Devices, methods, and medium for communication
By training a model cluster with individual models from PRUs, the solution ensures data consistency, enhancing the accuracy and reliability of AI/ML-based positioning by aligning UE-side conditions, addressing inconsistencies in UE-side additional conditions.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- NEC CORP
- Filing Date
- 2025-01-26
- Publication Date
- 2026-07-30
AI Technical Summary
Ensuring consistency between training data and inference data is critical for AI/ML-based positioning models deployed at user equipment (UE) to maintain accurate positioning, as UE-side additional conditions may differ from training data conditions, leading to potential inaccuracies.
A model cluster is trained based on a training dataset associated with a set of positioning reference units (PRUs), comprising multiple individual models, each trained from different PRUs, to align UE-side additional conditions during inference, ensuring consistency and improving positioning accuracy.
The solution effectively addresses inconsistencies in UE-side conditions, enhancing the accuracy and reliability of AI/ML-based positioning by aligning training and inference data, thereby improving positioning precision.
Smart Images

Figure CN2025075180_30072026_PF_FP_ABST
Abstract
Description
DEVICES, METHODS, AND MEDIUM FOR COMMUNICATIONFIELD
[0001] Example embodiments of the present disclosure generally relate to the field of communication techniques and in particular, to devices, methods, and a computer readable medium for communication.BACKGROUND
[0002] Supporting various positioning methods to provide reliable, timely, and accurate user equipment (UE) location is one of key features of the third generation partnership project (3GPP) standard. It has been agreed to investigate the potential for artificial intelligence (AI) / machine learning (ML) in air interface to improve comprehensive performance in 5G-adcanced (5G-A) . AI / ML based mechanism to improve the positioning accuracy is one of the use cases to apply AI / ML in air interface.
[0003] A consistency between training data and inference data is critical to ensure the good performance for a trained model. In case the AI / ML positioning model is deployed at UE side, how to enable the consistency between training data and inference data should be studied.SUMMARY
[0004] In general, example embodiments of the present disclosure provide devices, methods, and a computer storage medium for communication.
[0005] In a first aspect, there is provided a communication device. The communication device comprises at least one processor configured to cause the communication device at least to: obtain a training dataset comprising a plurality of samples, wherein the training dataset is generated at least based on measurements from a set of positioning reference units (PRU) ; and determine at least one individual model within a model cluster based on at least part of the training dataset, wherein a first individual model of the at least one individual model has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model.
[0006] In a second aspect, there is provided a terminal device. The terminal device comprises at least one processor configured to cause the terminal device at least to: activate at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; perform a model inference by using the activated at least one individual model based on a model input, to determine an inference result; and transmit, to a location server, a first message comprising the inference result.
[0007] In a third aspect, there is provided a location server. The location server comprises at least one processor configured to cause the location server at least to: transmit, to a terminal device, a second message comprising an indication that a model cluster should be activated, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; and receive, from the terminal device, a first message comprising an inference result associated with model inference of at least one individual model within the model cluster.
[0008] In a fourth aspect, there is provided a communication device. The communication device comprises at least one processor configured to cause the communication device at least to: obtain at least one model output of at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; and determine at least one monitoring metric of the at least one individual model or a monitoring metric of the model cluster based on the at least one model output.
[0009] In a fifth aspect, there is provided a method of communication performed by a communication device. The method comprises: obtaining a training dataset comprising a plurality of samples, wherein the training dataset is generated at least based on measurements from a set of PRUs; and determining at least one individual model within a model cluster based on at least part of the training dataset, wherein a first individual model of the at least one individual model has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model.
[0010] In a sixth aspect, there is provided a method of communication performed by a terminal device. The method comprises: activating at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; performing a model inference by using the activated at least one individual model based on a model input, to determine an inference result; and transmitting, to a location server, a first message comprising the inference result.
[0011] In a seventh aspect, there is provided a method of communication performed by a location server. The method comprises: transmitting, to a terminal device, a second message comprising an indication that a model cluster should be activated, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; and receiving, from the terminal device, a first message comprising an inference result associated with model inference of at least one individual model within the model cluster.
[0012] In an eighth aspect, there is provided a method of communication performed by a communication device. The method comprises: obtaining at least one model output of at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; and determining at least one monitoring metric of the at least one individual model or a monitoring metric of the model cluster based on the at least one model output.
[0013] In a ninth aspect, there is provided a computer readable medium having instructions stored thereon, the instructions, when executed on at least one processor, cause the at least one processor to carry out the method according to any of the fifth to the eighth aspects above.
[0014] It is to be understood that the summary section is not intended to identify key or essential features of embodiments of the present disclosure, nor is it intended to be used to limit the scope of the present disclosure. Other features of the present disclosure will become easily comprehensible through the following description.BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Through the more detailed description of some example embodiments of the present disclosure in the accompanying drawings, the above and other objects, features and advantages of the present disclosure will become more apparent, wherein:
[0016] FIG. 1A an example communication network in which some embodiments of the present disclosure can be implemented;
[0017] FIGS. 1B-1E illustrate some example schematics for AI / ML based positioning;
[0018] FIG. 2 illustrates a signalling chart illustrating a communication process for model training in accordance with some example embodiments of the present disclosure;
[0019] FIG. 3 illustrates a signalling chart illustrating a communication process for model inference in accordance with some example embodiments of the present disclosure;
[0020] FIG. 4 illustrates an example for determining an inference result in accordance with some example embodiments of the present disclosure;
[0021] FIG. 5 illustrates an example for model monitoring in accordance with some example embodiments of the present disclosure;
[0022] FIG. 6 illustrates a flowchart of an example method implemented at a communication device in accordance with some embodiments of the present disclosure;
[0023] FIG. 7 illustrates a flowchart of an example method implemented at a terminal device in accordance with some embodiments of the present disclosure;
[0024] FIG. 8 illustrates a flowchart of an example method implemented at a location server in accordance with some embodiments of the present disclosure;
[0025] FIG. 9 illustrates a flowchart of an example method implemented at a communication device in accordance with some embodiments of the present disclosure; and
[0026] FIG. 10 illustrates a simplified block diagram of a device that is suitable for implementing embodiments of the present disclosure.
[0027] Throughout the drawings, the same or similar reference numerals represent the same or similar element.DETAILED DESCRIPTION
[0028] Principle of the present disclosure will now be described with reference to some example embodiments. It is to be understood that these embodiments are described only for the purpose of illustration and help those skilled in the art to understand and implement the present disclosure, without suggesting any limitation as to the scope of the disclosure. Embodiments described herein can be implemented in various manners other than the ones described below.
[0029] In the following description and claims, unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skills in the art to which this disclosure belongs.
[0030] References in the present disclosure to “one embodiment, ” “an embodiment, ” “an example embodiment, ” and the like indicate that the embodiment described may include a particular feature, structure, or characteristic, but it is not necessary that every embodiment includes the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to affect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.
[0031] It shall be understood that although the terms “first” and “second” etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could be termed a second element, and similarly, a second element could be termed a first element, without departing from the scope of example embodiments. As used herein, the term “and / or” includes any and all combinations of one or more of the listed terms.
[0032] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a” , “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” , “comprising” , “has” , “having” , “includes” and / or “including” , when used herein, specify the presence of stated features, elements, and / or components etc., but do not preclude the presence or addition of one or more other features, elements, components and / or combinations thereof.
[0033] In some examples, values, procedures, or apparatus are referred to as “best, ” “lowest, ” “highest, ” “minimum, ” “maximum, ” or the like. It will be appreciated that such descriptions are intended to indicate that a selection among many used functional alternatives can be made, and such selections need not be better, smaller, higher, or otherwise preferable to other selections.
[0034] As used herein, the term “communication network” refers to a network following any suitable communication standards or technologies, such as New Radio (NR) , Long Term Evolution (LTE) , LTE-Advanced (LTE-A) , Code Divided Multiple Address (CDMA) , Frequency Divided Multiple Address (FDMA) , Time Divided Multiple Address (TDMA) , Frequency Divided Duplexer (FDD) , Time Divided Duplexer (TDD) , Multiple-Input Multiple-Output (MIMO) , Orthogonal Frequency Divided Multiple Access (OFDMA) , cdma2000, Wideband Code Division Multiple Access (WCDMA) , High-Speed Packet Access (HSPA) , Global System for Mobile Communications (GSM) , Narrow Band Internet of Things (NB-IoT) and so on. Furthermore, the communications between a terminal device and a network device in the communication network may be performed according to any suitable generation communication protocols, including, but not limited to, the first generation (1G) , the second generation (2G) , 2.5G, 2.75G, the third generation (3G) , the fourth generation (4G) , 4.5G, the fifth generation (5G) , 5.5G, 5G-Advanced networks, beyond 5G (B5G) , the sixth generation (6G) communication protocols, wireless local network communication protocols such as Institute for Electrical and Electronics Engineers (IEEE) 802.11 and the like, and / or any other protocols either currently known or to be developed in the future. The techniques described herein may be used for the wireless networks and radio technologies mentioned above as well as other wireless networks and radio technologies. Embodiments of the present disclosure may be applied in various communication systems. Given the rapid development in communications, there will of course also be future type communication technologies and systems with which the present disclosure may be embodied. It should not be seen as limiting the scope of the present disclosure to only the aforementioned system.
[0035] As used herein, the term “terminal device” refers to any device having wireless or wired communication capabilities. Examples of terminal device include, but not limited to, user equipment (UE) , personal computers, desktops, mobile phones, cellular phones, smart phones, personal digital assistants (PDAs) , portable computers, tablets, wearable devices, internet of things (IoT) devices, Ultra-reliable and Low Latency Communications (URLLC) devices, Internet of Everything (IoE) devices, machine type communication (MTC) devices, device on vehicle for V2X communication where X means pedestrian, vehicle, or infrastructure / network, devices for Integrated Access and Backhaul (IAB) , Space borne vehicles or Air borne vehicles in Non-terrestrial networks (NTN) including Satellites and High Altitude Platforms (HAPs) encompassing Unmanned Aircraft Systems (UAS) , eXtended Reality (XR) devices including different types of realities such as Augmented Reality (AR) , Mixed Reality (MR) and Virtual Reality (VR) , the unmanned aerial vehicle (UAV) commonly known as a drone which is an aircraft without any human pilot, devices on high speed train (HST) , or image capture devices such as digital cameras, sensors, gaming devices, music storage and playback appliances, or Internet appliances enabling wireless or wired Internet access and browsing and the like. The ‘terminal device’ can further has ‘multicast / broadcast’ feature, to support public safety and mission critical, V2X applications, transparent IPv4 / IPv6 multicast delivery, IPTV, smart TV, radio services, software delivery over wireless, group communications and IoT applications. It may also be incorporated one or multiple Subscriber Identity Module (SIM) as known as Multi-SIM. The term “terminal device” can be used interchangeably with a UE, a mobile station, a subscriber station, a mobile terminal, a user terminal or a wireless device. The term “target UE” refer to a UE whose distance, direction and / or position is measured.
[0036] As used herein, the term “network device” refers to a device which is capable of providing or hosting a cell or coverage where terminal devices can communicate. Examples of a network device include, but not limited to, a satellite, an unmanned aerial systems (UAS) platform, a Node B (NodeB or NB) , an evolved NodeB (eNodeB or eNB) , a next generation NodeB (gNB) , a transmission reception point (TRP) , a remote radio unit (RRU) , a radio head (RH) , a remote radio head (RRH) , an IAB node, a low power node such as a femto node, a pico node, a reconfigurable intelligent surface (RIS) , and the like.
[0037] As used herein, the term “TRP” may refer to an antenna port or an antenna array (with one or more antenna elements) available to the network device located at a specific geographical location, or a set of geographically co-located antennas (e.g. antenna array (with one or more antenna elements) ) supporting transmission point (TP) and / or reception point (RP) functionality. For example, a network device may be coupled with multiple TRPs in different geographical locations to achieve better coverage. Alternatively, or in addition, multiple TRPs may be incorporated into a network device, or in other words, the network device may comprise the multiple TRPs. The term “TRP” may be also referred to as a cell, such as a macro-cell, a micro-cell, a small cell, a pico-cell, a femto-cell, a remote radio head, a relay node, a gNB, etc. It is to be understood that the term “TRP” may refer to a logical concept which may be physically implemented by various manners. There may be an explicit TRP identification for a TRP.
[0038] As used herein, the term “location device” or “location server” refers to a device which is capable to manage the support of different location services for target UEs, including positioning of UEs and delivery of assistance data to UEs. The location device may interact with the serving gNB or serving ng-eNB for a target UE in order to obtain position measurements for the UE, including uplink measurements made by an NG-RAN and downlink measurements made by the UE that were provided to an NG-RAN as part of other functions such as for support of handover. Examples of a location device include, but not limited to, Location Management Function (LMF) , which is located in the access network or in a core network.
[0039] In one embodiment, the terminal device may be connected with a first network device and a second network device. One of the first network device and the second network device may be a master node (MN) and the other one may be a secondary node (SN) . The first network device and the second network device may use different radio access technologies (RATs) . In one embodiment, the first network device may be a first RAT device and the second network device may be a second RAT device. In one embodiment, the first RAT device is eNB and the second RAT device is gNB. Information related with different RATs may be transmitted to the terminal device from at least one of the first network device and the second network device. In one embodiment, first information may be transmitted to the terminal device from the first network device and second information may be transmitted to the terminal device from the second network device directly or via the first network device. In one embodiment, information related with configuration for the terminal device configured by the second network device may be transmitted from the second network device via the first network device. Information related with reconfiguration for the terminal device configured by the second network device may be transmitted to the terminal device from the second network device directly or via the first network device.
[0040] The terminal device or the network device may have Artificial intelligence (AI) or machine learning capability. It generally includes a model which has been trained from numerous collected data for a specific function, and can be used to predict some information.
[0041] The terminal device or the network device may work on several frequency ranges, e.g. frequency range 1 (FR1) (410 MHz –7125 MHz) , frequency range 2 (FR2) (24.25GHz to 71GHz) , frequency band larger than 100GHz as well as Tera Hertz (THz) . It can further work on licensed / unlicensed / shared spectrum. The terminal device may have more than one connection with the network device under Multi-Radio Dual Connectivity (MR-DC) application scenario. The terminal device or the network device can work on full duplex, flexible duplex and cross division duplex modes.
[0042] The embodiments of the present disclosure may be performed in test equipment, e.g., signal generator, signal analyzer, spectrum analyzer, network analyzer, test terminal device, test network device, or channel emulator.
[0043] The embodiments of the present disclosure may be performed according to any generation communication protocols either currently known or to be developed in the future. Examples of the communication protocols include, but not limited to, the 1G, 2G, 2.5G, 2.75G, 3G, 4G, 4.5G, 5G, 5.5G, 5G-Advanced networks, or 6G networks.
[0044] The term “circuitry” used herein may refer to hardware circuits and / or combinations of hardware circuits and software. For example, the circuitry may be a combination of analog and / or digital hardware circuits with software / firmware. As a further example, the circuitry may be any portions of hardware processors with software including digital signal processor (s) , software, and memory (ies) that work together to cause an apparatus, such as a terminal device or a network device, to perform various functions. In a still further example, the circuitry may be hardware circuits and or processors, such as a microprocessor or a portion of a microprocessor, that requires software / firmware for operation, but the software may not be present when it is not needed for operation. As used herein, the term circuitry also covers an implementation of merely a hardware circuit or processor (s) or a portion of a hardware circuit or processor (s) and its (or their) accompanying software and / or firmware.
[0045] As used herein, the singular forms “a” , “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. The term “includes” and its variants are to be read as open terms that mean “includes, but is not limited to. ” The term “based on” is to be read as “based at least in part on. ” The term “one embodiment” and “an embodiment” are to be read as “at least one embodiment. ” The term “another embodiment” is to be read as “at least one other embodiment. ” The terms “first, ” “second, ” and the like may refer to different or same objects. Other definitions, explicit and implicit, may be included below.
[0046] In some examples, values, procedures, or apparatus are referred to as “best, ” “lowest, ” “highest, ” “minimum, ” “maximum, ” or the like. It will be appreciated that such descriptions are intended to indicate that a selection among many used functional alternatives can be made, and such selections need not be better, smaller, higher, or otherwise preferable to other selections.
[0047] The terminal device or the network device may have AI or ML capability. It generally includes a model which has been trained from numerous collected data for a specific function, and can be used to predict some information.
[0048] As used herein, a model may be equivalent to at least one of the following: an AI / ML model, an ML model, an AI model, a data-driven, a data processing model, an algorithm, a functionality, a procedure, a method, a process, an entity, a function, a feature, a feature group, a model identifier (ID) , an ID, a functionality ID, a configuration ID, a scenario ID, a site ID, an area ID, an associated ID, or a dataset ID. As a result, the above terms may be used interchangeably. An “ID” may refer to an identifier, an identity, an identification, etc.
[0049] In some embodiments, the model may be represented by or associated with a channel, a resource, a resource set, a reference signal (RS) resource, an RS resource set, an RS port, a set of RS ports, an RS port ID, or a set of RS port IDs.
[0050] In some embodiments, the model may comprise a set of weights values that may be learned during training, e.g., for a specific architecture or configuration, where a set of weights values may also be called a parameter set.
[0051] In some embodiments, the model may be used to predict a target cell, or measurements of a set of beams of a set of candidate cells in future based on at least historical measurements (e.g., layer 1 (L1) -reference signal received power (RSRP) , L1-signal to interference plus noise ratio (SINR) ) of a set of beams of a set of candidate cells.
[0052] In some embodiments, an input of the AI / ML model (i.e., AI input) may refer to the input of a model and indicate data inputted into the model, which may be equivalent to data.
[0053] In some embodiments, an output of AI / ML model (i.e., AI output) may refers to the output of a model and indicate result (s) outputted by the model, which is equivalent to label / data.
[0054] In some embodiments, “ground truth” , “ground truth label” , “ground truth label of data” , “input label” , “input data” , “label” and “data” can be used interchangeably.
[0055] In some embodiments, “UE” , “terminal device” , “target UE” , “UE deployed with AI / ML model” , and “UE with UE-side model” can be used interchangeably.
[0056] In some embodiments, a ground truth label of data (or ground-truth label) for monitoring or training the ML model (i.e., AI output) may refers to the authoritative, accepted data, or true answer or outcome for AI / ML model.
[0057] In some embodiments, “model training” and “model developing” can be used interchangeably; and “model updating” , “mode fine-tuning” , and “model retraining” can be used interchangeably.
[0058] In some embodiments, the ground truth can be interpreted as actual / factual (i.e. actual / factual measured) data / values / results / collections / parameters, which can be used as reference, compared to prediction or inference.
[0059] In some embodiments, the positioning reference unit (PRU) is a normal terminal device with known location at some network device (e.g., a location server or gNB) and can report their measurements to a location server.
[0060] In some embodiments, the term “network-side (NW-side) additional conditions” can be used interchangeably with any of “additional conditions at network side” , “TRP’s additional conditions” , “TRP-side additional conditions” , “gNB’s additional conditions” , “gNB-side additional conditions” , etc.
[0061] In some embodiments, the term “UE-side additional conditions” can be used interchangeably with any of “PRU-side additional conditions” , “additional conditions at UE side” , “additional conditions at PRU side” , “additional conditions at terminal devices” , etc.
[0062] AI / ML techniques play a significant role in enhancing the accuracy and reliability of positioning, which is particularly useful in indoor environments where global position system (GPS) signals might be weak or unavailable.
[0063] An AI / ML model may be deployed at a terminal device (such as a UE) , a network device (such as one or more gNBs or TRPs) , or a core network entity (such as a location management function (LMF) ) . The AI / ML model may be used for positioning, e.g. determining a positon (or location) of a UE. Some cases (case 1, case 2b, and case 3b below) are discussed as direct AI / ML positioning, and some other cases (case 2a, and case 3a below) are discussed as AI / ML assisted positioning: · Case 1: UE-based positioning with UE-side model, direct AI / ML positioning. · Case 2b: UE-assisted / LMF-based positioning with LMF-side model, direct AI / ML positioning. · Case 3b: NG-RAN node assisted positioning with LMF-side model, direct AI / ML positioning. · Case 2a: UE-assisted / LMF-based positioning with UE-side model, AI / ML assisted positioning. · Case 3a: NG-RAN node assisted positioning with gNB-side model, AI / ML assisted positioning.
[0064] An enhancement for AI / ML based positioning accuracy has been agreed as a work item (WI) in release 19. An AI / ML model can be deployed at UE side, gNB side, or LMF side. A model input may be integrated information of timing, power and phase, such as delay of path (DP) , power delay profile (PDP) , or channel impulse response (CIR) . A model output may be a UE location (i.e., direct AI / ML positioning) or an intermediate measurement (i.e., AI / ML assisted positioning) .
[0065] For an AI / ML assisted positioning, the candidate output may include a line of sight (LOS) / non-line of sight (NLOS) indicator, angle information, or timing information like reference signal time difference (RSTD) , downlink reference signal time of arrival (DL-RTOA) , or UE Rx-Tx time difference for DL positioning, or uplink reference signal time of arrival (UL-RTOA) or gNB Rx-Tx time difference for UL positioning.
[0066] For a model targeting for positioning cases specified in 3GPP, it has demonstrated that the generalization of a model affects the positioning accuracy heavily. Therefore, the consistency between model training and model inference should be considered, that is, the data for inference should be consistent with the data for training this model. The additional conditions at network side may be changed, for example, the additional conditions when generating inference data are different from that when generating training data.
[0067] Ensuring consistency between model training and model inference can significantly improve the accuracy of model inference. For UE-side model, if the dataset for training the model is generated by PRU, it is necessary to align UE’s additional conditions during inference data collection to PRU’s additional conditions during training data collection, otherwise the positioning accuracy may be poor due to the inconsistency. However, UE-side additional conditions may be regarded as proprietary information that should not be disclosed to other entities. In this event, a solution for eliminating the inconsistency of UE-side additional conditions should be proposed.
[0068] Embodiments of the present disclosure provide a solution of communication. In the solution, a model cluster is trained based on a training dataset associated with a set of PRUs, the model cluster may include a plurality of individual models, and different individual models may be trained from different training data which is generated by different PRUs. As such, the model cluster can be used to assist for eliminating the inconsistent UE-side additional conditions between training and inference. Principles and implementations of the present disclosure will be described in detail below with reference to the figures.
[0069] FIG. 1A illustrates an example communication network 100 in which some embodiments of the present disclosure can be implemented. The communication network 100 may also be called as a network environment, a network system, a communication environment, a communication system, or the like, the present disclosure does not limit for this aspect. The communication network 100 includes a terminal device 110, multiple network devices 120-1 to 120-N, a location server 130, and a set of PRUs 140 (shown as PRU 140-1 to PRU 140-M, where M is a positive integer) . For example, the location server 130 may be an LMF.
[0070] In the present disclosure, there may be at least one positioning model deployed at the terminal device 110, for example, the terminal device 110 may also be referred to as a target terminal device or a target UE or a target device.
[0071] The multiple network devices 120-1 to 120-N (N is a positive integer, e.g. N≥3) may be separately or collectively be referred to as a network device 120, which may be a gNB or a TRP. For example, N may equal to 18 or another integer.
[0072] In some examples, one of the network devices 120-1 to 120-N may be a serving network device (e.g., a serving gNB) of the terminal device 110, which can control and manage other network devices 120 (e.g., one or more TRPs) . In some other examples, there may be an independent serving gNB of the terminal device 110 which is different from any of the multiple TRPs.
[0073] In some examples, a location of the PRU 140 is known, and the PRU 140 can perform positioning measurements and report these measurements to a location server 130 such as the LMF. In some implementations, the PRU 140 can communicate with the location server 130 in a manner similar as that of the terminal device 110 and the location server 130, e.g., according to any proper communication protocol, such as LPP. It is to be noted that the PRU may be with a different name, such as a positioning-assisted-UE, and the present disclosure does not limit for this aspect. It should be noted that the number of PRUs is illustrated as M in FIG. 1A, however there may be a plurality of PRUs in the communication network 100, and the present disclosure does not limit a quantity of PRUs.
[0074] In the communication network 100, the network device 120 can communicate / transmit data and control information to the terminal device 110, and the terminal device 110 can also communicate / transmit data and control information to the network device 120. A link from the network device 120 to the terminal device 110 is referred to as a DL, while a link from the terminal device 110 to the network device 120 is referred to as a UL. DL may comprise one or more logical channels, including but not limited to a Physical Downlink Control Channel (PDCCH) and a Physical Downlink Shared Channel (PDSCH) . UL may comprise one or more logical channels, including but not limited to a Physical Uplink Control Channel (PUCCH) and a Physical Uplink Shared Channel (PUSCH) . As used herein, the term “channel” may refer to a carrier or a part of a carrier consisting of a contiguous set of resource blocks (RBs) on which a channel access procedure is performed in shared spectrum.
[0075] In the communication network 100, the terminal device 110 can communicate with the location server 130 according to any proper communication protocol, such as an LTE positioning protocol (LPP) . In the communication network 100, the network device 120 can communicate with the location server 130 according to any proper communication protocol, such as an NR positioning protocol A (NRPPa) . It is to be understood that other protocol may also be applied and will not be listed herein.
[0076] Embodiments of the present disclosure can be applied to any suitable scenarios. For example, embodiments of the present disclosure can be implemented at reduced capability NR devices. Alternatively, embodiments of the present disclosure can be implemented in one of the followings: NR multiple-input and multiple-output (MIMO) , NR sidelink enhancements, NR systems with frequency above 52.6GHz, an extending NR operation up to 71GHz, narrow band-Internet of Thing (NB-IOT) / enhanced Machine Type Communication (eMTC) over non-terrestrial networks (NTN) , NTN, UE power saving enhancements, NR coverage enhancement, NB-IoT and LTE-MTC, Integrated Access and Backhaul (IAB) , NR Multicast and Broadcast Services, or enhancements on Multi-Radio Dual-Connectivity.
[0077] It is to be understood that the numbers of devices (i.e., the terminal devices 110, the network device 120, and the PRU 140) and their connection relationships and types shown in FIG. 1A are only for the purpose of illustration without suggesting any limitation. The communication network 100 may include any suitable numbers of devices adapted for implementing embodiments of the present disclosure. The communication network 100 may include one or more entities which are not shown in FIG. 1A.
[0078] It is to be understood that although the terminal device 110 and the PRU 140 is illustrated as a mobile phone in FIG. 1A, the type of the terminal device 110 or the PRU 140 can be another type and the present disclosure does not limit for this aspect.
[0079] For AI / ML-assisted positioning, a “single-TRP construction” and a “multi-TRP construction” are being discussed. Single-TRP construction: the input of the ML model is the channel measurement between the target UE and a single TRP, and the output of the ML model is for the same pair of UE and TRP, which is shown in FIG. 1D and FIG. 1E. Multi-TRP construction: the input of the ML model contains N sets of channel measurements between the target UE and N (N>1) TRPs, and the output of the ML model contains the UE location or N sets of values, one for each of the N TRPs, which is shown in FIG. 1B and FIG. 1C.
[0080] In some cases, three constructions may be evaluated for the AI / ML assisted positioning: Single-TRP, same model for N TRPs; Single-TRP, N models for N TRPs; and Multi-TRP (i.e., one model for N TRPs) .
[0081] In some cases, there may be N TRPs (TRP 0, TRP 1, …, TRP (N-1) ) used for AI / ML based positioning, and direct AI / ML positioning (FIG. 1B) and AI / ML assisted positioning (FIGS. 1C-1E) may be evaluated. FIG. 1B illustrates an example schematic of direct AI / ML positioning with an output is the UE location. FIG. 1C illustrates an example schematic of AI / ML assisted positioning with multi-TRP construction for model input, FIG. 1D illustrates an example schematic of AI / ML assisted positioning with single-TRP construction for model input and one same model for N TRPs, and FIG. 1E illustrates an example schematic of AI / ML assisted positioning with single-TRP construction and N different models for N TRPs.
[0082] As shown in FIGS. 1C-1E, the AI / ML assisted positioning can be applied using input data (such as CIR, PDP, or DP) associated with one single TRP or multiple TRPs. For the former scenario, N models with different parameters or a single model may be deployed to estimate time information (e.g., time of arrival (TOA) ) for N TRPs, which may be regarded as a distributed model on each TRP. In the latter scenario, a single comprehensive model (i.e. a centralized model) utilizes the data from multiple TRPs as the input and produces the multiple TOAs corresponding to the multiple TRPs.
[0083] In the present disclosure, “AI / ML model” is a data driven algorithm that applies AI / ML techniques to generate a set of outputs based on a set of inputs; “data collection” is a process of collecting data by the network nodes, management entity, or UE for the purpose of AI / ML model training, data analytics and inference; “model training” is a process to train an AI / ML Model by learning the input / output relationship in a data driven manner and obtain the trained AI / ML Model for inference; “model inference” is a process of using a trained AI / ML model to produce a set of outputs based on a set of inputs; and “UE side model” means an AI / ML model whose inference is performed entirely at the UE.
[0084] In the present disclosure, an entity for transmitting a positioning reference signal (PRS) in air interface may be interchangeably used with one of: a TRP, a TP, a gNB, a network device, a PRS, a reference TRP, a neighbor TRP, a reference PRS, a reference entity, etc., and the present disclosure does not limit for this aspect.
[0085] For an AI / ML-enabled feature, additional conditions refer to any aspects that are assumed for the training of the model but are not a part of UE capability for the AI / ML-enabled feature. It does not imply that additional conditions are necessarily specified. Additional conditions can be divided into two categories: NW-side additional conditions and UE-side additional conditions.
[0086] For inference for UE-side models, to ensure consistency between training and inference regarding NW-side additional conditions, the following options can be taken as potential approaches: -Model identification to achieve alignment on the NW-side additional condition between NW-side and UE-side; -Model training at NW and transfer to UE, where the model has been trained under the additional condition; -Information and / or indication on NW-side additional conditions is provided to UE; -Consistency assisted by monitoring (by UE and / or NW, the performance of UE-side candidate models / functionalities to select a model / functionality) ; -Other approaches are not precluded.
[0087] For AI / ML positioning Case 1, regarding the assistance data provided from LMF to UE, for ensuring consistency between training and inference, for each of the existing assistance data IE of UE-based DL-TDOA and / or UE-based DL-AoD, study whether it should be: (a) explicitly indicated, (b) implicitly indicated and / or (c) other. For example, the additional conditions at NW side may indicate some or all information that listed in Table 1 below. Table 1
[0088] In order to obtain a reliable result from the model output, the consistency between inference data and training data should be ensured, which may require that a same or similar environment, e.g., additional conditions (AC) at UE side and NW side, should be kept during training data collection and inference data collection. For UE-side model, any change in NW-side additional conditions may be a source of inconsistency for the model. Therefore, aligning the NW-side additional conditions among training and inference is important to ensure the consistency for UE side model.
[0089] In the present disclosure, UE-side additional conditions (AC) are considered for the consistency between model training and model inference for UE-side model, e.g., case 1 and case 2a mentioned above, if the training dataset is generated by PRU and the inference data is generated by target UE. For model training, UE-side additional conditions are those of the PRUs for generating at least the measurements, which may be referred to as PRU-side additional conditions for generating training data. For model inference, UE-side additional conditions are those of the target UE for location or intermediate measurement determination, which may be referred to as UE-side additional conditions for generating inference data.
[0090] The PRU-side additional conditions or the UE-side additional conditions may relate to proprietary information that should not be provided / disclosed to other entities. In some implementations, the PRU-side additional conditions or the UE-side additional conditions may include some or all of the following: -implementation algorithm to identify the path from the channel measurement, e.g., take peak samples of time-domain channel measurement as paths, take the strongest Nt’ samples of time-domain channel measurement as paths, use Capon / MUSIC algorithm to extract path from time-domain channel measurement. -timing granularity of sampling in receiver, i.e., T=2k Tc, where Tc is the basic time unit for NR. Different UEs may employ different k to sample the samples in their implementation. -signal processing methods to reduce interference, e.g., mean / median / Gaussian / adaptive / ... filter, noise reduction algorithms, adaptive gain control (spectral analysis / wavelet transforms / …) , signal amplification, … -UE hardware capability and software version for performing mathematical operations / statistical analysis / machine learning model to calculate multipath reception response. -encrypt method to ensure the security of measurement data during transmission and storage within UE internally. -other implementation methods that is evolved.
[0091] In the present disclosure, a network entity may refer to an entity at network side, for example, the network entity may be any of the following: the location server 130 (such as LMF) , a network device (such as gNB, TRP) , a server (such as a UE server) , etc.
[0092] In the present disclosure, a term “model cluster” is used, which may be interchangeably used with any of the following: model group, model set, model bunch, model gang, etc. In some examples, the model cluster includes at least one individual model, such as a plurality of individual models.
[0093] In the present disclosure, the tem “individual model” may refer to a model within a model cluster or a model that belongs to a model cluster, which may be used for some specific purposes. As a non-limited example, some of the following embodiments are discussed based on an assumption that the individual model is a positioning model.
[0094] In the present disclosure, the term “identification information” may refer to any type of information that can be used for identifying an item, and the present disclosure does not limit for this aspect. In the present disclosure, PRU identification information may refer to identification information of a PRU; model identification information may refer to identification information of an individual model, and cluster identification information may refer to identification information of a model cluster. It should be noted that although some examples such as an identifier (ID) etc. are provided in some embodiments below for the identification information, the present disclosure does not limit for this aspect, some other type of information may also be used as the identification information.
[0095] In the present disclosure, a large amount of samples may be used as training data, that is, a training dataset includes a plurality of samples. The may be a plurality of PRUs (which is regarded as a set of PRUs in the present disclosure) that are used for generating the training dataset, that is, the training dataset is associated with the set of PRUs. In some examples, the set of PRUs may be assigned and / or configured by a network entity (such as the location server 130 in FIG. 1A or other network entities) to generate the samples.
[0096] In the present disclosure, a PRU (or each PRU) in the set of PRUs may have proprietary information that is used at least for generating the samples. In some examples, different PRUs in the set of PRUs may use different attributes, such as, different methods to identify the path of reference signal, different timing granularities to sample the channel measurement, different algorithms to process the signal, different hardware capabilities and software versions for signal processing, and / or other aspects for generating the data. In some examples, the PRUs involved in training data generation may transmit the generated training data, e.g., timing / power / phase measurement, to a location server via an LPP message, or to other network entities, or to the terminal device 110 directly. It should be noted that the proprietary information of a PRU that relates to training data collection may not be disclosed or provided.
[0097] Reference is now made to FIG. 2, which illustrates a signalling chart illustrating communication process 200 in accordance with some example embodiments of the present disclosure. The process 200 may involve a terminal device 110, a location server 130, and a PRU 140 as shown in FIG. 1A. It is to be understood that the process 200 may also be applied to another scenario different from that shown in FIG. 1A, the present disclosure does not limit this aspect. In the process 200, the model training is performed at UE side.
[0098] In the process 200, the terminal device 110 obtains a training dataset which includes a plurality of samples at 210. In some implementations, the training dataset is generated at least based on measurements from a set of PRUs as mentioned above, for example, the training dataset is associated with the set of PRUs.
[0099] In some example embodiments, the set of PRUs that involved in training data collection may transmit the generated data to a network entity (such as a location server 130 in FIG. 1A or other network entities) . In some examples, a PRU may transmit its generated data (e.g., timing / power / phase measurement) to the location server 130 via an LPP message, as illustrated at 204 in FIG. 2. In addition, the network entity (such as a location server 130 in FIG. 1A or other network entities) may generate the training dataset based on data from the set of PRUs, and then transmit the training dataset to the terminal device 110. As illustrated, the location server 130 may transmit, and the terminal device 110 may receive the training dataset at 205.
[0100] In some examples, when the training dataset including a plurality of samples is transmitted from the network entity (such as a location server 130 in FIG. 1A or other network entities) to the terminal device 110, each sample should be indicated with generating information associated with associated PRU (s) . For example, for a specific sample such as SAMPLE-A, which PRU or source generates the SAMPLE-Ashould be indicated to the terminal device 110. For example, the indication of the PRU identification information may be implemented explicitly or implicitly.
[0101] In some embodiments, PRU identification information of a PRU may be provided together with a sample. For example, the PRU identification information may be included in a message from the location server 130 to the terminal device 110, where the message may be an LPP NR-PRU-DL-Info message or an LPP NR-DL-TDOA-ProvideAssistanceData message or another message. In some examples, the PRU identification information may be an ID (such as a device ID, etc. ) or other type of information of the PRU. For example, the PRU identification information of a PRU may be a 5G Globally Unique Temporary Identifier (5G-GUTI) , a Subscriber Concealed Identifier (SUCI) , an International Mobile station Equipment Identity (IMEI) , an IMEI software version (IMEISV) , or a paging early indication (PEI) of the PRU. For example, samples that are generated by a same PRU may be provided to the terminal device 110 together, e.g., with shared PRU identification information. For instance, the network entity may classify samples that are generated by a sample PRU into a sample subset, and transmit the sample subset to the terminal device 110. In some examples, the PRU identification information may be a virtual identification of the PRU, and the virtual identification may be assigned or allocated by a network entity such as a location server 130. For example, similar PRUs (e.g., multiple PRUs in a same UE vendor may be regarded similar) may be assigned with a same virtual identification. For example, samples generated by a same PRU or by similar PRUs (e.g., with same or similar additional conditions) may have same PRU identification information such as a same virtual identification.
[0102] In some example embodiments, the set of PRUs that involved in training data collection may transmit the generated data to the terminal device 110 directly, as illustrated at 203 in FIG. 2. In some examples, the terminal device 110 may receive data from the set of PRUs, and then determine or generate the training dataset based on data from the set of PRUs at 210.
[0103] In the process 200, the terminal device 110 determine at least one individual model within a model cluster at 220. In some example embodiments, at least part of the training dataset may be used for training an individual model.
[0104] In some implementations, the model cluster may be associated with the set of PRUs, and the set of PRUs involved in training data collection for a same model cluster has a feature of at least one of the following: being provided by a same vendor, being located within a same area, having same or similar capabilities at least for training data collection, having same or similar additional conditions, or having a same sign that is marked by a network entity (such as the location server 130 or other network entities) . In some examples, the set of PRUs may be located in a same area, for example, configured with same assistanceDataValidityArea in LPP NR-DL-TDOA-ProvideAssistanceData message, or different assistanceDataValidityArea with overlapped cell list.
[0105] In some implementations, different individual models within (or belonging to) a same model cluster may have a same model structure, but with different model parameters, e.g., at least one model transfer. In some implementations, different individual models in a same model cluster may be trained based on data from same or different PRUs or PRU sets.
[0106] In some examples, different individual models may be trained based on data from different subsets of PRUs, each subset of PRUs include at least one PRU, and different subsets may or may not have an intersection. Take a first individual model and a second individual model for example, the first individual model may be trained based on data from a first subset of PRUs (for brevity, it is assumed that the first subset includes PRU-1) , and the second individual model may be trained based on data from a second subset of PRUs. For instance, the second subset may not have an intersection with the first subset, e.g., the second subset includes one or more PRUs, such as PRU-2 (or PRU-2 and PRU-3, etc. ) . For instance, the second subset may have an intersection with the first subset, e.g., the second subset includes PRU-1, PRU-2, and optional one or more than one other PRU. In some implementations, the first / second subset of PRUs may be replaced by first / second data with first / second PRU identification information. It should be noted that the first subset of PRUs and the second subset of PRUs are the different sets, i.e., their elements are not the same. For example, a number of elements in the first subset of PRUs does not equal to a number of elements in the second subset of PRUs. For example, an element in the first subset of PRUs does not belong to the second subset of PRUs, or an element in the second subset of PRUs does not belong to the first subset of PRUs.
[0107] In some examples, the first individual model may be trained based on samples generated from first PRU identification information, and the second individual model may be trained based on samples generated from second PRU identification information (optional and the first PRU identification information) .
[0108] In some implementations, the model cluster includes a plurality of individual models. In some example embodiment, a maximum value for the number of the plurality of individual models is associated with a number of PRUs in the set of PRUs. In some examples, a maximum number of the plurality of the individual models in a same model cluster may be determined based on a number of PRUs. For example, if the number of PRU is n, then the maximum number of individual models is 2n-1, which is determined / calculated by:
[0109] In some implementations, a network entity (such as the location server 130 or other network entities) may transmit, to the terminal device 110, information for determining the number of the at least one individual model to be trained. As illustrated, the location server 130 may transmit, and the terminal device 110 may receive, information about the number of the at least one individual model for a model cluster or about a range of the number of the at least one individual model for a model cluster at 215.
[0110] In some implementations, the terminal device may receive, from a network entity (such as the location server 130) , information about a number of the plurality of individual models in the model cluster. In some examples, a hard value of the number may be indicated, which may be taken as an actual number of individual models within the model cluster to be trained. In some examples, the network entity such as the location server 130 may determine the hard value which is smaller than the maximum number of individual models for a model cluster.
[0111] In some examples, the network entity such as the location server 130 may provide a specific PRU set with a number represented by m, which meets m ≤ n. In some examples, the actual number (e.g., hard value) may be determined based on both m and n. For example, the actual number is 2n-m, which is determined or calculated by:
[0112] For example, the number m may be determined by the network entity (such as the location server 130 or other network entities) based on UE reported capability and / or based on the data quality generated from each PRU. In some instances, the number m may be dynamically changed.
[0113] For instance, if n=4, and a provided specific PRU set is {PRU-1} , i.e., m=1, then the actual number is 24-1=8. For instance, 8 individual models may be trained based on training data generated from {PRU-1} , {PRU-1 and PRU-2} , {PRU-1 and PRU-3} , {PRU-1 and PRU-4} , {PRU-1, PRU-2, and PRU-4} , {PRU-1, PRU-3, and PRU-4} , {PRU-1, PRU-2, and PRU-3} , and {PRU-1, PRU-2, PRU-3, and PRU-4} respectively.
[0114] For instance, if n=4, and the provided specific PRU set is {PRU-1, PRU-2} , i.e., m=2, then the actual number is 24-2=4. For instance, 4 individual models may be trained based on training data generated from, {PRU-1 and PRU-2} , {PRU-1, PRU-2, and PRU-4} , {PRU-1, PRU-2, and PRU-3} , and {PRU-1, PRU-2, PRU-3, and PRU-4} respectively.
[0115] In some examples, the network entity such as the location server 130 may provide more than one specific PRU set, in some examples, the actual number may be determined based on respectively numbers for corresponding specific PRU set and a number of intersections between different PRU sets. In some examples, the actual number is: a sum of numbers of trained individual models for each PRU set, then subtract a number of an overlapped trained individual models.
[0116] For instance, if n=4, and the provided specific PRU sets are {PRU-1, PRU-2} (i.e., m=2) and {PRU-1} (m=1) , then the actual number is 24-2+24-1-4=8, where 24-2 is a number of trained individual models based on {PRU-1, PRU-2} , 24-1 is a number of trained individual models based on {PRU-1} , and “4” is the number of overlapped trained individual models.
[0117] For instance, if n=4, and the provided specific PRU sets are {PRU-1, PRU-2} (i.e., m=2) and {PRU-3} (m=1) , then the actual number is 24-2+24-1-2=10, where 24-2 is a number of trained individual models based on {PRU-1, PRU-2} , 24-1 is a number of trained individual models based on {PRU-3} , and “2” is the number of overlapped trained individual models.
[0118] In some implementations, the terminal device may receive, from a network entity (such as the location server 130) , information about a number of the plurality of individual models in the model cluster. In some examples, a range (e.g., a soft range) of the number may be indicated, and then the terminal device 110 may determine the number within the range. For example, the number determined by the terminal device 110 may be an actual number of individual models within the model cluster to be trained. For example, the determination of the actual number by the terminal device 110 may be made depend on the UE capability at least for model training.
[0119] In some examples, the network entity such as the location server 130 may provide a specific PRU set with a number represented by m, which meets m ≤ n. In some examples, the network entity such as the location server 130 may determine that an upper boundary of the range equals to or is smaller than the maximum number, i.e. 2n-1 , and a lower boundary of the range equals to or is larger than the number 2n-m. For instance, n may be a number of all PRUs, and m is a number of PRUs in a specific PRU set, as mentioned above.
[0120] For instance, if n=4, and a provided specific PRU set is {PRU-1} , i.e., m=1, then the range may be 8 to 15, in other words, a lower boundary (minimum value) of the range is 24-1=8, and an upper boundary (maximum value) of the range is 24-1=15.
[0121] For instance, if n=4, and a provided specific PRU set is {PRU-1, PRU-2} , i.e., m=2, then the range may be 4 to 15, in other words, a lower boundary (minimum value) of the range is 24-2=4, and an upper boundary (maximum value) of the range is 24-1=15.
[0122] In some examples, the network entity such as the location server 130 may provide more than one specific PRU set, in some examples, a lower boundary of the range equals to or is larger than the number that is determined based on respectively numbers for corresponding specific PRU set and a number of intersections between different PRU sets. In some examples, a lower boundary of the range equals to or is larger than: a sum of numbers of trained individual models for each PRU set, then subtract a number of an overlapped trained individual models.
[0123] For instance, if n=4, and the provided specific PRU sets are {PRU-1, PRU-2} (i.e., m=2) and {PRU-1} (m=1) , then a lower boundary (minimum value) of the range may be 24-2+24-1-4=8, and an upper boundary (maximum value) of the range may be 15.
[0124] For instance, if n=4, and the provided specific PRU sets are {PRU-1, PRU-2} (i.e., m=2) and {PRU-3} (m=1) , then a lower boundary (minimum value) of the range may be 24-2+24-1-2=10, and an upper boundary (maximum value) of the range may be 15.
[0125] It should be noted that n and m mentioned above may refer to a number of PRUs, or may refer to a number of pieces of PRU identification information (e.g., ID, or virtual identifier, etc. ) .
[0126] In some implementations, the terminal device 110 may determine the actual number of individual models within the model cluster to be trained, e.g., without any indication from the network entity. In some examples, the UE capability at least for model training may be considered by the terminal device 110 while determining the actual number.
[0127] Specifically, the terminal device 110 may determine at least one individual model within a model cluster, and a number of the at least one individual model may be the actual number indicated or determined as above.
[0128] In some implementations, an individual model may have model identification information, and a model cluster may have cluster identification information.
[0129] In some examples, the cluster identification information of a model cluster may be a combination or a function of PRU identification information of all PRUs that involved in training data collection for all individual models in the model cluster. In some examples, the cluster identification information may be model identification information of a main individual model in the model cluster or a function thereof. In some examples, the cluster identification information may be an ID, such as a model ID, a functionality ID, an associated ID for ensuring consistency of network-side additional conditions, a configuration ID, an area ID, a data set ID, etc., and the present disclosure does not limit for this aspect.
[0130] For example, the main individual model may be a model which is determined based on training data generated by all PRUs in the set of PRUs, or a model which is determined based on training data generated by a specific PRU or a specific PRU set (this specific PRU or specific PRU set may have same or similar additional conditions with the terminal device 110) , or a model that indicted by a network entity (such as the location server 130 or other network entities) .
[0131] In some examples, the model identification information of an individual model may be a combination or a function of PRU identification information of PRUs that involved in training data collection for this individual model. In some examples, the model identification information may be an extension or a function of the cluster identification information. In some examples, the model identification information may be an ID, such as a model ID, a functionality ID, an associated ID for ensuring consistency of network-side additional conditions, a configuration ID, an area ID, a data set ID, etc., and the present disclosure does not limit for this aspect.
[0132] As such, the terminal device 110 may determine the at least one individual model within a model cluster, and also determine the model identification information of the at least one individual model. For example, the at least one trained individual model and associated model identification information may be stored at the terminal device 110 for later use, e.g., for model inference.
[0133] It should be noted that the process 200 is discussed for a model training procedure at the terminal device 110. In some implementations, the model training may be performed at a network entity such as a location server 130 or another network entity (such as a TRP, a gNB, a UE server, etc. )
[0134] In some implementations, a network entity may determine / train at least one individual model within a model cluster, and transmit the at least one trained individual model to the terminal device 110. In some example embodiments, the number (an actual number) of the at least one individual model that trained by the network entity may be determined by the network entity, for example, the actual number may be within a number range with a lower boundary 1 (or another value) and an upper boundary 2n-1.
[0135] In some example embodiments, the network entity may transfer some or all trained individual models within the model cluster to the terminal device 110. In some example embodiments, model identification information may also be provided from the network entity to the terminal device 110.
[0136] For example, the model identification information of an individual model may be based on PRU identification information of the PRU (s) that involved in training data collection for this individual model, that is, this individual model is trained based on training data generated by this / these PRUs. For example, the model identification information of an individual model may be virtual identification information of the PRU (s) that involved in training data collection for this individual model, where virtual identification information for a PRU may be assigned by the network entity.
[0137] According to embodiments with reference to FIG. 2, at least one individual model within a model cluster can be trained by the terminal device 110 or by a network entity, and model identification information may be further determined for each individual model. As such, the model cluster can be used to assist for eliminating the inconsistency of UE-side additional conditions between training and inference.
[0138] FIG. 3 illustrates a signalling chart illustrating a communication process 300 for model inference in accordance with some example embodiments of the present disclosure. The process 300 may involve a terminal device 110 and a location server 130 as shown in FIG. 1A. It is to be understood that the process 300 may also be applied to another scenario different from that shown in FIG. 1A, the present disclosure does not limit this aspect.
[0139] In the process 300, the terminal device 110 activates at least one individual model within a model cluster 310. The model cluster 310 may include a plurality of individual models which are generated based on a training dataset. In some examples, details about the model cluster and an individual model may refer to those discussed above, and thus will not be repeat herein.
[0140] In some implementations, whether a model cluster should be activated is determined by a network entity such as the location server 130 or other network entities. As illustrated, the location server 130 may transmit, and the terminal device may receive, a second message at 305. In some example embodiments, the second message includes an indication indicating that the model cluster should be activated.
[0141] In some example embodiments, whether an individual model within the activated model cluster should be activated is determined by the terminal device 110. In some examples, the terminal device 110 may active at least one individual model within the activated model cluster that is activated based on the second message.
[0142] In some examples, the activated at least one individual model may include: a plurality of individual models (e.g., all individual models) within the model cluster, one or more individual models that are associated with at least one specific subset of PRUs, or a main individual model among the plurality of individual models.
[0143] For example, the terminal device 110 may determine to activate every individual model within the model cluster. For example, the terminal device 110 may determine to activate one or more individual models which are trained at least based on data generated by one or more specific PRU sets, where the one or more specific PRU sets may be determined / selected by the terminal device 110. For example, the terminal device 110 may determine to activate at least the main individual model (optional and one or more other individual models) .
[0144] In some instances, the main individual model is one of: a model which is determined based on training data generated by all PRUs in the set of PRUs, a model which is determined based on training data generated by a specific PRU, or a model that indicted by a network entity (such as the location server 130) . For instance, the specific PRU may be that with same or similar additional conditions with the terminal device 110. For instance, the second message may include an indication of the main individual model.
[0145] In the process 300, the terminal device 110 performs model inference using the activated at least one individual model to determine an inference result at 320.
[0146] In some implementations, a model input (i.e., inference data) is input into each of the activated at least one individual model, and accordingly at least one model output may be determined, in addition, the inference result may be generated based on the at least one model output.
[0147] In some examples, multiple individual models are activated, and accordingly multiple model outputs can be obtained. In some examples, the inference result may be an integration of the multiple model outputs. For example, an integration mode may be used for determining the inference result.
[0148] In some examples, the integration mode indicates one of: a weighted sum of the multiple model outputs, or an average of the multiple model outputs. For example, the inference result may be an average or a weighted sum of the multiple model outputs.
[0149] In some instances, a weight for an individual model may be determined by the terminal device 110 or may be indicated by a network entity such as the location server 130.
[0150] For instance, a weight for a specific individual model may be determined based on: a ratio of a number of the training samples for training the specific individual model to a total number of samples in the training dataset for training all individual models within the model cluster.
[0151] For instance, a weight for a specific individual model may be determined based on: a ratio of a number of samples generated by specific PRUs to a total number of samples in the training dataset. For example, the specific PRUs may be those generating data for training the specific individual model. For example, the specific PRUs may be those having same or similar UE-side additional conditions with the terminal device 110 (if the model training is performed at the terminal device 110) .
[0152] For instance, a weight for a specific individual model may be determined based on: a result of model monitoring for the specific individual model. Details on the result of model monitoring will be provided below with reference to FIG. 5.
[0153] In the process 300, the terminal device 110 transmits a first message to the location server 130 at 330. In some implementations, the first message may be an LPP message, such as an LPP Provide Location Information or other LPP message. In some implementations, the first message includes the inference result.
[0154] In some example embodiments, the first message may further include an indication of the at least one individual model. As such, the location server 130 can know which individual model (s) is / are activated for model inference. In some examples, the first message may include model identification information of the activated at least one individual model, or PRU identification information of at least one PRU associated with the activated at least one individual model. For example, the first message may also include an indication of the model cluster, e.g., cluster identification information.
[0155] In some example embodiments, the first message may further include an indication of the integration mode. In some examples, the first message may indicate that the inference result is determined based on an average of the multiple model outputs (e.g., averaged mode) . In some examples, the first message may indicate that the inference result is determined based on a weighted sum of the multiple model outputs (e.g., weighted mode) , in this case, the first message may optionally further include weights for the multiple model outputs. In some other example embodiments, a default integration mode may be defined and the default integration mode is used by the terminal device 110, in this case, the indication of the integration mode may be omitted. As such, the location server 130 can know how the inference result is determined in model inference.
[0156] In some implementations, whether a model cluster should be activated and whether an individual model should be activated are both determined by a network entity such as the location server 130 or other network entities. As illustrated, the location server 130 may transmit, and the terminal device may receive, a second message at 305. In some example embodiments, the second message includes: an indication indicating that the model cluster should be activated, and an indication indicating that the at least one individual model within the model cluster should be activated.
[0157] In some examples, the second message may include model identification information of the at least one individual model. In some examples, the second message may include an activation time for the at least one individual model. In some instances, the location server 130 may determine that the at least one individual model is to be activated, for example, the determination of the at least one individual model at the location server 130 may be similar with that made at the terminal device 110, thus details on which will not be repeat herein.
[0158] In some examples, the second message may include an indication for an activation mode, where multiple activation modes may be predefined or preconfigured. For example, if the second message indicates activation mode 1, then the terminal device 110 may know that all individual models within the model cluster should be activated. For example, if the second message indicates activation mode 2, then the terminal device 110 may know that individual model (s) trained based on data generated by one or more specific PRU sets should be activated. For example, if the second message indicates activation mode 3, then the terminal device 110 may know that the main individual model within the model cluster should be activated.
[0159] Accordingly, the terminal device 110 may activate the at least one individual model based on the first message.
[0160] In some examples, the first message may further indicate an integration mode. For example, the first message may further include the weights for multiple model outputs, e.g., if the integration mode is a weighted mode. As such, the terminal device 110 may determine the inference result based on the integration model indicated by the first message.
[0161] Accordingly, the terminal device 110 may perform the model inference to determine the inference result, and may further provide the inference result to the location server 130 via the first message.
[0162] FIG. 4 illustrates an example 400 for determining an inference result in accordance with some example embodiments of the present disclosure. As illustrated, the model cluster may include individual model 1, individual model 2, …, individual model x. The individual model 1 and individual model x are activated, and other individual models (including the individual model 2) are not activated. The model input 410 is input into the activated individual model 1 and individual model x respectively, to obtain model output 402 and model output 404. In addition, the integrator 420 may integrate the model output 402 and the model output 404 to determine the inference result 450.
[0163] According to embodiments with reference to FIGS. 3-4, at least one individual model within a model cluster may be activated for model inference. As such, the model cluster can be used to assist for eliminating the inconsistency between training and inference.
[0164] FIG. 5 illustrates an example for model monitoring 500 in accordance with some example embodiments of the present disclosure. In the present disclosure, model monitoring may also refer to model performance monitoring, model performance detection, model performance determination, model perform evaluation, etc., and the present disclosure does not limit for this aspect. In the present disclosure, a monitoring metric may also refer to a performance metric, an evaluation metric, a metric, etc., and the present disclosure does not limit for this aspect.
[0165] In some implementations, the model monitoring may be performed for an individual model and / or for a model cluster. In some examples, the model performance monitoring may be performed for the model cluster, and thus whether the model cluster is reliable is determined. In some embodiments, a monitoring metric for an individual model can be determined or calculated. In some embodiments, a monitoring metric for a model cluster can be determined or calculated.
[0166] In some examples, the determination of monitoring metrics for an individual model and for a model cluster is performed at the terminal device 110. In some examples, the determination of monitoring metrics for an individual model and for a model cluster is performed at a network entity such as the location server 130 or other entities. In some examples, the determination of a monitoring metric for an individual model is performed at the terminal device 110, and the determination of a monitoring metric for a model cluster is performed at a network entity such as the location server 130 or other entities.
[0167] As illustrated in FIG. 5, an inference result 504 may be obtained, where the inference result 504 is determined based on inference data 502 by model inference 510. In addition, a monitoring metric 550 is determined by model performance monitoring 520, e.g., based on a ground truth 530.
[0168] It should be noted that the model monitoring 500 is applied for an individual model or for a model cluster. For example, if the model monitoring 500 is applied for an individual model, the inference result may be replaced by a model output of the individual model. Accordingly, a monitoring metric for an individual model can be determined.
[0169] In some implementations, a model cluster includes a plurality of individual models, and a motoring metric of the model cluster may be determined based on: an average of one or more monitoring metrics of one or more individual models within the model cluster, or a weighted sum of the one or more monitoring metrics of the one or more individual models within the model cluster. In some examples, the one or more individual models comprise one of: all individual models within the model cluster, at least one individual model associated with a specific subset of PRUs, a predefined number of individual models with model outputs closest to a ground truth, a predefined number of individual models with model outputs furthest to a ground truth, or a main individual models within the model cluster. In some examples, a weight for a motoring metric of an individual model may be related to: a ratio of a number of the training samples for training the specific individual model to a total number of samples in the training dataset for training all individual models within the model cluster, a ratio of a number of samples generated by specific PRUs to a total number of samples in the training dataset, or a weight value indicated by a location server. For instance, the specific PRUs may be those generating data for training the individual model. For example, the specific PRUs may be those having same or similar UE-side additional conditions with the terminal device 110 (if the model training is performed at the terminal device 110) .
[0170] For example, the monitoring metric for the model cluster may be integrated by one of the following ways: Option-1: averaging the metrics of all individual models involved in model monitoring. Option-2: averaging the metrics of individual models which are trained at least based on the data generating by one or more specific PRU sets (or with specific PRU identification information) . Option-3: using the metric of a specific individual model as the metric of the model cluster, where the model output of the specific individual model is closest to the provided ground truth. Option-4: averaging the metrics of multiple individual models, where the model outputs of these individual models are closest to the provided ground truth, and the number of these mentioned individual models is a default value or is predefined. Option-5: using the metric of a specific individual model as the metric of model cluster, where the model output of this individual model is the furthest from the provided ground truth. Option-6: averaging the metrics of multiple individual models, where the model outputs of these individual models are furthest from the provided ground truth, and the number of these mentioned individual models is a default value or is predefined. Option-7: using the metric of the main individual model within the model cluster as the monitoring metric of the model cluster, where the main individual model is defined as discussed above. Option-8: weighted averaging the metrics of multiple individual models involved in model monitoring.
[0171] In some implementations, the monitoring metric may be further used to determine weights for model inference. As mentioned above with reference to FIG. 3, the inference result may be determined based on a weighted sum of multiple model outputs, and the weights may be determined based on a result of model monitoring. As such, whether an individual model is reliable can be determined.
[0172] In some example embodiments, a weight of an individual model (or for a model output of the individual model) with the model cluster may be determined, and the weight is used for determining an inference result in a model inference stage.
[0173] In some examples, the weight of an individual model may be a ratio of a monitoring metric of the individual model to a sum of monitoring metrics of all individual models (e.g., related individual models) within the model cluster. For example, if the monitoring metric is the similarity between the label and model output, and there are three individual models; supposing the metric for individual-model-1, individual-model-2, and individual-model-3 is 0.5, 0.2, and 0.1 respectively, then the weights for individual-model-1, individual-model-2, and individual-model-3 are 0.625, 0.25, 0.125 respectively.
[0174] In some examples, the weight of an individual model may be an inverse order of ratios of respective monitoring metrics to a sum of monitoring metrics of all individual models (e.g., related individual models) within the model cluster. For example, if the monitoring metric is the difference between the provided ground truth and model output, and there are three individual models; supposing the metric for individual-model-1, individual-model-2, and individual-model-3 is 0.5m, 0.2m, and 0.1m respectively, then the ratios are 0.625, 0.25, and 0.125 respectively. Thus, the weights for individual-model-1, individual-model-2, and individual-model-3 are be 0.125, 0.25, 0.625 respectively, which may be determined based on a reverse order of ratios for individual models.
[0175] In some instances, the related individual models are those involved in model performance monitoring, and / or those use same inference (e.g., both measurement and label) for model performance monitoring.
[0176] As such, the model performance monitoring may be performed and a model metric may be determined, therefore, whether the model cluster (and / or individual model) is reliable can be determined.
[0177] According to embodiments with reference to FIGS. 2-5, a solution related to model cluster is provided. In the solution, UE-side additional conditions are considered for the consistency of model training and model inference, and the model cluster can be used to assist for eliminating the inconsistent UE-side additional conditions between training and inference.
[0178] In the present disclosure, some embodiments above are provided with reference to PRU (s) , it should be understood that a same / similar PRU may be replaced by same / similar PRU identification information, and different PRUs may be replaced by different PRU identification information, and the related embodiments are still in the protection scope of the present disclosure.
[0179] It is to be appreciated that the processes described above are only for illustration without any limitation. In some examples, one or more steps may be omitted or combined or modified. In some examples, one or more additional steps may be added. One or more steps in a process may be combined into another process. It is to be understood that some further embodiments may be obtained and are still in the protection scope of the present disclosure.
[0180] FIG. 6 illustrates a flowchart of an example method 600 implemented at a communication device in accordance with some embodiments of the present disclosure. For example, the communication device may be the terminal device 110 in FIG. 1A or a network entity (such as the location server 130 in FIG. 1A or other network entities) .
[0181] At block 610, the communication device obtains a training dataset comprising a plurality of samples, wherein the training dataset is generated at least based on measurements from a set of PRUs. At block 620, the communication device determines at least one individual model within a model cluster based on at least part of the training dataset, wherein a first individual model of the at least one individual model has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model.
[0182] It should be noted that the method 600 may include various other operations which may be performed by the terminal device 110 or a network entity as described above with reference to FIG. 2.
[0183] FIG. 7 illustrates a flowchart of an example method 700 implemented at a terminal device in accordance with some embodiments of the present disclosure. For example, the terminal device may be the terminal device 110 in FIG. 1A.
[0184] At block 710, the terminal device activates at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model. At block 720, the terminal device performs a model inference by using the activated at least one individual model based on a model input, to determine an inference result. At block 730, the terminal device transmits, to a location server, a first message comprising the inference result.
[0185] It should be noted that the method 700 may include various other operations which may be performed by the terminal device 110 as described above with reference to FIGS. 2-6.
[0186] FIG. 8 illustrates a flowchart of an example method 800 implemented at a location server in accordance with some embodiments of the present disclosure. For example, the location server may be the location server 130 in FIG. 1A.
[0187] At block 810, the location server transmits, to a terminal device, a second message comprising an indication that a model cluster should be activated, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model. At block 820, the location server receives, from the terminal device, a first message comprising an inference result associated with model inference of at least one individual model within the model cluster.
[0188] It should be noted that the method 800 may include various other operations which may be performed by the location server 130 as described above with reference to FIGS. 2-6.
[0189] FIG. 9 illustrates a flowchart of an example method 900 implemented at a communication device in accordance with some embodiments of the present disclosure. For example, the communication device may be the terminal device 110 in FIG. 1A or a network entity (such as the location server 130 in FIG. 1A or other network entities) .
[0190] At block 910, the communication device obtains at least one model output of at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model. At block 920, the communication device determines at least one monitoring metric of the at least one individual model or a monitoring metric of the model cluster based on the at least one model output.
[0191] It should be noted that the method 900 may include various other operations which may be performed by the terminal device 110 or a network entity as described above with reference to FIG. 5.
[0192] Details of some embodiments according to the present disclosure have been described with reference to FIGS. 1A-9. Now an example implementation of the device deployed with at least one positioning model will be discussed below.
[0193] In some example embodiments, a communication device comprises circuitry configured to: obtain a training dataset comprising a plurality of samples, wherein the training dataset is generated at least based on measurements from a set of PRUs; and determine at least one individual model within a model cluster based on at least part of the training dataset, wherein a first individual model of the at least one individual model has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model. It should be noted that the communication device comprises circuitry configured to perform various other operations as described above with reference to FIG. 2.
[0194] In some example embodiments, a terminal device comprises circuitry configured to: activate at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; perform a model inference by using the activated at least one individual model based on a model input, to determine an inference result; and transmit, to a location server, a first message comprising the inference result. It should be noted that the terminal device comprises circuitry configured to perform various other operations as described above with reference to FIGS. 3-4.
[0195] In some example embodiments, a location server comprises circuitry configured to: transmit, to a terminal device, a second message comprising an indication that a model cluster should be activated, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; and receive, from the terminal device, a first message comprising an inference result associated with model inference of at least one individual model within the model cluster. It should be noted that the location server comprises circuitry configured to perform various other operations as described above with reference to FIGS. 3-4.
[0196] In some example embodiments, a communication device comprises circuitry configured to: obtain at least one model output of at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; and determine at least one monitoring metric of the at least one individual model or a monitoring metric of the model cluster based on the at least one model output. It should be noted that the communication device comprises circuitry configured to perform various other operations as described above with reference to FIG. 5.
[0197] FIG. 10 illustrates a simplified block diagram of a device 1000 that is suitable for implementing embodiments of the present disclosure. The device 1000 can be considered as a further example implementation of the terminal device or the location server as described above. Accordingly, the device 1000 can be implemented at or as at least a part of the terminal device 110 or the location server 130 as shown in FIG. 1A.
[0198] As shown, the device 1000 includes a processor 1010, a memory 1020 coupled to the processor 1010, a suitable transceiver 1040 coupled to the processor 1010, and a communication interface coupled to the transceiver 1040. The memory 1020 stores at least a part of a program 1030. The transceiver 1040 may be for bidirectional communications or a unidirectional communication based on requirements. The transceiver 1040 may include at least one of a transmitter and a receiver. The transmitter and the receiver may be functional modules or physical entities. The transceiver 1040 has at least one antenna to facilitate communication, though in practice an Access Node mentioned in this application may have several ones. The communication interface may represent any interface that is necessary for communication with other network elements, such as X2 / Xn interface for bidirectional communications between eNBs / gNBs, S1 / NG interface for communication between a Mobility Management Entity (MME) / Access and Mobility Management Function (AMF) / serving gateway (SGW) / user plane function (UPF) and the eNB / gNB, Un interface for communication between the eNB / gNB and a relay node (RN) , or Uu interface for communication between the eNB / gNB and a terminal device.
[0199] The program 1030 is assumed to include program instructions that, when executed by the associated processor 1010, enable the device 1000 to operate in accordance with the embodiments of the present disclosure, as discussed herein with reference to FIGS. 1A-9. The embodiments herein may be implemented by computer software executable by the processor 1010 of the device 1000, or by hardware, or by a combination of software and hardware. The processor 1010 may be configured to implement various embodiments of the present disclosure. Furthermore, a combination of the processor 1010 and memory 1020 may form processing means 1050 adapted to implement various embodiments of the present disclosure.
[0200] The memory 1020 may be of any type suitable to the local technical network and may be implemented using any suitable data storage technology, such as a non-transitory computer readable storage medium, semiconductor-based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory and removable memory, as non-limiting examples. While only one memory 1020 is shown in the device 1000, there may be several physically distinct memory modules in the device 1000. The processor 1010 may be of any type suitable to the local technical network, and may include one or more of general purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs) and processors based on multicore processor architecture, as non-limiting examples. The device 1000 may have multiple processors, such as an application specific integrated circuit chip that is slaved in time to a clock which synchronizes the main processor.
[0201] In summary, embodiments of the present disclosure may provide the following solutions.
[0202] The present disclosure provides a communication device, comprising at least one processor configured to cause the communication device at least to: obtain a training dataset comprising a plurality of samples, wherein the training dataset is generated at least based on measurements from a set of PRUs; and determine at least one individual model within a model cluster based on at least part of the training dataset, wherein a first individual model of the at least one individual model has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model.
[0203] In one embodiment, the communication device as above, two different individual models within a same model cluster have a same model structure with different model parameters.
[0204] In one embodiment, the communication device as above, the first individual model within the model cluster is determined based on first training data associated with the first subset of PRUs, and a second individual model within the model cluster is determined based second training data associated with a second subset of PRUs with or without an intersection with the first subset.
[0205] In one embodiment, the communication device as above, the model cluster is associated with the set of PRUs, and wherein the set of PRUs involved in training data collection for a same model cluster has a feature of at least one of the following: being provided by a same vendor, being located within a same area, having same or similar capabilities at least for training data collection, or having a same sign that is marked by a network entity.
[0206] In one embodiment, the communication device as above, the communication device comprises a terminal device, and wherein the at least one processor is further configured to cause the communication device to: receive, from a network entity, information about a number of the at least one individual model or about a range of the number of the at least one individual model.
[0207] In one embodiment, the communication device as above, the number of the at least one individual model or the range of the number of the at least one individual model is associated with at least one of: a number of PRUs in the set of PRUs, a number of PRUs in the first subset of PRUs, a number of PRUs in a second subset of PRUs, or an intersection of the first subset and the second subset.
[0208] In one embodiment, the communication device as above, a maximum value for the number of the at least one individual model or for the range is associated with a number of PRUs in the set of PRUs.
[0209] In one embodiment, the communication device as above, the at least one processor is further configured to cause the communication device to: determine a number of the at least one individual model based on an implementation of the communication device.
[0210] In one embodiment, the communication device as above, the communication device comprises a network entity, and wherein the at least one processor is further configured to cause the communication device to: transmit, to a terminal device, the at least one individual model within the model cluster, wherein each of the at least one individual model has corresponding model identification information.
[0211] In one embodiment, the communication device as above, PRU identification information of a PRU in the set of PRUs is: an identifier of the PRU, or a virtual identification of the PRU that is allocated by a network entity.
[0212] In one embodiment, the communication device as above, the model identification information of the first individual model is: a combination or a function of PRU identification information of PRUs in the first subset of PRUs, an extension or a function of cluster identification information of the model cluster, or a first ID that is predefined or assigned by a network entity.
[0213] In one embodiment, the communication device as above, the cluster identification information of the model cluster is determined based on at least one of: a combination or a function of PRU identification information of all PRUs in the set of PRUs, model identification information of a main individual model within the model cluster, or a second ID that is predefined or assigned by a network entity.
[0214] In one embodiment, the communication device as above, the main individual model is one of: a model which is determined based on training data generated by all PRUs in the set of PRUs, a model which is determined based on training data generated by a specific PRU, or a model that indicted by a network entity.
[0215] The present disclosure provides a terminal device, comprising at least one processor configured to cause the terminal device at least to: activate at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; perform a model inference by using the activated at least one individual model based on a model input, to determine an inference result; and transmit, to a location server, a first message comprising the inference result.
[0216] In one embodiment, the terminal device as above, the at least one processor is further configured to cause the terminal device to: receive, from the location server, a second message comprising an indication that the model cluster should be activated.
[0217] In one embodiment, the terminal device as above, the activated at least one individual model comprises: a plurality of individual models within the model cluster, one or more individual models that are associated with at least one specific subset of PRUs, or a main individual model among the plurality of individual models.
[0218] In one embodiment, the terminal device as above, the main individual model is one of: a model which is determined based on training data generated by all PRUs in the set of PRUs, a model which is determined based on training data generated by a specific PRU, or a model that indicted by a network entity.
[0219] In one embodiment, the terminal device as above, the first message further comprises: model identification information of the activated at least one individual model, or PRU identification information of at least one PRU associated with the activated at least one individual model.
[0220] In one embodiment, the terminal device as above, the second message further comprises an indication indicating to activate the at least one individual model.
[0221] In one embodiment, the terminal device as above, the second message further comprises at least one of: model identification information of the at least one individual model, or an activation time for the at least one individual model.
[0222] In one embodiment, the terminal device as above, the at least one individual model comprises multiple individual models, and wherein the at least one processor is configured to cause the terminal device to: determine multiple model outputs of the multiple individual models respectively based on the model input; and determine the inference result of the model cluster based on the multiple model outputs using an integration mode.
[0223] In one embodiment, the terminal device as above, the integration mode indicates one of: a weighted sum of the multiple model outputs, or an average of the multiple model outputs.
[0224] In one embodiment, the terminal device as above, a model output is generated by a second individual model that is determined based on training samples in training dataset, and the training samples are associated with a second subset of PRUs, and wherein, a weight for the model output is related to at least one of: a ratio of a number of the training samples to a total number of samples in the training dataset, a ratio of a number of samples associated with the second subset of PRUs to a total number of samples in the training dataset, or a result of model monitoring for the second individual model.
[0225] In one embodiment, the terminal device as above, the first message further comprises the integration mode.
[0226] In one embodiment, the terminal device as above, the second message further comprises information of an integration mode which is used for determining the inference result.
[0227] The present disclosure provides a location server, comprising at least one processor configured to cause the location server at least to: transmit, to a terminal device, a second message comprising an indication that a model cluster should be activated, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; and receive, from the terminal device, a first message comprising an inference result associated with model inference of at least one individual model within the model cluster.
[0228] In one embodiment, the location server as above, the first message further comprises: model identification information of the at least one individual model that is activated, or PRU identification information of at least one PRU associated with the at least one individual model.
[0229] In one embodiment, the location server as above, the second message further comprises an indication indicating to activate the at least one individual model.
[0230] In one embodiment, the location server as above, the second message further comprises at least one of: model identification information of the at least one individual model, or an activation time for the at least one individual model.
[0231] In one embodiment, the location server as above, the at least one individual model comprises: a plurality of individual models within the model cluster, one or more individual models that are associated with at least one specific subset of PRUs, or a main individual model among the plurality of individual models.
[0232] In one embodiment, the location server as above, the main individual model is one of: a model which is determined based on training data generated by all PRUs in the set of PRUs, a model which is determined based on training data generated by a specific PRU, or a model that indicted by a network entity.
[0233] In one embodiment, the location server as above, the at least one individual model comprises multiple individual models, and wherein the inference result is determined by the terminal device from multiple model outputs of the multiple individual models based on an integration mode.
[0234] In one embodiment, the location server as above, the first message or the second message further comprises the integration mode.
[0235] In one embodiment, the location server as above, the integration mode indicates one of: a weighted sum of the multiple model outputs, or an average of the multiple model outputs.
[0236] In one embodiment, the location server as above, a model output is generated from a second individual model that is determined based on training samples in training dataset, and the training samples are associated with a second subset of PRUs, and wherein, a weight for the model output is related to at least one of: a ratio of a number of the training samples to a total number of samples in the training dataset, a ratio of a number of samples associated with the second subset of PRUs to a total number of samples in the training dataset, or a result of model monitoring for the second individual model.
[0237] The present disclosure provides a communication device, comprising at least one processor configured to cause the communication device at least to: obtain at least one model output of at least one individual model within a model cluster, wherein the model cluster is associated with a set of PRUs, and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; and determine at least one monitoring metric of the at least one individual model or a monitoring metric of the model cluster based on the at least one model output.
[0238] In one embodiment, the communication device as above, the monitoring metric of the model cluster is determined based on one of: an average of one or more monitoring metrics of one or more individual models within the model cluster, or a weighted sum of the one or more monitoring metrics of the one or more individual models within the model cluster.
[0239] In one embodiment, the communication device as above, the one or more individual models comprise one of: all individual models within the model cluster, at least one individual model associated with a specific subset of PRUs, a predefined number of individual models with model outputs closest to a ground truth, a predefined number of individual models with model outputs furthest to a ground truth, or a main individual models within the model cluster.
[0240] In one embodiment, the communication device as above, the main individual model is one of: a model which is determined based on training data generated by all PRUs in the set of PRUs, a model which is determined based on training data generated by a specific PRU, or a model that indicted by a network entity.
[0241] In one embodiment, the communication device as above, a second individual model in the one or more individual models is determined based on training samples in training dataset, and the training samples are associated with a second subset of PRUs, wherein a weight for a monitoring metric of the second individual model in the one or more individual models is related to at least one of: a ratio of a number of the training samples to a total number of samples in the training dataset, a ratio of a number of samples associated with the second subset of PRUs to a total number of samples in the training dataset, or a weight value indicated by a location server.
[0242] In one embodiment, the communication device as above, at least one processor is further configured to cause the communication device to: determine a weight of a second individual model within the model cluster, wherein the weight is used for determining an inference result in a model inference stage.
[0243] In one embodiment, the communication device as above, the weight of the second individual model is determined based on at least one of: a ratio of a monitoring metric of the second individual model to a sum of monitoring metrics of all individual models within the model cluster, or an inverse order of ratios of respective monitoring metrics to a sum of monitoring metrics of all individual models within the model cluster.
[0244] The present disclosure provides a method of communication, comprising the operations implemented at one of: the communication device, the terminal device or the location server.
[0245] The present disclosure provides a device, comprising: a processor; and a memory storing computer program codes; the memory and the computer program codes configured to, with the processor, cause the device to perform the method implemented at one of: the communication device, the terminal device or the location server discussed above.
[0246] The present disclosure provides a non-transitory computer readable storage medium having instructions stored thereon, the instructions, when executed by a processor of an apparatus, cause the apparatus to perform the method implemented at one of: the communication device, the terminal device or the location server discussed above.
[0247] The present disclosure provides a computer program product having instructions stored thereon, the instructions, when executed by a processor of an apparatus, cause the apparatus to perform the method implemented at one of: the communication device, the terminal device or the location server discussed above.
[0248] Generally, various embodiments of the present disclosure may be implemented in hardware or special purpose circuits, software, logic or any combination thereof. Some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software which may be executed by a controller, microprocessor or other computing device. While various aspects of embodiments of the present disclosure are illustrated and described as block diagrams, flowcharts, or using some other pictorial representation, it will be appreciated that the blocks, apparatus, systems, techniques or methods described herein may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof.
[0249] The present disclosure also provides at least one computer program product tangibly stored on a non-transitory computer readable storage medium. The computer program product includes computer-executable instructions, such as those included in program modules, being executed in a device on a target real or virtual processor, to carry out the process or method as described above. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, or the like that perform particular tasks or implement particular abstract data types. The functionality of the program modules may be combined or split between program modules as desired in various embodiments. Machine-executable instructions for program modules may be executed within a local or distributed device. In a distributed device, program modules may be located in both local and remote storage media.
[0250] Program code for carrying out methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may execute entirely on a machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
[0251] The above program code may be embodied on a machine readable medium, which may be any tangible medium that may contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine readable medium may be a machine readable signal medium or a machine readable storage medium. A machine readable medium may include but not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine readable storage medium would include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM) , a read-only memory (ROM) , an erasable programmable read-only memory (EPROM or Flash memory) , an optical fiber, a portable compact disc read-only memory (CD-ROM) , an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0252] Further, while operations are depicted in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Likewise, while several specific implementation details are contained in the above discussions, these should not be construed as limitations on the scope of the present disclosure, but rather as descriptions of features that may be specific to particular embodiments. Certain features that are described in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment may also be implemented in multiple embodiments separately or in any suitable sub-combination.
[0253] Although the present disclosure has been described in language specific to structural features and / or methodological acts, it is to be understood that the present disclosure defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.
Claims
1.A communication device comprising at least one processor configured to cause the communication device to:obtain a training dataset comprising a plurality of samples, wherein the training dataset is generated at least based on measurements from a set of positioning reference units (PRU) ; anddetermine at least one individual model within a model cluster based on at least part of the training dataset, wherein a first individual model of the at least one individual model has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model.2.The communication device of claim 1, wherein the first individual model within the model cluster is determined based on first training data associated with the first subset of PRUs, and a second individual model within the model cluster is determined based second training data associated with a second subset of PRUs with or without an intersection with the first subset.3.The communication device of claim 1, wherein the model cluster is associated with the set of PRUs, and wherein the set of PRUs involved in training data collection for a same model cluster has a feature of at least one of the following:being provided by a same vendor,being located within a same area,having same or similar capabilities at least for training data collection, orhaving a same sign that is marked by a network entity.4.The communication device of claim 1, wherein the communication device comprises a terminal device, and wherein the at least one processor is further configured to cause the communication device to:receive, from a network entity, information about a number of the at least one individual model or about a range of the number of the at least one individual model.5.The communication device of claim 4, wherein the number of the at least one individual model or the range of the number of the at least one individual model is associated with at least one of:a number of PRUs in the set of PRUs,a number of PRUs in the first subset of PRUs,a number of PRUs in a second subset of PRUs, oran intersection of the first subset and the second subset.6.The communication device of claim 4, wherein a maximum value for the number of the at least one individual model or for the range is associated with a number of PRUs in the set of PRUs.7.The communication device of claim 1, wherein the communication device comprises a network entity, and wherein the at least one processor is further configured to cause the communication device to:transmit, to a terminal device, the at least one individual model within the model cluster, wherein each of the at least one individual model has corresponding model identification information.8.The communication device of claim 1, wherein PRU identification information of a PRU in the set of PRUs is:an identifier of the PRU, ora virtual identification of the PRU that is allocated by a network entity.9.The communication device of claim 1, wherein the model identification information of the first individual model is:a combination or a function of PRU identification information of PRUs in the first subset of PRUs,an extension or a function of cluster identification information of the model cluster, ora first identifier (ID) that is predefined or assigned by a network entity.10.The communication device of claim 9, wherein the cluster identification information of the model cluster is determined based on at least one of:a combination or a function of PRU identification information of all PRUs in the set of PRUs,model identification information of a main individual model within the model cluster, ora second ID that is predefined or assigned by a network entity.11.The communication device of claim 10, wherein the main individual model is one of:a model which is determined based on training data generated by all PRUs in the set of PRUs,a model which is determined based on training data generated by a specific PRU, ora model that indicted by a network entity.12.A terminal device comprising at least one processor configured to cause the terminal device to:activate at least one individual model within a model cluster, wherein the model cluster is associated with a set of positioning reference units (PRU) , and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model;perform a model inference by using the activated at least one individual model based on a model input, to determine an inference result; andtransmit, to a location server, a first message comprising the inference result.13.The terminal device of claim 12, wherein the at least one processor is further configured to cause the terminal device to:receive, from the location server, a second message comprising an indication that the model cluster should be activated.14.The terminal device of claim 12, wherein the activated at least one individual model comprises:a plurality of individual models within the model cluster,one or more individual models that are associated with at least one specific subset of PRUs, ora main individual model among the plurality of individual models.15.The terminal device of claim 14, wherein the main individual model is one of:a model which is determined based on training data generated by all PRUs in the set of PRUs,a model which is determined based on training data generated by a specific PRU, ora model that indicted by a network entity.16.The terminal device of claim 12, wherein the first message further comprises:model identification information of the activated at least one individual model, orPRU identification information of at least one PRU associated with the activated at least one individual model.17.The terminal device of claim 13, wherein the second message further comprises an indication indicating to activate the at least one individual model.18.The terminal device of claim 17, wherein the second message further comprises at least one of:model identification information of the at least one individual model, oran activation time for the at least one individual model.19.The terminal device of claim 12, wherein the at least one individual model comprises multiple individual models, and wherein the at least one processor is configured to cause the terminal device to:determine multiple model outputs of the multiple individual models respectively based on the model input; anddetermine the inference result of the model cluster based on the multiple model outputs using an integration mode.20.The terminal device of claim 19, wherein the integration mode indicates one of:a weighted sum of the multiple model outputs, oran average of the multiple model outputs.21.The terminal device of claim 20, wherein a model output is generated by a second individual model that is determined based on training samples in training dataset, and the training samples are associated with a second subset of PRUs, and wherein,a weight for the model output is related to at least one of:a ratio of a number of the training samples to a total number of samples in the training dataset,a ratio of a number of samples associated with the second subset of PRUs to a total number of samples in the training dataset, ora result of model monitoring for the second individual model.22.The terminal device of claim 19, wherein the first message further comprises the integration mode.23.The terminal device of claim 13, wherein the second message further comprises information of an integration mode which is used for determining the inference result.24.A location server comprising at least one processor configured to cause the location server to:transmit, to a terminal device, a second message comprising an indication that a model cluster should be activated, wherein the model cluster is associated with a set of positioning reference units (PRU) , and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; andreceive, from the terminal device, a first message comprising an inference result associated with model inference of at least one individual model within the model cluster.25.The location server of claim 24, wherein the first message further comprises:model identification information of the at least one individual model that is activated, orPRU identification information of at least one PRU associated with the at least one individual model.26.The location server of claim 24, wherein the second message further comprises an indication indicating to activate the at least one individual model.27.The location server of claim 26, wherein the second message further comprises at least one of:model identification information of the at least one individual model, oran activation time for the at least one individual model.28.The location server of claim 24, wherein the at least one individual model comprises:a plurality of individual models within the model cluster,one or more individual models that are associated with at least one specific subset of PRUs, ora main individual model among the plurality of individual models.29.The location server of claim 28, wherein the main individual model is one of:a model which is determined based on training data generated by all PRUs in the set of PRUs,a model which is determined based on training data generated by a specific PRU, ora model that indicted by a network entity.30.The location server of claim 24, wherein the at least one individual model comprises multiple individual models, and wherein the inference result is determined by the terminal device from multiple model outputs of the multiple individual models based on an integration mode.31.The location server of claim 30, wherein the first message or the second message further comprises the integration mode.32.The location server of claim 30, wherein the integration mode indicates one of:a weighted sum of the multiple model outputs, oran average of the multiple model outputs.33.The location server of claim 32, wherein a model output is generated from a second individual model that is determined based on training samples in training dataset, and the training samples are associated with a second subset of PRUs, and wherein,a weight for the model output is related to at least one of:a ratio of a number of the training samples to a total number of samples in the training dataset,a ratio of a number of samples associated with the second subset of PRUs to a total number of samples in the training dataset, ora result of model monitoring for the second individual model.34.A communication device comprising at least one processor configured to cause the communication device to:obtain at least one model output of at least one individual model within a model cluster, wherein the model cluster is associated with a set of positioning reference units (PRU) , and wherein a first individual model within the model cluster has model identification information that is associated with a first subset of PRUs involved in training data collection of the first individual model; anddetermine at least one monitoring metric of the at least one individual model or a monitoring metric of the model cluster based on the at least one model output.35.The communication device of claim 34, wherein the monitoring metric of the model cluster is determined based on one of:an average of one or more monitoring metrics of one or more individual models within the model cluster, ora weighted sum of the one or more monitoring metrics of the one or more individual models within the model cluster.36.The communication device of claim 35, wherein the one or more individual models comprise one of:all individual models within the model cluster,at least one individual model associated with a specific subset of PRUs,a predefined number of individual models with model outputs closest to a ground truth,a predefined number of individual models with model outputs furthest to a ground truth, ora main individual models within the model cluster.37.The communication device of claim 36, wherein the main individual model is one of:a model which is determined based on training data generated by all PRUs in the set of PRUs,a model which is determined based on training data generated by a specific PRU, ora model that indicted by a network entity.38.The communication device of claim 35, wherein a second individual model in the one or more individual models is determined based on training samples in training dataset, and the training samples are associated with a second subset of PRUs,wherein a weight for a monitoring metric of the second individual model in the one or more individual models is related to at least one of:a ratio of a number of the training samples to a total number of samples in the training dataset,a ratio of a number of samples associated with the second subset of PRUs to a total number of samples in the training dataset, ora weight value indicated by a location server.39.The communication device of claim 34, wherein at least one processor is further configured to cause the communication device to:determine a weight of a second individual model within the model cluster, wherein the weight is used for determining an inference result in a model inference stage.40.The communication device of claim 39, wherein the weight of the second individual model is determined based on at least one of:a ratio of a monitoring metric of the second individual model to a sum of monitoring metrics of all individual models within the model cluster, oran inverse order of ratios of respective monitoring metrics to a sum of monitoring metrics of all individual models within the model cluster.