CSI, such as h (k+4) , may also be applied to predict other network parameters like throughput or latency, offering broader insights into network performance. Predicting throughput and latency can guide network settings to enhance user experience and resource management. Additionally, or alternatively, instead of predicting a single future time instance (e.g., h (k+4) ) , the UE-side model may predict a range of future instances, providing a forecast over a longer duration for better planning of network resources. Such a forecast can assist network operators in proactively adjusting resources for anticipated demand fluctuations. Additionally, or alternatively, the AI / ML model 206 may include mechanisms for continual learning, allowing it to adapt to new patterns in CSI data as they emerge over time. Continual learning ensures that the model remains accurate and relevant, reflecting the dynamic nature of wireless environments. Additionally, or alternatively, an alternative approach may involve using a distributed AI / ML model, where multiple models collaborate to predict future CSI, leveraging different aspects of the historical data for a more robust prediction. Collaborating models can capitalize on diverse data aspects, enhancing prediction reliability and robustness.
[0068] Fig. 3 is a diagram illustrating an example 300 of SGCS reporting. Example 300 shows a UE 320 (e.g., UE 120) that may initiate inference triggering for CSI prediction. As shown by reference number 325, the network entity 310 may transmit a request for a CSI report of a predicted CSI using AI / ML. In some aspects, inference triggering may include the UE 320 receiving a pre-configured signal from the network (e.g., network entity 310) to start the prediction process for CSI. Such a signal may be configured to optimize the timing of prediction initiation based on network conditions or priorities. Additionally, or alternatively, inference triggering may include the UE 320 employing a statistical prediction model based on time-series analysis of historical CSI data to anticipate future network conditions. Time-series analysis can enhance prediction accuracy by leveraging temporal patterns in the historical data. Additionally, or alternatively, inference triggering may include the prediction focusing on estimating SNR or other relevant network performance metrics for future time intervals. This approach can provide insights into expected communication quality and reliability.
[0069] For example, the UE 320 may start the process by utilizing historical CSI data to perform prediction using an AI / ML model. This prediction involves using previous CSI measurements, the measurements of the CSI using the measurement resource configured in the inference report, or a combination thereof, to forecast CSI for future slots. Example 300 shows measurement resources 312 for the inference report. The predicted CSI 316 may include channel matrices or precoding vectors, offering insights into future network conditions.
[0070] The UE 320 may generate an inference CSI report 304 at slot n1. In some aspects, the inference CSI report 304 at slot n1 may include additional data, such as confidence intervals or prediction uncertainty measures. These additional metrics can provide the network with more detailed insights into expected channel conditions and help in decision-making. Additionally, or alternatively, the inference CSI report 304 may be stored locally at the UE for later synchronization with the network, depending on network congestion or priority settings. This ensures efficient data management and transmission based on current network status.
[0071] For example, the inference CSI report 304 may include the predicted CSI 316, which is subsequently transmitted to the network entity 310. This report enables the network to understand expected channel conditions and optimize operations accordingly, such as resource allocation and beamforming.
[0072] The UE 320 may receive a monitoring trigger 306 from the network entity 310. In some aspects, receiving the monitoring trigger 306 may include the UE 320 initiating a different type of performance monitoring, such as evaluating the impact of predicted interference patterns on network operations. This alternative monitoring can provide insights into how interference affects network performance and guide adjustments. Additionally, or alternatively, receiving the monitoring trigger may include receiving alternative metrics, such as mean squared error (MSE) or cross-correlation, instead of the SGCS. These metrics offer different perspectives on prediction reliability and can be tailored to specific network requirements.
[0073] For example, the monitoring trigger 306 prompts the UE 320 to perform performance monitoring by comparing the predicted CSI 316 with an actual measured ground-truth 314. As shown by reference number 330, the network entity 310 may transmit CSI-RS for channel and interference measurements. As shown by reference number 335, the UE 320 may perform CSI measurement, CSI prediction, and PMI calculation. The UE 320 may later compare the predicted CSI 316 with the ground-truth CSI 314. This comparison involves evaluating the accuracy of the prediction model using metrics like SGCS, which quantifies prediction reliability.
[0074] The UE 320 may transmit a CSI report, as shown by reference number 340. As shown by reference number 345, the network entity 310 may transmit a request for a monitoring report. As shown by reference number 350, the network entity 310 may transmit a CSI-RS for ground-truth CSI measurement.
[0075] The UE may then perform a ground-truth measurement and PMI calculation, as shown by reference number 355. In some aspects, ground-truth measurement and PMI calculation may include additional steps, such as filtering out noise or outliers in the measured data to enhance the accuracy of the prediction assessment. Such filtering can improve the precision of measurement results and ensure reliable comparison with predicted values.
[0076] For example, the UE 320 may measure the current channel state and calculate the PMIs to assess the ground-truth 314 against the predicted CSI 316. This step is important for verifying prediction accuracy and may involve additional data processing to ensure precise measurement results.
[0077] As shown by reference number 360, the UE 320 may then transmit an SGCS report 308 or monitoring output at slot n2. In some aspects, instead of transmitting an SGCS report 308 at slot n2, the UE 320 may transmit a comprehensive performance report that includes both prediction accuracy and recommendations for network adjustments. This comprehensive report can guide network entities in implementing informed strategies for optimization. Additionally, or alternatively, the SGCS report 308 may be sent to multiple network entities, such as base stations and network controllers. This approach facilitates a coordinated effort in network management and optimization.
[0078] For example, the SGCS report 308 conveys the calculated metrics to the network entity 310, enabling the network entity 310 to evaluate prediction accuracy and make informed decisions regarding network configuration and optimization. This reporting ensures continuous monitoring and adjustment of network operations, facilitating enhanced communication quality and resource efficiency.
[0079] In a first alternative, a single SGCS metric is calculated based on the predicted CSI 316 for one inference reporting and the corresponding ground-truth CSI 314. This approach focuses on evaluating the accuracy of the predicted CSI 316 by directly comparing it with the actual measured CSI, providing a clear and straightforward assessment of the prediction model’s performance.
[0080] A second alternative involves a more comprehensive approach by calculating two SGCS metrics. The first SGCS is based on the predicted CSI 316 for one inference reporting and the ground-truth CSI 314, similar to the first alternative. The second SGCS, however, is calculated by comparing the ground-truth CSI 314 with the non-predicted CSI corresponding to the latest CSI-RS transmission occasion, which occurs no later than the CSI reference resource of the inference reporting instance. This dual comparison offers a deeper insight into the prediction model’s accuracy by evaluating both predicted and non-predicted CSI against the ground truth.
[0081] In a third alternative, the focus shifts to statistical analysis of the reporting contents, such as calculating the mean or a specific percentile of the cumulative distribution function (CDF) of SGCS values across multiple inference reporting instances within a configured monitoring window. This approach allows for a more aggregated view of prediction performance over time. The monitoring window can be configured by the network, with detailed signaling to determine whether to report per prediction instance, selected instances, or averaged over instances. Additionally, the report frequency can be adjusted for granularity, such as per wideband, per subband, or averaged over selected subbands. Reporting can also be configured per layer, for the first layer, or averaged over layers, with mechanisms in place to quantize and report the data effectively.
[0082] Fig. 4 is a diagram illustrating an example 400 of buffering times for calculating SGCS. Example 400 shows a network entity 310 that may initiate inference triggering for CSI prediction.
[0083] In wireless communication systems, the inference report and the monitoring report serve distinct purposes, necessitating separate handling by the UE. After generating an inference report, the UE may need to buffer both measurement results and inference results to accurately calculate the SGCS with respect to the ground truth when the monitoring report is triggered. This buffering ensures that the UE retains necessary data for comparison and validation, facilitating a comprehensive assessment of prediction accuracy. The process of buffering is important for maintaining the integrity of the prediction model and ensuring reliable performance monitoring. Example 400 shows a buffer time 402.
[0084] For alternatives one and three, the UE focuses on buffering the predicted precoding matrix indicator (PMI) . This approach allows the UE to retain essential prediction data for subsequent analysis and comparison with ground-truth measurements. By preserving the predicted PMI, the UE can efficiently calculate the SGCS and evaluate the accuracy of the prediction model. In contrast, alternative two requires the UE to buffer both the predicted PMI and the measured PMI. This dual buffering strategy provides a more detailed dataset for performance monitoring, enabling a thorough comparison between predicted and actual channel states. The additional data enhances the reliability of the SGCS calculation and offers deeper insights into prediction accuracy.
[0085] The need for additional buffering time arises from the requirement to retain both predicted and measured data until the monitoring report is triggered. This extended buffering period, shown by additional buffering time 404, ensures that the UE has access to all necessary information for accurate performance assessment. However, this requirement introduces challenges related to memory management and processing efficiency. The UE must allocate sufficient resources to accommodate the increased data volume, which can strain its capabilities and impact overall performance.
[0086] To address these challenges, it is important for the UE to determine when buffering is necessary. If there is a monitoring request, the UE must engage in buffering to prepare for the subsequent performance evaluation. However, if no monitoring request is present, buffering may be unnecessary, allowing the UE to conserve resources. Therefore, it is advantageous for the UE to have a mechanism to ascertain whether buffering is required, optimizing resource allocation and enhancing efficiency.
[0087] When buffering is needed, it is important to enhance the CSI processing criteria to reflect the additional buffering time and quantity. This enhancement ensures that the UE can handle the increased data load effectively, maintaining processing efficiency and accuracy. By adapting the CSI processing criteria, the UE can better manage memory resources and processing capabilities, ensuring reliable performance monitoring and prediction validation.
[0088] Fig. 5 is a diagram illustrating an example 500 of CPU durations. Example 500 shows a legacy CPU that manages the computational resources required for processing CSI reports. For periodic and semi-persistent CSI (excluding the initial report of semi-persistent CSI on the PUSCH) , the CPU is engaged from the first symbol of the earliest CSI-RS, CSI interference measurement (CSI-IM) , or synchronization signal block (SSB) resource used for channel or interference measurement. This engagement continues through the respective latest CSI-RS / CSI-IM / SSB occasion, which occurs no later than the corresponding CSI reference resource, until the last symbol of the configured PUSCH or PUCCH carrying the report. This structured approach ensures that the CPU is efficiently utilized for processing CSI data, maintaining the integrity and accuracy of the reports.
[0089] For aperiodic CSI reports, the CPU is occupied from the first symbol following the PDCCH that triggers the CSI report, extending until the last symbol of the scheduled PUSCH carrying the report. This allocation allows for dynamic processing of CSI data, accommodating the variability inherent in aperiodic reporting. Similarly, an initial semi-persistent CSI report on PUSCH after the PDCCH trigger occupies the CPU from the first symbol after the PDCCH, continuing until the last symbol of the scheduled PUSCH carrying the report. These mechanisms ensure that the CPU is effectively engaged in processing CSI data, adapting to the specific requirements of each reporting type. Example 500 shows a legacy CPU duration 502.
[0090] One solution is to extend the ending time until the reporting of the monitoring report addresses the need for prolonged CPU engagement. Example 500 shows an extended CPU duration 504. The extended CPU duration 504 may accommodate additional buffering time for prediction results. In some aspects, the extended CPU duration 504 may include accommodating additional computational tasks, such as recalibrating the prediction model or adjusting prediction intervals based on recent data trends. These tasks can enhance the accuracy and reliability of predictions by ensuring that models are up-to-date and responsive to changes in network conditions.
[0091] By extending the CPU’s occupation period, the solution accommodates additional processing requirements associated with monitoring reports, ensuring that the CPU can handle increased data loads and maintain processing efficiency. This extension helps support comprehensive performance monitoring and prediction validation, facilitating enhanced network operations.
[0092] In parallel, legacy active CSI-RS resource and port counting mechanisms may be employed to manage resource allocation. For periodic and semi-persistent CSI-RS, the active resource is counted from its radio resource control (RRC) configuration, MAC-CE, or DCI activation, continuing until its RRC reconfiguration, MAC-CE deactivation, or DCI deactivation. This counting ensures that resources are effectively allocated for processing CSI data, optimizing network performance.
[0093] For aperiodic CSI-RS, the active resource is counted from its DCI triggering until the PUSCH that carries the aperiodic CSI report. This dynamic counting mechanism accommodates the variability of aperiodic reporting, ensuring that resources are efficiently utilized. The solution to extend the ending time until the reporting of the monitoring report further enhances resource management, allowing for prolonged engagement of active resources to support comprehensive monitoring and performance evaluation. This extension helps maintain resource efficiency and optimizing network operations, particularly in dynamic and high-mobility environments.
[0094] Fig. 6 is a diagram illustrating examples 600 and 602 of restrictions on a timing for monitoring.
[0095] According to various aspects described herein, a UE may utilize a timing for ensuring that there is enough processing time for an SGCS report while not wasting time and CPUs. In some aspects, a UE (e.g., UE 320) may anticipate receiving an aperiodic monitoring report trigger, which can occur under several specified conditions. A first option may involve receiving the monitoring trigger 306 within a defined period (marked by a deadline for monitoring triggering 606) , such as the next available downlink slot, following the inference report trigger 302. This timing ensures that the monitoring process is closely aligned with the inference CSI report 304, facilitating timely evaluation of prediction accuracy.
[0096] A second option may specify that the monitoring trigger 306 is to be received no later than the first symbol of the inference report measurement resource 312, which is used to measure the historical CSI as the input to the ML model, ensuring that monitoring begins promptly as the measurement process is initiated.
[0097] When the monitoring trigger occurs as early as the inference trigger, the UE 320 is aware that the measurement and computation results from the inference CSI report 304 are to be buffered for use in the SGCS calculation for the monitoring report. Specifically, the UE 320 is expected to receive the monitoring trigger 306 in the next available downlink slot following the inference report trigger 302, or no later than the measurement resource 312 configured for model input in the inference CSI report 304. Consequently, the CPU and active resource / port counting are extended until the completion of the monitoring report. If the monitoring trigger 306 is not received within the defined time period (e.g., the next available downlink slot, no later than the first symbol of the measurement resource for model input configured in the inference report, or no later than the inference report) , the UE 320 does not anticipate a monitoring report and will discard the measurement and computation results after the inference report. These restrictions prevent the UE 320 from having to buffer results indefinitely without knowing when to discard them, ensuring efficient resource management.
[0098] A third option may allow for the monitoring trigger 306 to be received no later than the inference CSI report 304, providing flexibility in timing while still maintaining a connection to the inference process. A fourth option may involve setting an artificial deadline by specification, which can be used to standardize the timing of monitoring triggers across different network configurations.
[0099] Example 600 is used to show the first three options. Example 602 is used to show the fourth option.
[0100] For each of these four options, if the trigger is received according to the option’s rule, the UE 320 may perform the monitoring and the CPU occupation and active resource port counting for the inference resource (for historical CSI) extended to the end of the monitoring report. Otherwise, if the UE 320 is not receiving monitoring trigger with the option’s rule, the UE 320 is not expected to perform the monitoring report.
[0101] In scenarios where the monitoring trigger is received, the active resource and port counting, as well as the CPU duration of the measurement resource of the inference report, are extended to accommodate the monitoring report. This extension ensures that the UE 320 can effectively manage the increased computational and resource demands associated with monitoring, thus maintaining processing efficiency and accuracy. However, if no aperiodic monitoring report trigger is expected, the active resource and port counting, along with the CPU duration of the measurement resource, remain at their legacy settings, concluding at the end of the inference report. This approach allows the UE 320 to conserve resources and optimize efficiency when monitoring is not required, unless an artificial deadline extends the duration beyond the inference report.
[0102] For aperiodic inference, monitoring is exclusively expected to be conducted through aperiodic monitoring. This alignment ensures that the monitoring process is tailored to the specific characteristics of aperiodic inference, optimizing the evaluation of prediction accuracy. In contrast, semi-persistent inference offers two alternatives for monitoring. The first alternative may involve monitoring solely through semi-persistent monitoring, providing a consistent and predictable evaluation framework. The second alternative may allow for monitoring through either semi-persistent or aperiodic monitoring, but not both, offering flexibility in monitoring approaches while ensuring that the evaluation process remains focused and efficient.
[0103] These options and alternatives depicted in Fig. 6 provide a structured framework for managing monitoring triggers and processes, ensuring that the UE 320 can effectively evaluate prediction accuracy and optimize network operations. By accommodating different timing and monitoring strategies, the system can adapt to varying network conditions and requirements, enhancing overall performance and reliability.
[0104] Fig. 7 is a diagram illustrating an example 700 of a timeline for performance monitoring.
[0105] In some aspects, the UE 320 may use an extended timeline for monitoring activities with respect to the first and second options described in connection with Fig. 6. This approach ensures that the UE 320 can thoroughly process and evaluate the inference measurement resource alongside the ground-truth measurement resource within the monitoring timeline. As a result, the UE 320 is able to handle inference measurement, prediction, and ground-truth measurement cohesively, enhancing the accuracy and reliability of the monitoring process.
[0106] In some aspects, where implementation proceeds without buffering inference results, the UE 320 may calculate at least two PMIs. This requirement underscores the importance of comprehensive data analysis, ensuring that both predicted and non-predicted historical CSI are evaluated. The gap between the last symbol of the DCI and the monitoring report encompasses several critical operations, including DCI decoding, locating CSI-RSs for target and historical CSI, channel measurement, CSI prediction using AI / ML models, and PMI and SGCS calculations for both predicted and historical CSI. This sequence of operations builds upon the time requirements established for AI / ML-based CSI prediction, incorporating additional time for channel measurement, CSI-RS location, and PMI / SGCS processing.
[0107] The time requirement 702 for these operations may be defined as greater than or equal to Zml + w + Z 'or Zml + w + 2Z', where w accounts for various factors such as the number of measurement instances, and prediction delays. The term w may also represent 14 (K-1) m + nref + predictionDelay + 14 (N4-1) d. 14 refers to the number of symbols in one slot. K is the number of aperiodic burst or periodic / semi-persistent occasions for the measurement resource of the historical CSI for model input that is configured in the inference report, m is the gap in the units of slots, N4 is the number of prediction instances in the inference report (i.e., the number of RS bursts or occasions in the ground-truth measurement) , and d is the gap between the units of slots. This comprehensive time allocation ensures that the UE can efficiently manage and process the necessary data, maintaining accuracy and reliability in its predictions and evaluations.
[0108] In some aspects, the timeline between monitoring trigger and SGCS report may include both the measurement resource 312 for ground-truth and the measurement resource for model input. More specifically, since the PMI representation of the precoder is used in SGCS calculation, the timeline formulation of a codebook can be used. The time requirement may be given by Zaiml-csi-pred+w+Z′, where w=14 (K-1) m+predictionDelay+14 (N4-1) d. Moreover, Zaiml-csi-pred is the AI / ML timeline in inference because prediction inference should be taken into account, and Z′is added as a relaxed timeline because two PMI calculations are needed (one for predicted and one for ground-truth) .
[0109] Similarly, the gap between the last symbol of the measurement resource for ground-truth or target CSI and the monitoring report involves essential operations such as channel measurements, PMI calculation, and SGCS calculation. These operations align with potential activities in CSI calculation, including channel measurements, CSI prediction, and PMI calculation. The solution leverages the time requirements defined for a codebook, utilizing values of a time requirement 704, such as Z'or 2Z'up to the UE’s capability. This strategic reuse of established time requirements ensures that the UE can effectively manage its resources and processing capabilities, optimizing performance and enhancing the accuracy of its monitoring and prediction activities.
[0110] Fig. 8 is a diagram illustrating an example 800 of a timeline for performance monitoring. In some aspects, the UE 320 may buffer the results in the inference CSI report 304 and remain prepared to receive a monitoring trigger 306 at any time. Consequently, a focus may be on the ground-truth measurement resource, excluding the inference measurement resource from consideration by the UE 320. This approach simplifies the implementation by narrowing the scope to essential components, allowing for a more streamlined process in handling ground-truth data. The UE 320 may need to know whether buffering is required in order to optimize resource allocation and efficiency.
[0111] In an example where implementation involves buffering inference results, the UE 320 may calculate only one PMI. This reduces computational complexity and resource usage, ensuring that the UE 320 can efficiently manage its processing capabilities. The gap between the last symbol of the measurement resource for ground-truth or target CSI and the monitoring report encompasses several key operations, including channel measurements, PMI calculation of the target CSI, and SGCS calculation. These operations align with potential activities in CSI calculation, such as channel measurements, CSI prediction, and PMI calculation. The solution leverages the time requirements defined for the codebook, utilizing values like Z'or 2 × Z 'up to the UE’s capability. This strategic reuse of established time requirements ensures that the UE can effectively manage its resources and processing capabilities, optimizing performance and enhancing the accuracy of its monitoring and prediction activities. CSI processing criteria may be enhanced to reflect additional buffering time and quantity. This enhancement is helpful for managing increased data loads and maintaining processing efficiency
[0112] Similarly, the gap between the last symbol of the DCI and the monitoring report involves critical operations such as DCI decoding, locating CSI-RSs for target CSI, channel measurement, PMI calculation of the target CSI, and SGCS calculation. These operations are consistent with potential activities in CSI calculation, including DCI decoding, locating CSI-RS for measurement, channel measurement, CSI prediction, and PMI calculation. The solution employs time requirement logic, using formulas for a time requirement 802, such as Z2 + w or Z2 + w + Z', where w=14 (N4-1) d with N4 being the number of prediction instances in the inference report, the number of aperiodic bursts for the ground-truth measurement, or the number of measurement occasions for the ground-truth measurement if periodic / semi-persistent CSI-RS is used, d being the gap between prediction instances, or bursts or occasions. This comprehensive time allocation ensures that the UE can efficiently manage and process the necessary data, maintaining accuracy and reliability in its predictions and evaluations. By adhering to these time requirements, the UE 320 facilitates effective resource management and processing efficiency, supporting the UE’s ability to deliver accurate and reliable performance monitoring and prediction outcomes.
[0113] Fig. 9 is a diagram illustrating examples 900 and 902 of reporting SGCSs.
[0114] In some aspects, the UE 320 may report two SGCS metrics to provide a comprehensive evaluation of CSI prediction accuracy. One SGCS metric is calculated for the predicted CSI, while the other is derived from non-predicted CSI. This dual reporting approach allows for a nuanced assessment of the prediction model’s performance, providing insights into both the predicted and non-predicted aspects of CSI. The non-predicted CSI is based on a Type 1 codebook or an enhanced Type 2 codebook, which is configured by the network and subject to the UE capability. This configuration ensures that the evaluation process is tailored to the specific characteristics and constraints of the network and UE, optimizing the accuracy and reliability of the SGCS calculations.
[0115] In an example the UE 320 may double the CPU and active resource / port counting. Examples 900 and 902 show a CPU duration 904 for predicted CSI and a CPU duration 906 for non-predicted CSI. This increase in computational and resource allocation is essential to accommodate the additional processing demands associated with calculating two SGCS metrics. By doubling these resources, the UE 320 can efficiently manage the increased data load and maintain processing efficiency, ensuring that both predicted and non-predicted CSI are thoroughly evaluated. This approach underscores the importance of comprehensive data analysis, allowing for a detailed comparison between predicted and non-predicted CSI and enhancing the reliability of the SGCS calculations.
[0116] Buffering both predicted and non-predicted CSI data ensures that all necessary information is retained for accurate performance monitoring. This buffering process is helpful for maintaining the integrity of the prediction model and ensuring reliable evaluation outcomes. By accommodating both predicted and non-predicted CSI, the UE 320 facilitates a holistic assessment of prediction accuracy, providing deeper insights into the model’s performance and guiding network optimization efforts.
[0117] Examples 900 and 902 show a strategic approach to CSI prediction evaluation, leveraging dual SGCS reporting and increased resource allocation to deliver accurate and reliable performance monitoring. By tailoring the evaluation process to the specific characteristics of the network entity 310 and the UE 320, the accuracy and reliability of the SGCS calculations improves, supporting informed decision-making and optimization strategies.
[0118] For performance monitoring, the CPU counting rule and active resource / port counting rule may follow a legacy rule of the CSI report using a 3GPP Release 18 Doppler CSI codebook. Specifically, for the CPU counting rule, O_CPU may be the number of CPU allocated for a report. If aperiodic CSI-RS with K=12 bursts, the UE 320 may support O_CPU = 8. If aperiodic CSI-RS with K < 12 bursts, the UE 320 may support O_CPU = Y1 × K where Y1 = {1, 2, 3} is a UE capability. K is the N4 value configured in the inference report. If periodic / semi-persistent CSI-RS, the UE 320 may support O_CPU = max (4, Y2 × N4) and Y2 = {1, 2, 3} is a UE capability. N4 is the number of prediction instances in the inference report and also the number of measurement occasions in the monitoring report.
[0119] For an active resource counting rule, if aperiodic CSI-RS with K bursts, each resource and port is counted once, and a total of K times are counted. If periodic / semi-persistent CSI-RS, each resource and port is counted Kp times, Kp = {1, 2, 4} up to a UE capability.
[0120] Alternatively, for a CPU counting rule, O_CPU = max (4, Y2 × N4) and Y2 ={1, 2, 3} up to a UE capability is used for all of aperiodic CSI-RS, periodic CSI-RS, and semi-persistent CSI-RS. For active resource counting, if periodic / semi-persistent CSI-RS is used, each resource and port is counted N4 times, where N4 is the number of prediction instances in the inference report and also the number of measurement occasions in the monitoring report.
[0121] Fig. 10 is a diagram illustrating an example process 1000 performed, for example, at a UE or an apparatus of a UE. Example process 1000 is an example where the apparatus or the UE (e.g., UE 320) performs operations associated with timing for CSI prediction and monitoring.
[0122] As shown in Fig. 10, in some aspects, process 1000 may include predicting future CSI using historical CSI data and an ML model (block 1010) . For example, the UE (e.g., using communication manager 1106, depicted in Fig. 11) may predict future CSI using historical CSI data and an ML model, as described above.
[0123] As further shown in Fig. 10, in some aspects, process 1000 may include determining an SGCS metric by comparing the predicted CSI with actual ground-truth CSI within a configurable monitoring window associated with an extended CPU duration (block 1020) . For example, the UE (e.g., using communication manager 1106, depicted in Fig. 11) may determine an SGCS metric by comparing the predicted CSI with actual ground-truth CSI within a configurable monitoring window associated with an extended CPU duration, as described above.
[0124] As further shown in Fig. 10, in some aspects, process 1000 may include receiving a monitoring trigger (block 1030) . For example, the UE (e.g., using reception component 1102 or communication manager 1106, depicted in Fig. 11) may receive a monitoring trigger, as described above.
[0125] As further shown in Fig. 10, in some aspects, process 1000 may include transmitting an indication of the SGCS metric in an SGCS report to a network entity (block 1040) . For example, the UE (e.g., using transmission component 1104 or communication manager 1106, depicted in Fig. 11) may transmit an indication of the SGCS metric in an SGCS report to a network entity, as described above.
[0126] Process 1000 may include additional aspects, such as any single aspect or any combination of aspects described below or in connection with one or more other processes described elsewhere herein.
[0127] In a first aspect, predicting the future CSI includes monitoring and active resource or port counting for an inference report measurement resource for the historical CSI data during the extended CPU duration and until the SGCS report or a monitoring report, based at least in part on receiving the monitoring trigger.
[0128] In a second aspect, alone or in combination with the first aspect, receiving the monitoring trigger includes receiving the monitoring trigger within a specified time period after an inference trigger that triggers the predicting of the future CSI.
[0129] In a third aspect, alone or in combination with one or more of the first and second aspects, receiving the monitoring trigger includes receiving the monitoring trigger no later than a first symbol of the inference report measurement resource.
[0130] In a fourth aspect, alone or in combination with one or more of the first through third aspects, receiving the monitoring trigger includes receiving the monitoring trigger no later than an inference report that carries the predicted CSI.
[0131] In a fifth aspect, alone or in combination with one or more of the first through fourth aspects, receiving the monitoring trigger includes receiving the monitoring trigger before a specified deadline.
[0132] In a sixth aspect, alone or in combination with one or more of the first through fifth aspects, the determining is performed based at least in part on receiving a monitoring trigger prior to the SGCS report by a time requirement, where the time requirement includes at least one of a duration of receiving a measurement resource of historical CSI for model input and a measurement resource of ground-truth CSI.
[0133] In a seventh aspect, alone or in combination with one or more of the first through sixth aspects, the time requirement is based at least in part on a time requirement for performing prediction inference via the ML model or performing PMI calculation using codebooks based on Doppler-domain compression.
[0134] In an eighth aspect, alone or in combination with one or more of the first through seventh aspects, the SGCS metric comprises calculating two SGCS values, one based on predicted CSI and ground-truth CSI, and another based on non-predicted CSI and ground-truth CSI, where a CPU calculation is doubled.
[0135] In a ninth aspect, alone or in combination with one or more of the first through eighth aspects, the extended CPU duration accommodates additional buffering time for prediction results.
[0136] In a tenth aspect, alone or in combination with one or more of the first through ninth aspects, process 1000 includes extending an active resource and port counting duration to accommodate SGCS calculation and reporting.
[0137] In an eleventh aspect, alone or in combination with one or more of the first through tenth aspects, if an aperiodic resource is used for monitoring, a number of CPUs of the extended CPU duration is based at least in part on a number of aperiodic reference signal bursts, where the number of aperiodic reference signal bursts is based at least in part on a number of prediction instances configured in an inference report.
[0138] In a twelfth aspect, alone or in combination with one or more of the first through eleventh aspects, if a periodic or a semi-persistent resource is used, a number of CPUs of the extended CPU duration is based at least in part on a number of reference signal occasions, where the number of reference signal occasions is based at least in part on a number of prediction instances configured in an inference report.
[0139] In a thirteenth aspect, alone or in combination with one or more of the first through twelfth aspects, if an aperiodic resource is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted once.
[0140] In a fourteenth aspect, alone or in combination with one or more of the first through thirteenth aspects, if a periodic or a semi-persistent reference signal is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted multiple times up to a UE capability or based at least in part on a number of prediction instances configured in an inference report.
[0141] In a fifteenth aspect, alone or in combination with one or more of the first through fourteenth aspects, an extended CPU occupation duration for a measurement resource of the historical CSI ends at the last symbol of the physical uplink channel (e.g., PUSCH, PUCCH) that carries the SGCS report.
[0142] In a sixteenth aspect, alone or in combination with one or more of the first through fifteenth aspects, an active duration of a resource and port counting for a measurement resource of the historical CSI ends at the last symbol of the physical uplink channel (e.g., PUSCH, PUCCH) that carries the SGCS report.
[0143] Although Fig. 10 shows example blocks of process 1000, in some aspects, process 1000 may include additional blocks, fewer blocks, different blocks, or differently arranged blocks than those depicted in Fig. 10. Additionally, or alternatively, two or more of the blocks of process 1000 may be performed in parallel.
[0144] Fig. 11 is a diagram of an example apparatus 1100 for wireless communication. The apparatus 1100 may be a UE, or a UE may include the apparatus 1100. In some aspects, the apparatus 1100 includes a reception component 1102, a transmission component 1104, or a communication manager 1106, which may be in communication with one another (for example, via one or more buses or one or more other components) . In some aspects, the communication manager 1106 is the communication manager 150 described in connection with Fig. 1. As shown, the apparatus 1100 may communicate with another apparatus 1108, such as a UE or a network node (such as a CU, a DU, an RU, or a base station) , using the reception component 1102 and the transmission component 1104. The communication manager 1106 may be included in, or implemented via, a processing system (for example, the processing system 140 described in connection with Fig. 1) of the UE.
[0145] In some aspects, the apparatus 1100 may be configured to perform one or more operations described herein in connection with Figs. 1-9. Additionally, or alternatively, the apparatus 1100 may be configured to perform one or more processes described herein, such as process 1000 of Fig. 10. In some aspects, the apparatus 1100 or one or more components shown in Fig. 11 may include one or more components of the UE described in connection with Fig. 1. Additionally, or alternatively, one or more components shown in Fig. 11 may be implemented within one or more components described in connection with Fig. 1. Additionally, or alternatively, one or more components of the set of components may be implemented at least in part as software stored in one or more memories. For example, a component (or a portion of a component) may be implemented as instructions or code stored in a non-transitory computer-readable medium and executable by one or more controllers or one or more processors to perform the functions or operations of the component.
[0146] The reception component 1102 may receive communications, such as reference signals, control information, data communications, or a combination thereof, from the apparatus 1108. The reception component 1102 may provide received communications to one or more other components of the apparatus 1100. In some aspects, the reception component 1102 may perform signal processing on the received communications, and may provide the processed signals to the one or more other components of the apparatus 1100. In some aspects, the reception component 1102 may include one or more components of the UE described above in connection with Fig. 1, such as a radio, one or more RF chains, one or more transceivers, or one or more modems, each of which may in turn be coupled with one or more antennas of the UE.
[0147] The transmission component 1104 may transmit communications, such as reference signals, control information, data communications, or a combination thereof, to the apparatus 1108. In some aspects, one or more other components of the apparatus 1100 may generate communications and may provide the generated communications to the transmission component 1104 for transmission to the apparatus 1108. In some aspects, the transmission component 1104 may perform signal processing on the generated communications, and may transmit the processed signals to the apparatus 1108. In some aspects, the transmission component 1104 may include one or more components of the UE described above in connection with Fig. 1, such as a radio, one or more RF chains, one or more transceivers, or one or more modems, each of which may in turn be coupled with one or more antennas of the UE described in connection with Fig. 1. In some aspects, the transmission component 1104 may be co-located with the reception component 1102.
[0148] The communication manager 1106 may support operations of the reception component 1102 or the transmission component 1104. For example, the communication manager 1106 may receive information associated with configuring reception of communications by the reception component 1102 or transmission of communications by the transmission component 1104. Additionally, or alternatively, the communication manager 1106 may generate or provide control information to the reception component 1102 or the transmission component 1104 to control reception or transmission of communications.
[0149] The communication manager 1106 may predict future CSI using historical CSI data and an ML model. The communication manager 1106 may determine an SGCS metric by comparing the predicted CSI with actual ground-truth CSI within a configurable monitoring window associated with an extended CPU duration. The transmission component 1104 may transmit an indication of the SGCS metric in an SGCS report to a network entity.
[0150] The communication manager 1106 may extend an active resource and port counting duration to accommodate SGCS calculation and reporting.
[0151] The number and arrangement of components shown in Fig. 11 are provided as an example. In practice, there may be additional components, fewer components, different components, or differently arranged components than those shown in Fig. 11. Furthermore, two or more components shown in Fig. 11 may be implemented within a single component, or a single component shown in Fig. 11 may be implemented as multiple, distributed components. Additionally, or alternatively, a set of (one or more) components shown in Fig. 11 may perform one or more functions described as being performed by another set of components shown in Fig. 11.
[0152] The following provides an overview of some Aspects of the present disclosure:
[0153] Aspect 1: A method for wireless communication performed by user equipment (UE) , comprising: predicting future channel state information (CSI) using historical CSI data and a machine learning (ML) model; determining a squared generalized cosine similarity (SGCS) metric by comparing the predicted CSI with actual ground-truth CSI; and transmitting an indication of the SGCS metric in an SGCS report to a network entity.
[0154] Aspect 2: The method of Aspect 1, wherein predicting the future CSI includes monitoring and active resource or port counting for an inference report measurement resource for the historical CSI data during the extended CPU duration and until the SGCS report or a monitoring report, based at least in part on receiving a monitoring trigger.
[0155] Aspect 3: The method of Aspect 1 or 2, wherein receiving the monitoring trigger includes receiving the monitoring trigger within a specified time period after an inference trigger that triggers the predicting of the future CSI.
[0156] Aspect 4: The method of any of Aspects 1-3, wherein receiving the monitoring trigger includes receiving the monitoring trigger no later than a first symbol of the inference report measurement resource.
[0157] Aspect 5: The method of any of Aspects 1-4, wherein receiving the monitoring trigger includes receiving the monitoring trigger no later than an inference report that carries the predicted CSI.
[0158] Aspect 6: The method of any of Aspects 1-5, wherein receiving the monitoring trigger includes receiving the monitoring trigger before a specified deadline.
[0159] Aspect 7: The method of any of Aspects 1-6, wherein the determining is performed based at least in part on receiving a monitoring trigger prior to the SGCS report by a time requirement, wherein the time requirement includes at least one of a duration of receiving a measurement resource of historical CSI for model input and a measurement resource of ground-truth CSI.
[0160] Aspect 8: The method of Aspect 7, wherein the time requirement is based at least in part on a time requirement for performing prediction inference via the ML model or performing precoding matrix indicator calculation using codebooks based on Doppler-domain compression.
[0161] Aspect 9: The method of any of Aspects 1-8, wherein the SGCS metric comprises: calculating two SGCS values, one based on predicted CSI and ground-truth CSI, and another based on non-predicted CSI and ground-truth CSI, wherein a CPU calculation is doubled.
[0162] Aspect 10: The method of any of Aspects 1-9, wherein an extended CSI processing unit (CPU) duration for a measurement resource of the historical CSI ends at the last symbol of the physical uplink channel that carries the SGCS report.
[0163] Aspect 11: The method of any of Aspects 1-10, wherein an active duration of a resource and port counting for a measurement resource of the historical CSI ends at the last symbol of the physical uplink channel that carries the SGCS report.
[0164] Aspect 12: The method of any of Aspects 1-11, wherein if an aperiodic resource is used for monitoring, a number of CSI processing units (CPUs) of an extended CPU duration is based at least in part on a number of aperiodic reference signal bursts, where the number of aperiodic reference signal bursts is based at least in part on a number of prediction instances configured in an inference report.
[0165] Aspect 13: The method of any of Aspects 1-12, wherein if a periodic or a semi-persistent resource is used, a number of extended CSI processing units (CPUs) of the extended CPU duration is based at least in part on a number of reference signal occasions, where the number of reference signal occasions is based at least in part on a number of prediction instances configured in an inference report.
[0166] Aspect 14: The method of any of Aspects 1-13, if an aperiodic resource is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted once.
[0167] Aspect 15: The method of any of Aspects 1-14, wherein if a periodic or a semi-persistent reference signal is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted multiple times up to a UE capability or based at least in part on a number of prediction instances configured in an inference report.
[0168] Aspect 16: The method of any of Aspects 1-15, wherein the inference report measurement resource is no later than a CSI reference resource of the associated inference report.
[0169] Aspect 17: The method of any of Aspects 1-16, further comprising measuring ground-truth CSI based at least in part on the monitoring trigger and determining the monitoring trigger before a specified deadline.
[0170] Aspect 18: The method of any of Aspects 1-17, wherein an active resource and port counting time for an aperiodic resource of the inference report is extended to an end of the monitoring report.
[0171] Aspect 19: The method of Aspect 18, wherein the aperiodic resource comprises an aperiodic CSI reference signal (CSI-RS) resource associated with the inference report.
[0172] Aspect 20: The method of any of Aspects 1-19, further comprising calculating calculate two SGCS values, one based on the predicted CSI and the ground-truth CSI, and another based on non-predicted CSI and the ground-truth CSI.
[0173] Aspect 21: The method of any of Aspects 1-20, wherein an extended CSI processing unit (CPU) duration for a measurement resource of the historical CSI ends at the last symbol of a physical uplink channel that carries the SGCS report.
[0174] Aspect 22: The method of any of Aspects 1-21, wherein an active duration of a resource and port counting for a measurement resource of the historical CSI ends at the last symbol of a physical uplink channel that carries the SGCS report.
[0175] Aspect 23: The method of any of Aspects 1-22, wherein if an aperiodic resource is used for monitoring, a number of CSI processing units (CPUs) of an extended CPU duration is based at least in part on a number of aperiodic reference signal bursts, where the number of aperiodic reference signal bursts is based at least in part on a number of prediction instances configured in the inference report.
[0176] Aspect 24: The method of any of Aspects 1-23, wherein if a periodic or a semi-persistent resource is used, a number of CSI processing units (CPUs) of an extended CPU duration is based at least in part on a number of reference signal occasions, where the number of reference signal occasions is based at least in part on a number of prediction instances configured in the inference report.
[0177] Aspect 25: The method of any of Aspects 1-24, wherein if an aperiodic resource is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted once.
[0178] Aspect 26: The method of any of Aspects 1-25, wherein if a periodic or a semi-persistent reference signal is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted multiple times up to a UE capability or based at least in part on a number of prediction instances configured in the inference report.
[0179] Aspect 27: An apparatus for wireless communication at a device, the apparatus comprising one or more processors; one or more memories coupled with the one or more processors; and instructions stored in the one or more memories and executable by the one or more processors to cause the apparatus to perform the method of one or more of Aspects 1-26.
[0180] Aspect 28: An apparatus for wireless communication at a device, the apparatus comprising one or more memories and one or more processors coupled to the one or more memories, the one or more processors configured to cause the device to perform the method of one or more of Aspects 1-26.
[0181] Aspect 29: An apparatus for wireless communication, the apparatus comprising at least one means for performing the method of one or more of Aspects 1-26.
[0182] Aspect 30: A non-transitory computer-readable medium storing code for wireless communication, the code comprising instructions executable by one or more processors to perform the method of one or more of Aspects 1-26.
[0183] Aspect 31: A non-transitory computer-readable medium storing a set of instructions for wireless communication, the set of instructions comprising one or more instructions that, when executed by one or more processors of a device, cause the device to perform the method of one or more of Aspects 1-26.
[0184] Aspect 32: A device for wireless communication, the device comprising a processing system that includes one or more processors and one or more memories coupled with the one or more processors, the processing system configured to cause the device to perform the method of one or more of Aspects 1-26.
[0185] Aspect 33: An apparatus for wireless communication at a device, the apparatus comprising one or more memories and one or more processors coupled to the one or more memories, the one or more processors individually or collectively configured to cause the device to perform the method of one or more of Aspects 1-26.
[0186] Aspect 34: A device comprising a processing system that includes one or more processors and one or more code-storing memories coupled with the one or more processors, the processing system configured to cause the device to perform the method of one or more of Aspects 1-26.
[0187] Aspect 35: A device comprising a processing system that includes processor circuitry and code-storing memory circuitry, the processing system configured to cause the device to perform the method of one or more of Aspects 1-26.
[0188] It will be apparent that systems or methods described herein may be implemented in different forms of hardware or a combination of hardware and software. A component being configured to perform a function means that the component has a capability to perform the function, and does not require the function to be actually performed by the component, unless noted otherwise.
[0189] As used herein, the term “determine” or “determining” can encompass one or more of a wide variety of actions. For example, “determining” can include one or more of calculating, computing, processing, deriving, detecting, estimating, investigating, looking up, inferring, ascertaining, measuring, resolving, selecting, choosing, obtaining, identifying, interpreting, demodulating, decoding, reading, establishing, forming or generating, among other examples. In some such examples, determining can involve a processor performing some type of calculating, computing, deriving, estimating, inferring, ascertaining, resolving, predicting or other processing to obtain one or more numerical values, sets, elements or other information or results. In some other such examples, determining can involve a processor identifying, looking up, investigating or otherwise obtaining some type of value, set, element or other information or result from a table, a data structure, a database or other memory device or location. In some other such examples, determining can involve a processor identifying, interpreting, demodulating, decoding, detecting, reading or otherwise obtaining some type of value, set, element or other information or result signaled in, for example, a received wireless packet. In some other such examples, determining can involve a processor selecting or choosing one or more values, sets, elements or other information or results from a larger set of values, sets elements or other information or results. In some other such examples, determining can involve a processor performing a measurement, such as on a received signal.
[0190] As used herein, the articles “a” and “an” are intended to refer to one or more items and may be used interchangeably with “one or more” or “at least one. ” As used herein, a phrase referring to “at least one of” or “one or more of” a list of items refers to any combination of those items, including single members. As an example, “at least one of: a, b, or c” is intended to cover: a, b, c, a-b, a-c, b-c, and a-b-c. Additionally, as used herein, a phrase referring to “a” or “an” element refers to one or more of such elements acting individually or collectively to perform the recited function (s) . Additionally, as used herein, a “set” can refer to one or more items, and a “subset” can refer to a whole set or less than the whole set, but not an empty set. “Set, ” “group, ” and similar terms are intended to include one or more items and may be used interchangeably with “one or more. ” Furthermore, as used herein, the term “or” is intended to be interpreted in the inclusive sense (such as when referring to a series) and may be used interchangeably with “and / or, ” unless otherwise explicitly indicated (for example, if used in conjunction with “either” or “only one of” ) . For example, “A or B” may include A only, B only, or a combination of A and B. Also, as used herein, the terms “has, ” “have, ” “having, ” “comprise, ” “comprising, ” “include” and “including, ” and derivatives thereof or similar terms are intended to be open-ended terms that do not limit an element that they modify (for example, an element “having” A also may have B) .
[0191] As used herein, the phrase “associated with” is intended to be interpreted in the inclusive sense, unless otherwise explicitly indicated. For example, the phrase “associated with” is not to be construed as a reference to a closed set of conditions, factors, criteria, elements, components, or actions, among other examples. Specifically, unless a phrase refers to “associated with only ‘a, ’ ” or the equivalent in context, whatever it is that is “associated with ‘a, ’ ” may be associated with “a” alone or associated with a combination of “a” and one or more other conditions, factors, criteria, elements, components, or actions, among other examples.
[0192] As used herein, “satisfying a threshold” may, depending on the context, refer to a value being greater than the threshold, greater than or equal to the threshold, less than the threshold, less than or equal to the threshold, equal to the threshold, or not equal to the threshold, among other examples.
[0193] Even though particular combinations of features are recited in the claims or disclosed in the specification, these combinations are not intended to limit the scope of all aspects described herein. Many of these features may be combined in ways not specifically recited in the claims or disclosed in the specification. The disclosure of various aspects includes each dependent claim in combination with every other claim in the claim set.
Claims
1.A user equipment (UE) , comprising:a processing system that includes one or more processors and one or more code-storing memories coupled with the one or more processors, the processing system configured to cause the UE to:predict future channel state information (CSI) using historical CSI data and a machine learning (ML) model;determine a squared generalized cosine similarity (SGCS) metric by comparing the predicted CSI with actual ground-truth CSI;receive a monitoring trigger associated with an inference report; andtransmit an indication of the SGCS metric in an SGCS report to a network entity.2.The UE of claim 1, wherein the processing system is configured to cause the UE to receive the monitoring trigger within a specified time period after an inference trigger that triggers prediction of the future CSI.3.The UE of claim 1, wherein to receive the monitoring trigger, the processing system is configured to cause the UE to receive the monitoring trigger no later than a first symbol of an inference report measurement resource.4.The UE of claim 3, wherein the inference report measurement resource is no later than a CSI reference resource of the associated inference report.5.The UE of claim 1, wherein the processing system is configured to cause the UE to measure ground-truth CSI based at least in part on the monitoring trigger and determine the monitoring trigger before a specified deadline.6.The UE of claim 1, wherein an active resource and port counting time for an aperiodic resource of the inference report is extended to an end of the monitoring report.7.The UE of claim 6, wherein the aperiodic resource comprises an aperiodic CSI reference signal (CSI-RS) resource associated with the inference report.8.The UE of claim 1, wherein the processing system is configured to cause the UE to calculate two SGCS values, one based on the predicted CSI and the ground-truth CSI, and another based on non-predicted CSI and the ground-truth CSI.9.The UE of claim 1, wherein an extended CSI processing unit (CPU) duration for a measurement resource of the historical CSI ends at the last symbol of a physical uplink channel that carries the SGCS report.10.The UE of claim 1, wherein an active duration of a resource and port counting for a measurement resource of the historical CSI ends at the last symbol of a physical uplink channel that carries the SGCS report.11.The UE of claim 1, wherein if an aperiodic resource is used for monitoring, a number of CSI processing units (CPUs) of an extended CPU duration is based at least in part on a number of aperiodic reference signal bursts, where the number of aperiodic reference signal bursts is based at least in part on a number of prediction instances configured in the inference report.12.The UE of claim 1, wherein if a periodic or a semi-persistent resource is used, a number of CSI processing units (CPUs) of an extended CPU duration is based at least in part on a number of reference signal occasions, where the number of reference signal occasions is based at least in part on a number of prediction instances configured in the inference report.13.The UE of claim 1, wherein if an aperiodic resource is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted once.14.The UE of claim 1, wherein if a periodic or a semi-persistent reference signal is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted multiple times up to a UE capability or based at least in part on a number of prediction instances configured in the inference report.15.A method for wireless communication performed by user equipment (UE) , comprising:predicting future channel state information (CSI) using historical CSI data and a machine learning (ML) model;determining a squared generalized cosine similarity (SGCS) metric by comparing the predicted CSI with actual ground-truth CSI;receiving a monitoring trigger associated with an inference report; andtransmitting an indication of the SGCS metric in an SGCS report to a network entity.16.The method of claim 15, wherein receiving the monitoring trigger includes receiving the monitoring trigger no later than a first symbol of an earliest occasion of a set of inference measurement resources that are no later than a CSI reference resource of the associated inference report.17.The method of claim 15, wherein the measurement resource for the inference report is used to measure historical CSI as input to the machine learning model.18.The method of claim 15, wherein an active resource and port counting for a measurement resource of the inference report is extended to an end of a monitoring report.19.The method of claim 15, wherein if an aperiodic resource is used for monitoring, a number of CSI processing units (CPUs) in an extended CPU duration is based at least in part on a number of aperiodic reference signal bursts, where the number of aperiodic reference signal bursts is based at least in part on a number of prediction instances configured in an inference report, andwherein if a periodic or a semi-persistent resource is used, a number of CSI processing units (CPUs) in an extended CPU duration is based at least in part on a number of reference signal occasions, where the number of reference signal occasions is based at least in part on a number of prediction instances configured in the inference report.20.The method of claim 15, wherein if an aperiodic resource is used for active resource and port counting for a monitoring report, a number of active resources and ports are counted once, andwherein if a periodic or a semi-persistent reference signal is used for active resource and port counting for a monitoring report, the number of active resources and ports are counted multiple times up to a UE capability or based at least in part on a number of prediction instances configured in the inference report.
